初始化项目,由ModelHub XC社区提供模型

Model: W-61/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725
Source: Original Platform
This commit is contained in:
ModelHub XC
2026-05-29 21:07:24 +08:00
commit d9d1982718
501 changed files with 166650 additions and 0 deletions

677
margin_logs/margins.jsonl Normal file
View File

@@ -0,0 +1,677 @@
{"epoch": 0.0, "step": 1, "batch_size": 128, "mean": 0.01704716682434082, "std": 0.7752009630203247, "min": -2.2227783203125, "p10": -0.8393455505371094, "median": 0.0678863525390625, "p90": 0.8599296569824219, "max": 2.196014404296875, "pos_frac": 0.5546875, "sample": [0.6739997863769531, -0.3828582763671875, -1.298828125, 0.071014404296875, 0.8088531494140625, 0.6873779296875, -0.081512451171875, 1.77093505859375, -1.4920806884765625, 1.769927978515625, 0.0503692626953125, -0.1540679931640625, 0.15589141845703125, -0.27809906005859375, -0.009723663330078125, 0.3800621032714844, 0.111541748046875, 0.2325439453125, 1.35693359375, -0.482269287109375, 0.2686614990234375, 1.01312255859375, -0.5781745910644531, -0.1348876953125, 0.15979766845703125, -1.054656982421875, 0.01500701904296875, 0.2495269775390625, 0.5402507781982422, 1.80511474609375, -0.8189926147460938, -0.17108154296875, 0.2899322509765625, -0.8205108642578125, 0.650146484375, -0.6688079833984375, 0.080841064453125, -1.4810791015625, 0.07396697998046875, -1.242919921875, 0.086151123046875, 1.62188720703125, -0.38132476806640625, 0.317474365234375, 2.196014404296875, 0.5562095642089844, 0.1424732208251953, -0.7183074951171875, -0.6070556640625, -0.6271820068359375, -0.5325164794921875, 0.67987060546875, -1.229156494140625, -1.91693115234375, -0.143280029296875, 0.2532958984375, -0.349273681640625, 0.42700958251953125, -0.262420654296875, 1.2039794921875, 0.364410400390625, 0.8435211181640625, 0.00079345703125, -0.19837188720703125, -0.5950164794921875, 0.362579345703125, 0.7775421142578125, -0.015625, -0.0785369873046875, -0.6810150146484375, 0.31806182861328125, -0.27123260498046875, 0.00982666015625, 0.098541259765625, -1.4693603515625, 0.20909881591796875, 0.87872314453125, 0.0, 0.4891510009765625, 0.100677490234375, -0.3092498779296875, 0.839263916015625, -0.5568389892578125, -0.5927276611328125, 1.82940673828125, -1.8787841796875, 0.4121856689453125, 0.1029205322265625, -0.05619049072265625, 0.41302490234375, -0.20123291015625, -1.11163330078125, 0.11328125, -0.6293182373046875, -0.4356689453125, 1.026641845703125, -0.4438629150390625, 0.06475830078125, 0.15139389038085938, 0.4244384765625, -0.3410911560058594, -0.8832931518554688, 0.38405609130859375, -0.6134033203125, 0.14745521545410156, -2.2227783203125, -0.3340606689453125, -0.4396820068359375, 0.543609619140625, 0.310455322265625, -0.6277847290039062, 0.28015899658203125, 0.1476898193359375, 0.3895263671875, 0.0498046875, -0.1731414794921875, 0.24071502685546875, -0.351348876953125, -0.0234375, 0.6951370239257812, 0.0606689453125, 0.8756256103515625, 0.680145263671875, 0.14440345764160156, 0.8532028198242188, -1.6363525390625, 1.4149551391601562, -0.476959228515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000001.npy"}
{"epoch": 0.0020942408376963353, "step": 2, "batch_size": 128, "mean": 0.08979731798171997, "std": 1.0063953399658203, "min": -2.3612060546875, "p10": -1.1013275146484374, "median": 0.015979766845703125, "p90": 1.2868972778320311, "max": 4.49609375, "pos_frac": 0.53125, "sample": [0.006134033203125, 0.6754608154296875, 0.06409072875976562, 0.3348197937011719, 0.104522705078125, 0.848785400390625, -1.3365631103515625, 0.40827178955078125, 1.093963623046875, -0.233673095703125, 0.2210693359375, 0.34949493408203125, 1.295623779296875, -0.01953887939453125, -0.0232696533203125, -0.88238525390625, 1.92608642578125, -0.0975494384765625, -0.0151214599609375, -0.137054443359375, 0.18506240844726562, -0.7067947387695312, -1.04547119140625, 0.01041412353515625, 0.3020477294921875, -0.03971099853515625, -0.283477783203125, 0.6910400390625, 0.15057373046875, -0.1564178466796875, 0.561614990234375, 0.05670166015625, -0.48236846923828125, 1.7666015625, -0.0575408935546875, -1.0196533203125, -2.3612060546875, 2.6371917724609375, -0.6959686279296875, 0.836669921875, -0.16046142578125, -0.033050537109375, 0.0088958740234375, 1.2831573486328125, 1.604461669921875, -0.9659347534179688, 0.1966876983642578, -1.5263671875, -0.4609222412109375, -0.87158203125, 4.49609375, -0.16178131103515625, -1.54376220703125, 0.0208740234375, 0.0, 0.33994483947753906, 0.185943603515625, -0.30196380615234375, 3.5068359375, 1.463623046875, 0.9294891357421875, -0.02225494384765625, -1.140228271484375, 0.1556549072265625, -0.2645263671875, 0.90374755859375, 0.5412216186523438, -1.59906005859375, -0.3666229248046875, -0.4569854736328125, -0.6967010498046875, 0.36395263671875, 0.13336181640625, -0.068084716796875, -0.563751220703125, 0.5245513916015625, 0.082733154296875, -0.73480224609375, 0.570953369140625, 0.2091522216796875, -0.021331787109375, -0.547821044921875, 1.6211090087890625, 0.8912696838378906, 0.40868377685546875, 2.679595947265625, 0.1652679443359375, -0.8451919555664062, -0.22149658203125, -0.6046524047851562, -0.001617431640625, 0.72979736328125, -1.3320388793945312, -0.18963623046875, 1.436309814453125, 0.23449325561523438, 0.4606475830078125, 0.095184326171875, 0.49684906005859375, -0.4890289306640625, 0.9012298583984375, 1.9844970703125, 1.4336700439453125, -0.7718658447265625, 0.1373443603515625, -1.352508544921875, -1.7950439453125, 0.286163330078125, -0.2695465087890625, 0.72283935546875, 0.463165283203125, -1.27398681640625, 0.45489501953125, 0.01837158203125, 0.01849365234375, 0.36487579345703125, -1.4290771484375, 0.8643035888671875, -0.195465087890625, 0.240478515625, -0.7327880859375, -1.255859375, -0.87091064453125, -1.08465576171875, 1.1551513671875, 0.01358795166015625, -0.0644378662109375, -1.950225830078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000002.npy"}
{"epoch": 0.004188481675392671, "step": 3, "batch_size": 128, "mean": 0.09388026595115662, "std": 0.8561777472496033, "min": -2.092803955078125, "p10": -1.0053253173828125, "median": 0.05345344543457031, "p90": 1.0742858886718747, "max": 2.891204833984375, "pos_frac": 0.515625, "sample": [1.579620361328125, 0.06438827514648438, -0.1837158203125, -0.24857330322265625, 0.2940826416015625, -1.113006591796875, 0.5953369140625, 1.041351318359375, 0.0, -0.2435302734375, 1.27264404296875, -0.237274169921875, -0.36572265625, 0.7158203125, -0.33465576171875, 0.208099365234375, -1.142486572265625, 2.274658203125, 0.9187164306640625, -0.4959716796875, -0.6342010498046875, 0.79620361328125, 0.1468505859375, 0.18906402587890625, 0.578277587890625, -0.12957763671875, 1.76129150390625, -1.76507568359375, 0.79443359375, 0.657806396484375, -0.34604644775390625, 0.08338546752929688, 1.13848876953125, 0.8751220703125, -1.01226806640625, -0.30718994140625, -0.11798095703125, -0.24468994140625, -0.449554443359375, 1.018951416015625, -0.23712158203125, -1.13116455078125, -0.1181640625, 1.0224609375, 0.09332275390625, -1.401214599609375, -0.0321044921875, 1.052734375, -0.462921142578125, -0.592559814453125, -0.48675537109375, 0.22867965698242188, 0.857330322265625, 0.34612274169921875, 0.457275390625, 1.12457275390625, 0.5429840087890625, 0.59381103515625, 0.3558235168457031, 0.9729423522949219, 0.0, 0.4595947265625, 2.12347412109375, -1.209320068359375, -0.6443252563476562, -1.71588134765625, -1.223114013671875, -1.362640380859375, 0.09748458862304688, 0.053462982177734375, -0.671630859375, 0.246795654296875, 2.891204833984375, -0.88104248046875, -0.24415206909179688, 0.15576171875, 0.5763397216796875, -0.03057861328125, -1.002349853515625, -0.545318603515625, -0.0377655029296875, 0.1877593994140625, 0.9709091186523438, -0.3429679870605469, -0.128173828125, 1.754241943359375, -2.092803955078125, 0.4043731689453125, -0.1960296630859375, 0.3595123291015625, 0.262420654296875, -0.480743408203125, -0.3489990234375, -1.684326171875, 0.3990592956542969, 0.272796630859375, 0.12740325927734375, 0.7347030639648438, 0.6725616455078125, 0.21966171264648438, -0.2896232604980469, -0.486083984375, 0.38018035888671875, 0.583831787109375, -0.390777587890625, 1.17974853515625, -0.24166107177734375, 0.7689208984375, -0.227203369140625, -0.19376373291015625, -0.89105224609375, -0.631744384765625, 1.430084228515625, 0.6939697265625, 0.165985107421875, 0.10103607177734375, -0.9041748046875, -0.1480865478515625, -0.5684814453125, -0.111053466796875, 2.095458984375, 0.05344390869140625, 0.0230560302734375, -0.047695159912109375, -1.4273605346679688, 1.837615966796875, 0.43817901611328125, -0.12255859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000003.npy"}
{"epoch": 0.0062827225130890054, "step": 4, "batch_size": 128, "mean": 0.1393764317035675, "std": 1.0018424987792969, "min": -3.38116455078125, "p10": -1.0416038513183592, "median": 0.0640411376953125, "p90": 1.2722869873046874, "max": 2.75872802734375, "pos_frac": 0.578125, "sample": [-0.455413818359375, 0.8123016357421875, 0.0003566741943359375, 1.7396240234375, 1.45245361328125, -0.052276611328125, 0.5045928955078125, 0.4036865234375, -0.7919921875, 0.925384521484375, -1.27154541015625, -0.73638916015625, -0.0814056396484375, -0.4066314697265625, 1.0106201171875, 1.01446533203125, 0.95867919921875, -1.588836669921875, 0.983734130859375, 1.328125, -0.08355712890625, 0.4932861328125, 0.6616058349609375, -0.18963623046875, 0.010833740234375, 0.2007598876953125, 2.75872802734375, 0.6265754699707031, 0.9425048828125, -0.481689453125, -1.44732666015625, -0.0075206756591796875, -1.62799072265625, -0.706146240234375, 0.798675537109375, 0.3099365234375, 0.4140777587890625, -1.41082763671875, 0.9796295166015625, -0.199554443359375, -0.0014019012451171875, 0.07720947265625, -0.3257770538330078, -0.080108642578125, 0.27581787109375, -0.447174072265625, 0.3703765869140625, -0.3725738525390625, -0.305511474609375, -0.69366455078125, 1.295166015625, -0.18569183349609375, -1.8675537109375, -1.440032958984375, 0.991607666015625, -0.02564239501953125, -0.22564697265625, 1.262481689453125, 0.40814208984375, 0.636077880859375, 0.0401611328125, -0.9462738037109375, 0.1271209716796875, 1.359100341796875, 0.904754638671875, 0.0404052734375, 1.074432373046875, -0.9797210693359375, -0.0700531005859375, -0.615234375, 0.7659149169921875, 1.45672607421875, 1.21533203125, -1.677001953125, 0.833892822265625, -0.3087158203125, -3.38116455078125, 1.165069580078125, -0.90179443359375, -0.255615234375, 0.0235137939453125, -0.35005950927734375, 1.181640625, 0.0186767578125, -0.6519927978515625, 0.277313232421875, 0.447021484375, -0.740234375, 0.57550048828125, 0.019317626953125, 0.01514434814453125, 0.08502197265625, -1.0196380615234375, -0.252960205078125, -0.448638916015625, 0.21441650390625, 2.4063720703125, -0.3287086486816406, -1.0928573608398438, 1.2525634765625, -0.59765625, 0.050872802734375, 0.6868095397949219, -0.057342529296875, 1.026702880859375, -3.307708740234375, 0.02288055419921875, -1.3895111083984375, 0.84893798828125, 1.40716552734375, 0.8185577392578125, 0.328094482421875, 0.14864349365234375, 2.395263671875, 0.60369873046875, 0.66204833984375, -1.99774169921875, 0.37469482421875, 1.9969482421875, -0.6828842163085938, -0.3160247802734375, 0.725341796875, 1.651092529296875, 1.11993408203125, -0.1273193359375, 1.329559326171875, 1.05682373046875, 0.445556640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000004.npy"}
{"epoch": 0.008376963350785341, "step": 5, "batch_size": 128, "mean": 0.008981779217720032, "std": 0.8337785005569458, "min": -2.77252197265625, "p10": -0.9393585205078124, "median": 0.08347129821777344, "p90": 0.8959831237792969, "max": 2.5108642578125, "pos_frac": 0.5390625, "sample": [-0.810211181640625, -1.22369384765625, 0.39415740966796875, -0.017364501953125, 0.24839019775390625, 0.8973770141601562, 0.670074462890625, -0.4011688232421875, 0.2622528076171875, 0.441650390625, -0.4049072265625, -0.948394775390625, -0.5725250244140625, -1.2859344482421875, -0.350860595703125, -0.368194580078125, -0.24896240234375, 0.436859130859375, 0.464599609375, -2.77252197265625, 0.258819580078125, -0.05828857421875, -0.24291229248046875, -0.47265625, -0.689178466796875, 0.179901123046875, -0.0040435791015625, 0.5162773132324219, -0.44181060791015625, -0.12175750732421875, 1.287200927734375, -0.88623046875, 0.117706298828125, 0.1282501220703125, 0.9406890869140625, 0.75933837890625, 1.36810302734375, 0.2733154296875, -2.59161376953125, 1.0950927734375, 0.86676025390625, 1.11480712890625, 0.6025390625, 0.374786376953125, -0.816162109375, -0.09228706359863281, -1.868316650390625, 0.3828125, -0.47046661376953125, 1.62298583984375, 0.77239990234375, -0.98388671875, -0.0533447265625, -0.3490486145019531, 0.411651611328125, -0.56292724609375, -0.1337890625, -0.2046356201171875, 0.5103759765625, 1.4818572998046875, -1.4117431640625, -0.040924072265625, 0.4114227294921875, -0.17586135864257812, -0.1492462158203125, 0.06953048706054688, 1.2446441650390625, 0.86700439453125, -0.425994873046875, -0.544342041015625, -1.582977294921875, 0.0415802001953125, 0.16254425048828125, 0.0974884033203125, 0.05422401428222656, -0.93548583984375, -0.8572998046875, 1.51092529296875, 0.557464599609375, 0.887481689453125, -0.426361083984375, 0.1654052734375, 1.2938232421875, -2.244964599609375, 0.557861328125, -1.310394287109375, 0.8953857421875, -0.77294921875, 0.50067138671875, 0.1568145751953125, -1.876129150390625, -0.05225944519042969, -0.872955322265625, 0.7779541015625, 0.38189697265625, 0.0497589111328125, -1.8399658203125, -0.02874755859375, -0.1989288330078125, -0.118804931640625, 0.6500244140625, 0.83489990234375, 0.28278350830078125, 0.136932373046875, -0.6753768920898438, 0.097412109375, 0.664794921875, 0.22077560424804688, -0.27573394775390625, 0.240264892578125, -0.2295684814453125, 0.28856658935546875, 0.13112640380859375, 0.6578216552734375, 0.4985198974609375, -0.3686561584472656, 0.0612640380859375, 0.1621856689453125, 0.672515869140625, -0.120086669921875, 0.92413330078125, -0.5507354736328125, 0.848419189453125, -0.231536865234375, 0.28973388671875, 0.437225341796875, -0.2593841552734375, 2.5108642578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000005.npy"}
{"epoch": 0.010471204188481676, "step": 6, "batch_size": 128, "mean": -0.08054298162460327, "std": 0.9391193389892578, "min": -2.328277587890625, "p10": -1.357867431640625, "median": 0.025381088256835938, "p90": 0.8278564453124997, "max": 2.9314422607421875, "pos_frac": 0.515625, "sample": [0.2197723388671875, 0.762298583984375, -0.6783676147460938, 0.78271484375, 0.1318206787109375, 0.6714019775390625, 0.706451416015625, 0.01807403564453125, -0.32852935791015625, -0.48773193359375, -1.0948028564453125, 1.7124481201171875, 0.1591339111328125, 0.4195556640625, -0.1827392578125, 0.20941925048828125, 0.694091796875, 0.032688140869140625, -0.579681396484375, 0.072418212890625, 1.410614013671875, 0.32794189453125, 0.592376708984375, 0.46881103515625, 0.3541259765625, 0.551116943359375, 0.343017578125, -0.41082763671875, 0.760040283203125, 0.34601593017578125, -0.15515899658203125, -0.8931427001953125, -1.595794677734375, 0.22509765625, -0.1197967529296875, -1.629058837890625, -0.99871826171875, 0.157958984375, -0.923004150390625, 2.311767578125, -1.095123291015625, -0.566802978515625, -1.490692138671875, -2.10089111328125, -0.1817169189453125, -1.78289794921875, 0.10721588134765625, 0.3743896484375, 0.153961181640625, -0.00244140625, -0.4952392578125, 0.0099639892578125, 0.791839599609375, -0.670257568359375, 0.732177734375, 0.06702423095703125, -0.68597412109375, -0.87774658203125, -0.523712158203125, 1.5269775390625, 0.28743934631347656, 0.14227294921875, 1.0218582153320312, 0.38513946533203125, -1.4500732421875, 0.192840576171875, -0.04648780822753906, 1.4281005859375, 1.10784912109375, 0.059902191162109375, 2.9314422607421875, 0.1297149658203125, 0.911895751953125, -1.0472412109375, -0.07006072998046875, 0.150054931640625, 0.2042999267578125, -1.3216476440429688, 0.0, 0.274169921875, 0.360748291015625, -0.84246826171875, -0.4842376708984375, -0.5683937072753906, -0.6961212158203125, 0.63775634765625, 0.4258270263671875, -1.3365478515625, -0.14361953735351562, 0.03845977783203125, -0.5078125, 0.03436279296875, 0.5563201904296875, 0.156280517578125, -1.8001708984375, 0.03717041015625, -0.4578399658203125, -0.5415191650390625, -0.834869384765625, -1.3912353515625, 1.48150634765625, -1.00140380859375, -0.7577972412109375, -0.4273681640625, -0.9534072875976562, 0.6095123291015625, -2.328277587890625, -1.80426025390625, -0.492950439453125, -0.26604461669921875, -1.34356689453125, 0.46797943115234375, 2.801971435546875, -0.11460113525390625, 0.477081298828125, 1.3577880859375, 0.75860595703125, 0.469879150390625, -2.082550048828125, 0.2105255126953125, -0.2259063720703125, -1.5562744140625, -0.1906280517578125, -0.32659912109375, 1.607208251953125, -1.95831298828125, -0.15644073486328125, -0.1526031494140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000006.npy"}
{"epoch": 0.012565445026178011, "step": 7, "batch_size": 128, "mean": -0.07205848395824432, "std": 0.974696934223175, "min": -3.857086181640625, "p10": -1.3487228393554687, "median": -0.000598907470703125, "p90": 0.9837768554687499, "max": 3.02362060546875, "pos_frac": 0.4921875, "sample": [0.754425048828125, -0.953857421875, -3.857086181640625, -0.79754638671875, 0.2996368408203125, -0.5697784423828125, 0.24737548828125, -1.3787841796875, -1.0072021484375, 0.04787445068359375, -0.0133514404296875, 0.5090484619140625, -1.47149658203125, 0.7015762329101562, 0.1859130859375, -1.902557373046875, -0.7295684814453125, 0.700836181640625, 0.0, 0.0073947906494140625, 0.427154541015625, -0.5087127685546875, -0.6514511108398438, -0.2835845947265625, -0.231201171875, -0.462890625, -0.90673828125, -0.6053619384765625, 2.324066162109375, 0.7270965576171875, -0.0691070556640625, 0.032257080078125, -0.7901840209960938, 0.5626983642578125, 0.14239501953125, 3.02362060546875, 0.09476470947265625, 1.00238037109375, -0.5111083984375, 0.04157257080078125, -0.26971435546875, -0.13549041748046875, -0.925506591796875, 1.99560546875, -0.14984130859375, 0.729583740234375, 2.255340576171875, -0.008251190185546875, 0.04041099548339844, 0.60052490234375, 1.257354736328125, -1.6082763671875, 0.383575439453125, 0.2085723876953125, -1.58160400390625, 0.413116455078125, -0.0467529296875, -0.231353759765625, -0.615478515625, 1.6248016357421875, -1.3438568115234375, 1.00146484375, 0.2582283020019531, -0.267669677734375, -1.6103363037109375, 0.704925537109375, -2.089599609375, 0.24831390380859375, 1.12530517578125, -0.766387939453125, -1.0022125244140625, -0.5494384765625, 0.254669189453125, -0.35040283203125, -0.30066680908203125, -1.447235107421875, -0.04090118408203125, -1.762176513671875, -2.86962890625, -1.01727294921875, 0.47344970703125, 0.233673095703125, -0.1005706787109375, 0.1675872802734375, 0.30535888671875, -0.00119781494140625, -0.55767822265625, -0.831634521484375, 1.54803466796875, -0.34017181396484375, 0.23309326171875, 0.0855865478515625, -0.09946441650390625, -0.5320091247558594, 0.180694580078125, -0.2520904541015625, -0.064666748046875, 0.018266677856445312, 0.130462646484375, -1.360076904296875, 1.529327392578125, -0.1200103759765625, 0.106231689453125, 0.267913818359375, 0.79766845703125, 0.885467529296875, 0.25970458984375, 0.85943603515625, -1.28076171875, 0.9761962890625, -0.1285247802734375, -0.3565559387207031, 1.1785888671875, 0.2155303955078125, -0.2647857666015625, 0.137725830078125, -0.33258056640625, 0.2371978759765625, 0.827545166015625, 1.859375, -0.661224365234375, -0.19500732421875, 0.04770660400390625, 0.05819129943847656, -0.06862831115722656, -2.21978759765625, 0.2151641845703125, 0.4665069580078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000007.npy"}
{"epoch": 0.014659685863874346, "step": 8, "batch_size": 128, "mean": -0.07885238528251648, "std": 1.002476692199707, "min": -4.14361572265625, "p10": -1.3391586303710936, "median": 0.0030984878540039062, "p90": 0.9349853515624998, "max": 3.17205810546875, "pos_frac": 0.5, "sample": [-0.349365234375, 0.512908935546875, -0.561859130859375, -0.9130859375, 0.32843017578125, -1.609130859375, 0.76385498046875, 0.029756546020507812, -0.038936614990234375, -1.821502685546875, -0.44370269775390625, -1.208587646484375, 0.4720458984375, 0.48883056640625, 0.21826171875, 0.2404022216796875, -0.127593994140625, 0.71844482421875, -0.24451255798339844, 0.49884033203125, -0.26812744140625, 2.29296875, -1.35845947265625, 1.118408203125, 0.64898681640625, -0.2578125, 1.3790283203125, -0.115875244140625, -2.1978759765625, -2.516357421875, -1.4343719482421875, -0.28912353515625, 0.115966796875, -0.739501953125, 1.36492919921875, -1.083221435546875, -0.6681976318359375, -0.3173370361328125, -1.239990234375, -1.227508544921875, 1.2930908203125, -1.714630126953125, 1.095977783203125, -0.94024658203125, -0.158355712890625, -0.713348388671875, -0.13525772094726562, 0.6053466796875, 0.119384765625, 1.695526123046875, -0.4871978759765625, 0.383941650390625, 0.391326904296875, -0.2775135040283203, -0.2443695068359375, -0.8072052001953125, -1.4988861083984375, 2.40325927734375, 0.332366943359375, -4.14361572265625, -0.637939453125, -0.1348876953125, -0.3203125, 0.9200439453125, 0.2502593994140625, 0.902099609375, -1.3308868408203125, 0.2747039794921875, -1.02056884765625, 3.17205810546875, 0.5631103515625, 0.46484375, 0.1255035400390625, -0.538330078125, 1.674163818359375, -0.0235595703125, 0.786590576171875, 0.18833541870117188, 0.08075714111328125, -1.1883087158203125, 0.20166015625, 0.5677490234375, -0.6451416015625, 0.25051116943359375, 0.404937744140625, 0.21006393432617188, -1.3784942626953125, 0.9698486328125, 0.0555877685546875, 1.322906494140625, 0.33404541015625, -1.836669921875, 0.818389892578125, 0.20005035400390625, -0.0392608642578125, 0.66864013671875, 0.68328857421875, -0.07232666015625, 0.22289657592773438, -1.1649169921875, 0.6792716979980469, 0.737060546875, -0.2496337890625, -0.20367431640625, -0.15761184692382812, -0.055385589599609375, -0.197479248046875, 0.15279388427734375, -0.86163330078125, -2.011810302734375, -0.4711151123046875, -0.4776611328125, 1.26995849609375, 0.58172607421875, -0.665435791015625, 0.540985107421875, -1.25006103515625, -1.72515869140625, 0.39994049072265625, 0.629180908203125, -1.217193603515625, 0.680328369140625, 0.031036376953125, 0.09535980224609375, 0.15074920654296875, -0.2292633056640625, 0.623748779296875, -0.23319053649902344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000008.npy"}
{"epoch": 0.016753926701570682, "step": 9, "batch_size": 128, "mean": 0.0003725886344909668, "std": 1.0050214529037476, "min": -3.0142822265625, "p10": -1.1762359619140625, "median": 0.0, "p90": 1.1085388183593747, "max": 3.69854736328125, "pos_frac": 0.4921875, "sample": [-1.09197998046875, -0.39251708984375, -0.07622528076171875, -0.18659591674804688, -0.0660400390625, 0.243621826171875, -0.9307098388671875, -0.7250251770019531, 1.66119384765625, 0.021026611328125, 0.19586181640625, -1.04388427734375, 0.466033935546875, -0.7908935546875, -0.89306640625, -0.4652099609375, 1.250701904296875, -0.5164031982421875, 0.9379196166992188, -0.206085205078125, 0.9015312194824219, 2.610260009765625, 0.27600860595703125, 0.681884765625, 0.0735015869140625, -0.31629180908203125, -1.4136962890625, 0.4723358154296875, 0.8986358642578125, -2.79803466796875, -1.1005859375, 1.0423583984375, -0.3642120361328125, -0.2821197509765625, -0.450042724609375, -0.688812255859375, 1.93853759765625, -0.348114013671875, 0.5286712646484375, -0.0380859375, 0.1859130859375, -0.25647735595703125, 1.073089599609375, -0.0077056884765625, -0.978424072265625, -1.793548583984375, -1.17529296875, -1.178436279296875, 0.2894287109375, -0.5372505187988281, 0.03961181640625, -0.745941162109375, -0.83990478515625, 0.39996337890625, 0.003692626953125, 0.5644683837890625, 1.39813232421875, 0.921875, -1.8695068359375, 0.9984130859375, -1.3861541748046875, 0.07880401611328125, -3.0142822265625, -0.11806106567382812, 0.15262603759765625, 2.219970703125, -0.02205657958984375, 1.191253662109375, 0.418487548828125, 0.21772003173828125, -0.66796875, -0.0002994537353515625, -1.05517578125, 0.8602027893066406, -1.7510833740234375, 0.14971923828125, 0.025909423828125, 0.18206024169921875, 0.20013427734375, 0.55975341796875, -0.81475830078125, 0.850189208984375, -0.24468994140625, -0.1880645751953125, -0.50897216796875, -1.3354339599609375, 0.0, -0.3110198974609375, 0.441497802734375, 0.372406005859375, -0.2459716796875, 0.1081390380859375, 0.08544921875, 0.02518463134765625, 1.5562286376953125, 0.6107902526855469, -1.0225830078125, 0.29644775390625, 0.26258087158203125, 0.9161224365234375, 0.012725830078125, 0.0, 2.56878662109375, 0.563720703125, -0.36761474609375, -0.11578559875488281, 0.0, 1.303955078125, -0.4503593444824219, -0.6289482116699219, 1.4678955078125, 0.8880844116210938, -1.31561279296875, -0.719879150390625, -1.450042724609375, 0.08152008056640625, -1.599365234375, 0.300537109375, -0.725189208984375, -0.0421142578125, 0.456024169921875, 3.69854736328125, 0.42133331298828125, -0.06536865234375, -0.3609771728515625, -1.47039794921875, 2.17181396484375, 0.821746826171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000009.npy"}
{"epoch": 0.018848167539267015, "step": 10, "batch_size": 128, "mean": -0.03389531373977661, "std": 0.8098540306091309, "min": -2.00482177734375, "p10": -0.9598983764648437, "median": -0.028850555419921875, "p90": 0.8025588989257812, "max": 3.255615234375, "pos_frac": 0.453125, "sample": [0.39764404296875, 0.3185272216796875, 0.6139068603515625, 0.23050689697265625, 0.5730209350585938, 0.143402099609375, 0.17575836181640625, -1.513397216796875, -0.13098907470703125, 0.22188568115234375, 0.2011566162109375, -0.4320335388183594, 0.6546783447265625, 1.69012451171875, 0.295684814453125, -0.03044891357421875, -0.168975830078125, -0.929534912109375, -0.1160736083984375, 1.281036376953125, -0.4581298828125, -0.6894912719726562, -0.4105682373046875, -0.7711181640625, 0.1767730712890625, 0.20850372314453125, 1.5855712890625, 0.05896759033203125, -1.93719482421875, 0.2526397705078125, -0.14142608642578125, 0.30511474609375, -0.12005996704101562, -0.8525543212890625, -0.09612655639648438, -0.0698699951171875, -0.737640380859375, 0.17218017578125, 0.448150634765625, 0.957489013671875, -1.1221923828125, -0.01413726806640625, -0.1368885040283203, -1.051605224609375, -0.331787109375, 0.4021453857421875, -0.9456939697265625, -0.109893798828125, -0.04248046875, -0.8032379150390625, 0.11201667785644531, -0.803955078125, 0.05120849609375, 0.333221435546875, -0.7073211669921875, -0.3714752197265625, -0.8264617919921875, 1.36407470703125, 0.18761825561523438, 3.255615234375, 1.1678314208984375, -0.39715576171875, -0.34417724609375, 0.5004119873046875, -0.31751251220703125, -0.857696533203125, 0.6342620849609375, -0.027252197265625, 0.50018310546875, -0.768768310546875, -0.9938735961914062, -0.374053955078125, 0.81585693359375, 0.3461761474609375, -1.227996826171875, 0.05809783935546875, -0.20914077758789062, 1.500244140625, -1.808258056640625, 1.83367919921875, -0.9193267822265625, 0.6756591796875, -0.067626953125, 0.424560546875, 0.2508983612060547, -0.250244140625, 0.00830078125, -0.07990264892578125, -1.99169921875, -0.4527587890625, -0.3291015625, -0.01700592041015625, 0.49051666259765625, -1.0986328125, -0.68023681640625, 0.17864990234375, -0.112396240234375, -0.01751708984375, -0.30179595947265625, 0.3860626220703125, -0.281707763671875, -0.05014801025390625, 0.245086669921875, 0.5210723876953125, -0.540283203125, -0.3058319091796875, -1.3795547485351562, -0.51812744140625, 0.8895645141601562, 0.539337158203125, 0.31927490234375, -0.08514976501464844, 0.7968597412109375, 0.0, -1.343841552734375, -2.00482177734375, -0.3696746826171875, -0.537078857421875, 0.018161773681640625, 0.92242431640625, 0.168701171875, 0.38994598388671875, 2.7369384765625, 0.0973052978515625, -0.04156494140625, -0.9930419921875, -0.02490234375, 0.5693359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000010.npy"}
{"epoch": 0.020942408376963352, "step": 11, "batch_size": 128, "mean": 0.06075235456228256, "std": 0.8645458221435547, "min": -2.724395751953125, "p10": -0.9065185546875, "median": 0.054889678955078125, "p90": 1.0064331054687499, "max": 2.7443084716796875, "pos_frac": 0.53125, "sample": [-1.568115234375, 0.20550537109375, 1.3458251953125, 0.101043701171875, 0.8264617919921875, 0.05120086669921875, 0.538055419921875, 1.4886474609375, -0.07818603515625, 0.487335205078125, -0.5395965576171875, 0.05096435546875, 1.0499267578125, -0.373687744140625, 0.92938232421875, -0.079681396484375, 0.36883544921875, 0.21956920623779297, -0.568115234375, 0.3595123291015625, 0.2296123504638672, -0.9504852294921875, -0.45037841796875, 1.7431640625, -0.56353759765625, 0.3593292236328125, 0.2802734375, 0.4761962890625, 0.9499664306640625, -0.756072998046875, -0.328338623046875, 2.1214599609375, -0.116943359375, 0.3502197265625, 0.43853759765625, 0.98779296875, -0.93719482421875, 0.408416748046875, 2.38677978515625, -0.3616294860839844, 0.0585784912109375, -1.176025390625, -1.479766845703125, -1.49237060546875, 0.256927490234375, 0.618621826171875, -0.0463104248046875, 2.045745849609375, 2.312744140625, 0.1810302734375, -0.0308685302734375, -0.48004150390625, -0.39012908935546875, -0.849151611328125, -1.12127685546875, 1.14990234375, -1.3728179931640625, 0.3492431640625, 0.38507080078125, 0.210723876953125, 0.65924072265625, -0.7806396484375, -0.6378326416015625, 0.119384765625, 0.20477294921875, -0.060764312744140625, 0.4781494140625, -0.18926620483398438, 0.371551513671875, 0.11803817749023438, -0.467529296875, 0.52276611328125, -0.374237060546875, -0.89337158203125, 0.14398765563964844, 0.4623260498046875, 0.0618743896484375, -0.14988327026367188, 0.908050537109375, 0.538818359375, 0.024219512939453125, -0.07269287109375, -0.19710159301757812, -0.26190185546875, -0.41094970703125, 0.3257713317871094, -0.3203582763671875, 0.382293701171875, -1.28472900390625, 1.4412841796875, 0.300872802734375, -0.3306121826171875, -2.724395751953125, 0.08330345153808594, -0.577911376953125, -0.02593994140625, -0.4246826171875, -0.8411865234375, -0.17557525634765625, -0.3736896514892578, -0.6965255737304688, -1.78076171875, 0.72686767578125, 0.105316162109375, 0.01373291015625, -0.057525634765625, 1.941009521484375, 0.0612945556640625, 1.6368865966796875, 0.39279937744140625, 0.640655517578125, -0.2108306884765625, -1.07952880859375, -0.7702178955078125, -0.146087646484375, 0.970733642578125, -0.663116455078125, 0.4429931640625, -0.401702880859375, 2.7443084716796875, 0.856353759765625, -0.328338623046875, 0.5535430908203125, -0.6437225341796875, -0.705291748046875, 0.208892822265625, -1.2592926025390625, -0.559478759765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000011.npy"}
{"epoch": 0.023036649214659685, "step": 12, "batch_size": 128, "mean": -0.07305683195590973, "std": 0.880263090133667, "min": -3.14813232421875, "p10": -1.1908523559570312, "median": -0.0742950439453125, "p90": 0.9454650878906249, "max": 2.23077392578125, "pos_frac": 0.4609375, "sample": [-1.34332275390625, 0.70843505859375, 0.37371826171875, 0.01434326171875, -0.47641563415527344, -0.430419921875, 0.341339111328125, -0.2537384033203125, -0.51025390625, -0.0340423583984375, -0.0984039306640625, 0.376953125, -0.895538330078125, 0.975799560546875, -0.691680908203125, 0.09256744384765625, 0.5965232849121094, 0.0213470458984375, 1.68902587890625, 0.81549072265625, 0.265472412109375, -0.86285400390625, 1.427490234375, -0.07916259765625, 0.278472900390625, 0.1765289306640625, -1.202463150024414, -0.21343994140625, -1.22259521484375, 0.17108154296875, -2.412384033203125, 0.15996551513671875, -0.98126220703125, -2.0355224609375, -1.140960693359375, 1.420440673828125, -0.069427490234375, -1.188201904296875, 0.9119873046875, -0.30205535888671875, 0.0130767822265625, 0.473388671875, -0.334228515625, 1.018096923828125, -0.2428131103515625, -0.23553466796875, -0.3570556640625, 0.932464599609375, 1.203216552734375, 1.257110595703125, 1.2604827880859375, 0.4823455810546875, 0.4518280029296875, -0.138397216796875, -0.713348388671875, -0.28778076171875, 2.2113037109375, -0.211517333984375, -0.562042236328125, -0.4256134033203125, 0.56561279296875, -0.950042724609375, -0.102935791015625, 0.142303466796875, -0.14078140258789062, 0.494293212890625, 0.2085418701171875, -0.30584716796875, 0.5083541870117188, -0.2497406005859375, 0.030467987060546875, -1.319061279296875, 0.6305084228515625, 0.17803955078125, 0.3235015869140625, -0.892791748046875, 0.7427520751953125, -0.0172576904296875, -1.708221435546875, -1.380157470703125, 1.399078369140625, -0.134002685546875, -1.3619384765625, -0.0503997802734375, 0.11487388610839844, -0.811492919921875, -0.034008026123046875, 0.5232391357421875, 0.85693359375, -0.370880126953125, -0.8255615234375, -0.13728713989257812, -0.08994293212890625, 0.561798095703125, -1.453094482421875, 2.23077392578125, -0.4781494140625, 0.463836669921875, -1.1970367431640625, 0.7010498046875, -0.166290283203125, 0.16414642333984375, -0.9007568359375, 0.428619384765625, -1.18511962890625, -0.84521484375, -0.104888916015625, -0.31353759765625, 0.4264678955078125, 0.54388427734375, 1.309112548828125, 0.3177490234375, 0.837127685546875, -0.431243896484375, 2.0621337890625, -0.481475830078125, -0.4167060852050781, -3.14813232421875, -1.43328857421875, -0.526153564453125, 0.735992431640625, 0.14007568359375, -0.26824951171875, 0.218292236328125, 0.096221923828125, -1.1354293823242188, -1.1347198486328125, -0.9730377197265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000012.npy"}
{"epoch": 0.025130890052356022, "step": 13, "batch_size": 128, "mean": 0.012652680277824402, "std": 0.8657077550888062, "min": -2.34478759765625, "p10": -0.984307861328125, "median": -0.023698806762695312, "p90": 1.0760040283203125, "max": 3.72467041015625, "pos_frac": 0.4765625, "sample": [-0.2916412353515625, -0.029394149780273438, 0.18941497802734375, -1.88720703125, 0.13726806640625, -0.23345947265625, 0.9644775390625, 0.4909820556640625, 1.1621551513671875, 0.00431060791015625, 0.9259796142578125, -0.300567626953125, -0.2568321228027344, -0.4136848449707031, 0.05121421813964844, 0.12628555297851562, -0.57501220703125, 1.34375, 0.416656494140625, 0.3534393310546875, -0.016025543212890625, 0.224365234375, 3.72467041015625, 0.578521728515625, 0.571258544921875, -0.120941162109375, -0.018003463745117188, 0.093963623046875, -0.6072540283203125, -0.6590576171875, -0.839263916015625, 0.38236236572265625, 0.4786376953125, 1.00006103515625, 0.38299560546875, -0.685546875, -0.1940460205078125, -1.722137451171875, -0.28624725341796875, 0.0333251953125, 0.08888053894042969, 0.3542633056640625, 0.39141845703125, 1.69873046875, -0.086944580078125, 0.0635986328125, -0.95135498046875, -0.9710693359375, 0.06207275390625, 0.3868408203125, 0.136322021484375, -1.32977294921875, 0.12185287475585938, 0.917144775390625, -0.380615234375, 0.0, -0.06640625, -0.576324462890625, -0.03661346435546875, -0.623260498046875, 0.079193115234375, 0.1648101806640625, -1.3503189086914062, -0.1031341552734375, 2.2362060546875, -1.3460693359375, -0.74029541015625, -0.868499755859375, -0.0746917724609375, 0.06546592712402344, 1.1451416015625, -0.950347900390625, 0.014621734619140625, 1.203094482421875, -1.01519775390625, 1.6522369384765625, -0.03288078308105469, 2.57415771484375, -0.15903854370117188, -0.1699981689453125, 0.2640037536621094, 0.8459014892578125, -0.165924072265625, -0.62786865234375, 0.0782470703125, 0.94598388671875, -0.08929824829101562, -0.30974578857421875, 0.11407470703125, -0.351287841796875, 1.099822998046875, -0.10608673095703125, 0.5883026123046875, 0.0383758544921875, 0.110626220703125, -0.25194549560546875, -0.26569366455078125, 0.208892822265625, -1.450286865234375, -1.16766357421875, -0.243011474609375, -0.5448684692382812, -0.29984283447265625, -0.49814605712890625, 0.2817535400390625, 1.41241455078125, 1.26263427734375, -0.8349609375, 0.3363037109375, -1.029754638671875, -0.03779029846191406, -2.34478759765625, -1.38958740234375, 0.0609130859375, 1.82122802734375, 1.0657958984375, 0.88897705078125, -0.644287109375, 0.949462890625, -0.64093017578125, -0.185302734375, -1.135009765625, -1.4745025634765625, -0.15608978271484375, -0.11065673828125, 0.15755844116210938, -0.0601043701171875, -0.519287109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000013.npy"}
{"epoch": 0.027225130890052355, "step": 14, "batch_size": 128, "mean": 0.045512035489082336, "std": 0.9795951247215271, "min": -3.01776123046875, "p10": -1.0903900146484375, "median": -0.05613899230957031, "p90": 1.4317733764648437, "max": 2.90826416015625, "pos_frac": 0.4609375, "sample": [0.7276763916015625, 0.707244873046875, -0.709228515625, 1.54632568359375, 0.224456787109375, -1.215423583984375, -0.970672607421875, -0.596923828125, -0.06606864929199219, 1.4226226806640625, -0.089935302734375, -0.538726806640625, -2.465087890625, 1.176055908203125, 1.226593017578125, -0.7033538818359375, -0.38665771484375, -0.059722900390625, -0.2569694519042969, -1.0740509033203125, -0.23397445678710938, 0.23168182373046875, -0.47247314453125, -1.085052490234375, -0.268768310546875, -0.3218994140625, -0.07464599609375, 0.655303955078125, -0.97015380859375, 1.606689453125, 0.6177215576171875, -0.63232421875, -1.9351806640625, 0.3654937744140625, 0.00772857666015625, -1.63238525390625, -2.081298828125, 0.5106792449951172, 1.50946044921875, -1.58367919921875, 0.1167144775390625, -0.163116455078125, -0.01172637939453125, 1.636749267578125, 0.0, -0.057956695556640625, -0.4506988525390625, 0.388336181640625, 0.327392578125, -0.5375289916992188, -0.292236328125, 0.3222503662109375, 0.3269615173339844, 1.64935302734375, -0.303924560546875, 0.167572021484375, 1.080169677734375, 0.840667724609375, -1.408111572265625, 2.454620361328125, 0.04052734375, -3.01776123046875, 1.640411376953125, -0.280487060546875, 1.8673248291015625, -0.2315521240234375, -0.0237884521484375, -0.34759521484375, -0.226593017578125, 0.424896240234375, 0.08258056640625, -1.7089996337890625, -0.3043060302734375, 1.208160400390625, 0.80267333984375, 0.48569488525390625, -0.304962158203125, -0.574554443359375, 1.19354248046875, -0.2671356201171875, -0.3574981689453125, -1.2467193603515625, 0.13826751708984375, 0.282928466796875, -0.1649322509765625, 0.34767913818359375, -0.1367359161376953, -0.42236328125, -1.0387802124023438, -0.05999946594238281, 0.3364715576171875, 0.35634613037109375, 2.1513671875, 0.5255889892578125, -1.84747314453125, 0.577911376953125, -0.34502410888671875, -0.5838470458984375, 1.453125, -0.3604888916015625, 0.26763916015625, -0.22713470458984375, 0.04228973388671875, 0.8089599609375, -0.434326171875, 0.5781097412109375, 0.2070770263671875, -0.326568603515625, 0.2623138427734375, 0.572479248046875, 1.0216217041015625, -0.296295166015625, -1.772003173828125, -0.7488937377929688, 0.07321739196777344, -0.0345458984375, 0.72882080078125, 2.057342529296875, 0.9330596923828125, -0.30567169189453125, -0.0543212890625, 1.3888397216796875, -0.18878173828125, -1.10284423828125, -0.12770843505859375, -0.2083282470703125, 2.90826416015625, 1.5424652099609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000014.npy"}
{"epoch": 0.02931937172774869, "step": 15, "batch_size": 128, "mean": 0.01381191611289978, "std": 0.9074937105178833, "min": -2.587677001953125, "p10": -1.05745849609375, "median": -0.0014095306396484375, "p90": 1.1346099853515625, "max": 3.61749267578125, "pos_frac": 0.5, "sample": [0.1607666015625, -0.7354583740234375, 0.911102294921875, -1.567138671875, -0.9794921875, -0.6252365112304688, 1.4380645751953125, -0.38232421875, -0.1553955078125, 0.420166015625, 0.309051513671875, 1.155517578125, 0.616424560546875, -0.4438018798828125, -0.8944244384765625, -2.587677001953125, 0.18410491943359375, -0.32003021240234375, -0.42962646484375, -0.2001190185546875, 2.005615234375, -0.11959075927734375, 1.243316650390625, -0.322845458984375, 0.0688323974609375, 0.3452911376953125, 0.52691650390625, -1.9443359375, 0.037322998046875, 0.1422119140625, 1.144287109375, -0.60675048828125, -1.023529052734375, 0.297210693359375, 0.2553863525390625, 0.14301681518554688, 2.24896240234375, -0.0098876953125, 0.780364990234375, 0.5059661865234375, -0.701171875, -1.366424560546875, 0.5458984375, 0.00260162353515625, -0.0398101806640625, -1.16473388671875, 0.14410781860351562, 1.130462646484375, -0.77734375, 0.566436767578125, -0.4297504425048828, 0.633941650390625, 0.41180419921875, 0.612213134765625, 0.0516357421875, -1.15643310546875, -1.17230224609375, -0.309234619140625, -0.10133171081542969, 0.026336669921875, 0.150848388671875, 0.38947296142578125, -0.5172119140625, -0.109161376953125, -0.6089935302734375, -0.0633392333984375, 0.5968475341796875, -0.0630035400390625, 2.54339599609375, -1.398712158203125, -0.3070182800292969, 0.2883491516113281, -0.5347442626953125, -0.293060302734375, -0.66534423828125, 0.201812744140625, 0.8756561279296875, -0.516632080078125, -0.1580963134765625, 0.3531494140625, 1.242706298828125, 0.06652450561523438, 1.34991455078125, -0.2593994140625, -1.05108642578125, -0.2215576171875, -2.162139892578125, 1.00091552734375, 0.7375907897949219, 1.891143798828125, -0.23711395263671875, -0.5035858154296875, 0.0261383056640625, -1.360260009765625, -0.613067626953125, 3.61749267578125, -0.005420684814453125, 1.214752197265625, 0.03387451171875, 0.1097412109375, -0.8105316162109375, 2.347381591796875, -0.120391845703125, 1.1287078857421875, 0.305908203125, -0.097137451171875, -0.7454833984375, -0.3026123046875, -1.07232666015625, -0.1921234130859375, 0.8275375366210938, 0.18717193603515625, -0.472320556640625, 0.11672210693359375, 0.0124969482421875, 0.09774017333984375, -0.43798828125, 0.5394668579101562, -0.6124496459960938, -0.019256591796875, -0.438079833984375, 0.13216400146484375, -0.2426910400390625, -1.635223388671875, 0.45184326171875, 0.35565185546875, -1.1089019775390625, 0.032135009765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000015.npy"}
{"epoch": 0.031413612565445025, "step": 16, "batch_size": 128, "mean": 0.023690536618232727, "std": 1.027940273284912, "min": -2.88360595703125, "p10": -1.1592666625976562, "median": -0.030826568603515625, "p90": 1.1128082275390623, "max": 3.407318115234375, "pos_frac": 0.46875, "sample": [0.8132247924804688, 0.16716766357421875, 0.5088424682617188, -0.356719970703125, 0.828948974609375, 0.3209991455078125, 0.2141876220703125, 1.607391357421875, 0.1654052734375, 0.670166015625, 0.5207023620605469, -0.274444580078125, 0.479949951171875, -1.298309326171875, 1.00396728515625, -0.0090789794921875, 1.331451416015625, 2.376068115234375, -0.2161865234375, -0.2244110107421875, 0.4574737548828125, -0.03870391845703125, -0.300140380859375, 0.74774169921875, -0.157806396484375, -0.3425025939941406, 1.15435791015625, 2.34954833984375, -0.28253173828125, -1.1888427734375, 2.759490966796875, 1.0496826171875, -0.93731689453125, 0.157684326171875, -0.223876953125, -1.484832763671875, 2.79107666015625, -0.8377609252929688, -0.5699806213378906, 0.04467010498046875, 1.17108154296875, 0.654327392578125, -0.133575439453125, 0.170654296875, -0.825103759765625, -0.22287940979003906, -0.1881103515625, 0.5795440673828125, 1.85418701171875, -0.10462570190429688, -0.13224411010742188, 0.7900390625, 0.14144134521484375, -2.88360595703125, -1.591552734375, -0.6851806640625, 0.4591827392578125, -0.35036468505859375, 0.431488037109375, -0.62548828125, 0.693603515625, -0.013393402099609375, 0.090667724609375, -0.3143310546875, 0.0187530517578125, -2.50152587890625, -0.18814849853515625, 0.895965576171875, 0.79119873046875, -1.820281982421875, -0.051776885986328125, -0.0423583984375, -1.6154708862304688, 0.12994384765625, -0.292755126953125, -0.4432220458984375, 0.30255126953125, -0.02294921875, -0.18843841552734375, -2.4254150390625, -0.48990631103515625, -0.08225250244140625, -1.1465911865234375, -0.48358154296875, -0.257568359375, 1.190826416015625, 0.0, 0.5439224243164062, 1.095001220703125, -0.44329833984375, -0.59039306640625, 0.002593994140625, -1.9006805419921875, -0.776123046875, 1.093841552734375, -1.0237579345703125, 0.4258689880371094, 0.8087158203125, -0.5818328857421875, 0.10465621948242188, -0.152313232421875, 2.977813720703125, -0.93194580078125, 0.30291748046875, -0.19171142578125, -0.4952239990234375, 1.493316650390625, -0.480987548828125, 0.42388916015625, 0.771148681640625, -0.184844970703125, -0.7845611572265625, -1.555206298828125, 0.0714111328125, -2.02130126953125, 0.314056396484375, 0.4546356201171875, 0.02919769287109375, -1.46612548828125, 0.646759033203125, 0.12242889404296875, -0.33837890625, 0.08514404296875, -0.82159423828125, 3.407318115234375, -0.485015869140625, -0.80645751953125, -0.1300048828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000016.npy"}
{"epoch": 0.033507853403141365, "step": 17, "batch_size": 128, "mean": 0.012122467160224915, "std": 0.8861120939254761, "min": -2.293243408203125, "p10": -1.1562026977539062, "median": 0.012148857116699219, "p90": 1.0269256591796871, "max": 2.818695068359375, "pos_frac": 0.515625, "sample": [-0.6899642944335938, -0.7390899658203125, 2.099853515625, -0.917999267578125, 0.3209228515625, 0.73016357421875, -0.21710205078125, -0.1845703125, 0.00555419921875, -0.1352996826171875, 0.8793106079101562, -1.37359619140625, -2.1900634765625, 1.296051025390625, -0.3927001953125, -0.9310684204101562, 1.9278106689453125, 1.57891845703125, -0.6868896484375, 0.763671875, 1.1676025390625, 0.69873046875, -1.6999969482421875, 0.45035552978515625, 0.008045196533203125, 0.32940673828125, -1.163055419921875, -0.5548095703125, -0.21856689453125, -0.862548828125, 0.3498992919921875, 0.4981689453125, -0.041332244873046875, 0.83624267578125, -1.679290771484375, -0.10143280029296875, 0.9007568359375, -2.14697265625, 0.059650421142578125, 0.111480712890625, -0.2784423828125, -1.34271240234375, -0.6090316772460938, -0.4673118591308594, -0.108184814453125, 0.507080078125, 0.9368820190429688, 0.016252517700195312, 0.6534881591796875, 1.206512451171875, -0.18817138671875, 0.1831512451171875, -1.417877197265625, 2.818695068359375, 0.5667495727539062, -0.34246826171875, 0.02896881103515625, 0.178009033203125, -0.3471221923828125, -0.80718994140625, 0.5125579833984375, -0.43799591064453125, -0.36901092529296875, -0.0251312255859375, -0.796478271484375, -0.3497161865234375, -1.42718505859375, -0.33356475830078125, -0.21509552001953125, -0.457305908203125, -0.773162841796875, -0.67047119140625, -1.16253662109375, -0.945404052734375, 2.0910797119140625, 1.83209228515625, 0.41424560546875, -0.37036895751953125, 1.000457763671875, 0.16481399536132812, 0.3687744140625, 0.111236572265625, -0.09576416015625, 0.513153076171875, 0.2869415283203125, 0.239166259765625, 1.08868408203125, 0.38751220703125, -0.998748779296875, -0.346405029296875, 0.07775497436523438, 0.8084564208984375, 0.317626953125, 0.775665283203125, 0.26983642578125, -0.3303070068359375, -0.7033843994140625, -0.517913818359375, 0.8454360961914062, 0.98486328125, -0.5199546813964844, 0.06036376953125, 0.15625, 0.8604736328125, 0.28511810302734375, 0.60345458984375, 0.21450424194335938, -1.1534881591796875, -0.3149566650390625, -1.23114013671875, 0.07568359375, -0.26358795166015625, -0.0238037109375, 0.34871673583984375, 1.225830078125, 0.92120361328125, -0.2641448974609375, 1.30657958984375, 0.1765594482421875, -0.61163330078125, 0.6155052185058594, 0.2644805908203125, -1.345611572265625, 0.07206344604492188, -2.293243408203125, -0.1077728271484375, -0.1561279296875, 1.6124267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000017.npy"}
{"epoch": 0.0356020942408377, "step": 18, "batch_size": 128, "mean": -0.12205088138580322, "std": 0.8220311403274536, "min": -2.634033203125, "p10": -1.0757110595703125, "median": -0.10676193237304688, "p90": 0.8644294738769531, "max": 1.551666259765625, "pos_frac": 0.4296875, "sample": [-0.5412445068359375, -0.611724853515625, -0.12503623962402344, 0.6088104248046875, 0.09561920166015625, -1.21575927734375, -0.20244979858398438, -0.15700531005859375, -0.7470474243164062, 0.3015327453613281, -1.81512451171875, 0.8999176025390625, -0.43749237060546875, -0.10467529296875, 0.241455078125, -0.24379730224609375, 0.7017822265625, -0.9771728515625, 0.121002197265625, 0.4397125244140625, -0.0164031982421875, 0.6907119750976562, 1.210479736328125, -0.49704742431640625, 1.551666259765625, -0.2748756408691406, -1.312103271484375, 0.0109710693359375, 0.0009918212890625, -0.11597824096679688, 0.23388671875, -0.4356689453125, 0.7040481567382812, 0.3087921142578125, -1.2108306884765625, -0.0630950927734375, 0.70904541015625, 0.47440528869628906, -0.16849136352539062, -0.4378662109375, 0.28725433349609375, 0.385711669921875, -2.634033203125, 0.45794677734375, 0.8601150512695312, 0.5004501342773438, 0.94635009765625, -0.15964508056640625, -0.0972747802734375, -0.4341278076171875, 0.1275634765625, -0.045867919921875, 0.5431060791015625, 0.35012054443359375, -1.2757568359375, -0.41143798828125, -0.6361236572265625, 0.7718505859375, 1.29498291015625, -0.859161376953125, 1.4027099609375, -0.2574005126953125, 0.54022216796875, -2.554779052734375, -2.3699951171875, -0.589263916015625, -1.053466796875, 0.768341064453125, -0.65155029296875, -0.63092041015625, -0.011112213134765625, 0.02002716064453125, -0.7795562744140625, -0.41668701171875, -0.1672515869140625, 0.09540557861328125, -0.31719970703125, -0.2350311279296875, 0.5378265380859375, 0.934661865234375, 0.742645263671875, 0.6495132446289062, -0.25970458984375, 0.8744964599609375, -0.265625, 0.7309951782226562, -0.5823745727539062, -2.047882080078125, -1.067626953125, -0.10884857177734375, 0.250823974609375, 0.5484733581542969, 0.114471435546875, -0.076446533203125, -2.364501953125, 0.17474365234375, -0.4849891662597656, -1.0484771728515625, -0.9020538330078125, 0.7625961303710938, -0.8278350830078125, 1.03704833984375, 1.035186767578125, -0.30145263671875, 0.3718986511230469, 0.1722412109375, -0.7373046875, -0.97576904296875, -0.5454559326171875, 1.522064208984375, -1.094573974609375, -0.02077484130859375, -0.17431640625, -0.930267333984375, 1.1216278076171875, -0.7763671875, -1.14697265625, 0.6624755859375, -0.17291259765625, -0.707122802734375, 1.11920166015625, 0.4881439208984375, -0.724365234375, -0.13165283203125, -0.03155517578125, 0.52862548828125, -0.4320526123046875, -1.42144775390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000018.npy"}
{"epoch": 0.03769633507853403, "step": 19, "batch_size": 128, "mean": -0.044742852449417114, "std": 0.9383370280265808, "min": -2.5264892578125, "p10": -1.13751220703125, "median": -0.047222137451171875, "p90": 1.0984886169433592, "max": 3.6041259765625, "pos_frac": 0.4609375, "sample": [-0.07211685180664062, -0.92596435546875, 0.1561279296875, 1.1422119140625, 0.08026123046875, -0.320831298828125, -0.83526611328125, 0.345703125, -0.5421142578125, -2.412109375, 0.1758880615234375, -0.038356781005859375, 1.6882476806640625, -2.24822998046875, 0.21877670288085938, -0.5215873718261719, 0.0146942138671875, -1.64910888671875, 0.7325363159179688, 0.8649749755859375, -0.34635162353515625, -0.04052734375, -0.13627052307128906, 0.36676025390625, -0.4000396728515625, -0.2711181640625, 0.34433746337890625, -0.7073898315429688, 0.94818115234375, 0.5150833129882812, -0.86358642578125, -0.942779541015625, 0.067291259765625, -0.39520263671875, 1.1829833984375, 0.11598587036132812, -0.141815185546875, 0.92852783203125, 0.6739501953125, 0.17879867553710938, 0.44872283935546875, -0.12311363220214844, -1.061737060546875, -1.8491973876953125, -0.17138671875, 1.6798095703125, -0.031890869140625, 0.369293212890625, 0.813812255859375, 0.0052337646484375, 0.04555511474609375, -1.386932373046875, -1.12542724609375, -1.3656768798828125, -0.644073486328125, 0.91583251953125, -0.221954345703125, -0.51959228515625, 0.2335052490234375, -0.65020751953125, -0.7151107788085938, -1.2069625854492188, 0.1943511962890625, -0.39189910888671875, 0.029693603515625, -0.239410400390625, 1.344268798828125, 0.00037384033203125, -0.8509597778320312, 0.7205657958984375, 0.387451171875, -1.071075439453125, -0.115264892578125, -1.467620849609375, 0.5152740478515625, -0.197784423828125, 1.002838134765625, 0.106353759765625, 0.1853179931640625, 0.6502685546875, -0.293365478515625, -0.7235107421875, 0.3564128875732422, -0.447601318359375, 0.3027496337890625, -0.8627471923828125, -0.05391693115234375, -0.69036865234375, -0.48679351806640625, -0.383056640625, 1.6590576171875, -1.45196533203125, 0.120269775390625, -0.25799560546875, -0.1514892578125, 1.486785888671875, 1.9451828002929688, -0.16824722290039062, -0.4336662292480469, 0.7104167938232422, -0.34607696533203125, -0.81915283203125, -0.2817840576171875, 0.8218994140625, -2.5264892578125, -0.61163330078125, 0.25701904296875, 0.29925537109375, -0.00579833984375, -0.290679931640625, 1.33306884765625, -0.5747756958007812, 1.231842041015625, -0.86309814453125, -1.392486572265625, 0.458038330078125, -2.127227783203125, -1.16571044921875, 3.6041259765625, 1.0797500610351562, 2.098419189453125, 0.79833984375, 1.94073486328125, 0.24578857421875, -0.09758758544921875, 0.292266845703125, -0.00403594970703125, -0.429046630859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000019.npy"}
{"epoch": 0.039790575916230364, "step": 20, "batch_size": 128, "mean": 0.09984631836414337, "std": 0.8715108036994934, "min": -2.188934326171875, "p10": -0.7735977172851559, "median": 0.04901885986328125, "p90": 0.9402893066406249, "max": 4.08306884765625, "pos_frac": 0.546875, "sample": [0.8285064697265625, 1.0081939697265625, 1.7130126953125, 0.86737060546875, 0.6938018798828125, 1.075164794921875, 0.0604400634765625, -1.789703369140625, -0.03668212890625, -0.26528167724609375, -0.40069580078125, 0.117095947265625, 0.8368072509765625, 0.585174560546875, 0.423675537109375, 2.70281982421875, -0.38787841796875, -0.3843994140625, 0.4353446960449219, -0.10357284545898438, 0.376800537109375, 0.567535400390625, 1.06317138671875, -0.31067657470703125, -0.2635173797607422, 0.93060302734375, 0.181121826171875, 0.2498779296875, 0.11735916137695312, -0.2225341796875, 0.73077392578125, -0.005222320556640625, -0.04441070556640625, -1.005218505859375, -0.2080535888671875, 0.128997802734375, -0.482818603515625, -1.030792236328125, -0.269805908203125, -0.53997802734375, -2.026824951171875, 0.34295654296875, -1.41790771484375, 0.3048095703125, -0.6743316650390625, 0.3214111328125, -0.2736053466796875, 0.21710205078125, -0.1286468505859375, -1.0349349975585938, 0.551025390625, -0.09452056884765625, 0.0, 0.402740478515625, 1.8731689453125, -0.24985885620117188, -0.29770851135253906, 2.41253662109375, 0.007293701171875, 0.4302215576171875, 0.07972335815429688, -1.0103759765625, -0.243255615234375, 2.43115234375, 0.19775390625, 0.20623779296875, -0.2815704345703125, 0.31182861328125, -0.5503692626953125, 0.49359130859375, -1.753662109375, 0.0437469482421875, 0.7125701904296875, -0.2406005859375, 0.866912841796875, 0.5762939453125, 0.1841888427734375, -0.205413818359375, 1.55352783203125, 0.5221328735351562, 0.062286376953125, 0.437835693359375, -0.4224739074707031, 0.2732734680175781, -0.335845947265625, -0.0426025390625, -2.188934326171875, -1.26708984375, 0.21213531494140625, -0.42041015625, -0.11547470092773438, 0.0594482421875, 0.9039306640625, -1.91583251953125, 0.054290771484375, -0.4586181640625, -0.27294921875, -0.3089561462402344, 0.0319976806640625, -0.31296539306640625, 4.08306884765625, 0.76092529296875, -0.19677734375, -0.48621368408203125, -0.305877685546875, -0.377044677734375, -1.490020751953125, 0.824310302734375, 0.00048828125, 0.38507080078125, 0.962890625, -0.3645782470703125, 0.8874359130859375, 0.547332763671875, -1.335968017578125, 0.41193389892578125, 1.008026123046875, 0.0410003662109375, -0.24407958984375, -0.014299392700195312, 0.9844970703125, 0.2708740234375, 0.09442138671875, -0.01080322265625, 0.11507034301757812, -0.41213226318359375, 0.139862060546875, 0.026123046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000020.npy"}
{"epoch": 0.041884816753926704, "step": 21, "batch_size": 128, "mean": 0.05018548667430878, "std": 0.9268454909324646, "min": -3.0438232421875, "p10": -1.040367889404297, "median": 0.0421905517578125, "p90": 1.029902648925781, "max": 2.529632568359375, "pos_frac": 0.515625, "sample": [0.76263427734375, -0.4434356689453125, -0.8321685791015625, -0.0251007080078125, -1.033233642578125, -1.496673583984375, 0.99072265625, -0.39129638671875, 0.028900146484375, -0.1427459716796875, 0.6847076416015625, -1.55731201171875, 0.86151123046875, -0.728546142578125, -0.147918701171875, -0.186981201171875, 0.7896728515625, 0.7646636962890625, -1.73309326171875, -0.04907989501953125, 2.529632568359375, 0.20965003967285156, -3.0438232421875, 0.470123291015625, 0.33258056640625, 1.41558837890625, -0.251373291015625, 1.0955810546875, 0.26477813720703125, -0.7572860717773438, 0.2260589599609375, -0.52825927734375, 1.41546630859375, -0.39835357666015625, 0.5325050354003906, -0.22603607177734375, 0.24195098876953125, 0.184478759765625, -2.302947998046875, -0.204193115234375, -0.5646820068359375, 0.093841552734375, -0.255096435546875, 0.05548095703125, -0.2424468994140625, -0.55706787109375, -0.03896903991699219, -0.89263916015625, 1.0017547607421875, -0.149627685546875, -1.1799468994140625, 0.3935089111328125, 0.0, -0.30835723876953125, -0.312286376953125, 0.253173828125, -0.2972412109375, -1.14178466796875, -0.7658939361572266, 0.456146240234375, -2.11163330078125, 2.0882568359375, 0.14385604858398438, 1.60699462890625, 0.3245849609375, 0.5778656005859375, 0.3756866455078125, -0.14483642578125, -0.867889404296875, 1.493621826171875, -0.46783447265625, 0.1138458251953125, -1.487762451171875, -2.16473388671875, 0.5889205932617188, -0.0814208984375, -1.26019287109375, 0.22222900390625, -0.3253173828125, 1.31317138671875, -0.1600341796875, 0.014766693115234375, -0.3465576171875, -0.20562744140625, 0.148193359375, 0.365509033203125, 0.7080516815185547, -0.1335601806640625, -0.1930694580078125, -1.0570144653320312, -1.40081787109375, 0.5264739990234375, 2.4176025390625, 0.09065437316894531, 0.238006591796875, -0.6625022888183594, 0.13210296630859375, -0.251708984375, 2.302398681640625, 0.765869140625, 0.4365234375, -0.6581268310546875, 1.7691650390625, 0.94061279296875, -0.9512939453125, 0.352996826171875, -0.35976409912109375, 0.132904052734375, 0.813262939453125, 0.9279937744140625, 0.49481201171875, -0.0750732421875, 0.94091796875, 0.761566162109375, 0.5561981201171875, -0.499176025390625, 0.36908721923828125, 2.4171142578125, -0.12615966796875, -0.82122802734375, 0.5003662109375, 0.19818115234375, 0.4605827331542969, 0.695709228515625, -0.072540283203125, 1.1009979248046875, -0.20880889892578125, 0.2215576171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000021.npy"}
{"epoch": 0.04397905759162304, "step": 22, "batch_size": 128, "mean": 0.09043721854686737, "std": 0.7085484266281128, "min": -1.718505859375, "p10": -0.6759475708007813, "median": 0.08780288696289062, "p90": 0.8736110687255859, "max": 2.3580551147460938, "pos_frac": 0.5546875, "sample": [0.3102874755859375, 0.4515190124511719, 0.3273773193359375, -0.33306884765625, 0.703857421875, 0.08197021484375, 0.39312744140625, -0.29486083984375, 0.379730224609375, 0.472503662109375, 0.692291259765625, -0.14347076416015625, -1.318603515625, -0.02146148681640625, -0.5789031982421875, 0.34613037109375, -0.49334716796875, 0.295166015625, 0.065826416015625, -0.4678192138671875, -0.79180908203125, 0.33751678466796875, 0.33941650390625, -0.44677734375, 1.288726806640625, -0.6684722900390625, 0.89129638671875, -0.88922119140625, -0.12659454345703125, 0.33905792236328125, 0.654632568359375, -0.3934478759765625, 0.0, -0.168670654296875, -0.6493301391601562, -1.718505859375, -0.464447021484375, -0.3359527587890625, 0.8660316467285156, -1.0013427734375, -0.076324462890625, 0.2171630859375, -0.277191162109375, 1.92236328125, 0.08075714111328125, 0.48419189453125, 0.41473388671875, 0.3474273681640625, -1.03143310546875, -0.22021484375, -0.38323974609375, 0.5037460327148438, -1.52777099609375, 0.865020751953125, 0.1904144287109375, 0.5813369750976562, 0.4535980224609375, -0.89849853515625, 1.038726806640625, 1.0924072265625, -0.675567626953125, -0.039154052734375, -0.005950927734375, -0.2610015869140625, 0.12465858459472656, 0.255096435546875, 1.03424072265625, -0.636505126953125, 0.09363555908203125, 0.34759521484375, 0.041851043701171875, -1.09735107421875, 0.329742431640625, 0.7703857421875, 0.070648193359375, 0.3003692626953125, 0.277496337890625, -0.47660064697265625, 1.81170654296875, 0.2650146484375, -0.657257080078125, -0.1702423095703125, -0.233184814453125, -0.2623100280761719, 0.1157684326171875, 0.69384765625, 2.3580551147460938, -0.3382568359375, -0.343505859375, -0.3331298828125, 1.8673248291015625, -0.576263427734375, 0.04071807861328125, 1.0884017944335938, 0.252532958984375, -0.18438720703125, -1.47442626953125, -1.397613525390625, 1.6739501953125, 0.66925048828125, -0.518798828125, -0.11262130737304688, 0.72698974609375, 0.91107177734375, 0.38238525390625, 0.2259063720703125, 0.284881591796875, 0.74713134765625, 0.2362213134765625, 0.7912445068359375, 0.19367599487304688, -0.2835693359375, -0.20748138427734375, -0.1872406005859375, 0.05257987976074219, 0.4228515625, 0.355682373046875, 0.1502685546875, 0.419586181640625, 1.820068359375, -0.1040496826171875, -1.2281494140625, -0.2796516418457031, -0.6768341064453125, 0.3341865539550781, 0.370269775390625, -0.23507118225097656, -0.042724609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000022.npy"}
{"epoch": 0.04607329842931937, "step": 23, "batch_size": 128, "mean": 0.15511269867420197, "std": 0.8468354344367981, "min": -2.034881591796875, "p10": -1.0207603454589842, "median": 0.15500831604003906, "p90": 1.2057113647460938, "max": 2.61474609375, "pos_frac": 0.6015625, "sample": [-0.3208141326904297, 0.2641448974609375, -0.1891326904296875, 0.0834503173828125, 1.050872802734375, 2.27392578125, -0.45770263671875, 1.2694091796875, -1.1236724853515625, -0.87762451171875, 2.61474609375, 1.5096435546875, -0.01816558837890625, 0.703887939453125, -1.151580810546875, -0.11987686157226562, 1.74896240234375, 0.389007568359375, 0.5464820861816406, -0.8252105712890625, 0.35749053955078125, 0.3756103515625, 0.2823944091796875, -1.121734619140625, 0.37348175048828125, 0.450836181640625, 0.2872467041015625, 0.996337890625, -0.2418670654296875, -0.0574493408203125, 0.0, 0.32027435302734375, -0.1970367431640625, 0.3345184326171875, -2.034881591796875, 0.409454345703125, -0.14987564086914062, -0.167205810546875, 0.00567626953125, -0.78717041015625, -0.204132080078125, 0.23468017578125, 0.198516845703125, -0.009174346923828125, 0.92254638671875, 0.79498291015625, 0.1989898681640625, -0.8117446899414062, 1.02783203125, -0.21337890625, 0.7049636840820312, -0.06310272216796875, -1.3568115234375, 0.8829193115234375, 0.96942138671875, 0.5269775390625, 0.1090087890625, 0.06693267822265625, 0.865081787109375, 0.0938568115234375, -0.64788818359375, 0.427520751953125, 1.1654052734375, -0.06034088134765625, 1.08367919921875, -0.341522216796875, -2.01409912109375, 1.2596435546875, 0.284088134765625, 0.066131591796875, 2.0513458251953125, 0.27616119384765625, -0.8459014892578125, -0.390625, 0.49896240234375, -1.267547607421875, 1.9185791015625, 0.0839996337890625, 1.2066192626953125, 0.6899261474609375, -1.0019302368164062, 0.12971878051757812, 0.6030120849609375, 1.205322265625, 1.23687744140625, 0.1802978515625, -0.075714111328125, 0.82269287109375, -0.01324462890625, 0.0977783203125, -0.016632080078125, 0.2022552490234375, 0.210906982421875, 0.082122802734375, -0.7834320068359375, -1.064697265625, 0.727142333984375, 0.09967041015625, 0.1987762451171875, 0.989501953125, 1.3992919921875, 0.6403656005859375, 0.271331787109375, -0.2493438720703125, -0.93505859375, 1.803466796875, 0.31951904296875, 0.31972503662109375, -0.289764404296875, 1.217742919921875, 0.1132049560546875, -0.00836181640625, 0.19637680053710938, -1.11767578125, -0.14161300659179688, -1.0650634765625, 0.433563232421875, -0.400390625, 0.8694915771484375, -1.7735595703125, 0.606292724609375, -1.2414321899414062, -0.111846923828125, 0.0873870849609375, 1.0531158447265625, -1.82489013671875, -0.8499298095703125, -0.4873008728027344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000023.npy"}
{"epoch": 0.048167539267015703, "step": 24, "batch_size": 128, "mean": -0.10664233565330505, "std": 0.9941709041595459, "min": -2.92657470703125, "p10": -1.5597961425781248, "median": 0.010585784912109375, "p90": 0.9967063903808594, "max": 2.292755126953125, "pos_frac": 0.5078125, "sample": [0.8321304321289062, -0.9802703857421875, 0.37570953369140625, 1.8712310791015625, 0.248138427734375, -0.82623291015625, 1.2034912109375, 0.62274169921875, -0.14237594604492188, 2.292755126953125, -0.0111083984375, 0.30663299560546875, 0.31884765625, -0.26300048828125, 0.15929794311523438, -2.92657470703125, -0.1883544921875, -0.453216552734375, -0.1121978759765625, 0.41058349609375, -0.90899658203125, 2.07867431640625, -0.1492919921875, 0.0, 0.21685028076171875, 0.2948799133300781, -2.2110595703125, -0.886688232421875, 0.0999603271484375, 0.25946044921875, 0.7493896484375, 0.016693115234375, -0.594818115234375, 1.16278076171875, -0.191192626953125, -0.4661407470703125, -2.5699462890625, -1.0604705810546875, -0.45888519287109375, 0.90673828125, -1.0016632080078125, -0.352935791015625, -0.24371337890625, -0.51397705078125, 1.852264404296875, 0.3251953125, 0.0, -0.094970703125, -0.223663330078125, 0.5939788818359375, -1.099334716796875, 0.7278900146484375, 0.3602752685546875, 0.5555572509765625, 0.27838134765625, -0.75006103515625, -1.7001953125, 0.3642539978027344, 0.24139404296875, 2.0325927734375, -1.6627197265625, 0.0745849609375, -0.497314453125, 0.18073654174804688, -2.02349853515625, 0.9939346313476562, -0.31598663330078125, -0.1711883544921875, 0.0701904296875, 0.23382568359375, -0.7630615234375, -0.114898681640625, -1.3449325561523438, -1.51568603515625, -0.5332489013671875, 0.3212738037109375, 0.04315185546875, -1.4800567626953125, 1.003173828125, 0.6404190063476562, -0.126007080078125, -0.430908203125, -1.07489013671875, -0.7053375244140625, 0.0494384765625, -0.9271240234375, 0.35760498046875, 0.3795166015625, -2.293365478515625, 0.48590850830078125, -0.1261444091796875, -1.856597900390625, -1.89544677734375, -0.4622650146484375, 0.7300262451171875, 0.00447845458984375, 1.2388916015625, -0.5970916748046875, 0.112213134765625, 2.10009765625, 1.223388671875, 0.4945068359375, 0.1664886474609375, 1.661224365234375, 0.022918701171875, -1.77105712890625, 0.519195556640625, -0.704925537109375, 0.221649169921875, 0.631927490234375, 0.303466796875, -0.0560302734375, -0.1013031005859375, 0.757965087890625, 0.06819915771484375, -0.32073974609375, -0.429962158203125, 0.27777099609375, -2.1472015380859375, -0.66650390625, 1.752685546875, 0.3350830078125, 0.3658447265625, -2.413360595703125, -0.9151611328125, 0.020416259765625, -1.75360107421875, 0.32776641845703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000024.npy"}
{"epoch": 0.050261780104712044, "step": 25, "batch_size": 128, "mean": 0.07293592393398285, "std": 0.8414403200149536, "min": -2.1573486328125, "p10": -0.9318229675292968, "median": 0.12259292602539062, "p90": 0.9921234130859374, "max": 3.7274169921875, "pos_frac": 0.5703125, "sample": [-0.2863311767578125, 0.584625244140625, -1.57403564453125, 1.00457763671875, -0.69329833984375, 0.84271240234375, 0.148773193359375, 0.7799072265625, 3.7274169921875, -1.5931396484375, 1.0970306396484375, 0.2345123291015625, 0.0936431884765625, -0.115020751953125, 0.150970458984375, 0.02508544921875, -0.9861526489257812, -0.10980224609375, -0.232666015625, -0.109130859375, 0.4102325439453125, -0.3217182159423828, 0.1821136474609375, 0.4078521728515625, -0.0473785400390625, 0.1454315185546875, 0.8457183837890625, -0.5247650146484375, 0.20362091064453125, 0.267364501953125, 0.056396484375, -0.21143341064453125, 0.0, -0.8217620849609375, 1.2354736328125, 0.12022972106933594, 0.12050628662109375, -0.0778045654296875, -1.93963623046875, -0.707611083984375, -0.582000732421875, 0.0, 0.8135986328125, -1.3404541015625, 0.8004608154296875, -0.25494384765625, 0.243072509765625, -1.07562255859375, 0.320343017578125, -0.908538818359375, 0.14594268798828125, 0.64849853515625, 1.26177978515625, -0.256317138671875, 0.39044189453125, -0.0747528076171875, 0.3061981201171875, 0.1246795654296875, 1.597930908203125, -0.0318603515625, 0.4615325927734375, 1.2320556640625, 0.727508544921875, -1.2874755859375, -0.7997283935546875, 0.4483814239501953, 0.10638427734375, -0.5877685546875, -1.731170654296875, 0.2833404541015625, 0.986785888671875, -0.02899169921875, -0.38800048828125, 1.470550537109375, -0.860321044921875, 1.53253173828125, -0.8080062866210938, 0.2605743408203125, 1.5684661865234375, 0.399566650390625, -0.59588623046875, -0.02896881103515625, 0.9746551513671875, 0.195098876953125, -0.2897491455078125, 0.408966064453125, -0.4926624298095703, 0.3896446228027344, 0.5228004455566406, 1.713623046875, 1.849853515625, -0.056976318359375, 0.445831298828125, 0.82281494140625, 0.64141845703125, -0.35650634765625, 0.20425033569335938, 0.858856201171875, 1.797576904296875, 0.47576904296875, -0.28436279296875, 0.3002166748046875, 0.631683349609375, 0.025972366333007812, -0.8612060546875, -1.29461669921875, 0.277587890625, 0.381866455078125, -0.828094482421875, -2.1573486328125, -1.228759765625, 0.2028045654296875, -0.3433837890625, 0.5276718139648438, 0.2626953125, 0.139862060546875, -0.526641845703125, 0.4224281311035156, 0.2032012939453125, 0.6401443481445312, -0.3296165466308594, 0.08290672302246094, -1.410858154296875, -0.77166748046875, -1.25970458984375, -0.175506591796875, 0.052825927734375, -0.27189064025878906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000025.npy"}
{"epoch": 0.05235602094240838, "step": 26, "batch_size": 128, "mean": 0.09368638694286346, "std": 0.8171731233596802, "min": -2.79559326171875, "p10": -0.8500503540039062, "median": 0.06462478637695312, "p90": 0.9043426513671875, "max": 4.23291015625, "pos_frac": 0.5390625, "sample": [1.637451171875, 0.34552574157714844, 4.23291015625, -0.2837982177734375, 0.11175537109375, 0.019603729248046875, 0.43599891662597656, 0.6524620056152344, 0.53369140625, 0.0026874542236328125, -0.017559051513671875, 0.0535888671875, 0.0, 0.0925750732421875, 0.902740478515625, 0.567596435546875, 0.290374755859375, -0.3779144287109375, 1.740966796875, 1.5522918701171875, 1.533294677734375, -0.544403076171875, -0.012187957763671875, 0.1475830078125, -0.231597900390625, 0.4116630554199219, -0.1501922607421875, 0.3369598388671875, -0.69580078125, -0.12139892578125, -0.0864105224609375, -0.79217529296875, -0.28197479248046875, 0.272491455078125, 1.0859375, -1.639892578125, 0.8023223876953125, -1.197906494140625, -0.791961669921875, 1.702728271484375, -0.196746826171875, -0.4178333282470703, -0.600311279296875, -0.836212158203125, 0.0767822265625, -1.36712646484375, -0.08935546875, -0.92437744140625, -0.3388824462890625, 0.2730827331542969, -0.983184814453125, 1.89654541015625, 0.210906982421875, -1.26641845703125, 0.718780517578125, -0.23126983642578125, -0.3141021728515625, -0.394927978515625, 1.0293426513671875, 0.3668212890625, 0.587066650390625, -0.712646484375, 0.738739013671875, 0.73260498046875, -0.5600128173828125, -0.0164794921875, -0.98834228515625, 0.98065185546875, -0.7305908203125, 0.6247329711914062, -0.547882080078125, -0.1858367919921875, -0.926971435546875, 0.54766845703125, 0.5253677368164062, 0.13567733764648438, -0.2550506591796875, -0.2882518768310547, 0.9080810546875, 0.14426040649414062, 1.2708740234375, 0.86083984375, 0.07566070556640625, 0.525421142578125, 0.4881591796875, 0.30815887451171875, -0.3466033935546875, -0.0126953125, -0.24451446533203125, -0.6273880004882812, 0.3650665283203125, 0.3723487854003906, 0.16455459594726562, 0.470062255859375, 0.882843017578125, -1.1599273681640625, 0.0357513427734375, 0.07988548278808594, -0.1490039825439453, 0.59112548828125, -0.06932830810546875, -0.11452865600585938, 0.251983642578125, -0.8823394775390625, 0.5537567138671875, 0.44036865234375, -0.4842681884765625, 0.0, -2.79559326171875, -0.19251632690429688, 0.814453125, 0.04128265380859375, 0.20831298828125, -0.7559967041015625, -0.435821533203125, 0.5662994384765625, 0.789093017578125, 1.01556396484375, -1.002838134765625, 0.8816070556640625, 0.579345703125, 0.42132568359375, -0.22660064697265625, 0.479888916015625, -0.7659149169921875, 0.4929656982421875, -0.3519287109375, -0.981658935546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000026.npy"}
{"epoch": 0.05445026178010471, "step": 27, "batch_size": 128, "mean": 0.14168088138103485, "std": 0.9567137360572815, "min": -3.8099365234375, "p10": -0.8894210815429687, "median": 0.14980316162109375, "p90": 1.1972961425781243, "max": 3.28765869140625, "pos_frac": 0.5859375, "sample": [0.46514892578125, 0.591064453125, -0.6948127746582031, -0.19765472412109375, 0.460906982421875, 0.10113525390625, -0.03662872314453125, 0.1621246337890625, 1.5342864990234375, 0.6027679443359375, 0.3686981201171875, -1.254852294921875, 0.460052490234375, -0.108428955078125, -2.102294921875, 0.54815673828125, -0.0088348388671875, -0.02288055419921875, 0.0489654541015625, 1.08380126953125, -0.00824737548828125, 0.18104171752929688, 1.957275390625, 0.6209869384765625, 0.936279296875, 1.87603759765625, 0.2997283935546875, 0.87664794921875, -0.26373291015625, 0.355682373046875, -0.84722900390625, -0.22577667236328125, 0.7200927734375, -0.14166259765625, 0.4326171875, -0.992431640625, -0.8882904052734375, 0.612884521484375, -0.892059326171875, 0.0349578857421875, 2.52960205078125, 1.075103759765625, -2.49932861328125, 0.34256744384765625, 1.049530029296875, 0.137481689453125, -0.19937515258789062, -0.52056884765625, 0.62615966796875, 0.6832962036132812, -0.7562255859375, 0.3516845703125, 0.40950775146484375, -0.31490325927734375, 1.742034912109375, 1.58013916015625, 0.09954071044921875, -0.313201904296875, 0.266357421875, 1.4276123046875, -0.7310791015625, -0.6952056884765625, -1.397796630859375, -0.16717529296875, -0.74359130859375, 0.24481201171875, 1.12799072265625, -0.04986572265625, 3.28765869140625, -0.43438720703125, -0.51141357421875, -1.021453857421875, -1.743316650390625, -0.26153564453125, 1.46575927734375, 0.7553329467773438, -2.078643798828125, 1.1209716796875, 0.25335693359375, 0.886322021484375, -0.04201507568359375, 0.4072265625, 0.659515380859375, -0.824493408203125, -1.2351837158203125, 0.037689208984375, 0.553192138671875, -0.24927330017089844, 0.4703102111816406, 1.10906982421875, -0.5287017822265625, -1.1462554931640625, -3.8099365234375, 0.073272705078125, 0.256988525390625, -0.62451171875, 0.952850341796875, -0.6675434112548828, 0.4552001953125, 2.15985107421875, -0.3475799560546875, 0.4852123260498047, 0.04218292236328125, 1.1019287109375, 0.5302886962890625, 1.37127685546875, 0.314300537109375, 0.2744255065917969, 0.03179168701171875, 0.2726783752441406, 0.554840087890625, 0.7381801605224609, -0.10788536071777344, 1.3590087890625, 0.0999908447265625, -0.3056488037109375, 0.3290557861328125, -0.6694831848144531, -0.449371337890625, -0.582427978515625, 0.981048583984375, 1.3647613525390625, -1.056243896484375, -0.030303955078125, 0.229400634765625, 0.089202880859375, -0.10198974609375, -0.060016632080078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000027.npy"}
{"epoch": 0.05654450261780105, "step": 28, "batch_size": 128, "mean": 0.059473857283592224, "std": 0.8229645490646362, "min": -2.7592315673828125, "p10": -0.8416748046875, "median": 0.01318359375, "p90": 1.0595184326171874, "max": 2.546142578125, "pos_frac": 0.5078125, "sample": [-0.370361328125, 2.546142578125, -0.080841064453125, 0.2798919677734375, -0.06040191650390625, -0.5594882965087891, -0.27112579345703125, -0.11909103393554688, 0.3070487976074219, 0.5814208984375, 0.18886184692382812, -0.89825439453125, 1.3597412109375, 0.471923828125, 1.047607421875, 0.4430694580078125, 0.3902587890625, -0.11968231201171875, 1.320281982421875, 0.6609649658203125, 0.2141265869140625, -0.03953361511230469, -0.11382293701171875, 0.690338134765625, 0.0299072265625, -0.16054534912109375, -0.36859130859375, 0.0511016845703125, -0.163116455078125, -2.7592315673828125, 0.510986328125, -0.87554931640625, -2.330078125, 0.214508056640625, 0.11399459838867188, -0.83673095703125, -0.2434539794921875, -0.094482421875, 0.84027099609375, 0.0982513427734375, -1.306396484375, -1.42578125, -0.3321990966796875, -0.91796875, -0.0624237060546875, 1.98455810546875, 0.958099365234375, 1.53729248046875, 0.2989501953125, -0.052276611328125, -0.5533447265625, 0.09499359130859375, -0.451812744140625, -1.8646240234375, 2.09954833984375, -0.287445068359375, 0.2345428466796875, 0.00067138671875, 1.5899658203125, 1.478271484375, -0.938201904296875, 0.03771209716796875, 0.1069793701171875, -0.4039497375488281, 1.644744873046875, 0.1087646484375, -0.00888824462890625, 0.5348663330078125, -0.167572021484375, 1.1510009765625, -0.756317138671875, -1.28265380859375, -0.48264312744140625, -0.4421539306640625, -0.7130889892578125, 0.53948974609375, -1.32080078125, -0.7737464904785156, -0.00942230224609375, 0.3209228515625, 0.0906524658203125, -0.206390380859375, 1.2158203125, -0.183441162109375, 0.5213775634765625, 0.24808502197265625, 0.2279815673828125, -0.3842926025390625, -0.15689849853515625, -0.34819793701171875, -0.821807861328125, 1.0246429443359375, 0.02569580078125, -0.85321044921875, 0.04474449157714844, -0.2607078552246094, 0.4075927734375, 0.14020538330078125, 0.402252197265625, 0.92462158203125, 0.648681640625, 0.15635299682617188, -0.5962982177734375, 0.96905517578125, -0.154632568359375, -0.50726318359375, 1.087310791015625, -0.7399749755859375, 0.352386474609375, 0.50555419921875, 0.61993408203125, -0.535369873046875, 0.23956298828125, -0.07474327087402344, -0.755706787109375, 0.5451889038085938, 0.8865203857421875, 0.7143936157226562, -0.388519287109375, 0.7147445678710938, -0.0736236572265625, -0.122802734375, -0.35797119140625, 0.28607940673828125, 1.939208984375, -0.3245258331298828, 0.0, -1.5435943603515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000028.npy"}
{"epoch": 0.05863874345549738, "step": 29, "batch_size": 128, "mean": 0.09245166182518005, "std": 0.9887884855270386, "min": -3.09637451171875, "p10": -1.0198898315429688, "median": 0.056740760803222656, "p90": 1.250239562988281, "max": 3.284912109375, "pos_frac": 0.5546875, "sample": [0.05369758605957031, -1.04547119140625, 0.68890380859375, -0.62689208984375, -0.4216766357421875, -3.09637451171875, -0.0164794921875, 3.0994873046875, -0.896270751953125, 0.990264892578125, -0.656951904296875, 1.652923583984375, 0.9617919921875, 0.2110748291015625, 0.4016914367675781, 0.0, -2.01544189453125, 0.20458984375, 0.4256591796875, -0.14929962158203125, 0.99542236328125, 1.17608642578125, 0.01416015625, -2.61517333984375, 1.179046630859375, 1.628173828125, 3.284912109375, 0.5660552978515625, -0.07091522216796875, 0.5464401245117188, -0.2513427734375, -1.79559326171875, -1.0089263916015625, 0.100616455078125, -1.551544189453125, 0.8603515625, 0.489654541015625, 1.720947265625, 1.2206268310546875, -0.37439727783203125, -0.24749755859375, 0.018939971923828125, 0.320770263671875, 0.862945556640625, -0.234405517578125, 0.2764091491699219, -0.4826507568359375, 0.059783935546875, -0.7148895263671875, -0.0998382568359375, -0.24713134765625, 0.03009033203125, 0.03070068359375, -0.95281982421875, -0.0533294677734375, 0.442779541015625, 0.70953369140625, -0.05413818359375, -0.03094482421875, 0.5825042724609375, 0.956146240234375, -0.71295166015625, -1.3746337890625, 1.388671875, 0.0704345703125, 0.23717498779296875, 1.0643310546875, 0.629119873046875, 0.0, 0.01154327392578125, 1.3193359375, -1.274444580078125, 0.30230712890625, -0.8847808837890625, -0.10346221923828125, -0.17537689208984375, -1.720733642578125, 0.21129798889160156, 0.697906494140625, -0.258270263671875, -0.898834228515625, -0.6261749267578125, -2.18572998046875, 0.16095733642578125, 0.546417236328125, -0.051513671875, 0.46381378173828125, 1.033905029296875, -0.24688720703125, 1.721099853515625, 0.4279327392578125, 1.0467529296875, 0.9640121459960938, -1.3885498046875, 0.473175048828125, -0.1868743896484375, -0.4025726318359375, 0.2548065185546875, -0.5374832153320312, 1.523895263671875, 0.033966064453125, 0.10011672973632812, 1.45880126953125, -0.8179550170898438, 0.8016815185546875, 1.7653884887695312, -0.72869873046875, -0.083984375, 0.458251953125, -1.895904541015625, 0.391845703125, -0.02386474609375, -0.133544921875, -0.4290618896484375, 1.46343994140625, -1.67510986328125, 0.6031723022460938, -0.955535888671875, 1.77606201171875, -0.081207275390625, 0.094329833984375, 0.09243011474609375, 0.1820068359375, -0.22585296630859375, 0.491546630859375, 0.361968994140625, 0.3067626953125, -0.06964111328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000029.npy"}
{"epoch": 0.060732984293193716, "step": 30, "batch_size": 128, "mean": 0.06383557617664337, "std": 0.7669160962104797, "min": -1.793609619140625, "p10": -0.9168067932128906, "median": 0.030782699584960938, "p90": 0.8896682739257812, "max": 3.16058349609375, "pos_frac": 0.5234375, "sample": [-0.57757568359375, 1.4251251220703125, -1.147613525390625, -0.577362060546875, -0.9510345458984375, 0.8800048828125, 0.202667236328125, 0.1352081298828125, -0.08648681640625, 0.08216094970703125, 0.41391754150390625, -0.441619873046875, -0.97076416015625, -0.1973876953125, -0.0333099365234375, -1.240936279296875, -0.5361328125, 0.2867431640625, -0.35164642333984375, -0.1635894775390625, -0.044586181640625, 0.35870361328125, -0.4286041259765625, 0.010467529296875, -0.3598480224609375, -0.90655517578125, -0.7083740234375, 0.3722724914550781, 0.05275726318359375, 1.260833740234375, 0.872589111328125, -0.6621170043945312, 0.506317138671875, -0.40655517578125, -0.3941917419433594, -0.28643798828125, 0.934722900390625, 0.0509185791015625, 0.40262603759765625, -1.12127685546875, -0.6971435546875, -0.5612564086914062, 0.111846923828125, 1.0328369140625, -0.02166748046875, 1.25396728515625, -0.34564208984375, -1.1934051513671875, 0.678009033203125, -0.7914352416992188, 0.7153396606445312, 0.68194580078125, -0.205474853515625, 0.1775665283203125, 1.9793701171875, 0.12717437744140625, -0.6400299072265625, 0.38332366943359375, 0.416015625, 0.667999267578125, -0.2691650390625, 1.664825439453125, 0.4222412109375, -0.9407272338867188, 0.7909774780273438, -0.731475830078125, 0.1597900390625, -0.02734375, -0.15240097045898438, 0.784576416015625, -0.04328155517578125, -1.793609619140625, -0.32868194580078125, -0.11417770385742188, -0.33233642578125, 1.30487060546875, -0.36922454833984375, -1.107818603515625, -0.393798828125, -0.6235198974609375, 0.3493194580078125, 1.656280517578125, -0.406707763671875, 0.9122161865234375, 0.633087158203125, 0.27703857421875, 3.16058349609375, -0.5425262451171875, 1.7205657958984375, -1.2397003173828125, 0.4732666015625, 1.5635986328125, 0.5911712646484375, -0.52301025390625, 0.3880615234375, -1.31475830078125, 0.2124481201171875, 0.010646820068359375, -1.1539306640625, 0.453857421875, 0.24365234375, 0.364227294921875, 0.5201797485351562, 0.6359100341796875, 0.376220703125, 0.0796356201171875, -0.656005859375, 0.246185302734375, 0.7831878662109375, 0.70831298828125, 0.09417724609375, 0.4144134521484375, 0.0084228515625, -0.35284423828125, -0.21144485473632812, -0.04160308837890625, 0.41838836669921875, 0.6817169189453125, -0.4754791259765625, 0.740631103515625, 0.5340576171875, 0.5888748168945312, -0.5089111328125, -1.0909423828125, -0.35829734802246094, -0.17296600341796875, -0.8191986083984375, 0.8758544921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000030.npy"}
{"epoch": 0.06282722513089005, "step": 31, "batch_size": 128, "mean": 0.1599239706993103, "std": 0.8313660025596619, "min": -1.9276123046875, "p10": -0.7762870788574218, "median": 0.091827392578125, "p90": 1.0728546142578124, "max": 2.788818359375, "pos_frac": 0.59375, "sample": [0.311187744140625, -0.52886962890625, -0.7470703125, 0.98809814453125, -0.32021331787109375, 0.6405792236328125, -0.5911865234375, 1.08526611328125, -0.08270263671875, 1.609161376953125, 1.19622802734375, -0.8444595336914062, 0.339569091796875, -0.6575927734375, -0.01458740234375, -0.20245361328125, 1.3779563903808594, 0.068328857421875, 0.20891571044921875, 0.1808319091796875, 1.0126953125, 0.8450927734375, 0.184112548828125, 0.2432708740234375, 1.0382537841796875, 2.13140869140625, 0.090667724609375, 0.092987060546875, -1.400543212890625, 0.44976806640625, 0.5545501708984375, 1.9794464111328125, -0.17861175537109375, 0.018848419189453125, 1.0540771484375, -1.54888916015625, -0.0045318603515625, 0.145477294921875, -1.22369384765625, 0.0, 0.8134307861328125, 0.0150146484375, 0.0, 0.3421630859375, 0.491729736328125, -0.077972412109375, 0.19964599609375, 0.6841697692871094, -0.3267669677734375, -0.42449951171875, 0.872039794921875, -0.1956634521484375, 1.067535400390625, 1.1912078857421875, 0.004909515380859375, -0.0932159423828125, 1.805145263671875, 2.788818359375, 0.3879852294921875, -0.6394805908203125, 0.0, -0.05976104736328125, -1.722991943359375, -1.480010986328125, -0.3775634765625, 0.2919921875, -0.61590576171875, -0.34466552734375, 0.0111083984375, -0.032520294189453125, -1.692230224609375, -0.992706298828125, 0.11288833618164062, 0.596099853515625, -1.81982421875, 0.455230712890625, -0.9719696044921875, -0.0345001220703125, 0.983428955078125, -0.133453369140625, 0.2688255310058594, -0.15717315673828125, -0.579498291015625, 0.013998031616210938, 0.40288543701171875, 0.258331298828125, 1.27288818359375, 0.591888427734375, 0.02307891845703125, 0.644073486328125, -0.8522567749023438, -0.06048011779785156, -0.497894287109375, 0.508697509765625, 0.01953125, -0.32143402099609375, 1.02685546875, -0.455291748046875, 0.0227813720703125, 0.0, -0.515045166015625, 0.489959716796875, 0.1288604736328125, 1.0482864379882812, -0.20550537109375, 0.18701171875, -1.9276123046875, -0.18294715881347656, 0.692840576171875, 0.08587646484375, 0.90814208984375, -0.17444992065429688, -0.00579833984375, 2.362884521484375, 0.3748016357421875, 0.1251068115234375, 0.3481731414794922, 1.2791748046875, 2.12261962890625, 0.2535400390625, 0.3304443359375, 0.5248870849609375, 0.03866767883300781, -0.029296875, 0.32478904724121094, 0.6183319091796875, -1.9045562744140625, 0.4610595703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000031.npy"}
{"epoch": 0.06492146596858639, "step": 32, "batch_size": 128, "mean": 0.07146459817886353, "std": 0.8189705014228821, "min": -1.9454345703125, "p10": -1.0584625244140624, "median": 0.002475738525390625, "p90": 1.1524398803710938, "max": 2.50775146484375, "pos_frac": 0.5078125, "sample": [-1.146514892578125, -0.24234962463378906, -0.047515869140625, 0.377471923828125, -0.5806427001953125, -1.00030517578125, 0.00162506103515625, 1.5830841064453125, -0.0672454833984375, 0.3966064453125, -0.19842529296875, -0.6075820922851562, 1.3351821899414062, 0.5032539367675781, -0.682159423828125, -0.40311431884765625, 0.0830078125, -0.0513916015625, 1.3369140625, -0.18695068359375, 0.59637451171875, -0.46734619140625, 0.059906005859375, -1.1940155029296875, 1.21136474609375, -0.628814697265625, 0.8985443115234375, -0.32657432556152344, 0.1383819580078125, 0.01453399658203125, -0.8237380981445312, 0.20600128173828125, -1.1381568908691406, -0.71343994140625, -0.01251220703125, 0.049591064453125, -0.4532470703125, -0.33160400390625, -0.689483642578125, -0.1203460693359375, 0.74859619140625, -1.72216796875, 0.68731689453125, -0.18527984619140625, 0.6871185302734375, -0.3431549072265625, 0.986907958984375, -0.31999969482421875, 0.6510162353515625, 0.895172119140625, 0.2750244140625, -1.9454345703125, -1.101348876953125, 0.8304443359375, -0.269775390625, -0.647003173828125, 1.0623779296875, 0.48651123046875, 1.338958740234375, -1.2601318359375, 0.46640777587890625, -0.1162567138671875, 0.43096923828125, 1.2089691162109375, -0.0517578125, 0.2508087158203125, -1.549774169921875, 1.129852294921875, 1.2801513671875, 0.714569091796875, 2.093475341796875, 0.6641006469726562, 0.82843017578125, -1.0474853515625, 0.0522003173828125, 0.7018280029296875, -0.314849853515625, 0.555908203125, -0.03160858154296875, 1.158111572265625, 0.554443359375, -1.3980712890625, -0.02321624755859375, -0.34580230712890625, 1.2501983642578125, 0.372528076171875, 0.84075927734375, 0.0993804931640625, -0.0220184326171875, -0.27923583984375, -0.41900634765625, -0.19028472900390625, -0.0267333984375, 2.50775146484375, -1.084075927734375, 0.828155517578125, -0.91949462890625, -0.394683837890625, 0.46022796630859375, 1.1500091552734375, -0.39009857177734375, -0.345947265625, 0.4473114013671875, 0.04467201232910156, -0.28411865234375, 0.003326416015625, -0.17075729370117188, -0.280181884765625, 0.506103515625, 0.3003120422363281, 0.476318359375, -0.2318878173828125, -0.084808349609375, -1.303314208984375, 1.03363037109375, -0.7669219970703125, 0.2718505859375, -0.341796875, 0.0137939453125, -1.6316375732421875, -0.526611328125, 0.5650310516357422, 0.876617431640625, 1.81585693359375, 0.131927490234375, 0.225921630859375, -1.701995849609375, 1.57647705078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000032.npy"}
{"epoch": 0.06701570680628273, "step": 33, "batch_size": 128, "mean": 0.17560473084449768, "std": 0.9980047941207886, "min": -5.014984130859375, "p10": -0.79893798828125, "median": 0.3173980712890625, "p90": 1.2298736572265625, "max": 2.610107421875, "pos_frac": 0.6328125, "sample": [0.3991661071777344, -0.049591064453125, 0.702056884765625, -0.326171875, -0.5256805419921875, -0.8769683837890625, 0.8020553588867188, -2.691802978515625, -0.76776123046875, 0.4902191162109375, 0.3148193359375, -0.2784271240234375, 1.225799560546875, -1.480499267578125, 0.0, 0.09185791015625, 0.5220184326171875, 0.3037109375, -0.5035018920898438, -1.273681640625, -1.2978515625, 0.3948211669921875, 0.0117034912109375, -0.643035888671875, 0.56353759765625, -0.66876220703125, 0.21417236328125, 0.304779052734375, 1.2393798828125, 0.8138427734375, 1.26611328125, 0.4964599609375, 0.081207275390625, -2.119476318359375, -0.19411468505859375, 2.192291259765625, 0.44171142578125, 0.218231201171875, -0.49493408203125, 0.360076904296875, 0.320770263671875, -0.7760009765625, -0.2776947021484375, 0.74420166015625, 0.826416015625, 0.4391632080078125, 2.26043701171875, 0.7193222045898438, 0.82049560546875, -0.456512451171875, 0.9147186279296875, 0.2982940673828125, -0.7830810546875, 0.0, -0.6343231201171875, 0.53668212890625, 0.1701984405517578, 0.386138916015625, 0.978179931640625, 0.2628173828125, 0.454833984375, -0.386138916015625, 0.33956146240234375, 0.0530242919921875, -0.029767990112304688, -0.8359375, 0.025569915771484375, 0.542022705078125, -1.1856689453125, 1.34002685546875, -0.0324859619140625, 0.8501358032226562, -0.40484619140625, 0.597900390625, -1.392822265625, -0.355804443359375, -1.1488189697265625, 1.203887939453125, -0.6085205078125, -0.4552001953125, 0.98052978515625, 0.929412841796875, 2.2032470703125, 0.44959259033203125, 1.027313232421875, 0.385345458984375, 0.1132659912109375, -0.411773681640625, 0.3267059326171875, 0.9648590087890625, 0.83599853515625, 0.3663597106933594, 1.1219863891601562, 0.30689239501953125, 0.4275627136230469, 1.3214111328125, -0.753021240234375, 0.073333740234375, -5.014984130859375, -0.079833984375, -0.07868194580078125, 0.3955078125, 1.1610260009765625, 1.154815673828125, 0.34436798095703125, 0.35121917724609375, -0.38782501220703125, 1.58416748046875, 1.743804931640625, -0.004852294921875, 1.286712646484375, 1.1243896484375, -1.66522216796875, 0.5556640625, -0.2727203369140625, -0.350921630859375, 0.319976806640625, 0.33060455322265625, 2.610107421875, 0.347930908203125, 2.330230712890625, 0.3812713623046875, -2.16265869140625, -0.4659271240234375, 0.14215087890625, -0.2682228088378906, 0.3385200500488281, 1.47882080078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000033.npy"}
{"epoch": 0.06910994764397906, "step": 34, "batch_size": 128, "mean": 0.19889086484909058, "std": 1.0770306587219238, "min": -2.86224365234375, "p10": -1.0698211669921875, "median": 0.05853843688964844, "p90": 1.5198944091796873, "max": 3.5904541015625, "pos_frac": 0.546875, "sample": [-0.4540557861328125, 0.7423095703125, 1.33477783203125, -0.6234359741210938, -0.568695068359375, -0.2477874755859375, -0.7019500732421875, -0.10908889770507812, 0.90301513671875, 2.1251220703125, -1.646026611328125, -0.6689453125, 1.339569091796875, -1.06951904296875, -1.033233642578125, 1.84771728515625, 1.57568359375, 0.848052978515625, 0.8441925048828125, -0.3265380859375, 0.6709632873535156, -2.75628662109375, 1.16693115234375, -0.394134521484375, -0.3077392578125, 1.16973876953125, 2.13983154296875, 0.057445526123046875, 0.024550437927246094, -0.15948486328125, 0.5784912109375, 1.57757568359375, 0.8861083984375, -1.2161788940429688, -0.3594207763671875, -0.06009674072265625, 0.373260498046875, -0.6028900146484375, -2.09808349609375, 1.019012451171875, 0.8066558837890625, -1.51177978515625, 1.619964599609375, 3.5904541015625, 0.449188232421875, 0.3407440185546875, -0.45819091796875, 0.63446044921875, -1.69921875, -0.3889312744140625, -0.36102294921875, 0.0564117431640625, -0.2708148956298828, 1.39483642578125, -2.86224365234375, -0.3189849853515625, 0.7163238525390625, 1.755615234375, 3.10760498046875, -0.735931396484375, 0.9940185546875, 0.22040557861328125, 0.5149993896484375, -0.07965087890625, -1.070526123046875, -0.4014892578125, 0.0202178955078125, 0.014547348022460938, 0.68743896484375, 1.512176513671875, 0.32280731201171875, -0.154296875, 0.069305419921875, -0.1478118896484375, 0.292724609375, 1.1353607177734375, -0.19472885131835938, 0.12235260009765625, 0.434478759765625, -0.13024330139160156, 0.8824462890625, 1.53790283203125, -0.0224609375, -0.0424346923828125, -0.00567626953125, -0.341766357421875, -0.241485595703125, 1.1405029296875, 0.7103118896484375, 0.86572265625, -0.7839546203613281, -2.532501220703125, 1.02923583984375, -1.251556396484375, 0.047607421875, 0.7955322265625, 0.598236083984375, 0.379730224609375, -1.075164794921875, -0.42517852783203125, -0.7640380859375, 0.46881103515625, 0.05963134765625, -0.4786376953125, 0.9178466796875, -0.16328048706054688, -0.993896484375, -0.6327743530273438, -0.41991424560546875, 2.03564453125, 2.687652587890625, -1.35699462890625, 1.1829681396484375, -0.019744873046875, -0.33392333984375, 0.100555419921875, -1.301055908203125, 1.114501953125, -0.9607391357421875, 2.1690216064453125, 0.880615234375, 1.278411865234375, 0.40636444091796875, 0.549591064453125, 0.48223876953125, -0.055999755859375, 1.0457763671875, 0.4463653564453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000034.npy"}
{"epoch": 0.0712041884816754, "step": 35, "batch_size": 128, "mean": 0.043423622846603394, "std": 1.0108641386032104, "min": -3.576904296875, "p10": -1.01627197265625, "median": 0.042675018310546875, "p90": 1.2018981933593749, "max": 3.304534912109375, "pos_frac": 0.53125, "sample": [-0.6108245849609375, -0.48394775390625, -0.8944244384765625, 0.216644287109375, -0.19152069091796875, 1.82061767578125, 0.8648681640625, 0.0087432861328125, -1.886383056640625, -1.520416259765625, 1.1796875, 1.3225479125976562, -0.0059814453125, 0.1292572021484375, 0.7782058715820312, 0.259674072265625, -3.576904296875, -0.99609375, -0.81158447265625, -0.13945960998535156, 0.6035690307617188, -0.47698974609375, -0.615386962890625, 0.317596435546875, 1.1064453125, 0.3040618896484375, 0.25830078125, 0.04518890380859375, -0.1530303955078125, -0.31072998046875, 0.5146675109863281, -0.08956146240234375, -0.04510498046875, -0.4268798828125, 1.121734619140625, 2.861724853515625, 0.40566253662109375, 0.6849822998046875, 0.6758232116699219, 0.0, -0.660552978515625, -0.047607421875, 0.38653564453125, 1.295684814453125, 0.2338104248046875, -1.182708740234375, 1.68170166015625, -0.5357208251953125, 0.5894317626953125, -1.429351806640625, -3.00518798828125, -0.387725830078125, -0.5404052734375, -0.329864501953125, 1.8037109375, -0.0958404541015625, -0.593963623046875, 0.638214111328125, 0.2886505126953125, -0.16725921630859375, 1.1217041015625, 2.07275390625, 0.20263671875, 0.09348297119140625, 0.8040771484375, 0.032623291015625, 0.70166015625, 0.19346237182617188, 1.17486572265625, 0.15282058715820312, -0.7298736572265625, -0.74945068359375, -1.3148040771484375, 1.3077545166015625, -0.07958221435546875, 0.126373291015625, -0.1647491455078125, 1.1033172607421875, 0.50054931640625, 1.25372314453125, -0.14197921752929688, 0.5269088745117188, -0.35791015625, 0.2630157470703125, 0.48122406005859375, -1.6855926513671875, -0.06842422485351562, -0.642181396484375, -0.327667236328125, 0.0401611328125, -1.0633544921875, -0.054721832275390625, 0.32784461975097656, 0.43927001953125, 0.10030364990234375, 0.7158203125, -0.0196533203125, -0.779541015625, 0.7884521484375, 3.304534912109375, -1.9273681640625, 1.675506591796875, 0.0, -0.42266845703125, -0.788299560546875, 0.43914794921875, 0.50494384765625, 1.699920654296875, 0.294769287109375, 0.4262542724609375, 0.187774658203125, 0.020965576171875, -1.171722412109375, -0.2682952880859375, 0.5202217102050781, -0.343505859375, 0.06913375854492188, 0.617401123046875, -0.121734619140625, -0.9344100952148438, 0.75347900390625, -0.949005126953125, -0.3930206298828125, -0.9605712890625, 0.0821990966796875, -2.79986572265625, -2.0072479248046875, 1.518035888671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000035.npy"}
{"epoch": 0.07329842931937172, "step": 36, "batch_size": 128, "mean": 0.315809041261673, "std": 1.1578891277313232, "min": -2.70819091796875, "p10": -0.91646728515625, "median": 0.228271484375, "p90": 1.5447631835937499, "max": 5.18121337890625, "pos_frac": 0.6171875, "sample": [2.93499755859375, -1.826904296875, 0.83892822265625, -0.7858657836914062, 0.29486083984375, 0.29864501953125, 1.31005859375, 0.20532989501953125, 0.99298095703125, -0.061733245849609375, 1.2315673828125, -0.3119659423828125, -0.1278076171875, -0.380035400390625, -0.6321258544921875, 0.26596832275390625, -0.03891754150390625, 0.33868408203125, 1.4326171875, 0.18764495849609375, 0.04338836669921875, -2.35101318359375, 0.4584236145019531, 0.0, 0.0385589599609375, 0.047882080078125, 1.3267822265625, -1.1131591796875, -0.2039794921875, 0.08282852172851562, 5.18121337890625, 0.45782470703125, 0.11395263671875, 1.2066192626953125, 1.47662353515625, -0.2801513671875, 0.0313873291015625, -0.021453857421875, -0.028881072998046875, 0.461517333984375, 4.006866455078125, 0.13397979736328125, 0.6618194580078125, 0.57574462890625, -0.466278076171875, -0.24224853515625, 0.4755096435546875, 1.0424652099609375, -2.0557479858398438, 0.48828125, 0.76849365234375, 1.013702392578125, 0.38824462890625, -0.6074295043945312, -1.9156646728515625, 0.846771240234375, 0.1640167236328125, -0.019561767578125, 2.537017822265625, 1.3614501953125, 0.0, -1.803009033203125, 0.58062744140625, 0.9251937866210938, 1.3275299072265625, 0.6572265625, 0.0186767578125, 1.0300064086914062, -0.12591552734375, -2.031890869140625, 0.75421142578125, 1.55706787109375, -0.592041015625, 1.6331634521484375, -0.2145538330078125, -0.02242279052734375, -0.22040557861328125, 1.06097412109375, -2.37774658203125, 0.56964111328125, 1.09814453125, -0.9628448486328125, 0.239044189453125, 2.24481201171875, -1.82208251953125, -0.2940235137939453, 1.39691162109375, -1.5000152587890625, 0.141204833984375, 0.599517822265625, 1.53948974609375, 0.79998779296875, 0.41640472412109375, 0.9972763061523438, -2.70819091796875, -0.31744384765625, -0.14921188354492188, -0.023040771484375, -0.5666046142578125, 1.506500244140625, 1.8416748046875, 0.415283203125, 1.179901123046875, 1.7462005615234375, -0.598541259765625, 0.6218032836914062, 2.382232666015625, -0.26531982421875, -1.49896240234375, -0.0169525146484375, -0.28108787536621094, 0.24114990234375, 0.2407989501953125, 0.86968994140625, 0.0370635986328125, -0.1454010009765625, 0.822723388671875, 1.99365234375, 1.97857666015625, -0.132171630859375, 0.217498779296875, 1.75830078125, 0.306610107421875, 0.05472373962402344, 0.808868408203125, -0.0299072265625, -0.8965911865234375, -0.8431549072265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000036.npy"}
{"epoch": 0.07539267015706806, "step": 37, "batch_size": 128, "mean": 0.12415631860494614, "std": 1.045518398284912, "min": -2.477783203125, "p10": -1.3019119262695311, "median": 0.21428966522216797, "p90": 1.4300086975097654, "max": 3.26361083984375, "pos_frac": 0.578125, "sample": [-1.0696563720703125, 0.1737213134765625, 0.8463134765625, 1.4168701171875, 0.62261962890625, 0.13868141174316406, 0.3167724609375, 0.3948402404785156, 0.2481842041015625, -0.675506591796875, 0.3367919921875, 0.2974090576171875, -1.5748291015625, 0.42409515380859375, 0.3278656005859375, 1.0426025390625, 0.50689697265625, 0.0921630859375, -0.793426513671875, -0.071258544921875, -0.28619384765625, 0.3536376953125, 0.05180168151855469, -1.43365478515625, -1.45648193359375, 0.4082489013671875, 0.30129241943359375, -0.440521240234375, 0.9863662719726562, 0.6282196044921875, 0.3316802978515625, -0.37436676025390625, -0.23316001892089844, 0.276580810546875, 0.41530609130859375, 1.084564208984375, -1.2681884765625, 2.4364013671875, 2.60498046875, -0.143798828125, 0.22676658630371094, -1.57257080078125, -0.22356414794921875, 0.201812744140625, 0.9388275146484375, 0.126953125, 0.15510845184326172, -0.29596710205078125, 1.4464950561523438, -1.153594970703125, 0.0564727783203125, 0.4655303955078125, -1.72528076171875, 1.76788330078125, -1.953338623046875, -0.539764404296875, 0.71051025390625, 0.5555419921875, -0.3758544921875, -0.05078125, 1.422943115234375, 0.74005126953125, 0.28908538818359375, -0.94677734375, -1.760833740234375, 0.1678009033203125, -2.477783203125, 0.6171875, 0.58837890625, 1.822998046875, 0.57373046875, -0.217132568359375, 1.523162841796875, -0.193817138671875, 0.3159027099609375, 0.26165771484375, -0.22325897216796875, 0.472381591796875, -0.0645751953125, -1.8040771484375, 0.45526885986328125, 2.4423828125, 0.4952392578125, -1.0069580078125, 0.113922119140625, -0.555633544921875, -0.804229736328125, 0.2591400146484375, 1.327239990234375, 0.41070556640625, -1.464385986328125, 0.6571044921875, -0.483306884765625, -0.00299072265625, -0.23691558837890625, 0.7026100158691406, -0.03607177734375, -0.234130859375, -1.197967529296875, -0.88116455078125, -0.4495849609375, -0.5581245422363281, -1.00518798828125, -0.8719024658203125, 0.4967498779296875, 0.594696044921875, 3.26361083984375, -0.144134521484375, 1.4942474365234375, -0.16041946411132812, 1.476165771484375, -0.149993896484375, 0.6581268310546875, -2.2568359375, 1.8526611328125, 1.065826416015625, 0.527923583984375, -0.317047119140625, 0.3997955322265625, -1.3805999755859375, 0.793121337890625, -2.1271820068359375, 0.460205078125, 0.99444580078125, 2.45611572265625, -0.539031982421875, -0.615814208984375, 2.86224365234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000037.npy"}
{"epoch": 0.0774869109947644, "step": 38, "batch_size": 128, "mean": 0.17567716538906097, "std": 1.039191484451294, "min": -3.053985595703125, "p10": -1.0228057861328124, "median": 0.22185897827148438, "p90": 1.4870811462402342, "max": 3.76580810546875, "pos_frac": 0.59375, "sample": [0.91748046875, -0.1618194580078125, -0.15213394165039062, 0.7321243286132812, 0.24253082275390625, 0.06863784790039062, 0.7105636596679688, 2.447021484375, 0.3583984375, 1.601531982421875, 1.6474609375, -0.11656951904296875, 0.419952392578125, 3.344390869140625, -0.3013916015625, 1.800262451171875, -0.9345703125, 0.4090423583984375, 0.030853271484375, 0.8035430908203125, -0.267669677734375, -1.17425537109375, 0.28829193115234375, -0.6647567749023438, 0.48389434814453125, 0.7909088134765625, -0.795166015625, -0.7386474609375, -0.41400146484375, -0.15527915954589844, 0.3126983642578125, 0.2011871337890625, 0.0418701171875, -1.164306640625, 0.8759994506835938, 0.1264190673828125, 0.536285400390625, -0.1610870361328125, 0.637939453125, -0.996795654296875, -0.17877197265625, 0.2767295837402344, -0.455322265625, 0.37958526611328125, 0.3558349609375, 0.4787750244140625, 0.10589599609375, -1.08349609375, 0.6709747314453125, 0.4308929443359375, -0.07232666015625, -0.11576080322265625, -0.119049072265625, -0.12294769287109375, 0.4948768615722656, -0.7418060302734375, 0.3055419921875, -0.2292938232421875, -1.3017730712890625, -0.1961212158203125, 0.2454986572265625, -1.1243896484375, -0.187835693359375, 1.0650787353515625, 0.0096435546875, 0.0053157806396484375, 1.057830810546875, -2.04803466796875, 0.781036376953125, 0.098846435546875, -0.2587890625, -0.06048583984375, 1.136199951171875, -0.7169189453125, 0.34100341796875, -1.1669921875, 2.0159759521484375, 0.41713714599609375, -0.119415283203125, -3.053985595703125, 0.5065994262695312, -1.1154861450195312, 0.466949462890625, -0.29638671875, -0.45902252197265625, 2.673675537109375, 0.48919677734375, -0.5863494873046875, -0.011016845703125, 0.42357635498046875, 0.6820755004882812, 0.3658447265625, 3.76580810546875, 0.06610107421875, 0.79864501953125, 0.45781707763671875, 0.7137870788574219, -0.36492919921875, 1.877532958984375, -2.769317626953125, 0.13205718994140625, 0.0, 0.928009033203125, -2.95245361328125, -0.4458770751953125, -0.868408203125, 1.65545654296875, -0.05712890625, 0.2694091796875, 0.483367919921875, 1.517730712890625, -1.499176025390625, -0.880859375, 0.333740234375, 1.63232421875, 0.198638916015625, 1.55474853515625, 0.905609130859375, 1.0424575805664062, 0.9124755859375, 0.6866455078125, -0.4836273193359375, -2.22332763671875, 1.0416488647460938, 0.3812255859375, 1.4739456176757812, -0.18910789489746094, 0.30205535888671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000038.npy"}
{"epoch": 0.07958115183246073, "step": 39, "batch_size": 128, "mean": 0.4021110236644745, "std": 0.904716432094574, "min": -2.45098876953125, "p10": -0.6967819213867187, "median": 0.28021240234375, "p90": 1.58023681640625, "max": 3.25360107421875, "pos_frac": 0.6953125, "sample": [1.081634521484375, -0.296295166015625, -1.04412841796875, -1.043060302734375, 0.3668212890625, 0.0861053466796875, 0.7132110595703125, -0.0609588623046875, -1.13555908203125, 1.67010498046875, 0.747344970703125, 1.193572998046875, 0.863800048828125, 0.481292724609375, 1.602813720703125, 1.13128662109375, -0.4564666748046875, 1.853302001953125, 0.9900360107421875, -1.466888427734375, 0.6297760009765625, 1.96435546875, 0.3776512145996094, -0.08013916015625, 1.591217041015625, 0.7842330932617188, 1.0744552612304688, -0.3841705322265625, 0.175628662109375, 0.67266845703125, 1.738677978515625, 0.172027587890625, 1.36077880859375, -0.46368408203125, 1.29156494140625, -1.149658203125, -0.2450714111328125, 0.6314544677734375, 2.0621337890625, -0.0454864501953125, -0.0167236328125, -0.03146934509277344, 1.492095947265625, 0.8578529357910156, -0.5450172424316406, 0.49163818359375, 1.49688720703125, 0.0005359649658203125, 2.1065216064453125, 0.30462646484375, -2.152313232421875, -0.4285888671875, -0.62109375, 1.579833984375, 0.8737335205078125, 0.07568359375, 0.70281982421875, 1.261688232421875, -2.45098876953125, -0.935882568359375, -0.3708686828613281, -0.2207183837890625, 0.36743927001953125, 0.517791748046875, 0.679840087890625, 1.48736572265625, 0.092437744140625, -0.16596603393554688, 1.7357940673828125, -0.7059783935546875, 1.5811767578125, 0.22601318359375, -0.055820465087890625, 0.09661865234375, 0.02191162109375, 1.3253173828125, 1.11431884765625, 0.346282958984375, 0.8081741333007812, 0.15643310546875, 1.330657958984375, -0.185638427734375, 1.04583740234375, 0.200408935546875, 0.99029541015625, 0.1142425537109375, 2.27337646484375, 1.34161376953125, 0.5746307373046875, -0.9395065307617188, 0.10739898681640625, 0.04327392578125, -0.5350608825683594, -0.7159919738769531, 0.0, 0.736114501953125, 0.4677276611328125, -0.1320343017578125, 1.745574951171875, 1.222930908203125, 1.16119384765625, 0.205718994140625, 1.176300048828125, -0.0987548828125, 0.0946044921875, 0.78302001953125, -1.507904052734375, 0.9976348876953125, 1.078338623046875, -0.291290283203125, 0.5203704833984375, 0.18621826171875, -0.07560157775878906, 0.5814704895019531, -0.692840576171875, 0.17021560668945312, -0.264007568359375, 0.25579833984375, 0.0662689208984375, 0.06322479248046875, -0.884185791015625, 0.205596923828125, 0.1141510009765625, 3.25360107421875, 1.00030517578125, 0.237823486328125, 0.0644378662109375, 0.84686279296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000039.npy"}
{"epoch": 0.08167539267015707, "step": 40, "batch_size": 128, "mean": 0.4939156174659729, "std": 1.1126805543899536, "min": -1.957916259765625, "p10": -0.6107139587402344, "median": 0.36800384521484375, "p90": 1.927584838867187, "max": 4.6414794921875, "pos_frac": 0.640625, "sample": [-0.10442161560058594, -1.206756591796875, -0.1943511962890625, 0.2617645263671875, 0.07238197326660156, 1.2581787109375, -1.64642333984375, 3.410064697265625, 0.6152572631835938, 2.35028076171875, 0.878936767578125, 0.6076812744140625, -0.7950897216796875, 0.66009521484375, -0.4488067626953125, 0.5475654602050781, -0.00365447998046875, 1.0810546875, -0.10178756713867188, 0.28160858154296875, 0.452239990234375, 2.9002685546875, -0.20952987670898438, -0.5552215576171875, 0.17679214477539062, 2.132568359375, -1.4693603515625, 0.6061248779296875, 0.05511474609375, 1.325836181640625, -0.60772705078125, 0.79888916015625, -0.31749725341796875, 0.6474609375, 2.623748779296875, -0.68499755859375, 0.0713043212890625, 0.39483642578125, 3.6942138671875, 1.872314453125, 4.34906005859375, 1.579559326171875, -0.11669921875, -0.53448486328125, -0.381072998046875, -0.14297866821289062, 0.3913116455078125, 1.0608367919921875, 0.580078125, 1.7396697998046875, 0.6974563598632812, 0.3271636962890625, 0.525360107421875, -0.135955810546875, 1.067657470703125, 4.6414794921875, 0.38934326171875, 0.5869522094726562, 1.5225830078125, -0.271087646484375, -0.278656005859375, -0.0610504150390625, 1.185394287109375, -1.150146484375, 1.2708740234375, -0.5718994140625, 0.62420654296875, 0.8766326904296875, 1.20123291015625, 0.642120361328125, -0.6176834106445312, -0.5402679443359375, 0.3466644287109375, 2.421905517578125, 0.29949951171875, -0.8527717590332031, 0.12762451171875, 0.4025459289550781, 1.682891845703125, -0.0203857421875, 0.8536834716796875, 0.4827880859375, 0.34658050537109375, 0.10564804077148438, -0.584625244140625, 0.5180168151855469, -0.1293182373046875, 2.056549072265625, -0.23363113403320312, -0.24481201171875, -0.31329345703125, 0.61767578125, 1.118621826171875, 2.352630615234375, -1.42633056640625, 1.117156982421875, -1.40582275390625, -0.110992431640625, -0.52655029296875, 0.8395004272460938, 1.320465087890625, 0.51666259765625, 0.553985595703125, 0.21585845947265625, -0.0391387939453125, 0.23614501953125, -0.808135986328125, -0.07387161254882812, -0.33721923828125, 1.832427978515625, 0.50177001953125, 0.9835357666015625, 0.913818359375, 0.314117431640625, 2.216094970703125, 0.10052490234375, -1.957916259765625, -0.3677215576171875, 1.39752197265625, 0.98065185546875, 0.634490966796875, 0.3179206848144531, -0.0016937255859375, 0.4891357421875, 0.148040771484375, 2.369842529296875, -0.737640380859375, -0.227874755859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000040.npy"}
{"epoch": 0.08376963350785341, "step": 41, "batch_size": 128, "mean": 0.15818291902542114, "std": 0.9850521087646484, "min": -3.0408935546875, "p10": -0.95404052734375, "median": 0.14023971557617188, "p90": 1.4445785522460937, "max": 2.60955810546875, "pos_frac": 0.5546875, "sample": [1.883880615234375, -0.534210205078125, 0.487884521484375, -1.2361602783203125, 0.86480712890625, -1.85919189453125, -1.734283447265625, 0.6867828369140625, 0.0092926025390625, 0.20966339111328125, 0.8550567626953125, 1.098785400390625, 0.390838623046875, -0.5678558349609375, -1.1403350830078125, 1.112884521484375, -0.1626739501953125, 0.0318145751953125, 1.040771484375, 1.416168212890625, -0.70220947265625, 1.5789337158203125, -0.1691741943359375, 1.2664794921875, 0.556396484375, 0.273406982421875, 0.475616455078125, 0.35101318359375, -0.7867279052734375, 0.5667057037353516, 1.0474166870117188, -0.463958740234375, -0.052219390869140625, -0.63330078125, -0.33697509765625, -0.2476329803466797, 2.60955810546875, 1.680938720703125, 0.5718841552734375, 0.335968017578125, 0.5648193359375, 0.8799362182617188, 1.1022491455078125, 1.8558349609375, 0.4320526123046875, 0.1027069091796875, 0.184661865234375, 0.2583770751953125, -0.601318359375, 1.9525146484375, -0.1551666259765625, -0.33011627197265625, -0.49725341796875, -0.1882476806640625, 2.16082763671875, -2.7955322265625, -0.36086273193359375, -0.71435546875, 0.07403564453125, 0.6268844604492188, -0.03759765625, 0.0836181640625, -0.2641754150390625, 0.58837890625, 0.068634033203125, 1.4220733642578125, -0.768035888671875, -0.7074737548828125, 0.0, -0.37139892578125, -0.051876068115234375, -0.4412841796875, -1.942169189453125, -0.7720794677734375, -0.94671630859375, -0.33349609375, 1.2331771850585938, -0.5945205688476562, 0.544708251953125, -1.20538330078125, 0.9621429443359375, 0.4924278259277344, 0.11370086669921875, 0.7304458618164062, 1.431884765625, 1.072113037109375, 0.22150421142578125, -0.35772705078125, 1.4741973876953125, -0.0850372314453125, -0.28631591796875, 1.0802001953125, 0.5545196533203125, -0.4644927978515625, -0.453948974609375, 1.703094482421875, -1.107818603515625, -0.770263671875, -0.3674945831298828, -1.0410919189453125, -0.4036865234375, 0.166778564453125, 0.58258056640625, 0.43042755126953125, -1.325347900390625, 0.7913894653320312, 0.23077392578125, 0.8313522338867188, -1.525909423828125, -0.33563232421875, 1.1532459259033203, 0.22711181640625, 0.7193603515625, -0.456512451171875, 1.0579833984375, -0.6182861328125, 1.547393798828125, -0.52728271484375, 1.1408157348632812, -0.6522064208984375, -3.0408935546875, 0.25640869140625, 2.0169677734375, 1.590484619140625, 1.6883163452148438, -0.97113037109375, 0.8411712646484375, -0.90277099609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000041.npy"}
{"epoch": 0.08586387434554973, "step": 42, "batch_size": 128, "mean": 0.32735371589660645, "std": 1.0300960540771484, "min": -2.6441650390625, "p10": -0.6872661590576171, "median": 0.17301559448242188, "p90": 1.7215515136718749, "max": 3.22442626953125, "pos_frac": 0.625, "sample": [-0.4788475036621094, 0.17665863037109375, 0.81195068359375, 1.0789794921875, 0.46044921875, 1.456207275390625, -0.4972686767578125, 1.87274169921875, 0.09130859375, -0.39005279541015625, 0.643524169921875, 2.849761962890625, 0.392578125, 0.16049957275390625, -2.339447021484375, -0.2500648498535156, 0.146942138671875, 1.69683837890625, 0.2728118896484375, 2.764556884765625, -0.16284942626953125, 0.12003326416015625, -0.9097442626953125, 0.2798309326171875, 2.082244873046875, -0.046112060546875, -0.39129638671875, 0.039920806884765625, 0.93389892578125, -0.5149688720703125, -0.0771484375, 0.6719150543212891, 0.567718505859375, 0.41485595703125, 2.0548095703125, -0.024932861328125, 2.2817230224609375, -0.04764556884765625, -0.6694259643554688, 0.44891357421875, -0.0618896484375, 0.2793693542480469, 0.33673095703125, 0.2856292724609375, 0.49462890625, -0.1129913330078125, 0.6432723999023438, -0.008087158203125, 0.75531005859375, 0.06764602661132812, 0.2372589111328125, 0.47895050048828125, 0.6826019287109375, -0.049892425537109375, -0.13622665405273438, 0.00904083251953125, -0.174896240234375, -0.635833740234375, 0.14263916015625, -0.41693878173828125, 1.984649658203125, -0.311676025390625, -0.205657958984375, -1.1641845703125, 0.488677978515625, -0.77276611328125, 0.8381881713867188, 0.003387451171875, -0.7288932800292969, 0.3577251434326172, -0.197662353515625, 0.15205001831054688, 1.650054931640625, 1.471435546875, 0.572265625, 0.016038894653320312, -2.6441650390625, -0.44852447509765625, 1.2701416015625, -0.4466056823730469, 0.454498291015625, 1.15142822265625, -0.737335205078125, 0.3946075439453125, 1.286163330078125, -0.5661468505859375, 1.34942626953125, 0.5992889404296875, 0.11704826354980469, -0.945343017578125, 1.682464599609375, -1.76043701171875, 1.1904296875, 0.2523651123046875, 0.69744873046875, 0.16937255859375, -1.2955322265625, 2.87353515625, 0.81341552734375, 0.48184967041015625, 0.9970855712890625, 0.07358551025390625, 0.0, 2.1971435546875, -0.53900146484375, -0.17161941528320312, 1.7703857421875, 0.291229248046875, -0.1259918212890625, 0.7324066162109375, 0.14361572265625, -0.983551025390625, 1.0917434692382812, 0.11535835266113281, 1.9979248046875, 3.22442626953125, 1.70062255859375, -0.01828765869140625, 1.4605636596679688, 0.1945648193359375, -0.533538818359375, 2.341888427734375, -1.95611572265625, 0.4225921630859375, -2.452667236328125, -0.2214813232421875, -0.655731201171875, -0.07709503173828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000042.npy"}
{"epoch": 0.08795811518324607, "step": 43, "batch_size": 128, "mean": 0.44660818576812744, "std": 1.2831213474273682, "min": -4.303802490234375, "p10": -0.8493728637695312, "median": 0.28632354736328125, "p90": 2.255926513671875, "max": 3.473724365234375, "pos_frac": 0.6328125, "sample": [0.001800537109375, 2.4149169921875, 1.86834716796875, -1.51629638671875, 0.2651824951171875, 1.1382598876953125, 2.8753662109375, 0.10390472412109375, -0.16222381591796875, 0.28546905517578125, -0.2753639221191406, 1.070648193359375, -0.7267303466796875, 3.473724365234375, -0.2595672607421875, -1.6907958984375, -0.523193359375, -0.8205490112304688, 2.3717041015625, 0.051025390625, 2.15423583984375, 0.4591522216796875, 2.33062744140625, 2.0416107177734375, -0.744537353515625, -1.8638763427734375, -1.0537567138671875, 0.6439056396484375, -0.11513137817382812, 2.347747802734375, 1.64794921875, 0.8975372314453125, -0.44476318359375, -1.5257568359375, 3.17108154296875, 0.5725746154785156, 0.28717803955078125, 0.12067413330078125, 0.3576393127441406, 0.5534210205078125, -0.8407440185546875, 1.469482421875, -0.2278289794921875, 0.7802734375, 0.3554229736328125, -0.11452102661132812, -0.0997467041015625, -0.5409317016601562, -4.303802490234375, -0.4961700439453125, 0.556610107421875, 0.7375717163085938, 0.22301101684570312, 1.0339813232421875, 3.1260986328125, 1.100341796875, 2.5400390625, 1.4938812255859375, -0.380706787109375, 2.025390625, 0.3782196044921875, -0.422760009765625, 1.720916748046875, 0.1397552490234375, 2.87042236328125, 2.23382568359375, -0.097076416015625, -0.042236328125, 0.7449798583984375, 2.5311126708984375, -0.081329345703125, 0.819183349609375, 0.445892333984375, 0.13775634765625, 0.99786376953125, -0.049816131591796875, -2.3777008056640625, -0.228057861328125, 0.23388671875, 0.68218994140625, 1.0080184936523438, 1.64599609375, -0.60064697265625, 0.281219482421875, 0.8297882080078125, -0.063934326171875, 0.9752960205078125, -0.1931915283203125, 0.146240234375, 0.178985595703125, 1.1046295166015625, -0.13836669921875, -1.036865234375, 1.5299453735351562, 2.3074951171875, -0.05574798583984375, 0.09342193603515625, 2.43792724609375, -0.8695068359375, 0.6013412475585938, 0.35784912109375, 1.3255691528320312, 1.8477783203125, -0.3238792419433594, -1.46533203125, -1.10308837890625, 1.9547119140625, -3.07208251953125, 1.8775482177734375, 1.58856201171875, 0.304473876953125, 1.73138427734375, 0.01200103759765625, -2.5176010131835938, 0.15289306640625, -0.6683502197265625, 0.036224365234375, 0.6231536865234375, -0.34881591796875, 0.8321533203125, -0.3574676513671875, -0.53997802734375, 0.31638526916503906, 1.944000244140625, 1.361572265625, -0.31357574462890625, 0.6680450439453125, -0.09815597534179688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000043.npy"}
{"epoch": 0.09005235602094241, "step": 44, "batch_size": 128, "mean": 0.5962188839912415, "std": 1.3836408853530884, "min": -3.582916259765625, "p10": -0.8829238891601561, "median": 0.5068321228027344, "p90": 2.3228271484375, "max": 4.629150390625, "pos_frac": 0.734375, "sample": [0.857940673828125, 0.40093994140625, 2.2152099609375, -0.33946990966796875, -0.6201171875, 2.4119873046875, 0.221099853515625, 0.07316970825195312, -0.2641448974609375, 0.021244049072265625, 0.3745880126953125, 1.099273681640625, 0.5169677734375, -0.26202392578125, 0.843109130859375, 0.899871826171875, -0.838836669921875, 0.49071502685546875, -2.03240966796875, 2.53729248046875, 1.241607666015625, 1.0675048828125, 0.533416748046875, 0.3927497863769531, 0.8978996276855469, 0.03078460693359375, 1.822998046875, 0.4221057891845703, -3.461822509765625, 0.0083465576171875, 1.831329345703125, -0.600799560546875, -0.3442840576171875, -0.6908721923828125, 0.3780326843261719, 0.103271484375, 1.032257080078125, -0.8069438934326172, 3.5931396484375, 0.845611572265625, 4.629150390625, -0.26724815368652344, 1.008544921875, 2.431976318359375, 3.097808837890625, 2.206512451171875, 0.767913818359375, 2.795318603515625, 1.6192626953125, 0.0158843994140625, 0.6961898803710938, 1.0615234375, -1.6340408325195312, 0.723785400390625, -0.0322113037109375, -0.3650665283203125, -0.546051025390625, 0.9097900390625, 0.5126495361328125, 1.1981201171875, 0.7910919189453125, 1.972991943359375, 0.7480659484863281, 3.66986083984375, -0.160980224609375, 0.7309112548828125, -0.52459716796875, -3.582916259765625, 1.4752197265625, 2.8441162109375, -1.147796630859375, 0.6714591979980469, -0.082275390625, 0.89532470703125, 1.751800537109375, 1.084808349609375, 0.9356689453125, -1.532135009765625, 0.4313812255859375, -2.133026123046875, 3.152008056640625, 0.4907684326171875, 1.99267578125, 0.15445899963378906, 0.1236572265625, 0.5688858032226562, 0.319854736328125, 2.06195068359375, 2.295654296875, 0.5010147094726562, -1.01513671875, 2.38623046875, 1.599609375, 0.072845458984375, 0.0460052490234375, 2.21148681640625, 0.5009765625, -0.2852783203125, 0.533721923828125, 0.206573486328125, 1.4725341796875, 0.0095367431640625, -1.045654296875, -0.24755859375, 0.3637199401855469, -0.74749755859375, -1.494781494140625, 2.233673095703125, 0.0354156494140625, -2.348541259765625, 2.999420166015625, 0.245697021484375, 0.28066253662109375, 4.285675048828125, 0.07485198974609375, 0.627288818359375, 0.7716522216796875, -0.052825927734375, -2.67578125, 0.4059181213378906, -0.9857940673828125, 0.7957534790039062, 0.8206329345703125, 1.944427490234375, 1.87554931640625, -0.48873138427734375, 0.9981536865234375, 1.673126220703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000044.npy"}
{"epoch": 0.09214659685863874, "step": 45, "batch_size": 128, "mean": 0.6198710799217224, "std": 1.4243115186691284, "min": -3.790771484375, "p10": -0.7343368530273436, "median": 0.37808799743652344, "p90": 2.34696044921875, "max": 5.40826416015625, "pos_frac": 0.6875, "sample": [-0.135284423828125, 0.069854736328125, 1.6450958251953125, -0.31580352783203125, -0.06761741638183594, 0.16925048828125, 0.2414989471435547, 0.895782470703125, 0.32605743408203125, 2.39447021484375, 3.10052490234375, 2.8848876953125, 1.97979736328125, 0.18997955322265625, 1.2288970947265625, 0.7346572875976562, -0.06627655029296875, 0.8778076171875, 0.0, -0.8619384765625, 1.04229736328125, 0.0951080322265625, -0.7786102294921875, 0.8424606323242188, -0.715362548828125, 3.28436279296875, -0.0681915283203125, 0.379638671875, 0.818603515625, 0.07332611083984375, -0.2171173095703125, -0.3402442932128906, 0.20330810546875, 0.6488189697265625, -0.6160850524902344, 1.24346923828125, 0.647064208984375, 0.3011474609375, 2.189697265625, 1.84259033203125, 4.9613037109375, -0.94122314453125, 0.33571624755859375, 0.41064453125, 3.278411865234375, 0.2135009765625, 0.548370361328125, 0.17264556884765625, -0.5208740234375, 0.5410385131835938, -0.954742431640625, -0.1439208984375, -3.650146484375, 0.854705810546875, 0.19234848022460938, -0.313232421875, 0.2439727783203125, 1.1895294189453125, -0.4718360900878906, -3.790771484375, -0.6395263671875, 1.6849365234375, -0.26911163330078125, 0.278076171875, -1.51593017578125, 0.970916748046875, 0.34759521484375, 0.17072296142578125, 1.04229736328125, 0.1330108642578125, -0.059234619140625, 1.859832763671875, 0.2224273681640625, 0.8794326782226562, 2.76214599609375, 0.9335479736328125, 0.647552490234375, -0.514251708984375, 1.02911376953125, -0.070953369140625, 0.847076416015625, 2.02667236328125, 0.5772132873535156, 2.9349517822265625, 0.3765373229980469, -1.298980712890625, -0.315399169921875, 0.582977294921875, -0.8971481323242188, -2.5228271484375, 0.9908447265625, 2.54083251953125, 0.0031890869140625, -1.1117095947265625, 4.9061279296875, 1.4180374145507812, 1.82464599609375, -0.3629150390625, 1.33551025390625, -0.510528564453125, -1.1275634765625, 0.587066650390625, 0.7849655151367188, -0.209442138671875, 1.666351318359375, 3.27825927734375, 0.9238739013671875, -1.1782684326171875, 0.8519973754882812, 0.2874755859375, -0.3559417724609375, 0.107025146484375, 1.4517669677734375, 2.32659912109375, 0.49367523193359375, 0.227020263671875, 0.7573394775390625, 1.204803466796875, 5.40826416015625, 1.349151611328125, 1.414398193359375, 2.301361083984375, 0.6031875610351562, -0.6638336181640625, 5.0118408203125, -0.03448486328125, -0.2421875, 0.5577239990234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000045.npy"}
{"epoch": 0.09424083769633508, "step": 46, "batch_size": 128, "mean": 0.6784305572509766, "std": 1.2125465869903564, "min": -2.864288330078125, "p10": -0.6742889404296875, "median": 0.5849380493164062, "p90": 2.27982177734375, "max": 3.96563720703125, "pos_frac": 0.6796875, "sample": [3.310791015625, 3.740234375, 0.7303619384765625, 0.7159500122070312, 0.7082061767578125, 1.77337646484375, 3.00811767578125, -0.4881591796875, -0.9471435546875, -0.036651611328125, -1.77935791015625, 0.10260009765625, 1.28936767578125, -1.443511962890625, 0.866729736328125, -0.548370361328125, -0.5738525390625, -0.29986000061035156, 0.6321563720703125, -0.047393798828125, 1.28900146484375, 1.5782623291015625, -0.158203125, 1.5460968017578125, -0.101409912109375, -0.0627288818359375, -0.673583984375, -0.0773162841796875, 3.96563720703125, -0.2585887908935547, -0.03712654113769531, 0.1159515380859375, 1.946990966796875, 1.154022216796875, 1.578277587890625, 2.18084716796875, 0.8692626953125, 0.250030517578125, 0.8509063720703125, 1.27655029296875, -0.199737548828125, 1.239105224609375, 1.6920013427734375, 1.3383636474609375, -0.025146484375, -2.52545166015625, 1.986724853515625, 0.23301315307617188, 0.6981964111328125, 0.26908111572265625, 2.51544189453125, 2.5758819580078125, 1.0850677490234375, 0.2293548583984375, -0.2666473388671875, 0.85980224609375, 1.25738525390625, 2.13543701171875, -0.6893081665039062, 1.951873779296875, -0.028350830078125, 0.3010139465332031, -0.7214584350585938, 0.0598907470703125, 2.250579833984375, -1.21832275390625, 0.2685813903808594, -0.60504150390625, -0.938232421875, 1.7454833984375, -0.869232177734375, 0.22161865234375, 2.23992919921875, 0.666351318359375, -0.636199951171875, 1.58642578125, -0.6529350280761719, -0.18912506103515625, 0.43128204345703125, -0.675933837890625, 1.0884475708007812, 2.669891357421875, 1.8072509765625, -0.28814697265625, 1.67138671875, 0.46002960205078125, -2.864288330078125, 0.6546630859375, 0.8864288330078125, 1.246795654296875, -0.7494659423828125, -0.25445556640625, 2.4794921875, 0.16583251953125, 0.186370849609375, 2.12176513671875, 0.26029205322265625, 0.049163818359375, 0.6756591796875, 1.2064666748046875, 0.17899513244628906, 1.114349365234375, 0.9189910888671875, 0.1815643310546875, 3.055908203125, 1.003387451171875, 0.9062957763671875, 0.7873382568359375, 0.84173583984375, 1.02734375, 2.348052978515625, -1.2279052734375, -0.03711700439453125, 0.3110485076904297, -0.367645263671875, 2.5267333984375, -0.05426025390625, 1.7032852172851562, 0.4949951171875, 1.73272705078125, 0.0, 0.5377197265625, 0.52587890625, -0.5134201049804688, 0.0753173828125, 2.67291259765625, 2.12286376953125, 2.9552001953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000046.npy"}
{"epoch": 0.09633507853403141, "step": 47, "batch_size": 128, "mean": 0.3133854269981384, "std": 1.4404650926589966, "min": -4.397705078125, "p10": -1.550018310546875, "median": 0.4456939697265625, "p90": 1.7712867736816396, "max": 4.109771728515625, "pos_frac": 0.6328125, "sample": [0.72930908203125, 0.5847148895263672, 1.6700363159179688, -0.002593994140625, 3.5926513671875, -2.32373046875, -4.05072021484375, 0.8850555419921875, 2.56732177734375, 0.943695068359375, 0.080902099609375, 0.8064002990722656, 0.20281982421875, -0.7729263305664062, -1.82781982421875, -0.3654327392578125, 1.1808013916015625, -0.3400421142578125, 0.2867431640625, -1.581085205078125, 0.4742431640625, -1.96185302734375, 0.7423095703125, -0.495880126953125, -0.0703125, 1.357635498046875, 0.2773284912109375, -2.03607177734375, -4.397705078125, -1.509735107421875, 1.343597412109375, 0.0, 1.308807373046875, -0.09877777099609375, -0.5947265625, 2.1239471435546875, 3.488800048828125, 2.06890869140625, 0.1688232421875, 0.7030963897705078, -2.373382568359375, 0.441375732421875, 1.098602294921875, 3.0888671875, 2.7246856689453125, 0.678314208984375, -1.563262939453125, 0.02880859375, 0.0247802734375, 0.5672550201416016, 0.5419769287109375, 0.011951446533203125, 3.904754638671875, 0.53924560546875, -1.718292236328125, 1.037384033203125, -2.7872314453125, 2.266510009765625, 2.3012542724609375, 0.7841644287109375, -2.1140289306640625, 1.33709716796875, -0.35115814208984375, -0.0068359375, 1.219696044921875, -0.44774627685546875, 0.45001220703125, 0.0563812255859375, 0.5643787384033203, 1.536102294921875, 0.21887969970703125, -0.04877662658691406, -0.7967453002929688, 0.0001220703125, 1.2000885009765625, 0.643310546875, 0.0, 0.9788055419921875, 4.109771728515625, 1.5950927734375, 0.928802490234375, -0.6322860717773438, -0.29615020751953125, -1.544342041015625, 1.438201904296875, -0.3487701416015625, 0.08291053771972656, -0.236785888671875, 0.24217987060546875, 1.1656875610351562, 0.9572601318359375, -0.13641357421875, -0.4408836364746094, 1.308807373046875, 1.0357818603515625, -1.4058837890625, 2.007537841796875, 0.759490966796875, 1.21337890625, 0.8752670288085938, -1.313720703125, 1.1660537719726562, -0.0770111083984375, 0.176727294921875, 0.537750244140625, -1.35693359375, 0.6058425903320312, -0.795379638671875, -0.023040771484375, -0.20365524291992188, -0.2783203125, 0.56817626953125, 3.46588134765625, 0.2855033874511719, 1.3634185791015625, 1.57684326171875, -1.3181915283203125, 0.7690582275390625, -0.6223220825195312, 0.679931640625, -0.7505950927734375, 0.865203857421875, 1.0954437255859375, 0.8887939453125, 0.09893798828125, -3.491668701171875, 1.53582763671875, 0.7902450561523438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000047.npy"}
{"epoch": 0.09842931937172775, "step": 48, "batch_size": 128, "mean": 0.5563240647315979, "std": 1.4138561487197876, "min": -2.77740478515625, "p10": -0.8520858764648437, "median": 0.342498779296875, "p90": 2.4384857177734376, "max": 7.3564453125, "pos_frac": 0.609375, "sample": [-0.6560211181640625, 1.8959503173828125, 0.1103668212890625, 0.6158676147460938, 1.41180419921875, 1.17010498046875, 0.06146240234375, 2.2750244140625, 1.191986083984375, -0.12152862548828125, -0.5291976928710938, -0.0653076171875, -0.7179489135742188, -2.7090301513671875, -0.8181838989257812, 2.484130859375, -0.80438232421875, -0.9078521728515625, -1.620941162109375, 0.39051055908203125, 3.206695556640625, 0.6394271850585938, -0.46044921875, -0.0440673828125, 1.471343994140625, -0.1480712890625, -1.6688995361328125, 0.9124221801757812, 0.3546142578125, 1.0836639404296875, 0.5686302185058594, 1.810028076171875, 2.4058837890625, 0.89398193359375, 4.20660400390625, 0.7146759033203125, 0.12234687805175781, 3.11468505859375, 1.939605712890625, 1.794952392578125, -0.536407470703125, 2.526885986328125, 1.16424560546875, -0.5045623779296875, 0.6615371704101562, -0.82818603515625, -2.77740478515625, -0.7489013671875, 0.01220703125, 0.9196243286132812, 4.17193603515625, 1.1697731018066406, 0.08099365234375, -0.4033966064453125, -1.370208740234375, 1.23590087890625, 0.8950271606445312, 0.02326202392578125, -0.28474998474121094, 0.88397216796875, -0.598968505859375, 0.833587646484375, -0.00830078125, 0.3899993896484375, -0.340362548828125, -0.3939208984375, -0.778533935546875, -0.3138427734375, 2.2785110473632812, 0.6532440185546875, -1.494384765625, 1.6812744140625, 0.0, 0.2569427490234375, 0.241973876953125, -0.07200241088867188, 1.829833984375, 0.3838653564453125, 0.4333038330078125, 2.5, 0.917755126953125, 1.4886932373046875, -0.270355224609375, 2.43206787109375, -1.55633544921875, 2.453460693359375, 1.63787841796875, 1.09381103515625, -0.1319904327392578, 0.3719329833984375, -0.3804779052734375, 1.0019073486328125, -0.9693603515625, -1.0896682739257812, 0.753143310546875, 2.548095703125, 0.33038330078125, 1.192626953125, -0.993499755859375, 1.7148208618164062, -0.8021697998046875, 1.280914306640625, 0.0, 0.732666015625, 1.641693115234375, 0.11029052734375, 7.3564453125, -0.19031143188476562, -0.58563232421875, 1.2356338500976562, -0.212982177734375, 0.709136962890625, -0.667236328125, 0.03515625, -0.4949455261230469, 0.2936553955078125, -0.2725067138671875, -0.46141624450683594, 0.00494384765625, 0.16229248046875, -0.426177978515625, 2.811767578125, -1.0965576171875, 1.493377685546875, -0.9439697265625, 1.697052001953125, 3.1300048828125, 2.774810791015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000048.npy"}
{"epoch": 0.10052356020942409, "step": 49, "batch_size": 128, "mean": 0.6000461578369141, "std": 1.6870638132095337, "min": -4.56890869140625, "p10": -1.3155715942382813, "median": 0.4794502258300781, "p90": 2.8497024536132813, "max": 5.738800048828125, "pos_frac": 0.65625, "sample": [0.72235107421875, 1.5647430419921875, 0.7084541320800781, 0.47765350341796875, -1.32049560546875, -0.9571380615234375, -0.98895263671875, 4.96173095703125, 1.99346923828125, 0.7392482757568359, 0.45583343505859375, 1.1602783203125, -1.57470703125, 0.70654296875, 0.196380615234375, 4.447540283203125, 1.7194976806640625, 0.8190708160400391, 0.043613433837890625, 2.400787353515625, -1.602081298828125, -0.5533561706542969, 1.680908203125, 0.654327392578125, 1.473907470703125, -1.3134613037109375, -1.196380615234375, 5.738800048828125, 0.5981369018554688, 2.55621337890625, -0.0995941162109375, 2.203399658203125, 2.4201202392578125, 0.5854644775390625, 3.5992431640625, 1.0133838653564453, 0.164306640625, 0.11358642578125, 0.11592864990234375, -1.102020263671875, 0.10156631469726562, 0.9017333984375, -1.53082275390625, 0.689056396484375, 0.6791305541992188, 1.207305908203125, 1.4076385498046875, 2.2744598388671875, 3.218597412109375, -1.191070556640625, -0.58642578125, -1.711029052734375, -0.6044235229492188, 1.876983642578125, 0.8077201843261719, 3.55792236328125, 0.4812469482421875, 0.7965240478515625, -1.0845794677734375, 1.122039794921875, -0.05910682678222656, 3.848846435546875, -1.4197998046875, 3.08984375, 0.5517730712890625, 0.226959228515625, 0.86822509765625, -2.252971649169922, 2.849395751953125, 1.18524169921875, 1.5347900390625, 2.2921905517578125, 0.9883575439453125, -0.5224990844726562, -0.1573314666748047, 2.494384765625, -3.374755859375, -0.32061767578125, 2.1425933837890625, -0.234710693359375, 0.14436912536621094, 2.545654296875, 0.223785400390625, -1.1924362182617188, 0.44028472900390625, -0.0382080078125, -3.12274169921875, 2.024566650390625, 4.31561279296875, 2.464263916015625, 0.99810791015625, -0.677276611328125, 0.4670562744140625, 0.96728515625, -0.20375823974609375, 0.160308837890625, -0.894561767578125, 0.233245849609375, 0.6253166198730469, -1.8685302734375, -1.60546875, -0.84710693359375, -2.299163818359375, 1.3737640380859375, 3.68585205078125, -1.1341400146484375, -0.61041259765625, 3.684814453125, 1.6383209228515625, 0.43088340759277344, 3.1697540283203125, -0.16419219970703125, 0.916107177734375, -0.3041839599609375, -0.0462646484375, 0.40633392333984375, 0.16223907470703125, 2.8504180908203125, 1.376068115234375, 0.7707595825195312, -4.56890869140625, 0.07147979736328125, -0.49664306640625, 0.754180908203125, 0.2899169921875, -0.730560302734375, -0.441070556640625, -0.6103286743164062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000049.npy"}
{"epoch": 0.10261780104712041, "step": 50, "batch_size": 128, "mean": 0.5013490319252014, "std": 1.4507752656936646, "min": -5.05584716796875, "p10": -1.1747146606445313, "median": 0.5333366394042969, "p90": 2.1824203491210934, "max": 5.162353515625, "pos_frac": 0.6796875, "sample": [0.5301361083984375, -0.553955078125, 1.675537109375, -0.004978179931640625, 0.235107421875, -0.28685760498046875, 0.769775390625, -0.7650718688964844, -0.6467437744140625, 4.80596923828125, 0.81854248046875, -0.2485332489013672, 1.05145263671875, 1.9407958984375, 2.1031494140625, 1.8364028930664062, 1.6620635986328125, -2.2335205078125, 1.3034210205078125, 0.16983795166015625, -1.19390869140625, -0.6165771484375, -2.3855133056640625, 0.95458984375, 0.523406982421875, 3.380096435546875, 1.8680038452148438, 1.2216567993164062, 0.93756103515625, 0.900054931640625, -1.8109130859375, 3.28369140625, 0.9163970947265625, -0.717620849609375, 0.36045074462890625, -1.1963348388671875, -0.5842742919921875, 0.07379150390625, 0.552215576171875, -0.389739990234375, -0.5223236083984375, -0.319976806640625, 1.7259368896484375, 1.26593017578125, -0.2320098876953125, -0.4019775390625, 0.1248779296875, 1.601348876953125, -0.07977104187011719, 0.162322998046875, 1.43756103515625, 0.7142791748046875, 0.5365371704101562, -0.08945846557617188, 0.7853775024414062, 2.1421051025390625, 2.45977783203125, -0.5754852294921875, 1.785858154296875, 2.2764892578125, 1.2476806640625, 1.337310791015625, -5.05584716796875, 0.08339309692382812, -2.676544189453125, -1.17230224609375, 0.050811767578125, 0.6786537170410156, 0.979888916015625, -0.45850372314453125, 0.4447021484375, -0.3962745666503906, -0.5207138061523438, 0.550079345703125, 0.7109527587890625, -1.3895263671875, 1.399078369140625, 0.561676025390625, 1.412322998046875, -0.335601806640625, -0.74884033203125, 0.4239501953125, -2.562774658203125, -1.1803436279296875, 0.4964637756347656, 3.04168701171875, -0.806182861328125, -2.53271484375, 1.7483596801757812, -3.157318115234375, 0.11883544921875, 0.1539306640625, 3.29754638671875, 0.9265594482421875, 1.0673065185546875, 0.6524658203125, 1.1507644653320312, 0.563720703125, 1.4708251953125, 0.4793243408203125, 1.02618408203125, 2.77838134765625, 0.8403472900390625, 1.0875701904296875, 0.687774658203125, 0.1208953857421875, 0.124053955078125, 1.363983154296875, 0.38950347900390625, -1.665374755859375, 1.080078125, 2.4620361328125, 5.162353515625, 0.172027587890625, 0.714447021484375, -0.15228271484375, 2.47613525390625, 1.428802490234375, -0.189208984375, 0.6738300323486328, 2.35174560546875, 0.2084178924560547, 1.25689697265625, -0.2596435546875, 0.490875244140625, 0.258026123046875, -0.285980224609375, 2.479034423828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000050.npy"}
{"epoch": 0.10471204188481675, "step": 51, "batch_size": 128, "mean": 0.6440699100494385, "std": 1.9014743566513062, "min": -6.5673828125, "p10": -1.2640174865722655, "median": 0.4887199401855469, "p90": 3.066781616210937, "max": 5.565673828125, "pos_frac": 0.6171875, "sample": [1.6815185546875, -0.207916259765625, 1.3885078430175781, 0.5667953491210938, 1.3033199310302734, -2.298095703125, 1.8614501953125, 4.8736572265625, -0.11614990234375, 0.3631591796875, 1.6192626953125, 1.741973876953125, 1.18292236328125, 3.02996826171875, -1.2286758422851562, -0.90704345703125, -0.1163330078125, -1.07830810546875, 1.2921295166015625, 3.0100860595703125, -6.5673828125, 0.1108856201171875, 0.9837646484375, 0.2634735107421875, 1.019989013671875, 5.565673828125, -0.5657234191894531, -1.03448486328125, 3.82135009765625, 0.07801055908203125, -1.574859619140625, 3.3594970703125, 4.79052734375, 1.3111572265625, 1.0503158569335938, 1.5462646484375, 3.637054443359375, 1.4630584716796875, -0.53936767578125, -0.020538330078125, 0.640289306640625, 1.974029541015625, -1.4531402587890625, 3.27362060546875, -0.09780120849609375, 3.231414794921875, 0.6467742919921875, 0.3285789489746094, 0.3184814453125, -4.08319091796875, 4.6552734375, 0.3840789794921875, -4.226318359375, 1.137237548828125, -0.120452880859375, 2.7286376953125, 0.122833251953125, 0.40010833740234375, -0.31646728515625, 3.011016845703125, -0.31781005859375, 1.36328125, -0.769775390625, 1.2637405395507812, -2.789215087890625, -1.023345947265625, -0.1420440673828125, -0.661865234375, 1.31524658203125, 2.291961669921875, 0.3531951904296875, 1.1839599609375, -1.3464813232421875, 2.701446533203125, -1.574371337890625, 0.0, 3.701904296875, -0.956207275390625, 1.7432403564453125, 0.9969482421875, 0.632354736328125, -0.9931373596191406, 0.6911468505859375, 0.5812530517578125, -2.91534423828125, 0.041717529296875, -0.650177001953125, -0.31156158447265625, -0.8648681640625, 2.018280029296875, -0.06531906127929688, -0.432342529296875, 0.319000244140625, -0.41873931884765625, -0.34195709228515625, 0.7033233642578125, -0.444793701171875, 1.4416656494140625, 2.9653778076171875, -1.707489013671875, -2.479705810546875, -1.172119140625, -0.4978485107421875, 3.152679443359375, 5.10205078125, 4.5318603515625, 1.6436080932617188, 2.59796142578125, 2.159332275390625, 1.815673828125, 0.69525146484375, 0.2632408142089844, -0.177215576171875, 1.7947807312011719, 2.15087890625, 3.01611328125, -0.9588470458984375, -2.43048095703125, 0.41064453125, 0.762420654296875, 0.6047210693359375, -0.20751953125, 0.0995635986328125, -0.1460723876953125, 1.0950927734375, 0.98455810546875, -0.06719970703125, 0.8994293212890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000051.npy"}
{"epoch": 0.1068062827225131, "step": 52, "batch_size": 128, "mean": 0.8648929595947266, "std": 1.9779517650604248, "min": -3.882110595703125, "p10": -1.5651718139648436, "median": 0.6451530456542969, "p90": 3.6711303710937484, "max": 7.429718017578125, "pos_frac": 0.6640625, "sample": [1.042449951171875, 1.67657470703125, 0.58953857421875, -2.6612548828125, -0.18801116943359375, -1.9698486328125, 3.358917236328125, 1.1217575073242188, -0.11496734619140625, 4.08453369140625, -0.3797607421875, -0.93365478515625, 0.6724700927734375, -2.59918212890625, 2.17498779296875, 5.45660400390625, 0.579010009765625, 1.143798828125, -0.323638916015625, -1.5333023071289062, 1.349090576171875, 0.8485641479492188, 1.0183639526367188, 0.6522140502929688, 0.638092041015625, -0.5213088989257812, 0.290863037109375, 3.21380615234375, 1.4462890625, -1.5809478759765625, -1.55841064453125, -0.8131027221679688, -0.08575820922851562, 0.8636703491210938, -1.408447265625, 0.7121696472167969, -0.5597305297851562, -3.11962890625, 0.9412269592285156, -1.6883544921875, 4.49456787109375, 0.9952926635742188, -1.6457977294921875, 2.261322021484375, 2.2440338134765625, 1.8187179565429688, 1.24951171875, 4.3128662109375, 0.58770751953125, 3.44366455078125, 1.370391845703125, 2.4383087158203125, 1.4161376953125, 1.1635894775390625, -0.1616363525390625, 1.951416015625, 0.32064247131347656, -1.173553466796875, 0.251434326171875, -0.1976318359375, 0.5103530883789062, -0.447845458984375, 2.59613037109375, -0.069580078125, 4.44549560546875, 1.577972412109375, -2.3284912109375, 0.4766197204589844, 1.5658798217773438, -0.09050655364990234, -0.81939697265625, 1.2376251220703125, 2.6246490478515625, 1.4344329833984375, 1.44970703125, -3.091217041015625, 0.8676624298095703, 0.631103515625, 1.3974151611328125, 1.46600341796875, 4.559814453125, 5.26043701171875, 0.8964080810546875, 0.9033050537109375, 3.2293701171875, -2.47894287109375, -0.3205451965332031, 4.09234619140625, 5.1068115234375, 1.34600830078125, 2.0010299682617188, -0.8524169921875, -0.27233123779296875, 4.8334197998046875, -1.61346435546875, 0.19774246215820312, 0.17383575439453125, 5.62603759765625, 0.5437393188476562, -0.5896682739257812, 0.14849853515625, 1.809234619140625, 1.2181243896484375, -0.2532958984375, 5.4703369140625, 1.5892868041992188, -1.9586181640625, 0.4332122802734375, 7.429718017578125, 1.773651123046875, 0.17974853515625, 1.90826416015625, -0.11174774169921875, -0.5729904174804688, 1.8720550537109375, -0.24525070190429688, 0.396331787109375, 0.3186359405517578, 0.202606201171875, 1.3380126953125, 0.557769775390625, 1.0996551513671875, -3.882110595703125, -0.7804183959960938, 0.4066314697265625, -0.05144500732421875, 3.49395751953125, -0.13716888427734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000052.npy"}
{"epoch": 0.10890052356020942, "step": 53, "batch_size": 128, "mean": 0.7730849981307983, "std": 2.024153232574463, "min": -5.56231689453125, "p10": -1.5754791259765624, "median": 0.7724838256835938, "p90": 3.4092468261718745, "max": 6.42572021484375, "pos_frac": 0.640625, "sample": [0.820770263671875, -1.3819732666015625, 2.1492462158203125, -1.4347763061523438, 0.8828125, 0.578521728515625, 4.72308349609375, -2.283843994140625, -0.79754638671875, 0.5805587768554688, 2.063232421875, -0.097808837890625, 0.231597900390625, -4.11181640625, -1.6825408935546875, 2.115692138671875, 1.492706298828125, -0.6644287109375, 1.388153076171875, -0.188385009765625, -0.30853271484375, -0.5709075927734375, 0.848907470703125, 1.23828125, 2.18878173828125, 6.20068359375, 2.8139114379882812, 4.61285400390625, 1.20294189453125, 4.6326904296875, 1.0792236328125, 4.6339111328125, -0.24122238159179688, 0.851593017578125, 0.8313846588134766, -0.285064697265625, 1.689666748046875, -3.719268798828125, -0.31247520446777344, 1.7261428833007812, 1.979888916015625, 0.3615608215332031, 0.217987060546875, -1.7748565673828125, -0.060901641845703125, -3.6824951171875, -1.595367431640625, -0.20114898681640625, 3.33624267578125, -1.68243408203125, 0.200897216796875, 2.8099365234375, 2.068634033203125, -1.56695556640625, -0.3244895935058594, -1.22515869140625, 2.3668212890625, 1.0127677917480469, -0.5781173706054688, 3.57958984375, 3.663787841796875, 0.5723800659179688, -0.6578598022460938, 1.879730224609375, 1.618927001953125, -0.40007781982421875, -0.425537109375, -0.40308380126953125, 0.33879852294921875, 0.5201339721679688, 0.411895751953125, 1.323455810546875, 5.116180419921875, 1.2398223876953125, 5.456512451171875, 2.6473388671875, -0.44525146484375, -0.13336181640625, 1.84576416015625, 0.91571044921875, 1.527740478515625, 0.6770439147949219, 0.73199462890625, -0.52740478515625, -0.3292694091796875, 4.90155029296875, -4.151458740234375, 2.193084716796875, 0.39589881896972656, -0.312744140625, 1.61846923828125, -1.7161407470703125, 2.84954833984375, -1.791778564453125, 1.2666015625, 6.42572021484375, 0.06630325317382812, -2.2279052734375, 0.5652389526367188, -0.781341552734375, -0.5001487731933594, 1.8367767333984375, -0.9263687133789062, -1.250457763671875, 0.11150360107421875, -0.8867645263671875, 1.3135986328125, 2.2765960693359375, 0.8129730224609375, -5.56231689453125, 1.199920654296875, 0.34759521484375, 1.86981201171875, 3.08154296875, 1.082244873046875, -0.1102294921875, 3.65673828125, 0.0, 2.13238525390625, 0.668548583984375, 1.130889892578125, 1.4261627197265625, 1.2753410339355469, 1.033538818359375, 2.0228271484375, 0.8821563720703125, 1.049713134765625, 3.77471923828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000053.npy"}
{"epoch": 0.11099476439790576, "step": 54, "batch_size": 128, "mean": 0.6549609899520874, "std": 2.116523504257202, "min": -5.254669189453125, "p10": -1.891604614257812, "median": 0.41012001037597656, "p90": 3.2568939208984373, "max": 6.93157958984375, "pos_frac": 0.640625, "sample": [2.620361328125, 4.1806640625, 2.50323486328125, 2.1390380859375, 1.042510986328125, 0.3946113586425781, 0.425628662109375, 2.827392578125, -0.2842578887939453, -2.4217529296875, 0.6827392578125, -1.4851531982421875, -0.6653213500976562, -1.1204681396484375, 4.754150390625, -0.33863067626953125, 5.6190185546875, 0.46148681640625, 0.352081298828125, 0.84033203125, 0.1529388427734375, 4.76544189453125, 0.7220458984375, 0.561187744140625, -1.540008544921875, -0.03209686279296875, 2.3697509765625, -0.008575439453125, 1.47222900390625, 0.3817863464355469, 0.3911285400390625, 1.307647705078125, -3.122467041015625, 0.1590576171875, 0.954254150390625, 1.97235107421875, 6.93157958984375, -3.82501220703125, 2.9798583984375, -0.10039901733398438, 0.7922821044921875, 0.5913486480712891, 3.27618408203125, -1.18231201171875, 3.770294189453125, -0.1563262939453125, -3.360382080078125, -0.797119140625, -1.0744972229003906, -0.97357177734375, 2.763031005859375, 4.057403564453125, 0.9418716430664062, 0.23431396484375, 0.010410308837890625, 0.3876495361328125, -0.38677978515625, 1.6627655029296875, 0.437957763671875, 0.20101165771484375, 2.477031707763672, 1.0109710693359375, -1.73736572265625, -0.046783447265625, 2.053436279296875, 1.474853515625, -0.3499755859375, 1.400604248046875, 1.8247222900390625, 3.248626708984375, -0.38817596435546875, 2.1518402099609375, 0.46820068359375, 2.5078125, -0.49908447265625, 0.9050064086914062, 2.094573974609375, -3.140289306640625, 0.1305084228515625, 0.21451950073242188, -0.08500289916992188, -0.9210433959960938, 0.5893096923828125, 0.35919189453125, -0.1978759765625, 1.81787109375, -1.058868408203125, -0.508209228515625, 1.6610946655273438, -0.6072235107421875, 0.019989013671875, 4.265380859375, -2.739013671875, 5.492889404296875, 1.5193939208984375, 0.08807373046875, -2.49615478515625, 5.6380615234375, 4.9541015625, 2.8485107421875, 0.664581298828125, 3.040557861328125, 3.9517669677734375, -0.55712890625, 3.083526611328125, -0.4206504821777344, -3.96868896484375, 0.46441650390625, -5.254669189453125, 2.951446533203125, -0.73150634765625, -2.7372894287109375, -0.48883056640625, -0.652313232421875, 2.525177001953125, -1.084014892578125, -2.6517257690429688, -0.4798583984375, 0.324066162109375, 0.119659423828125, 0.151702880859375, 0.7994384765625, 0.9407844543457031, -3.0281982421875, -2.251495361328125, 0.560882568359375, 1.1576957702636719, 0.748260498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000054.npy"}
{"epoch": 0.1130890052356021, "step": 55, "batch_size": 128, "mean": 0.5787118077278137, "std": 1.8997976779937744, "min": -5.02777099609375, "p10": -1.2521255493164063, "median": 0.4255943298339844, "p90": 2.8465972900390626, "max": 7.91748046875, "pos_frac": 0.6015625, "sample": [-1.23870849609375, 0.855499267578125, 0.44580078125, 0.435546875, -0.572418212890625, -0.2579345703125, 0.6037483215332031, 0.006500244140625, 1.78863525390625, -0.577484130859375, 1.7831878662109375, 3.6997222900390625, 2.9038238525390625, 2.04315185546875, 0.310333251953125, 0.27912330627441406, 1.0805511474609375, 0.49627685546875, 0.68719482421875, 1.7139472961425781, -0.26068878173828125, 0.7818603515625, -0.19439697265625, 2.07421875, 2.7181549072265625, -0.5799598693847656, 0.932647705078125, 2.19586181640625, 0.219818115234375, -0.187713623046875, -1.9595947265625, 0.14774322509765625, -1.2502899169921875, 2.58984375, 0.078826904296875, 0.6409111022949219, -0.18121337890625, 0.441925048828125, -0.7417678833007812, 4.91473388671875, -2.571624755859375, -0.14861297607421875, -0.3416748046875, 1.642974853515625, 0.69873046875, 0.6509857177734375, 0.6012420654296875, -0.93585205078125, 1.0274658203125, 4.614044189453125, 0.22039794921875, -1.773590087890625, -2.1893310546875, -0.8420486450195312, 1.12017822265625, -1.7395248413085938, -4.090240478515625, 4.4068603515625, -1.0020294189453125, 1.6282958984375, -1.25640869140625, 1.0672607421875, 5.675689697265625, 0.19830703735351562, 1.094573974609375, -0.36370849609375, -1.6213607788085938, 4.577201843261719, 0.519744873046875, -0.6760902404785156, 0.033599853515625, 0.5749282836914062, -0.6819000244140625, -1.2021484375, -0.8708152770996094, -0.17075538635253906, 0.117431640625, 2.443084716796875, 0.4781608581542969, 1.2224578857421875, -0.39565277099609375, 2.72918701171875, -0.9493408203125, 0.4509735107421875, -0.101165771484375, 0.16042327880859375, -0.5173816680908203, 2.83367919921875, 2.876739501953125, -0.6391754150390625, -0.6105194091796875, 2.7895660400390625, 1.76873779296875, -0.2158203125, 7.91748046875, -0.2793426513671875, 3.197601318359375, -0.99298095703125, -0.11858367919921875, 3.4027099609375, -0.924102783203125, 1.388153076171875, 0.4479484558105469, 0.88116455078125, 0.6896896362304688, -3.794097900390625, 1.94354248046875, 1.4166717529296875, 2.2675323486328125, -1.0622711181640625, -5.02777099609375, -0.18505859375, 2.16680908203125, 0.05144500732421875, -0.4687652587890625, -0.8047256469726562, 1.7394256591796875, -2.0137939453125, 1.4932899475097656, -0.933990478515625, 1.80029296875, 0.41564178466796875, -1.844970703125, 2.067962646484375, -1.7003173828125, 5.1361541748046875, 3.7161102294921875, 0.9026870727539062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000055.npy"}
{"epoch": 0.11518324607329843, "step": 56, "batch_size": 128, "mean": 1.0370157957077026, "std": 2.3817548751831055, "min": -5.746612548828125, "p10": -1.348597717285156, "median": 0.7320709228515625, "p90": 3.923237609863281, "max": 8.005889892578125, "pos_frac": 0.703125, "sample": [0.63134765625, -2.8951568603515625, 3.41546630859375, 5.4833984375, 0.51861572265625, 1.6969757080078125, 1.02069091796875, -0.64898681640625, -0.31674957275390625, 7.557373046875, 1.236419677734375, 1.42755126953125, -0.2210540771484375, -1.2971038818359375, 0.598876953125, -0.149658203125, -0.7399978637695312, -0.3943634033203125, 3.8619384765625, -1.46875, 1.6663818359375, 0.6624755859375, 2.650787353515625, -0.288421630859375, -0.49135589599609375, 4.282989501953125, 0.38043212890625, 4.9990234375, 3.71929931640625, 0.64080810546875, 2.5865478515625, 3.06640625, 1.072479248046875, -0.540191650390625, 5.420074462890625, 2.2964324951171875, -2.6763763427734375, 1.1444931030273438, 1.2717819213867188, 0.091033935546875, 6.1811981201171875, 0.102142333984375, 0.24344635009765625, -0.2450389862060547, 0.3456401824951172, 6.80419921875, 2.7181396484375, 0.052616119384765625, -1.0272216796875, -0.8324661254882812, -1.550445556640625, -0.9311447143554688, 0.83172607421875, -5.34814453125, 0.35626220703125, -5.746612548828125, 0.1272602081298828, -0.5380096435546875, 1.92474365234375, 0.2551422119140625, 2.5118637084960938, 6.09783935546875, 2.79705810546875, 3.083251953125, -2.12353515625, -1.288909912109375, 1.81976318359375, 2.1142578125, 3.88262939453125, 0.9073410034179688, 0.12725067138671875, -0.3000640869140625, -4.627227783203125, 0.6385746002197266, 0.17237472534179688, 0.801666259765625, -2.544036865234375, 3.70831298828125, -0.1903076171875, 1.890167236328125, -4.74176025390625, 1.635284423828125, 2.4700164794921875, 0.498870849609375, 1.2465667724609375, 8.005889892578125, 1.52099609375, 2.363616943359375, -2.473358154296875, 2.1247711181640625, -0.0598907470703125, 6.1605224609375, 3.18121337890625, 1.3960494995117188, -1.226806640625, 4.14593505859375, 1.7250213623046875, 0.3603057861328125, 0.08692169189453125, 0.8348007202148438, 1.314697265625, -0.1759490966796875, 1.1521072387695312, -1.7158966064453125, -2.99969482421875, 5.62530517578125, 0.32550811767578125, 2.27410888671875, -0.2563934326171875, 1.525665283203125, 1.926116943359375, 0.53399658203125, 2.076812744140625, -0.2940826416015625, 0.060230255126953125, 0.649139404296875, -0.26971435546875, 0.18151473999023438, 2.0704498291015625, 1.78973388671875, 2.04547119140625, -0.7787094116210938, 1.9839019775390625, 2.2906494140625, 0.152191162109375, 4.0179901123046875, 2.5741119384765625, 0.9301605224609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000056.npy"}
{"epoch": 0.11727748691099477, "step": 57, "batch_size": 128, "mean": 1.4406650066375732, "std": 2.396458148956299, "min": -3.175933837890625, "p10": -1.24647216796875, "median": 0.9294233322143555, "p90": 4.1222900390624995, "max": 15.1700439453125, "pos_frac": 0.7265625, "sample": [-1.6061782836914062, 3.83294677734375, -1.791595458984375, 2.056610107421875, 0.03800201416015625, -2.0008544921875, 4.37408447265625, 0.393585205078125, 0.80950927734375, 1.15380859375, 0.3616485595703125, -1.88031005859375, 2.859893798828125, 0.5164794921875, 3.1988677978515625, 0.9242458343505859, 3.0993804931640625, 2.829864501953125, 2.60858154296875, 1.388580322265625, 0.9522361755371094, 4.2594146728515625, 0.37727928161621094, -0.0955352783203125, 0.412078857421875, 0.934600830078125, -1.84210205078125, -1.48779296875, -0.2506256103515625, -0.7369613647460938, 2.984222412109375, 3.035980224609375, 1.5618133544921875, 0.6477813720703125, 8.274169921875, 15.1700439453125, 0.71868896484375, -0.83819580078125, -1.080657958984375, 2.975006103515625, 3.5169830322265625, 0.8507423400878906, -1.576995849609375, 6.45904541015625, 1.435821533203125, 1.9417648315429688, -0.21044921875, -3.175933837890625, -0.76788330078125, 1.8382568359375, 0.575531005859375, 0.312835693359375, 1.7398681640625, 6.207977294921875, 0.6729278564453125, 1.07073974609375, 0.9215850830078125, 1.0507965087890625, 3.91802978515625, 2.64154052734375, -0.31329345703125, 0.3526611328125, 0.5572624206542969, 3.26495361328125, 1.54864501953125, 5.3294677734375, 5.47607421875, 1.23443603515625, -1.833770751953125, 0.98480224609375, 2.727996826171875, 0.66143798828125, -0.2549896240234375, -0.5154705047607422, 3.55633544921875, 2.669342041015625, -0.51483154296875, 3.9610595703125, 0.59332275390625, 1.8785400390625, -0.04290771484375, 0.48102569580078125, 5.637542724609375, 5.756072998046875, -1.323455810546875, -1.2314300537109375, -1.2815704345703125, 4.31292724609375, -1.5606689453125, -1.030120849609375, 1.6951828002929688, 3.58837890625, 0.2398357391357422, 0.48430633544921875, -0.27691650390625, -0.11979293823242188, 2.43408203125, 0.5777664184570312, 0.408843994140625, -0.021881103515625, -0.4716796875, 1.0218963623046875, 4.4970703125, 2.849822998046875, 0.2760200500488281, 3.942840576171875, 0.8176689147949219, 3.587371826171875, 0.1456451416015625, 0.2026214599609375, 1.1572265625, 1.023529052734375, 4.0635223388671875, -0.1945953369140625, 2.9803085327148438, 2.72930908203125, 2.532379150390625, -0.3441295623779297, 0.7870769500732422, -0.17050552368164062, 3.11505126953125, -2.453094482421875, 3.096923828125, 1.60235595703125, 2.1768875122070312, -0.00933837890625, 2.51910400390625, 4.270843505859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000057.npy"}
{"epoch": 0.1193717277486911, "step": 58, "batch_size": 128, "mean": 1.028170108795166, "std": 2.1674654483795166, "min": -9.431182861328125, "p10": -1.299884796142578, "median": 0.8276824951171875, "p90": 3.393213653564452, "max": 7.33880615234375, "pos_frac": 0.734375, "sample": [0.5294189453125, 4.875213623046875, 1.5732421875, 2.264923095703125, 2.3032302856445312, -0.3753776550292969, 2.330608367919922, -0.997314453125, 0.50634765625, 3.272705078125, 1.8291397094726562, 3.872528076171875, 1.6430282592773438, 5.4327392578125, 2.301910400390625, -1.4197006225585938, 3.6744003295898438, 3.21246337890625, 2.9810867309570312, -2.121826171875, 7.33880615234375, 2.98492431640625, 0.30828857421875, -1.009674072265625, -0.655853271484375, 0.9817962646484375, 2.5215301513671875, 2.23468017578125, 0.801788330078125, -1.57232666015625, -0.021663665771484375, 0.7552947998046875, 0.00701904296875, 0.2397747039794922, 2.2613525390625, 0.3686676025390625, 1.448944091796875, 1.8660507202148438, -0.2988929748535156, 0.1903076171875, -3.7247314453125, 1.131103515625, 0.8582916259765625, 0.570770263671875, 1.950927734375, -0.428955078125, 0.5324268341064453, 6.377593994140625, -0.03433799743652344, -9.431182861328125, 0.1527557373046875, 5.845947265625, -0.983367919921875, -1.6643104553222656, -0.0711669921875, 0.4664764404296875, 0.08217620849609375, 1.382822036743164, 0.151824951171875, 3.959228515625, 1.8268890380859375, 0.8782958984375, 3.057098388671875, -1.24853515625, 2.72247314453125, 0.8604736328125, 0.30511474609375, 3.04876708984375, 0.85357666015625, 1.919036865234375, 0.567535400390625, -1.751556396484375, -1.1075897216796875, 2.308502197265625, 1.4154739379882812, 0.423095703125, 6.224456787109375, 0.3205833435058594, 0.504791259765625, 0.471893310546875, 1.0200424194335938, -0.69207763671875, 5.98626708984375, -3.363983154296875, 1.281707763671875, 0.9405441284179688, 0.10353851318359375, 0.46730804443359375, 0.6588592529296875, 1.4112548828125, 0.13496780395507812, 2.2942352294921875, 2.52978515625, 0.6921539306640625, 2.056884765625, -0.9013671875, 3.70123291015625, -1.53387451171875, 1.2393798828125, 1.369171142578125, 0.4858245849609375, -0.2304821014404297, -0.05588531494140625, -0.984283447265625, 1.966888427734375, 0.9146347045898438, 0.7788162231445312, 1.2130508422851562, 4.027587890625, 1.99658203125, -2.595367431640625, 3.20660400390625, -0.22916412353515625, 4.888671875, 0.7802658081054688, 0.3470458984375, -0.636260986328125, -0.155029296875, 2.3704071044921875, 3.11053466796875, 2.2668304443359375, 1.1646499633789062, -1.4647216796875, 1.550079345703125, -1.1822891235351562, -1.566680908203125, 2.748687744140625, -1.6715087890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000058.npy"}
{"epoch": 0.12146596858638743, "step": 59, "batch_size": 128, "mean": 0.9450608491897583, "std": 2.6759653091430664, "min": -5.7520751953125, "p10": -1.628729248046875, "median": 0.548797607421875, "p90": 4.657403564453123, "max": 8.7730712890625, "pos_frac": 0.6015625, "sample": [-0.334197998046875, 2.888946533203125, 0.4976806640625, -5.27838134765625, 1.7577285766601562, 1.0388336181640625, -0.6741943359375, 3.305908203125, 1.072052001953125, 1.05059814453125, -0.0216522216796875, 0.7595977783203125, 2.0567626953125, 2.948333740234375, -3.35430908203125, 0.39046478271484375, -0.24029541015625, -0.1687469482421875, -1.553497314453125, 1.063751220703125, -0.5935821533203125, 1.0146026611328125, 0.2479724884033203, -0.1959991455078125, 3.0730743408203125, 0.6823272705078125, 1.8336639404296875, 5.163818359375, -3.399749755859375, 1.3226318359375, 0.033203125, 1.4207725524902344, 0.4382171630859375, 2.3197784423828125, 0.0, 5.197998046875, -0.32731056213378906, 1.8599090576171875, -0.06429481506347656, -2.976531982421875, 0.4437446594238281, 3.0457763671875, 6.131439208984375, 8.16314697265625, -0.396392822265625, -5.7520751953125, 1.8449172973632812, 8.06964111328125, 2.436065673828125, 7.449951171875, 1.7689018249511719, 0.6754150390625, 5.8057861328125, 5.68902587890625, 0.33113861083984375, 1.0655517578125, -0.1954498291015625, 1.5700836181640625, -4.728363037109375, -1.654876708984375, -0.22137451171875, 2.347747802734375, 0.773895263671875, -0.7269287109375, 1.324859619140625, 4.20733642578125, 5.59674072265625, -0.5699462890625, 3.3547210693359375, 3.0687484741210938, -0.47905731201171875, 1.329498291015625, 0.2900390625, 7.63531494140625, -1.879974365234375, -4.12030029296875, 1.047616958618164, -3.94342041015625, -1.2437744140625, 0.06603050231933594, 3.84136962890625, -0.05596923828125, 2.4073028564453125, 1.2421875, 0.9045028686523438, -0.1115264892578125, 3.024444580078125, -1.1356964111328125, -2.1626739501953125, -0.16265869140625, 1.421630859375, 3.008026123046875, -0.09934234619140625, -0.3868122100830078, -1.4704742431640625, 5.4456634521484375, 1.54315185546875, -1.43255615234375, -0.6911468505859375, -1.065185546875, -0.1146392822265625, 0.1965179443359375, 0.5841293334960938, 1.7193145751953125, -0.15288162231445312, 0.280120849609375, -1.617523193359375, -4.4193115234375, -1.49652099609375, 1.03900146484375, 1.8280029296875, 2.85528564453125, 6.38336181640625, -0.149017333984375, -0.4912147521972656, 0.060699462890625, 8.7730712890625, 1.562896728515625, 1.889495849609375, -0.3332366943359375, -3.722076416015625, 0.5134658813476562, 4.44036865234375, 2.3966064453125, -0.09402847290039062, -0.0925140380859375, 1.3953857421875, -0.2082977294921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000059.npy"}
{"epoch": 0.12356020942408377, "step": 60, "batch_size": 128, "mean": 1.2792444229125977, "std": 2.3629910945892334, "min": -6.65374755859375, "p10": -0.9848648071289061, "median": 0.84490966796875, "p90": 4.4834350585937495, "max": 8.435455322265625, "pos_frac": 0.7265625, "sample": [-5.26422119140625, 0.593963623046875, 4.406505584716797, -0.3573760986328125, 2.632415771484375, 4.52752685546875, -0.39732933044433594, 2.3122634887695312, 2.275726318359375, 0.4080810546875, 0.32833099365234375, -0.6099109649658203, -0.13281631469726562, 2.6282501220703125, 0.39947509765625, 0.70703125, 0.80841064453125, -0.454437255859375, 1.1421966552734375, 4.5355224609375, 1.310638427734375, -2.077911376953125, 5.2205810546875, 0.0219268798828125, -1.4775066375732422, 0.13341522216796875, 1.837982177734375, 0.0, -0.3371257781982422, 4.58245849609375, -0.564178466796875, 5.26300048828125, 1.496551513671875, 1.577056884765625, 0.6620635986328125, 0.0804595947265625, 3.597381591796875, -3.292816162109375, 2.559539794921875, -0.20885848999023438, -0.26009368896484375, 2.1619110107421875, 0.54644775390625, 3.2602081298828125, 1.0071182250976562, -0.0067138671875, -1.267660140991211, -0.46795654296875, 0.808807373046875, 0.3981132507324219, 0.023284912109375, 0.5562744140625, 2.179656982421875, 5.98443603515625, 0.20343017578125, -0.626983642578125, 2.8173675537109375, -0.501953125, 0.881011962890625, 3.9903411865234375, 0.6924285888671875, 3.1116943359375, 3.0480499267578125, 0.06234550476074219, -6.65374755859375, 2.229644775390625, 0.2381591796875, 0.4289703369140625, -0.035400390625, -1.9229583740234375, -1.1079864501953125, 4.2393798828125, 3.5631790161132812, 5.382293701171875, 2.4586181640625, 0.318878173828125, 0.01712799072265625, 2.5262298583984375, 1.4688568115234375, 8.182258605957031, 2.713653564453125, 0.22467803955078125, -0.07842636108398438, 0.7982940673828125, 5.250640869140625, 0.9427490234375, 2.304901123046875, -2.690399169921875, 1.4266090393066406, 2.3558807373046875, 2.053192138671875, 0.09610366821289062, -4.2847900390625, 0.12245559692382812, 2.995361328125, 3.0596466064453125, 0.1880950927734375, 5.1044921875, -0.7178268432617188, 0.039886474609375, 1.817901611328125, 4.46453857421875, -0.0423736572265625, -1.494720458984375, 5.0716705322265625, 2.6002960205078125, -0.08919525146484375, 2.4447250366210938, 8.435455322265625, 0.0, 6.79974365234375, -1.256591796875, 1.8422698974609375, -0.2668304443359375, -0.932098388671875, 1.055511474609375, 1.0753173828125, 0.15422439575195312, 2.5002288818359375, -2.8170166015625, 2.08050537109375, 2.7437744140625, 1.3779830932617188, 4.410400390625, 2.2376708984375, 1.2432403564453125, 2.987640380859375, 2.58447265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000060.npy"}
{"epoch": 0.1256544502617801, "step": 61, "batch_size": 128, "mean": 1.212318778038025, "std": 2.6480553150177, "min": -4.749053955078125, "p10": -1.8548934936523438, "median": 0.9013452529907227, "p90": 4.690396118164062, "max": 9.442901611328125, "pos_frac": 0.6953125, "sample": [-2.104522705078125, -0.6483917236328125, 1.0472602844238281, -0.650634765625, 7.1103515625, 2.51177978515625, -0.7937164306640625, 0.04327392578125, -0.22046661376953125, 1.0097427368164062, 1.8752899169921875, -4.71881103515625, -2.2657470703125, -0.0970458984375, 0.90728759765625, 0.8954029083251953, -1.689910888671875, -1.18231201171875, -3.4508056640625, -1.2116317749023438, 0.5656051635742188, 0.9757843017578125, 7.145355224609375, 3.915557861328125, 5.05780029296875, 1.005126953125, 3.389312744140625, -0.289459228515625, 4.278045654296875, 2.7704315185546875, 0.37897491455078125, 0.5342445373535156, 0.26349830627441406, -0.615509033203125, 5.724700927734375, 1.08319091796875, 0.09421539306640625, 3.8371429443359375, 0.389251708984375, 0.7652168273925781, -0.4094085693359375, 0.68475341796875, 2.6013946533203125, 0.07822418212890625, 1.802215576171875, 0.0030059814453125, 4.714813232421875, 2.836212158203125, 1.876617431640625, 2.966094970703125, -0.24127197265625, 3.5875396728515625, 7.532073974609375, 4.38330078125, 4.2085723876953125, 3.31719970703125, 1.39251708984375, 1.0231990814208984, 6.00787353515625, 0.098236083984375, 0.27587890625, 2.5996780395507812, 1.30029296875, -0.8811264038085938, 0.6352081298828125, 4.1558837890625, 4.679931640625, 5.484222412109375, 3.356903076171875, 0.5775299072265625, 1.771728515625, 0.9978179931640625, 2.5531845092773438, -0.354095458984375, -0.7629776000976562, 0.66632080078125, 0.0064907073974609375, -0.4205169677734375, 1.1932525634765625, -1.889404296875, 7.349212646484375, 1.5544586181640625, 2.99871826171875, 3.184356689453125, 1.82147216796875, 1.6327667236328125, -0.0363922119140625, -3.0157470703125, -0.5581817626953125, 1.171112060546875, -0.09418487548828125, -4.1762542724609375, -3.1905517578125, 1.9476318359375, -0.00420379638671875, 0.826934814453125, 1.1070060729980469, 4.850860595703125, -0.8980865478515625, 9.1790771484375, 0.51177978515625, 9.442901611328125, 2.12274169921875, 0.0, 0.34173583984375, 1.0838394165039062, 3.490997314453125, 0.454010009765625, 1.7296066284179688, -4.749053955078125, 3.62872314453125, -2.432891845703125, -1.8401031494140625, 0.27764892578125, -3.177947998046875, 2.1162872314453125, -4.58087158203125, -0.581634521484375, 1.36517333984375, 0.840057373046875, -0.53155517578125, 0.07205963134765625, -0.7344436645507812, 2.0166854858398438, 2.0465621948242188, 5.057861328125, 1.7303924560546875, -2.2180099487304688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000061.npy"}
{"epoch": 0.12774869109947645, "step": 62, "batch_size": 128, "mean": 1.166268229484558, "std": 2.6713297367095947, "min": -5.30670166015625, "p10": -2.2150344848632812, "median": 0.8896369934082031, "p90": 4.640638732910156, "max": 8.567138671875, "pos_frac": 0.671875, "sample": [4.819915771484375, 0.783905029296875, 7.497772216796875, 2.445770263671875, 4.1094970703125, 0.5202255249023438, 2.081939697265625, 2.2039794921875, -2.88763427734375, -1.011138916015625, -1.157958984375, -1.6469497680664062, -0.6933059692382812, 2.609222412109375, 3.01678466796875, 0.7690200805664062, 1.986602783203125, 2.6941375732421875, -0.07447052001953125, -2.30267333984375, 2.4829254150390625, 1.2627105712890625, 2.5710220336914062, 0.8881454467773438, -3.69085693359375, 6.95294189453125, -4.152801513671875, 0.183502197265625, 0.6419677734375, 0.97344970703125, -0.952423095703125, 1.253448486328125, 1.5120391845703125, 1.8057403564453125, 0.12127876281738281, -0.03656768798828125, 2.31463623046875, -0.5057830810546875, 1.3695869445800781, 0.111907958984375, 1.35845947265625, 5.494140625, 2.7213363647460938, 0.08687591552734375, -2.1774749755859375, -0.57708740234375, 0.4794960021972656, 3.3367919921875, 3.27130126953125, 8.398651123046875, -5.30670166015625, 3.4443359375, -1.535491943359375, 2.531005859375, 4.75860595703125, -2.7256927490234375, 6.744659423828125, 0.0, 3.1119384765625, -0.00887298583984375, -2.580169677734375, -1.3794441223144531, 4.6062469482421875, -0.12569427490234375, 2.2982101440429688, -0.1191253662109375, 0.6653060913085938, -2.1717529296875, 0.118255615234375, 0.316650390625, 4.13623046875, 0.8501739501953125, 0.44676780700683594, -2.333965301513672, 1.8503570556640625, 4.533111572265625, 1.213531494140625, -2.4830474853515625, 0.8911285400390625, 2.1851577758789062, 0.1903533935546875, -1.1961212158203125, 2.947509765625, 2.6668701171875, -3.03961181640625, 2.270233154296875, 8.567138671875, 2.0899658203125, 0.51708984375, 4.46209716796875, 0.9115581512451172, -1.893585205078125, 5.639190673828125, 3.3411102294921875, 5.4288330078125, 2.40472412109375, -2.5058441162109375, 5.9364013671875, -1.4127197265625, -2.7383270263671875, 4.068084716796875, -1.609039306640625, 1.1633453369140625, 3.208099365234375, 6.718719482421875, 3.8760223388671875, 1.9669952392578125, 0.8836097717285156, 0.4834747314453125, 4.72088623046875, 2.732666015625, -0.9211196899414062, 0.463623046875, -2.9033203125, 0.641326904296875, -1.0541915893554688, -1.9481201171875, -1.108978271484375, 2.2352447509765625, 0.34496307373046875, 1.3431396484375, -1.217498779296875, 4.308837890625, 2.0719757080078125, -0.7252960205078125, 1.264801025390625, -2.14422607421875, -1.3642654418945312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000062.npy"}
{"epoch": 0.12984293193717278, "step": 63, "batch_size": 128, "mean": 1.3046636581420898, "std": 2.879157781600952, "min": -7.41650390625, "p10": -1.8823364257812498, "median": 1.1374588012695312, "p90": 4.530044555664062, "max": 9.9932861328125, "pos_frac": 0.703125, "sample": [4.291656494140625, 2.9232177734375, 0.1488189697265625, -4.7867431640625, -4.1903076171875, -0.8492488861083984, 3.378204345703125, 0.05322265625, -0.46550750732421875, 1.467132568359375, -4.853271484375, -2.7338485717773438, -2.716949462890625, -2.07391357421875, 3.9900970458984375, 0.02466583251953125, 3.08514404296875, 3.79632568359375, 7.6470947265625, 7.836395263671875, -1.648681640625, -1.697479248046875, 9.93218994140625, -1.16229248046875, 1.346923828125, 2.265045166015625, 3.07537841796875, -2.4043197631835938, 0.33002471923828125, 4.205596923828125, -0.3091869354248047, 1.534149169921875, 0.0713653564453125, 0.4302940368652344, 1.4603404998779297, 9.9932861328125, -3.587432861328125, 2.2056121826171875, -2.19525146484375, 5.4215087890625, -1.2110595703125, 0.9701652526855469, 3.210723876953125, -3.0458984375, 1.1020660400390625, 2.7408447265625, 7.0540771484375, -7.41650390625, -0.594482421875, 5.297210693359375, 0.37461090087890625, 4.659820556640625, 3.008026123046875, 0.0, -0.08245086669921875, 1.1728515625, 3.136138916015625, -1.197906494140625, -0.015369415283203125, 0.9789009094238281, 1.2852706909179688, 0.803131103515625, 2.564239501953125, 5.96661376953125, -1.67681884765625, 0.31143951416015625, 3.2904052734375, 1.7415771484375, -1.1770248413085938, 4.26678466796875, 5.68218994140625, 1.2411651611328125, 3.355377197265625, 0.40869140625, 4.47442626953125, -2.507659912109375, 3.662933349609375, 0.6658859252929688, 1.5300750732421875, 0.312835693359375, 0.0101318359375, 8.11572265625, 1.3377685546875, -1.55072021484375, 2.2451248168945312, 0.5639495849609375, 2.640911102294922, 0.3078117370605469, 1.6975784301757812, 2.1095352172851562, 2.0535736083984375, -1.49951171875, -1.80023193359375, 3.94805908203125, 0.3767662048339844, 3.0272216796875, 0.31542205810546875, 4.10540771484375, 3.4229583740234375, 2.9503173828125, 3.4495849609375, 2.1683807373046875, 2.48541259765625, -0.4718780517578125, 2.2323837280273438, -0.6054153442382812, -0.802734375, -0.0686798095703125, 4.09881591796875, 0.2787322998046875, 0.35943603515625, -1.3051910400390625, 3.097137451171875, 1.7726898193359375, 1.7169342041015625, 4.665435791015625, 0.6199798583984375, 1.71624755859375, -0.362548828125, 0.8345718383789062, -4.331512451171875, 0.1548614501953125, 2.2877197265625, 0.828887939453125, -1.605621337890625, -0.42043113708496094, 2.9586181640625, 5.284759521484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000063.npy"}
{"epoch": 0.1319371727748691, "step": 64, "batch_size": 128, "mean": 1.7037553787231445, "std": 3.096487045288086, "min": -6.4594879150390625, "p10": -1.409552001953125, "median": 1.2048187255859375, "p90": 6.096261596679687, "max": 11.182220458984375, "pos_frac": 0.7265625, "sample": [4.7500762939453125, -0.61669921875, -4.02496337890625, 7.722686767578125, -2.00823974609375, 3.64068603515625, 2.1786956787109375, 2.668701171875, 7.314453125, 1.3566131591796875, 3.040802001953125, 2.00811767578125, 7.819091796875, 1.77239990234375, 0.93536376953125, -2.9514007568359375, 7.49578857421875, 2.7334747314453125, 0.279052734375, -1.499786376953125, 0.0571746826171875, 1.2389678955078125, -1.370880126953125, -2.6944580078125, 1.1489181518554688, -6.4594879150390625, 2.4725303649902344, 1.9549636840820312, 3.83538818359375, 4.511688232421875, 0.934478759765625, 2.15545654296875, 4.30413818359375, 0.9211158752441406, 0.4990234375, 0.5849227905273438, 1.34893798828125, 2.3365097045898438, 2.0551300048828125, 1.1706695556640625, 3.837554931640625, 0.07254791259765625, -0.5397491455078125, 2.6711578369140625, 0.272369384765625, 10.260833740234375, 6.55023193359375, -1.7618980407714844, 5.855377197265625, 1.8562164306640625, 3.578857421875, -0.3736114501953125, 0.17345428466796875, 0.21783447265625, 0.5781230926513672, -0.98828125, -2.8301849365234375, 3.307525634765625, 2.9017715454101562, 3.1331939697265625, 3.16571044921875, 2.9276161193847656, 0.39630126953125, 5.418212890625, 1.112213134765625, 0.45709228515625, -0.13153076171875, -5.72467041015625, -0.81011962890625, 0.11566162109375, -0.030029296875, 1.03106689453125, 9.4462890625, 2.2000656127929688, 0.3740234375, 2.42950439453125, 3.1409454345703125, -0.0356597900390625, -1.018096923828125, 0.583099365234375, -1.687896728515625, 4.782196044921875, -0.4706077575683594, 0.12709999084472656, 3.36865234375, 6.566680908203125, 1.4808120727539062, 1.74359130859375, 0.8714332580566406, -0.67938232421875, 1.8828125, 11.182220458984375, 5.11767578125, 6.50933837890625, 0.6707916259765625, -0.053497314453125, -5.11248779296875, 0.8367156982421875, -0.92724609375, 0.53692626953125, 3.432281494140625, 6.210357666015625, 2.449291229248047, 0.0, 0.29126739501953125, 0.5948657989501953, -0.16619873046875, 2.16998291015625, 4.855339050292969, -1.071990966796875, 6.04736328125, 1.0657081604003906, -0.8134536743164062, 7.03155517578125, 1.4484100341796875, -0.7073173522949219, -0.1294097900390625, 7.74688720703125, -4.597625732421875, 2.2478485107421875, 0.0, 3.56365966796875, 5.34710693359375, 1.8261489868164062, 4.4078521728515625, 3.2414093017578125, -4.12451171875, -0.4730987548828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000064.npy"}
{"epoch": 0.13403141361256546, "step": 65, "batch_size": 128, "mean": 1.16643226146698, "std": 2.8224687576293945, "min": -8.59381103515625, "p10": -1.8624816894531246, "median": 0.8062305450439453, "p90": 5.153990173339844, "max": 6.76312255859375, "pos_frac": 0.6484375, "sample": [-0.043521881103515625, -2.927398681640625, 0.7529296875, -0.022060394287109375, 6.5019378662109375, 1.0992088317871094, 0.47287750244140625, 2.4363250732421875, 3.6232452392578125, -5.439453125, -3.399627685546875, -5.919525146484375, 0.5057373046875, 3.815460205078125, -1.0873794555664062, 4.09307861328125, 4.631378173828125, -0.2999267578125, 0.7683753967285156, -1.066558837890625, -3.158111572265625, -0.034450531005859375, 1.364471435546875, 5.1566925048828125, 2.485555648803711, 2.19793701171875, 5.064453125, 1.6767959594726562, -3.0250244140625, 0.5672988891601562, 1.5785064697265625, -1.2477798461914062, -0.5897293090820312, 3.143524169921875, 6.31866455078125, -1.5070648193359375, 0.20932579040527344, 5.938262939453125, -0.450653076171875, 0.6035308837890625, 3.18548583984375, -3.3632965087890625, 1.43463134765625, 1.8373260498046875, 6.2760009765625, -0.54376220703125, 4.5516204833984375, 3.3271636962890625, 4.208160400390625, 0.5350723266601562, 1.6181716918945312, 2.916290283203125, -2.04632568359375, 1.36224365234375, 2.8402633666992188, -0.6207427978515625, 6.426605224609375, 1.2416954040527344, 0.17950439453125, 2.863677978515625, 0.0716552734375, 0.0339202880859375, 0.6942596435546875, -0.40645599365234375, 2.2616424560546875, -0.1934051513671875, 1.3603897094726562, -0.100250244140625, -1.78369140625, -1.027435302734375, 3.6282501220703125, 5.1993408203125, -0.133270263671875, 0.6826019287109375, 3.189239501953125, 0.1935577392578125, -0.906585693359375, 0.3343963623046875, 0.844085693359375, -0.5146636962890625, -4.630096435546875, 0.15765380859375, 0.0, 0.984832763671875, 4.0636749267578125, 2.4063491821289062, -1.4613037109375, 0.99566650390625, 6.2275238037109375, 2.54583740234375, 0.675079345703125, 3.2265777587890625, 5.15283203125, 2.5009613037109375, 5.2704010009765625, -0.34747314453125, 3.7599029541015625, 1.844451904296875, -0.4223613739013672, 6.76312255859375, -1.65960693359375, 3.820953369140625, -8.59381103515625, 1.6834716796875, 1.6229095458984375, 5.527008056640625, -0.1556549072265625, 6.1602783203125, 4.24871826171875, 6.52288818359375, 0.7223587036132812, -0.0535125732421875, 3.2088470458984375, 1.4906005859375, -4.92120361328125, -1.1370658874511719, 4.643798828125, -2.79913330078125, 0.0765380859375, 1.1119537353515625, -2.45684814453125, -1.0771408081054688, 2.47381591796875, -1.16351318359375, 3.622833251953125, -0.3531036376953125, 1.0028915405273438, -0.42424774169921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000065.npy"}
{"epoch": 0.13612565445026178, "step": 66, "batch_size": 128, "mean": 1.319630742073059, "std": 3.77875018119812, "min": -10.396728515625, "p10": -2.555314636230469, "median": 1.0244951248168945, "p90": 6.138592529296875, "max": 12.14178466796875, "pos_frac": 0.6640625, "sample": [1.0537261962890625, 1.0420551300048828, 5.827056884765625, -1.9312744140625, -1.0174407958984375, 6.1500244140625, -0.6392364501953125, 0.9415016174316406, 1.4748077392578125, 0.9585418701171875, -0.0041103363037109375, -0.3227195739746094, 2.0929527282714844, 0.7060317993164062, 1.5126495361328125, -0.307586669921875, -0.5040454864501953, 0.571990966796875, 8.1002197265625, 1.6114654541015625, -0.483428955078125, -6.78045654296875, 4.9586181640625, 5.41107177734375, 3.2825927734375, 1.170806884765625, -0.7315216064453125, 1.64324951171875, 0.00054168701171875, 5.6626129150390625, 2.8468017578125, 0.5569534301757812, 12.14178466796875, -2.43084716796875, 0.9034576416015625, -0.4122505187988281, -0.8994140625, 10.787109375, 4.252471923828125, -2.6097564697265625, -6.3455352783203125, 2.2396373748779297, 6.66876220703125, 1.649871826171875, 1.0069351196289062, 2.5683746337890625, -0.13677978515625, 8.63580322265625, 7.1334228515625, -0.6669921875, -8.2926025390625, 1.105194091796875, 0.4947662353515625, 11.808837890625, 7.84222412109375, -0.275726318359375, 0.8619537353515625, 4.3787689208984375, -10.396728515625, 1.12481689453125, -1.3699798583984375, 0.8995819091796875, 9.59771728515625, 0.8999176025390625, 3.4853515625, -8.402984619140625, 0.8313751220703125, -1.6115875244140625, 2.700469970703125, -0.4495697021484375, 0.16302490234375, 0.116668701171875, 6.588775634765625, -4.1011962890625, 0.63458251953125, 1.0000686645507812, 3.84796142578125, -1.1183910369873047, 3.265228271484375, -5.1587677001953125, 2.9538421630859375, 1.4586944580078125, -3.0831146240234375, 4.1116943359375, -0.30879974365234375, 1.5658493041992188, -0.115753173828125, 0.731048583984375, 4.62554931640625, 2.455413818359375, 1.22021484375, 4.22235107421875, 2.4783935546875, -9.09454345703125, -0.160491943359375, -2.055511474609375, -3.4472198486328125, -2.854248046875, 3.591094970703125, 1.06085205078125, 6.14752197265625, 1.7880706787109375, 2.701568603515625, 0.0, 3.1173362731933594, -0.978973388671875, -0.9781341552734375, 6.7867889404296875, 1.463531494140625, 3.595703125, 0.371063232421875, 2.1646728515625, -0.08213615417480469, -3.467803955078125, 0.69720458984375, 2.56378173828125, -1.252716064453125, 6.134765625, 4.1361083984375, 4.5151519775390625, -2.531982421875, 0.4862060546875, 1.1481781005859375, -2.5045242309570312, 4.852020263671875, 1.0865249633789062, 5.13726806640625, 2.64996337890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000066.npy"}
{"epoch": 0.1382198952879581, "step": 67, "batch_size": 128, "mean": 1.213550329208374, "std": 3.4359731674194336, "min": -8.1878662109375, "p10": -3.160035705566406, "median": 1.2805309295654297, "p90": 5.406491088867187, "max": 10.43023681640625, "pos_frac": 0.6640625, "sample": [-0.9475555419921875, 4.000762939453125, -1.46405029296875, 9.659271240234375, 0.7684650421142578, -4.09088134765625, 0.5623626708984375, 4.452728271484375, 4.20733642578125, -0.3319091796875, 1.7008438110351562, 10.428802490234375, 0.34967041015625, 1.15228271484375, 2.367645263671875, 3.5450439453125, 3.613372802734375, 5.037628173828125, -0.214508056640625, -1.3222198486328125, 1.6295013427734375, -1.1052322387695312, 3.9002227783203125, 0.46533203125, 1.390106201171875, 5.407623291015625, -3.016357421875, 0.272430419921875, 2.2278289794921875, -0.65399169921875, -4.407196044921875, 0.6932430267333984, 1.6319580078125, 1.5278167724609375, 0.773895263671875, -3.74713134765625, 0.0, 1.2504158020019531, 4.4315185546875, 1.8997840881347656, -1.5464935302734375, 3.874542236328125, 3.6029510498046875, 6.8143310546875, 1.3106460571289062, -1.7067508697509766, 2.5323944091796875, -2.825439453125, -5.6312255859375, -0.26996612548828125, -4.3175048828125, 4.9547271728515625, -0.01335906982421875, 6.75311279296875, 1.94384765625, 2.615570068359375, 8.74114990234375, -3.72296142578125, 1.7850341796875, 0.7046070098876953, 1.62774658203125, 2.606822967529297, 5.406005859375, -3.4952850341796875, 2.2244720458984375, -0.11268997192382812, 0.3010444641113281, -7.320068359375, -0.56695556640625, -0.10495185852050781, 5.509979248046875, 1.09051513671875, 0.7268180847167969, 4.986724853515625, -6.93121337890625, 10.43023681640625, 7.225982666015625, 2.73443603515625, -0.15505218505859375, 1.8792724609375, 1.787872314453125, -1.7258930206298828, 1.35968017578125, -0.29205322265625, -5.37908935546875, -0.05864524841308594, 5.7867889404296875, 0.23410797119140625, 0.37189483642578125, 4.3623504638671875, 5.5089111328125, -8.1878662109375, 6.1002197265625, 2.9527130126953125, -6.6488037109375, 4.8807220458984375, 0.081939697265625, 2.38232421875, 2.796783447265625, 5.764007568359375, 4.338836669921875, 1.3362808227539062, -1.1547164916992188, 1.6585845947265625, 0.4812164306640625, -0.59228515625, -1.3723182678222656, -6.917266845703125, -1.6077423095703125, 0.7028884887695312, 1.946502685546875, 4.299224853515625, -0.08673095703125, -1.2963943481445312, -0.9115180969238281, 4.5350341796875, 1.5133056640625, 0.4788932800292969, 4.33758544921875, 1.4232177734375, 2.223846435546875, 1.9365577697753906, 3.007080078125, 1.090179443359375, 0.42903900146484375, -1.3777313232421875, 2.43023681640625, -1.30523681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000067.npy"}
{"epoch": 0.14031413612565444, "step": 68, "batch_size": 128, "mean": 1.6295251846313477, "std": 3.9582180976867676, "min": -7.6822509765625, "p10": -2.954060363769531, "median": 0.8787155151367188, "p90": 7.351165771484374, "max": 15.154327392578125, "pos_frac": 0.671875, "sample": [7.632415771484375, -0.06898307800292969, 0.29498291015625, -1.602630615234375, -0.3771247863769531, 0.2358417510986328, 0.278289794921875, 3.406097412109375, -0.5595893859863281, 2.017974853515625, 1.3415298461914062, -0.150390625, 0.327972412109375, 5.1793212890625, 4.77557373046875, -4.4830322265625, 0.01638031005859375, 0.51214599609375, 1.358154296875, 2.399517059326172, 5.3177490234375, -0.4666748046875, -2.341693878173828, 0.768157958984375, 15.154327392578125, 0.5145721435546875, 1.347574234008789, 2.8939208984375, 0.25333595275878906, -6.432373046875, 4.88763427734375, 2.91998291015625, 9.05865478515625, 7.584075927734375, 0.005950927734375, 7.296417236328125, 0.5598602294921875, -0.250701904296875, -3.41650390625, -5.06427001953125, -1.837127685546875, -1.5433731079101562, 3.815380096435547, 2.4862060546875, 0.8311996459960938, -4.54632568359375, -0.1449737548828125, 2.4391250610351562, 4.4298095703125, 6.259674072265625, 2.458831787109375, -2.8664627075195312, 3.2705078125, 4.4857177734375, 10.283599853515625, -0.0399169921875, 3.54705810546875, 8.85546875, -0.9385337829589844, 1.39324951171875, 4.261604309082031, 2.262725830078125, -0.07481002807617188, 2.0977783203125, 0.05465126037597656, -0.20769882202148438, -6.602996826171875, 4.559478759765625, 0.8807373046875, 12.4302978515625, 0.5035934448242188, -3.287506103515625, 1.1894378662109375, 0.7854080200195312, -6.049072265625, 2.37823486328125, -0.70416259765625, -3.2073974609375, 10.0399169921875, -0.5771827697753906, -2.948333740234375, 1.9752273559570312, -0.45749664306640625, 3.377838134765625, 2.04937744140625, 0.216644287109375, -1.395599365234375, 6.311614990234375, -7.6822509765625, 9.191535949707031, 3.768463134765625, 1.08148193359375, 0.0318603515625, -0.45928955078125, -1.100433349609375, 5.23406982421875, -5.3191680908203125, -4.531951904296875, 4.253387451171875, 2.0810470581054688, 5.0648345947265625, 0.7829303741455078, 2.6168785095214844, 8.44189453125, 7.8836669921875, 1.048248291015625, 6.5027008056640625, 7.478912353515625, 2.155029296875, 3.397125244140625, 2.4063720703125, -2.9674224853515625, 1.8343353271484375, 1.8541240692138672, -0.8256454467773438, 1.1644020080566406, 4.283477783203125, 0.6243743896484375, -1.428375244140625, 0.8766937255859375, -0.19179534912109375, -1.6018829345703125, 0.0291900634765625, 12.314041137695312, 4.6953125, -1.77764892578125, -1.1130752563476562, 0.8558921813964844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000068.npy"}
{"epoch": 0.1424083769633508, "step": 69, "batch_size": 128, "mean": 2.1001274585723877, "std": 3.8891777992248535, "min": -11.48553466796875, "p10": -1.8825836181640623, "median": 1.703023910522461, "p90": 7.782009887695311, "max": 13.10626220703125, "pos_frac": 0.75, "sample": [5.386566162109375, -0.351715087890625, 1.9208908081054688, 4.88385009765625, 0.685516357421875, -7.7474365234375, 3.5005264282226562, 2.2112350463867188, -2.31890869140625, -0.3096199035644531, 4.565208435058594, 1.7588996887207031, 9.596649169921875, 0.1989898681640625, 7.979827880859375, -0.18471145629882812, 7.7040252685546875, 0.6116123199462891, 2.85528564453125, 0.795379638671875, -1.3776092529296875, 0.035465240478515625, -3.6179962158203125, -6.12298583984375, 5.78887939453125, 9.7066650390625, 5.259735107421875, 4.66888427734375, 0.3792457580566406, 0.686920166015625, -11.48553466796875, 4.63275146484375, -0.15677642822265625, 1.90289306640625, 2.2581939697265625, -1.24371337890625, 0.8838272094726562, 0.48824310302734375, 6.9959716796875, 1.481414794921875, 10.12890625, -4.93121337890625, 0.04165458679199219, 2.004608154296875, 1.6471481323242188, 1.9736824035644531, -3.277069091796875, 7.636566162109375, 2.5204925537109375, 0.2801361083984375, 6.48504638671875, 0.89190673828125, 8.083770751953125, 4.7623291015625, 5.42205810546875, -4.283599853515625, -4.2394256591796875, 0.38459014892578125, 2.45062255859375, -0.6119422912597656, 0.822265625, 4.79119873046875, 0.3582801818847656, 3.441070556640625, 1.5731964111328125, 4.0609283447265625, 0.585205078125, 4.66827392578125, 5.142242431640625, 5.10748291015625, 9.1104736328125, 3.5945587158203125, 1.3524169921875, 0.5910491943359375, 2.068115234375, -0.8050079345703125, 8.640472412109375, 1.3957290649414062, 2.7684478759765625, 1.473541259765625, -1.7453937530517578, -0.259124755859375, 1.946523666381836, 1.0869979858398438, -1.5869598388671875, 2.125030517578125, 2.83575439453125, 0.636505126953125, 3.6405792236328125, 9.123016357421875, 8.16302490234375, 7.9639739990234375, 1.21307373046875, 6.678924560546875, -2.0317840576171875, -2.418426513671875, -0.4240570068359375, 1.628387451171875, 9.931396484375, -0.42931365966796875, 3.9738616943359375, 2.197998046875, 13.10626220703125, 1.990966796875, 4.667724609375, 3.5926971435546875, -1.9869384765625, 0.51300048828125, -0.94708251953125, 2.342498779296875, -1.7692108154296875, -1.5960540771484375, 0.569732666015625, 5.639007568359375, 7.330322265625, -6.485870361328125, -0.9481925964355469, -1.837860107421875, 3.3416900634765625, 8.610794067382812, 1.107086181640625, 3.390655517578125, 0.5029754638671875, -0.5592155456542969, 0.22386550903320312, 3.35595703125, 5.566558837890625, 1.8282051086425781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000069.npy"}
{"epoch": 0.14450261780104712, "step": 70, "batch_size": 128, "mean": 2.09108829498291, "std": 4.001116752624512, "min": -13.19940185546875, "p10": -1.406229400634765, "median": 1.4785652160644531, "p90": 7.08997802734375, "max": 15.5628662109375, "pos_frac": 0.7109375, "sample": [-0.9052734375, 3.500762939453125, 2.0292739868164062, -1.2826766967773438, -0.376068115234375, 0.7685546875, 3.25079345703125, 2.7046737670898438, -6.24273681640625, 7.158935546875, -1.931488037109375, 0.10313796997070312, -0.0283966064453125, 3.15765380859375, 6.09124755859375, 5.123924255371094, 2.2589874267578125, 3.570648193359375, 0.31211090087890625, 4.0855712890625, 1.1067428588867188, 6.46551513671875, 0.644378662109375, 1.7051734924316406, -5.4466552734375, 4.02294921875, 0.40114593505859375, 3.3629608154296875, 2.7265090942382812, -1.7677230834960938, 6.8704833984375, 0.313201904296875, 0.31713104248046875, -0.8016357421875, 1.00115966796875, 0.183074951171875, 7.88671875, -3.11151123046875, -4.35760498046875, -0.1129150390625, 1.809967041015625, 0.0344085693359375, 10.81011962890625, 1.4712142944335938, 1.7468414306640625, -0.48095703125, -1.69451904296875, 0.9722747802734375, -0.69158935546875, -2.9278564453125, 0.98974609375, 0.6186141967773438, -0.00726318359375, 0.28314208984375, -0.4295005798339844, 6.171661376953125, 10.102203369140625, 4.337890625, 7.0604248046875, 1.4554367065429688, 2.14849853515625, 5.38409423828125, -0.8994827270507812, 0.1224365234375, 2.55828857421875, 2.471405029296875, 0.242218017578125, 4.760955810546875, 1.4181137084960938, -0.149566650390625, -5.54052734375, 0.8527984619140625, -0.4152507781982422, 2.3635406494140625, 1.5377426147460938, 7.6226806640625, 0.803558349609375, 1.8250885009765625, 8.6014404296875, 2.38763427734375, 12.089752197265625, 6.994354248046875, 2.0676651000976562, 3.86688232421875, 5.9346466064453125, 2.9503173828125, -1.8555145263671875, 3.35394287109375, -0.6559066772460938, 13.370849609375, 5.1029205322265625, 8.193572998046875, 0.0, 0.8373947143554688, 1.958251953125, 9.06988525390625, 6.4435577392578125, 4.63482666015625, 0.3229217529296875, 11.3277587890625, 4.247802734375, 6.44964599609375, -0.02032470703125, -2.181720733642578, 3.3925094604492188, -0.630889892578125, 7.27557373046875, 1.4859161376953125, 4.5789031982421875, -1.0817413330078125, -0.0411529541015625, -0.8395957946777344, 1.1687698364257812, 2.8603668212890625, -7.42791748046875, -13.19940185546875, 1.8220672607421875, -0.37421417236328125, 1.934417724609375, 5.43768310546875, 2.6636962890625, 2.453643798828125, 1.26904296875, -0.0803375244140625, 15.5628662109375, 0.0, 0.7307815551757812, -0.29183197021484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000070.npy"}
{"epoch": 0.14659685863874344, "step": 71, "batch_size": 128, "mean": 2.0354151725769043, "std": 4.485329627990723, "min": -14.95257568359375, "p10": -3.3837585449218746, "median": 1.9865570068359375, "p90": 7.591867065429687, "max": 12.829559326171875, "pos_frac": 0.7265625, "sample": [-0.42305755615234375, 10.855377197265625, -0.8598403930664062, -4.24957275390625, -1.708343505859375, -6.51898193359375, -0.8280868530273438, 10.75201416015625, -6.1778411865234375, 3.57696533203125, 1.8408546447753906, 3.2856788635253906, -2.65032958984375, 3.71124267578125, 1.8534164428710938, 5.1856231689453125, 5.5707855224609375, 0.428253173828125, 3.4669647216796875, 7.35308837890625, 2.2908401489257812, -0.770111083984375, 3.257293701171875, 12.829559326171875, 2.5975799560546875, -1.434906005859375, 1.413818359375, 7.282989501953125, 2.7741470336914062, 3.54278564453125, -4.4088134765625, 4.8619384765625, 1.1435165405273438, 9.411224365234375, -2.94915771484375, 4.481781005859375, 8.05487060546875, 0.5324077606201172, 8.98614501953125, -4.8744964599609375, -1.76953125, -0.8223114013671875, -3.28424072265625, 2.5953445434570312, 1.7714385986328125, 0.2086639404296875, 1.0130615234375, -12.32855224609375, -3.615966796875, -0.8838272094726562, 2.9647216796875, 4.79107666015625, -1.037628173828125, 4.678005218505859, 1.1690559387207031, 4.9078369140625, 11.37249755859375, 2.03277587890625, -0.4195232391357422, 1.12847900390625, 5.32574462890625, -1.26690673828125, 6.4774169921875, 7.660888671875, 0.425201416015625, 0.46533203125, 0.6284065246582031, -2.51483154296875, 4.4803466796875, -0.6328887939453125, 1.5118789672851562, 10.692047119140625, 6.48529052734375, -4.61883544921875, -4.705562591552734, 6.111114501953125, 8.29486083984375, -6.8377685546875, -3.2633056640625, 0.5199508666992188, -2.3580322265625, -1.070526123046875, 1.272979736328125, 2.670654296875, 7.096588134765625, 7.290740966796875, 1.544464111328125, 4.9334716796875, 5.955352783203125, 3.33856201171875, 1.5941123962402344, 1.992645263671875, 2.02313232421875, 0.16387557983398438, 0.602935791015625, 9.867706298828125, -0.7099838256835938, 7.562286376953125, 3.6393814086914062, 4.92291259765625, 6.7850341796875, -3.7217483520507812, 0.4378185272216797, 4.56689453125, 9.45904541015625, 0.560943603515625, 1.98046875, 1.1974639892578125, -3.6654052734375, 1.00201416015625, 2.6328582763671875, 1.4149723052978516, 4.8486328125, 2.6385955810546875, 2.29107666015625, 2.2423534393310547, 2.5654983520507812, 5.11932373046875, 2.094158172607422, 2.4782867431640625, 2.0198516845703125, 1.9687347412109375, 3.3370361328125, 0.9163169860839844, -0.029819488525390625, 11.500946044921875, -14.95257568359375, 3.3137359619140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000071.npy"}
{"epoch": 0.1486910994764398, "step": 72, "batch_size": 128, "mean": 2.0696425437927246, "std": 4.100938320159912, "min": -10.386962890625, "p10": -2.471840667724609, "median": 1.9037284851074219, "p90": 8.082211303710936, "max": 13.59814453125, "pos_frac": 0.6796875, "sample": [3.939605712890625, 2.17120361328125, -2.2427139282226562, -0.658172607421875, 10.9642333984375, -2.343902587890625, 1.7276611328125, 4.7175140380859375, 1.8775634765625, -0.6409530639648438, -0.3259429931640625, 1.34326171875, 2.348846435546875, 2.2010040283203125, 3.319580078125, 9.01361083984375, -2.0900115966796875, 1.410552978515625, -3.118896484375, 1.985015869140625, 3.0315628051757812, 8.0423583984375, 0.818634033203125, 9.514404296875, 4.4974822998046875, 0.5833377838134766, -0.9065628051757812, 3.596832275390625, -3.5195388793945312, 0.33014678955078125, 3.5496368408203125, -10.386962890625, -1.175750732421875, -6.049461364746094, 2.6986312866210938, 2.588897705078125, 2.7400970458984375, -1.0180740356445312, 1.446502685546875, 0.8724441528320312, 3.68341064453125, -1.690185546875, 7.882904052734375, 5.355804443359375, 8.198471069335938, 2.6927452087402344, 3.473785400390625, 0.2735595703125, -4.548828125, -2.319091796875, -2.7192001342773438, 4.50830078125, -2.109954833984375, 0.0, 7.4918212890625, 8.41168212890625, 8.369293212890625, -0.8535346984863281, 5.158935546875, 9.57891845703125, 0.5133781433105469, -2.9524993896484375, 3.644672393798828, 5.721473693847656, 2.96673583984375, 5.7665863037109375, 0.9041748046875, 1.8798141479492188, -0.6793937683105469, 3.07855224609375, -4.09942626953125, 1.82855224609375, 0.6830062866210938, 3.3924102783203125, 8.175201416015625, 1.4234390258789062, 0.14418411254882812, 4.51739501953125, 0.64019775390625, 8.563468933105469, 1.71734619140625, 2.1869258880615234, -2.1713027954101562, -1.91729736328125, -2.3658294677734375, -1.31134033203125, 1.3667144775390625, -0.2751922607421875, -2.0118961334228516, 7.23748779296875, -4.802490234375, 8.4671630859375, 0.3056488037109375, 4.128082275390625, 2.758087158203125, 6.121826171875, 6.6958465576171875, -1.2443084716796875, 2.2692718505859375, 3.166046142578125, 2.99566650390625, 7.337554931640625, 0.0, 5.2678070068359375, -3.0265235900878906, 1.3974571228027344, 7.2510986328125, 13.59814453125, 5.8188018798828125, 6.766510009765625, 4.102836608886719, -1.46051025390625, -1.20867919921875, -7.54534912109375, 11.70684814453125, -0.2815399169921875, -1.7375640869140625, 0.0, 1.927642822265625, 3.1720733642578125, 3.7630233764648438, 2.1805419921875, -5.637603759765625, 3.603485107421875, 2.7790145874023438, -3.097564697265625, 1.54437255859375, 9.571487426757812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000072.npy"}
{"epoch": 0.15078534031413612, "step": 73, "batch_size": 128, "mean": 1.9992163181304932, "std": 4.436720848083496, "min": -15.4456787109375, "p10": -2.8119453430175776, "median": 1.5149192810058594, "p90": 6.932080078125, "max": 13.86004638671875, "pos_frac": 0.7109375, "sample": [0.92352294921875, 2.460540771484375, 7.377471923828125, 5.016387939453125, -0.7740478515625, 2.4990234375, 0.911041259765625, -1.468780517578125, -0.7234954833984375, 5.049201965332031, 3.98443603515625, 4.29345703125, 1.4748992919921875, 10.23822021484375, 0.053524017333984375, 0.184539794921875, -7.03173828125, -7.74462890625, 0.6753692626953125, 0.553314208984375, 2.112701416015625, 0.32224273681640625, -0.21576309204101562, 2.634857177734375, 1.3283920288085938, 0.42733001708984375, 1.8920974731445312, 1.5520286560058594, 1.6391372680664062, 4.68389892578125, -6.458465576171875, -2.5404052734375, 5.1529541015625, 10.1553955078125, -2.0504150390625, 1.4641036987304688, -3.729766845703125, 13.27947998046875, 5.508857727050781, 4.40643310546875, 3.9649658203125, 1.4778099060058594, 5.3619232177734375, -2.3101425170898438, 13.86004638671875, -1.2568511962890625, -2.0536346435546875, -1.6424560546875, 1.7623672485351562, 9.985565185546875, -6.10302734375, 2.642822265625, 10.23858642578125, -15.4456787109375, 3.903106689453125, 1.4392471313476562, 2.6972198486328125, -3.8502960205078125, 5.75018310546875, -3.1458969116210938, 2.7396240234375, 0.5203857421875, 4.724052429199219, -0.068939208984375, -1.722412109375, 6.82489013671875, 4.596870422363281, -4.295867919921875, 8.028533935546875, 0.645111083984375, -1.1039695739746094, -2.54534912109375, 4.783660888671875, 5.7788848876953125, 0.754425048828125, 1.3359851837158203, -6.0724334716796875, -5.060455322265625, -2.0567169189453125, 1.232696533203125, 6.6005859375, 4.8941192626953125, 5.213897705078125, 0.7667236328125, -2.0597686767578125, 3.1597824096679688, -0.5088119506835938, 2.7453765869140625, -1.0678024291992188, 5.335784912109375, 0.9357528686523438, 1.386505126953125, 5.152923583984375, 4.032691955566406, 6.9041748046875, 6.9971923828125, 2.370870590209961, 3.50579833984375, -5.17218017578125, 5.69927978515625, -0.05596160888671875, 4.6890411376953125, 3.7959213256835938, 4.67779541015625, 6.84027099609375, 5.431182861328125, 1.2231826782226562, -1.125, 12.373016357421875, 6.7618408203125, 3.7982177734375, -1.196929931640625, 1.4359588623046875, -0.39322853088378906, 9.893707275390625, -2.6688232421875, 0.7108707427978516, 7.236297607421875, -2.5833816528320312, 0.64501953125, 9.0108642578125, 1.8638839721679688, 6.074005126953125, 3.6262855529785156, 1.16015625, 0.7859039306640625, 2.92950439453125, -3.7649993896484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000073.npy"}
{"epoch": 0.15287958115183245, "step": 74, "batch_size": 128, "mean": 1.9281835556030273, "std": 4.660504341125488, "min": -13.082794189453125, "p10": -3.968623352050781, "median": 1.7634201049804688, "p90": 7.917648315429687, "max": 13.35968017578125, "pos_frac": 0.6875, "sample": [2.49560546875, 8.9984130859375, 3.254425048828125, 0.00140380859375, 10.248382568359375, -0.35784912109375, -4.5892181396484375, 1.49334716796875, 6.282012939453125, 6.8980712890625, -1.16131591796875, 9.383987426757812, 0.0, 3.5887451171875, 0.682830810546875, 7.81048583984375, 3.7498779296875, 9.66259765625, -2.1378936767578125, -0.205810546875, 3.952117919921875, 4.0546875, 2.18798828125, 1.6712646484375, 7.759307861328125, -5.452789306640625, -4.80731201171875, -0.0547027587890625, -1.5921630859375, -3.0286102294921875, -1.183349609375, -0.1844482421875, -4.32098388671875, -0.3090476989746094, 4.59442138671875, 1.252899169921875, 0.8867263793945312, 7.401508331298828, 0.07732009887695312, 7.7283782958984375, -0.5531196594238281, -2.13653564453125, -7.536529541015625, -7.02508544921875, 0.6890411376953125, 4.2267913818359375, 1.252593994140625, -1.375396728515625, -1.4429931640625, 5.2155609130859375, -8.654541015625, -9.64080810546875, -0.278350830078125, 2.726318359375, 1.44097900390625, -1.6976814270019531, 1.2281913757324219, 1.691925048828125, -7.483184814453125, 6.7008056640625, 5.4411773681640625, 4.0258636474609375, 1.759857177734375, 8.78863525390625, 5.613037109375, 2.8504638671875, 9.23284912109375, 3.785247802734375, 1.6575775146484375, 13.35968017578125, 1.832733154296875, 0.50506591796875, -1.7764129638671875, 3.92364501953125, -0.9019088745117188, 3.861785888671875, 2.4948272705078125, 2.86688232421875, 0.65069580078125, 5.2788543701171875, 3.282958984375, 0.1114959716796875, 1.7669830322265625, -13.082794189453125, 11.9180908203125, 0.797332763671875, 1.089111328125, 2.02264404296875, 1.0194091796875, -0.8475418090820312, -0.1982269287109375, -3.04296875, -2.4775390625, 5.839637756347656, 3.14959716796875, 5.32989501953125, 5.333831787109375, -9.12353515625, -5.68798828125, 2.84930419921875, 5.0008697509765625, -3.8914031982421875, 1.9901123046875, 8.167694091796875, 2.29144287109375, 5.9291839599609375, 4.0659942626953125, -4.1488037109375, 6.59600830078125, 0.04766845703125, -0.3302192687988281, 1.802886962890625, 4.104705810546875, 2.8405838012695312, 11.9979248046875, -0.275177001953125, 1.74005126953125, 4.366058349609375, 3.3936309814453125, 0.4639892578125, 3.148651123046875, 2.7720565795898438, 0.7782516479492188, 8.610313415527344, 13.269744873046875, 4.4208526611328125, -0.215362548828125, 8.490234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000074.npy"}
{"epoch": 0.1549738219895288, "step": 75, "batch_size": 128, "mean": 2.635268211364746, "std": 4.498477458953857, "min": -10.528045654296875, "p10": -2.5457733154296873, "median": 2.2148218154907227, "p90": 8.194482421874996, "max": 14.0345458984375, "pos_frac": 0.765625, "sample": [1.9378280639648438, 2.8602066040039062, 1.9913482666015625, 1.1119232177734375, -1.7069091796875, 7.8193359375, 5.560699462890625, -2.60479736328125, 2.9167022705078125, 3.3712310791015625, 0.8226852416992188, -3.576202392578125, 2.07818603515625, -0.7858123779296875, 9.06982421875, 0.1861572265625, 14.0345458984375, 0.22149658203125, 6.944610595703125, 3.8262176513671875, 7.657989501953125, 6.483909606933594, 6.35809326171875, -3.683868408203125, 0.34090423583984375, 10.66607666015625, -3.461181640625, 6.4507293701171875, -0.59478759765625, 1.2682037353515625, 2.725341796875, 7.01885986328125, 3.86676025390625, 2.44439697265625, 0.3370513916015625, -1.6337528228759766, 3.730804443359375, 4.62567138671875, 3.7995376586914062, 1.20623779296875, 11.808319091796875, 6.659992218017578, 1.63623046875, -10.528045654296875, 1.0324859619140625, -1.6954345703125, 0.44605064392089844, 0.15770721435546875, 9.0853271484375, 1.4814605712890625, 3.881805419921875, 4.397064208984375, 2.5131988525390625, 0.1418914794921875, 5.515625, 5.5258941650390625, -5.1121826171875, 10.757965087890625, -0.0261383056640625, 6.970703125, 3.66021728515625, 2.159719467163086, 1.65264892578125, 6.779052734375, 1.6488990783691406, 1.3286895751953125, 6.520782470703125, -2.001312255859375, 2.2699241638183594, 0.61358642578125, 1.866729736328125, -1.684051513671875, 1.6982765197753906, 10.63885498046875, 9.3251953125, 0.1632404327392578, -8.16925048828125, -7.742431640625, -1.00836181640625, 4.157146453857422, 12.014617919921875, 5.8127593994140625, 6.668212890625, -4.770271301269531, -0.10105133056640625, 9.22784423828125, 0.13444900512695312, 0.037750244140625, 2.80712890625, 6.3131866455078125, 4.514617919921875, -1.607177734375, 6.24017333984375, 2.95068359375, 7.17364501953125, 12.147552490234375, -0.6942596435546875, 3.61761474609375, -0.295867919921875, 6.7580108642578125, -9.084014892578125, 0.68341064453125, 5.157989501953125, 9.99285888671875, 5.861724853515625, 3.490264892578125, -0.24420166015625, 3.499359130859375, 1.5211410522460938, -3.893768310546875, 6.89971923828125, -0.00653839111328125, 0.001434326171875, 6.011421203613281, 4.681774139404297, 5.8734130859375, 0.53045654296875, 1.3601875305175781, 4.164886474609375, -2.520477294921875, 3.5355224609375, 1.6019439697265625, 5.533905029296875, 10.36395263671875, -1.79022216796875, -3.1580657958984375, -6.513725280761719, 0.5966339111328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000075.npy"}
{"epoch": 0.15706806282722513, "step": 76, "batch_size": 128, "mean": 2.9896059036254883, "std": 5.087985038757324, "min": -8.823486328125, "p10": -2.5939254760742183, "median": 2.1610069274902344, "p90": 9.822764587402343, "max": 22.043502807617188, "pos_frac": 0.7578125, "sample": [7.6810302734375, -3.7884521484375, 1.7895469665527344, 2.161468505859375, 2.7675704956054688, 5.9114990234375, 1.531097412109375, 1.3419380187988281, 2.7946090698242188, 3.999298095703125, 9.3885498046875, 0.03440093994140625, 0.1651611328125, -0.52301025390625, 13.593505859375, 4.883758544921875, -0.8831100463867188, -2.9386749267578125, 2.4430389404296875, -1.618560791015625, 11.887115478515625, 1.08795166015625, 5.452880859375, 2.0364532470703125, 2.8819427490234375, 13.5538330078125, 1.746002197265625, 3.1648025512695312, 1.0727424621582031, -1.2041015625, 2.9530029296875, -0.30928802490234375, 5.143646240234375, 8.545547485351562, -8.823486328125, 16.66143798828125, 3.5482330322265625, 4.61688232421875, 2.96466064453125, 4.52374267578125, 11.813873291015625, 5.38800048828125, 6.958160400390625, 7.225830078125, -6.8681640625, 1.0944404602050781, 3.1529006958007812, 3.3049468994140625, 0.3958282470703125, 9.994613647460938, 0.0, 4.3359222412109375, 0.1922454833984375, 0.8147125244140625, 7.534004211425781, -3.37274169921875, 11.4677734375, -7.085205078125, -1.631317138671875, -4.217926025390625, 6.1643829345703125, 11.596710205078125, -3.29071044921875, 1.3756103515625, 3.7720909118652344, 1.0718917846679688, 2.4476470947265625, 4.077323913574219, 5.778633117675781, -1.8929595947265625, 0.0050201416015625, 8.468719482421875, -2.89593505859375, 0.8056488037109375, -3.484710693359375, 0.0, 11.64715576171875, 4.658119201660156, 4.560516357421875, 12.279388427734375, 3.267974853515625, -0.05148124694824219, 1.33990478515625, 1.9858245849609375, 0.0809326171875, 2.6630706787109375, 1.8767280578613281, 8.71185302734375, 1.4222412109375, 3.5565948486328125, 4.361053466796875, 1.0485153198242188, -0.18657684326171875, 2.3734893798828125, -0.24852371215820312, -1.083648681640625, -4.8971710205078125, -0.270904541015625, 1.735382080078125, 0.01438140869140625, 2.3533935546875, 1.74127197265625, 7.627960205078125, 14.259735107421875, 9.598541259765625, 16.857025146484375, 9.749114990234375, 1.309173583984375, 0.89349365234375, -3.6800384521484375, 2.2100067138671875, -1.5811767578125, 22.043502807617188, -2.4644927978515625, -8.411468505859375, 1.1507568359375, 3.7203369140625, 3.0298004150390625, 7.42742919921875, 1.185577392578125, 2.4162139892578125, 8.244384765625, 2.1605453491210938, 1.132415771484375, -0.7810821533203125, 2.9788131713867188, 0.0681610107421875, -0.150604248046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000076.npy"}
{"epoch": 0.15916230366492146, "step": 77, "batch_size": 128, "mean": 2.4400293827056885, "std": 5.224828720092773, "min": -13.189971923828125, "p10": -3.603438568115234, "median": 1.6256866455078125, "p90": 9.120008850097655, "max": 18.9329833984375, "pos_frac": 0.6953125, "sample": [-1.0604019165039062, 2.7862396240234375, 1.4192657470703125, 2.981964111328125, 1.404052734375, 0.3545989990234375, 8.343963623046875, 3.84246826171875, 0.09839630126953125, 3.9423599243164062, -3.99267578125, -0.12295150756835938, 9.67724609375, 5.2359619140625, 10.843093872070312, 1.604736328125, 1.131317138671875, 3.2056350708007812, -13.189971923828125, 6.648529052734375, -0.593841552734375, -0.544921875, -2.0880126953125, -1.9251708984375, 5.66558837890625, 8.348251342773438, 5.706207275390625, 2.713268280029297, 4.463287353515625, 4.282958984375, -3.174072265625, -0.903594970703125, 0.907806396484375, 7.82183837890625, 0.30560302734375, 1.0173988342285156, 1.4670562744140625, 5.54278564453125, 8.2760009765625, -1.5320358276367188, 0.693145751953125, 0.7296829223632812, 7.132537841796875, 1.146026611328125, 4.753074645996094, 4.108367919921875, 0.8479881286621094, 18.46087646484375, -1.1539764404296875, 3.675323486328125, 7.7628173828125, -2.114501953125, 4.428497314453125, 9.593475341796875, 8.501754760742188, 6.626457214355469, 4.7484588623046875, -2.063568115234375, 1.168792724609375, -0.012411117553710938, 0.4759674072265625, -3.6477813720703125, 8.5069580078125, 18.9329833984375, -1.286407470703125, 7.51800537109375, 3.9862594604492188, -0.4271240234375, 10.06488037109375, 3.2694625854492188, 10.242523193359375, 3.186920166015625, -6.58172607421875, -4.26153564453125, -3.433013916015625, -1.9957427978515625, 5.506927490234375, 1.2829742431640625, 1.1371307373046875, 0.9754638671875, -3.27874755859375, 9.75830078125, -2.8769912719726562, -2.80938720703125, 1.0151214599609375, 3.64776611328125, 9.557968139648438, -9.15936279296875, -5.353363037109375, 4.0330352783203125, 2.3183364868164062, 1.8156280517578125, -1.5815505981445312, 3.0469207763671875, 5.0991363525390625, -4.40972900390625, -0.97540283203125, -7.387451171875, 8.93231201171875, 4.983917236328125, -6.0421142578125, 5.65478515625, -0.6516456604003906, 3.03143310546875, 6.223167419433594, 1.3558349609375, 0.257415771484375, 12.2991943359375, 11.6781005859375, -0.5129547119140625, 2.0770416259765625, -4.884857177734375, -5.4655303955078125, 16.0863037109375, 0.018951416015625, 3.641143798828125, -4.3989410400390625, 1.646636962890625, -2.1215744018554688, 0.02618408203125, 5.63250732421875, 2.071136474609375, 4.1084747314453125, 4.84954833984375, 5.609405517578125, 13.6923828125, -3.5844345092773438, 0.2535400390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000077.npy"}
{"epoch": 0.1612565445026178, "step": 78, "batch_size": 128, "mean": 2.836503505706787, "std": 5.230582237243652, "min": -12.964950561523438, "p10": -2.9833335876464844, "median": 2.442699432373047, "p90": 9.531832885742187, "max": 22.8262939453125, "pos_frac": 0.7265625, "sample": [-1.4913177490234375, 4.319244384765625, -5.0908203125, 5.0430908203125, 9.2098388671875, 2.258544921875, 3.393280029296875, 4.656005859375, -3.2908782958984375, 5.7363433837890625, 0.00791168212890625, 1.9349594116210938, 4.46173095703125, 3.609954833984375, -1.0656509399414062, 2.1639404296875, 3.5198211669921875, 5.93487548828125, 9.55377197265625, -0.1580810546875, 0.0, 9.19403076171875, -4.30340576171875, 2.926116943359375, 0.4111480712890625, 1.660614013671875, 5.565765380859375, 2.7574119567871094, -3.37652587890625, 5.2043914794921875, -2.9532241821289062, 5.62225341796875, 9.2772216796875, 0.8893585205078125, 7.02008056640625, -2.849517822265625, 0.9252777099609375, 14.261932373046875, 1.4957275390625, -5.07672119140625, 4.312568664550781, 4.20233154296875, -0.9604034423828125, 22.8262939453125, 12.3255615234375, -2.372161865234375, 2.6697540283203125, 3.728546142578125, 0.62713623046875, 4.3075103759765625, 7.06488037109375, -12.964950561523438, 1.37646484375, 3.44989013671875, -0.5273284912109375, -0.5407047271728516, 5.2619476318359375, 3.3777618408203125, 2.7759628295898438, -1.272491455078125, 10.66448974609375, -0.2196044921875, 4.4429473876953125, 6.2469024658203125, 2.436046600341797, -3.313201904296875, 1.1713485717773438, 2.4901123046875, 4.37371826171875, -10.417083740234375, 14.36474609375, 3.0447998046875, 2.50634765625, 1.9027099609375, 10.5050048828125, 6.893310546875, -0.10170936584472656, 2.3571701049804688, -1.3754501342773438, 0.9940948486328125, -4.571075439453125, -1.6054859161376953, 9.522430419921875, 15.45562744140625, 0.3558197021484375, 1.91864013671875, -0.06026458740234375, 2.9769287109375, -4.6296539306640625, 5.88543701171875, 2.9381752014160156, -2.6730804443359375, 7.9352874755859375, -0.3354949951171875, -7.998382568359375, 1.5941543579101562, 0.8707275390625, -2.8991546630859375, 10.2137451171875, 1.64788818359375, 0.3875770568847656, 1.4207344055175781, -1.7015838623046875, 9.172607421875, 5.7386474609375, 3.6509552001953125, 6.5816802978515625, -1.0399322509765625, 1.8065643310546875, 2.3734054565429688, 13.351470947265625, 0.760345458984375, 2.960479736328125, 6.0235443115234375, 3.172119140625, 2.449352264404297, 4.523834228515625, 10.5345458984375, 9.59698486328125, -3.8348541259765625, -3.0535888671875, 0.247344970703125, 17.921417236328125, 0.9569664001464844, 1.433685302734375, 3.5821151733398438, -1.593048095703125, 7.113037109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000078.npy"}
{"epoch": 0.16335078534031414, "step": 79, "batch_size": 128, "mean": 3.0607354640960693, "std": 6.954108715057373, "min": -40.166748046875, "p10": -3.337930297851562, "median": 2.123821258544922, "p90": 10.2615234375, "max": 25.37017822265625, "pos_frac": 0.6875, "sample": [-1.1627044677734375, 6.000083923339844, 9.760299682617188, -2.575286865234375, 5.75860595703125, -0.982513427734375, 2.0959625244140625, 0.2694244384765625, 2.8210601806640625, -3.43023681640625, 12.3941650390625, -1.6341400146484375, 0.0, 8.60723876953125, -0.1756725311279297, -0.7029037475585938, -0.51470947265625, -3.0542068481445312, -5.379692077636719, 0.19281005859375, 2.9217071533203125, -1.147186279296875, -3.298370361328125, -1.8497467041015625, -2.65106201171875, 4.20074462890625, 3.140869140625, 0.0, 9.609100341796875, 0.23483848571777344, -6.495574951171875, 11.972991943359375, 5.64581298828125, 6.089057922363281, 3.2868576049804688, 1.6804580688476562, 2.68133544921875, 1.2013397216796875, 2.877532958984375, -1.759857177734375, 5.029937744140625, -7.768440246582031, 9.00726318359375, 0.47845458984375, 0.3002471923828125, 6.6350555419921875, 6.23919677734375, 15.73681640625, 2.1363754272460938, 2.934051513671875, -0.334930419921875, -2.5388641357421875, -40.166748046875, -6.672088623046875, 2.11126708984375, 7.83013916015625, 0.03900146484375, 13.882659912109375, 9.501785278320312, 9.1209716796875, -0.772857666015625, 4.319793701171875, 2.177032470703125, 7.1692047119140625, 10.248046875, 0.73040771484375, 3.5253067016601562, 2.443317413330078, -1.7725067138671875, 0.576202392578125, -0.074249267578125, 4.467132568359375, 0.9712715148925781, 1.46038818359375, -4.94049072265625, -5.034027099609375, 8.9818115234375, -0.5640869140625, 9.5797119140625, 2.4118270874023438, -3.6176681518554688, 18.55316162109375, 10.29296875, 7.36181640625, 4.12261962890625, 8.70947265625, 4.4747772216796875, 0.25426483154296875, 0.000751495361328125, 1.410879135131836, 0.3496074676513672, 5.605113983154297, 0.6635513305664062, 0.842864990234375, 1.7551307678222656, 25.37017822265625, -0.5750598907470703, 4.7649078369140625, -3.6957244873046875, 9.348236083984375, 15.573150634765625, 10.14849853515625, 3.63153076171875, 4.4907379150390625, -4.332489013671875, 8.62076187133789, 7.93743896484375, -0.08019256591796875, 8.675247192382812, -1.2223739624023438, -3.6094970703125, 16.624786376953125, 0.182464599609375, -0.7068138122558594, 4.686243057250977, 6.414497375488281, 6.914306640625, 5.11346435546875, 1.20257568359375, 5.65301513671875, -0.5343990325927734, 11.14666748046875, 12.0643310546875, 20.74261474609375, -0.2607269287109375, -4.1779022216796875, 0.8950881958007812, 13.9554443359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000079.npy"}
{"epoch": 0.16544502617801046, "step": 80, "batch_size": 128, "mean": 2.582939624786377, "std": 6.195615768432617, "min": -23.13861083984375, "p10": -3.405010986328125, "median": 1.954803466796875, "p90": 11.068084716796875, "max": 23.5328369140625, "pos_frac": 0.6640625, "sample": [2.0692901611328125, 4.595458984375, -2.0998992919921875, 1.3370742797851562, 1.994110107421875, -0.9335174560546875, 1.4000701904296875, -5.266845703125, 6.22698974609375, -3.3701019287109375, 0.18042755126953125, -2.47528076171875, 6.9822998046875, 6.3629150390625, 4.99786376953125, -2.7039260864257812, -1.2444000244140625, 10.608306884765625, 3.5128402709960938, 4.5584716796875, 16.692169189453125, -0.8246231079101562, 2.120147705078125, 5.24066162109375, 0.388214111328125, 11.308258056640625, 5.494873046875, -3.4864654541015625, 3.513988494873047, 9.3887939453125, 3.059295654296875, 0.4218330383300781, 3.2314453125, -1.6380176544189453, -11.962081909179688, 1.26708984375, -23.13861083984375, 6.44854736328125, 0.0, 13.50555419921875, -1.88238525390625, 8.9835205078125, -0.94586181640625, -2.174041748046875, 1.763214111328125, -7.683837890625, -0.43255615234375, 6.43499755859375, -1.08770751953125, -1.7045707702636719, 1.4325370788574219, 1.54156494140625, 2.6139297485351562, 6.84381103515625, 8.6998291015625, 0.11046791076660156, 0.1798095703125, -0.366729736328125, -2.0287322998046875, 11.27850341796875, 4.589111328125, 1.003875732421875, 6.064029693603516, 3.067626953125, -4.170806884765625, 1.1848907470703125, 4.685882568359375, 0.1308746337890625, 6.321800231933594, 2.585947036743164, -2.4935760498046875, 4.2623443603515625, 3.140195846557617, -4.684604644775391, 17.98565673828125, -1.6929931640625, -0.8324661254882812, -0.20709991455078125, 5.7225341796875, -7.47369384765625, 13.582275390625, 3.8836517333984375, -0.471160888671875, 10.316864013671875, 4.79583740234375, 2.965728759765625, -5.4369964599609375, 1.915496826171875, -1.760284423828125, 15.160675048828125, 6.075920104980469, 2.3029937744140625, 12.645034790039062, 3.41485595703125, 4.320375442504883, 12.51104736328125, 7.9080657958984375, 1.1155242919921875, -10.004104614257812, 0.47144317626953125, -8.574462890625, 10.9779052734375, -3.582061767578125, 4.64581298828125, 2.79949951171875, 13.208740234375, 0.3900909423828125, -0.8320999145507812, 3.0776519775390625, 7.08740234375, -1.86572265625, -0.62530517578125, -0.25127410888671875, -0.8039703369140625, 1.183074951171875, -5.5690155029296875, 3.3758506774902344, -1.243560791015625, 1.1224422454833984, -2.9631195068359375, 6.8354644775390625, 10.218292236328125, 12.53289794921875, 3.5990142822265625, 23.5328369140625, 0.16652679443359375, 3.8453750610351562, 14.08624267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000080.npy"}
{"epoch": 0.16753926701570682, "step": 81, "batch_size": 128, "mean": 2.8234469890594482, "std": 6.631261825561523, "min": -19.38458251953125, "p10": -3.5557693481445307, "median": 2.513613700866699, "p90": 11.911993408203124, "max": 16.068695068359375, "pos_frac": 0.703125, "sample": [-1.292266845703125, 0.2510414123535156, 9.331642150878906, -11.568374633789062, 4.6920013427734375, -1.0462875366210938, -10.706939697265625, 3.0021743774414062, 16.068695068359375, 5.892974853515625, 15.792709350585938, 11.72509765625, -2.278900146484375, -3.3797454833984375, 7.857666015625, 0.17096328735351562, 4.22833251953125, 7.28521728515625, 4.4229888916015625, 3.2454833984375, 0.6354751586914062, 12.5223388671875, 0.116607666015625, -2.6584014892578125, 5.0872039794921875, 3.7598342895507812, 1.277130126953125, -0.4342193603515625, 8.779083251953125, 9.171661376953125, 13.22930908203125, 3.56805419921875, 8.272430419921875, 4.10693359375, -2.202056884765625, 3.144378662109375, -5.13787841796875, 15.62554931640625, 13.518646240234375, 2.8066253662109375, -4.64306640625, 0.70391845703125, 10.428436279296875, -0.0052032470703125, -0.6479034423828125, 2.148834228515625, -15.994720458984375, -1.085906982421875, 5.445648193359375, 9.84722900390625, 0.7509765625, 2.6288299560546875, -6.5820770263671875, 8.317535400390625, -4.344024658203125, -0.8408432006835938, 5.644378662109375, 5.0327911376953125, 1.9291229248046875, -14.74798583984375, -3.96649169921875, 3.208770751953125, 3.769622802734375, 1.6587677001953125, 0.04293251037597656, 4.104316711425781, 1.94390869140625, 0.27597808837890625, 3.2027587890625, 3.5152587890625, -0.2996368408203125, -1.9693527221679688, -1.207916259765625, 12.574493408203125, 13.3182373046875, 9.809600830078125, -1.136993408203125, 5.2208404541015625, 2.011251449584961, 5.853851318359375, -12.2789306640625, 6.140632629394531, 10.96270751953125, -3.2399444580078125, 1.23455810546875, 1.7492904663085938, -0.8669853210449219, -4.356964111328125, 0.07757568359375, 2.9162368774414062, -19.38458251953125, -1.178070068359375, -14.762664794921875, 0.715576171875, 2.513946533203125, 1.5384063720703125, -0.3050537109375, 0.06378555297851562, 15.42041015625, 13.621841430664062, 12.34808349609375, 13.48553466796875, 4.857627868652344, -2.072784423828125, 9.83941650390625, 2.5132808685302734, -0.858154296875, 1.6195068359375, 0.3886280059814453, 7.12371826171875, -3.220287322998047, 4.699928283691406, 0.26316070556640625, -0.1298828125, 2.883880615234375, 10.577049255371094, 8.452285766601562, -1.8263320922851562, 10.51617431640625, 4.689849853515625, 5.671600341796875, 10.94012451171875, 14.44183349609375, 10.215705871582031, 1.722412109375, 0.786376953125, 6.916526794433594, -0.825164794921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000081.npy"}
{"epoch": 0.16963350785340314, "step": 82, "batch_size": 128, "mean": 3.107499122619629, "std": 6.019745349884033, "min": -15.533355712890625, "p10": -4.032452392578125, "median": 3.2203826904296875, "p90": 11.056741333007812, "max": 23.66351318359375, "pos_frac": 0.7265625, "sample": [1.261016845703125, 5.264892578125, 0.15814208984375, -6.691558837890625, -1.1583099365234375, 3.262451171875, 2.08929443359375, 10.9595947265625, 7.6689300537109375, -0.0681304931640625, 1.3017120361328125, 7.237945556640625, -4.273078918457031, -1.752288818359375, -3.277099609375, 6.7393035888671875, 5.25, 3.29901123046875, 11.767868041992188, 6.4911956787109375, 1.1776962280273438, -2.913543701171875, 6.798912048339844, -1.2662811279296875, 0.42723846435546875, 11.935394287109375, -9.594390869140625, 4.78387451171875, 4.677642822265625, 3.9091796875, 18.89447021484375, 2.5367584228515625, 10.057342529296875, 4.320648193359375, 10.770660400390625, 13.20880126953125, -0.8786239624023438, 5.173187255859375, 4.36572265625, -15.533355712890625, -4.79461669921875, -0.1577606201171875, 2.1139068603515625, 3.622222900390625, -2.5462188720703125, 11.283416748046875, -1.3043212890625, 7.44921875, 16.73809814453125, -1.786468505859375, -1.067626953125, 20.153793334960938, 4.723407745361328, 4.4952239990234375, 2.25115966796875, 0.5907135009765625, 11.3172607421875, -3.99713134765625, -0.7021484375, 2.6526336669921875, 3.540313720703125, -4.88702392578125, -12.94781494140625, 3.5522327423095703, -4.470947265625, 0.6280593872070312, -4.1148681640625, 2.0143356323242188, 11.30718994140625, 4.3450775146484375, -0.5594959259033203, 7.7623138427734375, -0.29376220703125, 3.5721435546875, 0.08135986328125, 1.8151397705078125, 1.79376220703125, 1.4116935729980469, 4.8908843994140625, 2.5217208862304688, 1.3846244812011719, 23.66351318359375, -3.067230224609375, 8.41998291015625, 0.3179473876953125, 3.178314208984375, 5.49163818359375, 10.739871978759766, 2.327545166015625, 3.3041534423828125, 3.4757461547851562, -2.949249267578125, 3.3740005493164062, 4.652587890625, 7.316471099853516, 6.64404296875, -5.3050537109375, 2.0699310302734375, 5.296417236328125, 9.5108642578125, 6.088653564453125, -0.0727386474609375, 1.853729248046875, 4.7927398681640625, 6.932586669921875, 3.517486572265625, 0.42290496826171875, -0.762908935546875, -5.07037353515625, 1.516632080078125, -2.892181396484375, 3.69573974609375, 6.688205718994141, 8.487205505371094, -4.49859619140625, 12.03759765625, 13.59027099609375, 3.1128616333007812, 4.0630035400390625, 6.056640625, 4.683441162109375, 5.66546630859375, 12.693267822265625, 6.37286376953125, 0.7787017822265625, 0.6297721862792969, -10.67047119140625, -3.1780166625976562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000082.npy"}
{"epoch": 0.17172774869109947, "step": 83, "batch_size": 128, "mean": 3.5128588676452637, "std": 6.926236152648926, "min": -15.25604248046875, "p10": -3.4332298278808593, "median": 2.5459117889404297, "p90": 12.458987426757812, "max": 29.2843017578125, "pos_frac": 0.6875, "sample": [7.47686767578125, 0.0, 10.4052734375, 2.02557373046875, 3.70355224609375, -0.6529388427734375, -1.6467437744140625, 7.465423583984375, 0.90496826171875, 0.03780364990234375, -1.119476318359375, -15.25604248046875, 0.155487060546875, -11.977569580078125, -0.3905487060546875, 1.82379150390625, 4.4748687744140625, 9.873138427734375, 8.083106994628906, 2.4999961853027344, 8.978118896484375, 6.525840759277344, -0.9736785888671875, 4.1165924072265625, 12.5054931640625, -4.11749267578125, -1.596435546875, -0.8308296203613281, -1.642578125, 12.439056396484375, 6.445091247558594, 1.17010498046875, 4.42364501953125, 1.797698974609375, -7.1939697265625, 16.0504150390625, 3.6606674194335938, 2.92218017578125, 19.745071411132812, 2.1764373779296875, 4.496574401855469, -1.5716323852539062, 2.7162704467773438, 3.486480712890625, -3.5253143310546875, 11.87353515625, 5.202789306640625, 5.102317810058594, -1.4644775390625, -3.7509536743164062, 6.609100341796875, 4.4287109375, 19.4644775390625, 11.71044921875, -12.536529541015625, 1.7007904052734375, 10.49713134765625, 2.3028564453125, 29.2843017578125, 0.67156982421875, -3.4240493774414062, 11.841148376464844, 5.103851318359375, 6.538421630859375, 1.2496109008789062, -2.094818115234375, 0.93792724609375, -0.9134597778320312, -3.2136993408203125, 7.7629241943359375, -7.20599365234375, 7.1652069091796875, 0.27301788330078125, -0.8109893798828125, -3.4796142578125, 3.421356201171875, 15.839599609375, -3.2478485107421875, 3.49267578125, 8.561935424804688, -3.096771240234375, 13.65374755859375, 4.13299560546875, 4.5855712890625, -11.59979248046875, -5.13775634765625, 7.078929901123047, 13.79107666015625, -9.20684814453125, 10.5321044921875, -0.2788238525390625, -1.4029388427734375, 1.287506103515625, -2.5096435546875, 4.964801788330078, 15.345458984375, 16.0010986328125, 2.591827392578125, 2.052734375, 1.535736083984375, 12.972412109375, -1.22967529296875, 6.0689697265625, 2.4113922119140625, -2.059234619140625, 0.26671600341796875, 7.6203460693359375, 5.764739990234375, 3.255523681640625, -1.675140380859375, 9.095855712890625, -3.45465087890625, 8.354537963867188, 10.507049560546875, 1.634033203125, 4.1075439453125, 0.92669677734375, 17.280426025390625, 20.219329833984375, -3.1506195068359375, 1.7484169006347656, -0.2004547119140625, 9.605133056640625, -1.1655426025390625, 3.9850921630859375, 7.075050354003906, 4.2902984619140625, 2.087066650390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000083.npy"}
{"epoch": 0.17382198952879582, "step": 84, "batch_size": 128, "mean": 2.7905383110046387, "std": 7.070040225982666, "min": -22.46380615234375, "p10": -3.903207397460937, "median": 2.7456817626953125, "p90": 11.527346801757812, "max": 19.71270751953125, "pos_frac": 0.6796875, "sample": [-1.3256778717041016, 7.49456787109375, 17.1607666015625, -0.12548065185546875, -3.84368896484375, -2.7353668212890625, 6.59991455078125, 4.091102600097656, 5.501564025878906, -4.691925048828125, 1.5427322387695312, 5.58251953125, 1.030548095703125, -2.145782470703125, 3.4920654296875, 4.115943908691406, 6.163414001464844, 1.9481353759765625, -0.3007774353027344, 10.142990112304688, -3.1944808959960938, 7.314727783203125, 2.618560791015625, 3.87255859375, 8.554931640625, 1.437744140625, -7.7585296630859375, 5.732269287109375, 6.2584228515625, 1.96380615234375, -1.418212890625, -20.0208740234375, -22.46380615234375, 13.410888671875, 2.7855377197265625, -0.7480697631835938, 8.52777099609375, 2.3809890747070312, 2.967559814453125, -0.1125335693359375, 5.668060302734375, 1.2764892578125, 12.99346923828125, -3.659210205078125, 11.67169189453125, -2.46514892578125, 1.384521484375, -3.3562393188476562, 12.27923583984375, 3.5954132080078125, 19.3265380859375, -17.63983154296875, 9.373992919921875, 2.604217529296875, 3.7839279174804688, -4.527191162109375, 3.3914642333984375, -3.078216552734375, 7.0206756591796875, 7.847511291503906, 5.829376220703125, 3.7218170166015625, 0.7792205810546875, 5.081939697265625, 12.715728759765625, 1.757110595703125, 5.032379150390625, -0.33351898193359375, -1.836334228515625, 2.673553466796875, -3.3452911376953125, 3.76861572265625, 2.047760009765625, -0.08856201171875, 2.7058258056640625, 11.465484619140625, 8.09415054321289, 8.912834167480469, -11.751708984375, 3.823883056640625, 3.4487457275390625, 2.12451171875, -2.7219085693359375, 2.3540420532226562, -4.830169677734375, -0.46082496643066406, -2.22845458984375, 9.910797119140625, 16.922149658203125, -0.572265625, 11.252716064453125, 4.6524200439453125, -5.3870849609375, 4.7843017578125, -0.8205108642578125, -0.7129058837890625, 4.7089080810546875, 2.611663818359375, 3.53253173828125, 3.5369110107421875, 6.463226318359375, -1.0572357177734375, 2.070709228515625, 1.7703628540039062, 8.896648406982422, 5.527126312255859, -2.89862060546875, 19.71270751953125, -12.2122802734375, 10.50948715209961, 16.809326171875, 2.0096511840820312, 6.668998718261719, 3.08538818359375, 17.883270263671875, 8.615966796875, -6.8385009765625, -16.48907470703125, 5.363128662109375, -0.68109130859375, 0.16486740112304688, 11.840072631835938, 13.0521240234375, -3.559814453125, 4.167243957519531, 5.8441009521484375, 2.055145263671875, -4.042083740234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000084.npy"}
{"epoch": 0.17591623036649215, "step": 85, "batch_size": 128, "mean": 1.981445550918579, "std": 5.758843898773193, "min": -14.9237060546875, "p10": -5.676556396484374, "median": 2.4173736572265625, "p90": 9.20503234863281, "max": 15.710205078125, "pos_frac": 0.6484375, "sample": [4.1647796630859375, -2.478912353515625, -1.989471435546875, 15.710205078125, 6.75872802734375, 5.095703125, -0.9146270751953125, -2.328399658203125, 4.692901611328125, 4.0946044921875, 2.1971092224121094, -3.0996170043945312, 1.8466072082519531, 8.857536315917969, 6.586517333984375, 2.682708740234375, 6.476806640625, 5.4182891845703125, 5.8448028564453125, 3.2830352783203125, 1.061767578125, -0.34173583984375, 9.5225830078125, 13.284072875976562, 1.6466751098632812, -1.688873291015625, -0.533477783203125, -5.501800537109375, -7.215087890625, 4.029937744140625, 6.833282470703125, 4.030303955078125, -2.06378173828125, 5.615478515625, 9.691802978515625, 11.1214599609375, -4.947288513183594, 5.182098388671875, -0.17330169677734375, -2.11297607421875, 10.160430908203125, 5.681610107421875, -0.030731201171875, 5.2705078125, 10.0830078125, -14.9237060546875, 2.2815284729003906, 3.0459060668945312, 3.5269851684570312, 15.40740966796875, -0.78948974609375, -6.084320068359375, 5.063121795654297, 1.04010009765625, 2.405914306640625, -4.45947265625, 2.4288330078125, 1.4235763549804688, 5.1589508056640625, 1.634368896484375, -4.25335693359375, 4.163551330566406, 3.5682449340820312, -1.827362060546875, 7.71600341796875, 6.529449462890625, 6.195953369140625, 2.8355712890625, -5.267631530761719, -9.7027587890625, -2.893341064453125, 5.982002258300781, -8.54876708984375, -3.281494140625, -0.4317474365234375, -7.68048095703125, 10.372467041015625, 8.999755859375, 0.17527389526367188, 5.8427276611328125, 2.517913818359375, 9.068939208984375, 1.7780609130859375, 6.5742034912109375, 15.685089111328125, -12.00054931640625, -1.91790771484375, 8.181060791015625, 4.017181396484375, -4.7334747314453125, -10.266578674316406, 1.6144485473632812, 3.7713623046875, 3.197385787963867, 9.672637939453125, 2.3201465606689453, -1.4998779296875, 2.9543304443359375, -6.76214599609375, 4.233062744140625, -0.3140220642089844, 5.677581787109375, 0.6693153381347656, -1.660308837890625, -1.3567047119140625, 0.4420166015625, -6.367401123046875, 0.64764404296875, 9.6474609375, 3.4933242797851562, -6.0906829833984375, -7.7711181640625, 0.341827392578125, 6.219512939453125, -4.2632904052734375, 3.12225341796875, 6.626457214355469, 1.2883377075195312, 15.512619018554688, -7.339210510253906, -1.7721099853515625, 4.499481201171875, -5.00909423828125, 1.4761962890625, 3.8780670166015625, -2.0643482208251953, 3.0867462158203125, 5.440149307250977], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000085.npy"}
{"epoch": 0.17801047120418848, "step": 86, "batch_size": 128, "mean": 3.563296318054199, "std": 7.588063716888428, "min": -16.76495361328125, "p10": -4.694116210937498, "median": 3.6398725509643555, "p90": 12.770367431640624, "max": 24.42529296875, "pos_frac": 0.6875, "sample": [3.57720947265625, -11.86358642578125, 4.089874267578125, -7.67919921875, 4.763526916503906, 1.834686279296875, 5.1170654296875, 3.1424484252929688, 6.4528656005859375, 6.0424957275390625, 24.32415771484375, -3.8045654296875, 0.9799308776855469, 1.50885009765625, 7.42364501953125, -10.688505172729492, -11.6600341796875, -3.7743072509765625, 1.2524795532226562, -0.7034988403320312, 2.1986160278320312, 5.77668571472168, -1.8960838317871094, -0.7982635498046875, 14.123504638671875, -1.5867919921875, -14.45782470703125, 1.158447265625, 16.89501953125, 14.207229614257812, -8.200714111328125, 11.69940185546875, -0.1211395263671875, 9.3350830078125, 19.911865234375, 9.648773193359375, 13.232391357421875, -4.1124267578125, 1.3533935546875, 13.256103515625, 4.478096008300781, -2.8497467041015625, 9.27972412109375, -2.2080078125, 3.0867919921875, 0.419281005859375, 8.605823516845703, 12.1436767578125, 10.587615966796875, 4.903633117675781, -14.385589599609375, 10.43182373046875, 1.659149169921875, 7.3318634033203125, 4.959136962890625, -2.3798828125, 11.65985107421875, -16.76495361328125, 1.9228286743164062, 10.179168701171875, 22.04254150390625, 4.324188232421875, 4.1072998046875, 12.641006469726562, 7.76287841796875, -1.14300537109375, 0.0, -2.29229736328125, 3.2075424194335938, 6.373931884765625, 5.1894073486328125, 13.00531005859375, -11.34979248046875, 4.7199249267578125, 1.998342514038086, 1.2472763061523438, -0.658538818359375, -1.558349609375, 11.027175903320312, 9.578826904296875, -3.4569320678710938, 7.7188873291015625, 9.467132568359375, -0.5035552978515625, 4.878326416015625, -9.552093505859375, -10.759292602539062, 8.630599975585938, -2.14093017578125, 1.1050567626953125, -0.4378471374511719, 5.560302734375, 1.4303741455078125, 2.074085235595703, 4.40472412109375, -6.0513916015625, 5.913177490234375, 4.17718505859375, 3.7703857421875, 3.702535629272461, 0.696807861328125, -6.807373046875, -2.88739013671875, 11.2655029296875, 12.669677734375, 10.209625244140625, -1.513214111328125, 24.42529296875, 2.5993690490722656, 3.042266845703125, -2.2326622009277344, 6.7678680419921875, 13.51519775390625, 4.41705322265625, 12.30010986328125, -1.2530670166015625, 6.1581878662109375, 14.984207153320312, -0.4360198974609375, 6.028228759765625, 4.998321533203125, 9.752071380615234, 9.79705810546875, 13.27362060546875, -3.132904052734375, -0.874114990234375, 1.95611572265625, 1.2086029052734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000086.npy"}
{"epoch": 0.18010471204188483, "step": 87, "batch_size": 128, "mean": 4.162985801696777, "std": 6.7205963134765625, "min": -17.984161376953125, "p10": -2.4643241882324216, "median": 3.4552574157714844, "p90": 14.05802764892578, "max": 25.9488525390625, "pos_frac": 0.734375, "sample": [4.235969543457031, 9.109848022460938, 7.078887939453125, 9.651611328125, 16.106475830078125, 0.0, -0.2552032470703125, -2.3990859985351562, 9.279922485351562, 15.7711181640625, -4.726806640625, 7.539085388183594, -2.616546630859375, 9.140474319458008, 25.9488525390625, -0.09582901000976562, 0.6415939331054688, 7.2408447265625, 15.96405029296875, 4.587127685546875, -1.31787109375, -7.34625244140625, 0.966461181640625, 6.2867279052734375, 2.111358642578125, -1.7999420166015625, -0.0395355224609375, 5.47393798828125, 15.016754150390625, 0.8571109771728516, 9.51519775390625, -9.296417236328125, 6.762420654296875, 16.605712890625, 6.6223907470703125, 4.640815734863281, 4.355125427246094, 3.81671142578125, 4.41619873046875, 3.31622314453125, -0.8480930328369141, 0.02020263671875, 8.022216796875, -3.414398193359375, -2.0653305053710938, -1.1587371826171875, 9.3394775390625, -5.79132080078125, 10.930084228515625, 3.577667236328125, 7.019844055175781, 0.0702362060546875, 13.975341796875, 15.2447509765625, 11.00225830078125, 9.9912109375, 6.18719482421875, 8.116455078125, 9.587432861328125, 17.250091552734375, 0.85589599609375, 0.43548583984375, 2.0156402587890625, 10.552375793457031, 3.33868408203125, 23.1033935546875, -0.7487335205078125, 0.0, 4.80291748046875, 5.125732421875, 3.5718307495117188, 5.2995147705078125, 15.824371337890625, 5.8641204833984375, 14.250961303710938, 4.138336181640625, 4.2225341796875, -1.2418441772460938, 9.35931396484375, 0.5028228759765625, -8.265716552734375, -2.8310394287109375, -1.434173583984375, 0.7659912109375, -1.724822998046875, 10.4071044921875, -2.712860107421875, 6.210626602172852, 3.8327102661132812, 7.5511474609375, 4.16571044921875, 1.5694503784179688, 2.6956253051757812, -0.161224365234375, -0.01922607421875, -8.42279052734375, 1.6847763061523438, 1.3071842193603516, -6.477996826171875, 14.6800537109375, 0.1943511962890625, 2.76123046875, -0.4091987609863281, 1.536529541015625, 13.4947509765625, -17.984161376953125, 4.2872467041015625, -5.989349365234375, 0.14234352111816406, 1.6375274658203125, 12.35595703125, 16.734375, 0.526031494140625, 6.9338836669921875, -1.07769775390625, 0.0680999755859375, 10.20953369140625, 2.6651039123535156, 2.720947265625, 0.5030975341796875, 3.87628173828125, -0.4514617919921875, -2.1988067626953125, 2.842531204223633, 5.8418121337890625, 2.1324920654296875, 13.940567016601562, 1.252187728881836], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000087.npy"}
{"epoch": 0.18219895287958116, "step": 88, "batch_size": 128, "mean": 3.2734498977661133, "std": 7.416155815124512, "min": -27.26947021484375, "p10": -3.5429168701171876, "median": 1.9829349517822266, "p90": 13.093527221679688, "max": 25.601318359375, "pos_frac": 0.6875, "sample": [8.885299682617188, 15.286956787109375, 13.11187744140625, 0.6022529602050781, 17.798248291015625, 1.5984573364257812, -2.4097137451171875, 0.7734451293945312, 7.146148681640625, 2.414600372314453, 1.9657173156738281, -3.5174102783203125, 3.5285186767578125, 21.307952880859375, 0.02042388916015625, -1.71771240234375, 0.3371257781982422, 8.1661376953125, 0.934112548828125, -3.012451171875, 12.845733642578125, -4.589752197265625, 9.159332275390625, -3.6024322509765625, -1.162109375, 6.962860107421875, -5.22650146484375, 0.3584442138671875, 15.74920654296875, -6.15216064453125, -7.97210693359375, 9.998207092285156, 4.0856170654296875, 5.6837158203125, 2.5214385986328125, -1.2358818054199219, 20.073211669921875, 0.0, 0.09176063537597656, 1.877532958984375, 7.40509033203125, 4.025299072265625, 10.172134399414062, -0.65618896484375, 13.8668212890625, 13.085662841796875, 2.01922607421875, -6.4659271240234375, 2.000152587890625, 3.837066650390625, 0.0, 2.819488525390625, 3.510467529296875, 6.03948974609375, 13.313140869140625, 0.8988494873046875, -0.7627906799316406, 1.10052490234375, -1.109893798828125, 5.927215576171875, 9.6253662109375, -0.5381908416748047, 0.4686431884765625, -5.556068420410156, 5.771247863769531, 2.099212646484375, -3.993682861328125, 11.193634033203125, 0.09783935546875, 12.04248046875, 2.8955078125, -4.0225830078125, 12.70068359375, 0.3035888671875, -1.707244873046875, -21.645843505859375, 4.7473907470703125, 1.4927253723144531, -0.0077972412109375, -2.8659286499023438, 6.9216461181640625, 7.8198394775390625, 8.52197265625, -2.081899642944336, 1.35052490234375, -3.21112060546875, -2.5719375610351562, -7.52618408203125, 5.2483062744140625, 7.579986572265625, 0.35619163513183594, 4.23907470703125, 1.7877655029296875, 2.94189453125, 5.469146728515625, 15.973297119140625, 1.75091552734375, -2.0313720703125, -1.0627593994140625, 20.08819580078125, 4.08544921875, 3.3291397094726562, -1.35467529296875, 10.109161376953125, -1.8218994140625, 25.601318359375, 14.7880859375, -0.12484169006347656, 8.995941162109375, -7.44854736328125, 7.4095001220703125, 0.2754364013671875, -0.13592529296875, 3.646331787109375, 21.70831298828125, 3.7908363342285156, -1.0625133514404297, 0.36812591552734375, 2.518524169921875, 0.459228515625, -27.26947021484375, 5.805908203125, -3.156005859375, 7.70977783203125, 1.9239959716796875, 6.8248291015625, 4.03375244140625, -0.38455963134765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000088.npy"}
{"epoch": 0.18429319371727748, "step": 89, "batch_size": 128, "mean": 3.873971939086914, "std": 6.509339332580566, "min": -16.59600830078125, "p10": -2.9415946960449215, "median": 3.5538101196289062, "p90": 12.145318603515625, "max": 24.512451171875, "pos_frac": 0.78125, "sample": [0.5283889770507812, -9.845840454101562, 13.50311279296875, 2.068878173828125, 5.889007568359375, 15.52716064453125, -1.2819499969482422, 1.3499603271484375, 2.4599761962890625, 9.466537475585938, 3.3355712890625, 3.9402618408203125, 3.542083740234375, 12.1646728515625, 2.7537994384765625, 12.020889282226562, 1.4859333038330078, -3.2296295166015625, 5.4431610107421875, 3.0929718017578125, 4.3973388671875, 12.779571533203125, -2.8546218872070312, 0.6252326965332031, 1.323486328125, 2.0278549194335938, 4.668212890625, -0.7755126953125, -1.39825439453125, 3.5892868041992188, 2.5211334228515625, 0.9796676635742188, -0.286590576171875, 12.9483642578125, 13.202972412109375, 24.512451171875, 11.9656982421875, -7.093170166015625, 2.3983612060546875, 3.950042724609375, -16.59600830078125, -2.12005615234375, 3.68218994140625, -8.91436767578125, -1.303619384765625, 5.23486328125, 7.182464599609375, -11.399444580078125, 1.2705230712890625, -2.4793853759765625, -0.3368206024169922, 7.643341064453125, 3.64453125, -0.1782989501953125, 9.183380126953125, 10.008880615234375, 5.732757568359375, -3.5226306915283203, 7.007804870605469, 10.237823486328125, 2.180805206298828, 14.642364501953125, 7.693939208984375, 8.01239013671875, 6.85089111328125, 1.2150192260742188, -7.03448486328125, 1.198577880859375, 1.1614875793457031, -3.626953125, 4.4214630126953125, 5.815399169921875, 4.361114501953125, 7.0890045166015625, -5.2343292236328125, -1.089019775390625, 3.6088905334472656, 6.8958740234375, 22.32891845703125, -5.843719482421875, 1.139617919921875, 5.8290252685546875, 3.4282913208007812, 8.211380004882812, 0.707122802734375, 10.642547607421875, 3.0331954956054688, 14.04754638671875, 9.139801025390625, 19.23974609375, 5.753776550292969, -2.0340576171875, 12.294090270996094, 1.8194999694824219, 2.602497100830078, 3.67510986328125, 24.45318603515625, -1.2519454956054688, 1.9111251831054688, 3.5793228149414062, 7.18243408203125, 1.586029052734375, 3.5655364990234375, 0.391326904296875, 3.870809555053711, 4.886375427246094, -2.3708114624023438, -0.7382965087890625, 2.75146484375, 4.2086181640625, 3.91082763671875, 3.8177337646484375, 9.806655883789062, 12.13702392578125, 6.58349609375, 4.0934906005859375, 4.85113525390625, 2.845855712890625, 8.157699584960938, -12.14898681640625, 1.8159599304199219, 12.033294677734375, 5.6421661376953125, 1.5088043212890625, 2.3030242919921875, 3.379791259765625, 2.3985748291015625, -3.14453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000089.npy"}
{"epoch": 0.18638743455497384, "step": 90, "batch_size": 128, "mean": 3.3384389877319336, "std": 7.251960277557373, "min": -20.30120849609375, "p10": -4.794340515136718, "median": 2.563678741455078, "p90": 11.864056396484374, "max": 24.65045166015625, "pos_frac": 0.6953125, "sample": [8.907318115234375, 4.18316650390625, 1.8025054931640625, 19.0106201171875, -10.324722290039062, 0.5885009765625, -3.494354248046875, 1.0691375732421875, -2.61260986328125, 11.4674072265625, 2.1963424682617188, 2.8456802368164062, -8.08026123046875, -0.6238441467285156, -1.3583984375, 6.19384765625, 22.120468139648438, 18.824127197265625, 10.43963623046875, 1.9133758544921875, 7.758026123046875, -0.165863037109375, 3.3897247314453125, 2.7266082763671875, -12.26837158203125, -4.6033477783203125, 11.437313079833984, 7.482421875, 5.84893798828125, 3.40142822265625, -1.408782958984375, 6.324371337890625, 8.277664184570312, 2.8758544921875, 8.072311401367188, 2.6641082763671875, -4.267364501953125, -5.417327880859375, 0.42626953125, 5.842185974121094, -5.267299652099609, 2.6652984619140625, -0.8193359375, -11.13385009765625, 7.597991943359375, 11.806396484375, 4.166908264160156, 7.7985076904296875, 5.8572845458984375, 1.653167724609375, 7.5749053955078125, -3.5993576049804688, 0.3067626953125, 1.79705810546875, 0.54901123046875, 0.5482635498046875, -9.3428955078125, -0.03627777099609375, 0.42256927490234375, 6.266727447509766, 1.436981201171875, 0.0, 11.40399169921875, 6.5125885009765625, 10.231048583984375, -1.634033203125, 8.6370849609375, 3.227874755859375, -0.87127685546875, 11.12158203125, 1.1089706420898438, 1.807525634765625, 6.62188720703125, 15.47650146484375, 12.412445068359375, 1.114776611328125, 0.27651405334472656, 3.8625717163085938, 3.4182662963867188, 6.7578125, 11.23046875, -20.30120849609375, 9.40240478515625, -10.1224365234375, -0.5871353149414062, 3.1883087158203125, 4.268306732177734, 2.4632492065429688, -0.613525390625, -3.694244384765625, -0.662139892578125, -5.5740966796875, 10.869476318359375, 11.6500244140625, -2.982513427734375, 0.87060546875, -4.4218597412109375, 5.873779296875, 8.78497314453125, 11.5576171875, -2.13873291015625, 11.076484680175781, 9.67169189453125, 0.247833251953125, -1.7900848388671875, 24.65045166015625, -1.4322052001953125, 4.343292236328125, -6.02740478515625, -13.02215576171875, 4.4986114501953125, 1.3188018798828125, -5.239990234375, 13.96026611328125, 1.376312255859375, -0.3749217987060547, 11.99859619140625, 1.90777587890625, 13.557601928710938, 1.50634765625, -0.5819244384765625, -2.5002384185791016, 20.604248046875, 12.351837158203125, 3.1219959259033203, 12.9002685546875, 1.1756591796875, 13.760650634765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000090.npy"}
{"epoch": 0.18848167539267016, "step": 91, "batch_size": 128, "mean": 3.3870153427124023, "std": 7.652322292327881, "min": -13.062286376953125, "p10": -5.380777740478515, "median": 3.349517822265625, "p90": 13.961373901367187, "max": 25.3360595703125, "pos_frac": 0.625, "sample": [3.62664794921875, 1.1749153137207031, -1.1366043090820312, 14.789413452148438, 7.901275634765625, 14.259735107421875, -0.07528877258300781, 3.340576171875, 4.982391357421875, -2.7467727661132812, 1.754364013671875, 5.4080963134765625, 6.295265197753906, 7.024345397949219, 6.143592834472656, 7.879669189453125, 1.3127593994140625, -13.062286376953125, -4.2576446533203125, 0.12491798400878906, 7.9082489013671875, 3.603790283203125, 5.390510559082031, 7.573089599609375, -1.7773723602294922, 6.35003662109375, 0.35709190368652344, -1.07366943359375, -0.8580436706542969, -7.4360504150390625, 12.616180419921875, -0.3686981201171875, -2.267578125, -5.641326904296875, -3.7221145629882812, 8.837982177734375, -5.309547424316406, 6.46044921875, 6.033241271972656, 6.2603912353515625, -9.21484375, -0.74114990234375, 4.4321441650390625, -1.863555908203125, 4.04608154296875, 20.0919189453125, -4.98199462890625, -4.7951202392578125, 0.0, -11.503707885742188, -5.2572479248046875, -4.3567047119140625, -10.764923095703125, 2.5178070068359375, 6.678955078125, 5.920867919921875, -0.0006103515625, -5.5469818115234375, 14.212188720703125, 5.2328948974609375, 9.29742431640625, -8.460540771484375, 4.3895416259765625, -0.048412322998046875, -2.5270843505859375, -1.499237060546875, 25.3360595703125, -2.16168212890625, -0.4230194091796875, 10.610488891601562, -7.25787353515625, -4.583831787109375, 25.09039306640625, 3.983165740966797, 15.332489013671875, 0.7707862854003906, 4.4265594482421875, 1.0256977081298828, 4.186927795410156, 14.29052734375, -8.621772766113281, -4.22747802734375, 22.583892822265625, 10.051513671875, 12.43621826171875, 0.0, 8.708450317382812, 4.2584381103515625, 8.36749267578125, 13.8538818359375, 3.7594070434570312, 9.28125, 4.111572265625, -7.183837890625, -0.02093505859375, 13.539337158203125, -8.392913818359375, 2.53814697265625, 3.35845947265625, 20.40435791015625, 8.213592529296875, 8.996902465820312, 7.758941650390625, 3.538116455078125, 4.47607421875, -10.628379821777344, -2.535247802734375, 11.772598266601562, -1.70709228515625, 18.3822021484375, -3.39056396484375, 17.305419921875, 20.05029296875, -0.168121337890625, -1.5335235595703125, 10.665054321289062, 1.6038627624511719, -1.727020263671875, 5.339630126953125, 1.0172119140625, 9.536605834960938, -3.228057861328125, 1.8357391357421875, 1.6759624481201172, 4.80621337890625, 2.01025390625, 6.5059967041015625, 0.5994491577148438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000091.npy"}
{"epoch": 0.1905759162303665, "step": 92, "batch_size": 128, "mean": 3.746002197265625, "std": 8.030708312988281, "min": -27.63079833984375, "p10": -5.279675292968749, "median": 3.5588912963867188, "p90": 13.807017517089843, "max": 21.410003662109375, "pos_frac": 0.71875, "sample": [-7.06768798828125, 0.18482208251953125, -4.48419189453125, 3.447418212890625, 2.576751708984375, 5.2451629638671875, 5.4244842529296875, 5.86334228515625, 13.390716552734375, 6.1361846923828125, 1.5211639404296875, 2.3645248413085938, 4.36566162109375, -0.701080322265625, 19.587188720703125, 21.288330078125, 19.66912841796875, -7.64013671875, 7.4192352294921875, 1.9830780029296875, -6.259223937988281, 7.70849609375, 14.9189453125, 12.889251708984375, 1.958251953125, -2.8189544677734375, 2.545867919921875, 13.664337158203125, 1.480926513671875, -0.8160858154296875, 4.6495361328125, 0.7200775146484375, 12.207305908203125, 6.09027099609375, 0.8644561767578125, 0.13385009765625, 0.6180419921875, 5.7132415771484375, 14.139938354492188, 7.40533447265625, 1.828887939453125, 1.299530029296875, 0.2222766876220703, -27.63079833984375, 9.797500610351562, 3.47784423828125, 5.352447509765625, 2.618682861328125, 0.675079345703125, 1.7601890563964844, -4.932708740234375, -8.859100341796875, -9.596542358398438, -1.071502685546875, 17.49456787109375, -5.000091552734375, -10.613922119140625, -0.312896728515625, -7.327789306640625, 6.431640625, 5.20379638671875, 6.7484130859375, 9.000732421875, 5.121337890625, 6.760772705078125, 4.064971923828125, 9.78311538696289, 0.0, -2.1942138671875, 13.29388427734375, -2.8850250244140625, 3.81219482421875, 0.61419677734375, -0.5668869018554688, -3.2160491943359375, 8.124404907226562, -1.0761833190917969, 1.185302734375, 10.930404663085938, -0.891474723815918, -8.163360595703125, 13.4813232421875, 7.0896148681640625, -21.993377685546875, -2.1946983337402344, 0.40557861328125, 6.336090087890625, 19.85919189453125, 20.72003173828125, 6.381355285644531, 7.8559112548828125, 6.58636474609375, -0.24149131774902344, -0.75506591796875, -0.29017162322998047, 21.410003662109375, -0.748016357421875, -5.932037353515625, 20.10443115234375, 5.1276397705078125, 10.362579345703125, 9.732330322265625, -11.8140869140625, 14.853240966796875, 8.242889404296875, 4.5577545166015625, 2.989044189453125, 3.8355560302734375, 3.1863861083984375, 4.191314697265625, 10.24688720703125, 4.531883239746094, 9.137451171875, -10.844573974609375, 5.288421630859375, 7.96282958984375, 0.507080078125, 0.9558258056640625, -1.761138916015625, 7.490447998046875, -4.41522216796875, 9.119735717773438, 3.6399383544921875, 2.3079891204833984, 16.48956298828125, 17.83148193359375, 9.239120483398438, -1.2307357788085938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000092.npy"}
{"epoch": 0.19267015706806281, "step": 93, "batch_size": 128, "mean": 3.6038029193878174, "std": 7.4708452224731445, "min": -18.03411865234375, "p10": -4.758338928222656, "median": 2.6118850708007812, "p90": 14.539135742187499, "max": 24.830810546875, "pos_frac": 0.6875, "sample": [0.03487396240234375, 14.75250244140625, 1.3000946044921875, 21.87591552734375, -4.8227996826171875, 8.320594787597656, -2.6318283081054688, 12.94189453125, 5.5780029296875, 4.839544296264648, 3.309234619140625, -7.7235260009765625, 1.84344482421875, 2.55352783203125, 4.1960906982421875, 12.25714111328125, 2.6373291015625, 18.124114990234375, -3.760711669921875, 5.298789978027344, 5.7051544189453125, 4.9422149658203125, 3.2842540740966797, 3.9480743408203125, 14.44769287109375, 1.679595947265625, -0.7711639404296875, 2.50823974609375, 11.20965576171875, 10.723358154296875, 2.7968826293945312, -7.424224853515625, 1.1174430847167969, -1.4894561767578125, 2.5864410400390625, 6.9658355712890625, 4.427143096923828, 4.9930419921875, -4.8466796875, 13.57073974609375, 2.4107666015625, 6.62451171875, 3.3085479736328125, -4.730712890625, -0.5321044921875, 4.10369873046875, 0.356597900390625, 15.183624267578125, 1.84063720703125, -1.3567619323730469, 1.5375518798828125, 5.648468017578125, 16.49951171875, -5.317230224609375, 9.03997802734375, 1.312957763671875, -1.0936203002929688, -0.9399795532226562, 10.85504150390625, 0.7823944091796875, 7.65570068359375, 4.602264404296875, 15.88818359375, -3.1267242431640625, -1.4556121826171875, -3.7176513671875, 9.801025390625, 16.49176025390625, -6.9180755615234375, 1.3158111572265625, 3.7639846801757812, 3.2110652923583984, -4.477973937988281, -1.3808822631835938, 7.721778869628906, -2.6223297119140625, 14.286224365234375, 0.276458740234375, 20.00390625, -18.03411865234375, 20.72674560546875, 8.21368408203125, 16.38507080078125, -7.15740966796875, 6.6424713134765625, 1.463134765625, 0.0, -9.86737060546875, 2.0081825256347656, 4.334877014160156, 18.621673583984375, 11.4066162109375, -3.83026123046875, 2.714019775390625, 5.817840576171875, -12.9737548828125, 7.1747894287109375, -0.9504146575927734, -0.216522216796875, -6.666860580444336, 9.731048583984375, 6.3648223876953125, 3.293548583984375, 2.80474853515625, 5.3352203369140625, 0.0, 1.8323211669921875, -2.0396652221679688, -5.722053527832031, -0.6682891845703125, -0.6822891235351562, 0.3258953094482422, 6.567192077636719, 2.5433120727539062, -11.969970703125, 22.27056884765625, 24.830810546875, -2.758453369140625, -1.8722457885742188, -1.3968963623046875, 4.211029052734375, 2.66912841796875, -0.26708984375, 13.1214599609375, 0.6169643402099609, 1.71405029296875, 1.3098526000976562, 5.1580810546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000093.npy"}
{"epoch": 0.19476439790575917, "step": 94, "batch_size": 128, "mean": 4.159952640533447, "std": 9.072989463806152, "min": -19.431304931640625, "p10": -6.965495300292969, "median": 2.876593589782715, "p90": 15.978341674804687, "max": 29.79437255859375, "pos_frac": 0.6953125, "sample": [-1.4859161376953125, 9.94083023071289, -7.3878021240234375, 17.204620361328125, 5.357391357421875, 10.97540283203125, 19.657958984375, 2.722555160522461, 4.202545166015625, 9.478317260742188, 2.1739501953125, -4.437675476074219, -13.768768310546875, 0.3212928771972656, 11.709487915039062, 0.7353515625, 14.023529052734375, 0.9016799926757812, -0.2345733642578125, 0.52032470703125, 13.55291748046875, 14.543548583984375, 4.306884765625, -2.2364425659179688, 19.939239501953125, 13.63250732421875, 2.28265380859375, 4.6182861328125, -10.141021728515625, -2.2071533203125, 12.8004150390625, 22.0308837890625, -12.250701904296875, 24.29681396484375, 7.5469970703125, 7.098358154296875, 7.5628662109375, 1.7957134246826172, -5.953338623046875, 21.26031494140625, 5.0531463623046875, 29.79437255859375, 14.169189453125, -0.6319961547851562, 16.0089111328125, 4.95648193359375, 13.659515380859375, -6.45855712890625, 7.340461730957031, -2.4375228881835938, 1.390960693359375, 0.7965011596679688, -0.8389415740966797, 6.657867431640625, 3.0306320190429688, -6.14617919921875, 8.548721313476562, -2.93731689453125, 8.69354248046875, -9.54412841796875, -0.3720703125, -2.9544677734375, 2.3239974975585938, -3.82086181640625, 2.126983642578125, 12.608551025390625, 16.17828369140625, 1.10107421875, -6.9353790283203125, 11.434051513671875, 0.74310302734375, 1.3769340515136719, 7.13372802734375, 13.247329711914062, 19.158447265625, -6.6590576171875, 5.829437255859375, -1.6080894470214844, 4.263671875, 5.34698486328125, 3.9437408447265625, -0.557373046875, -5.854118347167969, -4.441253662109375, -11.5557861328125, 22.148406982421875, 13.139556884765625, -7.63800048828125, 15.965240478515625, -11.784820556640625, -19.431304931640625, 5.3556365966796875, 2.4612159729003906, 9.047748565673828, 5.1858367919921875, 12.94775390625, 8.1463623046875, 9.272506713867188, 17.641571044921875, -7.254669189453125, 6.74420166015625, -6.507171630859375, 12.070159912109375, 1.6718559265136719, 3.0377960205078125, 7.98309326171875, 2.1469039916992188, 9.800773620605469, -0.00726318359375, 1.4410781860351562, 4.0965576171875, 5.7430419921875, 1.5283355712890625, -17.6236572265625, -0.26904296875, 0.91851806640625, 8.929229736328125, 2.677734375, 1.6318588256835938, -7.0357666015625, 9.3480224609375, 0.062957763671875, 27.336181640625, 11.940658569335938, -10.771087646484375, 1.47137451171875, -0.2596282958984375, -1.089599609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000094.npy"}
{"epoch": 0.1968586387434555, "step": 95, "batch_size": 128, "mean": 4.5131425857543945, "std": 9.15200138092041, "min": -25.548675537109375, "p10": -6.973925781249998, "median": 4.288768768310547, "p90": 17.35021667480468, "max": 27.306304931640625, "pos_frac": 0.7109375, "sample": [-3.0222835540771484, -0.8602294921875, 3.5765762329101562, 24.450225830078125, 7.8797607421875, 3.4652099609375, 1.5229644775390625, 8.29534912109375, 25.333038330078125, 8.07891845703125, -0.478912353515625, 0.308990478515625, 2.4035263061523438, 3.569263458251953, -8.987045288085938, 6.177032470703125, 3.26593017578125, 6.3303375244140625, 11.628875732421875, 1.5851898193359375, 12.572402954101562, 6.232398986816406, -12.25927734375, 8.629150390625, -0.55743408203125, -1.3325042724609375, 27.306304931640625, -3.644927978515625, 16.68890380859375, -25.548675537109375, 19.241943359375, -1.89691162109375, 9.457508087158203, -6.07427978515625, 24.842041015625, 11.6815185546875, 3.09136962890625, 6.9916229248046875, 2.86590576171875, 10.639480590820312, 3.9650802612304688, 6.8438720703125, -3.773468017578125, -0.21557998657226562, -0.00604248046875, 7.807861328125, 4.612457275390625, 9.390487670898438, 21.19232177734375, 2.674530029296875, -0.6443634033203125, 8.990005493164062, 5.8100433349609375, 0.2861919403076172, 6.29754638671875, 8.397430419921875, 4.817619323730469, 4.772529602050781, 6.6730499267578125, 10.791015625, 7.85888671875, -1.843597412109375, 1.99822998046875, 1.301361083984375, 3.1508026123046875, 7.202667236328125, -4.34759521484375, 12.239959716796875, 5.131874084472656, 3.76495361328125, -6.5877838134765625, 19.77227783203125, 8.96258544921875, 18.893280029296875, 0.445831298828125, -1.8126373291015625, -0.07367897033691406, 15.956787109375, 2.7247772216796875, 5.7543487548828125, -8.408416748046875, 10.010601043701172, 3.9070587158203125, 1.660552978515625, 5.4174346923828125, 22.611572265625, 6.5448150634765625, 8.667526245117188, -2.62640380859375, -14.950958251953125, 4.930900573730469, 6.7710723876953125, -2.633758544921875, 1.4298858642578125, -14.38916015625, 2.57025146484375, 8.765380859375, 5.56768798828125, -8.282958984375, 4.834014892578125, -3.7842559814453125, -2.7358837127685547, -7.8749237060546875, 1.8761978149414062, 11.148677825927734, -8.668464660644531, 4.678653717041016, -0.595245361328125, 24.530548095703125, -11.119171142578125, 3.8115234375, 6.23583984375, -0.6267242431640625, 10.00665283203125, 1.6603851318359375, 15.464141845703125, 1.702392578125, 9.291671752929688, 6.0611572265625, -12.9132080078125, -8.32232666015625, -0.41021156311035156, 6.0185394287109375, 21.39190673828125, 4.776397705078125, 25.497100830078125, -10.7025146484375, 26.2611083984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000095.npy"}
{"epoch": 0.19895287958115182, "step": 96, "batch_size": 128, "mean": 4.337155342102051, "std": 8.377815246582031, "min": -19.238525390625, "p10": -4.491148376464843, "median": 3.67608642578125, "p90": 15.975607299804684, "max": 30.138916015625, "pos_frac": 0.703125, "sample": [-2.424041748046875, -10.126220703125, -7.43896484375, 1.7549591064453125, 16.7557373046875, 9.022048950195312, -13.338119506835938, -4.69561767578125, 2.426666259765625, 2.580047607421875, 3.6941299438476562, -4.4035186767578125, -6.421600341796875, -1.646881103515625, 8.937797546386719, 2.019002914428711, 4.06842041015625, -6.69866943359375, 1.240692138671875, 4.535400390625, 7.9805908203125, 2.930267333984375, 9.689361572265625, 6.2031707763671875, -3.653961181640625, 3.559722900390625, 18.039398193359375, 11.155242919921875, 1.9920196533203125, 5.7388916015625, 26.478790283203125, 4.818206787109375, 10.661590576171875, 11.037078857421875, 6.4358367919921875, 3.4618759155273438, 4.040534973144531, 4.0652923583984375, 1.499542236328125, 0.8424072265625, 7.0406646728515625, 4.6490478515625, -2.999298095703125, 19.36883544921875, -5.359344482421875, 5.92413330078125, -8.111228942871094, -0.5382156372070312, 3.6580429077148438, -0.37896728515625, 5.9490814208984375, -0.759002685546875, 8.99798583984375, -3.4856414794921875, 5.1353759765625, -2.24127197265625, 5.334800720214844, 14.544921875, 4.341522216796875, 12.903045654296875, -1.3306655883789062, 2.6045989990234375, 13.201171875, 11.051666259765625, 6.014129638671875, 20.9979248046875, 30.138916015625, 5.67205810546875, -0.715728759765625, 15.641265869140625, 3.825347900390625, 14.902839660644531, 1.9623489379882812, -10.870162963867188, 24.9925537109375, 4.383636474609375, 9.459213256835938, -2.030181884765625, 18.5281982421875, 0.8412628173828125, 8.122055053710938, 1.70721435546875, 4.9461822509765625, -7.10162353515625, 0.9645843505859375, -1.6780128479003906, 0.399627685546875, 10.5804443359375, 17.643280029296875, 0.2548980712890625, -1.1448650360107422, -0.6480636596679688, 4.53489875793457, 18.754364013671875, 17.791259765625, -6.603031158447266, 4.286346435546875, 7.691997528076172, 10.222803115844727, 13.61920166015625, -17.287933349609375, 19.838775634765625, 2.2172698974609375, -1.7896404266357422, -0.28057098388671875, 10.3001708984375, -0.532958984375, 26.15472412109375, 1.0514373779296875, 3.44488525390625, 9.91021728515625, 0.0, 1.129058837890625, 2.941638946533203, 0.48016357421875, 4.6558074951171875, -2.97515869140625, 7.3260498046875, 5.2860107421875, 10.82025146484375, -4.320930480957031, 7.179420471191406, -2.3894805908203125, -2.96453857421875, 2.5610198974609375, 8.361358642578125, -1.1281585693359375, -19.238525390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000096.npy"}
{"epoch": 0.20104712041884817, "step": 97, "batch_size": 128, "mean": 4.6048736572265625, "std": 10.052743911743164, "min": -16.414886474609375, "p10": -7.524682235717774, "median": 3.917388916015625, "p90": 17.34297180175781, "max": 31.3763427734375, "pos_frac": 0.6875, "sample": [5.518882751464844, 5.48809814453125, 8.052108764648438, -7.181343078613281, 7.990264892578125, 2.1325149536132812, -1.6643409729003906, 6.1254425048828125, 9.24468994140625, 8.75701904296875, -1.4275684356689453, 30.6087646484375, 10.5980224609375, 14.972869873046875, 0.39654541015625, 3.9915313720703125, 16.896881103515625, -0.9084014892578125, -5.06903076171875, 11.709625244140625, 3.8432464599609375, 2.3639450073242188, -3.2996826171875, -1.632598876953125, 9.437347412109375, -8.8997802734375, 3.745147705078125, -6.070075988769531, 8.651729583740234, 9.276092529296875, -7.4973297119140625, 1.1220321655273438, -4.1141204833984375, -8.932220458984375, 7.1278228759765625, 26.502120971679688, 10.847366333007812, -5.1998138427734375, -6.44659423828125, -11.241119384765625, 2.50665283203125, 0.93780517578125, 3.4233245849609375, 23.2689208984375, -7.9453277587890625, 6.193603515625, 5.656280517578125, 9.518058776855469, 1.007110595703125, 9.539230346679688, 13.239578247070312, -7.11395263671875, 24.91595458984375, -1.541778564453125, 8.143035888671875, 11.711776733398438, -7.0743408203125, 8.074905395507812, 1.6614990234375, 0.3427734375, 9.443672180175781, 6.3067169189453125, -1.6929473876953125, 8.010112762451172, 5.6990509033203125, 0.0, 28.84619140625, 8.684310913085938, 9.339935302734375, 31.3763427734375, 1.1170101165771484, 11.315963745117188, 9.158920288085938, 5.6882476806640625, -0.6125869750976562, 28.81915283203125, 3.765380859375, -12.97052001953125, -7.588504791259766, 18.55914306640625, 7.50128173828125, 0.9168701171875, -6.861663818359375, 4.018806457519531, 12.999114990234375, 6.010120391845703, -0.12980270385742188, 30.346466064453125, 4.74761962890625, -2.960540771484375, 7.408740997314453, -9.7017822265625, 5.721099853515625, 1.8426437377929688, 5.761444091796875, -16.414886474609375, 0.0, 9.0244140625, 12.751014709472656, 0.5432586669921875, 1.5238037109375, 28.7454833984375, -10.03009033203125, 1.9400405883789062, 3.0106201171875, 2.3488922119140625, -1.0328826904296875, -5.869659423828125, 26.28607177734375, 7.0594482421875, -5.150146484375, 25.20526123046875, -8.129650115966797, -11.672454833984375, 0.593505859375, 12.51959228515625, 3.10662841796875, 6.704925537109375, -0.8443450927734375, -13.574172973632812, 9.249763488769531, -4.7170867919921875, 0.2422943115234375, 13.554351806640625, 18.38385009765625, -13.161376953125, 4.256561279296875, 9.8035888671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000097.npy"}
{"epoch": 0.2031413612565445, "step": 98, "batch_size": 128, "mean": 4.8256635665893555, "std": 10.406757354736328, "min": -28.35626220703125, "p10": -6.207757568359375, "median": 3.687765121459961, "p90": 17.52878265380859, "max": 34.913787841796875, "pos_frac": 0.671875, "sample": [12.001667022705078, 13.535385131835938, 22.35162353515625, 3.3268775939941406, 30.663330078125, 17.974411010742188, 14.540847778320312, -6.7832183837890625, 4.81005859375, -1.0839614868164062, 15.947601318359375, -12.409149169921875, -0.761383056640625, 5.08770751953125, 8.068161010742188, 8.47515869140625, 2.49163818359375, -12.8363037109375, -5.652061462402344, 2.4461746215820312, 11.60858154296875, -6.17071533203125, -6.794830322265625, 4.166259765625, 3.7216415405273438, 3.620553970336914, -0.75347900390625, -4.2297821044921875, -8.35992431640625, -17.069854736328125, 1.8661975860595703, -28.35626220703125, -3.769989013671875, -7.578125, 3.653888702392578, -1.2471923828125, -5.462665557861328, 14.64752197265625, 2.4127655029296875, 1.9752349853515625, 2.299468994140625, 7.763916015625, 0.233673095703125, 6.887939453125, -1.5498046875, 17.17474365234375, 0.44641876220703125, 19.7022705078125, -5.0059356689453125, 13.353767395019531, 32.1861572265625, 16.05902099609375, -1.0049362182617188, 2.45001220703125, -0.2266845703125, -0.21038818359375, 31.7227783203125, 13.470184326171875, -3.704071044921875, 24.670806884765625, -16.59307861328125, -3.4485244750976562, -1.72723388671875, -6.294189453125, 23.473480224609375, -1.7788848876953125, 11.907073974609375, 5.918243408203125, 6.359884262084961, -2.0306854248046875, 8.433197021484375, 17.337799072265625, -13.85430908203125, 2.9329681396484375, 21.3790283203125, 13.120315551757812, 5.7008056640625, 13.7750244140625, 0.208709716796875, 16.9656982421875, 6.0461273193359375, 18.486740112304688, -7.550445556640625, -1.5079154968261719, 14.558197021484375, -4.081413269042969, -0.44281005859375, 1.2153301239013672, 2.60003662109375, 0.34002685546875, 21.8408203125, 9.166107177734375, 5.373802185058594, -1.679656982421875, 7.280921936035156, 8.06134033203125, 0.7350845336914062, 7.317474365234375, 3.5740280151367188, 5.736480712890625, 2.2677001953125, 16.655471801757812, 34.913787841796875, 4.573883056640625, 7.317352294921875, -5.045768737792969, -3.545379638671875, 6.1725006103515625, 5.737449645996094, 4.716888427734375, -0.9033203125, 2.379913330078125, 4.6616363525390625, 7.402700424194336, -1.1948432922363281, 8.431007385253906, 2.4279022216796875, -2.04583740234375, 7.458106994628906, 13.7598876953125, 12.66152572631836, -5.13641357421875, 14.35626220703125, 7.5997161865234375, 21.03875732421875, 4.483978271484375, -18.70062255859375, 9.591400146484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000098.npy"}
{"epoch": 0.20523560209424083, "step": 99, "batch_size": 128, "mean": 5.088243007659912, "std": 10.713027954101562, "min": -24.485382080078125, "p10": -8.144102478027344, "median": 4.895835876464844, "p90": 18.359442138671874, "max": 25.830718994140625, "pos_frac": 0.734375, "sample": [-17.538726806640625, 0.4983978271484375, -11.665618896484375, 25.305862426757812, 4.924079895019531, 12.20941162109375, -1.1377410888671875, 17.47698974609375, 2.33599853515625, -5.376495361328125, 5.5355072021484375, -11.243942260742188, 0.1114501953125, 6.4994354248046875, -8.014862060546875, 15.578216552734375, 19.24560546875, 9.03805160522461, 8.4217529296875, -2.6463775634765625, 5.141720771789551, 3.2318878173828125, 11.344970703125, -18.0357666015625, 25.830718994140625, 7.8712615966796875, 4.84490966796875, -11.44415283203125, 6.77081298828125, 16.9114990234375, 15.02557373046875, -4.101318359375, 21.58831787109375, 1.9630889892578125, 7.3055419921875, -0.2469482421875, 4.522186279296875, 7.19500732421875, 24.48504638671875, 4.873565673828125, 10.559280395507812, 16.766204833984375, 6.219268798828125, 18.09088134765625, 6.6543426513671875, -4.9503021240234375, 12.484386444091797, 2.2170772552490234, -0.0041351318359375, 0.612823486328125, -13.81146240234375, -21.45208740234375, -19.556060791015625, 11.423118591308594, 12.330795288085938, -4.102684020996094, 14.23187255859375, -19.461639404296875, 15.236480712890625, 8.519561767578125, 4.9181060791015625, 2.0370407104492188, 18.986083984375, 1.4554367065429688, -5.07928466796875, 16.9002685546875, -24.485382080078125, 7.399810791015625, 2.6886825561523438, 20.242340087890625, 1.73779296875, 17.17413330078125, -1.049041748046875, 0.8151683807373047, -2.3155517578125, 10.09564208984375, 5.089912414550781, 18.01043701171875, 6.005928039550781, 3.084514617919922, 11.157867431640625, 1.50848388671875, -3.009063720703125, 4.6023101806640625, 17.615676879882812, 3.9296798706054688, 12.962188720703125, -3.885986328125, 7.506927490234375, 3.6854400634765625, -8.445663452148438, 3.0475425720214844, 0.123687744140625, 7.33062744140625, 19.02484130859375, 1.4566268920898438, 16.156005859375, -11.604461669921875, 1.5158824920654297, -4.11358642578125, 21.267791748046875, -0.02631378173828125, 0.0, 5.193515777587891, 0.68634033203125, 17.15167236328125, 2.3927154541015625, -3.450836181640625, 17.400100708007812, 23.1986083984375, 22.831390380859375, 12.6566162109375, -23.9588623046875, 16.10833740234375, 2.060842514038086, 6.4758453369140625, 11.70709228515625, 22.695098876953125, -3.683258056640625, 1.897125244140625, 2.634033203125, -2.439483642578125, 13.702362060546875, -0.7648468017578125, 12.12982177734375, 7.072479248046875, 22.304962158203125, 3.132232666015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000099.npy"}
{"epoch": 0.20732984293193718, "step": 100, "batch_size": 128, "mean": 4.568582057952881, "std": 9.878426551818848, "min": -17.0338134765625, "p10": -6.80193176269531, "median": 3.6075353622436523, "p90": 18.591171264648434, "max": 29.46441650390625, "pos_frac": 0.6953125, "sample": [7.86285400390625, 3.77069091796875, 11.679916381835938, 3.99951171875, 15.926025390625, 4.92803955078125, -11.428619384765625, 18.225830078125, 0.51959228515625, 9.507888793945312, -3.38470458984375, 1.51904296875, -0.7386627197265625, 4.896697998046875, 14.687942504882812, 27.618377685546875, 1.120391845703125, 2.454132080078125, -8.807754516601562, 9.01910400390625, 0.266357421875, 6.44085693359375, -9.9273681640625, -1.7682762145996094, 1.883575439453125, 11.347900390625, 2.562530517578125, -8.9971923828125, 0.1824493408203125, 16.93975830078125, -12.09564208984375, 1.254119873046875, -1.8552398681640625, 4.445213317871094, 3.37713623046875, 4.2714691162109375, -2.325164794921875, 19.443634033203125, 3.173858642578125, 15.277847290039062, -3.356719970703125, 11.868026733398438, 6.8758544921875, 2.0325927734375, -1.4695892333984375, 5.7176513671875, 7.8803863525390625, 21.18841552734375, 23.94366455078125, 13.294906616210938, -1.0128250122070312, -5.8112335205078125, -5.392742156982422, 5.777032852172852, 3.5882701873779297, 11.371047973632812, -13.8271484375, 0.373291015625, -17.0338134765625, 2.8027801513671875, 0.9067916870117188, 27.261993408203125, -5.5799102783203125, 4.72552490234375, -0.507598876953125, 5.818389892578125, -4.9028167724609375, -14.427032470703125, 4.0231475830078125, 6.74560546875, 29.46441650390625, 10.703857421875, 4.285003662109375, 5.548112869262695, -1.08660888671875, -8.3675537109375, -5.357025146484375, 7.6930084228515625, 0.57220458984375, 11.632680892944336, -12.076492309570312, 5.329399108886719, 4.20794677734375, 3.447723388671875, 5.605377197265625, 3.4797611236572266, -3.8566436767578125, -0.07984161376953125, 14.768218994140625, 27.3853759765625, -0.1619873046875, 3.6722412109375, -6.130950927734375, 3.626800537109375, 6.292022705078125, 1.9115066528320312, 4.102577209472656, 2.5974864959716797, -0.152252197265625, 12.542938232421875, 22.632781982421875, -0.874725341796875, 24.48773193359375, 15.5391845703125, 5.266082763671875, -4.12957763671875, 15.9766845703125, 25.847122192382812, 0.7035846710205078, 1.11505126953125, 26.521026611328125, 5.290985107421875, -0.5703582763671875, -4.1448211669921875, -11.7215576171875, 16.642135620117188, 0.00418853759765625, 4.450958251953125, 22.3533935546875, -10.1300048828125, 0.5832195281982422, 10.141265869140625, 14.792633056640625, -3.6803627014160156, 9.369274139404297, -1.745849609375, 22.491668701171875, -14.1805419921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000100.npy"}
{"epoch": 0.2094240837696335, "step": 101, "batch_size": 128, "mean": 4.595433712005615, "std": 10.414432525634766, "min": -22.61175537109375, "p10": -6.987526702880859, "median": 4.8499755859375, "p90": 15.425973510742185, "max": 36.5018310546875, "pos_frac": 0.734375, "sample": [5.760093688964844, -22.61175537109375, 6.200225830078125, 16.4169921875, 3.6671524047851562, 2.394866943359375, 3.1866188049316406, 4.86102294921875, -4.35406494140625, 0.9537353515625, 15.981414794921875, 11.933013916015625, -3.26416015625, 21.4805908203125, 30.834228515625, 27.1541748046875, 2.4773311614990234, 5.2507476806640625, -1.1933155059814453, 12.156661987304688, 7.4703521728515625, 5.186286926269531, 11.214691162109375, 9.5528564453125, 1.3763675689697266, 20.031265258789062, 6.309722900390625, 1.059539794921875, 1.134735107421875, 27.80426025390625, -1.6776504516601562, 6.04449462890625, 3.900165557861328, 6.2592315673828125, 12.17706298828125, 9.28338623046875, 23.580307006835938, 14.582138061523438, 12.766822814941406, 1.60089111328125, -7.653656005859375, 3.8335189819335938, 12.7413330078125, 11.24493408203125, 18.96051788330078, 15.115142822265625, 7.458740234375, -5.1117401123046875, -0.21604156494140625, -11.6431884765625, -1.9878387451171875, 6.7008209228515625, -16.325225830078125, 11.510658264160156, 22.07684326171875, 6.61309814453125, 6.452423095703125, -17.886672973632812, -1.5874862670898438, 2.9033355712890625, -4.56121826171875, 7.904266357421875, 2.897369384765625, 12.176155090332031, -2.436920166015625, -4.391210556030273, 6.8625946044921875, 10.405563354492188, -1.0807418823242188, 14.972808837890625, 1.140777587890625, 3.6240081787109375, 0.8570556640625, 5.438392639160156, 0.9940605163574219, 2.279460906982422, 22.28326416015625, 1.4206619262695312, 3.0084304809570312, -22.41717529296875, -21.173309326171875, 8.424224853515625, -9.187849044799805, 15.18792724609375, 3.3892974853515625, 2.7576446533203125, 13.512969970703125, 14.658039093017578, 1.0509185791015625, -2.0449371337890625, 9.690704345703125, 5.504322052001953, 7.7719879150390625, 5.205841064453125, 8.6370849609375, 2.3375091552734375, -4.260009765625, 31.12139892578125, 12.90667724609375, 5.6526031494140625, -16.15496826171875, -21.205596923828125, 9.730775833129883, 1.358642578125, 5.0511474609375, -2.18414306640625, 6.3140869140625, 4.83892822265625, 36.5018310546875, -7.102012634277344, 9.3052978515625, -0.5115280151367188, 12.322723388671875, -1.88250732421875, 0.8044052124023438, 6.853485107421875, -2.89892578125, -6.9384613037109375, -5.6676025390625, 6.786460876464844, -10.699417114257812, -0.305877685546875, 3.1715240478515625, 0.5745201110839844, -13.483322143554688, 1.0916719436645508, 7.4874420166015625, 10.39425277709961], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000101.npy"}
{"epoch": 0.21151832460732983, "step": 102, "batch_size": 128, "mean": 6.431709289550781, "std": 9.558104515075684, "min": -18.495574951171875, "p10": -5.72793731689453, "median": 5.8612060546875, "p90": 18.447464752197266, "max": 33.0364990234375, "pos_frac": 0.734375, "sample": [-5.3936004638671875, -1.8864593505859375, 12.86709976196289, 5.629463195800781, 1.7666168212890625, 18.733726501464844, 9.73541259765625, -0.6755256652832031, 5.848541259765625, 13.291351318359375, 6.43182373046875, 6.552951812744141, 11.68817138671875, 7.816680908203125, -0.226959228515625, 16.150283813476562, 8.114715576171875, -8.816680908203125, -1.44744873046875, -0.9820556640625, -3.1175384521484375, 13.041351318359375, 12.77655029296875, 22.892547607421875, 16.78985595703125, -2.444244384765625, 6.151763916015625, -7.855743408203125, 5.497703552246094, -7.9681396484375, -6.730010986328125, 5.725799560546875, 3.3386993408203125, 8.182022094726562, -6.552947998046875, 20.6693115234375, 14.38623046875, -1.5937881469726562, 1.6520233154296875, 1.114034652709961, 0.58819580078125, 1.3046188354492188, -2.025808334350586, 16.972862243652344, 6.80029296875, 4.8516082763671875, 3.1629486083984375, -6.508056640625, -9.33392333984375, -6.760223388671875, 18.882415771484375, -4.7133026123046875, 6.090446472167969, 6.342132568359375, 4.943756103515625, 16.588653564453125, 7.7606353759765625, 33.0364990234375, 9.0894775390625, 16.992523193359375, 1.8164443969726562, -0.92645263671875, 17.811798095703125, 11.586456298828125, 8.30291748046875, -4.609771728515625, -9.332962036132812, 1.9421844482421875, -3.8774681091308594, 5.585990905761719, 10.85696029663086, 17.03009796142578, 22.55193328857422, 15.432647705078125, 5.827274322509766, 19.174392700195312, 18.46074676513672, 5.6491546630859375, 7.86737060546875, -13.818099975585938, 1.2885093688964844, 8.509033203125, 0.0, 4.999542236328125, -2.1686248779296875, 17.16412353515625, 0.0, 6.277366638183594, 13.92669677734375, -0.1445293426513672, 12.449447631835938, 13.836990356445312, -1.3488311767578125, 1.923065185546875, 16.89788818359375, 5.873870849609375, 13.333999633789062, 0.7207717895507812, 10.268951416015625, 15.796958923339844, 17.240203857421875, 4.9452667236328125, 2.8457870483398438, 24.57427978515625, 6.791473388671875, 11.43145751953125, 2.825338363647461, 18.794418334960938, 0.6012153625488281, 18.4417724609375, -18.495574951171875, 10.613372802734375, -1.6198272705078125, 3.3126068115234375, 12.951919555664062, -15.589996337890625, 15.128791809082031, 2.6269378662109375, 31.6429443359375, 4.27764892578125, 19.0850830078125, 5.4844970703125, 12.064468383789062, 8.6051025390625, 9.5894775390625, 29.252288818359375, -3.16302490234375, -13.155303955078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000102.npy"}
{"epoch": 0.2136125654450262, "step": 103, "batch_size": 128, "mean": 4.867948055267334, "std": 10.88575553894043, "min": -21.8006591796875, "p10": -6.41690979003906, "median": 2.5023040771484375, "p90": 19.64322509765625, "max": 46.488067626953125, "pos_frac": 0.7109375, "sample": [-20.554901123046875, -7.69866943359375, 6.8105010986328125, 9.782806396484375, 18.95263671875, 16.158939361572266, 0.29022216796875, -2.8660659790039062, 25.9447021484375, 1.305389404296875, -2.817230224609375, 15.921875, 18.4112548828125, 2.0240478515625, 0.0, -1.501220703125, -15.2283935546875, -11.712356567382812, -2.69854736328125, 8.0460205078125, -0.6687822341918945, 1.2058563232421875, 9.877426147460938, -21.8006591796875, 1.294464111328125, 7.62109375, 1.60601806640625, 1.2065887451171875, 6.094749450683594, 1.6218414306640625, 12.821075439453125, 9.301620483398438, 0.2652587890625, 5.940582275390625, -4.2110595703125, 7.204811096191406, -1.8413009643554688, 15.912139892578125, 10.595626831054688, 7.970806121826172, -1.6390352249145508, 14.5576171875, 0.774261474609375, 30.77069091796875, -1.9726943969726562, 14.474136352539062, 2.953460693359375, 9.782684326171875, 13.85693359375, 46.488067626953125, 19.38446044921875, 14.22393798828125, 8.475799560546875, -12.60498046875, 2.286651611328125, 0.7991180419921875, -11.063156127929688, 8.332061767578125, 10.3321533203125, 10.6695556640625, 20.24700927734375, 5.44537353515625, 1.2787017822265625, 10.920639038085938, 5.754547119140625, 21.056243896484375, 10.564132690429688, 12.413238525390625, 4.024772644042969, 7.08612060546875, 0.9983291625976562, -0.5999298095703125, 1.13592529296875, 3.34796142578125, 0.2508697509765625, 0.70068359375, -9.076164245605469, -19.66259765625, 2.71795654296875, 3.077789306640625, 22.381500244140625, 13.853965759277344, 0.8652286529541016, 11.739013671875, -1.1717338562011719, 9.1751708984375, 5.725494384765625, 1.8497161865234375, 27.015884399414062, -2.4881591796875, 2.18634033203125, 1.84295654296875, 16.040985107421875, 7.85626220703125, -1.6854476928710938, 2.1009368896484375, 23.4156494140625, -2.787689208984375, -5.867584228515625, -13.34661865234375, 0.0, 3.2206649780273438, 3.6516456604003906, 17.23699951171875, -0.7297210693359375, 20.76763916015625, -12.90386962890625, 0.28741455078125, 2.0300674438476562, 11.900196075439453, 11.999374389648438, -5.397834777832031, -3.508342742919922, 22.31768798828125, 0.8540802001953125, 0.30071067810058594, 22.086471557617188, 23.67047119140625, 9.41973876953125, 21.63653564453125, -2.091705322265625, 6.788230895996094, -2.92626953125, -15.35595703125, 0.9752349853515625, -19.086395263671875, -4.667457580566406, -1.2326202392578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000103.npy"}
{"epoch": 0.2157068062827225, "step": 104, "batch_size": 128, "mean": 4.362521171569824, "std": 11.770313262939453, "min": -29.2064208984375, "p10": -8.196797943115234, "median": 3.505840301513672, "p90": 19.650753784179685, "max": 47.43035888671875, "pos_frac": 0.6796875, "sample": [4.222900390625, -5.085113525390625, 8.366512298583984, 17.1846923828125, -10.513671875, 5.790771484375, 4.06085205078125, -0.5905323028564453, 19.53692626953125, -7.43853759765625, 17.52838134765625, -8.190025329589844, 5.035369873046875, -16.780487060546875, 7.425079345703125, 8.146221160888672, 28.194717407226562, 21.147857666015625, -3.09368896484375, -2.4452896118164062, 2.65301513671875, -2.499176025390625, 7.876129150390625, -22.04766845703125, 26.78082275390625, -3.785400390625, 3.9029998779296875, -2.256267547607422, 2.8058338165283203, 26.280029296875, 7.202278137207031, 16.39111328125, -14.97235107421875, -0.6440658569335938, 3.245819091796875, -8.121574401855469, 0.939117431640625, -8.212600708007812, 19.971893310546875, -1.94561767578125, 12.321014404296875, 0.120452880859375, 0.9706649780273438, -11.379486083984375, -10.305938720703125, 10.988967895507812, 0.6836700439453125, 7.3969879150390625, -2.1644287109375, -2.933441162109375, -3.8533935546875, 9.290374755859375, 8.862167358398438, 3.662933349609375, 26.564971923828125, 6.453025817871094, 19.916351318359375, -5.563385009765625, -4.096839904785156, 2.384397506713867, 4.837776184082031, 9.622398376464844, 3.17144775390625, 6.722846984863281, 4.987213134765625, 3.3487472534179688, -1.7128334045410156, 16.45849609375, 4.966835021972656, -14.293731689453125, 11.786102294921875, 18.197540283203125, 0.3442535400390625, -26.8013916015625, 47.43035888671875, 5.779670715332031, 22.238128662109375, 9.882835388183594, 11.365104675292969, 23.133026123046875, 6.1013641357421875, -29.2064208984375, -12.0032958984375, 11.782073974609375, 2.8041839599609375, -0.6236190795898438, 1.1204071044921875, 6.8820037841796875, -7.7412109375, 18.712066650390625, 25.0948486328125, 0.4388427734375, 4.5633544921875, 8.50244140625, 1.7566757202148438, 0.20480728149414062, -7.435028076171875, 0.5666942596435547, 24.7823486328125, 11.919708251953125, 15.193443298339844, 30.246826171875, -5.5593414306640625, 16.9884033203125, -18.305755615234375, 0.426239013671875, -6.729339599609375, 2.812652587890625, 6.012138366699219, -16.81610107421875, 0.0, 9.301620483398438, -1.2252616882324219, 14.078865051269531, 3.7301788330078125, 5.2330322265625, 11.10992431640625, -7.264671325683594, 8.4539794921875, 0.9889984130859375, -1.0195465087890625, 15.909149169921875, 2.884185791015625, 7.09063720703125, 18.43109130859375, -1.70452880859375, 0.116668701171875, 2.9458694458007812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000104.npy"}
{"epoch": 0.21780104712041884, "step": 105, "batch_size": 128, "mean": 8.211006164550781, "std": 11.36180305480957, "min": -24.758209228515625, "p10": -5.261918640136718, "median": 7.460666656494141, "p90": 22.67927703857422, "max": 34.20379638671875, "pos_frac": 0.78125, "sample": [5.409576416015625, 3.5847015380859375, 13.492324829101562, 1.624349594116211, -8.856201171875, 18.478271484375, 9.529754638671875, 15.777145385742188, 10.91485595703125, 2.1465682983398438, 14.614959716796875, 6.1175689697265625, 20.95367431640625, 19.373504638671875, -8.138427734375, 22.399505615234375, 6.2191162109375, -1.4240303039550781, 27.31781005859375, 9.3511962890625, -0.8540077209472656, -0.724517822265625, 8.863723754882812, 7.766937255859375, 24.367401123046875, 29.7923583984375, -21.048919677734375, -0.1024932861328125, -5.427337646484375, 29.396240234375, 5.19122314453125, 13.519668579101562, 0.114410400390625, 17.89422607421875, -2.124847412109375, 20.76019287109375, 23.192779541015625, 5.4555206298828125, 18.52569580078125, -11.854583740234375, 5.2263946533203125, 23.40301513671875, 15.321258544921875, 18.95849609375, 1.8162994384765625, 17.92919921875, 9.5623779296875, -2.11932373046875, 21.775192260742188, 20.535888671875, 3.0900802612304688, 13.4267578125, -0.859222412109375, 15.114288330078125, -2.598968505859375, 13.336029052734375, 9.62762451171875, 27.5494384765625, 9.275520324707031, -10.87481689453125, 3.50079345703125, -11.631439208984375, -0.148101806640625, 3.7578277587890625, 6.1304931640625, 8.379913330078125, 15.223114013671875, 0.21380615234375, 2.796781539916992, 2.137481689453125, 20.321319580078125, -19.6448974609375, 16.995635986328125, 22.69964599609375, 8.11712646484375, -4.2858428955078125, 12.913330078125, 17.08404541015625, 26.5146484375, 1.92010498046875, 8.36480712890625, 3.5977935791015625, 7.364265441894531, 0.6692047119140625, 7.55706787109375, 20.584930419921875, -13.80303955078125, 22.44964599609375, 6.512062072753906, -24.758209228515625, 3.5261383056640625, 10.465469360351562, 25.076126098632812, -12.42706298828125, 7.135017395019531, 4.7891998291015625, 9.451759338378906, -7.3339996337890625, 12.0699462890625, 0.0, 4.655620574951172, 6.574005126953125, 0.8007831573486328, -0.32592010498046875, 15.3892822265625, -3.2566375732421875, 33.47723388671875, 27.506591796875, 6.4554595947265625, 5.566734313964844, 7.63458251953125, 22.670547485351562, 0.921875, 4.948722839355469, 17.913909912109375, 11.397537231445312, -5.1910247802734375, 11.55145263671875, 2.92218017578125, -8.138336181640625, 8.575164794921875, 21.272377014160156, 4.846961975097656, 4.399169921875, -2.1280517578125, 34.20379638671875, 20.105682373046875, 10.8868408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000105.npy"}
{"epoch": 0.2198952879581152, "step": 106, "batch_size": 128, "mean": 5.4460906982421875, "std": 11.719334602355957, "min": -22.757904052734375, "p10": -9.071356201171874, "median": 5.016040802001953, "p90": 22.127566528320312, "max": 35.99835205078125, "pos_frac": 0.6875, "sample": [-5.6346588134765625, -14.996337890625, 14.513336181640625, -2.230865478515625, -3.857147216796875, 22.24609375, -6.4520263671875, 7.5671844482421875, -13.840972900390625, 14.427947998046875, -14.894927978515625, -0.3823585510253906, 4.1431884765625, 18.721893310546875, 13.062255859375, 31.820556640625, 27.56695556640625, -9.633056640625, -4.6938323974609375, -3.46820068359375, 14.350196838378906, 3.73541259765625, 5.105865478515625, 4.958839416503906, 12.037506103515625, 1.9105377197265625, 3.440426826477051, 6.675376892089844, 11.704910278320312, 25.771026611328125, 2.005767822265625, 9.151031494140625, 15.82427978515625, 1.3809051513671875, 0.45111846923828125, -6.051017761230469, 9.45294189453125, 2.895660400390625, 6.0371551513671875, 8.196380615234375, 22.086669921875, 1.3523406982421875, 6.724708557128906, -15.494140625, -2.7350006103515625, 14.63726806640625, -6.50146484375, -1.978424072265625, 13.250770568847656, 5.446403503417969, 6.79254150390625, 8.476577758789062, -5.033699035644531, 17.9351806640625, -18.058792114257812, -22.757904052734375, 10.540069580078125, 30.9932861328125, 28.760284423828125, 21.8426513671875, 7.0335693359375, 6.403350830078125, -11.001251220703125, -7.21124267578125, 11.3863525390625, 0.4344329833984375, -6.061279296875, -3.6379623413085938, 22.7406005859375, -12.1112060546875, 11.288482666015625, 6.075439453125, -2.67657470703125, 5.9957733154296875, -12.76031494140625, 0.9527912139892578, -2.9955215454101562, 3.970428466796875, 18.05694580078125, 2.427825927734375, -0.0376434326171875, 35.99835205078125, 8.816162109375, 2.0283355712890625, -6.764373779296875, 3.244232177734375, 22.73895263671875, -0.4488868713378906, 15.15313720703125, -8.83062744140625, 10.297698974609375, 4.159095764160156, 15.732513427734375, -0.4944610595703125, 2.33343505859375, 4.9576416015625, 12.90106201171875, 20.211578369140625, 5.0732421875, 7.969749450683594, 29.22607421875, -17.12457275390625, 1.1069717407226562, -3.7366180419921875, 6.8832855224609375, 12.597614288330078, 13.483551025390625, -12.547210693359375, 3.891204833984375, -5.605987548828125, -10.903533935546875, -2.938720703125, 32.512939453125, 7.244365692138672, -5.823028564453125, 5.8244781494140625, 11.680984497070312, -5.776206970214844, 19.927093505859375, 1.82000732421875, 22.222991943359375, 6.8846893310546875, 9.288482666015625, 23.214385986328125, 19.75738525390625, 0.08794403076171875, 10.050140380859375, 1.2023773193359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000106.npy"}
{"epoch": 0.22198952879581152, "step": 107, "batch_size": 128, "mean": 7.372006416320801, "std": 12.527613639831543, "min": -57.27874755859375, "p10": -4.940663909912109, "median": 5.6724090576171875, "p90": 23.86631774902343, "max": 34.164215087890625, "pos_frac": 0.71875, "sample": [-0.7735595703125, -3.9365692138671875, 4.248695373535156, 5.570159912109375, 10.76080322265625, -5.02899169921875, 13.088134765625, 27.97357177734375, 8.537200927734375, 4.74761962890625, -4.902809143066406, 5.774658203125, 1.4428253173828125, 13.56927490234375, 13.092388153076172, -1.55267333984375, 14.826141357421875, -8.738922119140625, 9.58648681640625, 19.70306396484375, -2.894390106201172, 23.122543334960938, 7.4437255859375, 14.22802734375, -12.241729736328125, 27.22857666015625, 8.270263671875, -13.47149658203125, 17.550674438476562, -7.8283538818359375, 17.3128662109375, 0.5730361938476562, 1.274810791015625, 19.317298889160156, 21.212158203125, 23.141998291015625, 11.22222900390625, 5.923393249511719, 33.5804443359375, 0.18017578125, 27.0264892578125, 7.229705810546875, 6.056671142578125, 12.77423095703125, 3.3947982788085938, 1.0815448760986328, -2.382965087890625, 4.761077880859375, 14.078153610229492, 7.2139739990234375, 8.340217590332031, -0.8370437622070312, 3.6172714233398438, 3.384429931640625, 22.561492919921875, -5.235984802246094, 5.1810760498046875, 1.5296516418457031, 19.48760986328125, 9.246292114257812, 4.7930908203125, 2.7785491943359375, -3.27294921875, 33.71087646484375, -0.2629547119140625, -5.590766906738281, 4.279205322265625, 31.135711669921875, 22.804519653320312, 18.65740966796875, -12.647979736328125, 25.556396484375, -4.5728607177734375, 14.763763427734375, 21.726806640625, 1.8462944030761719, 3.74078369140625, 19.361236572265625, 12.39208984375, 12.917926788330078, 29.356903076171875, 15.067138671875, -2.5491943359375, -11.542922973632812, -2.33001708984375, 5.4519195556640625, 7.572486877441406, -0.8470916748046875, 34.164215087890625, -8.1951904296875, 28.842864990234375, 12.27001953125, -0.9525527954101562, 7.463165283203125, 5.380035400390625, 1.4377994537353516, 16.151397705078125, 1.9705810546875, -0.23907470703125, 0.22385597229003906, -57.27874755859375, -9.621490478515625, 11.689208984375, 26.71649169921875, 18.22504425048828, -3.0836105346679688, 28.273712158203125, -1.487884521484375, 7.7279510498046875, 1.5605621337890625, 20.993865966796875, -1.8232498168945312, 9.20556640625, 1.98748779296875, 6.0240478515625, 16.38958740234375, -3.2778778076171875, -1.8245849609375, -5.80279541015625, -0.052001953125, 1.05426025390625, 8.515350341796875, -4.015625, 6.95989990234375, -0.6291923522949219, 31.710540771484375, 4.899993896484375, 16.1243896484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000107.npy"}
{"epoch": 0.22408376963350785, "step": 108, "batch_size": 128, "mean": 8.218379974365234, "std": 11.055420875549316, "min": -19.36572265625, "p10": -5.1185791015625, "median": 7.1119232177734375, "p90": 24.858834838867185, "max": 34.396087646484375, "pos_frac": 0.7578125, "sample": [25.285675048828125, 15.819351196289062, -1.7178192138671875, 6.944549560546875, 12.977706909179688, -7.3858184814453125, 19.669082641601562, 8.844608306884766, 3.459878921508789, 12.244598388671875, 1.9722137451171875, -5.64617919921875, -4.68115234375, 13.197265625, 5.053466796875, -8.042411804199219, 28.33306884765625, 5.1139068603515625, -2.5383148193359375, 10.484130859375, 13.7027587890625, 4.7197113037109375, 11.4591064453125, -8.637786865234375, -0.0698394775390625, 8.078155517578125, 27.1522216796875, -0.9392547607421875, 0.0, -19.36572265625, 13.615772247314453, 4.816986083984375, 10.00677490234375, 5.153598785400391, 16.9381103515625, 2.241180419921875, 27.0057373046875, 7.279296875, 17.13897705078125, -0.28908729553222656, 3.624574661254883, 10.215606689453125, 4.32550048828125, -5.086151123046875, 5.56524658203125, 26.595382690429688, 11.578369140625, -15.15631103515625, 5.69342041015625, 27.229705810546875, 14.403976440429688, 3.7356643676757812, 9.60174560546875, 34.396087646484375, 6.1799163818359375, 1.721923828125, 0.0, -5.194244384765625, 1.7396717071533203, -0.3415813446044922, 2.3041343688964844, 24.6759033203125, 14.595977783203125, 9.924789428710938, -19.26678466796875, 7.8379364013671875, 22.044830322265625, 11.200542449951172, -1.00848388671875, 11.86383056640625, 9.332489013671875, 6.094940185546875, -3.390228271484375, 2.191314697265625, -3.9787063598632812, 23.42071533203125, 21.6588134765625, 1.6314620971679688, 11.75823974609375, 17.92730712890625, 9.528656005859375, -6.655418395996094, 28.3016357421875, 10.780288696289062, 4.38763427734375, -0.4825592041015625, 21.88702392578125, 10.948966979980469, 9.99365234375, 29.745361328125, 6.06439208984375, 20.328369140625, 3.3121185302734375, 11.39837646484375, 8.532941818237305, -1.8082733154296875, 30.9088134765625, 15.916732788085938, 27.429168701171875, 21.758056640625, 16.07293701171875, 4.26043701171875, 8.318374633789062, 0.0, -2.0933990478515625, 9.860794067382812, 6.7132110595703125, 33.286651611328125, 5.834079742431641, 24.2528076171875, 6.6217041015625, -8.5145263671875, 3.8004913330078125, 0.9401626586914062, 16.50494384765625, 26.548431396484375, 21.52703857421875, 2.72772216796875, 3.0175933837890625, 11.429130554199219, -10.766845703125, -11.384025573730469, -11.09200668334961, 7.6241455078125, -0.01207733154296875, 11.89990234375, 4.035003662109375, 17.222015380859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000108.npy"}
{"epoch": 0.2261780104712042, "step": 109, "batch_size": 128, "mean": 8.334964752197266, "std": 13.680073738098145, "min": -28.7318115234375, "p10": -7.088311576843261, "median": 6.608467102050781, "p90": 24.391578674316406, "max": 56.2899169921875, "pos_frac": 0.7421875, "sample": [4.251838684082031, 11.564163208007812, -10.226173400878906, 24.287185668945312, 3.796356201171875, 8.280387878417969, 15.007766723632812, 16.750396728515625, -11.083572387695312, 6.113525390625, 4.5286865234375, 18.82373046875, 0.3955554962158203, -3.06573486328125, -8.060256958007812, -5.65216064453125, -0.277008056640625, 31.676513671875, 41.792236328125, 1.7802276611328125, 0.9358367919921875, -0.094207763671875, -4.078399658203125, -14.175048828125, -7.196603775024414, 11.579612731933594, 0.479461669921875, -0.9514198303222656, -0.13948822021484375, 11.6676025390625, 6.579551696777344, -0.3072052001953125, 15.197509765625, 36.277435302734375, 25.088226318359375, 10.332557678222656, 12.501693725585938, 19.80804443359375, 2.09832763671875, 21.951812744140625, 13.406158447265625, 18.04168701171875, 5.666786193847656, -0.2037982940673828, 15.35247802734375, 33.48223876953125, -1.250518798828125, 27.7674560546875, 8.248687744140625, 6.637382507324219, 12.6776123046875, 4.4215087890625, 11.1551513671875, 5.10601806640625, -21.863006591796875, 6.216705322265625, 8.57916259765625, 6.897148132324219, 12.8087158203125, 7.745819091796875, 16.193145751953125, 3.657276153564453, 0.7451171875, 24.635162353515625, 7.1031494140625, 10.789176940917969, -5.50469970703125, 8.31732177734375, 9.663673400878906, 11.855979919433594, 17.885894775390625, 3.1412353515625, -8.158493041992188, 15.917877197265625, 11.07822036743164, 22.804718017578125, 48.422943115234375, -4.72100830078125, 1.7136001586914062, 30.31256103515625, 4.89654541015625, 4.576385498046875, 19.00151824951172, 14.230728149414062, 0.947021484375, 0.9564437866210938, 14.100433349609375, 1.68182373046875, -10.066329956054688, 42.455322265625, 24.191070556640625, 17.73590087890625, 6.2359619140625, 56.2899169921875, 9.03658676147461, -0.946533203125, 2.0315399169921875, 34.496826171875, -11.363525390625, 4.285430908203125, -7.472900390625, 18.797119140625, 0.0, 1.9125289916992188, 30.6812744140625, 10.393630981445312, 13.529037475585938, 10.15338134765625, -0.549224853515625, 9.874969482421875, -1.4713134765625, 6.298728942871094, 1.1857757568359375, -2.260601043701172, 21.898834228515625, -1.0587310791015625, 13.2845458984375, 19.653564453125, -12.998321533203125, -28.7318115234375, -27.93555450439453, 23.97454833984375, 4.7403106689453125, 16.19214630126953, 6.1702880859375, -7.041900634765625, -6.995025634765625, 14.891937255859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000109.npy"}
{"epoch": 0.22827225130890053, "step": 110, "batch_size": 128, "mean": 8.079712867736816, "std": 13.642189979553223, "min": -39.15118408203125, "p10": -5.970778656005859, "median": 6.0443572998046875, "p90": 26.65087890625, "max": 44.01434326171875, "pos_frac": 0.734375, "sample": [-0.3008880615234375, -39.15118408203125, -11.894840240478516, 28.443603515625, -2.17938232421875, 1.6409225463867188, -3.1053390502929688, 1.383056640625, -1.38433837890625, 28.10638427734375, 4.364593505859375, 3.90045166015625, 2.8582115173339844, 24.65020751953125, -5.117664337158203, 0.180694580078125, 12.46343994140625, 6.1383209228515625, 17.136154174804688, 4.353424072265625, -10.236137390136719, 20.530029296875, 2.0737476348876953, 26.430084228515625, 8.326019287109375, 3.1233673095703125, -5.795173645019531, 14.298568725585938, 0.17645263671875, -4.301177978515625, 29.397857666015625, 13.672164916992188, 26.18865966796875, 2.447265625, 34.95477294921875, -1.97796630859375, 16.74462890625, 17.5771484375, 10.472564697265625, -3.98291015625, 5.791656494140625, 8.4417724609375, 10.829208374023438, 26.84100341796875, -9.636932373046875, 0.6253852844238281, 15.634613037109375, -8.053741455078125, 12.214309692382812, 0.2472991943359375, 22.252662658691406, -20.63055419921875, 12.8388671875, 1.6476898193359375, 31.702606201171875, 12.14642333984375, 3.44696044921875, -31.155303955078125, 16.16156005859375, 17.845458984375, 4.33123779296875, 5.8759613037109375, 2.839996337890625, -1.1735076904296875, 1.608001708984375, 27.421661376953125, -7.404693603515625, 27.4466552734375, 18.9208984375, -14.89727783203125, -0.7641048431396484, 16.99188232421875, 4.76812744140625, 2.3623428344726562, 11.684791564941406, 6.551544189453125, 0.5466384887695312, -2.8809280395507812, -1.80255126953125, 11.726730346679688, 0.0, 4.93280029296875, 22.266845703125, 14.025398254394531, -11.3756103515625, 13.93865966796875, 0.5389080047607422, -5.47528076171875, 6.534759521484375, 27.395729064941406, 15.425125122070312, 38.490081787109375, 15.393020629882812, 25.385910034179688, 6.599517822265625, 17.90997314453125, 24.7069091796875, -1.6109161376953125, -2.350383758544922, 22.26422119140625, 26.56939697265625, 6.731334686279297, 1.3129119873046875, 24.546844482421875, -16.33769989013672, 8.3887939453125, 4.509742736816406, -6.380523681640625, 25.196441650390625, 22.181015014648438, -0.9198837280273438, 30.439239501953125, -3.1684417724609375, 9.273178100585938, 10.661102294921875, 28.2781982421875, 44.01434326171875, -3.77301025390625, 18.722930908203125, 21.723236083984375, 1.3971405029296875, 20.752593994140625, 5.9503936767578125, 7.846343994140625, -13.488677978515625, 2.9624481201171875, -1.05718994140625, 6.925201416015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000110.npy"}
{"epoch": 0.23036649214659685, "step": 111, "batch_size": 128, "mean": 8.773923873901367, "std": 13.786818504333496, "min": -21.805679321289062, "p10": -7.748516464233398, "median": 7.340665817260742, "p90": 26.841630554199217, "max": 47.800079345703125, "pos_frac": 0.75, "sample": [4.5626068115234375, -10.9228515625, 13.168365478515625, -21.805679321289062, 47.800079345703125, -7.358772277832031, 10.138198852539062, 26.649566650390625, 0.8725433349609375, -0.60882568359375, 28.25689697265625, 5.364654541015625, 8.150070190429688, 6.0908203125, 6.02392578125, 14.982051849365234, 8.01165771484375, 36.437652587890625, -18.787002563476562, -13.417190551757812, 11.019180297851562, 5.768211364746094, 14.984344482421875, 4.06280517578125, 15.267547607421875, 2.86993408203125, 4.5989837646484375, 27.289779663085938, 1.7808914184570312, 11.530158996582031, 14.999305725097656, 18.34796142578125, -2.9123001098632812, 34.77197265625, -2.727783203125, 6.5949554443359375, -5.006866455078125, 15.96484375, -0.7869720458984375, 10.240509033203125, 2.1485366821289062, 10.466316223144531, -7.684604644775391, 22.78558349609375, 11.698837280273438, 2.2872314453125, 17.24285888671875, -1.07073974609375, 31.38250732421875, 7.384532928466797, 0.7488021850585938, 46.869903564453125, -12.601409912109375, -2.21783447265625, 0.29736328125, 18.7596435546875, 13.2603759765625, 23.709671020507812, 4.132347106933594, 29.437255859375, 5.02703857421875, 11.779983520507812, 5.286342620849609, 11.081634521484375, 4.59686279296875, -7.89764404296875, 3.51348876953125, 13.318374633789062, 41.0960693359375, 12.12298583984375, 7.429779052734375, -0.7712020874023438, 7.2967987060546875, 0.663726806640625, 5.01629638671875, -0.0024967193603515625, -0.82391357421875, 26.076202392578125, 9.977767944335938, -16.76556396484375, 0.9633865356445312, 10.16522216796875, -8.719207763671875, 3.62384033203125, -4.090606689453125, -6.235992431640625, 0.36319732666015625, 1.997802734375, 17.789405822753906, 29.122467041015625, 17.164794921875, 31.76641845703125, 18.75940704345703, -2.8172836303710938, 17.00347900390625, 9.94049072265625, 16.347164154052734, -2.848541259765625, 44.949737548828125, 3.56402587890625, 20.89531707763672, 10.124526977539062, -14.564208984375, -7.10540771484375, -2.96234130859375, 26.564422607421875, -12.074188232421875, 18.1260986328125, -2.464447021484375, 19.61896514892578, 15.729522705078125, 7.6688232421875, 16.68603515625, 37.926055908203125, 22.24090576171875, 17.89031982421875, 25.160858154296875, -17.478912353515625, 11.882476806640625, 14.01095962524414, 6.897499084472656, -11.62139892578125, 12.639190673828125, 5.58172607421875, 21.621124267578125, 1.347991943359375, -11.1578369140625, 1.743072509765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000111.npy"}
{"epoch": 0.2324607329842932, "step": 112, "batch_size": 128, "mean": 4.362063407897949, "std": 15.408896446228027, "min": -49.16754150390625, "p10": -13.400123596191404, "median": 4.036308288574219, "p90": 25.35478820800781, "max": 46.437225341796875, "pos_frac": 0.6171875, "sample": [15.704505920410156, 3.2113037109375, -4.56671142578125, 12.6451416015625, -0.36553955078125, 7.6635894775390625, -1.2508621215820312, -31.34625244140625, -5.15985107421875, 2.7400970458984375, -1.5579643249511719, 6.700809478759766, -10.12103271484375, 19.87591552734375, -1.318572998046875, 4.52447509765625, 12.88800048828125, 16.775222778320312, 7.039268493652344, 24.76263427734375, -8.845657348632812, 18.279937744140625, 14.026214599609375, 14.01220703125, -49.16754150390625, -13.0694580078125, -9.806182861328125, 12.113067626953125, 9.778106689453125, 8.647438049316406, -10.795913696289062, 5.350170135498047, -27.797149658203125, 5.62139892578125, 5.265983581542969, -17.242279052734375, 6.1773681640625, 5.763160705566406, 0.2212066650390625, 3.9082794189453125, 17.68878173828125, -1.9326171875, 6.4081878662109375, 14.1593017578125, 12.017951965332031, -2.5236434936523438, 14.99169921875, -0.440399169921875, 17.4022216796875, -6.15673828125, 10.4183349609375, 4.164337158203125, 0.818145751953125, 14.837646484375, -16.794097900390625, -2.896533966064453, -16.35247802734375, -9.189926147460938, -37.16802978515625, -8.705997467041016, -1.9191856384277344, -0.7026748657226562, 17.019546508789062, 35.344207763671875, -4.661102294921875, 3.3022308349609375, 30.33380126953125, 16.852859497070312, 1.4119873046875, 6.559700012207031, 7.23817253112793, -4.84516716003418, 18.054779052734375, 14.9654541015625, 6.8567047119140625, 37.8587646484375, -3.9781417846679688, 1.049560546875, 29.14471435546875, -18.580833435058594, -1.981201171875, 8.858566284179688, 0.26021575927734375, -0.8546466827392578, 20.331024169921875, 25.27587890625, 11.796066284179688, 0.21893310546875, -5.573600769042969, 2.5770721435546875, 25.538909912109375, -14.171676635742188, 30.34625244140625, 4.659610748291016, -4.595123291015625, 17.314102172851562, -2.5475540161132812, 0.7490291595458984, -37.06805419921875, -7.1669769287109375, -7.36181640625, 24.3585205078125, -1.287933349609375, -1.6405029296875, 27.350875854492188, 17.677413940429688, -16.15277099609375, 26.421295166015625, 4.71002197265625, 18.4063720703125, 34.66192626953125, -12.218780517578125, 5.047119140625, 28.11126708984375, 5.472084045410156, 1.2044525146484375, -18.060760498046875, -0.0909881591796875, 2.405975341796875, 0.24298095703125, 8.66748046875, -5.512359619140625, 27.1025390625, 27.40240478515625, 46.437225341796875, -17.7841796875, -4.370849609375, 9.842193603515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000112.npy"}
{"epoch": 0.23455497382198953, "step": 113, "batch_size": 128, "mean": 10.15865707397461, "std": 15.056642532348633, "min": -29.5384521484375, "p10": -7.787059020996093, "median": 8.514659881591797, "p90": 31.34191436767578, "max": 58.563995361328125, "pos_frac": 0.7421875, "sample": [4.523475646972656, 16.643936157226562, 9.7508544921875, -0.14596176147460938, 15.03057861328125, 1.6567459106445312, -6.786773681640625, 14.7154541015625, 26.23907470703125, 5.4815673828125, 8.048629760742188, 8.124954223632812, 18.692138671875, -11.053754806518555, -1.787384033203125, 3.872152328491211, 3.721220016479492, 2.0888442993164062, 5.8955078125, 14.005718231201172, 58.563995361328125, -4.64349365234375, -2.4219589233398438, 40.52886962890625, 13.7939453125, -29.5384521484375, 5.0727691650390625, 22.5455322265625, 13.2008056640625, -2.475860595703125, -2.88861083984375, -11.85662841796875, 23.133880615234375, 32.821319580078125, 39.59808349609375, -0.05279541015625, 16.149139404296875, -6.1625823974609375, 28.259674072265625, -9.5916748046875, 12.598770141601562, 30.741195678710938, 0.2234039306640625, 13.998748779296875, 9.426666259765625, 4.393482208251953, -10.67852783203125, 0.862945556640625, 0.5963516235351562, 31.93865966796875, 5.2462158203125, 18.322006225585938, -26.571563720703125, 27.518051147460938, 33.65582275390625, 24.065093994140625, 11.701896667480469, 34.306793212890625, 6.0215911865234375, -12.12384033203125, 8.562644958496094, -2.9145965576171875, -5.7422637939453125, 25.456451416015625, -0.13227081298828125, 8.4666748046875, 6.296611785888672, 7.1806182861328125, -16.245849609375, 18.4866943359375, 6.7576904296875, 32.040771484375, -1.8858833312988281, 11.508636474609375, -9.341354370117188, 4.178672790527344, 13.614044189453125, 31.086166381835938, 13.635711669921875, 16.427139282226562, 15.181770324707031, -11.50848388671875, 17.399032592773438, 29.74755859375, 34.98046875, 13.911170959472656, 13.113861083984375, -3.6853790283203125, -3.547637939453125, 7.25567626953125, 36.72369384765625, 26.620086669921875, 19.749191284179688, 13.119903564453125, -7.5250396728515625, 27.759429931640625, 1.4741058349609375, 12.840225219726562, 15.345703125, 7.87957763671875, 5.546112060546875, -2.13427734375, 13.1771240234375, 2.6469573974609375, 23.69757080078125, 38.03515625, 23.355300903320312, 41.21246337890625, -9.0045166015625, 1.329833984375, 22.86676025390625, 17.566680908203125, 12.33331298828125, 41.65899658203125, -5.6249542236328125, -8.3984375, 2.9877166748046875, 2.5867691040039062, 19.24700927734375, 16.545547485351562, 0.5887451171875, 17.2435302734375, -17.930038452148438, 10.34552001953125, 4.229927062988281, -1.8597564697265625, 19.85009765625, -5.058967590332031], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000113.npy"}
{"epoch": 0.23664921465968586, "step": 114, "batch_size": 128, "mean": 5.124326705932617, "std": 13.860980033874512, "min": -28.689453125, "p10": -12.692637634277343, "median": 4.287322998046875, "p90": 23.206402587890626, "max": 42.80426025390625, "pos_frac": 0.6640625, "sample": [3.7603206634521484, 21.246856689453125, 3.4630584716796875, 26.643474578857422, -0.0687103271484375, 6.4609375, 1.5255756378173828, 5.289043426513672, -1.35467529296875, -5.7654266357421875, -2.1953582763671875, 24.152572631835938, -0.872039794921875, 30.225250244140625, 15.666610717773438, 38.751129150390625, -11.172698974609375, 3.002716064453125, -21.89117431640625, 2.5645599365234375, 2.19171142578125, -15.959869384765625, 4.7853546142578125, 15.00885009765625, 1.8488311767578125, 17.288970947265625, 10.583915710449219, 9.96453857421875, 11.639617919921875, 14.468719482421875, 3.2763824462890625, 13.868759155273438, 4.09930419921875, 7.448204040527344, 17.17236328125, -12.277496337890625, 9.827323913574219, -23.705078125, 28.345611572265625, 5.16259765625, -5.560966491699219, -4.2464599609375, 2.3126754760742188, -28.689453125, 5.817169189453125, 1.7405853271484375, 0.011199951171875, 0.999542236328125, 6.872810363769531, -8.99407958984375, -0.41429901123046875, -13.667377471923828, 9.998062133789062, -10.07000732421875, -13.661300659179688, -19.196365356445312, 12.219680786132812, 42.80426025390625, 23.291595458984375, -7.4221038818359375, 26.720230102539062, -23.35540771484375, 19.029632568359375, -7.078643798828125, -3.106170654296875, 20.11248779296875, 37.42045593261719, -8.827606201171875, 0.5680961608886719, 15.96319580078125, -24.424407958984375, -0.05168914794921875, -17.324249267578125, 9.295364379882812, 2.89739990234375, 34.0723876953125, 0.9075546264648438, -0.3456878662109375, -2.2396392822265625, -15.50604248046875, 2.8069229125976562, 10.333152770996094, 7.900299072265625, -19.003082275390625, 1.4024639129638672, -0.8691921234130859, -18.988555908203125, -4.3800048828125, 0.0, 23.169891357421875, 12.226966857910156, -2.7561492919921875, 12.12017822265625, 5.6776123046875, -4.139617919921875, 21.688873291015625, -3.27996826171875, 17.674041748046875, 1.3951416015625, 4.776588439941406, 10.990135192871094, 32.3988037109375, 11.57785415649414, 9.04290771484375, 37.31341552734375, 16.732086181640625, 6.312591552734375, 4.984447479248047, -9.505950927734375, 0.0, 33.820068359375, -1.759979248046875, -6.716850280761719, 13.504486083984375, 7.317901611328125, 19.82459259033203, 6.18841552734375, 4.475341796875, 5.857608795166016, 5.794548034667969, 3.2178497314453125, 7.22906494140625, 6.215696334838867, -1.543426513671875, 15.418365478515625, 7.638641357421875, 3.3567123413085938, 15.129913330078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000114.npy"}
{"epoch": 0.2387434554973822, "step": 115, "batch_size": 128, "mean": 5.864140033721924, "std": 15.420974731445312, "min": -30.556488037109375, "p10": -11.808744812011719, "median": 5.091718673706055, "p90": 29.160845947265624, "max": 59.70355224609375, "pos_frac": 0.671875, "sample": [30.91754150390625, -14.3887939453125, 13.631134033203125, -10.987762451171875, -1.8896484375, -14.588203430175781, -8.745574951171875, 4.5836181640625, 2.571807861328125, 20.7039737701416, -5.1651153564453125, 9.539413452148438, 7.304656982421875, 8.322837829589844, 3.5253067016601562, -10.32464599609375, 10.178058624267578, -5.299957275390625, 15.57769775390625, 2.9320220947265625, -12.245243072509766, 2.2659454345703125, 6.650390625, 9.153564453125, 4.44420051574707, -3.5807952880859375, 2.54693603515625, 13.354339599609375, 14.55487060546875, 17.54876708984375, -3.8521881103515625, 13.367111206054688, 11.957275390625, 4.748512268066406, 2.9388465881347656, 8.761688232421875, -8.074249267578125, -19.971054077148438, -1.73553466796875, 33.4351806640625, 5.90545654296875, 29.4932861328125, 39.24609375, 7.8863983154296875, -23.33831787109375, -18.928482055664062, 24.47589874267578, -11.077163696289062, 26.791854858398438, 41.81219482421875, 6.313957214355469, 59.70355224609375, -8.819610595703125, -0.948944091796875, -11.367523193359375, 15.251190185546875, 30.486724853515625, 22.79803466796875, 13.061737060546875, -23.831512451171875, -16.88348388671875, -8.42694091796875, 29.01837158203125, 2.365234375, -4.3109130859375, 10.564208984375, 6.0157470703125, 0.767578125, 5.790367126464844, 32.46630859375, 0.8748416900634766, 8.6744384765625, 14.872272491455078, 9.311882019042969, -3.8323841094970703, 6.488533020019531, 43.8841552734375, 2.448822021484375, -18.489471435546875, -7.705230712890625, 2.18115234375, 2.66265869140625, -11.69671630859375, 34.93768310546875, 21.18981170654297, 4.64349365234375, -27.04315185546875, 9.624862670898438, 1.9863395690917969, 32.791046142578125, -4.25065803527832, 3.782745361328125, -7.767364501953125, 34.47174072265625, 16.376922607421875, -3.504913330078125, 2.0786819458007812, 5.85809326171875, -8.675384521484375, 6.924285888671875, 9.4171142578125, 5.1647796630859375, 9.465232849121094, 8.34166145324707, 8.67047119140625, 2.2691497802734375, 42.9266357421875, 5.9182891845703125, -8.7978515625, -12.070144653320312, 12.689605712890625, 5.018657684326172, 17.149818420410156, 6.4380950927734375, 12.681354522705078, -12.13568115234375, -6.4527740478515625, 13.325115203857422, -0.6129074096679688, 21.915802001953125, -30.556488037109375, -2.5193443298339844, 10.536972045898438, -0.08922576904296875, 12.807815551757812, 1.961181640625, -2.4383392333984375, 9.53753662109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000115.npy"}
{"epoch": 0.24083769633507854, "step": 116, "batch_size": 128, "mean": 8.168201446533203, "std": 14.695357322692871, "min": -41.659759521484375, "p10": -7.571015930175781, "median": 7.248046875, "p90": 25.108929443359372, "max": 56.54547119140625, "pos_frac": 0.7265625, "sample": [4.1820068359375, -4.078502655029297, 5.3751220703125, 3.3408432006835938, 6.774810791015625, 11.48956298828125, 9.519149780273438, 25.683212280273438, 42.387237548828125, 7.9847412109375, 19.539276123046875, 9.7154541015625, 20.840293884277344, 3.0811004638671875, 16.12054443359375, 12.008026123046875, -7.130767822265625, 12.590362548828125, 4.420864105224609, -9.714401245117188, -21.365692138671875, -6.092765808105469, 23.333099365234375, 23.87890625, 0.23554420471191406, -8.67401123046875, 11.588714599609375, -0.951904296875, 5.987846374511719, -0.187530517578125, 2.522785186767578, 23.72998046875, -10.57781982421875, 17.768035888671875, -5.4998626708984375, 1.7719573974609375, 22.834251403808594, 11.45733642578125, -7.935638427734375, -5.1469573974609375, -2.896343231201172, 18.202667236328125, 16.8280029296875, -9.580902099609375, 8.98150634765625, 3.4083633422851562, -3.53106689453125, 7.041259765625, 22.2816162109375, 33.19207763671875, -41.659759521484375, 3.880523681640625, 14.109390258789062, -16.96099853515625, -0.52984619140625, 26.2474365234375, 12.33673095703125, 24.72447967529297, -5.092742919921875, 1.4430580139160156, 10.441520690917969, 11.896438598632812, 8.178741455078125, 3.4005126953125, 8.944053649902344, 0.0, 8.467933654785156, 33.6534423828125, -2.8484649658203125, -9.12603759765625, 17.896453857421875, 33.484710693359375, -0.07696533203125, -5.879150390625, -7.2336578369140625, 1.5760955810546875, 1.2524337768554688, 2.0031890869140625, 7.052089691162109, 23.307769775390625, 3.4653854370117188, 36.52001953125, 16.65289306640625, 0.0, 12.275739669799805, 7.182243347167969, 19.246200561523438, 4.360107421875, 35.570098876953125, -0.669097900390625, -2.2187957763671875, 56.54547119140625, 13.865234375, -36.785400390625, 10.35870361328125, 31.325225830078125, 18.59410858154297, 4.436286926269531, 10.262907028198242, 9.80517578125, 11.487503051757812, 0.734649658203125, 7.7665557861328125, 28.361618041992188, 21.36712646484375, -21.061172485351562, 8.80181884765625, 13.341156005859375, 3.084228515625, -12.839752197265625, 11.335281372070312, 7.116119384765625, 45.76422119140625, -0.9370574951171875, 24.862808227539062, 17.193191528320312, 6.6913909912109375, 17.99530029296875, 22.937164306640625, 13.040069580078125, -7.4147491455078125, 7.313850402832031, 0.241973876953125, 29.376556396484375, -13.734832763671875, 6.275890350341797, -4.4820556640625, 18.4967041015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000116.npy"}
{"epoch": 0.24293193717277486, "step": 117, "batch_size": 128, "mean": 8.379829406738281, "std": 13.913829803466797, "min": -24.24554443359375, "p10": -7.622073364257813, "median": 6.904518127441406, "p90": 28.456852722167966, "max": 44.671112060546875, "pos_frac": 0.71875, "sample": [3.715991973876953, 7.061786651611328, 12.823616027832031, 16.827178955078125, 4.93756103515625, 13.94818115234375, 1.806610107421875, -1.713531494140625, 9.51300048828125, -4.038299560546875, 13.583450317382812, -7.7340545654296875, -1.6302490234375, 15.971939086914062, 18.26457977294922, 5.0225830078125, 9.396873474121094, 9.643966674804688, 24.469940185546875, 4.159027099609375, 18.110519409179688, -8.779953002929688, 7.593788146972656, 2.2049217224121094, 14.454618453979492, 28.311370849609375, 1.7381744384765625, -18.378280639648438, 13.03466796875, 44.671112060546875, 16.111934661865234, 28.796310424804688, 33.72454833984375, 14.279863357543945, 9.568513870239258, -10.55841064453125, 24.603134155273438, -0.378814697265625, 1.62939453125, -9.43536376953125, 35.42425537109375, 7.8341827392578125, 4.362030029296875, 27.9384765625, 0.0, 23.18316650390625, -5.8658447265625, 6.0407562255859375, 38.37396240234375, 24.487152099609375, 8.168228149414062, 34.884368896484375, -22.9544677734375, 17.7681884765625, 24.507110595703125, 22.2943115234375, -7.140380859375, 3.275064468383789, 3.8893585205078125, 6.599212646484375, 17.71783447265625, 16.18017578125, 12.137420654296875, 11.921375274658203, 5.74237060546875, 11.097015380859375, 14.028457641601562, -0.606719970703125, -7.5740814208984375, -8.149520874023438, 7.646240234375, -20.911163330078125, -0.33332252502441406, 9.420257568359375, 27.5079345703125, -13.763214111328125, 11.308929443359375, 12.968536376953125, 21.544677734375, 1.8836402893066406, -2.3900680541992188, -2.6083984375, 6.7672119140625, 34.16217041015625, 6.13946533203125, -6.7666015625, -6.2237701416015625, -14.070404052734375, -0.2520751953125, 3.7277889251708984, -24.24554443359375, 16.586669921875, 30.2940673828125, 29.204742431640625, 8.389453887939453, 5.153526306152344, -11.865890502929688, -2.320098876953125, -1.2014999389648438, 29.667083740234375, 3.27728271484375, -3.67755126953125, 1.748138427734375, -4.4492034912109375, -6.20062255859375, 7.0418243408203125, 30.477127075195312, 5.827384948730469, 25.2928466796875, -4.077369689941406, 16.16939926147461, 16.9302978515625, 21.6414794921875, 0.2489013671875, 0.5426788330078125, -4.06494140625, 4.836736679077148, 0.5553131103515625, 4.9045257568359375, -3.7577781677246094, -20.321426391601562, 36.50105285644531, 11.9632568359375, 14.411537170410156, 2.1638031005859375, 17.777084350585938, 38.637359619140625, 9.9029541015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000117.npy"}
{"epoch": 0.2450261780104712, "step": 118, "batch_size": 128, "mean": 9.169331550598145, "std": 15.382549285888672, "min": -26.18244171142578, "p10": -8.691111755371093, "median": 7.2857666015625, "p90": 27.859490966796873, "max": 52.4781494140625, "pos_frac": 0.75, "sample": [6.298576354980469, 9.041473388671875, 18.66448974609375, 4.2103271484375, 29.615982055664062, 24.80596923828125, -22.092620849609375, -13.1566162109375, 19.444061279296875, 27.746368408203125, 15.072265625, 5.65203857421875, 5.954067230224609, 5.96343994140625, -14.76568603515625, 35.022186279296875, 18.38568115234375, 20.151153564453125, 22.52672576904297, 6.6794586181640625, 11.035552978515625, 19.9215087890625, 32.95270538330078, 1.089874267578125, -25.60626220703125, -26.18244171142578, 16.761810302734375, 3.8294677734375, -19.622665405273438, 23.663360595703125, 17.661346435546875, 25.793968200683594, 43.91131591796875, 13.3280029296875, 3.6257171630859375, 5.6943817138671875, 2.4568634033203125, 6.449562072753906, 18.459152221679688, 20.263465881347656, -3.410797119140625, 23.8914794921875, 32.192779541015625, -3.1930484771728516, 2.7508697509765625, -3.341461181640625, 19.950546264648438, 16.73492431640625, 13.413818359375, 0.3966522216796875, 1.501556396484375, 2.5283203125, 28.123443603515625, 17.4307861328125, 28.1578369140625, 41.25128173828125, -7.436920166015625, 12.516571044921875, -0.5409412384033203, 21.145751953125, -5.7667999267578125, 0.0, -8.13226318359375, 10.5269775390625, -12.958282470703125, -8.121330261230469, -19.40203857421875, 6.873451232910156, 16.247772216796875, -11.016082763671875, 0.022144317626953125, 30.681060791015625, -0.0173797607421875, 17.202682495117188, 7.284210205078125, 25.777008056640625, 25.3189697265625, 17.0406494140625, -9.669815063476562, 26.15911865234375, 4.8927154541015625, 22.008590698242188, 7.287322998046875, -4.0566253662109375, 13.601875305175781, 25.296539306640625, -21.489227294921875, 1.08892822265625, -6.46649169921875, -7.7508392333984375, 51.30322265625, -4.8497467041015625, 24.332489013671875, 8.194976806640625, -1.7224197387695312, 9.933151245117188, 1.6049079895019531, -14.448356628417969, 13.1829833984375, -8.143386840820312, 5.110103607177734, 37.227874755859375, 8.9713134765625, 13.788299560546875, 0.448333740234375, 11.578201293945312, 4.827018737792969, 3.7059478759765625, 10.568368911743164, 32.296722412109375, -5.779041290283203, -8.27166748046875, 18.190044403076172, 13.990554809570312, 5.85577392578125, 1.8230209350585938, -22.580902099609375, 10.141357421875, 6.4931793212890625, 21.82342529296875, 2.267650604248047, 14.70172119140625, 3.897380828857422, 21.994064331054688, 0.0, 5.897918701171875, 15.609420776367188, 52.4781494140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000118.npy"}
{"epoch": 0.24712041884816754, "step": 119, "batch_size": 128, "mean": 7.728414535522461, "std": 18.457788467407227, "min": -40.07423400878906, "p10": -13.664340209960937, "median": 5.312740325927734, "p90": 31.05731811523437, "max": 68.47381591796875, "pos_frac": 0.6796875, "sample": [-40.07423400878906, -19.16216278076172, 28.17535400390625, 3.2226333618164062, 2.8050994873046875, 4.637931823730469, -8.884788513183594, -8.499832153320312, 8.486083984375, -3.92987060546875, 23.128997802734375, -11.708885192871094, -10.0123291015625, 34.202392578125, -17.3282470703125, 11.17242431640625, -13.279937744140625, 23.87664794921875, 27.078346252441406, -1.850921630859375, 6.83843994140625, 16.95501708984375, -1.5942707061767578, -8.614456176757812, 7.1366729736328125, 14.564048767089844, 18.901145935058594, 6.576416015625, 33.35693359375, 4.62945556640625, 4.31245231628418, 1.8164215087890625, -3.5465850830078125, 1.02728271484375, 13.355621337890625, 11.581146240234375, 11.909164428710938, 4.076957702636719, 8.222091674804688, 3.5819931030273438, 23.422088623046875, -9.133842468261719, -2.13873291015625, -0.028289794921875, 31.941162109375, 1.3424034118652344, 3.6145477294921875, -35.332275390625, -14.561279296875, -19.875, 50.362030029296875, 11.722335815429688, 3.542388916015625, 29.606781005859375, -4.771415710449219, -1.0992584228515625, 14.491241455078125, 2.123128890991211, 16.43218994140625, 22.819839477539062, 4.830413818359375, 0.6607666015625, 23.861404418945312, 14.69256591796875, 19.56964111328125, -8.1788330078125, 41.07269287109375, 12.853160858154297, 32.33221435546875, 18.043701171875, 0.6336498260498047, 7.285736083984375, 13.784317016601562, 0.0, 38.854095458984375, -2.362823486328125, -2.4231414794921875, -26.449676513671875, 19.364013671875, -24.05291748046875, 21.882152557373047, 13.12237548828125, 4.4801025390625, -17.049072265625, 12.776611328125, 0.4093780517578125, 30.67852783203125, -7.391450881958008, 16.627548217773438, 6.91693115234375, 25.3511962890625, -15.715423583984375, -2.19024658203125, 59.36761474609375, -3.7075576782226562, 68.47381591796875, 6.98052978515625, 0.214447021484375, 8.31744384765625, -29.2498779296875, 22.774993896484375, 32.512115478515625, 4.43096923828125, 18.22027587890625, -8.268096923828125, 20.21514892578125, -7.33497428894043, 6.363983154296875, 5.309501647949219, -12.70587158203125, 3.0790252685546875, -21.942779541015625, 51.90008544921875, 15.511009216308594, 57.469635009765625, 6.1016387939453125, -1.2549591064453125, 15.384735107421875, -0.604461669921875, 20.66180419921875, 7.392303466796875, 2.982208251953125, 28.9254150390625, -6.377105712890625, 39.4769287109375, 5.31597900390625, 5.482513427734375, -18.07171630859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000119.npy"}
{"epoch": 0.24921465968586387, "step": 120, "batch_size": 128, "mean": 9.059913635253906, "std": 17.28131675720215, "min": -32.68450927734375, "p10": -12.74334335327148, "median": 10.009925842285156, "p90": 31.8102294921875, "max": 57.129669189453125, "pos_frac": 0.6796875, "sample": [-10.75299072265625, 24.90618896484375, -6.689750671386719, 12.330795288085938, -4.017036437988281, 11.19818115234375, 23.7025146484375, -11.47821044921875, 7.1584625244140625, 13.624237060546875, -2.8311080932617188, -4.5230712890625, 21.27576446533203, 4.480010986328125, -15.781158447265625, 0.20465087890625, 32.759429931640625, -3.6517677307128906, 9.957504272460938, 17.917469024658203, -6.814453125, 32.53541564941406, 31.73724365234375, 27.8369140625, 26.89215087890625, 12.622146606445312, 5.9134368896484375, -2.7030029296875, -5.2730865478515625, 31.98052978515625, 26.5377197265625, -0.24932861328125, 6.088653564453125, 2.68646240234375, -27.045013427734375, 6.52410888671875, -0.6823883056640625, 15.773521423339844, 23.0374755859375, 21.63104248046875, 1.2307815551757812, -1.2631378173828125, 14.22198486328125, -25.988067626953125, 0.0, -2.109527587890625, -21.678802490234375, -2.0950927734375, -6.8591156005859375, 5.224578857421875, 11.293975830078125, 12.797130584716797, 52.68170166015625, 19.15484619140625, 15.06597900390625, 9.892379760742188, -15.695320129394531, 14.67034912109375, 9.049774169921875, 2.900054931640625, -2.2413482666015625, 0.0, 11.358978271484375, 20.369186401367188, 57.129669189453125, 15.2620849609375, 16.1820068359375, 39.29083251953125, -17.3951416015625, 20.05406951904297, 24.43145751953125, 10.47955322265625, 4.4163818359375, 17.2354736328125, 13.516998291015625, -7.497222900390625, 8.014678955078125, 33.23321533203125, 18.01495361328125, 18.2821044921875, 10.399017333984375, -16.75677490234375, -18.963775634765625, -8.68914794921875, -10.342460632324219, 36.252227783203125, 10.062347412109375, 38.510284423828125, 44.754425048828125, 16.405704498291016, 0.33316802978515625, 24.490631103515625, -28.070037841796875, 31.5301513671875, 37.66033935546875, 5.4913482666015625, 0.2268524169921875, 18.292770385742188, 11.963150024414062, 17.944732666015625, -22.229530334472656, 2.4634780883789062, 0.3135528564453125, 19.158531188964844, 27.160858154296875, -0.9577445983886719, 12.660232543945312, -2.4530258178710938, 40.24169921875, -32.68450927734375, 22.77191162109375, 21.825225830078125, 21.68793487548828, -17.8603515625, -23.53192138671875, 16.2666015625, 13.50042724609375, 19.697418212890625, 6.195564270019531, -1.06671142578125, -4.980583190917969, 14.786407470703125, -4.08184814453125, 50.144317626953125, 2.4061851501464844, 3.3096923828125, -3.5250701904296875, 15.503204345703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000120.npy"}
{"epoch": 0.2513089005235602, "step": 121, "batch_size": 128, "mean": 5.420945167541504, "std": 15.328969955444336, "min": -34.5689697265625, "p10": -12.823277282714843, "median": 4.195240020751953, "p90": 24.0266616821289, "max": 46.780517578125, "pos_frac": 0.6484375, "sample": [-1.8181838989257812, -0.76190185546875, 3.4462451934814453, 6.858467102050781, 0.869476318359375, 3.640899658203125, -2.883544921875, 6.4525299072265625, 8.608795166015625, 27.651214599609375, 3.305713653564453, -1.5173187255859375, 33.36412048339844, 4.592506408691406, 17.528907775878906, 0.3605194091796875, -2.814605712890625, 4.0278167724609375, 0.5360755920410156, 23.448226928710938, 18.104904174804688, -17.211334228515625, 19.718597412109375, 15.6278076171875, -25.446044921875, 0.0, 8.661384582519531, 2.837860107421875, 1.9987869262695312, 17.53106689453125, 10.423994064331055, 10.13128662109375, -34.5689697265625, 25.3763427734375, 5.581485748291016, -34.30999755859375, -1.8711624145507812, 5.1577911376953125, 23.051666259765625, 13.353195190429688, -4.0663299560546875, -9.576263427734375, 6.930538177490234, 2.121426582336426, 5.872062683105469, 0.764556884765625, 15.789398193359375, -12.4381103515625, -8.955078125, 0.0, 7.658023834228516, -4.326377868652344, -1.1017608642578125, -1.83721923828125, 16.734832763671875, -8.133079528808594, -0.845733642578125, 6.4662628173828125, 18.05572509765625, -13.722000122070312, -0.923370361328125, -17.556015014648438, -2.6146583557128906, 46.780517578125, 13.962760925292969, 9.177581787109375, 43.498504638671875, -5.87152099609375, -4.40478515625, -2.1286163330078125, -4.725116729736328, 13.9063720703125, -4.0899658203125, 14.384597778320312, -9.812889099121094, 0.05157470703125, 4.298362731933594, 7.825103759765625, 0.8602294921875, 16.099987030029297, 15.20166015625, -0.6034164428710938, -7.0071868896484375, 29.096145629882812, 46.255157470703125, 5.44183349609375, -25.754547119140625, -0.220916748046875, -19.776092529296875, 46.10333251953125, 17.154556274414062, 18.455520629882812, 18.116134643554688, 2.771697998046875, 34.014251708984375, 3.53839111328125, -23.2508544921875, 4.3555450439453125, 12.895263671875, 0.8380355834960938, 2.8621978759765625, 8.8492431640625, 14.149311065673828, 8.094879150390625, 4.0921173095703125, 16.024169921875, -1.31597900390625, 4.579345703125, -3.9834747314453125, -4.080623626708984, 7.338165283203125, 34.90447998046875, 8.66400146484375, 4.96282958984375, 7.531463623046875, 8.692253112792969, 29.6798095703125, -25.736801147460938, 28.236541748046875, 12.9378662109375, -0.15203094482421875, -14.620750427246094, 41.41615295410156, 13.111175537109375, 1.1067848205566406, -17.503799438476562, -23.42852783203125, 16.691513061523438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000121.npy"}
{"epoch": 0.2534031413612565, "step": 122, "batch_size": 128, "mean": 7.355593681335449, "std": 18.516237258911133, "min": -38.925445556640625, "p10": -16.795544433593747, "median": 6.758968353271484, "p90": 30.11563110351562, "max": 61.414794921875, "pos_frac": 0.65625, "sample": [21.558349609375, -17.76470947265625, 26.895339965820312, -15.6488037109375, 14.949188232421875, 23.188674926757812, 8.2252197265625, 3.488189697265625, -35.56048583984375, 29.5916748046875, 25.148040771484375, 0.0, 7.763092041015625, -26.698394775390625, 0.134490966796875, -38.925445556640625, 11.471023559570312, 16.454193115234375, -13.132843017578125, 14.725921630859375, -2.876220703125, 22.63311767578125, -22.928237915039062, -31.8431396484375, 8.18255615234375, 34.99365234375, -12.937408447265625, 11.12611198425293, 47.793304443359375, 2.4794235229492188, 9.537506103515625, -3.0968170166015625, 8.314239501953125, 7.882843017578125, 18.1175537109375, -4.600257873535156, 4.816822052001953, -23.1783447265625, 0.137908935546875, 10.854576110839844, -29.057891845703125, 11.084747314453125, 6.52276611328125, 46.56671142578125, -2.6868629455566406, -23.310867309570312, -10.410903930664062, 9.517547607421875, 41.103271484375, -8.437347412109375, -6.2974853515625, 9.534378051757812, 13.078277587890625, -22.433609008789062, 10.6925048828125, 15.035472869873047, 0.15580368041992188, 23.477035522460938, -30.833038330078125, 10.612060546875, 14.048126220703125, 21.0516357421875, 18.021835327148438, -0.748687744140625, -7.393669128417969, 33.287322998046875, 15.5872802734375, 24.3681640625, 19.20428466796875, 29.35345458984375, 45.376800537109375, 13.156953811645508, 1.389312744140625, 35.8302001953125, 61.414794921875, 6.995170593261719, 5.9727935791015625, -2.25146484375, 31.743408203125, 22.81304931640625, -3.737640380859375, -16.38018798828125, 4.22509765625, -9.522933959960938, -5.846435546875, -4.52301025390625, -0.704376220703125, 13.310821533203125, 5.8438262939453125, -5.916015625, 17.384979248046875, 0.0518798828125, 2.489412307739258, -18.56957244873047, -3.0993881225585938, -18.90667724609375, -0.7362213134765625, 26.302032470703125, 23.725189208984375, -1.0985107421875, 6.354339599609375, -6.362945556640625, 23.7701416015625, 18.1209716796875, 2.6670455932617188, 23.450927734375, 20.85430908203125, 28.89806365966797, 4.7186126708984375, 2.3404159545898438, -1.272003173828125, -0.296478271484375, -8.570442199707031, 14.818572998046875, 51.41070556640625, 1.506591796875, 33.1829833984375, -4.86614990234375, 3.448822021484375, 25.797500610351562, 13.02294921875, 20.33453369140625, -4.113958358764648, 37.52813720703125, -4.984001159667969, 1.616668701171875, 31.33819580078125, 8.10394287109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000122.npy"}
{"epoch": 0.2554973821989529, "step": 123, "batch_size": 128, "mean": 10.59537410736084, "std": 18.8223819732666, "min": -47.5244140625, "p10": -11.64691619873047, "median": 8.859619140625, "p90": 33.99426422119141, "max": 67.386474609375, "pos_frac": 0.7109375, "sample": [25.917037963867188, -10.752685546875, 29.025100708007812, -11.59783935546875, 7.273529052734375, 5.685558319091797, 9.961959838867188, 2.736419677734375, 13.554443359375, 9.86688232421875, -0.3721275329589844, 4.33740234375, -2.419464111328125, 1.823333740234375, 6.146345138549805, -10.020538330078125, -11.761428833007812, 16.532806396484375, 45.626922607421875, 8.206573486328125, -23.90765380859375, -23.60528564453125, 8.750244140625, 16.98798942565918, 50.51312255859375, 4.1510162353515625, 16.73455810546875, 29.639373779296875, 30.4827880859375, 1.1740036010742188, 28.17626953125, 23.89794921875, 20.9041748046875, 25.486785888671875, 8.378021240234375, 23.127685546875, 24.46484375, -24.05084228515625, 1.07879638671875, 28.90789794921875, 17.593841552734375, 4.34808349609375, 1.028961181640625, 34.4127197265625, 9.785354614257812, 56.481292724609375, -0.10752677917480469, 3.2078933715820312, 19.893157958984375, 30.071044921875, 3.1852378845214844, 18.684173583984375, 21.918731689453125, 67.386474609375, 46.87872314453125, 13.489242553710938, -7.62101936340332, -2.3057861328125, 1.08135986328125, 47.09796142578125, -2.5462188720703125, 5.42534065246582, -47.5244140625, 33.81492614746094, 8.90338134765625, -17.348419189453125, 21.783355712890625, 6.754974365234375, 37.51629638671875, 7.69378662109375, 8.81585693359375, -0.2597198486328125, 8.45406723022461, -19.141685485839844, 11.156990051269531, 26.530853271484375, 22.9296875, -3.14166259765625, -21.186492919921875, 43.25787353515625, -4.139862060546875, 11.580795288085938, 20.24911117553711, -7.5909423828125, 7.606498718261719, 19.638092041015625, -17.074951171875, 18.539993286132812, 18.579132080078125, -7.882904052734375, 0.0, -4.5126190185546875, 38.188201904296875, 11.588699340820312, 10.28607177734375, 0.0, 14.481552124023438, 9.954055786132812, 0.5657539367675781, 23.046630859375, -6.292530059814453, -13.390869140625, 8.42376708984375, 26.86920166015625, -0.2792205810546875, 12.519966125488281, 11.336700439453125, 35.39447021484375, 17.41522216796875, -16.812088012695312, -1.74615478515625, -4.6461181640625, 32.49591064453125, 8.80224609375, -30.994720458984375, 50.07086181640625, 14.921356201171875, -9.20489501953125, 14.02557373046875, -2.856689453125, -0.16920852661132812, 10.995086669921875, 7.269683837890625, 33.297882080078125, 45.480560302734375, -18.0826416015625, 17.924606323242188, 20.87591552734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000123.npy"}
{"epoch": 0.25759162303664923, "step": 124, "batch_size": 128, "mean": 9.606401443481445, "std": 17.348731994628906, "min": -49.01751708984375, "p10": -7.629048538208005, "median": 7.883923530578613, "p90": 34.17905120849609, "max": 57.852874755859375, "pos_frac": 0.7734375, "sample": [-0.524658203125, 29.42200469970703, -49.01751708984375, 17.4180908203125, -4.342487335205078, 2.7095565795898438, 25.189666748046875, 8.579330444335938, 36.469451904296875, 45.934478759765625, 4.8895263671875, -29.3614501953125, 6.2444915771484375, -7.0179901123046875, 22.2393798828125, 5.258819580078125, 6.3940277099609375, 20.012176513671875, -6.417205810546875, 13.022857666015625, 4.9556884765625, 5.6810760498046875, 0.0341644287109375, 9.52313232421875, 19.75787353515625, 3.3018341064453125, -0.60400390625, -4.894161224365234, 4.064140319824219, 9.032318115234375, 29.599853515625, -13.535858154296875, 5.022357940673828, 24.67974853515625, 44.019317626953125, 38.27226257324219, 18.67418670654297, 12.523117065429688, 5.537750244140625, 22.156951904296875, 21.955673217773438, 17.73529815673828, 39.613433837890625, 8.26654052734375, 3.4097843170166016, -2.4357452392578125, 0.287872314453125, 14.037338256835938, 47.398773193359375, -9.054851531982422, 10.371917724609375, -18.03805923461914, 6.7239837646484375, 0.0, 2.584268569946289, -2.8215484619140625, 24.46881103515625, 5.904571533203125, 3.52093505859375, -11.392570495605469, 34.14485168457031, -37.7916259765625, -27.419403076171875, 41.303955078125, 12.44635009765625, 24.3355712890625, 8.488616943359375, 31.86700439453125, 25.05150604248047, -0.216400146484375, 11.14697265625, 5.3845672607421875, -19.599945068359375, -27.22161865234375, 22.45276641845703, 1.9282798767089844, 3.882171630859375, 13.769973754882812, 34.25885009765625, 21.967926025390625, 10.74322509765625, 29.823974609375, 8.773361206054688, -17.143463134765625, -6.378753662109375, 6.493827819824219, 3.4045562744140625, 15.596511840820312, 34.81280517578125, 16.371219635009766, 2.75799560546875, 13.836341857910156, -2.38470458984375, 12.491043090820312, 12.595443725585938, 8.002695083618164, -1.235198974609375, 57.852874755859375, 12.475580215454102, 21.424407958984375, -3.2433013916015625, 2.031463623046875, 15.46441650390625, 1.2672576904296875, 2.934051513671875, 52.79266357421875, -10.435928344726562, 0.11773681640625, 8.249454498291016, 15.027099609375, -2.4309234619140625, 0.209075927734375, 5.778564453125, 11.129974365234375, 7.7651519775390625, 21.18239974975586, 8.759429931640625, 34.984832763671875, 0.2794208526611328, -2.2261734008789062, 5.203094482421875, 14.877838134765625, 23.55401611328125, 43.850860595703125, 3.67352294921875, 10.94287109375, 3.88336181640625, -13.93770980834961], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000124.npy"}
{"epoch": 0.25968586387434556, "step": 125, "batch_size": 128, "mean": 9.400023460388184, "std": 19.0484561920166, "min": -33.3873291015625, "p10": -11.578530883789062, "median": 6.499109268188477, "p90": 34.32494201660156, "max": 63.49945068359375, "pos_frac": 0.6484375, "sample": [6.1103363037109375, -9.51882553100586, -33.3873291015625, 40.44818115234375, 1.5823535919189453, 13.444427490234375, 15.82147216796875, -11.64947509765625, 3.2967891693115234, -16.3228759765625, 4.46221923828125, -2.0701828002929688, 34.135986328125, -0.7677001953125, 11.25445556640625, 24.767837524414062, -28.733734130859375, 15.174514770507812, 12.928268432617188, 14.776161193847656, -8.8306884765625, -3.4548606872558594, 21.99652099609375, -5.665580749511719, 14.41644287109375, 14.976516723632812, 10.515480041503906, 15.2725830078125, -4.4331512451171875, 45.3131103515625, 5.6713104248046875, -7.115333557128906, 11.360885620117188, -1.7309799194335938, 6.748538970947266, 33.62440490722656, 23.7445068359375, 34.765838623046875, 14.071395874023438, -4.028633117675781, 0.4426422119140625, -21.983612060546875, -11.480361938476562, -7.092166900634766, -5.500885009765625, 43.1751708984375, 0.998687744140625, -7.2620849609375, -32.60844421386719, 16.035987854003906, 15.60009765625, -16.05072021484375, 24.8070068359375, 63.49945068359375, 7.264244079589844, 13.827018737792969, 8.980857849121094, 20.1512451171875, 20.345062255859375, 28.1505126953125, -1.392364501953125, 38.310394287109375, 13.347152709960938, -0.9706554412841797, 2.4017333984375, 12.791229248046875, 3.4559326171875, 0.7806587219238281, 6.9896240234375, 5.7557830810546875, 54.275634765625, 10.613914489746094, 29.9794921875, 42.35186767578125, -12.418197631835938, -21.093734741210938, 6.2496795654296875, 23.91754150390625, 33.8057861328125, 24.750999450683594, 32.610992431640625, 4.568084716796875, -0.343658447265625, 43.477142333984375, -0.160888671875, 13.561279296875, -14.354537963867188, -7.14581298828125, 20.74925994873047, 32.110015869140625, 28.30634307861328, 3.038787841796875, -5.9682769775390625, 14.58673095703125, -10.214874267578125, -1.4695358276367188, -10.3028564453125, 28.697479248046875, 27.68646240234375, -0.9515190124511719, -7.00225830078125, 0.0, 3.117502212524414, -3.269012451171875, 41.635528564453125, -9.142379760742188, 1.184234619140625, 4.742942810058594, 5.8649139404296875, 44.297149658203125, -7.173366546630859, -11.548126220703125, 20.42510986328125, 30.897247314453125, 0.6427974700927734, 26.454925537109375, 50.388397216796875, 10.758453369140625, -15.36114501953125, 17.38983154296875, 28.2322998046875, -32.402130126953125, -7.741172790527344, 48.6551513671875, 14.266120910644531, 18.92572021484375, -2.2699813842773438, -15.4197998046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000125.npy"}
{"epoch": 0.2617801047120419, "step": 126, "batch_size": 128, "mean": 10.664873123168945, "std": 19.377689361572266, "min": -45.9986572265625, "p10": -11.23092041015625, "median": 8.417743682861328, "p90": 35.340325927734376, "max": 51.68359375, "pos_frac": 0.6953125, "sample": [21.298583984375, 50.549163818359375, 13.391403198242188, 17.88287353515625, 29.078857421875, 42.2052001953125, 13.357803344726562, 41.41828918457031, 3.6598663330078125, -2.2452392578125, -42.60302734375, 24.274032592773438, 2.6040306091308594, 18.579635620117188, 27.13772201538086, 5.177154541015625, 26.46533203125, -4.744873046875, -22.1815185546875, -7.589996337890625, -14.276756286621094, 30.554641723632812, -11.54229736328125, -6.697834014892578, 4.609130859375, -9.247039794921875, -5.710113525390625, -3.0237960815429688, 5.384674072265625, -8.89447021484375, 7.22613525390625, -2.9053497314453125, 17.980308532714844, 14.320480346679688, 31.5159912109375, -9.9276123046875, -8.557479858398438, -0.60113525390625, -1.683349609375, 4.935300827026367, 8.373786926269531, 23.285388946533203, 35.660675048828125, -1.2311019897460938, 16.176483154296875, 23.483642578125, 6.757747650146484, 17.248016357421875, -15.676498413085938, -3.425201416015625, 34.114471435546875, -0.141082763671875, 11.72174072265625, 50.130859375, 3.3521881103515625, 38.89781188964844, 16.796546936035156, 7.2050933837890625, 4.211700439453125, 21.583189010620117, 11.31671142578125, 29.93011474609375, 20.939697265625, 25.051788330078125, -12.83184814453125, 17.343780517578125, 4.721412658691406, 24.427719116210938, 22.617172241210938, 49.316650390625, 40.609893798828125, -4.47998046875, 26.754638671875, 28.693016052246094, -2.7500877380371094, 7.031982421875, 10.61187744140625, -8.710113525390625, 0.21145248413085938, -26.44580078125, 27.98004913330078, 10.84490966796875, -3.3767547607421875, 44.46087646484375, 24.982879638671875, 3.1703033447265625, -30.839508056640625, 4.63824462890625, 21.2882080078125, -35.81243896484375, 13.04754638671875, -9.566474914550781, 26.040771484375, 31.11279296875, -45.9986572265625, -9.1845703125, 4.515260696411133, 8.461700439453125, 5.59722900390625, -14.68896484375, 23.59881591796875, 27.769638061523438, 34.79362487792969, 42.507537841796875, 7.71533203125, 19.79925537109375, -6.674072265625, 46.477813720703125, 23.73974609375, 38.71148681640625, -11.99493408203125, 19.787017822265625, 17.594444274902344, 35.203033447265625, 20.213077545166016, 7.056976318359375, 4.320732116699219, -2.398162841796875, 10.71875, -0.4466705322265625, 13.849395751953125, -11.09747314453125, 1.7667236328125, 4.384002685546875, 29.783493041992188, 51.68359375, 4.525550842285156, -19.018707275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000126.npy"}
{"epoch": 0.2638743455497382, "step": 127, "batch_size": 128, "mean": 12.382560729980469, "std": 19.415616989135742, "min": -44.5009765625, "p10": -10.111090087890625, "median": 9.06610107421875, "p90": 38.82235412597656, "max": 56.5440673828125, "pos_frac": 0.734375, "sample": [-0.6553955078125, -0.983978271484375, -9.2320556640625, 24.53973388671875, 10.14453125, 3.447174072265625, 18.774749755859375, 12.250133514404297, -11.341842651367188, -44.5009765625, 29.1513671875, 3.1173858642578125, 20.443206787109375, 16.793142318725586, 8.113311767578125, -32.773651123046875, -8.128082275390625, 13.269195556640625, -17.276763916015625, 6.43768310546875, 7.272613525390625, 7.086517333984375, 20.568115234375, 5.217864990234375, 17.356979370117188, 45.8734130859375, 21.45208740234375, 35.04766845703125, 12.433677673339844, 25.15345001220703, 38.7374267578125, 14.807405471801758, 30.74884033203125, 4.335348129272461, 10.33477783203125, 46.04302978515625, -9.253387451171875, -19.041946411132812, -1.8480453491210938, 10.377769470214844, 49.27008056640625, 7.421852111816406, 8.927444458007812, 5.3919677734375, -15.33837890625, -23.738906860351562, 35.031005859375, 36.52085876464844, 13.539276123046875, -23.35784912109375, 18.547515869140625, 35.803497314453125, 22.057357788085938, -6.8796844482421875, 29.453704833984375, 40.1400146484375, 25.271820068359375, 15.9893798828125, -3.4183349609375, 3.3187484741210938, 2.58074951171875, 2.2203369140625, 36.78834533691406, 21.61065673828125, 9.204757690429688, 34.087646484375, -7.2164764404296875, 5.9840087890625, 31.487884521484375, 39.25408935546875, 26.75994873046875, -2.2965011596679688, -11.742050170898438, -2.5118980407714844, 10.324325561523438, -10.448379516601562, -0.6356201171875, -14.826473236083984, 6.8485870361328125, -3.441387176513672, 4.0350341796875, 23.459442138671875, 54.4461669921875, -1.6677742004394531, 39.42950439453125, 18.001785278320312, 3.821319580078125, 56.5440673828125, 33.0888671875, 20.96148681640625, 46.8436279296875, -8.669723510742188, 6.43798828125, 7.6837158203125, 4.036334991455078, -1.9657955169677734, -21.179304122924805, 8.055908203125, 10.519647598266602, 31.446304321289062, -5.441131591796875, 7.6407012939453125, -9.966537475585938, 30.46082305908203, 23.49498748779297, 4.341438293457031, 39.916015625, 50.1981201171875, 1.477996826171875, -4.1175994873046875, 6.4536285400390625, 2.927947998046875, 25.45196533203125, 35.323974609375, -17.695701599121094, 4.953300476074219, 38.889495849609375, 38.7935791015625, -2.3773841857910156, 11.40899658203125, 31.124847412109375, 24.61212158203125, 2.23236083984375, 31.769821166992188, 1.76470947265625, -8.41259765625, 23.823898315429688, 48.31292724609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000127.npy"}
{"epoch": 0.26596858638743454, "step": 128, "batch_size": 128, "mean": 13.766580581665039, "std": 20.95404624938965, "min": -43.13037109375, "p10": -10.044042968749999, "median": 13.162226676940918, "p90": 38.45906448364258, "max": 74.100830078125, "pos_frac": 0.7109375, "sample": [27.78973388671875, 3.839080810546875, 19.909027099609375, 25.9390869140625, -32.537078857421875, -33.187164306640625, -20.32757568359375, 6.70318603515625, 26.265655517578125, 29.975265502929688, 23.30194854736328, 3.088287353515625, 35.508209228515625, -0.25611114501953125, 38.84959411621094, 0.9996089935302734, 3.0106468200683594, 25.416580200195312, 26.217605590820312, 25.734939575195312, -3.8863372802734375, -0.577667236328125, 36.60558319091797, 67.677734375, 5.10125732421875, 8.223953247070312, -3.0726318359375, 4.37567138671875, 28.828460693359375, 20.892486572265625, 9.952186584472656, 11.84528923034668, -19.87255859375, 34.24635314941406, -9.043954849243164, 31.895675659179688, -16.873443603515625, 6.718681335449219, -0.453887939453125, -21.662567138671875, 24.84937286376953, 2.914764404296875, -2.303192138671875, 17.716094970703125, 33.41288757324219, -2.149627685546875, 0.5108909606933594, 24.449356079101562, 19.94134521484375, 38.29169464111328, 35.8619384765625, 26.60107421875, -3.6663742065429688, 34.9952392578125, -2.084228515625, -5.1270904541015625, 3.8924560546875, 25.16360855102539, 19.389434814453125, 41.7265625, 28.14754867553711, 4.00775146484375, 14.481922149658203, -0.259246826171875, 1.292144775390625, 8.092849731445312, 42.02392578125, 29.793701171875, 28.053985595703125, 14.479164123535156, 15.778564453125, 29.797073364257812, 50.89012145996094, 19.5985107421875, 47.43690490722656, 37.4656982421875, 9.953971862792969, -5.0089111328125, -3.3587646484375, -1.1340522766113281, 40.48577880859375, 32.101806640625, -13.73822021484375, 19.668792724609375, 29.600555419921875, -43.13037109375, 20.58941650390625, -21.709686279296875, 19.51605987548828, 74.100830078125, 8.908340454101562, -9.75927734375, 35.90892028808594, 15.5252685546875, -6.841400146484375, 0.8650016784667969, -16.956382751464844, -8.310195922851562, 7.53173828125, 11.496688842773438, 36.485992431640625, -6.035186767578125, 7.75152587890625, 45.961326599121094, 22.974830627441406, 34.694580078125, -7.564643859863281, -6.497467041015625, 55.53985595703125, 1.0521316528320312, 8.66448974609375, 0.0, -22.177230834960938, -7.3585205078125, 26.1890869140625, -10.99896240234375, -10.70849609375, 15.880859375, 24.79437255859375, 43.089447021484375, 3.91595458984375, 18.9140625, 11.395782470703125, 41.609954833984375, 61.448272705078125, 25.909423828125, 19.761474609375, -1.5042495727539062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000128.npy"}
{"epoch": 0.2680628272251309, "step": 129, "batch_size": 128, "mean": 12.544189453125, "std": 21.704347610473633, "min": -48.766876220703125, "p10": -13.019439697265623, "median": 10.790639877319336, "p90": 42.20004577636718, "max": 78.92355346679688, "pos_frac": 0.765625, "sample": [-48.766876220703125, 28.2763671875, -23.96148681640625, -14.48101806640625, 12.274742126464844, 45.459781646728516, 27.030670166015625, 36.65467071533203, 30.936187744140625, 2.255767822265625, 31.253173828125, 1.639657974243164, 15.24591064453125, -1.003997802734375, -6.6704254150390625, -11.351776123046875, 29.063323974609375, 34.13531494140625, 12.798110961914062, -11.356094360351562, 11.775672912597656, 12.543716430664062, 1.3066329956054688, 18.257904052734375, 43.103240966796875, -5.5909423828125, 17.493194580078125, -2.3745269775390625, 78.92355346679688, 61.1641845703125, 5.871337890625, 23.360855102539062, 52.989410400390625, 15.79473876953125, 1.5290603637695312, -44.90203857421875, 35.00213623046875, 4.96038818359375, 62.700439453125, 10.771038055419922, -4.9382781982421875, 14.450653076171875, 40.875732421875, 45.67169189453125, 10.81024169921875, 42.774627685546875, -24.153350830078125, 9.94035530090332, 0.9034385681152344, 4.304801940917969, 26.994110107421875, 17.337989807128906, 9.498199462890625, 26.5184326171875, -12.44036865234375, 8.974651336669922, 5.049896240234375, 43.9256591796875, 3.366790771484375, 28.393539428710938, 1.7010574340820312, 3.2649307250976562, 41.95379638671875, 22.59381103515625, 60.641693115234375, -8.385986328125, 16.378448486328125, 17.9246826171875, 26.096405029296875, 4.728271484375, 5.6883544921875, 6.480995178222656, 3.0354766845703125, -5.467033386230469, 3.116943359375, -14.37060546875, 27.587051391601562, 22.647933959960938, -2.6069602966308594, 60.497406005859375, 27.690536499023438, 10.375961303710938, 19.93365478515625, 3.5589447021484375, 2.561786651611328, 2.31463623046875, 30.485122680664062, 3.8127288818359375, 20.587738037109375, 38.3804931640625, -11.45269775390625, -29.2650146484375, -2.481159210205078, 0.153900146484375, -7.056671142578125, 12.8941650390625, -10.3536376953125, -32.6470947265625, 16.501876831054688, 10.75030517578125, 46.35125732421875, -15.300521850585938, 13.500396728515625, -3.5673294067382812, -19.569610595703125, -22.23504638671875, 11.44781494140625, 3.41461181640625, 14.446006774902344, 22.48272705078125, 27.626388549804688, 6.39857292175293, 15.7593994140625, 9.06964111328125, 10.180225372314453, 46.77508544921875, 10.891159057617188, 31.147476196289062, -18.06512451171875, 19.961746215820312, 10.465614318847656, -0.38535118103027344, 2.8575897216796875, 30.291580200195312, 13.857925415039062, -14.760971069335938, 27.5145263671875, 32.47736358642578], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000129.npy"}
{"epoch": 0.27015706806282724, "step": 130, "batch_size": 128, "mean": 11.6906099319458, "std": 19.707441329956055, "min": -38.9683837890625, "p10": -11.013728332519529, "median": 10.085762023925781, "p90": 37.50721969604491, "max": 56.41712951660156, "pos_frac": 0.71875, "sample": [9.32110595703125, -3.2347869873046875, 31.0140380859375, 5.0963134765625, 5.2846221923828125, 41.65850830078125, 11.641204833984375, 29.388671875, 39.59784698486328, -9.159027099609375, -3.192413330078125, 8.78558349609375, 22.998367309570312, 41.130889892578125, -1.084197998046875, 5.5516357421875, -3.444366455078125, -18.40338134765625, 6.604301452636719, 20.55340576171875, 16.708740234375, 40.37969970703125, -1.5685043334960938, 10.193649291992188, -2.0303688049316406, 56.41712951660156, 2.74371337890625, 21.11181640625, 46.265777587890625, 49.948883056640625, -31.4918212890625, -0.8485107421875, 0.0, -10.037185668945312, 26.1297607421875, 11.342453002929688, 24.83160400390625, -36.492431640625, 8.629676818847656, 22.449310302734375, 25.5802001953125, 14.597808837890625, 51.142578125, -10.137893676757812, -38.9683837890625, -36.275421142578125, -4.90692138671875, 34.48280334472656, 3.534332275390625, -0.11057281494140625, 35.50128173828125, 2.508068084716797, 10.285362243652344, 33.382362365722656, -16.24896240234375, -1.8218994140625, 12.84455680847168, 18.861083984375, 36.611236572265625, -3.806884765625, 9.977874755859375, 0.9202919006347656, 22.457000732421875, 17.341583251953125, 5.792938232421875, 43.934051513671875, 27.24188232421875, -17.1170654296875, -13.057342529296875, -1.1310577392578125, 39.6397705078125, 8.555351257324219, 30.0859375, -23.32672119140625, 4.985076904296875, 18.16217041015625, -15.452377319335938, 29.48919677734375, 10.56817626953125, 15.925304412841797, 28.452423095703125, 16.59814453125, 29.94549560546875, 16.484375, 33.03564453125, 0.8852939605712891, 31.118072509765625, 8.244796752929688, 28.108367919921875, -3.6415557861328125, 13.12554931640625, 8.30279541015625, 6.853546142578125, 3.1439247131347656, 23.288497924804688, -28.306793212890625, -5.76318359375, 5.0748748779296875, 3.673736572265625, 48.4027099609375, 43.20947265625, 35.65643310546875, 21.1961669921875, 1.3037643432617188, -25.92987060546875, 27.147064208984375, 6.241035461425781, 4.764892578125, 19.467742919921875, -6.2826080322265625, 35.53125, 7.297649383544922, 27.705810546875, -6.601287841796875, 19.80572509765625, 19.270977020263672, 41.3099365234375, 19.05010986328125, -0.8804168701171875, 16.867416381835938, 15.529518127441406, -7.032684326171875, -1.2028350830078125, 2.4475784301757812, 29.84814453125, -25.364837646484375, 5.199739456176758, 26.977020263671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000130.npy"}
{"epoch": 0.27225130890052357, "step": 131, "batch_size": 128, "mean": 9.831412315368652, "std": 22.23431968688965, "min": -46.52604675292969, "p10": -18.229405212402344, "median": 9.107967376708984, "p90": 39.38919677734375, "max": 71.99053955078125, "pos_frac": 0.640625, "sample": [31.584503173828125, 9.238418579101562, 37.519439697265625, -22.81060791015625, 13.171585083007812, 2.8318634033203125, 2.2037391662597656, 24.480499267578125, -18.792953491210938, 40.547576904296875, 31.165847778320312, 30.926025390625, 42.741546630859375, 19.5924072265625, -16.834320068359375, 1.4486236572265625, 1.90777587890625, -15.539306640625, 9.087364196777344, 47.69367980957031, 31.010162353515625, 11.104156494140625, -6.2796630859375, -5.413352966308594, 30.320213317871094, -6.4595794677734375, -10.020965576171875, 44.04315185546875, 10.862457275390625, 14.412872314453125, -1.8742294311523438, 60.00227355957031, -6.561798095703125, 1.26715087890625, 20.908172607421875, -2.6340465545654297, -23.31450653076172, 43.66143798828125, -2.900054931640625, 30.53924560546875, 7.247314453125, 3.2825927734375, -1.7511329650878906, -16.093109130859375, 24.9739990234375, 6.642250061035156, 57.775726318359375, -17.09698486328125, 19.862991333007812, 25.9356689453125, -2.386566162109375, -3.59698486328125, -17.987884521484375, 15.9630126953125, 13.807373046875, 39.023406982421875, -24.16595458984375, -5.758453369140625, 26.003902435302734, 36.52375793457031, -9.039566040039062, -17.96673583984375, -20.28411865234375, 31.68706512451172, 8.11212158203125, 6.539104461669922, 16.394004821777344, 27.68377685546875, 71.99053955078125, 9.14776611328125, -46.52604675292969, 4.141960144042969, 10.1319580078125, -8.890565872192383, -3.697345733642578, 18.937347412109375, -27.754745483398438, 62.7816162109375, -7.280529022216797, 36.02287292480469, 10.97930908203125, -22.54876708984375, 2.0484085083007812, -13.343666076660156, 11.335662841796875, 20.27056884765625, -15.59130859375, 38.665618896484375, 42.90606689453125, -24.374176025390625, 0.8862228393554688, -3.37335205078125, 44.128662109375, -20.794429779052734, 26.81072998046875, 5.2296905517578125, 1.6918830871582031, 9.786567687988281, -7.3383026123046875, 17.39946746826172, 11.698463439941406, -19.02001953125, 28.53802490234375, -3.6990127563476562, 11.39453125, 36.627716064453125, -5.354400634765625, -33.351318359375, 52.36090087890625, 8.958236694335938, 34.58050537109375, 40.242706298828125, 14.687713623046875, 21.52923583984375, -11.7225341796875, -3.1834869384765625, 20.642303466796875, -8.5174560546875, 28.081209182739258, 9.128570556640625, 23.19512939453125, 15.197235107421875, 24.92913818359375, 17.710479736328125, -6.76092529296875, -21.089130401611328, -10.854095458984375, 2.523937225341797], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000131.npy"}
{"epoch": 0.2743455497382199, "step": 132, "batch_size": 128, "mean": 13.099861145019531, "std": 24.94384765625, "min": -46.536346435546875, "p10": -16.969223022460938, "median": 10.582038879394531, "p90": 44.95895385742187, "max": 95.828369140625, "pos_frac": 0.6875, "sample": [-7.99383544921875, 26.3570556640625, 38.50605773925781, 13.913978576660156, -0.04766845703125, -10.551513671875, 27.451873779296875, 10.828582763671875, -2.7896728515625, 37.750946044921875, 4.04925537109375, -10.573722839355469, -20.6883544921875, 34.0989990234375, 22.839096069335938, 9.802139282226562, 4.185577392578125, 46.1165771484375, 41.94041442871094, 60.056549072265625, 34.986968994140625, 31.45147705078125, 71.5897216796875, 48.8380126953125, 15.572036743164062, 1.4487800598144531, -0.73419189453125, -33.31196594238281, 14.707061767578125, 20.288116455078125, -2.1010284423828125, 47.615875244140625, 3.1580963134765625, 20.5916748046875, -24.02020263671875, -17.4462890625, -16.50403594970703, 10.335494995117188, 4.5993194580078125, 2.732208251953125, 43.364227294921875, 15.352279663085938, -11.695968627929688, -14.784744262695312, -23.227401733398438, -0.27710723876953125, 6.176422119140625, -17.221832275390625, 12.297271728515625, -12.951988220214844, 17.781082153320312, -10.510520935058594, 16.489864349365234, 30.774497985839844, 4.162078857421875, -0.0103759765625, 4.56610107421875, 18.928985595703125, 95.828369140625, -0.284912109375, 26.380615234375, 30.338882446289062, 22.467071533203125, -24.4207763671875, 36.689903259277344, -2.34051513671875, -2.9503440856933594, 13.21063232421875, -20.55023193359375, 1.584991455078125, 17.01416015625, 4.295536041259766, 14.740447998046875, 35.87225341796875, -4.7406158447265625, -0.87542724609375, 0.0, 36.044952392578125, 13.161331176757812, -19.0125732421875, 9.290573120117188, 38.32916259765625, 51.316162109375, -32.9521484375, -46.536346435546875, 56.490234375, 13.922225952148438, 19.13018798828125, 2.955078125, 10.09844970703125, 22.8642578125, 12.996551513671875, 33.43556213378906, 24.536880493164062, 51.81877136230469, 91.59521484375, -15.30010986328125, 27.478897094726562, 0.96319580078125, 14.002685546875, 26.58001708984375, 41.241363525390625, 1.71759033203125, 0.8683013916015625, 11.180908203125, 51.44390869140625, 4.484832763671875, 5.9845428466796875, 0.0, -16.77093505859375, 32.667877197265625, 41.18864440917969, -9.767059326171875, 1.391082763671875, -10.318145751953125, 9.996307373046875, -4.6326904296875, 56.109954833984375, -27.7242431640625, 44.46282958984375, 34.575477600097656, 28.60919189453125, -23.667312622070312, 15.066207885742188, -16.8609619140625, 16.352325439453125, 1.098846435546875, 64.35189819335938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000132.npy"}
{"epoch": 0.2764397905759162, "step": 133, "batch_size": 128, "mean": 13.54984188079834, "std": 22.529659271240234, "min": -66.2938232421875, "p10": -12.958106231689449, "median": 11.786773681640625, "p90": 42.69796142578125, "max": 69.83294677734375, "pos_frac": 0.7578125, "sample": [20.5628662109375, 1.6961193084716797, -34.158538818359375, 0.0, 1.230743408203125, 33.34291076660156, -1.179117202758789, 13.973098754882812, 4.6579437255859375, 29.983322143554688, 16.85308837890625, 17.162811279296875, 6.16900634765625, 25.30535888671875, 39.22857666015625, 51.320526123046875, -9.480941772460938, -0.21051025390625, -1.4084091186523438, -17.27923583984375, 33.121246337890625, 48.4073486328125, 69.83294677734375, 5.571464538574219, 25.372222900390625, 45.056671142578125, 22.678306579589844, 53.1641845703125, 0.0, -1.3439788818359375, 21.018524169921875, 23.559234619140625, -4.746013641357422, 1.769195556640625, 26.413070678710938, 32.86036682128906, 23.159683227539062, 25.224624633789062, -25.671600341796875, 34.49188232421875, 42.22645568847656, 6.78582763671875, -5.1711578369140625, -6.760498046875, 10.9693603515625, 0.71112060546875, -17.18805694580078, 8.847061157226562, 5.4536590576171875, 8.63079833984375, 16.988739013671875, 39.567626953125, 3.98712158203125, 29.5010986328125, 48.919708251953125, 7.8992462158203125, 6.5918426513671875, 42.9097900390625, -3.9490203857421875, 31.606658935546875, -23.4949951171875, 12.006072998046875, 35.903076171875, -0.169677734375, 18.00526237487793, 7.5841064453125, 25.752487182617188, 43.0118408203125, 32.0545654296875, 29.805870056152344, -38.355712890625, 0.2998199462890625, 11.567474365234375, 8.495254516601562, 51.51641845703125, 9.410039901733398, -6.770545959472656, -21.889190673828125, 4.419513702392578, 45.99859619140625, -6.806232452392578, 0.944366455078125, 9.330230712890625, -9.0296630859375, 20.91986083984375, -11.603874206542969, 7.9888916015625, 26.821434020996094, 25.441162109375, 42.607177734375, -5.0798797607421875, 46.678802490234375, 29.736480712890625, 14.373214721679688, 15.807144165039062, 3.8919677734375, 7.2001953125, 41.226318359375, 25.30389404296875, 45.34010314941406, 7.386039733886719, 30.492599487304688, 21.42328643798828, 25.938709259033203, -28.248291015625, 21.153961181640625, 22.440521240234375, 27.087203979492188, -25.7503662109375, 48.68975830078125, 21.119842529296875, 8.314002990722656, 5.51708984375, -16.11798095703125, -49.90220642089844, 17.072341918945312, -66.2938232421875, 4.039653778076172, 2.8040924072265625, 37.8623046875, 5.708221435546875, 24.275238037109375, 36.553749084472656, -5.3478851318359375, -25.28179931640625, 2.2094478607177734, 28.802001953125, 39.95384979248047], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000133.npy"}
{"epoch": 0.27853403141361255, "step": 134, "batch_size": 128, "mean": 9.49384880065918, "std": 21.254802703857422, "min": -44.343505859375, "p10": -15.371340942382812, "median": 7.639434814453125, "p90": 33.86039733886719, "max": 92.9512939453125, "pos_frac": 0.6796875, "sample": [40.18901062011719, 26.399749755859375, -17.9129638671875, -5.493438720703125, -24.84490966796875, -16.566909790039062, 8.43896484375, -15.7698974609375, -4.81427001953125, 25.3795166015625, -35.7158203125, -5.579681396484375, 11.02862548828125, 15.64471435546875, 28.253036499023438, -0.6471710205078125, -5.9394378662109375, 9.163753509521484, 23.4461669921875, 7.283016204833984, 12.326416015625, 60.467041015625, 0.0, 19.340087890625, 24.15606689453125, 22.967498779296875, 32.5135498046875, -3.005645751953125, 6.0806884765625, 46.556976318359375, -29.44147300720215, -5.625091552734375, 15.412567138671875, 3.216533660888672, 29.016754150390625, 3.3150177001953125, 26.762619018554688, 16.144195556640625, 15.2498779296875, 5.51080322265625, -10.185409545898438, 25.694297790527344, 14.447998046875, 26.07244873046875, 92.9512939453125, 33.98590087890625, 4.37261962890625, -13.60626220703125, 70.02886962890625, 28.216644287109375, -21.92218017578125, 34.515811920166016, 9.625457763671875, -13.76815414428711, 0.09600067138671875, 2.3871688842773438, 31.652984619140625, 20.48150634765625, 39.91876220703125, 8.276008605957031, -10.435516357421875, -3.56842041015625, 8.859588623046875, 7.93438720703125, -29.13934326171875, 42.39338684082031, -2.7065887451171875, 15.01080322265625, 24.0523681640625, 10.576530456542969, 1.9746036529541016, 15.6861572265625, -17.62036895751953, -4.417184829711914, 1.337249755859375, 42.80816650390625, 10.75726318359375, 26.99700927734375, 7.344482421875, -33.73289489746094, 17.106689453125, 32.8182373046875, 29.000701904296875, -11.0286865234375, -2.16473388671875, -3.34771728515625, 11.207366943359375, 14.605764389038086, 44.7769775390625, 0.8161067962646484, -0.7083206176757812, -4.8713226318359375, 9.674358367919922, 0.9766845703125, -44.343505859375, 26.994537353515625, 20.047088623046875, 13.7415771484375, 7.2900390625, -9.2344970703125, 6.576709747314453, 14.7294921875, -14.62628173828125, -41.666046142578125, -16.593658447265625, 20.870849609375, -15.200531005859375, 1.111602783203125, 35.63063049316406, 2.76214599609375, 7.113861083984375, 32.399078369140625, 0.43631744384765625, 20.057861328125, 38.54095458984375, 26.306026458740234, 33.57099151611328, 2.320068359375, 4.87860107421875, -7.707557678222656, -1.78582763671875, 0.4931640625, -3.4767227172851562, -3.9017333984375, 33.806610107421875, 19.506385803222656, 5.442413330078125, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000134.npy"}
{"epoch": 0.2806282722513089, "step": 135, "batch_size": 128, "mean": 13.281157493591309, "std": 21.137256622314453, "min": -50.8642578125, "p10": -10.04712333679199, "median": 9.803031921386719, "p90": 45.13445434570312, "max": 60.42340087890625, "pos_frac": 0.78125, "sample": [0.4453582763671875, -26.04736328125, 11.070762634277344, 22.678054809570312, -9.55966567993164, 43.16973876953125, 0.9853830337524414, 13.770751953125, 30.703414916992188, 3.6226158142089844, 53.876739501953125, 16.977142333984375, -6.3492279052734375, -3.247344970703125, 11.008087158203125, 4.3753662109375, 9.172080993652344, 25.92938232421875, 24.296737670898438, 2.5801010131835938, 1.459869384765625, 7.0183563232421875, -0.21492767333984375, 2.045520782470703, -31.82403564453125, 0.6534385681152344, 34.260345458984375, 26.004112243652344, 19.63097381591797, 46.591888427734375, 33.325164794921875, 0.6406364440917969, 29.934326171875, 5.5992889404296875, -16.71734619140625, 41.1400146484375, 3.267364501953125, 16.96417236328125, 27.33050537109375, 8.874008178710938, 4.763427734375, 12.945465087890625, 14.44952392578125, 38.251953125, 57.901123046875, 18.384902954101562, 43.689178466796875, 10.789810180664062, 1.00872802734375, -3.27825927734375, 19.584556579589844, 21.247039794921875, 2.1825485229492188, -11.184524536132812, -17.950775146484375, -8.43939208984375, 44.61163330078125, 5.8497772216796875, 40.70947265625, 16.954193115234375, 23.3912353515625, 15.542388916015625, 33.60902404785156, -5.48565673828125, -2.5283966064453125, 9.300613403320312, -3.840911865234375, 20.05010986328125, 19.586273193359375, 3.6766510009765625, 16.438873291015625, 6.944149017333984, 50.07916259765625, 17.002639770507812, 34.17615509033203, 31.681663513183594, 16.507965087890625, 5.850372314453125, -5.714385986328125, 51.831390380859375, 19.63775634765625, -18.44830322265625, -0.9498367309570312, 0.50860595703125, 8.3258056640625, 50.93804931640625, 0.74163818359375, 53.86749267578125, 36.230194091796875, 49.80126953125, 20.208526611328125, 42.1580810546875, 4.287628173828125, -40.1099853515625, 46.3543701171875, 7.0164642333984375, 6.825111389160156, 7.838991165161133, 1.328216552734375, 21.850677490234375, 7.808647155761719, 12.44329833984375, 29.857284545898438, 60.42340087890625, 2.4601669311523438, 0.392303466796875, 5.205272674560547, 15.33221435546875, 48.437347412109375, 58.77093505859375, -16.5010986328125, -15.515777587890625, 1.724395751953125, 10.305450439453125, 0.0, 21.607513427734375, -20.240234375, -50.8642578125, -1.1808242797851562, 11.142036437988281, -6.03155517578125, 20.297897338867188, 5.0963134765625, 48.1024169921875, 29.576263427734375, -13.797637939453125, -18.49658203125, -0.789215087890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000135.npy"}
{"epoch": 0.28272251308900526, "step": 136, "batch_size": 128, "mean": 12.740806579589844, "std": 23.151615142822266, "min": -54.1614990234375, "p10": -9.85479278564453, "median": 10.834369659423828, "p90": 42.183824157714845, "max": 83.17678833007812, "pos_frac": 0.75, "sample": [44.61065673828125, 17.363258361816406, 1.9813232421875, 15.04312515258789, 33.74903106689453, 38.042877197265625, 46.009185791015625, 11.765167236328125, 33.23699188232422, 58.686981201171875, 6.2938385009765625, 4.17912483215332, 42.55303955078125, 6.257682800292969, 26.182666778564453, 40.44160461425781, 14.548797607421875, 38.831207275390625, 7.961944580078125, -0.11270904541015625, -44.7032470703125, 36.643592834472656, 8.4136962890625, 24.320648193359375, -19.39508056640625, 4.665626525878906, -7.159759521484375, -42.11737060546875, 7.605415344238281, 12.649703979492188, 4.51849365234375, -0.26914215087890625, -5.902587890625, 9.08917236328125, 25.331787109375, -23.745773315429688, -7.58355712890625, 29.789947509765625, 15.918342590332031, -5.2490386962890625, 27.36492919921875, -21.208038330078125, 59.10076904296875, -54.1614990234375, 12.210845947265625, -6.264190673828125, 5.36492919921875, 5.294830322265625, 18.7735595703125, -4.511756896972656, 15.717010498046875, -34.5689697265625, 42.16218566894531, 17.845458984375, 33.4129638671875, 71.5257568359375, -8.785400390625, 7.2201995849609375, 10.08514404296875, 32.47593688964844, 43.84814453125, 19.95751953125, 37.061126708984375, -13.601364135742188, 10.254730224609375, 0.7200164794921875, -1.161163330078125, 13.6990966796875, 34.259368896484375, -2.8536376953125, 3.3881072998046875, 16.192626953125, 8.232967376708984, 11.971038818359375, 5.1136932373046875, 55.262939453125, -1.0900917053222656, 15.166748046875, 11.751850128173828, -9.579315185546875, -0.109649658203125, -26.87847900390625, 23.669097900390625, 59.548919677734375, 35.47772216796875, 26.240829467773438, 2.189451217651367, 17.561668395996094, -8.031896591186523, 47.243804931640625, 18.14264678955078, 8.66845703125, -10.497573852539062, 11.501396179199219, 4.7266845703125, 0.412506103515625, 58.633453369140625, 13.709358215332031, 28.77440643310547, 22.836334228515625, -0.21544647216796875, 5.500072479248047, 34.359130859375, 1.2520332336425781, 30.68505859375, -13.169403076171875, -8.32440185546875, 2.4984893798828125, 36.5794677734375, 42.23431396484375, 35.58094787597656, 11.414009094238281, -20.10455322265625, -7.079742431640625, 4.217071533203125, 8.8182373046875, 83.17678833007812, 4.751720428466797, -0.6590576171875, 8.69635009765625, 12.48114013671875, 15.19744873046875, 13.079963684082031, 3.0786895751953125, 12.91455078125, -51.54266357421875, 28.947662353515625, 4.540504455566406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000136.npy"}
{"epoch": 0.2848167539267016, "step": 137, "batch_size": 128, "mean": 11.197776794433594, "std": 22.226604461669922, "min": -49.71575927734375, "p10": -16.3223934173584, "median": 8.446483612060547, "p90": 37.17569427490234, "max": 75.4429931640625, "pos_frac": 0.765625, "sample": [-30.20458984375, 2.19256591796875, 37.85009765625, -7.0250701904296875, 11.392990112304688, 19.66558837890625, 37.52241516113281, 12.730743408203125, 3.7601470947265625, 1.7366790771484375, 16.790611267089844, 15.201385498046875, 25.7498779296875, 56.3123779296875, 16.730514526367188, 5.088775634765625, 0.1800079345703125, 32.5740966796875, 0.66741943359375, -15.09771728515625, 29.897506713867188, 20.699066162109375, 8.5721435546875, 10.037017822265625, -43.588348388671875, 24.14410400390625, 4.35205078125, -17.21099853515625, 30.845245361328125, 7.7842864990234375, 47.496795654296875, 14.1151123046875, -4.2687530517578125, 38.79449462890625, -18.193756103515625, -1.58868408203125, 4.4749908447265625, 7.953269958496094, 1.5584564208984375, -9.260551452636719, 7.29974365234375, -4.123106002807617, -20.347618103027344, 0.62603759765625, -49.71575927734375, 5.71588134765625, 8.120658874511719, 16.186660766601562, -45.61116027832031, 16.876018524169922, 8.28094482421875, 45.664215087890625, 7.9056396484375, 11.0399169921875, 17.6124267578125, 34.998687744140625, -18.0872802734375, 39.3807373046875, -17.98199462890625, 1.1247711181640625, 6.22369384765625, -1.2451324462890625, 27.7772216796875, -0.3842315673828125, 12.606842041015625, 58.096923828125, -27.4886474609375, 26.204267501831055, 8.6715087890625, -3.858978271484375, 1.51239013671875, 24.344390869140625, 30.03460693359375, 11.35073471069336, 35.166316986083984, 2.228687286376953, 0.8164882659912109, 75.4429931640625, 28.491470336914062, 14.771163940429688, 5.2873992919921875, 58.98577880859375, -2.2668991088867188, 35.761749267578125, 17.54815673828125, 19.239990234375, 57.425537109375, 18.536117553710938, -4.961692810058594, 34.511505126953125, 0.8256988525390625, -12.544967651367188, -0.8353805541992188, 8.080108642578125, 1.6080398559570312, 65.8807373046875, 25.2255859375, 3.385894775390625, 18.749958038330078, 12.550064086914062, -38.15155029296875, -31.02935791015625, 7.901336669921875, -33.518341064453125, -11.646369934082031, 53.15531921386719, -15.94156265258789, 31.47650146484375, 37.027099609375, 33.481239318847656, 25.994171142578125, 33.145263671875, 0.7271728515625, 23.464935302734375, 15.2266845703125, 24.373180389404297, -2.7832794189453125, 5.468841552734375, 2.386627197265625, 6.17022705078125, 8.320823669433594, 21.38146209716797, 4.56146240234375, 13.723724365234375, 15.138336181640625, 21.80950927734375, 27.716094970703125, -5.4179840087890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000137.npy"}
{"epoch": 0.2869109947643979, "step": 138, "batch_size": 128, "mean": 11.132314682006836, "std": 24.747941970825195, "min": -49.1978759765625, "p10": -17.908493041992188, "median": 8.444190979003906, "p90": 42.246345520019524, "max": 81.2147216796875, "pos_frac": 0.6796875, "sample": [69.51177978515625, 41.666748046875, 43.59873962402344, 17.481231689453125, -12.147567749023438, 5.024993896484375, -6.19891357421875, 66.77590942382812, 7.9462890625, 23.589569091796875, 10.710006713867188, 8.86358642578125, 4.7860870361328125, 14.439407348632812, 12.184089660644531, -31.185073852539062, -18.855072021484375, 27.75408935546875, -13.0322265625, -29.002777099609375, 21.241989135742188, 8.154754638671875, -13.519378662109375, -8.542449951171875, -49.1978759765625, 25.837478637695312, -4.176643371582031, 35.4268798828125, 2.4191131591796875, 81.2147216796875, -6.76251220703125, 7.08721923828125, 3.739044189453125, 43.62859344482422, 1.0354194641113281, 25.128921508789062, -27.975555419921875, 11.923622131347656, 40.76922607421875, -1.329681396484375, 11.167938232421875, -2.360219955444336, 29.561782836914062, 36.20068359375, 15.46148681640625, 33.20164489746094, 6.9422607421875, 3.502246856689453, 13.96966552734375, -16.611801147460938, 10.042068481445312, 1.0292282104492188, 38.390228271484375, -16.611541748046875, 44.696319580078125, 12.29608154296875, 37.872833251953125, -3.054443359375, 19.27395248413086, -19.66796875, -17.70142364501953, 2.6471099853515625, -1.0247802734375, -0.9335174560546875, 21.64801025390625, -28.525711059570312, 68.42205810546875, 30.939170837402344, 30.989532470703125, -1.203125, -28.9212646484375, 16.0054931640625, -6.9532470703125, 54.762115478515625, 15.024078369140625, -12.026123046875, -3.323760986328125, 21.467529296875, 12.780799865722656, 2.2806472778320312, 15.3978271484375, 33.9505615234375, 27.4393310546875, 34.10752868652344, -39.790496826171875, -11.795883178710938, 14.129180908203125, 0.078277587890625, -15.71405029296875, -17.733123779296875, 27.92669677734375, 56.658538818359375, 43.64323425292969, 4.3575439453125, -2.5202178955078125, 27.52801513671875, 26.461380004882812, 24.981842041015625, 1.778533935546875, -3.2034454345703125, 27.780670166015625, -15.251861572265625, 75.5576171875, -5.499320983886719, 8.579437255859375, 13.2666015625, -8.487754821777344, -18.31768798828125, 4.9347381591796875, 40.73345184326172, -20.25250244140625, 3.819732666015625, 2.0533676147460938, 20.172821044921875, 28.76678466796875, 3.0451278686523438, 55.884735107421875, -32.6834716796875, 33.10888671875, -37.1539306640625, 5.066719055175781, 22.89349365234375, 6.3339385986328125, 20.56744384765625, 50.52105712890625, 26.656036376953125, 8.308944702148438, -14.8199462890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000138.npy"}
{"epoch": 0.28900523560209423, "step": 139, "batch_size": 128, "mean": 8.150290489196777, "std": 24.325456619262695, "min": -45.415283203125, "p10": -22.513835144042968, "median": 6.182342529296875, "p90": 41.353765869140624, "max": 68.48370361328125, "pos_frac": 0.6328125, "sample": [-37.33717346191406, -3.9158477783203125, -30.270477294921875, 6.450714111328125, -32.5848388671875, 13.778900146484375, 31.99517822265625, 19.23541259765625, -2.9379730224609375, -11.7738037109375, 48.457244873046875, 4.002105712890625, 21.86260986328125, 11.480743408203125, -28.407852172851562, 46.45207214355469, 54.3773193359375, 10.14202880859375, -15.353172302246094, 10.726486206054688, 5.283805847167969, 26.6881103515625, 32.38298034667969, -14.525253295898438, 14.720062255859375, -15.193023681640625, 8.195629119873047, 36.05615234375, -6.430030822753906, -45.415283203125, 0.5994453430175781, -8.038917541503906, 35.84684753417969, 39.26017761230469, 25.4808349609375, 0.35099315643310547, 12.386932373046875, -7.49200439453125, 3.37823486328125, 30.245208740234375, -2.6894378662109375, -21.031707763671875, -0.493072509765625, 21.544082641601562, 13.773849487304688, -7.57330322265625, -43.041961669921875, 0.9349517822265625, -10.226402282714844, 38.290069580078125, -8.6500244140625, 0.21258544921875, -22.949722290039062, -22.3270263671875, 7.403999328613281, 43.4189453125, -1.783843994140625, 1.2302703857421875, -24.13226318359375, -19.182861328125, 42.19488525390625, -13.691276550292969, 43.60264587402344, 25.118967056274414, 27.48052978515625, 15.51446533203125, 0.4089374542236328, 0.4666595458984375, 37.81988525390625, 22.172836303710938, -17.712120056152344, 18.516265869140625, 13.64385986328125, 42.94316101074219, -4.699165344238281, -9.816192626953125, 6.220489501953125, 68.48370361328125, -11.36492919921875, 40.9932861328125, 23.62030792236328, 21.978546142578125, 15.278564453125, 51.91981506347656, 4.63909912109375, -15.432174682617188, 54.72869873046875, -36.571075439453125, 18.039077758789062, -4.428232192993164, 27.6448974609375, 47.41162109375, 17.94921875, 16.44244384765625, -20.720916748046875, 13.60498046875, -1.3642730712890625, -39.51582336425781, -7.33917236328125, 2.3803024291992188, 25.141189575195312, 6.144195556640625, -8.709869384765625, 3.7209625244140625, 12.787017822265625, -10.7723388671875, 27.994644165039062, 67.8985595703125, 9.82958984375, 58.654022216796875, 38.29998779296875, 19.985748291015625, 5.308868408203125, -34.3953857421875, 7.1544952392578125, -32.40875244140625, 0.0, 3.9095001220703125, 13.415878295898438, 33.36782455444336, -7.9303436279296875, -33.1102294921875, 20.406600952148438, -5.638877868652344, 21.3697509765625, -12.6082763671875, 36.585113525390625, 5.386787414550781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000139.npy"}
{"epoch": 0.29109947643979056, "step": 140, "batch_size": 128, "mean": 15.066213607788086, "std": 22.795324325561523, "min": -43.876556396484375, "p10": -9.891073608398436, "median": 10.619026184082031, "p90": 46.70220031738281, "max": 81.39434814453125, "pos_frac": 0.7109375, "sample": [-7.581298828125, 24.195220947265625, 32.23875427246094, -9.535659790039062, -11.258735656738281, 51.155487060546875, -5.889610290527344, 41.125518798828125, -3.4852294921875, 0.2777862548828125, -15.718772888183594, 10.643417358398438, 30.20751953125, 22.514198303222656, 5.4937591552734375, 9.643970489501953, 0.90179443359375, 20.581954956054688, 45.77545166015625, 9.6646728515625, 7.7364349365234375, -17.137252807617188, 29.146385192871094, -27.759033203125, 21.04656982421875, 39.568359375, -3.294952392578125, 6.7341766357421875, 43.756103515625, 42.41045379638672, 28.288330078125, 33.81883239746094, 65.76229858398438, 37.78669738769531, -8.518356323242188, 56.015899658203125, -2.4881591796875, 3.5213623046875, 49.25982666015625, 49.021392822265625, -1.3537216186523438, -25.784530639648438, 10.594635009765625, -0.37049102783203125, -2.900360107421875, 37.25634765625, 50.695404052734375, 8.693328857421875, -7.9193115234375, 81.39434814453125, 9.47210693359375, 24.328887939453125, 31.122360229492188, 11.93768310546875, 2.6110153198242188, 21.874710083007812, 21.73712158203125, 6.419944763183594, -8.154861450195312, 23.005393981933594, 26.8388671875, 0.15134429931640625, 3.97137451171875, 22.482025146484375, 6.3749542236328125, 0.936920166015625, -4.353656768798828, -10.62249755859375, 0.0, -43.876556396484375, -0.365478515625, -27.63836669921875, 67.1640625, 1.3075637817382812, 5.556236267089844, 13.894676208496094, 24.325836181640625, 32.744293212890625, 2.6314315795898438, 46.933319091796875, 11.555572509765625, 4.462158203125, 0.629730224609375, 43.6541748046875, 51.9754638671875, 35.4107666015625, 29.5108642578125, -3.067136764526367, 36.81962585449219, 38.09028625488281, 6.517435073852539, 6.2711181640625, 18.380096435546875, -3.1922454833984375, 29.68262481689453, 30.525131225585938, 0.0, -5.797979354858398, 19.56280517578125, -0.981201171875, -9.4176025390625, 48.322601318359375, -17.33447265625, 46.6031494140625, 10.051406860351562, -18.575218200683594, 43.97625732421875, 22.196121215820312, 23.347442626953125, 23.30126190185547, 44.272064208984375, 21.0084228515625, 52.34234619140625, 19.986923217773438, 27.546905517578125, -9.577606201171875, -2.8710479736328125, -16.651046752929688, 8.98272705078125, -26.678817749023438, 12.740325927734375, 56.177947998046875, 10.8333740234375, -2.7822189331054688, 6.054168701171875, -14.77545166015625, 27.578887939453125, 23.065155029296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000140.npy"}
{"epoch": 0.2931937172774869, "step": 141, "batch_size": 128, "mean": 14.428566932678223, "std": 25.458316802978516, "min": -77.06216430664062, "p10": -11.584572410583496, "median": 13.120357513427734, "p90": 46.22508239746093, "max": 90.66033935546875, "pos_frac": 0.7265625, "sample": [40.50923156738281, 17.854736328125, -25.725936889648438, -10.668777465820312, 7.43048095703125, 13.087287902832031, -5.904693603515625, 21.61145782470703, -46.942138671875, 90.66033935546875, 5.7342987060546875, 11.246917724609375, 30.896018981933594, 30.37298583984375, 13.406112670898438, 44.444549560546875, 27.267959594726562, 18.01885986328125, 18.71502113342285, 23.964263916015625, 58.52587890625, -4.659423828125, 1.1218490600585938, 45.6700439453125, -0.036865234375, -16.540430068969727, 39.303470611572266, 13.153427124023438, 2.502655029296875, 51.233245849609375, 34.005615234375, -17.350616455078125, -10.978240966796875, 68.378662109375, 2.00250244140625, 17.420013427734375, -25.622146606445312, -11.519540786743164, 12.032424926757812, -18.37640380859375, 44.04443359375, 37.1209716796875, 14.765203475952148, 18.340347290039062, -7.375709533691406, 8.144561767578125, 1.938507080078125, -7.7882232666015625, 60.467132568359375, 8.23553466796875, 11.732269287109375, 24.221588134765625, 3.586515426635742, -7.924713134765625, 28.49481201171875, 24.165924072265625, 1.44207763671875, 1.1386032104492188, 25.770294189453125, 2.13763427734375, 35.867401123046875, 34.20917510986328, -14.75408935546875, -42.79731750488281, -6.588165283203125, 68.62188720703125, -1.0495471954345703, 27.475013732910156, -4.8171844482421875, 37.912261962890625, 4.8902740478515625, 12.388511657714844, 8.039909362792969, 23.38903045654297, 28.08709716796875, 38.32194519042969, 31.94287109375, -7.30926513671875, 0.0, -16.152587890625, 41.85969543457031, -3.0208740234375, -0.004913330078125, 59.7454833984375, 6.408782958984375, 55.71246337890625, -77.06216430664062, 17.1046142578125, 20.638153076171875, -2.485565185546875, 24.813766479492188, 19.35144805908203, 32.82689666748047, -45.22450256347656, -6.557342529296875, 17.353500366210938, -0.319427490234375, 55.624755859375, -0.82183837890625, 32.19181823730469, -7.418243408203125, 11.0323486328125, 18.696334838867188, 37.258544921875, 16.430023193359375, 4.514518737792969, 22.93939208984375, 49.395599365234375, 42.855316162109375, -11.736312866210938, 9.192024230957031, 13.67828369140625, 58.41473388671875, -36.585479736328125, 1.6429290771484375, 10.101638793945312, 9.57147216796875, 13.384490966796875, 19.39471435546875, 21.157012939453125, -9.175430297851562, 47.520172119140625, 44.74688720703125, 8.474563598632812, 11.330066680908203, 55.939544677734375, 16.22796630859375, 3.158660888671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000141.npy"}
{"epoch": 0.29528795811518327, "step": 142, "batch_size": 128, "mean": 14.220439910888672, "std": 25.74233627319336, "min": -68.21356201171875, "p10": -14.35975036621093, "median": 10.003974914550781, "p90": 48.17914733886719, "max": 80.8045654296875, "pos_frac": 0.671875, "sample": [-4.84747314453125, -2.5665054321289062, 9.289791107177734, 27.710357666015625, 1.9630279541015625, -19.18017578125, 11.033905029296875, -0.87945556640625, 75.3114013671875, 40.82218933105469, 43.98199462890625, -3.884796142578125, -7.719390869140625, 35.480865478515625, 18.452606201171875, 26.81719970703125, -0.039813995361328125, -2.5933837890625, 51.74395751953125, 26.581764221191406, -1.6827239990234375, 26.75434112548828, 0.0, 20.260032653808594, 20.64208984375, 48.267181396484375, 10.916259765625, 3.64337158203125, -3.223186492919922, 0.024444580078125, -1.353912353515625, 30.33740234375, -3.0000228881835938, -25.80767822265625, 2.96929931640625, 2.5964736938476562, 25.88909912109375, 13.844093322753906, -24.278839111328125, 28.855682373046875, 20.36102294921875, 51.80096435546875, 43.423828125, 17.906143188476562, 19.20733642578125, -10.382316589355469, -1.0836963653564453, 44.61921691894531, 19.759056091308594, 70.87957763671875, 15.97314453125, -6.404897689819336, 3.2298583984375, 13.810867309570312, -12.293853759765625, 2.624431610107422, 1.191864013671875, 35.654457092285156, -20.78436279296875, -20.352203369140625, 22.4840087890625, 32.3487548828125, -11.7662353515625, -8.55859375, 48.14141845703125, 37.68025207519531, 13.5013427734375, 0.2852287292480469, 46.79692077636719, -3.9831809997558594, 8.600631713867188, 20.63189697265625, 2.38330078125, 6.787233352661133, -23.42437744140625, 34.24073791503906, -40.0894775390625, 28.415130615234375, 44.42852783203125, -26.304061889648438, -3.91754150390625, 69.75616455078125, 80.8045654296875, 0.179595947265625, -1.7236251831054688, 16.539085388183594, 10.021484375, 50.47320556640625, 9.986465454101562, 6.475307464599609, -3.0029239654541016, -3.749439239501953, 31.47650146484375, 42.30827331542969, 59.486083984375, 6.6245269775390625, 37.366180419921875, 28.559234619140625, 20.741363525390625, 54.994659423828125, -68.21356201171875, 53.46221923828125, -31.23065185546875, 0.7994976043701172, -0.4766521453857422, -25.636505126953125, 35.58702087402344, 15.845855712890625, 32.45904541015625, -4.4566192626953125, 37.735076904296875, -36.65570068359375, 3.0581512451171875, 5.41082763671875, 23.55914306640625, -6.564544677734375, 30.897979736328125, -1.386932373046875, 6.2571563720703125, 2.24853515625, -22.122459411621094, 62.82427978515625, -0.18886947631835938, -0.69952392578125, 31.598175048828125, 29.72998809814453, 66.49113464355469, 41.61376190185547], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000142.npy"}
{"epoch": 0.2973821989528796, "step": 143, "batch_size": 128, "mean": 16.291526794433594, "std": 27.845394134521484, "min": -78.4254150390625, "p10": -11.732349014282224, "median": 13.428955078125, "p90": 48.63379516601562, "max": 99.72662353515625, "pos_frac": 0.75, "sample": [27.84564971923828, 50.259124755859375, 56.994049072265625, 29.286865234375, 42.10044860839844, 15.99273681640625, 31.243560791015625, 8.928863525390625, -18.344635009765625, -8.996326446533203, 28.683746337890625, -0.3894195556640625, 20.31072998046875, 15.431289672851562, 2.1002464294433594, -13.051773071289062, 47.91748046875, 4.9404296875, -22.264068603515625, 99.72662353515625, -19.285598754882812, -9.511749267578125, 45.42350769042969, 46.42027282714844, 23.18085479736328, 45.989013671875, 37.87750244140625, 63.12469482421875, 89.8951416015625, 2.41595458984375, -2.04400634765625, 42.783416748046875, 29.19952392578125, 30.135509490966797, 25.984580993652344, -6.27215576171875, -78.4254150390625, -3.225006103515625, 27.95868682861328, 8.98211669921875, 8.563705444335938, -31.648712158203125, 8.548164367675781, 39.904632568359375, 5.734870910644531, 6.60791015625, 10.123725891113281, 21.878311157226562, 17.493045806884766, -68.58209228515625, 5.35064697265625, 26.87335205078125, 3.6484222412109375, 39.093162536621094, 19.693359375, 22.78570556640625, 1.26007080078125, 22.793752670288086, -14.525009155273438, -0.2749366760253906, -8.956695556640625, 1.376138687133789, 70.19537353515625, 21.74493408203125, 5.543458938598633, -2.5131607055664062, 60.9434814453125, 31.84222412109375, 40.234771728515625, 2.493377685546875, 31.099700927734375, 35.229583740234375, -11.166881561279297, 3.9025630950927734, 33.76292419433594, -8.9892578125, -39.510009765625, -0.15662384033203125, 11.298141479492188, 13.05340576171875, 3.1589202880859375, 1.337432861328125, 7.636970520019531, 17.754928588867188, 12.913238525390625, -2.17730712890625, -18.54217529296875, 3.3934974670410156, -0.00850677490234375, 13.897857666015625, 35.17463684082031, 16.358917236328125, -2.636474609375, 21.901870727539062, 20.1627197265625, 23.595687866210938, 0.46468162536621094, 17.011932373046875, -4.492916107177734, 2.5323028564453125, 83.754150390625, 30.965484619140625, -4.4288482666015625, 11.923439025878906, -29.840057373046875, 74.84613037109375, 35.46617889404297, 3.5198402404785156, 3.618204116821289, -1.4104156494140625, 28.190093994140625, 55.953948974609375, 28.67456817626953, 67.80419921875, 26.446533203125, 62.033721923828125, 43.678192138671875, 44.02825927734375, -41.41326904296875, 26.589351654052734, -16.735809326171875, 50.9405517578125, -7.69573974609375, 12.7279052734375, 47.937225341796875, 7.198211669921875, 13.80450439453125, 1.2286701202392578], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000143.npy"}
{"epoch": 0.2994764397905759, "step": 144, "batch_size": 128, "mean": 13.559566497802734, "std": 25.499650955200195, "min": -68.62942504882812, "p10": -14.179824066162109, "median": 7.089539527893066, "p90": 47.219970703125, "max": 92.0673828125, "pos_frac": 0.71875, "sample": [-1.4275093078613281, 17.123268127441406, 45.5350341796875, 5.79229736328125, 6.514801025390625, 12.98394775390625, -19.875038146972656, -10.820655822753906, 1.2491397857666016, 92.0673828125, 2.3946990966796875, -14.1649169921875, 7.3811798095703125, 57.6385498046875, 3.9822235107421875, 33.818511962890625, 1.264190673828125, 47.391082763671875, 9.129905700683594, -0.998870849609375, 3.81744384765625, 19.86187744140625, -14.067352294921875, 2.6498565673828125, 4.052001953125, -6.188934326171875, -21.807640075683594, -3.583629608154297, 28.793289184570312, 5.249334335327148, -39.2757568359375, 6.79789924621582, -1.1396331787109375, 24.5863037109375, 51.199005126953125, 39.60276794433594, 31.64404296875, 73.031494140625, 8.770980834960938, 59.172821044921875, -0.9542770385742188, -8.84747314453125, 5.482513427734375, 21.370452880859375, -0.7590198516845703, -18.33258056640625, 6.685333251953125, 36.21478271484375, -17.263702392578125, 22.23626708984375, 12.504753112792969, 57.803619384765625, 37.53123474121094, -19.37457275390625, -68.62942504882812, -27.703277587890625, 15.7508544921875, -14.214607238769531, -10.977508544921875, 2.8367919921875, 34.9962158203125, -0.91192626953125, -7.55230712890625, 6.3285675048828125, 26.09998321533203, 13.88348388671875, 67.07763671875, 65.57220458984375, 38.81999206542969, 38.6502685546875, 1.8025360107421875, 20.036895751953125, 48.14190673828125, -28.978240966796875, 21.781631469726562, 29.726287841796875, 4.49163818359375, 10.180534362792969, 47.146636962890625, 44.2982177734375, 44.95616149902344, 37.89864730834961, 5.7926483154296875, -8.915573120117188, 15.699981689453125, 26.209564208984375, -0.2806396484375, 6.699188232421875, 2.1675033569335938, 33.70172119140625, 69.69891357421875, 32.27392578125, 13.231437683105469, 0.6239242553710938, 11.946287155151367, 47.512542724609375, 3.13232421875, -4.27520751953125, 7.68511962890625, 35.256256103515625, 11.471221923828125, 0.18832015991210938, 0.0, 2.4412384033203125, -13.76141357421875, -12.7332763671875, 3.50201416015625, -35.2401123046875, 26.177520751953125, 24.984233856201172, 18.317459106445312, -1.693817138671875, 43.92584228515625, -6.780181884765625, 27.705078125, 2.151611328125, 17.753204345703125, 6.205909729003906, 17.930694580078125, -26.04217529296875, 15.0416259765625, 37.21319580078125, 8.1812744140625, 46.903564453125, -17.610809326171875, -0.1631011962890625, 70.42794799804688, 5.015045166015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000144.npy"}
{"epoch": 0.30157068062827225, "step": 145, "batch_size": 128, "mean": 17.516250610351562, "std": 26.674806594848633, "min": -55.47285461425781, "p10": -13.184542846679687, "median": 16.1532039642334, "p90": 54.060308837890624, "max": 84.20098876953125, "pos_frac": 0.7109375, "sample": [-13.058868408203125, 38.737754821777344, -42.31915283203125, 38.345184326171875, 13.309259414672852, -27.804229736328125, 43.53997802734375, 19.032867431640625, 81.82089233398438, 24.733139038085938, 18.564300537109375, 12.214141845703125, 23.620925903320312, 33.819793701171875, 30.34137725830078, 64.68524169921875, 4.409912109375, 29.194976806640625, 7.105812072753906, 5.663825988769531, 6.0897369384765625, -7.584228515625, 53.97467041015625, -12.687484741210938, -8.149169921875, 67.94949340820312, -27.561874389648438, 51.97412109375, 0.6767425537109375, 20.2293701171875, 1.240142822265625, 31.23786163330078, 8.320236206054688, 36.55110168457031, 56.83831787109375, 23.1112060546875, 78.27468872070312, -55.47285461425781, 6.4610137939453125, -1.6285400390625, 21.23419952392578, 19.72991943359375, -14.903083801269531, 65.64529418945312, 70.23739624023438, 14.00933837890625, 38.529754638671875, 36.07769775390625, 10.838798522949219, 21.598419189453125, 16.034343719482422, -7.1211700439453125, -7.0008392333984375, 33.736244201660156, 47.07197570800781, 46.89363098144531, 0.2847900390625, -4.460182189941406, -15.241859436035156, 3.342744827270508, 16.272064208984375, 8.4700927734375, -9.085479736328125, 42.4122314453125, 62.99371337890625, -5.3864898681640625, 56.539398193359375, -9.545120239257812, 55.18310546875, 31.605405807495117, 54.2601318359375, -9.69732666015625, 6.346473693847656, 35.646942138671875, 48.22283935546875, 11.469970703125, 1.5843696594238281, 11.194168090820312, 28.280616760253906, -19.703842163085938, -24.22803497314453, 5.6606597900390625, 9.413867950439453, 29.27667236328125, 84.20098876953125, 21.894027709960938, 5.492393493652344, 24.874099731445312, -24.82402801513672, -12.012001037597656, -3.118743896484375, 0.0, 7.0831146240234375, 19.130950927734375, 9.652931213378906, 3.8449630737304688, 41.597076416015625, 71.80926513671875, 42.864593505859375, 24.55303955078125, -2.191783905029297, -8.87591552734375, 27.4893798828125, 24.54559326171875, 17.03961181640625, -13.477783203125, 24.527198791503906, 29.74395751953125, -2.715351104736328, -1.0609407424926758, -16.4442138671875, 49.54412841796875, 28.69927215576172, 44.667236328125, -2.262420654296875, -2.720672607421875, 4.053009033203125, 30.704559326171875, 44.101837158203125, 50.945037841796875, 31.756195068359375, -15.373016357421875, -1.5296554565429688, -3.1085052490234375, 27.65325927734375, -22.077392578125, -7.6324310302734375, 19.511520385742188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000145.npy"}
{"epoch": 0.3036649214659686, "step": 146, "batch_size": 128, "mean": 13.90401554107666, "std": 25.32842254638672, "min": -61.593780517578125, "p10": -14.792405700683592, "median": 8.598808288574219, "p90": 45.284794616699216, "max": 78.62738037109375, "pos_frac": 0.734375, "sample": [5.099541664123535, 2.7757186889648438, 19.166828155517578, -33.6309814453125, 13.5333251953125, 39.39959716796875, -26.828704833984375, 11.33758544921875, 34.218162536621094, -25.932464599609375, 78.62738037109375, 73.91058349609375, 10.148162841796875, -9.098297119140625, 19.66192626953125, 3.3980445861816406, 42.084686279296875, 23.268646240234375, 2.8853683471679688, 3.45257568359375, 29.489151000976562, 2.7427825927734375, 9.796554565429688, 13.473617553710938, -18.846160888671875, -9.648681640625, 4.134073257446289, 13.46099853515625, 17.046096801757812, 21.212188720703125, 8.546981811523438, 1.5245914459228516, 40.6275634765625, 0.47064208984375, 5.391937255859375, 19.334716796875, 16.44287109375, -22.29552459716797, 14.690299987792969, 3.3814010620117188, 6.40179443359375, 48.86064147949219, 0.044158935546875, 36.17523193359375, 0.0, 46.44776916503906, -4.79571533203125, -1.4798603057861328, 28.416946411132812, 24.55889892578125, 11.176177978515625, -15.7349853515625, 10.305318832397461, -1.691253662109375, 52.07464599609375, 62.086517333984375, -61.593780517578125, 30.57763671875, 5.460113525390625, 34.835693359375, 36.950897216796875, 6.32818603515625, 4.274314880371094, 47.37451171875, 36.48358154296875, -14.388442993164062, -7.15802001953125, -13.745208740234375, 7.153282165527344, -19.4503173828125, 52.27796936035156, -21.884796142578125, 4.4525146484375, 43.177886962890625, -21.022964477539062, 61.95916748046875, 8.650634765625, 8.427894592285156, -11.56390380859375, 71.34030151367188, 6.102569580078125, 33.868560791015625, 3.2736568450927734, -8.3204345703125, 34.659698486328125, 0.07723808288574219, -7.4670867919921875, 3.6036529541015625, 0.0, 38.182838439941406, -2.95758056640625, -5.374755859375, 25.51312255859375, -19.5205078125, 1.03369140625, 58.31675720214844, 21.8311767578125, 39.879302978515625, 7.099903106689453, 53.15461730957031, -22.199813842773438, -14.367950439453125, 3.7310638427734375, 11.070709228515625, 40.3680419921875, -9.010025024414062, 33.103607177734375, 5.6990966796875, 51.78987121582031, 44.786376953125, 44.165130615234375, -60.37890625, 31.092723846435547, 43.966575622558594, 13.23773193359375, 36.55059814453125, -7.400543212890625, 43.1273193359375, 2.5859718322753906, 38.90150451660156, -0.8741455078125, 6.8741607666015625, 23.584381103515625, 44.043670654296875, 30.021331787109375, -9.38888931274414, 29.796234130859375, -8.33544921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000146.npy"}
{"epoch": 0.3057591623036649, "step": 147, "batch_size": 128, "mean": 11.920906066894531, "std": 28.805540084838867, "min": -77.0943603515625, "p10": -20.07015380859375, "median": 9.81597900390625, "p90": 50.01272277832031, "max": 89.30996704101562, "pos_frac": 0.6640625, "sample": [0.1439361572265625, -3.8250732421875, 7.727100372314453, 30.026275634765625, -10.217002868652344, -2.490631103515625, 84.29806518554688, -15.4183349609375, 0.0, 71.6568603515625, -15.777938842773438, 20.113548278808594, 14.119903564453125, 5.442535400390625, 25.42254066467285, 4.692939758300781, -20.06341552734375, 47.9315185546875, 53.8485107421875, 3.7535018920898438, -3.88446044921875, 20.76214599609375, -0.3697967529296875, 36.66185760498047, -1.56890869140625, 50.67779541015625, 1.1872901916503906, 44.20484924316406, 9.265777587890625, -0.273834228515625, 69.7392578125, 35.393310546875, 17.556018829345703, 22.859100341796875, 11.212066650390625, -29.717987060546875, -43.25702667236328, -6.886138916015625, -17.01202392578125, 17.754722595214844, 10.56365966796875, 36.21580505371094, 32.01674270629883, 6.022502899169922, -31.102367401123047, 27.69610595703125, -27.813369750976562, -5.892038345336914, -0.3361835479736328, 0.9395790100097656, -2.9134368896484375, 51.114501953125, 10.132431030273438, 12.553909301757812, 27.33014678955078, 30.0234375, 79.00006103515625, 13.014404296875, 5.49163818359375, -4.81695556640625, 23.40814208984375, 22.356842041015625, 60.10203552246094, 10.604785919189453, 55.9771728515625, 1.1051025390625, 2.05242919921875, 59.887481689453125, 40.57883071899414, 3.5185546875, -44.591766357421875, 11.723907470703125, 1.32855224609375, 0.1644287109375, 29.2677001953125, 5.45269775390625, 14.588544845581055, 4.384941101074219, 0.1723918914794922, -0.0477142333984375, 27.622772216796875, -9.2159423828125, -0.5653533935546875, 35.88691711425781, 18.485694885253906, 37.191253662109375, 9.95550537109375, 32.926025390625, -14.76043701171875, 33.23591613769531, 14.537872314453125, 42.5860595703125, -26.7103271484375, -77.0943603515625, -38.80656433105469, -1.8312835693359375, 12.044403076171875, 33.48649597167969, -17.925872802734375, -41.91944885253906, -8.702194213867188, 3.6994552612304688, 54.939849853515625, 15.22540283203125, -20.08587646484375, -7.949237823486328, 45.46949768066406, -2.7744598388671875, -40.371368408203125, -13.758529663085938, -65.7597885131836, 9.67645263671875, 6.8777618408203125, 39.723304748535156, 49.727691650390625, 36.26995849609375, 12.704383850097656, 55.2398681640625, -5.08184814453125, -7.55084228515625, 16.938995361328125, 44.07427978515625, 45.1705322265625, 18.5557861328125, 89.30996704101562, 12.387985229492188, -4.752458572387695, -29.42242431640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000147.npy"}
{"epoch": 0.3078534031413613, "step": 148, "batch_size": 128, "mean": 12.571126937866211, "std": 27.066743850708008, "min": -58.680877685546875, "p10": -24.476315307617188, "median": 11.939952850341797, "p90": 48.88619384765625, "max": 93.51870727539062, "pos_frac": 0.65625, "sample": [14.521728515625, 13.295783996582031, 18.516311645507812, -14.9622802734375, -1.4122962951660156, 68.8641357421875, -0.05256462097167969, 20.4476318359375, 73.667236328125, 26.718276977539062, 5.630947113037109, 20.693466186523438, -4.2286224365234375, 56.35798645019531, 14.550384521484375, 36.63618469238281, 9.7830810546875, -19.84320068359375, 16.817474365234375, 27.123126983642578, -22.430938720703125, 43.25592041015625, -46.91020202636719, 0.7497215270996094, 8.1300048828125, -26.62664794921875, 13.541671752929688, -4.4542694091796875, 15.158233642578125, -9.276931762695312, 20.96558380126953, 15.677688598632812, -8.970191955566406, -58.680877685546875, -9.33612060546875, -15.840606689453125, 60.80523681640625, 7.1630706787109375, -28.216156005859375, 21.50177001953125, 58.08062744140625, 1.7702198028564453, -15.740478515625, -30.99383544921875, 1.5263214111328125, 6.483919143676758, 2.0907211303710938, -16.833343505859375, 23.70458984375, 33.287811279296875, 14.000755310058594, 8.207389831542969, 7.519203186035156, 18.26531982421875, -30.498214721679688, 71.65237426757812, 9.288902282714844, 37.79194641113281, 10.825569152832031, 47.8399658203125, 23.143951416015625, -6.9280853271484375, 0.0944366455078125, -1.1876220703125, 0.0, 7.940441131591797, 55.4102783203125, 23.350433349609375, 36.78547668457031, 39.59063720703125, -3.486743927001953, -29.108078002929688, -5.7332763671875, -36.35075378417969, 39.736083984375, 11.031089782714844, -29.040206909179688, 32.52333068847656, -2.402496337890625, 29.698989868164062, -1.0058422088623047, 21.04700469970703, -24.356536865234375, 28.783233642578125, 56.044281005859375, 18.494407653808594, 2.780170440673828, 16.727813720703125, -10.19219970703125, -15.49310302734375, 27.832794189453125, 21.915935516357422, -0.5670032501220703, 28.495697021484375, 23.256378173828125, 48.985870361328125, -26.223175048828125, 6.43865966796875, -12.512825012207031, 63.261962890625, 3.47467041015625, 27.059600830078125, 93.51870727539062, 37.684814453125, 2.03326416015625, -7.6156158447265625, 30.414566040039062, -27.983245849609375, 32.25506591796875, 23.01232147216797, 48.843475341796875, 44.73534393310547, 17.365753173828125, 30.94158935546875, -26.917442321777344, -1.4029312133789062, -24.75579833984375, 16.0999755859375, 61.55828857421875, 0.0, -9.60772705078125, 14.124908447265625, 34.533470153808594, -0.82012939453125, 12.84881591796875, 44.615875244140625, 60.63179016113281, -3.9271469116210938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000148.npy"}
{"epoch": 0.3099476439790576, "step": 149, "batch_size": 128, "mean": 15.678827285766602, "std": 26.60759735107422, "min": -42.185028076171875, "p10": -20.709765624999996, "median": 12.489273071289062, "p90": 47.564166259765614, "max": 97.1917724609375, "pos_frac": 0.7578125, "sample": [50.552093505859375, -23.22308349609375, 25.552490234375, 33.652244567871094, 43.849029541015625, 46.61053466796875, 22.566329956054688, 5.454219818115234, 5.48529052734375, 7.4362945556640625, 17.98968505859375, 4.509033203125, 43.62640380859375, 15.62518310546875, 30.83734130859375, -4.2309722900390625, 45.1395263671875, 10.700485229492188, -3.6099853515625, -9.439521789550781, 17.095802307128906, 4.6233673095703125, 19.22906494140625, 4.073646545410156, 27.690876007080078, 12.696533203125, -41.02111053466797, 37.419525146484375, 49.789306640625, 16.82562255859375, 7.6814117431640625, 21.45733642578125, 33.743255615234375, 11.912567138671875, 1.8514785766601562, -0.041595458984375, 51.78746032714844, 2.573322296142578, -19.63262939453125, 0.0334930419921875, 35.1392822265625, 69.75726318359375, -11.363677978515625, 1.9336166381835938, 76.88580322265625, 45.037200927734375, 11.339035034179688, -24.906631469726562, 39.77252197265625, 50.802642822265625, 20.810230255126953, -5.198333740234375, 8.950286865234375, 25.28997039794922, 5.37359619140625, 78.915771484375, -30.66357421875, 15.11932373046875, 17.7720947265625, 15.38232421875, -7.561302185058594, 23.66663360595703, -24.98712158203125, 38.05003356933594, -28.827423095703125, 5.13153076171875, 7.397064208984375, -6.51458740234375, 24.617172241210938, 23.727548599243164, 41.07978820800781, 21.338790893554688, 34.214683532714844, 12.282012939453125, 41.998321533203125, 13.951774597167969, 52.492095947265625, 9.555488586425781, 57.336517333984375, 0.17163848876953125, 43.200286865234375, 0.07833099365234375, 33.25708770751953, 28.14483642578125, 13.05218505859375, -10.703125, 55.73236083984375, 44.80950927734375, -40.26593017578125, 3.1398239135742188, 11.80328369140625, 34.385986328125, -5.576263427734375, -25.41168212890625, -1.908355712890625, -4.68096923828125, 66.33889770507812, 5.9910888671875, 43.56927490234375, 4.906585693359375, -6.9241180419921875, 4.0667572021484375, 32.7203369140625, 0.307861328125, 28.942626953125, -2.1175537109375, -29.654815673828125, -24.379974365234375, 1.2613372802734375, 97.1917724609375, -3.47979736328125, -5.266082763671875, 8.00140380859375, 28.5145263671875, 5.21044921875, -12.667037963867188, 42.609153747558594, 29.298690795898438, -41.67674255371094, -42.185028076171875, 20.347305297851562, 38.22625732421875, -25.47003173828125, 75.27047729492188, 39.974510192871094, 24.230255126953125, 3.600250244140625, 6.9330291748046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000149.npy"}
{"epoch": 0.31204188481675393, "step": 150, "batch_size": 128, "mean": 14.798748970031738, "std": 25.60117530822754, "min": -49.2730712890625, "p10": -13.493789672851559, "median": 12.529394149780273, "p90": 46.66218948364258, "max": 83.71414184570312, "pos_frac": 0.7734375, "sample": [22.344390869140625, 33.96061706542969, -4.163547515869141, 35.62689208984375, 10.100204467773438, -29.361602783203125, -3.3091888427734375, 20.44927978515625, -30.43585205078125, 30.88970947265625, 26.957412719726562, 24.759979248046875, 45.27728271484375, 15.3414306640625, -15.9454345703125, 3.683807373046875, 8.241058349609375, 31.285926818847656, 6.20281982421875, 38.936309814453125, 25.216049194335938, -16.755035400390625, -1.821258544921875, 4.238983154296875, -5.820030212402344, 14.33026123046875, 7.8551025390625, 6.8216705322265625, 66.45414733886719, 34.421165466308594, 68.697265625, 0.6734733581542969, 20.91015625, 8.706506729125977, 16.874893188476562, 27.74310302734375, 22.785507202148438, -39.073486328125, 39.097808837890625, 9.970321655273438, 21.235565185546875, 43.77186584472656, 17.006683349609375, 40.036956787109375, -7.0893096923828125, 2.2914886474609375, 1.833770751953125, 5.10894775390625, 14.029541015625, 63.99365234375, 16.323522567749023, 76.30535888671875, 0.325592041015625, 32.901214599609375, 47.583251953125, 62.89666748046875, 52.324676513671875, 6.049413681030273, -1.04534912109375, 0.44576454162597656, 0.915618896484375, -0.5155086517333984, -32.581146240234375, 46.927162170410156, 4.57080078125, -0.1083526611328125, 6.409637451171875, 1.2830963134765625, 43.93714141845703, 44.95648193359375, 20.56317138671875, 39.286376953125, 2.0709228515625, 46.54862976074219, 21.25436782836914, -9.526351928710938, -43.509002685546875, 8.662750244140625, 2.539989471435547, 24.091773986816406, 12.995891571044922, -48.5311279296875, -1.42333984375, -24.009384155273438, -5.52978515625, 14.56072998046875, -23.47161865234375, 83.71414184570312, 36.6427001953125, 12.062896728515625, 14.93353271484375, -12.443084716796875, 7.023162841796875, 24.919158935546875, -49.2730712890625, 9.919330596923828, 42.04774475097656, -3.16748046875, -27.160308837890625, 1.1480960845947266, 20.03160858154297, 32.96739959716797, 9.11492919921875, 54.40962219238281, 10.826339721679688, 16.8177490234375, 36.3587646484375, 11.066925048828125, 42.898101806640625, 31.87994384765625, 2.20404052734375, 14.2239990234375, 4.263814926147461, 53.020233154296875, 7.4117279052734375, 0.6300048828125, 16.64362335205078, -10.27313232421875, 57.58660888671875, -11.989166259765625, 0.0, 27.0189208984375, 76.15277099609375, 0.584625244140625, 13.35433578491211, 17.904739379882812, 19.449806213378906, -31.551551818847656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000150.npy"}
{"epoch": 0.31413612565445026, "step": 151, "batch_size": 128, "mean": 22.58228874206543, "std": 29.22235679626465, "min": -54.71746826171875, "p10": -16.76954116821289, "median": 25.982269287109375, "p90": 58.61131286621093, "max": 84.15045166015625, "pos_frac": 0.765625, "sample": [39.63038635253906, 19.165557861328125, 1.0580329895019531, -21.670608520507812, 4.57763671875e-05, 84.15045166015625, 53.588836669921875, -25.74810791015625, 1.6013641357421875, 41.665740966796875, -1.310516357421875, -5.856697082519531, 25.8717041015625, 37.3958740234375, -7.6878204345703125, -16.138328552246094, -2.3698883056640625, 40.78898620605469, 37.75103759765625, 23.370468139648438, -54.71746826171875, 3.0423583984375, -9.679290771484375, 26.283737182617188, 4.66522216796875, 38.7342529296875, 2.674591064453125, -3.8955078125, 27.8134765625, 44.6051025390625, 54.278167724609375, 2.8903045654296875, -3.2593231201171875, 20.21833038330078, 40.56233215332031, 11.67401123046875, 17.5347900390625, 18.710906982421875, -5.534351348876953, 80.09506225585938, 30.658828735351562, 32.342864990234375, 14.020462036132812, 18.1954345703125, 45.598968505859375, -23.395492553710938, 27.58563232421875, 9.273895263671875, 47.556182861328125, 45.82238006591797, 44.67536926269531, -10.170166015625, 4.196041107177734, 0.4366607666015625, 38.770904541015625, 47.88859558105469, 16.1322021484375, 26.09283447265625, -28.45806884765625, 21.781570434570312, -4.12554931640625, 82.9371337890625, 49.009765625, -9.465347290039062, 31.780914306640625, 45.20098876953125, 27.6416015625, -23.85504150390625, 40.383636474609375, 59.89935302734375, -35.796112060546875, -29.4488525390625, 45.91326904296875, 36.459739685058594, 32.070152282714844, 25.7847900390625, 58.059295654296875, 39.84959411621094, 74.94793701171875, -19.639404296875, -26.4183349609375, 45.724853515625, 72.55166625976562, 49.072509765625, 78.87689208984375, 7.649421691894531, 40.844207763671875, -0.640289306640625, 0.0, 13.409454345703125, 29.74212646484375, 42.07318115234375, 34.292022705078125, -11.25421142578125, 36.18365478515625, 32.57145690917969, 51.74884033203125, 69.28878784179688, -14.985542297363281, 72.99237060546875, -36.802886962890625, 35.84246826171875, 2.70928955078125, 36.45330810546875, 63.12629699707031, 47.655914306640625, 12.450935363769531, -18.24237060546875, 31.409175872802734, 34.505889892578125, 18.657073974609375, 67.91644287109375, 15.810333251953125, -15.347297668457031, 76.05349731445312, 19.934356689453125, 52.728790283203125, 37.38005065917969, 67.17349243164062, 25.0235595703125, 40.301788330078125, 9.391876220703125, 14.6903076171875, 31.670654296875, 39.62730026245117, 0.50225830078125, 12.656494140625, -35.03533172607422], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000151.npy"}
{"epoch": 0.3162303664921466, "step": 152, "batch_size": 128, "mean": 17.08238983154297, "std": 29.14360237121582, "min": -57.031097412109375, "p10": -14.43829345703125, "median": 15.722530364990234, "p90": 57.46125183105469, "max": 100.0875244140625, "pos_frac": 0.6953125, "sample": [35.22796630859375, 48.52447509765625, -4.9101715087890625, 17.420089721679688, 7.017669677734375, 35.90106201171875, 12.47344970703125, 35.21247863769531, 58.282135009765625, 28.6065673828125, 19.127166748046875, 37.374725341796875, 70.20086669921875, 6.32391357421875, 35.75933837890625, -14.723907470703125, -0.625579833984375, 2.733428955078125, 1.897735595703125, -23.496063232421875, -57.031097412109375, 29.029296875, -53.818878173828125, 23.48504638671875, -10.730522155761719, 31.80450439453125, 15.884605407714844, -0.5091171264648438, 53.1265869140625, -1.457611083984375, -0.6733818054199219, 47.05558776855469, 57.43670654296875, 28.7808837890625, 32.85137939453125, 35.94903564453125, 52.28564453125, 43.24284362792969, 60.061767578125, 30.059173583984375, -42.56269836425781, 100.0875244140625, -27.320526123046875, 4.89678955078125, 34.804100036621094, -2.2616043090820312, -11.723365783691406, 22.669769287109375, -8.827314376831055, 0.378387451171875, 6.5288848876953125, -7.6025390625, 16.583572387695312, -8.069900512695312, -7.49278450012207, 27.272552490234375, 10.938018798828125, -25.7291259765625, -8.817886352539062, -0.5930023193359375, 63.17022705078125, 15.560455322265625, -18.791351318359375, 18.11175537109375, 69.5931396484375, -22.5389404296875, 18.8692626953125, -12.512428283691406, -7.225517272949219, 45.92059326171875, 6.6713409423828125, -9.92364501953125, -5.066375732421875, 12.53533935546875, 2.1538162231445312, 65.11883544921875, 40.5299072265625, 20.80572509765625, 72.05670166015625, 32.98090362548828, 57.518524169921875, -32.39044189453125, 2.043947219848633, -4.612640380859375, 1.206390380859375, 4.684974670410156, 40.472991943359375, 35.88287353515625, 68.90325927734375, 13.100473403930664, 6.320915222167969, -50.52153015136719, -10.863983154296875, 64.53411865234375, 35.924720764160156, 30.198524475097656, -17.0120849609375, 5.564544677734375, 24.681541442871094, -14.315887451171875, 20.675567626953125, 47.44287109375, 29.437774658203125, 21.46600341796875, -12.811767578125, 61.6168212890625, 4.911502838134766, 12.343490600585938, -7.833892822265625, -6.3790283203125, 43.5378532409668, 11.288459777832031, -42.5645751953125, 26.3798828125, -1.5039520263671875, 14.020034790039062, 34.546653747558594, 0.8359909057617188, 44.452423095703125, 62.07586669921875, 47.236328125, 54.67431640625, 47.77558135986328, 37.610801696777344, 7.912471771240234, 25.914031982421875, -11.6820068359375, 45.50676727294922], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000152.npy"}
{"epoch": 0.3183246073298429, "step": 153, "batch_size": 128, "mean": 24.543630599975586, "std": 27.842716217041016, "min": -59.101715087890625, "p10": -3.8133514404296873, "median": 20.24249267578125, "p90": 63.79037170410156, "max": 91.8238525390625, "pos_frac": 0.84375, "sample": [15.386962890625, -0.003520965576171875, 75.69085693359375, 16.388580322265625, 62.806640625, 48.201271057128906, 47.380401611328125, 0.868865966796875, 91.8238525390625, 7.27447509765625, 46.891357421875, 3.76800537109375, -18.021240234375, 22.859649658203125, 35.83631134033203, 5.1585540771484375, -12.928176879882812, 3.4559478759765625, 30.188232421875, 69.08415222167969, 63.299102783203125, 73.32818603515625, 36.37144470214844, 3.6220855712890625, 3.3245849609375, -4.148681640625, 35.532325744628906, -3.128175735473633, 1.3610954284667969, 2.602783203125, -17.03375244140625, 4.195335388183594, 49.6011962890625, 4.7990875244140625, 4.913360595703125, 14.51763916015625, 55.09014892578125, 45.17372131347656, 30.04347038269043, 0.0, 63.427978515625, 40.770843505859375, 58.35115051269531, 22.490997314453125, 47.46014404296875, 38.78826904296875, 47.337890625, -3.771209716796875, 18.128021240234375, 19.94463348388672, 18.32952880859375, 53.62298583984375, -59.101715087890625, 3.6299667358398438, -3.91168212890625, 35.26702880859375, 51.08135986328125, 37.9075927734375, -0.108673095703125, 70.84506225585938, 18.772613525390625, 6.18023681640625, 20.567398071289062, 65.3175048828125, -4.515296936035156, 52.804443359375, 9.08203125, 12.774505615234375, 19.5491943359375, 18.6007080078125, -6.5736846923828125, 22.250534057617188, 4.94659423828125, 40.12602233886719, 1.14825439453125, -0.37725067138671875, 8.229827880859375, 33.21728515625, 15.423660278320312, 36.57867431640625, 36.906707763671875, -12.673171997070312, 49.099151611328125, 72.25137329101562, 26.789199829101562, 20.264495849609375, 8.849929809570312, 14.288330078125, 41.30896759033203, -0.0625152587890625, 2.7913055419921875, 21.457672119140625, 67.29786682128906, 47.506317138671875, 56.40959167480469, 10.659862518310547, 6.04747200012207, 20.220489501953125, 10.53009033203125, 7.63720703125, 47.889434814453125, 46.16845703125, 41.17498779296875, 7.457740783691406, 3.6970977783203125, 2.9419021606445312, 30.586380004882812, 78.92996215820312, 30.003292083740234, 74.5382080078125, 66.34263610839844, 24.884750366210938, 6.09930419921875, 64.63595581054688, -24.259300231933594, 29.322052001953125, 10.40561294555664, 62.80841064453125, 34.09239196777344, 5.879547119140625, -20.38385009765625, 71.33699035644531, 59.458404541015625, 13.449554443359375, -49.13954162597656, -40.698638916015625, 33.43718719482422, 32.775970458984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000153.npy"}
{"epoch": 0.3204188481675393, "step": 154, "batch_size": 128, "mean": 17.88668441772461, "std": 32.72853469848633, "min": -79.71685791015625, "p10": -23.833319091796874, "median": 18.564178466796875, "p90": 62.09308013916015, "max": 87.85638427734375, "pos_frac": 0.71875, "sample": [25.444137573242188, 51.05406188964844, -27.6019287109375, 64.2103271484375, 33.860931396484375, 7.664909362792969, -6.233903884887695, 60.719696044921875, 6.9104766845703125, 16.7388916015625, -51.26054382324219, 39.80999755859375, 20.605026245117188, 14.422149658203125, 35.34187316894531, -79.71685791015625, 36.71708679199219, -27.97705078125, 35.62549591064453, 5.98541259765625, 11.660324096679688, 0.0017547607421875, 7.407989501953125, -24.855056762695312, -5.636993408203125, 5.262855529785156, 35.53778076171875, 8.897720336914062, 4.1221923828125, 76.44601440429688, 12.43572998046875, -6.72808837890625, 59.670684814453125, 67.42071533203125, 4.5395050048828125, 51.222198486328125, 29.674530029296875, 70.576171875, 41.13623046875, 67.2886962890625, 2.27001953125, 57.610595703125, 66.87057495117188, 30.3336181640625, -55.03828430175781, 18.134765625, 80.5169677734375, -1.1864089965820312, -8.4395751953125, -28.340728759765625, -4.5531158447265625, -10.086311340332031, 19.724090576171875, -23.7762451171875, 26.53985595703125, 44.20515441894531, 35.442108154296875, -6.253143310546875, 3.572906494140625, -21.0311279296875, -2.8117103576660156, 11.519866943359375, 48.840789794921875, 24.32939910888672, 26.12078857421875, -55.020843505859375, -18.06989097595215, -0.1617584228515625, -2.7408828735351562, -9.474945068359375, -10.768810272216797, -47.753326416015625, 54.76609802246094, 70.30099487304688, -3.9380950927734375, -8.297401428222656, 11.3218994140625, -3.89349365234375, 36.310791015625, 9.740325927734375, -38.366363525390625, 12.003715515136719, -1.4692001342773438, 23.664825439453125, 38.228118896484375, 10.97003173828125, 59.325042724609375, 35.695892333984375, -7.086578369140625, 34.012351989746094, 24.2474365234375, 21.173736572265625, 61.78172302246094, 50.41234588623047, 37.56024169921875, -16.06878662109375, 22.806365966796875, 3.2210693359375, 87.85638427734375, 15.940765380859375, 21.0760498046875, 50.392333984375, -23.96649169921875, -67.41632080078125, 41.414459228515625, 28.644180297851562, 36.28184509277344, 33.128082275390625, 18.62567138671875, 24.965133666992188, 69.9984130859375, 35.80201721191406, 34.543853759765625, 57.57560729980469, 1.27984619140625, 4.391412734985352, 81.24041748046875, 18.14862823486328, 18.502685546875, 24.1268310546875, -44.98382568359375, -12.296531677246094, 62.819580078125, 33.69029235839844, 3.386383056640625, 22.198974609375, 38.96269226074219, 65.81771087646484], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000154.npy"}
{"epoch": 0.3225130890052356, "step": 155, "batch_size": 128, "mean": 13.252790451049805, "std": 29.83599281311035, "min": -62.603759765625, "p10": -22.34083786010742, "median": 11.935765266418457, "p90": 50.68471755981444, "max": 86.6175537109375, "pos_frac": 0.6875, "sample": [69.8421630859375, 26.720596313476562, 16.9639892578125, -5.7396697998046875, 38.609649658203125, -0.7359180450439453, 11.3734130859375, -2.0089111328125, 7.50225830078125, 49.10105895996094, 44.44443893432617, 4.199127197265625, 29.62469482421875, -22.928680419921875, 25.628509521484375, 32.250946044921875, -16.634857177734375, 74.31399536132812, 86.6175537109375, -6.8797454833984375, 11.994155883789062, 28.521041870117188, 47.6556396484375, 85.618896484375, 6.558967590332031, 43.29241943359375, -43.841644287109375, 0.0, -36.007537841796875, -11.196868896484375, -21.774524688720703, 55.42681884765625, 22.377975463867188, 49.330467224121094, 1.6890869140625, -15.943756103515625, -22.088905334472656, 42.44950485229492, -8.908935546875, 0.0896759033203125, 7.75347900390625, 30.086044311523438, 17.490234375, 2.662334442138672, 37.23236083984375, 56.620849609375, 37.072418212890625, 15.14068603515625, 5.2945709228515625, 36.727699279785156, -10.682060241699219, -52.885009765625, 8.538421630859375, 8.524566650390625, 3.441659927368164, -37.586761474609375, 0.483123779296875, 37.93951416015625, 75.24649047851562, 53.844635009765625, 3.4974822998046875, 4.620693206787109, 23.06787109375, -8.398796081542969, -43.996421813964844, -8.625680923461914, 46.660888671875, 60.18809509277344, 13.899761199951172, 17.317825317382812, -1.6301422119140625, -31.014907836914062, 40.224761962890625, 46.0556640625, -15.873809814453125, 32.9859619140625, 8.708526611328125, -7.371845245361328, 42.487640380859375, -27.70257568359375, 34.48866271972656, 56.370513916015625, -28.037933349609375, -14.507461547851562, 1.1058807373046875, 7.6148834228515625, 40.58740234375, -40.7447509765625, 1.7388153076171875, 42.87255859375, 19.085922241210938, 27.771224975585938, 23.44561767578125, -2.157684326171875, -43.5452880859375, 16.351058959960938, 14.472869873046875, 65.76206970214844, 12.248504638671875, -15.265419006347656, -15.41070556640625, -17.329151153564453, -4.16436767578125, 11.877374649047852, 17.50555419921875, 42.68878173828125, 5.328193664550781, -6.0610198974609375, -62.603759765625, 0.9304428100585938, -43.7393798828125, 13.685394287109375, 38.045654296875, -12.316802978515625, 56.427490234375, 32.23553466796875, 38.694427490234375, 26.411636352539062, 15.480499267578125, 61.555877685546875, -20.95166015625, 0.5939254760742188, -0.5833969116210938, 13.504608154296875, 18.78863525390625, 23.045562744140625, 14.85113525390625, 2.6558685302734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000155.npy"}
{"epoch": 0.32460732984293195, "step": 156, "batch_size": 128, "mean": 22.433757781982422, "std": 34.127906799316406, "min": -79.358642578125, "p10": -23.21388702392578, "median": 25.064956665039062, "p90": 62.61811218261719, "max": 97.6214599609375, "pos_frac": 0.765625, "sample": [39.85597229003906, 41.927101135253906, 30.97142791748047, 21.844390869140625, -1.849945068359375, 4.824577331542969, 48.977020263671875, 48.66029357910156, 33.009613037109375, 28.13712501525879, 44.230918884277344, 27.85302734375, 56.7891845703125, 35.056549072265625, 2.5783233642578125, -22.8240966796875, -7.007987976074219, 34.26995849609375, 7.005462646484375, 43.75830841064453, 3.091552734375, 46.037628173828125, -11.081314086914062, -32.750572204589844, 28.68011474609375, 10.295883178710938, -3.35736083984375, 22.468246459960938, 0.8691482543945312, 46.197021484375, 72.92849731445312, -6.701812744140625, 72.9139404296875, -6.18170166015625, 13.416540145874023, -26.903182983398438, -39.38435363769531, 5.392578125, 52.87188720703125, 3.3948516845703125, 24.11810302734375, 62.8460693359375, 27.8211669921875, 10.451263427734375, 6.4646148681640625, 2.743011474609375, 46.04493713378906, -36.82379150390625, 47.28496170043945, 33.162994384765625, 3.4270973205566406, 37.89112091064453, 22.058135986328125, 26.011810302734375, 88.49932861328125, -25.41436767578125, -40.018280029296875, 89.3106689453125, -1.706024169921875, 9.701522827148438, -7.6923065185546875, 86.96063232421875, 36.57359313964844, 53.00767517089844, 34.10191345214844, 60.5533447265625, 53.418670654296875, 15.262863159179688, 61.830718994140625, 5.9147796630859375, 22.47418212890625, 0.5851898193359375, -6.37725830078125, 68.34967041015625, 41.481842041015625, 32.026519775390625, 97.6214599609375, 17.703521728515625, 0.4272422790527344, -3.665773391723633, 85.5946044921875, 31.379638671875, 73.23187255859375, 3.312591552734375, 9.330169677734375, 27.334205627441406, 2.832273483276367, 36.50956726074219, 31.35120391845703, -28.43377685546875, 47.068359375, 52.862548828125, 38.39915466308594, 49.1239013671875, 27.33624267578125, 14.739990234375, -19.66192626953125, 62.520416259765625, 27.928466796875, -24.123397827148438, 61.316162109375, 12.941162109375, 23.756179809570312, -26.7371826171875, -79.358642578125, 55.673797607421875, -3.2630863189697266, 88.74801635742188, -2.0912017822265625, -40.889068603515625, 7.363677978515625, 32.4268798828125, 89.62615966796875, 11.360626220703125, -77.75927734375, -38.671573638916016, 1.397216796875, 47.265533447265625, 54.3634033203125, 59.073974609375, -1.0805168151855469, 27.548248291015625, 43.93086242675781, 63.910888671875, -20.563125610351562, 60.92132568359375, -8.047378540039062, 5.024139404296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000156.npy"}
{"epoch": 0.3267015706806283, "step": 157, "batch_size": 128, "mean": 15.518184661865234, "std": 30.723033905029297, "min": -64.40483093261719, "p10": -22.83521041870117, "median": 11.754791259765625, "p90": 56.163937377929685, "max": 100.23675537109375, "pos_frac": 0.6953125, "sample": [28.421676635742188, -0.23954010009765625, 24.717681884765625, 65.52418518066406, -1.393798828125, -25.00342559814453, 14.148681640625, 18.2491455078125, 58.47784423828125, -7.3177490234375, 25.307220458984375, 0.01474761962890625, -45.933807373046875, 26.925369262695312, 1.4650344848632812, 37.13775634765625, 11.608673095703125, -1.2396316528320312, 3.0534095764160156, -12.656204223632812, -17.278732299804688, 67.41619873046875, 30.475662231445312, -10.824127197265625, 49.87295913696289, -1.8030223846435547, -36.3646240234375, -0.2191009521484375, 5.3405914306640625, 16.65545654296875, 8.487701416015625, -5.29034423828125, 10.852249145507812, -50.841796875, 36.48369598388672, 55.989593505859375, 22.1209716796875, 33.53822326660156, -45.54014587402344, -64.40483093261719, 83.62469482421875, 16.37464141845703, 34.208892822265625, 19.70068359375, 82.54345703125, 40.82795715332031, 14.347877502441406, -11.088027954101562, 22.47406005859375, 0.0, 19.212600708007812, 50.81922912597656, 5.4237213134765625, -4.856781005859375, 68.84307861328125, 12.603271484375, 42.338958740234375, -8.156219482421875, -4.775177001953125, 1.3693695068359375, 53.74786376953125, 25.0367431640625, 55.688453674316406, 11.900909423828125, 33.632301330566406, -3.1235580444335938, 53.030975341796875, -4.1058349609375, -50.76861572265625, 41.49418640136719, 1.8175048828125, -17.161346435546875, -46.079620361328125, 7.32330322265625, -13.55645751953125, 18.672943115234375, -34.75360107421875, 45.00201416015625, 6.927055358886719, 56.57073974609375, 21.27032470703125, 41.73992919921875, -1.083740234375, -7.152618408203125, 3.1163406372070312, 8.390167236328125, 6.7837677001953125, 14.711273193359375, 25.278919219970703, 0.0, -27.32257080078125, 61.551849365234375, 22.134490966796875, 15.473556518554688, 5.7740020751953125, -3.109405517578125, 29.225690841674805, 57.47022247314453, 99.66510009765625, 56.72172546386719, 9.936248779296875, -21.905975341796875, 39.938812255859375, 6.957403182983398, 4.771728515625, 3.5812149047851562, 36.71044921875, 37.212039947509766, 28.270172119140625, -8.53399658203125, -25.62371826171875, 42.408050537109375, 4.753326416015625, 40.766998291015625, 29.979896545410156, 42.835418701171875, -5.54876708984375, 1.1372718811035156, -29.19158935546875, 47.86370849609375, -27.92547607421875, 7.960319519042969, 6.559812545776367, 39.689788818359375, 7.27789306640625, 20.437240600585938, 100.23675537109375, 62.06755065917969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000157.npy"}
{"epoch": 0.3287958115183246, "step": 158, "batch_size": 128, "mean": 17.02230453491211, "std": 30.808704376220703, "min": -61.2291259765625, "p10": -25.397113037109374, "median": 17.17669677734375, "p90": 55.43851013183593, "max": 86.17718505859375, "pos_frac": 0.71875, "sample": [5.081277847290039, -2.3408966064453125, 22.30987548828125, 56.561309814453125, -34.398162841796875, 5.1186065673828125, -1.7258148193359375, 23.19915771484375, -19.184555053710938, 56.29656982421875, -18.08819580078125, 0.5107421875, -43.08746337890625, -37.895965576171875, 24.367584228515625, 20.317474365234375, 35.508792877197266, 31.358245849609375, 9.972900390625, 31.603408813476562, 23.45014190673828, 62.74859619140625, 41.447601318359375, 3.8293914794921875, 16.96446990966797, 43.88240051269531, 11.81787109375, -61.2291259765625, 48.78582763671875, 22.649124145507812, -10.926651000976562, 34.13017272949219, -13.74310302734375, 56.31024169921875, 21.15380859375, 6.452415466308594, 41.479644775390625, -41.51076126098633, 19.966102600097656, -43.6063232421875, 35.54396057128906, 5.92913818359375, 14.38409423828125, -27.20604705810547, 17.1572265625, -1.6326065063476562, -6.2177886962890625, 17.852401733398438, 30.935516357421875, 25.230846405029297, 66.93350219726562, -4.402412414550781, 2.146146774291992, 9.120124816894531, 36.95783996582031, 48.96734619140625, -38.10497283935547, -21.66204833984375, 66.2816162109375, 17.962890625, 63.53411865234375, 7.348014831542969, -4.9405517578125, 10.706787109375, 35.344970703125, 14.255828857421875, -22.126144409179688, 0.0, 32.284934997558594, 15.5308837890625, 71.6484375, -6.693084716796875, -3.0977783203125, 50.129638671875, 25.611572265625, 54.574005126953125, 7.768524169921875, -26.949981689453125, -13.765975952148438, -2.2650527954101562, 53.31871032714844, 9.4715576171875, -49.173187255859375, 16.316238403320312, 33.12507629394531, 36.73028564453125, 34.757293701171875, 0.1143798828125, 44.9395751953125, 40.132568359375, 49.777557373046875, 30.12646484375, 11.986045837402344, 18.256988525390625, 72.0809326171875, 47.03472900390625, 82.14517211914062, -6.597087860107422, -31.40838623046875, -24.731597900390625, 14.99591064453125, 85.47923278808594, -18.862213134765625, 36.97557067871094, 50.74189758300781, 49.29827880859375, 56.883636474609375, 36.427093505859375, 41.399871826171875, -2.9901161193847656, 17.1961669921875, 33.905548095703125, 38.20238494873047, 8.67034912109375, 29.880081176757812, -16.9501953125, 55.070770263671875, 3.8165817260742188, -35.16523742675781, -10.4161376953125, 32.468994140625, 51.1268310546875, -47.39404296875, 12.23992919921875, 86.17718505859375, 17.125274658203125, 28.442962646484375, 1.09259033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000158.npy"}
{"epoch": 0.3308900523560209, "step": 159, "batch_size": 128, "mean": 13.638815879821777, "std": 36.00312805175781, "min": -92.70745849609375, "p10": -32.19681396484375, "median": 10.173904418945312, "p90": 62.40482788085937, "max": 101.16310119628906, "pos_frac": 0.6171875, "sample": [-22.510711669921875, 58.98534393310547, 25.2864990234375, 34.8673095703125, -4.469970703125, -18.852081298828125, -40.1982421875, 51.1322021484375, -4.453094482421875, 64.37030029296875, -55.42036437988281, 9.800006866455078, -92.70745849609375, -12.199699401855469, 8.214168548583984, 41.15526580810547, 12.495208740234375, -5.9565582275390625, -37.603851318359375, 46.55357360839844, 25.33758544921875, 4.2645263671875, -13.690505981445312, -27.958160400390625, 23.909027099609375, 62.269622802734375, 23.604843139648438, 61.1373291015625, 36.21490478515625, -7.7716217041015625, 24.52125358581543, 0.40203857421875, 46.91204833984375, 18.15149688720703, 21.10675811767578, -20.429397583007812, -0.211761474609375, -28.326904296875, -4.626651763916016, -39.6484375, 49.09745788574219, 16.69854736328125, 27.44873046875, 44.00889587402344, -11.6126708984375, 19.50152587890625, -30.98883056640625, 3.6717376708984375, -40.28839111328125, -20.019256591796875, 1.405609130859375, -8.547065734863281, 52.251007080078125, 10.659393310546875, 18.02685546875, 53.16123962402344, 2.3714828491210938, 51.01220703125, -10.089950561523438, -3.20989990234375, 101.16310119628906, 10.154327392578125, -30.62158966064453, 13.60772705078125, 79.07818603515625, 14.5797119140625, 54.650054931640625, 4.6435546875, -10.404022216796875, 12.115081787109375, 2.6286163330078125, 36.96337890625, -5.151729583740234, 49.99925994873047, 39.2935791015625, -35.69525146484375, 2.870941162109375, 62.950592041015625, -18.269882202148438, 19.071502685546875, -18.8931884765625, -40.12640380859375, 32.002342224121094, 74.33340454101562, 82.71868896484375, -10.437511444091797, -14.085006713867188, -37.80039978027344, -9.277984619140625, 62.720306396484375, 28.276649475097656, 18.847488403320312, -23.308197021484375, 24.111480712890625, 11.958770751953125, 32.91595458984375, -36.06182861328125, 65.9334716796875, -12.01776123046875, 44.08575439453125, 35.83331298828125, -28.6137638092041, -35.01544189453125, -1.6409454345703125, 98.36233520507812, 80.26446533203125, 23.46636962890625, -55.73500061035156, 46.19261932373047, 7.657436370849609, -5.78857421875, -11.4344482421875, 65.37519836425781, -50.378753662109375, 78.40701293945312, -8.359375, 53.4605712890625, 38.22467041015625, 56.05830383300781, 51.505126953125, 10.1934814453125, 8.680789947509766, -4.59564208984375, 75.60055541992188, 9.732643127441406, 7.3182373046875, -12.503814697265625, 45.7634162902832], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000159.npy"}
{"epoch": 0.33298429319371725, "step": 160, "batch_size": 128, "mean": 20.96258544921875, "std": 31.36819076538086, "min": -50.508758544921875, "p10": -13.484561157226562, "median": 15.861442565917969, "p90": 60.47094268798827, "max": 112.9111328125, "pos_frac": 0.7578125, "sample": [-5.130657196044922, 47.287994384765625, 0.0, 91.606201171875, 28.918258666992188, 26.692138671875, 31.97808837890625, 18.27667236328125, 83.022216796875, -2.800750732421875, 57.69708251953125, 69.72119140625, 32.030906677246094, 72.7396240234375, 18.33514404296875, -2.96551513671875, 74.869873046875, 28.205230712890625, 8.084014892578125, 15.577224731445312, 6.86865234375, 9.770263671875, 31.346298217773438, 36.473724365234375, 0.2986297607421875, -8.766250610351562, -6.8490753173828125, 81.76954650878906, 2.2045955657958984, -50.508758544921875, 14.95526123046875, 110.79443359375, -37.840553283691406, 52.27300262451172, 19.027236938476562, 2.4963607788085938, 10.073265075683594, 3.595670700073242, 30.805831909179688, 10.791942596435547, 37.13737487792969, 52.73052978515625, 9.1956787109375, 0.0, 43.80902099609375, -14.0194091796875, 36.105491638183594, -25.352066040039062, -12.821182250976562, 6.3954315185546875, 24.25286865234375, 36.94834899902344, 112.9111328125, 74.23831176757812, -21.904708862304688, 12.647102355957031, 29.3189697265625, -15.757171630859375, 10.496833801269531, -13.035903930664062, -9.9068603515625, 55.68585205078125, 9.912353515625, -36.15667724609375, -9.85247802734375, 59.260467529296875, 57.82684326171875, 58.98258972167969, 5.174081802368164, 15.114990234375, 32.95130157470703, 2.4199981689453125, -0.15619468688964844, 12.13641357421875, 29.9014892578125, 35.606231689453125, 71.88409423828125, 51.665611267089844, 35.83404541015625, 18.40423583984375, -27.43719482421875, -4.7527313232421875, 44.28345489501953, -35.21111297607422, -13.255340576171875, 64.82196044921875, 33.775062561035156, 9.0667724609375, 43.8341064453125, -12.738540649414062, 9.23785400390625, 10.560417175292969, 1.2677764892578125, -29.054656982421875, 31.279083251953125, -49.02777099609375, -1.7474212646484375, 29.810516357421875, 64.89584350585938, 2.03717041015625, 14.264408111572266, 14.258781433105469, -17.328351974487305, 26.506103515625, 18.39990234375, 8.606010437011719, 9.133232116699219, -8.063713073730469, 43.50921630859375, 26.000198364257812, 56.46343994140625, 3.1271209716796875, 16.145660400390625, 45.412330627441406, -5.767242431640625, 28.301231384277344, 56.729248046875, 11.953155517578125, 33.058685302734375, 59.60920715332031, 62.481658935546875, 31.953994750976562, -35.564971923828125, 38.101806640625, 17.271217346191406, 56.328033447265625, 2.1841278076171875, 4.781074523925781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000160.npy"}
{"epoch": 0.33507853403141363, "step": 161, "batch_size": 128, "mean": 21.996055603027344, "std": 34.466636657714844, "min": -72.74468994140625, "p10": -16.1713623046875, "median": 19.599868774414062, "p90": 71.83347778320312, "max": 90.68951416015625, "pos_frac": 0.7109375, "sample": [-4.500968933105469, 1.4331855773925781, -3.149383544921875, -71.49464416503906, 0.7814159393310547, 29.884658813476562, 3.880126953125, 16.446441650390625, 7.14776611328125, 6.280517578125, 4.086910247802734, 80.88031005859375, 30.903533935546875, -37.1265869140625, 3.9763946533203125, 2.4533462524414062, 78.28326416015625, 22.1417236328125, 37.72584915161133, -4.848052978515625, 49.162078857421875, 14.09600830078125, 59.44743347167969, 81.29522705078125, -24.76439666748047, 3.63421630859375, 5.389984130859375, 79.32901000976562, 8.088104248046875, 34.16219711303711, 6.631378173828125, 30.63979148864746, -7.3428802490234375, -15.15570068359375, 28.403793334960938, 26.471832275390625, 38.448822021484375, 56.0576171875, -1.7385406494140625, 14.0322265625, -5.740234375, -45.13975524902344, -34.69537353515625, -9.486190795898438, 13.943840026855469, 46.4815673828125, -16.011932373046875, -2.580718994140625, 0.0, 42.230812072753906, 19.4910888671875, 40.412200927734375, 8.063789367675781, -0.00630950927734375, 12.259521484375, -0.667724609375, 55.472564697265625, 58.659019470214844, 43.145851135253906, 27.219757080078125, 60.652069091796875, 68.10659790039062, -20.954833984375, 19.708648681640625, -12.141998291015625, 70.12457275390625, 59.507904052734375, 77.25885009765625, -17.146591186523438, 15.59332275390625, 81.8687744140625, 26.82369613647461, 40.60504150390625, 70.18386840820312, -38.400665283203125, 74.3787841796875, 33.187828063964844, -2.3065643310546875, 0.18695068359375, 40.21833038330078, -16.595291137695312, -1.3807754516601562, 90.68951416015625, 72.55654907226562, 30.452972412109375, 77.08828735351562, 2.224273681640625, 54.072998046875, 85.85382080078125, -1.5891799926757812, 0.12738037109375, 35.82234191894531, -18.328338623046875, 41.953582763671875, -15.131622314453125, 29.73541259765625, 70.13662719726562, -9.037574768066406, 40.2261962890625, -8.97869873046875, 26.235939025878906, 62.96907043457031, 22.800079345703125, -16.543365478515625, 18.270126342773438, 44.58604431152344, -13.85931396484375, 29.9727783203125, 42.209556579589844, 53.690399169921875, 61.92950439453125, 71.52359008789062, -14.78753662109375, 38.884857177734375, 29.877405166625977, -54.14398193359375, 74.931396484375, 10.605621337890625, 10.680534362792969, 0.0, -5.93426513671875, 74.03744506835938, 60.86968994140625, -72.74468994140625, 29.3221435546875, 20.388534545898438, 1.5570220947265625, 56.2877197265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000161.npy"}
{"epoch": 0.33717277486910996, "step": 162, "batch_size": 128, "mean": 19.879669189453125, "std": 33.29869079589844, "min": -80.42835998535156, "p10": -21.058114624023435, "median": 16.794082641601562, "p90": 63.78313598632812, "max": 90.440673828125, "pos_frac": 0.7421875, "sample": [-8.045547485351562, 7.256355285644531, 67.76235961914062, 41.990013122558594, 3.542236328125, 18.10791015625, -20.89239501953125, 27.5511474609375, 7.125881195068359, 25.07990264892578, 55.84135437011719, 87.8499755859375, 14.378471374511719, 29.64911651611328, -5.62158203125, -15.153335571289062, 24.90972900390625, 28.0230712890625, 1.7824630737304688, 4.339618682861328, -33.4710693359375, 43.5601806640625, -36.97142028808594, 33.05320739746094, 16.48089599609375, 36.639305114746094, 67.92816162109375, 11.9637451171875, -43.49848937988281, 7.406982421875, -0.4288787841796875, 17.8975830078125, 40.767578125, -32.67814636230469, 82.21484375, 7.8368682861328125, 16.19757080078125, -66.42019653320312, -10.076642990112305, 39.53691101074219, 4.023906707763672, 57.9639892578125, 4.348772048950195, 69.81927490234375, 65.13336181640625, -80.42835998535156, 62.645751953125, 34.95738220214844, 28.788330078125, 66.3216552734375, -3.416675567626953, 44.5986328125, 43.582672119140625, 1.20794677734375, 28.608047485351562, -3.5405502319335938, -14.615921020507812, 45.63047790527344, 73.6861572265625, 27.311767578125, 73.04400634765625, 90.440673828125, -4.2877197265625, -0.003265380859375, 40.11973571777344, 57.12274169921875, 58.34590148925781, 71.45012664794922, -19.108795166015625, 50.43952941894531, -0.4278717041015625, 9.846691131591797, -31.936004638671875, 18.8905029296875, 89.9600830078125, -25.032859802246094, -25.571945190429688, 1.440826416015625, 17.107269287109375, 59.219573974609375, -6.497842788696289, 15.3541259765625, 45.55767822265625, 31.521224975585938, 61.68218994140625, 59.315338134765625, 6.670124053955078, -0.49407958984375, 33.25831604003906, 29.014205932617188, 13.0711669921875, 12.45654296875, 8.536209106445312, -62.07084655761719, 6.4700927734375, 33.55458068847656, 58.30640411376953, 39.275390625, 17.4129638671875, 62.3699951171875, 46.31748962402344, -34.07489013671875, 50.8333740234375, 9.0087890625, 2.3399810791015625, 69.9403076171875, -19.196670532226562, 7.5012969970703125, -3.4026031494140625, 49.59588623046875, -34.67401123046875, 5.164154052734375, -10.43505859375, 40.78125, 29.425567626953125, 38.695831298828125, 14.139400482177734, -2.6256885528564453, 27.159820556640625, 19.88592529296875, 7.387979507446289, -21.444793701171875, 0.7760467529296875, -1.569061279296875, 5.143547058105469, 63.2044677734375, 29.330413818359375, 9.523590087890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000162.npy"}
{"epoch": 0.3392670157068063, "step": 163, "batch_size": 128, "mean": 23.116073608398438, "std": 29.136882781982422, "min": -51.903076171875, "p10": -13.286125183105469, "median": 24.41803741455078, "p90": 63.0544448852539, "max": 89.6552734375, "pos_frac": 0.75, "sample": [24.71826171875, -3.3080291748046875, 44.45684051513672, -13.649551391601562, 27.715240478515625, -12.577293395996094, 27.24169921875, 24.690200805664062, 38.2412109375, 4.5557098388671875, 2.68212890625, 67.7603988647461, 44.778778076171875, 15.3067626953125, 75.0703125, 19.71551513671875, 12.825332641601562, -15.6077880859375, 8.854690551757812, -10.882774353027344, -17.99969482421875, 39.058597564697266, 25.24139404296875, -0.8762741088867188, 14.93499755859375, 57.77485656738281, -6.626291275024414, 16.483047485351562, 51.2545166015625, -3.573638916015625, -5.6984100341796875, 62.756591796875, -1.79583740234375, 46.740814208984375, 36.74485778808594, 25.83282470703125, 13.360992431640625, 51.0567626953125, -12.8974609375, -5.697113037109375, 38.580108642578125, 36.31890869140625, -25.634140014648438, 10.990432739257812, 11.33819580078125, 27.456756591796875, -46.012847900390625, 64.1219482421875, 47.27827453613281, -34.99017333984375, 37.68988800048828, 6.26580810546875, 41.00140380859375, 34.678375244140625, -3.5032577514648438, 58.93794250488281, 20.6507568359375, 65.712646484375, 65.01722717285156, 39.077392578125, 3.570404052734375, -4.4950408935546875, -13.13037109375, 38.882408142089844, -9.587356567382812, -10.17071533203125, 9.666683197021484, 51.083038330078125, 6.197235107421875, 34.07600402832031, -51.903076171875, 61.86761474609375, 28.870262145996094, 23.89727783203125, 14.573699951171875, -16.245086669921875, 75.22738647460938, -15.203887939453125, 67.11276245117188, 69.778076171875, 35.23217010498047, 47.443084716796875, 24.1458740234375, 16.40096664428711, 44.717124938964844, 42.44103240966797, 36.94856262207031, -0.31048583984375, 64.98564147949219, 50.47099304199219, 5.572134017944336, 5.791259765625, 23.61927032470703, 89.6552734375, 37.5936279296875, -31.96368408203125, -26.37273406982422, 28.17659568786621, 77.45474243164062, 32.420806884765625, 55.686187744140625, 51.328704833984375, -19.257003784179688, 48.96624755859375, -1.9906692504882812, 15.842864990234375, 63.74943542480469, 17.774917602539062, 20.938018798828125, 29.823204040527344, 60.6708984375, 38.1680908203125, 17.37164306640625, 1.596963882446289, 17.799522399902344, 11.902557373046875, 37.08587646484375, 56.71217346191406, 26.17230224609375, 66.62914276123047, -3.39605712890625, 2.820892333984375, -35.515869140625, 60.65229797363281, 51.52964782714844, 33.675148010253906, 9.285551071166992, -9.293502807617188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000163.npy"}
{"epoch": 0.3413612565445026, "step": 164, "batch_size": 128, "mean": 19.850873947143555, "std": 31.897022247314453, "min": -55.3580322265625, "p10": -12.698802185058591, "median": 16.44231128692627, "p90": 64.15668487548828, "max": 101.20175170898438, "pos_frac": 0.7265625, "sample": [-8.942840576171875, 23.410755157470703, -2.820058822631836, 56.429840087890625, -0.049530029296875, -1.309661865234375, -7.182624816894531, -2.80902099609375, 27.907577514648438, 30.881103515625, 35.290008544921875, -12.128204345703125, 44.03009033203125, 43.110137939453125, 25.09796142578125, -48.81272888183594, -2.0150833129882812, 55.765777587890625, 18.279460906982422, 67.12295532226562, 1.172821044921875, -47.8861083984375, 49.0897216796875, 13.72259521484375, 12.542274475097656, -26.755069732666016, 30.5462646484375, 7.005828857421875, -10.1295166015625, 4.218376159667969, 38.426971435546875, -0.024261474609375, 4.14093017578125, 62.16423034667969, -6.5196380615234375, 52.71124267578125, -37.51202392578125, 2.6424713134765625, 1.6485366821289062, -44.91490173339844, -2.11627197265625, 70.50975036621094, 46.463409423828125, 34.93701171875, 67.85989379882812, 8.403030395507812, 55.888519287109375, -2.2440185546875, 1.170654296875, 3.118316650390625, -35.47734069824219, 13.049880981445312, 18.887359619140625, 29.369775772094727, -15.849563598632812, 65.85498046875, -1.3684539794921875, 43.805389404296875, 30.08514404296875, 33.5814208984375, 0.224273681640625, 8.577789306640625, 64.89579772949219, 67.42684936523438, -14.030197143554688, 65.95188903808594, 18.911788940429688, -10.416717529296875, 46.32464599609375, 13.872100830078125, 83.502197265625, 31.34954833984375, 70.9892578125, 4.738800048828125, -8.815532684326172, 0.5390701293945312, 23.856826782226562, 5.7458648681640625, -50.104766845703125, 23.441940307617188, 27.708740234375, 57.18719482421875, -23.776832580566406, 3.16259765625, -0.0982513427734375, 40.58876037597656, 55.890380859375, 65.693359375, 41.256805419921875, 25.861846923828125, -6.352720260620117, 48.67076873779297, 0.42452239990234375, 57.699127197265625, 12.20257568359375, 50.3990478515625, 50.958274841308594, 64.284912109375, -11.6085205078125, 1.56011962890625, -43.149383544921875, 30.481170654296875, 0.4399261474609375, 30.522994995117188, 7.470527648925781, -7.5859375, 55.75384521484375, -8.59368896484375, 49.25390625, 5.8016815185546875, 35.16221618652344, 43.987060546875, 8.902885437011719, 16.38421630859375, -55.3580322265625, -23.007568359375, 7.8275909423828125, -11.03759765625, 101.09814453125, 20.26959228515625, 30.14053726196289, 47.11907958984375, 14.18017578125, 32.603057861328125, 48.27191162109375, 64.10173034667969, 101.20175170898438, 16.50040626525879], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000164.npy"}
{"epoch": 0.34345549738219894, "step": 165, "batch_size": 128, "mean": 16.975677490234375, "std": 30.35817527770996, "min": -64.26591491699219, "p10": -15.343307495117188, "median": 10.174072265625, "p90": 57.150254821777345, "max": 93.66091918945312, "pos_frac": 0.6640625, "sample": [5.375030517578125, 58.015106201171875, -6.618415832519531, 50.285400390625, -9.39837646484375, 4.620628356933594, 37.58893585205078, 1.653076171875, 48.83416748046875, 46.22723388671875, 66.73651123046875, 3.5269546508789062, -60.470977783203125, 46.453338623046875, 66.06527709960938, 15.8443603515625, -7.27410888671875, 69.84552001953125, -6.53106689453125, 47.41094970703125, 43.03877258300781, -43.944732666015625, 56.31871032714844, 42.62212371826172, 7.6917572021484375, -25.25516128540039, -25.235107421875, -12.4569091796875, 18.844085693359375, 35.178375244140625, 6.528228759765625, -8.60137939453125, -2.4952392578125, 3.3929901123046875, -0.09520339965820312, 13.821044921875, 29.894485473632812, -4.1648712158203125, -6.682371139526367, -15.296134948730469, 19.35302734375, 13.567115783691406, 56.326202392578125, -20.439590454101562, 15.53521728515625, 54.48248291015625, -1.2120246887207031, -1.4471664428710938, -15.129608154296875, 11.320480346679688, 87.2880859375, 4.2934417724609375, 5.0401153564453125, 2.107269287109375, 64.54107666015625, 31.165618896484375, -10.339385986328125, 4.508441925048828, -5.344646453857422, 13.111846923828125, -2.946746826171875, 70.43792724609375, -11.084892272949219, 46.997955322265625, 28.604522705078125, -16.73297119140625, 24.590530395507812, -4.493682861328125, -5.34967041015625, -7.194488525390625, 0.264495849609375, -2.9257278442382812, 25.342010498046875, 20.21455192565918, 32.43083190917969, 4.1683502197265625, 40.652191162109375, -24.67706298828125, 43.64984130859375, 16.6123046875, -30.911514282226562, 24.738365173339844, 70.32272338867188, -12.7197265625, 8.780742645263672, 41.779815673828125, -3.1664352416992188, 65.58807373046875, 55.06500244140625, 20.3399658203125, 13.6453857421875, 0.9364166259765625, 39.87744140625, 45.29141616821289, 54.33506774902344, -17.5037841796875, -15.453376770019531, 24.14251708984375, 47.36505126953125, 5.668144226074219, 93.66091918945312, -2.0626220703125, 35.200164794921875, 42.64884948730469, 64.03756713867188, -33.65818786621094, -8.830436706542969, 15.809051513671875, 57.14747619628906, 1.785491943359375, 0.5828895568847656, 5.962318420410156, 61.653411865234375, 32.551300048828125, 52.863433837890625, 48.59735107421875, 5.4366455078125, 14.845695495605469, 9.027664184570312, 49.62981414794922, -1.0928421020507812, 48.55677795410156, -64.26591491699219, -31.152496337890625, -3.9186477661132812, 57.15673828125, -9.139419555664062, -6.8487701416015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000165.npy"}
{"epoch": 0.34554973821989526, "step": 166, "batch_size": 128, "mean": 15.455398559570312, "std": 36.67471694946289, "min": -80.31024169921875, "p10": -27.320558166503904, "median": 11.3089599609375, "p90": 64.55638961791992, "max": 108.34576416015625, "pos_frac": 0.671875, "sample": [13.322418212890625, 3.418243408203125, -28.719436645507812, 15.440109252929688, -2.747161865234375, 11.816741943359375, 66.3211669921875, 60.73907470703125, -3.4438838958740234, 3.4451656341552734, 11.226760864257812, -3.0660629272460938, 107.04165649414062, -19.711822509765625, 2.7468490600585938, -6.817230224609375, 24.894302368164062, 26.343353271484375, 7.4585113525390625, 1.7103652954101562, -1.2377052307128906, -31.784088134765625, 27.715484619140625, 69.20744323730469, -60.187225341796875, -26.721038818359375, 1.6875534057617188, 32.55378723144531, -58.25408935546875, -20.797332763671875, -9.900970458984375, 19.67047119140625, 29.928848266601562, 75.25140380859375, 63.34832763671875, 80.1927490234375, 3.4851531982421875, -10.1746826171875, -17.341697692871094, -75.2322998046875, 6.065740585327148, -30.54821014404297, 3.7686920166015625, -1.9361572265625, -13.821380615234375, 72.93173217773438, 32.18463897705078, 19.91338348388672, -14.7073974609375, 35.5045166015625, -2.2334747314453125, 91.12081909179688, -48.405242919921875, 70.8343505859375, 96.15267944335938, -54.034637451171875, -36.56536865234375, 94.646240234375, 36.39642333984375, 45.17486572265625, -28.798004150390625, -10.67803955078125, -22.67535400390625, 39.819366455078125, 51.3536376953125, -25.218048095703125, 53.4501953125, 39.0732421875, 7.2520751953125, -21.343124389648438, 21.11444091796875, 26.55828857421875, 74.2713623046875, -8.367563247680664, -16.380630493164062, 23.0657958984375, 15.463294982910156, -0.6781177520751953, 13.089286804199219, 11.1142578125, -4.64759635925293, 48.1397705078125, 18.07707977294922, 9.156021118164062, 11.37841796875, 76.00059509277344, 1.99273681640625, 42.98439025878906, 17.399078369140625, -40.03248596191406, 5.531982421875, -8.536293029785156, 13.79428482055664, 59.09429931640625, 52.843963623046875, 28.18914794921875, 19.847686767578125, 53.67017364501953, 30.42359161376953, 6.13812255859375, 22.894912719726562, 1.5562191009521484, -1.9965438842773438, 50.73870849609375, 1.7398681640625, 32.61354064941406, 54.45684814453125, 48.9581298828125, -10.8909912109375, 63.80005645751953, -46.45684814453125, -80.31024169921875, 15.39892578125, 43.968353271484375, 58.73712158203125, -21.62078094482422, 12.160331726074219, -0.6651611328125, 19.68157958984375, 0.5071563720703125, 62.12445068359375, -1.2447967529296875, 108.34576416015625, 14.189844131469727, 3.81610107421875, 35.14707946777344, 11.199188232421875, 11.239501953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000166.npy"}
{"epoch": 0.34764397905759165, "step": 167, "batch_size": 128, "mean": 18.496248245239258, "std": 33.243865966796875, "min": -63.40510559082031, "p10": -15.370959472656248, "median": 13.9024658203125, "p90": 64.98301544189452, "max": 100.13839721679688, "pos_frac": 0.71875, "sample": [47.810752868652344, 5.2124481201171875, 16.794677734375, 44.602264404296875, 14.1878662109375, 55.083251953125, 44.91705322265625, 75.14755249023438, 32.56974792480469, 16.08551025390625, 59.59095764160156, -53.816864013671875, 68.05120849609375, 0.7873611450195312, 13.6170654296875, 79.171630859375, 15.647918701171875, 71.51968383789062, -1.309478759765625, 11.784423828125, 3.0693359375, 34.280426025390625, -57.28440856933594, -63.36695861816406, -1.5837326049804688, 1.659088134765625, -26.671218872070312, 54.41242980957031, 30.910369873046875, 57.121612548828125, 30.03173828125, 65.90782165527344, 21.20355987548828, 64.586669921875, 43.37603759765625, 7.82720947265625, 16.672767639160156, -34.15190124511719, 22.35443115234375, -9.155311584472656, 71.1563720703125, -4.769721984863281, 12.886199951171875, 2.642120361328125, 0.03961181640625, -63.40510559082031, 0.0, 1.324493408203125, 0.484405517578125, 34.63555908203125, 15.728424072265625, 74.66192626953125, -32.06158447265625, 67.59112548828125, -11.623672485351562, 14.5899658203125, 47.320274353027344, 13.009368896484375, 33.39398193359375, 1.875213623046875, -34.421844482421875, 2.697509765625, 86.224365234375, 0.2679462432861328, 28.142669677734375, 12.562347412109375, 18.798187255859375, -14.762161254882812, 37.27415466308594, 3.002676010131836, -0.506103515625, 1.0475654602050781, 40.91039276123047, -36.27497863769531, 48.7674560546875, 40.670135498046875, 58.21672058105469, -0.738525390625, 27.161865234375, 2.113494873046875, -2.8661041259765625, 48.87016296386719, 47.47850036621094, -28.305679321289062, 57.669677734375, -6.284393310546875, 19.445968627929688, -5.129665374755859, 33.21832275390625, 40.46742248535156, 36.709014892578125, 7.227630615234375, 40.83030700683594, 17.203857421875, 40.395172119140625, 83.9051513671875, -5.6041107177734375, -16.791488647460938, -14.51934814453125, -7.0956573486328125, -5.1173858642578125, 5.756040573120117, -11.711715698242188, 70.35726928710938, -9.324050903320312, 10.1702880859375, 100.13839721679688, 22.302734375, 46.3214111328125, -13.0867919921875, 7.2186279296875, 44.081298828125, -31.0670166015625, -3.7976303100585938, -3.1135711669921875, 4.319450378417969, 41.115234375, 4.3255615234375, -3.15728759765625, 52.09661102294922, 48.12422180175781, 34.046653747558594, 9.655502319335938, -2.8147506713867188, -57.9296875, 54.71356201171875, 6.767730712890625, 81.01655578613281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000167.npy"}
{"epoch": 0.34973821989528797, "step": 168, "batch_size": 128, "mean": 23.579952239990234, "std": 33.237403869628906, "min": -69.08917236328125, "p10": -10.945532226562499, "median": 19.730101585388184, "p90": 67.02071075439453, "max": 97.49786376953125, "pos_frac": 0.7578125, "sample": [26.883689880371094, 17.917694091796875, 77.01089477539062, 66.90374755859375, 44.2381591796875, 19.821744918823242, 46.09913635253906, 20.562217712402344, 50.43328857421875, -7.674247741699219, 51.146728515625, -12.108795166015625, 7.975494384765625, 11.699592590332031, 0.46411895751953125, 3.5874176025390625, 23.805130004882812, 30.1478271484375, -8.179180145263672, 53.41552734375, 15.803504943847656, 63.64404296875, 19.638458251953125, 30.0821533203125, 62.5704345703125, -2.118593215942383, 18.583297729492188, 48.5521354675293, -3.3744277954101562, 0.18608474731445312, 0.0, 13.989761352539062, 77.024169921875, 17.533798217773438, 66.51351928710938, 36.1446533203125, 17.386566162109375, 8.1270751953125, 30.665374755859375, 2.9781494140625, 68.7481689453125, 16.23614501953125, -1.61895751953125, 44.741424560546875, -6.89459228515625, 97.32659912109375, 19.335906982421875, 34.096405029296875, 83.19302368164062, -13.995849609375, 56.98529052734375, -3.885009765625, 37.6748161315918, 14.05511474609375, -44.681182861328125, 37.041542053222656, 51.38613510131836, 50.9844970703125, 50.172950744628906, 4.8678131103515625, -22.4468994140625, -22.938499450683594, 0.76470947265625, -0.463958740234375, 46.261199951171875, -0.6107635498046875, 58.230743408203125, 49.099456787109375, 97.49786376953125, 24.265045166015625, 77.06787109375, 5.877288818359375, -69.08917236328125, 45.66949462890625, 67.29362487792969, 0.7278060913085938, 2.758636474609375, 51.3385009765625, 6.40777587890625, -3.204681396484375, -5.271759033203125, 8.127777099609375, 35.98011779785156, 21.04620361328125, 64.65957641601562, 56.6561279296875, 67.81849670410156, 41.362762451171875, 80.9749755859375, 53.66131591796875, -21.16196060180664, -10.036956787109375, 61.22381591796875, -13.014007568359375, 7.462646484375, 43.130035400390625, -10.446990966796875, 20.049072265625, 31.683731079101562, -58.37599182128906, 16.02337646484375, 9.3671875, 54.631980895996094, 36.32745361328125, 27.367446899414062, 79.1510009765625, -4.337989807128906, -3.3069915771484375, 5.205085754394531, -1.7487716674804688, 7.0411376953125, 1.659820556640625, -29.0635986328125, -39.2808837890625, 56.61846160888672, 24.118133544921875, 49.916748046875, 71.00274658203125, 65.96920776367188, 46.78712463378906, 0.590423583984375, 31.72686767578125, 13.6348876953125, -8.5103759765625, -38.995697021484375, 11.477981567382812, 67.59555053710938, -66.62002563476562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000168.npy"}
{"epoch": 0.3518324607329843, "step": 169, "batch_size": 128, "mean": 27.496299743652344, "std": 34.647911071777344, "min": -58.72755432128906, "p10": -8.42832336425781, "median": 19.501174926757812, "p90": 76.09975891113281, "max": 111.92889404296875, "pos_frac": 0.7890625, "sample": [-2.016204833984375, 47.80140686035156, 88.51565551757812, 37.04461669921875, -9.616790771484375, 12.22805404663086, 1.788949966430664, 6.9830169677734375, 2.66021728515625, 27.427932739257812, 24.978103637695312, 58.04840087890625, 58.34781265258789, 56.61981201171875, -31.714111328125, -1.3816452026367188, 54.889007568359375, 8.9993896484375, 5.429172515869141, -3.251220703125, -3.31982421875, 9.318313598632812, 52.56779479980469, 15.319618225097656, 53.893585205078125, 66.58517456054688, 29.379852294921875, -15.69830322265625, 41.1566162109375, 48.19110107421875, 63.61083984375, 67.957275390625, 57.50347900390625, -7.8891448974609375, 4.038414001464844, 9.238899230957031, -14.736104965209961, -58.72755432128906, 27.838912963867188, -1.220703125, -9.633094787597656, 68.60995483398438, -11.814476013183594, 14.657470703125, 0.0, -47.208717346191406, 8.533998489379883, 29.98358154296875, 20.293670654296875, 16.447891235351562, -0.00341796875, 9.730804443359375, 111.92889404296875, 65.86276245117188, 39.87671661376953, 54.96484375, -50.69971466064453, 54.67625427246094, 0.7907314300537109, -1.7047653198242188, 76.36279296875, 50.3634033203125, 28.532135009765625, -9.3511962890625, 41.26629638671875, 77.74154663085938, 63.42601013183594, 89.13973999023438, 18.70867919921875, 63.72209167480469, 0.8543243408203125, 90.2918701171875, 8.313270568847656, 2.828521728515625, 41.22039794921875, -36.44081115722656, 7.9375, 78.69854736328125, 44.356597900390625, -6.0479736328125, 53.90217590332031, 9.157295227050781, 9.559860229492188, 51.052459716796875, 21.84613037109375, 56.371124267578125, 67.07720947265625, -8.032806396484375, 41.950775146484375, 37.601715087890625, 0.14003753662109375, 37.43170166015625, 6.51177978515625, 103.26969909667969, 32.533782958984375, 10.13406753540039, 14.665924072265625, 6.387939453125, 85.17491149902344, 44.71595764160156, 7.363670349121094, 52.64979553222656, 94.15206909179688, 17.2628173828125, -32.7347412109375, -30.275466918945312, 9.320388793945312, -1.5367889404296875, 96.048583984375, 21.384796142578125, 49.681976318359375, 0.03897857666015625, 67.91937255859375, 53.47454833984375, 55.932220458984375, 0.0, 14.13531494140625, 3.282989501953125, 18.34210205078125, 44.39636993408203, 20.7769775390625, 0.4022331237792969, 13.965530395507812, 16.526287078857422, 89.19512939453125, 89.78839111328125, 75.98703002929688, -5.412628173828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000169.npy"}
{"epoch": 0.3539267015706806, "step": 170, "batch_size": 128, "mean": 24.098995208740234, "std": 32.746490478515625, "min": -58.064239501953125, "p10": -8.663731384277343, "median": 20.972915649414062, "p90": 67.43450317382812, "max": 115.12213134765625, "pos_frac": 0.78125, "sample": [4.791015625, 15.0137939453125, 53.31794738769531, -8.585540771484375, 51.227996826171875, 9.12677001953125, 0.9375991821289062, 108.296142578125, 16.347862243652344, 9.702545166015625, 30.100494384765625, 91.1527099609375, 43.058135986328125, 25.420066833496094, -39.955810546875, 104.74932861328125, -29.01519775390625, 32.00898742675781, 0.6500320434570312, 15.962846755981445, -34.06866455078125, 7.23028564453125, 38.456077575683594, 43.93975830078125, 4.196464538574219, -25.40932846069336, -17.131515502929688, 3.3876867294311523, 70.4976806640625, -7.22503662109375, 30.14776611328125, 18.90089988708496, 38.39866638183594, 8.330392837524414, 16.743919372558594, -2.156646728515625, 1.62261962890625, 5.2498931884765625, 62.824188232421875, 55.945220947265625, 2.2248077392578125, 37.47532653808594, 5.864776611328125, 65.9561767578125, 52.467323303222656, 38.26930236816406, 30.75494384765625, -2.36376953125, -22.794586181640625, 68.84014892578125, 67.35585021972656, 54.59674835205078, 54.725341796875, -47.42747497558594, 21.402801513671875, -8.14739990234375, 71.74569702148438, -11.590652465820312, 67.61802673339844, 43.407493591308594, 23.8099365234375, 75.731689453125, 58.61088562011719, -6.7490234375, 44.90684127807617, 42.644554138183594, -4.903106689453125, 14.736572265625, 61.31596374511719, 2.668670654296875, 32.19468688964844, 56.260589599609375, 7.418590545654297, 20.707122802734375, 47.198814392089844, 37.30908203125, 0.0, 70.68389892578125, 18.633628845214844, 49.50822448730469, -30.802276611328125, 48.40667724609375, 4.32586669921875, 46.159027099609375, 58.5606689453125, 34.060699462890625, -34.18597412109375, 11.4134521484375, 42.4078369140625, -58.064239501953125, 28.411376953125, 25.693939208984375, -5.656982421875, 35.97003173828125, 1.4773788452148438, -2.620664596557617, 69.49533081054688, -8.846176147460938, 30.60321044921875, 115.12213134765625, 2.035675048828125, -8.551895141601562, 34.958274841308594, -3.48870849609375, -6.274749755859375, 13.694976806640625, 25.904434204101562, 11.589637756347656, 1.7502670288085938, 12.735992431640625, -3.1007003784179688, 4.800750732421875, 88.140625, 21.23870849609375, -22.556922912597656, 22.334823608398438, 8.005935668945312, 64.37445068359375, 23.10845184326172, 4.436910629272461, 1.809814453125, 38.98760986328125, 74.73472595214844, 61.35993194580078, 5.711585998535156, 32.810935974121094, -2.316253662109375, 67.24838256835938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000170.npy"}
{"epoch": 0.35602094240837695, "step": 171, "batch_size": 128, "mean": 24.39261245727539, "std": 33.49990463256836, "min": -37.159751892089844, "p10": -17.807541656494138, "median": 20.66900634765625, "p90": 74.25401611328124, "max": 94.99212646484375, "pos_frac": 0.75, "sample": [89.02670288085938, 17.6402587890625, 26.90826416015625, 12.92437744140625, 60.57696533203125, 67.421630859375, 62.987388610839844, -33.367462158203125, -5.552595138549805, 21.573532104492188, -22.782867431640625, 0.1943817138671875, 77.46694946289062, -18.54901123046875, 2.8786773681640625, 2.1152076721191406, -13.664680480957031, 17.530227661132812, 29.678009033203125, 78.63775634765625, 55.21009826660156, -6.388580322265625, 19.465179443359375, 82.45864868164062, 66.57565307617188, 30.058349609375, -24.119049072265625, 11.451156616210938, -24.50469970703125, 21.12896728515625, 13.356124877929688, 11.462890625, -11.844085693359375, 56.54835510253906, 55.97686767578125, 76.15203857421875, 55.94523620605469, 2.150360107421875, 24.260387420654297, -6.151344299316406, 17.02301025390625, -13.009735107421875, -6.707576751708984, -6.4073638916015625, -17.489768981933594, 90.09527587890625, 83.73773193359375, 10.766427993774414, -36.842071533203125, 5.3512115478515625, 69.03280639648438, 28.0301513671875, 9.876129150390625, 65.73892211914062, 30.30999755859375, 31.833053588867188, 40.83795166015625, 75.99658203125, 78.89990234375, -13.966705322265625, 50.282318115234375, 27.20288848876953, -10.610076904296875, 57.638389587402344, 37.7831916809082, 61.80250549316406, 30.46832275390625, 9.803041458129883, 56.788421630859375, 44.63172149658203, 23.605117797851562, 78.94180297851562, -6.025238037109375, 4.10821533203125, -32.63622283935547, -16.66107177734375, 76.31672668457031, 2.5497169494628906, -28.4044189453125, 13.474380493164062, -0.055164337158203125, 2.4251861572265625, -9.5933837890625, -25.473495483398438, 17.902984619140625, -28.65386962890625, 53.53591537475586, 44.9129638671875, -1.282379150390625, 24.815399169921875, 49.9235954284668, 94.98396301269531, -37.159751892089844, 13.549598693847656, 32.634674072265625, 60.28045654296875, 55.348175048828125, 2.667633056640625, 69.30624389648438, 24.8011474609375, 21.2537841796875, 47.059814453125, -24.247360229492188, 11.5411376953125, 25.50537109375, 15.0496826171875, 21.49573516845703, 62.437469482421875, 19.12230682373047, 1.1474838256835938, 20.20904541015625, -19.921524047851562, 59.95637512207031, 56.157562255859375, 52.1356201171875, 15.82843017578125, 24.387451171875, 5.3514404296875, 22.62982177734375, 10.471572875976562, 94.99212646484375, 50.086021423339844, -10.656929016113281, 1.0635738372802734, -4.479911804199219, 73.5072021484375, 27.188262939453125, -8.888824462890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000171.npy"}
{"epoch": 0.3581151832460733, "step": 172, "batch_size": 128, "mean": 16.40755271911621, "std": 35.160091400146484, "min": -88.99871826171875, "p10": -17.522356796264646, "median": 14.612678527832031, "p90": 64.05447692871093, "max": 88.00146484375, "pos_frac": 0.6796875, "sample": [16.15576171875, -12.238616943359375, 3.0993499755859375, 23.30487060546875, 34.46095275878906, 13.149635314941406, -2.29327392578125, -9.790946960449219, 5.983375549316406, -0.347747802734375, -2.6413726806640625, -1.1477527618408203, -62.1158447265625, 5.399430274963379, 52.62034606933594, 12.45945930480957, 34.953216552734375, 74.0758056640625, 38.595367431640625, 10.956840515136719, -7.726104736328125, 9.78070068359375, 24.437774658203125, 82.13970947265625, 5.78948974609375, 43.9649658203125, 38.33952331542969, -6.2301025390625, -13.319801330566406, 60.699005126953125, 4.2220458984375, 27.833465576171875, -69.65545654296875, 8.03692626953125, 19.96142578125, 51.09564208984375, -14.641448974609375, 74.94577026367188, -21.726821899414062, -5.606285095214844, -4.116472244262695, 15.947799682617188, -2.25909423828125, -0.9810752868652344, 76.72755432128906, -88.99871826171875, -5.8848876953125, 37.65740966796875, 76.64581298828125, 58.135833740234375, -4.0696258544921875, -27.839736938476562, 22.54827880859375, -1.344146728515625, 32.896636962890625, 16.329978942871094, 61.4686279296875, 67.37110900878906, -17.04840087890625, -4.9081878662109375, -4.4549560546875, 19.48626708984375, 23.368362426757812, -1.748870849609375, 59.92388916015625, 43.88555908203125, -3.72998046875, 6.751617431640625, 20.759262084960938, 8.350135803222656, 85.5904541015625, -61.6077880859375, 2.3190078735351562, 41.729705810546875, 71.8507080078125, 59.403076171875, 0.9609375, 1.4054718017578125, 22.04052734375, -70.30389404296875, 28.15936279296875, 15.784675598144531, 19.290054321289062, -60.437744140625, 16.24555206298828, 47.754730224609375, 5.7003631591796875, 66.44677734375, 14.86920166015625, -46.2257080078125, 76.29541015625, -3.4816436767578125, 69.12640380859375, -37.21484375, -34.47505187988281, 23.217742919921875, 25.241485595703125, 18.376129150390625, -18.628253936767578, 14.356155395507812, -5.841278076171875, 88.00146484375, 55.996246337890625, 45.07144546508789, 26.78515625, 27.700149536132812, -8.933296203613281, 19.762359619140625, 37.89873123168945, 7.931671142578125, 12.5421142578125, 53.54188537597656, 38.10504913330078, 63.093292236328125, 58.610107421875, 38.408905029296875, -8.11865234375, 7.9285888671875, 7.567079544067383, 47.76495361328125, -76.5860595703125, 66.2972412109375, 0.49224853515625, 10.917221069335938, 0.0, 38.6458740234375, 35.142303466796875, -10.19635009765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000172.npy"}
{"epoch": 0.36020942408376966, "step": 173, "batch_size": 128, "mean": 20.424579620361328, "std": 33.83799362182617, "min": -68.89297485351562, "p10": -17.792593383789054, "median": 18.121685028076172, "p90": 63.021766662597656, "max": 120.7677001953125, "pos_frac": 0.7109375, "sample": [-2.15240478515625, -68.89297485351562, -47.73902893066406, 21.903564453125, 27.61688232421875, -13.779525756835938, 48.43939208984375, -39.915557861328125, 47.45111083984375, 111.00848388671875, -11.666763305664062, 27.648284912109375, -10.012426376342773, 39.58001708984375, 49.1932373046875, 62.37153625488281, 11.600067138671875, 4.911323547363281, 22.27166748046875, 78.77093505859375, 66.31636047363281, 14.430389404296875, 8.350128173828125, 37.300689697265625, 20.678512573242188, 13.60888671875, 0.1296234130859375, 7.743743896484375, 41.93023681640625, 52.38182067871094, -0.48956298828125, 47.162017822265625, -3.0903396606445312, 5.883029937744141, 6.1667938232421875, 33.803192138671875, -35.46539306640625, 26.495147705078125, 26.88809585571289, 40.860504150390625, -7.08148193359375, 26.55828857421875, 39.624481201171875, 43.38041687011719, 52.61956787109375, 34.9334716796875, -3.689483642578125, -8.484756469726562, 3.7729873657226562, 11.88189697265625, -13.915672302246094, -41.2218132019043, -26.25518798828125, 120.7677001953125, 13.338348388671875, -29.52532958984375, 7.603538513183594, -22.670013427734375, 32.13136291503906, 99.7587890625, 60.077728271484375, 65.54571533203125, 3.017498016357422, 55.82926940917969, 59.99774169921875, 63.5684814453125, -50.344970703125, 35.26336669921875, 22.392196655273438, 18.94073486328125, 24.938522338867188, 1.2282028198242188, 70.69247436523438, 48.11268615722656, 3.0572967529296875, 47.847900390625, 11.824714660644531, 69.147216796875, -3.2756614685058594, 17.302635192871094, 31.69183349609375, -29.1353759765625, 3.0840492248535156, 5.322784423828125, 55.84455871582031, 5.873565673828125, 37.5555419921875, 62.78746032714844, -4.7611083984375, 7.93402099609375, 19.19561767578125, -4.829010009765625, 53.94493103027344, 27.499465942382812, 67.63372802734375, 27.876739501953125, 64.0035400390625, 34.60443115234375, 43.202728271484375, 57.5982666015625, 71.61885070800781, 16.5074462890625, -3.0767669677734375, -12.969757080078125, -6.48914909362793, 59.974220275878906, -29.962677001953125, 53.59754943847656, 38.061981201171875, 20.70428466796875, -6.63189697265625, 3.205718994140625, -15.7022705078125, 9.548248291015625, 0.13218116760253906, -6.32342529296875, -3.8855133056640625, -35.006324768066406, 79.86273193359375, -5.055519104003906, 60.385040283203125, -10.941802978515625, 0.0, 16.066673278808594, 37.2398681640625, -37.126861572265625, -7.2548065185546875, 32.55767822265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000173.npy"}
{"epoch": 0.362303664921466, "step": 174, "batch_size": 128, "mean": 25.622657775878906, "std": 33.62141036987305, "min": -49.74836730957031, "p10": -15.27051773071289, "median": 20.360572814941406, "p90": 73.31336669921875, "max": 115.53659057617188, "pos_frac": 0.765625, "sample": [53.86473846435547, 1.9491004943847656, 6.347190856933594, 1.1717681884765625, 87.8916015625, 63.006866455078125, 74.89996337890625, -36.6826171875, 8.287689208984375, -1.10296630859375, 69.4840087890625, 52.581336975097656, 1.73687744140625, 50.782249450683594, 13.84063720703125, 30.363433837890625, 67.33584594726562, 100.41006469726562, 1.7090301513671875, 15.579910278320312, -8.178848266601562, 73.29315185546875, 0.17919921875, 83.36422729492188, 78.1978759765625, -5.42376708984375, 45.91612243652344, 8.465255737304688, 78.49505615234375, 51.10906982421875, -4.8719329833984375, 54.462188720703125, 115.53659057617188, -35.34039306640625, -22.166603088378906, -9.474456787109375, 60.833709716796875, 73.36053466796875, 89.25283813476562, 4.201568603515625, 16.68901252746582, 36.26713562011719, 38.816566467285156, 29.868896484375, -2.5255603790283203, 74.51028442382812, 49.6986083984375, 37.53192138671875, -3.384979248046875, 106.72755432128906, -10.98828125, -18.109405517578125, 59.38732147216797, 38.98431396484375, -13.898735046386719, 4.14154052734375, 13.46023178100586, -0.758758544921875, 24.898223876953125, 19.55230712890625, 20.770462036132812, 5.6876068115234375, 74.89990234375, 66.02597045898438, 28.2718505859375, -22.065597534179688, -4.773468017578125, 41.97662353515625, 11.614944458007812, 35.14582824707031, 22.234085083007812, 50.210723876953125, 61.364173889160156, 56.991668701171875, 46.99427032470703, 30.841232299804688, 14.16265869140625, 17.495819091796875, 61.002899169921875, -17.516021728515625, -38.74281311035156, 42.781829833984375, 1.5115203857421875, 42.2728271484375, 20.460952758789062, 1.582498550415039, 66.74755859375, 8.238149642944336, -21.54351806640625, 11.195503234863281, 9.083244323730469, 16.595413208007812, 34.68788146972656, 39.91734313964844, 43.912567138671875, 50.3046875, 2.4302978515625, 85.893310546875, -1.6460552215576172, 21.834205627441406, 56.79423522949219, -19.538330078125, 46.04278564453125, 21.847991943359375, 10.133056640625, 13.758255004882812, -15.991661071777344, 0.7335700988769531, -6.0302734375, -25.085357666015625, 45.23365783691406, -2.275787353515625, 3.2852859497070312, 36.55229949951172, 25.631317138671875, -16.5064697265625, 68.82229614257812, -14.961456298828125, 8.71337890625, -49.74836730957031, 7.3656005859375, 17.968505859375, 49.260040283203125, 20.26019287109375, 51.533477783203125, -0.0389556884765625, 22.2613525390625, -10.709548950195312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000174.npy"}
{"epoch": 0.3643979057591623, "step": 175, "batch_size": 128, "mean": 20.223865509033203, "std": 36.81006622314453, "min": -95.22630310058594, "p10": -23.421302795410153, "median": 18.928546905517578, "p90": 66.24710083007811, "max": 121.88134765625, "pos_frac": 0.75, "sample": [55.528076171875, 28.505859375, 39.972412109375, -2.391368865966797, 1.65087890625, -45.95960998535156, 23.765899658203125, 6.417808532714844, 91.9892578125, -58.809356689453125, 18.0987548828125, 7.779853820800781, 21.479705810546875, 26.95458221435547, 49.80661392211914, 121.88134765625, 5.432096481323242, 52.5244140625, 22.553573608398438, 73.5589599609375, -32.57623291015625, 58.03218078613281, 59.06056213378906, 46.490875244140625, 41.939727783203125, -10.504119873046875, -1.8860931396484375, -38.0057373046875, 18.970565795898438, -42.49188232421875, -3.8446807861328125, 0.4319915771484375, 21.32525634765625, 55.1702880859375, -5.0526580810546875, 56.805389404296875, 10.051910400390625, 3.8883819580078125, 77.54910278320312, 43.62548828125, 4.5086212158203125, 12.255096435546875, 15.00982666015625, 17.4786376953125, 10.02239990234375, 70.71047973632812, 53.950836181640625, 53.82098388671875, -15.23968505859375, 83.3516845703125, 13.954025268554688, 11.386405944824219, -0.7760009765625, 10.540435791015625, -17.063323974609375, -40.6512451171875, 18.2103271484375, -4.739814758300781, -0.6627883911132812, 71.67550659179688, 24.715362548828125, 70.41061401367188, 9.030227661132812, 40.096702575683594, 52.32501220703125, 8.488662719726562, 2.84442138671875, 4.214366912841797, -0.872711181640625, 6.99920654296875, 59.77001953125, -2.4783935546875, 22.011474609375, 58.0086669921875, 50.23723602294922, -29.48834228515625, -0.595733642578125, 64.46273803710938, 0.0, -95.22630310058594, 18.88652801513672, 42.92919921875, 78.99606323242188, 4.79693603515625, 20.822509765625, -2.62353515625, 25.169296264648438, -17.149253845214844, -35.8648681640625, 84.6307373046875, 42.921661376953125, 4.955169677734375, 51.600982666015625, 58.24945068359375, 47.77491760253906, 30.502716064453125, -22.7550048828125, 10.974319458007812, -24.975997924804688, 8.023101806640625, 51.52606201171875, 55.906394958496094, 9.861183166503906, 36.7197265625, 4.57220458984375, -74.55148315429688, 19.4241943359375, 28.015472412109375, -8.13409423828125, 24.255126953125, 34.277645111083984, 19.813308715820312, 26.605606079101562, 31.78937530517578, 77.87350463867188, 7.1820068359375, 42.30437469482422, 86.5455322265625, 55.4576416015625, 1.58740234375, 19.08929443359375, 24.856414794921875, 22.06683349609375, -71.94351196289062, -71.05496215820312, 0.0, 5.006984710693359, 89.3218994140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000175.npy"}
{"epoch": 0.36649214659685864, "step": 176, "batch_size": 128, "mean": 16.187786102294922, "std": 38.47663879394531, "min": -115.63211059570312, "p10": -29.65601577758789, "median": 13.168342590332031, "p90": 63.053153228759754, "max": 107.24154663085938, "pos_frac": 0.7265625, "sample": [-29.4462890625, 13.481597900390625, 2.564300537109375, -35.379730224609375, 21.52490234375, 53.56489562988281, -50.25929260253906, -54.235626220703125, 9.265632629394531, 17.8665771484375, -83.005615234375, 6.173004150390625, 24.568565368652344, 79.74862670898438, 10.790328979492188, 53.9349365234375, 46.97554016113281, 5.245639801025391, 2.09857177734375, -9.108428955078125, -42.827362060546875, 7.6500244140625, 11.223001480102539, -69.94598388671875, 26.360580444335938, 72.72235107421875, 30.99224090576172, 99.26303100585938, 47.674041748046875, 1.92791748046875, 95.34719848632812, -31.551605224609375, -4.166717529296875, 18.908416748046875, 14.373661041259766, -2.371429443359375, -8.9093017578125, 23.784744262695312, 15.624420166015625, 11.713897705078125, -50.06031036376953, 13.002639770507812, 48.108184814453125, -3.738555908203125, 7.017486572265625, 61.681846618652344, 6.137458801269531, 3.2299346923828125, 28.625381469726562, 79.65374755859375, -30.14537811279297, 53.38530731201172, 46.86640930175781, -21.085693359375, 13.33404541015625, 52.392333984375, 2.0769500732421875, 47.127830505371094, 30.676361083984375, 5.053596496582031, 3.6043701171875, 1.4105072021484375, -6.012176513671875, -18.333770751953125, 46.65007019042969, 60.223876953125, 35.11724853515625, -14.2518310546875, 70.9024658203125, 2.59600830078125, -20.207992553710938, 77.08905029296875, 21.478302001953125, 24.820953369140625, 107.24154663085938, 71.51083374023438, 1.5777454376220703, -15.58099365234375, 18.311508178710938, 19.69122314453125, 22.779815673828125, 44.75581359863281, 54.02939224243164, 45.356834411621094, 43.865089416503906, 56.987022399902344, -3.8346710205078125, 20.085041046142578, 2.790313720703125, 84.4908447265625, 22.105361938476562, -2.1886749267578125, -31.37542724609375, 28.966064453125, -16.38385009765625, 5.727092742919922, 44.528594970703125, -70.91680908203125, -25.46441650390625, 49.988739013671875, -24.537673950195312, 53.68653869628906, 3.6997222900390625, -10.6715087890625, 7.290794372558594, -68.16586303710938, 30.608978271484375, 15.106414794921875, -17.084152221679688, 4.930324554443359, 55.76763916015625, -7.426719665527344, 83.04025268554688, 66.25286865234375, 86.407470703125, 34.34904479980469, 5.44140625, 29.05406951904297, -7.52752685546875, -115.63211059570312, 61.412384033203125, 6.47357177734375, 0.27575111389160156, 31.580223083496094, 38.86798095703125, 8.813751220703125, 41.65772247314453, -3.2665786743164062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000176.npy"}
{"epoch": 0.36858638743455496, "step": 177, "batch_size": 128, "mean": 14.613849639892578, "std": 42.154457092285156, "min": -109.61981201171875, "p10": -38.27832031249999, "median": 10.320154190063477, "p90": 71.30319061279297, "max": 104.8056640625, "pos_frac": 0.671875, "sample": [51.72999572753906, -0.9025726318359375, 2.02423095703125, 84.07572937011719, -26.96612548828125, 71.78128051757812, 10.352985382080078, -24.20501708984375, 45.452484130859375, 6.999763488769531, 68.95468139648438, -46.27314758300781, -94.45428466796875, -15.46499252319336, 3.660186767578125, -37.17510986328125, 45.39720153808594, 2.424285888671875, 44.15867614746094, 13.1920166015625, 46.62969970703125, 23.325119018554688, 59.066619873046875, 45.74676513671875, -31.2633056640625, 26.53173828125, -22.593658447265625, 88.47811889648438, 52.84991455078125, 97.14324951171875, 5.8109893798828125, 27.017555236816406, -75.2144775390625, 91.79119873046875, 7.715057373046875, 29.459182739257812, 73.16506958007812, 2.742259979248047, -9.300411224365234, 7.23681640625, 53.783966064453125, -23.975059509277344, -3.5496654510498047, 3.1598739624023438, -10.07635498046875, -35.276039123535156, 4.674655914306641, 14.758804321289062, 50.704139709472656, 52.41798400878906, -11.25396728515625, -3.416748046875, 92.13406372070312, 25.565475463867188, -3.5679168701171875, 63.54344177246094, 25.8712158203125, -6.007476806640625, 79.52421569824219, 5.2122802734375, -40.85247802734375, 79.81402587890625, 0.8261260986328125, -42.95831298828125, 76.60708618164062, -3.79229736328125, -46.68408203125, -9.545125961303711, 5.712406158447266, 66.82473754882812, -15.050811767578125, 99.123291015625, -4.5563507080078125, 7.673847198486328, -5.608070373535156, 50.83390808105469, 49.667694091796875, -88.79011535644531, 24.140289306640625, 10.287322998046875, 21.94537353515625, -62.49163818359375, -46.77904510498047, 35.645294189453125, -71.80941772460938, 11.376220703125, 35.18872833251953, 0.3780517578125, -22.72955322265625, 12.27294921875, 18.637969970703125, 69.8251953125, 32.446903228759766, 40.42677307128906, -1.38494873046875, -16.970184326171875, 13.174057006835938, 9.850341796875, 58.453277587890625, 2.0361175537109375, -24.548751831054688, -6.2356414794921875, 14.8323974609375, 44.022247314453125, 9.224246978759766, -80.14254760742188, 7.6108856201171875, 0.5739288330078125, -55.42291259765625, -109.61981201171875, 26.98773956298828, 48.12916564941406, 71.15486145019531, 25.63299560546875, 17.770233154296875, 71.6492919921875, 38.2513427734375, 38.10198974609375, 104.8056640625, 11.913803100585938, 26.684417724609375, -0.7078475952148438, -2.546630859375, -6.044624328613281, 18.674850463867188, 4.7610015869140625, 34.95831298828125, 61.61005401611328], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000177.npy"}
{"epoch": 0.3706806282722513, "step": 178, "batch_size": 128, "mean": 25.139781951904297, "std": 39.541229248046875, "min": -96.08880615234375, "p10": -17.906863403320312, "median": 19.982032775878906, "p90": 78.16408081054686, "max": 144.70199584960938, "pos_frac": 0.7421875, "sample": [62.487030029296875, 18.396949768066406, 5.373863220214844, -9.56195068359375, -6.5535125732421875, 7.62965202331543, 42.66314697265625, 65.29772186279297, 34.67113494873047, 39.75471878051758, 144.70199584960938, 61.864349365234375, 13.570953369140625, 38.40069580078125, -15.855072021484375, 87.566650390625, 0.0, 60.79148864746094, 58.62371826171875, 28.623680114746094, 14.260833740234375, 92.04531860351562, 7.502710342407227, 76.1712646484375, 53.240325927734375, -13.510520935058594, -8.449600219726562, 98.32955932617188, -23.275421142578125, 66.14785766601562, 5.7693328857421875, -6.042198181152344, 3.1068115234375, 32.96051025390625, -65.47001647949219, 18.1986083984375, 42.03131103515625, 70.01661682128906, -30.39111328125, 56.468727111816406, 79.3018798828125, 19.926589965820312, 57.18218994140625, 53.524322509765625, 77.67645263671875, 59.0709228515625, 12.946548461914062, -48.23765563964844, -11.549102783203125, 90.81660461425781, 61.26507568359375, 90.65487670898438, -58.15045166015625, -96.08880615234375, -7.0069580078125, 10.965705871582031, 18.9029541015625, 35.87812042236328, -34.89527893066406, 23.590179443359375, 0.44559478759765625, 100.21188354492188, 11.65118408203125, 44.683685302734375, 63.885498046875, 22.838729858398438, 27.830535888671875, 33.489044189453125, 34.295013427734375, 2.04852294921875, 94.13876342773438, 27.73943328857422, -13.494766235351562, -1.1805419921875, 12.696252822875977, 83.80215454101562, -18.491912841796875, 12.543609619140625, -10.976280212402344, 2.96844482421875, 40.382537841796875, 15.359100341796875, 56.9234619140625, -28.8621826171875, -32.146820068359375, -5.8540802001953125, 74.1527099609375, -6.08392333984375, 35.80751037597656, 32.05517578125, 15.023551940917969, 4.834478378295898, 0.2171630859375, 20.0374755859375, 30.942138671875, 98.49090576171875, 19.4901123046875, 9.502708435058594, 0.7899703979492188, 68.99072265625, 9.202949523925781, -13.651617050170898, -26.52777099609375, 95.58477783203125, 21.474281311035156, -17.6561279296875, 3.1724929809570312, 65.4423828125, 68.98866271972656, -15.0902099609375, -4.520053863525391, 97.2989501953125, -39.94519805908203, 0.518585205078125, 60.63800048828125, 20.04949951171875, -18.60654067993164, 27.234603881835938, 47.37660217285156, 4.3878173828125, 16.1363525390625, 31.592620849609375, 25.142242431640625, -5.84454345703125, -6.6467132568359375, 31.28955078125, 54.95869064331055, 41.382049560546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000178.npy"}
{"epoch": 0.37277486910994767, "step": 179, "batch_size": 128, "mean": 19.80405044555664, "std": 37.558780670166016, "min": -63.32177734375, "p10": -32.855204772949214, "median": 15.910736083984375, "p90": 71.2334587097168, "max": 100.880859375, "pos_frac": 0.7109375, "sample": [45.65269470214844, 4.000898361206055, 60.17999267578125, 10.8941650390625, 15.385467529296875, 67.45880126953125, 9.715156555175781, 10.629371643066406, -59.902069091796875, 1.51678466796875, 18.615570068359375, -2.2332763671875, 23.941864013671875, 1.550323486328125, 73.51199340820312, -15.40789794921875, 46.2066650390625, 27.741943359375, 6.844207763671875, -4.837898254394531, 25.5484619140625, 43.77252197265625, 2.542266845703125, 50.91131591796875, 31.36785888671875, -50.856685638427734, -41.54652404785156, -7.358057022094727, -15.786827087402344, -52.671112060546875, 52.178123474121094, 16.05176544189453, -46.94929504394531, 24.92376708984375, -20.935821533203125, 47.06486511230469, 25.648834228515625, 76.64756774902344, 22.764862060546875, -6.2942352294921875, -7.906707763671875, -38.50689697265625, -2.208984375, 36.306640625, -20.232757568359375, 57.61537170410156, 5.7139892578125, -10.254125595092773, 68.6759033203125, -40.8843994140625, -30.976364135742188, -30.5638427734375, 39.412559509277344, 16.380599975585938, 2.834453582763672, 80.07666015625, 72.1921615600586, 64.00286865234375, 8.058837890625, -3.6414794921875, -63.32177734375, -12.475265502929688, 92.7481689453125, 11.573699951171875, -17.619766235351562, 37.09710693359375, 70.82258605957031, -8.678298950195312, 0.2453460693359375, 15.105926513671875, 26.239246368408203, 43.9453125, 57.12908935546875, 35.466461181640625, 1.048858642578125, 10.561498641967773, 81.66802978515625, 11.88671875, 13.979194641113281, 57.169105529785156, 14.581645965576172, 9.378158569335938, 93.39935302734375, -17.709121704101562, 14.142295837402344, -43.4781494140625, 82.20886993408203, 66.5029296875, 1.2409172058105469, -38.12495422363281, 51.245506286621094, 20.40387725830078, 78.355712890625, 45.56097412109375, 34.66990661621094, 21.106842041015625, 47.25883483886719, 37.96580505371094, 15.769706726074219, 46.80987548828125, 86.58505249023438, 20.27996826171875, 66.59237670898438, 22.40137481689453, -39.9505615234375, 55.251495361328125, -54.08451843261719, 4.5782470703125, 45.36213684082031, 64.6191635131836, 24.007652282714844, 43.03680419921875, -37.239166259765625, 100.880859375, -15.08941650390625, -9.8123779296875, 44.874324798583984, 50.73860168457031, 0.19036865234375, -9.648681640625, 81.49853515625, 52.16212463378906, -3.0883636474609375, 13.59243392944336, -14.4656982421875, -19.33544921875, 75.58204650878906, 54.93598937988281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000179.npy"}
{"epoch": 0.374869109947644, "step": 180, "batch_size": 128, "mean": 24.731735229492188, "std": 36.36339569091797, "min": -63.056549072265625, "p10": -16.031838226318357, "median": 18.65045738220215, "p90": 77.13649597167968, "max": 106.59835815429688, "pos_frac": 0.7421875, "sample": [-1.1408348083496094, -4.2112579345703125, -32.808013916015625, -7.9035186767578125, 1.3644256591796875, 15.567108154296875, 67.14889526367188, 10.065673828125, -30.32086181640625, 68.80674743652344, -3.33502197265625, 54.34649658203125, 102.65802001953125, -58.70747375488281, 8.17529296875, 67.51425170898438, 14.115570068359375, 40.30328369140625, 8.176078796386719, -63.056549072265625, 59.521881103515625, 106.59835815429688, 36.40088653564453, 37.393310546875, 26.332427978515625, 51.39756774902344, 17.537261962890625, 66.84370422363281, 97.40484619140625, -8.370964050292969, -30.413894653320312, 2.474334716796875, 76.53547668457031, 103.10076904296875, 76.78970336914062, 30.158798217773438, 15.77081298828125, 18.58367919921875, 7.108406066894531, 23.823165893554688, 68.4171142578125, -12.989875793457031, 47.13640594482422, 35.31078338623047, 89.9791259765625, 23.6920166015625, -32.774658203125, -2.2986221313476562, 20.289794921875, 1.4746246337890625, -36.54443359375, -5.5504608154296875, 3.048004150390625, 31.143798828125, 83.98992919921875, 3.95489501953125, 18.717235565185547, 54.95330810546875, 14.235015869140625, 8.599273681640625, 80.98272705078125, 42.066261291503906, -2.30889892578125, 94.04425048828125, 50.50342559814453, -14.741119384765625, 43.837890625, 34.469635009765625, -42.541168212890625, 37.412200927734375, 26.742431640625, 3.6779556274414062, 38.45948028564453, 91.97744750976562, 73.52243041992188, 72.71722412109375, -0.031280517578125, 0.33778953552246094, 77.9456787109375, 42.06343078613281, 100.91168212890625, 10.113975524902344, -13.461334228515625, 41.44207763671875, 56.759212493896484, 48.08351135253906, 41.67079162597656, -16.83130645751953, 3.6644287109375, 21.87030029296875, 0.581146240234375, 14.596038818359375, 7.6186370849609375, 32.968292236328125, 54.200836181640625, 15.764724731445312, 27.72882080078125, 30.85174560546875, -0.7724838256835938, -2.0446319580078125, 4.7370452880859375, 29.836578369140625, 68.21307373046875, -1.823974609375, 4.043264389038086, -21.398448944091797, 5.2543182373046875, -21.750335693359375, 78.44158935546875, 11.135780334472656, -15.689208984375, 34.19004821777344, -6.597011566162109, -8.47760009765625, 52.41260528564453, 30.733795166015625, 31.470291137695312, -1.184661865234375, -20.120880126953125, 55.755828857421875, 0.14389991760253906, 44.616668701171875, 2.1624069213867188, 4.228359222412109, -18.510902404785156, 89.95010375976562, -12.074325561523438, 32.5775146484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000180.npy"}
{"epoch": 0.3769633507853403, "step": 181, "batch_size": 128, "mean": 23.150653839111328, "std": 41.12436294555664, "min": -81.418212890625, "p10": -29.289382934570312, "median": 24.401748657226562, "p90": 75.30729980468749, "max": 131.09881591796875, "pos_frac": 0.7265625, "sample": [0.8822288513183594, 7.838218688964844, 2.1905899047851562, 46.03424072265625, 0.0, 27.31446075439453, 80.38259887695312, 50.1781005859375, 9.916084289550781, 28.08279037475586, 63.60089111328125, -11.42087173461914, 52.45649719238281, -7.7764434814453125, 89.84930419921875, 43.035797119140625, 0.0, 73.85047912597656, 27.536300659179688, -8.640243530273438, 37.2655029296875, 1.642364501953125, 78.66314697265625, 32.76806640625, -72.04312133789062, 94.16105651855469, -44.0230712890625, 39.61812973022461, 8.492622375488281, 30.397186279296875, 61.6502799987793, 2.6219100952148438, 14.416923522949219, 63.596343994140625, 26.89825439453125, 48.940155029296875, 34.977203369140625, 21.033111572265625, 24.67529296875, 40.58934020996094, 17.568328857421875, 65.33929443359375, -23.669647216796875, -5.865396499633789, 13.9241943359375, 14.775398254394531, 9.934112548828125, 12.9725341796875, 95.62015533447266, -19.56983184814453, 33.70794677734375, -7.921714782714844, 54.808746337890625, 63.12237548828125, -51.48724365234375, -62.853485107421875, -4.380451202392578, 16.7213134765625, -2.746337890625, -11.90362548828125, 35.9764404296875, -11.505500793457031, -22.8193359375, 78.51937866210938, 48.10137939453125, 44.299503326416016, 7.29681396484375, 65.88783264160156, 19.03082275390625, 59.98297119140625, 17.325775146484375, 24.646400451660156, 11.980606079101562, 86.25233459472656, 113.80712890625, 131.09881591796875, -81.418212890625, 13.366409301757812, -57.814727783203125, 68.69955444335938, -34.452239990234375, 30.0751953125, -6.689481735229492, -55.25840759277344, -18.222579956054688, 60.83323669433594, 45.889892578125, -11.674049377441406, 20.7303466796875, 34.9637451171875, -17.425506591796875, 29.60546875, 73.93069458007812, -51.628387451171875, 24.15709686279297, 55.126312255859375, 9.656875610351562, 46.46436309814453, 38.913818359375, 86.82546997070312, 80.244873046875, -10.051910400390625, 44.77244567871094, 13.700225830078125, -28.798065185546875, -13.368827819824219, 9.749557495117188, 89.746337890625, -16.7069091796875, -79.15765380859375, 59.919708251953125, 34.171844482421875, 71.4991455078125, -31.6962890625, 50.5794677734375, 82.05494689941406, -30.435791015625, 11.977279663085938, 13.282081604003906, 4.418487548828125, 3.920867919921875, 29.214370727539062, 60.4071044921875, -47.0576171875, 62.51171875, 48.81829833984375, 44.91838073730469, 60.365081787109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000181.npy"}
{"epoch": 0.37905759162303665, "step": 182, "batch_size": 128, "mean": 29.074859619140625, "std": 34.7405891418457, "min": -61.657501220703125, "p10": -12.166083908081049, "median": 27.89138412475586, "p90": 72.23136138916016, "max": 110.06231689453125, "pos_frac": 0.7890625, "sample": [39.03558349609375, 62.91532897949219, 13.9775390625, -1.281412124633789, -4.7154388427734375, 38.065582275390625, 70.76335144042969, 58.149810791015625, 24.134445190429688, 6.7936553955078125, 61.460205078125, 33.3505859375, 27.102874755859375, 51.42523193359375, 58.590850830078125, 4.23895263671875, 35.45000457763672, 53.447998046875, 48.312347412109375, 47.08898162841797, 29.47449493408203, 17.217723846435547, 48.88683319091797, 82.88130187988281, -16.043243408203125, 73.52682495117188, -6.656822204589844, 51.099334716796875, 96.970947265625, 60.326385498046875, 52.876190185546875, 42.696319580078125, 55.67536926269531, 63.92852783203125, 41.09895706176758, -6.583965301513672, -10.069671630859375, -7.87054443359375, 1.415130615234375, 20.08428955078125, -57.21246337890625, 26.370254516601562, -61.657501220703125, 47.41071319580078, 34.29046630859375, -10.391754150390625, 69.12481689453125, -28.597488403320312, 64.69627380371094, 110.06231689453125, 46.396697998046875, 60.83819580078125, 67.9200439453125, 75.81552124023438, 80.45306396484375, -8.47210693359375, 25.86982536315918, 6.945091247558594, 7.1148681640625, 46.99835205078125, 38.4180908203125, 71.67616271972656, 22.370513916015625, 16.042098999023438, 63.988800048828125, 31.919822692871094, 28.679893493652344, 70.29635620117188, 98.7479248046875, 37.53898239135742, -27.606796264648438, 9.20549201965332, 24.57391357421875, 13.578475952148438, 0.020660400390625, -0.071929931640625, 23.934326171875, 47.76585388183594, 7.419807434082031, 13.613922119140625, 8.879486083984375, -3.057373046875, 0.6689300537109375, 55.79240417480469, -27.336883544921875, -1.5818557739257812, 12.377365112304688, 9.138389587402344, 11.60821533203125, -16.244155883789062, 80.1593017578125, 53.69091796875, 79.75772094726562, 74.14913940429688, -27.267318725585938, -10.504444122314453, 78.02127075195312, 29.011817932128906, 56.7625732421875, 23.091705322265625, -33.19139099121094, -2.43682861328125, 25.127464294433594, 19.388423919677734, 19.347137451171875, 49.35483932495117, 21.18243408203125, 32.58320617675781, 82.008056640625, 7.0609130859375, 0.751556396484375, 58.06756591796875, 70.42910766601562, 2.0269031524658203, 1.0974960327148438, 12.010467529296875, 37.04637908935547, 64.72599792480469, 6.6316986083984375, -19.94086456298828, -21.198822021484375, 67.594970703125, 69.00970458984375, -0.37054443359375, 74.76516723632812, -36.74024200439453, 58.26605224609375, -45.460693359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000182.npy"}
{"epoch": 0.381151832460733, "step": 183, "batch_size": 128, "mean": 24.128345489501953, "std": 41.345218658447266, "min": -77.30471801757812, "p10": -20.98746871948242, "median": 18.309768676757812, "p90": 76.87247619628906, "max": 110.37020874023438, "pos_frac": 0.6953125, "sample": [-24.399658203125, 73.1585693359375, -17.440780639648438, 54.23272705078125, -44.414703369140625, 75.74136352539062, -50.83604431152344, 11.5142822265625, 56.36305236816406, 4.0492401123046875, -18.716461181640625, 36.877708435058594, 67.755615234375, 0.9923095703125, 2.84716796875, -64.1812744140625, -4.223304748535156, 74.557373046875, 15.972442626953125, -12.332080841064453, -1.411834716796875, 31.09417724609375, 57.16807556152344, -16.962295532226562, 12.366302490234375, -14.943084716796875, 69.25576782226562, 37.41374206542969, 9.223526000976562, 1.8704681396484375, -5.216156005859375, -67.76168823242188, 17.285247802734375, 44.84735107421875, -0.8215179443359375, 39.864471435546875, 19.33428955078125, 69.02943420410156, 31.1771240234375, -8.458526611328125, 95.87738037109375, -9.599441528320312, 47.12562561035156, -23.4578857421875, 66.327880859375, 12.324638366699219, 60.123382568359375, -13.475204467773438, 82.2098388671875, 34.669979095458984, 1.7886962890625, 76.30960083007812, 59.78153991699219, -49.38494873046875, 0.0, -0.08003997802734375, -0.27044677734375, 7.9024505615234375, -19.92871856689453, 85.07958984375, 11.691520690917969, 63.68360900878906, 1.3209075927734375, 88.43505859375, 75.09146118164062, -40.4427490234375, 35.22563171386719, 67.81576538085938, 58.0089111328125, -16.794097900390625, 44.36023712158203, 94.06631469726562, 110.37020874023438, 44.329185485839844, -18.678848266601562, 42.98207092285156, -59.12359619140625, 25.449127197265625, -4.803009033203125, 28.6968994140625, 38.248870849609375, 31.843231201171875, 29.397491455078125, 8.716039657592773, -1.8018722534179688, 7.119659423828125, 68.9774169921875, 69.55111694335938, 12.59930419921875, 14.363983154296875, -13.967536926269531, 14.032623291015625, 6.9058380126953125, 3.963895797729492, -31.327728271484375, 62.41278076171875, 79.43907165527344, 84.72021484375, 76.36883544921875, -0.17137908935546875, -68.00958251953125, -0.572967529296875, 2.45001220703125, 16.12420654296875, 3.494312286376953, 57.66789245605469, 1.822906494140625, 69.32011413574219, -5.853515625, 91.04275512695312, 37.462158203125, 84.94390869140625, 78.04763793945312, 27.39349365234375, 49.24847412109375, -42.692588806152344, 20.37774658203125, 47.375823974609375, -77.30471801757812, 107.14218139648438, -18.600799560546875, 58.376075744628906, 71.31425476074219, 36.42425537109375, -7.441009521484375, 31.257293701171875, 48.031890869140625, 81.31329345703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000183.npy"}
{"epoch": 0.3832460732984293, "step": 184, "batch_size": 128, "mean": 36.48895263671875, "std": 44.64841842651367, "min": -83.43023681640625, "p10": -22.719698333740233, "median": 41.96681213378906, "p90": 89.82746276855467, "max": 116.27984619140625, "pos_frac": 0.8125, "sample": [64.04460144042969, 111.22140502929688, 81.713134765625, 4.95831298828125, 19.535415649414062, -15.840492248535156, 47.427581787109375, -7.621044158935547, 49.39324951171875, 31.68511962890625, 27.246231079101562, 6.55126953125, -55.9764404296875, 41.441864013671875, -34.273406982421875, 0.076873779296875, 100.9443359375, 78.47756958007812, 60.557777404785156, 55.42060089111328, 61.814361572265625, 88.43731689453125, -83.43023681640625, 87.59274291992188, -6.1539459228515625, 82.918212890625, 75.197265625, 9.422164916992188, -50.40681457519531, 98.3753662109375, 39.219207763671875, 94.8565673828125, 55.90118408203125, -51.488861083984375, 23.69195556640625, 84.0870361328125, 6.302703857421875, 85.69549560546875, 59.59051513671875, 102.84884643554688, 27.048309326171875, 95.7919921875, -44.32330322265625, 43.00770950317383, 28.0948486328125, 105.33151245117188, -75.96176147460938, 104.62615966796875, -11.1748046875, 43.5816650390625, 33.681148529052734, -22.644805908203125, 21.78662109375, 5.243804931640625, -58.608734130859375, 70.27142333984375, 67.4405517578125, 43.19134521484375, -4.69818115234375, 104.72109985351562, 8.683990478515625, 4.23084831237793, 73.49569702148438, -41.51580810546875, 74.80657958984375, 75.50382995605469, 71.26535034179688, 83.38731384277344, 24.3450927734375, 70.2742919921875, 10.260894775390625, 42.03582763671875, 41.938995361328125, 37.25273895263672, 41.99462890625, 38.18357849121094, 29.75201416015625, 44.4759521484375, 34.35487365722656, 86.02401733398438, 116.27984619140625, 79.97726440429688, 2.041900634765625, 76.69842529296875, 28.538536071777344, -46.98490905761719, 8.17376708984375, 80.22372436523438, 4.540863037109375, 10.627197265625, 47.28863525390625, 86.691650390625, 50.93455505371094, 45.42411804199219, 71.26519775390625, -7.483070373535156, -19.372587203979492, 21.741073608398438, 1.2792396545410156, 83.36529541015625, 56.17839813232422, 53.0865478515625, 49.730682373046875, 74.5784912109375, 1.0238838195800781, 3.2258033752441406, 70.08453369140625, 50.173248291015625, -22.894447326660156, 93.07113647460938, 28.160987854003906, -41.79736328125, 38.9266357421875, 13.117568969726562, 71.69085693359375, -8.50897216796875, 93.69560241699219, 12.39923095703125, -8.575637817382812, 68.74588012695312, 9.747406005859375, -29.94146728515625, 81.21762084960938, 19.943115234375, 106.77752685546875, 49.107391357421875, -15.536048889160156, 53.294029235839844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000184.npy"}
{"epoch": 0.38534031413612563, "step": 185, "batch_size": 128, "mean": 22.95440673828125, "std": 40.109371185302734, "min": -69.69601440429688, "p10": -25.843983459472657, "median": 18.904273986816406, "p90": 75.61446990966796, "max": 111.92724609375, "pos_frac": 0.703125, "sample": [17.691452026367188, -58.360015869140625, 50.0831298828125, -17.278717041015625, -36.09934997558594, 61.24407958984375, 27.85216522216797, 80.90785217285156, 89.30938720703125, -7.3279571533203125, 15.91046142578125, 44.52070617675781, 72.3597412109375, 56.08389663696289, 43.03101348876953, 94.35296630859375, 47.469879150390625, -15.65966796875, 9.738868713378906, 64.26824951171875, 40.44512939453125, 9.1185302734375, 49.787288665771484, 18.017227172851562, 30.250335693359375, 20.2420654296875, 82.5130615234375, 9.474311828613281, 1.3938789367675781, 85.30853271484375, -69.69601440429688, 56.24961853027344, 3.0650558471679688, -9.132293701171875, 22.195335388183594, 66.07733154296875, 27.2974853515625, 46.889244079589844, 36.97271728515625, 105.06692504882812, 7.381256103515625, 55.14923095703125, -4.5234375, 22.33642578125, 12.7066650390625, 81.84403991699219, -2.50146484375, 35.385414123535156, 33.27508544921875, 62.898193359375, -48.10533142089844, -25.591018676757812, -24.279510498046875, -15.80145263671875, 18.960617065429688, 0.42913055419921875, 13.725128173828125, 0.0, 20.583084106445312, 62.46173095703125, -26.643966674804688, -26.434234619140625, 11.7841796875, 1.47137451171875, 34.54315185546875, 18.847930908203125, 4.672798156738281, 16.5533447265625, 11.697509765625, -4.348056793212891, -28.854156494140625, 31.758880615234375, 53.23884582519531, 74.56764221191406, -16.08470916748047, 0.6331634521484375, -4.319183349609375, 16.11669921875, 24.315919876098633, 69.04585266113281, 111.37820434570312, 10.789932250976562, 11.159332275390625, 0.0, -24.225662231445312, 61.080955505371094, 72.1646728515625, -22.574493408203125, 64.00384521484375, -6.963409423828125, 49.69822692871094, 50.5283203125, -22.022811889648438, -5.702312469482422, 78.05706787109375, 57.934326171875, -49.531524658203125, 68.57315063476562, 15.541999816894531, 21.249755859375, -21.98932647705078, 44.19036865234375, 51.337127685546875, -37.4515380859375, 95.16226196289062, -55.79071044921875, -13.07025146484375, 13.942413330078125, 102.48086547851562, 111.92724609375, 61.2392578125, 35.55961608886719, 13.452346801757812, -3.64678955078125, 80.84616088867188, -64.61514282226562, 40.433837890625, 70.404052734375, 53.87425231933594, -40.54144287109375, 30.36853790283203, 9.033905029296875, -9.529769897460938, -21.359786987304688, -38.51654052734375, -22.01221466064453, 30.500713348388672, 40.2642822265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000185.npy"}
{"epoch": 0.387434554973822, "step": 186, "batch_size": 128, "mean": 22.61076545715332, "std": 38.05802917480469, "min": -79.09616088867188, "p10": -17.380426025390623, "median": 21.959808349609375, "p90": 73.13591003417969, "max": 129.9495849609375, "pos_frac": 0.6953125, "sample": [27.608535766601562, -2.6255950927734375, 10.961669921875, 51.931884765625, 19.57781982421875, 55.23529052734375, 27.7779541015625, -2.956512451171875, 15.174896240234375, 14.728267669677734, -2.8264694213867188, 6.663818359375, 27.151473999023438, 49.04710388183594, -8.95660400390625, 69.48138427734375, 60.24512481689453, 2.745546340942383, 72.35507202148438, 7.32489013671875, 10.371826171875, -11.728813171386719, -27.328201293945312, 3.7140960693359375, 1.4832229614257812, 54.04069519042969, 39.11375427246094, 27.398406982421875, 27.902618408203125, -0.456512451171875, 8.61782455444336, 35.66510009765625, 94.7410888671875, 33.03724670410156, -9.154190063476562, -51.728355407714844, 35.015716552734375, 90.64410400390625, 29.83929443359375, 31.67913818359375, 2.4119644165039062, 62.4444580078125, 60.990966796875, 51.77420425415039, 57.530704498291016, 26.405517578125, -50.48779296875, 3.6934452056884766, 52.85979461669922, 24.28424072265625, 30.52264404296875, 21.640472412109375, 75.76400756835938, -7.68707275390625, -20.601638793945312, -13.610702514648438, 75.473876953125, 26.91229248046875, -16.37640380859375, 9.04620361328125, -8.954090118408203, -8.018829345703125, 82.35223388671875, -36.81396484375, 72.47402954101562, 23.43682861328125, 39.33453369140625, -3.94622802734375, 18.51776123046875, 78.07826232910156, -6.606468200683594, 22.279144287109375, 97.91775512695312, 28.464248657226562, -13.385284423828125, 15.867889404296875, 61.17729949951172, -73.7900390625, -0.1464214324951172, 82.14617919921875, 56.84698486328125, 9.021575927734375, 33.9241943359375, 123.91473388671875, 62.89324951171875, -62.94676208496094, 24.231658935546875, -23.181228637695312, 6.01812744140625, -49.731201171875, -7.6571502685546875, 19.394134521484375, 41.583045959472656, -9.084999084472656, 96.4920654296875, 93.00588989257812, 6.419593811035156, 13.518783569335938, -79.09616088867188, 27.324993133544922, 42.8395881652832, 67.23622131347656, 10.170326232910156, -6.908050537109375, 38.4376220703125, 42.20855712890625, 24.74639892578125, -0.45843505859375, -0.374359130859375, 47.17158889770508, 52.21525573730469, -13.114028930664062, 33.612884521484375, 3.9834671020507812, 129.9495849609375, 0.0, 74.6802978515625, 59.19914627075195, -19.72314453125, 31.1356201171875, -20.68622589111328, -22.09510040283203, -1.5604400634765625, 2.3761520385742188, 31.171661376953125, -15.510261535644531, 62.0308837890625, -8.338623046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000186.npy"}
{"epoch": 0.38952879581151834, "step": 187, "batch_size": 128, "mean": 22.741168975830078, "std": 45.22842788696289, "min": -84.37188720703125, "p10": -37.728948974609374, "median": 17.51544189453125, "p90": 87.06201782226563, "max": 116.34857177734375, "pos_frac": 0.6875, "sample": [49.152587890625, 72.00447082519531, 28.597900390625, 39.97235107421875, -55.7965087890625, 8.55316162109375, 93.7703857421875, 0.8219928741455078, -39.774017333984375, 116.34857177734375, 79.83554077148438, 1.2221145629882812, -31.2156982421875, 18.493820190429688, 67.23921203613281, 114.591552734375, 51.00602722167969, 47.52904510498047, 87.01840209960938, 13.859283447265625, 68.18267822265625, -32.07185363769531, 41.032562255859375, -4.57012939453125, -11.396820068359375, -12.57916259765625, 69.00718688964844, -70.33547973632812, -46.896881103515625, -67.466064453125, -9.6534423828125, 77.5994873046875, 53.83056640625, -1.8624420166015625, -25.00421905517578, 35.12959289550781, -36.957763671875, -10.676727294921875, 38.64530944824219, 40.42181396484375, 11.54559326171875, -8.177978515625, 100.33636474609375, 19.175949096679688, -14.266845703125, 49.344696044921875, 88.67987060546875, 3.7110595703125, -25.030487060546875, 46.1632080078125, 1.6934814453125, 29.108177185058594, 26.442054748535156, 63.548431396484375, -84.37188720703125, -1.101837158203125, 55.015869140625, -36.5311279296875, 58.24177169799805, 91.68110656738281, 62.8199462890625, -8.142440795898438, 50.207183837890625, -42.337799072265625, 111.11451721191406, 71.84700012207031, -60.59710693359375, -54.512451171875, 12.775787353515625, 18.916259765625, 90.4202880859375, -47.30295944213867, -6.2674102783203125, -39.52838134765625, -9.541778564453125, -14.37725830078125, 71.31613159179688, 18.51018524169922, 79.25259399414062, 14.050628662109375, 2.838653564453125, 10.136871337890625, -50.063743591308594, 4.4991912841796875, 8.223648071289062, -3.6870346069335938, 21.652740478515625, 10.039688110351562, -62.57110595703125, 74.64263916015625, 56.469329833984375, 74.86575317382812, 72.01141357421875, 21.753517150878906, 27.82745361328125, 92.23356628417969, 0.9420928955078125, 59.83258056640625, 59.097412109375, 14.518424987792969, -32.5015869140625, 26.203460693359375, 103.333740234375, 8.1634521484375, 0.294647216796875, 94.15155029296875, -0.19146728515625, 24.917266845703125, -15.87255859375, 87.16378784179688, 55.067779541015625, 36.8133544921875, 6.3424530029296875, 16.537063598632812, 7.755393981933594, 77.50906372070312, 89.12171936035156, 0.7649307250976562, 5.062950134277344, 11.127334594726562, -10.544464111328125, 46.970489501953125, 34.132049560546875, -7.3416595458984375, 0.0, 78.46282958984375, -11.529937744140625, 32.25413513183594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000187.npy"}
{"epoch": 0.39162303664921466, "step": 188, "batch_size": 128, "mean": 23.515459060668945, "std": 44.63573455810547, "min": -94.06532287597656, "p10": -30.326617431640617, "median": 19.322467803955078, "p90": 84.80086059570313, "max": 107.3358154296875, "pos_frac": 0.7109375, "sample": [7.392827987670898, 47.061370849609375, 84.38748168945312, 0.0, 36.686119079589844, 22.379486083984375, 3.878326416015625, 89.1533203125, -12.328964233398438, 57.12543487548828, 43.15898132324219, 8.983856201171875, 98.17987060546875, -9.37045669555664, 27.15771484375, -0.4617462158203125, 97.38995361328125, 55.24761962890625, 14.603622436523438, 106.47259521484375, 64.404296875, 61.579742431640625, -19.164459228515625, 0.8399124145507812, -11.930557250976562, -22.19818115234375, -28.389892578125, 19.077651977539062, -26.169784545898438, 62.42997741699219, 76.31378173828125, -13.012588500976562, 87.70697021484375, 46.84607696533203, 1.4839935302734375, 21.196502685546875, 83.66616821289062, 54.533233642578125, 67.31356811523438, 8.921783447265625, 19.81475830078125, 36.454132080078125, 7.969879150390625, 35.812469482421875, 98.575439453125, -78.75848388671875, 34.63055419921875, 59.9288330078125, 60.432342529296875, 10.567977905273438, 12.63397216796875, 67.8314208984375, -21.5518798828125, -4.3111114501953125, 40.819671630859375, 88.01171875, 107.3358154296875, 9.10891342163086, 60.057708740234375, 95.29638671875, 43.5809326171875, 29.58710479736328, 51.69841766357422, 71.63716125488281, 4.980251312255859, 4.160797119140625, 11.010696411132812, 75.44346618652344, 48.51832580566406, 15.978607177734375, -0.02716064453125, 85.76541137695312, 71.84344482421875, -1.5755901336669922, -37.36956787109375, 2.7830810546875, 70.14404296875, 90.94235229492188, -1.5717315673828125, 4.765830993652344, -65.09210205078125, 60.593414306640625, 19.567283630371094, 75.24386596679688, -0.1658172607421875, -5.279821395874023, 38.87615966796875, 68.1320571899414, 0.0262451171875, -2.177001953125, -94.06532287597656, 29.522720336914062, -65.90626525878906, -69.1766357421875, 37.49278259277344, -8.1932373046875, -12.101448059082031, -34.84564208984375, 27.88861083984375, 64.90010833740234, 22.34698486328125, 3.0022048950195312, -53.058563232421875, -92.2183837890625, 46.4459228515625, 14.019332885742188, 41.803443908691406, 71.03070831298828, 93.31494140625, -11.836666107177734, -69.00613403320312, 1.5438098907470703, 52.147064208984375, 46.431060791015625, 79.9400634765625, -15.869476318359375, 90.18753051757812, -5.224475860595703, 3.780548095703125, -38.44293212890625, -41.61065673828125, 6.104888916015625, 0.0, 15.837188720703125, 3.812835693359375, 32.088497161865234, 5.078948974609375, -52.41180419921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000188.npy"}
{"epoch": 0.393717277486911, "step": 189, "batch_size": 128, "mean": 29.001649856567383, "std": 37.583858489990234, "min": -71.6851806640625, "p10": -12.705353546142575, "median": 26.501785278320312, "p90": 79.86887207031249, "max": 110.01666259765625, "pos_frac": 0.8125, "sample": [80.90567016601562, 18.9598388671875, -20.022918701171875, 11.384536743164062, 10.66064453125, 44.253021240234375, 56.85345458984375, 102.27752685546875, 52.0516242980957, 25.29150390625, 43.91527557373047, 53.2989501953125, 96.80044555664062, 32.92559814453125, 63.737762451171875, 39.1136474609375, 74.16937255859375, -0.16773605346679688, -8.950714111328125, 5.41217041015625, 51.21575927734375, 85.54583740234375, -15.156394958496094, 96.05715942382812, 37.910064697265625, 30.13897705078125, 36.59222412109375, 73.5849609375, 4.5089874267578125, -5.618827819824219, -26.98858642578125, 5.248863220214844, 23.842323303222656, 24.61815643310547, 2.0899124145507812, 25.06102752685547, 21.389019012451172, 3.1795120239257812, 15.697372436523438, 19.45330810546875, 1.5537891387939453, 86.04534912109375, 67.58592224121094, 30.10693359375, 71.0064697265625, 50.334320068359375, 34.174072265625, -16.97412109375, 75.44160461425781, -71.6851806640625, -32.2288818359375, 10.31500244140625, -11.6549072265625, 63.67547607421875, 1.789398193359375, 29.6905517578125, 31.935546875, 3.0190887451171875, 10.42120361328125, 13.490234375, 74.58175659179688, 44.83135986328125, 76.76922607421875, -3.9906005859375, -7.026519775390625, 14.805904388427734, 8.822050094604492, 4.533714294433594, 4.686317443847656, 8.073478698730469, 48.83027648925781, 5.094291687011719, -28.236854553222656, 53.38336181640625, 7.193748474121094, 3.1672897338867188, 29.073455810546875, 6.056552886962891, 3.9251708984375, 15.532470703125, 7.05645751953125, 69.36151123046875, 37.298858642578125, -29.271484375, 46.37841796875, 54.97666931152344, 105.90167236328125, -30.278488159179688, 0.13449859619140625, 0.6719551086425781, 7.953643798828125, 84.39181518554688, 29.21856689453125, -56.2161865234375, 2.7227935791015625, 0.0, 109.6527099609375, 27.712066650390625, 33.16009521484375, 84.7257080078125, 71.4481201171875, 38.410430908203125, 77.85137176513672, 41.21235656738281, 19.763381958007812, 0.0, 27.720001220703125, 102.214599609375, -18.82904052734375, 64.170654296875, 73.86466979980469, 14.837051391601562, 39.80535888671875, 10.090856552124023, 27.997406005859375, -59.223175048828125, 43.47991943359375, 79.42453002929688, 62.85128402709961, 40.770294189453125, 33.43035888671875, 96.7001953125, -29.990516662597656, 110.01666259765625, 74.40557861328125, -8.478622436523438, -5.4632415771484375, -11.208332061767578], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000189.npy"}
{"epoch": 0.3958115183246073, "step": 190, "batch_size": 128, "mean": 27.868839263916016, "std": 42.243202209472656, "min": -83.68081665039062, "p10": -20.04705581665039, "median": 26.635854721069336, "p90": 81.45291137695312, "max": 127.13287353515625, "pos_frac": 0.765625, "sample": [-27.891563415527344, 13.2578125, 99.01739501953125, 43.14622497558594, 0.692596435546875, 27.19900894165039, -64.85333251953125, -3.9071788787841797, -21.366867065429688, 32.046417236328125, 52.17973327636719, -28.259681701660156, -19.82469940185547, 83.04327392578125, 32.48118591308594, 22.450912475585938, 7.73638916015625, -10.321456909179688, 42.04371643066406, -82.04351806640625, -10.752071380615234, -14.846351623535156, -41.973472595214844, 73.17179870605469, 72.56291198730469, 5.859954833984375, 39.496490478515625, 2.162139892578125, 48.433441162109375, 92.26119995117188, 117.59490966796875, -20.565887451171875, 14.210159301757812, 39.395790100097656, 52.177001953125, -5.0615997314453125, 57.488922119140625, 4.989192962646484, 61.58061981201172, 22.995086669921875, 81.03067016601562, -81.42706298828125, 4.86724853515625, 99.90754699707031, 67.76554107666016, 0.81890869140625, 63.7265625, 24.474735260009766, 58.411285400390625, 102.890380859375, 16.243194580078125, 6.524848937988281, 55.32274627685547, -0.914093017578125, 42.65852355957031, 26.07270050048828, 5.652727127075195, -2.218292236328125, 24.990875244140625, -9.45050048828125, -2.010498046875, 71.84303283691406, 35.746490478515625, 23.777542114257812, 46.75074768066406, 33.925384521484375, 27.30535888671875, -16.349822998046875, 39.322540283203125, 56.649803161621094, 59.795387268066406, 38.79678726196289, 58.92408752441406, 0.0, 69.5321044921875, 79.97550964355469, 44.30949783325195, 51.49761199951172, 111.01931762695312, 17.20813751220703, 3.603515625, 39.64178466796875, 31.252670288085938, 51.298004150390625, 73.9779052734375, 13.98065185546875, 17.71129608154297, 8.564910888671875, 59.032928466796875, -1.7286376953125, 17.085189819335938, 127.13287353515625, 54.5458984375, 92.67727661132812, -2.7759933471679688, -83.68081665039062, 14.798446655273438, -0.28173828125, 79.80716705322266, -31.8154296875, 48.316650390625, 21.606903076171875, 67.74169921875, 82.43814086914062, 20.165771484375, 51.97735595703125, 15.138511657714844, 11.50537109375, -7.81878662109375, 28.47491455078125, 3.7982521057128906, 112.77056884765625, 38.24986267089844, 41.3880615234375, 4.265857696533203, 9.38909912109375, -8.637237548828125, 54.45377731323242, -67.41995239257812, 57.943023681640625, 45.537384033203125, 0.8747100830078125, 5.885265350341797, 54.42497253417969, -73.00077819824219, 87.46783447265625, -35.967254638671875, 88.03935241699219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000190.npy"}
{"epoch": 0.39790575916230364, "step": 191, "batch_size": 128, "mean": 26.622127532958984, "std": 45.14908981323242, "min": -113.774658203125, "p10": -25.392286872863767, "median": 15.694038391113281, "p90": 87.05148620605468, "max": 122.7000732421875, "pos_frac": 0.765625, "sample": [95.84219360351562, 65.73582458496094, -10.281463623046875, -12.176162719726562, 49.93457794189453, 97.7313232421875, 14.433883666992188, 14.3367919921875, 1.74505615234375, -60.780487060546875, 25.186676025390625, 51.74877166748047, 5.45256233215332, 5.6003875732421875, 6.748443603515625, -17.57354736328125, 50.86271286010742, 0.41192626953125, 0.9995937347412109, 55.259674072265625, 43.77427673339844, 88.42971801757812, 12.039497375488281, 2.764190673828125, -0.17082977294921875, 15.476333618164062, -10.1690673828125, -75.97418212890625, 64.917236328125, 50.13169860839844, 11.610214233398438, 3.7816314697265625, -17.638809204101562, 77.05838012695312, 5.48248291015625, -29.462158203125, 53.56768798828125, -6.57659912109375, -113.774658203125, -4.0833587646484375, 28.637428283691406, 74.49591064453125, 105.31411743164062, 12.608406066894531, 3.343090057373047, 105.48454284667969, 14.7777099609375, 10.546539306640625, 10.028543472290039, 15.05596923828125, 2.015045166015625, 69.75469970703125, 106.114990234375, 85.62338256835938, 10.038116455078125, -6.9769287109375, 51.57097625732422, 86.13600158691406, 26.71649169921875, 24.167343139648438, 7.832754135131836, 5.689971923828125, 45.82972717285156, -14.629600524902344, 76.30838012695312, 6.7397613525390625, 108.13352966308594, 65.0063247680664, 22.835205078125, 3.9432239532470703, 2.572479248046875, 63.012596130371094, 7.030235290527344, -37.189453125, 23.18029022216797, -63.741546630859375, 53.215087890625, 26.15045166015625, 66.7601318359375, -7.275672912597656, 3.315399169921875, 39.100738525390625, 2.051607131958008, 93.63424682617188, -49.11090087890625, 122.7000732421875, 85.12948608398438, -58.802642822265625, 80.28628540039062, 91.09732055664062, 24.7017822265625, 74.36883544921875, 15.9117431640625, 0.16132354736328125, -1.5072250366210938, -40.84184265136719, 63.0057373046875, 33.673980712890625, 101.90628051757812, 33.50831604003906, 29.729034423828125, 9.242919921875, 71.97445678710938, 54.71519470214844, 115.51593017578125, 86.4608154296875, 3.6832351684570312, 4.1877899169921875, -52.07916259765625, 16.7449951171875, -0.708282470703125, -67.29345703125, 54.71202087402344, 59.19111633300781, 64.59172058105469, 57.298095703125, -5.63323974609375, -1.7363052368164062, -27.093582153320312, 72.15951538085938, 43.733612060546875, 45.69456481933594, 22.01988983154297, -33.7545166015625, 0.0, 76.64083862304688, -24.66316032409668, 108.74911499023438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000191.npy"}
{"epoch": 0.4, "step": 192, "batch_size": 128, "mean": 20.348949432373047, "std": 46.750579833984375, "min": -114.08731079101562, "p10": -41.32026214599608, "median": 14.161909103393555, "p90": 83.28738861083983, "max": 108.83927917480469, "pos_frac": 0.671875, "sample": [8.049163818359375, 31.236358642578125, 87.33908081054688, -80.335205078125, 78.55889892578125, 20.012847900390625, -60.682830810546875, 15.919052124023438, -5.792915344238281, 92.9248046875, 46.545928955078125, 35.639434814453125, 61.72467803955078, 81.9515380859375, 86.8719482421875, -6.8543853759765625, -50.99324035644531, 54.18084716796875, -114.08731079101562, 108.83927917480469, -80.52302551269531, 102.12762451171875, -11.790679931640625, -68.49653625488281, 7.65350341796875, 80.18667602539062, 13.302490234375, 9.586631774902344, -14.056129455566406, 62.200103759765625, -91.1385498046875, 32.312339782714844, 58.46549987792969, 58.3253173828125, -0.146026611328125, 2.369138717651367, -26.067779541015625, 61.423126220703125, 20.793563842773438, 16.780364990234375, -2.84332275390625, -10.12158203125, -14.474594116210938, 63.67120361328125, 85.89862060546875, -46.86579895019531, -3.587646484375, -67.83047485351562, 0.0, 0.0, 2.7449417114257812, -5.09332275390625, -67.56945037841797, 33.77685546875, -53.03807067871094, 61.77159118652344, 4.1605377197265625, 49.32684326171875, -5.1904754638671875, -20.07342529296875, 38.66969299316406, 40.14014434814453, -12.839447021484375, 3.46343994140625, 74.9512939453125, 86.83599853515625, 72.48844909667969, 82.16828918457031, -0.6123847961425781, 70.79473876953125, -38.943603515625, 4.35809326171875, 9.988433837890625, -0.240997314453125, 16.57659149169922, 54.186058044433594, 42.83197021484375, 86.55685424804688, 87.34768676757812, 10.051048278808594, 6.958103179931641, 77.31417846679688, -21.55584716796875, -34.924774169921875, -7.5785369873046875, 20.4996337890625, 15.02132797241211, 11.9945068359375, 66.01608276367188, 67.32131958007812, 0.03692626953125, 19.162841796875, -73.81826782226562, 107.212158203125, 8.827674865722656, 7.272369384765625, -2.0255966186523438, -8.990398406982422, 90.45877075195312, 48.27716064453125, 11.304420471191406, 23.790496826171875, 16.27850341796875, -18.019847869873047, 2.1094970703125, 30.94329833984375, 36.19163513183594, 10.6317138671875, 68.58126831054688, 48.979888916015625, 67.32332611083984, -11.22601318359375, 61.28870391845703, 11.96832275390625, 45.91569519042969, 10.705375671386719, 68.51016235351562, -6.651744842529297, -73.84967041015625, -0.38128662109375, 4.717872619628906, 89.53497314453125, 51.75599670410156, 22.180877685546875, 18.435653686523438, 90.17678833007812, -11.636978149414062, 79.83680725097656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000192.npy"}
{"epoch": 0.40209424083769635, "step": 193, "batch_size": 128, "mean": 22.861392974853516, "std": 43.00570297241211, "min": -79.26251220703125, "p10": -29.69102325439453, "median": 18.48354721069336, "p90": 82.0852066040039, "max": 107.42898559570312, "pos_frac": 0.703125, "sample": [76.60493469238281, -3.7411270141601562, -4.184478759765625, -7.985984802246094, -6.3597869873046875, 59.5721435546875, 16.61585235595703, 50.0570068359375, 29.105464935302734, 32.13372802734375, 23.183135986328125, 3.4546127319335938, 31.533660888671875, 76.515380859375, 3.179595947265625, 63.004241943359375, -23.2281494140625, 30.17279815673828, -0.9061279296875, 3.7829036712646484, 69.41268920898438, -20.87518310546875, 62.55433654785156, 9.22625732421875, 29.20172119140625, 33.942413330078125, 7.7745208740234375, -60.75421142578125, 62.000518798828125, -18.43359375, -44.594200134277344, -1.53509521484375, 22.36688232421875, 23.32708740234375, 59.23502731323242, -20.90802001953125, 90.83331298828125, -56.4173583984375, 3.6843318939208984, 33.541465759277344, -69.6556396484375, -9.882232666015625, 1.446685791015625, 42.6414794921875, 10.759162902832031, 44.520172119140625, 82.53741455078125, 33.652557373046875, 83.54373168945312, 94.53353881835938, 0.449676513671875, 9.875579833984375, 63.666893005371094, 5.535213470458984, 55.32676696777344, 5.290122985839844, 96.79315948486328, 26.98492431640625, -17.90447998046875, 37.135009765625, -8.937362670898438, 33.766448974609375, 67.22512817382812, 6.762298583984375, 98.8211669921875, -27.465911865234375, 47.21662521362305, 43.287689208984375, 21.40972900390625, -8.4190673828125, -45.3714599609375, -29.400833129882812, -2.383636474609375, 102.94439697265625, -30.368133544921875, 56.6014404296875, 77.27615356445312, 87.19613647460938, 12.721893310546875, 95.3779296875, 72.26242065429688, 44.6239013671875, 72.5206298828125, -53.837860107421875, 95.33892822265625, 100.21832275390625, 9.760452270507812, 1.5327529907226562, -6.86309814453125, 19.485595703125, 60.65679931640625, 0.9443206787109375, 107.42898559570312, 75.42378234863281, 13.374412536621094, 19.2296142578125, 27.55963134765625, 5.334564208984375, 23.526657104492188, 66.95480346679688, 57.008209228515625, 81.89140319824219, 8.914711952209473, -43.29823303222656, 39.682952880859375, -18.532257080078125, 98.45162963867188, -27.1033935546875, -67.03033447265625, -4.538818359375, -2.3732070922851562, -5.028564453125, -79.26251220703125, 56.3116455078125, 48.662109375, 17.73748016357422, 76.76101684570312, -32.21588134765625, 9.94073486328125, -4.53192138671875, 1.02764892578125, -5.687713623046875, -59.76690673828125, 1.9783477783203125, 68.24600219726562, -44.91130065917969, 3.4696044921875, 63.37702941894531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000193.npy"}
{"epoch": 0.4041884816753927, "step": 194, "batch_size": 128, "mean": 18.00137710571289, "std": 42.76070022583008, "min": -85.81369018554688, "p10": -22.311566162109376, "median": 8.553153991699219, "p90": 77.5181640625, "max": 116.73495483398438, "pos_frac": 0.640625, "sample": [114.84890747070312, 19.740493774414062, 20.451416015625, -49.83738708496094, 25.018218994140625, 109.56765747070312, 45.08702087402344, -8.810089111328125, 2.216217041015625, -1.8877220153808594, 72.96443939208984, -22.286270141601562, -3.1206836700439453, 5.6941070556640625, 11.886383056640625, 3.0843677520751953, -10.457206726074219, 46.53276824951172, 86.91249084472656, -0.2674560546875, 17.14202880859375, -4.143545150756836, 19.215328216552734, 5.6866607666015625, 12.847869873046875, -3.5023193359375, 1.25604248046875, 29.261444091796875, -7.652618408203125, -85.74508666992188, -11.413253784179688, 24.92809295654297, -10.2203369140625, 10.713470458984375, 28.28411865234375, 15.104644775390625, 116.73495483398438, 55.60059356689453, 11.639923095703125, 65.64053344726562, -19.631546020507812, 102.580810546875, -7.8310699462890625, -12.630210876464844, -13.084548950195312, 90.95883178710938, -21.143264770507812, 77.67798614501953, 63.111968994140625, 8.992340087890625, -10.384017944335938, 32.8221435546875, 28.196975708007812, 35.800811767578125, 111.59912109375, 31.618118286132812, 39.32811737060547, 31.02471923828125, 59.45445251464844, -12.619140625, 18.9208984375, 12.17962646484375, 2.4544525146484375, 61.296417236328125, -55.98699951171875, -41.39959716796875, 9.339981079101562, -85.81369018554688, 38.128753662109375, -16.8953857421875, 105.7637939453125, -9.9058837890625, -29.5640869140625, 86.14533996582031, 0.792999267578125, 80.08157348632812, -3.6608123779296875, -18.62316131591797, 33.852752685546875, 70.38255310058594, -14.26446533203125, 6.353271484375, 69.81033325195312, 77.44966888427734, -3.5664806365966797, -61.954833984375, 54.48309326171875, -11.027519226074219, 65.6240234375, 73.80526733398438, 55.3741455078125, -66.22543334960938, 16.137359619140625, -44.85248947143555, 5.275508880615234, -0.53875732421875, 6.807945251464844, -75.09483337402344, 108.77386474609375, 7.6454010009765625, -5.193328857421875, 56.21539306640625, 0.5767936706542969, 7.2694549560546875, -48.904327392578125, 8.113967895507812, -12.1505126953125, 30.22412109375, -4.9185028076171875, -14.383056640625, -0.47698974609375, 91.47467041015625, 63.433319091796875, 9.962921142578125, 38.85969543457031, 40.230712890625, 68.2420654296875, 2.4998397827148438, -36.372047424316406, 3.8131866455078125, 3.1982421875, -6.442962646484375, 59.7005615234375, -18.398590087890625, 3.0269546508789062, -22.370590209960938, 26.399398803710938, 58.4825439453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000194.npy"}
{"epoch": 0.406282722513089, "step": 195, "batch_size": 128, "mean": 17.5473575592041, "std": 44.59526824951172, "min": -86.98333740234375, "p10": -28.73837203979492, "median": 8.889890670776367, "p90": 74.42343749999999, "max": 127.01715087890625, "pos_frac": 0.625, "sample": [-10.594863891601562, -1.1679000854492188, -1.9374542236328125, 66.7705078125, 75.14155578613281, 67.1513671875, -75.53448486328125, 2.9376220703125, 57.761932373046875, 8.622798919677734, 43.0919189453125, 9.567787170410156, 3.2609176635742188, -8.69131851196289, -5.466972351074219, 22.461700439453125, -10.943878173828125, -15.2034912109375, 19.3826904296875, 29.583251953125, -16.72555160522461, 15.799240112304688, 73.76133728027344, -13.694900512695312, -15.0350341796875, -2.5810089111328125, -0.18868255615234375, 2.22955322265625, -30.838623046875, 10.95892333984375, -11.78179931640625, 64.153076171875, -28.559494018554688, -8.660873413085938, 6.16387939453125, 1.6358489990234375, 88.31272888183594, 92.50527954101562, -4.22491455078125, -10.567535400390625, -22.327667236328125, -9.157455444335938, 13.489784240722656, 5.5307464599609375, 77.1038818359375, 5.440528869628906, 73.2615966796875, -71.09051513671875, -19.93115234375, 43.401885986328125, -42.53337097167969, 53.7501220703125, 34.52857971191406, 35.16099548339844, -83.17889404296875, 27.28369140625, -4.1695556640625, -69.84129333496094, 75.89156341552734, -5.4911651611328125, 29.216873168945312, -86.98333740234375, 4.331878662109375, 58.95970916748047, -17.889427185058594, -76.24679565429688, -73.32809448242188, -12.911224365234375, -1.2684478759765625, 51.19000244140625, 94.18795776367188, 38.97027587890625, 64.1524658203125, -18.47881317138672, 33.217803955078125, 88.80999755859375, 16.108184814453125, 2.90570068359375, 35.275390625, 16.311065673828125, 53.80381774902344, 46.722686767578125, 32.84271240234375, 51.91192626953125, -4.88653564453125, -1.7665252685546875, 6.3244781494140625, -56.578582763671875, -1.033930778503418, 29.786376953125, -24.058380126953125, 118.28857421875, 65.98196411132812, 1.6107177734375, 66.53833770751953, 48.24676513671875, 127.01715087890625, 7.8800201416015625, 67.65782165527344, 97.29818725585938, -22.109825134277344, 51.34483337402344, 9.9224853515625, 68.67967224121094, 4.5716552734375, 10.436904907226562, -9.99017333984375, 60.79229736328125, 51.70947265625, 75.00767517089844, 105.96697998046875, 7.09814453125, -81.85772705078125, 28.97723388671875, -29.15575408935547, 7.403228759765625, 87.88134765625, -63.7625732421875, 10.15582275390625, 62.55195617675781, 9.156982421875, -5.728233337402344, -3.17352294921875, 50.1614990234375, 74.17304992675781, 45.546783447265625, 70.207275390625, -16.001983642578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000195.npy"}
{"epoch": 0.4083769633507853, "step": 196, "batch_size": 128, "mean": 29.659008026123047, "std": 39.37150573730469, "min": -64.0206298828125, "p10": -12.283093261718749, "median": 16.47445297241211, "p90": 78.57133560180664, "max": 124.00509643554688, "pos_frac": 0.765625, "sample": [-0.20743370056152344, 9.631103515625, 15.796043395996094, 1.6112213134765625, 15.248533248901367, 10.224365234375, -21.187591552734375, 3.664182662963867, 63.28465270996094, 0.073883056640625, 4.900299072265625, 83.980712890625, 39.625, 47.30438232421875, 100.90264892578125, 4.76904296875, 2.01336669921875, 3.07769775390625, 13.829910278320312, 21.118072509765625, 52.53895568847656, 6.262706756591797, 53.079437255859375, 78.31986999511719, -1.1539459228515625, 61.70225524902344, -4.274078369140625, -0.4490032196044922, -55.46912384033203, 69.05865478515625, 106.633056640625, 81.87454986572266, 45.8804931640625, 57.7139892578125, 15.40447998046875, 53.082427978515625, 8.790451049804688, -40.099395751953125, 71.08090209960938, 49.965538024902344, -24.0439453125, 94.49264526367188, 37.43975830078125, -9.006378173828125, 15.149089813232422, 8.826656341552734, 15.7353515625, 97.16717529296875, 20.617950439453125, 64.06155395507812, -34.80865478515625, 0.5523166656494141, 1.673980712890625, 5.267974853515625, -1.6410446166992188, 80.91705322265625, 75.50332641601562, 3.7753257751464844, 7.521240234375, 59.168426513671875, 59.634521484375, 10.6387939453125, 78.41075897216797, -7.1871337890625, 21.52349853515625, 84.98536682128906, -15.952316284179688, 49.555389404296875, -18.451019287109375, -14.360214233398438, 15.655105590820312, 18.645355224609375, 1.3093700408935547, 61.810333251953125, 47.17304992675781, 72.25553894042969, 1.708251953125, 22.95648193359375, 42.2620849609375, 32.60400390625, 17.152862548828125, -47.67936706542969, 59.54576110839844, -3.64080810546875, -11.80328369140625, 67.35797119140625, 63.92718505859375, 72.75515747070312, 73.40469360351562, 0.941253662109375, 73.09625244140625, -64.0206298828125, 17.270660400390625, 106.73312377929688, 52.2930908203125, -4.143096923828125, -9.088569641113281, -13.40264892578125, 70.1474609375, 75.02424621582031, 62.189849853515625, -9.757888793945312, 9.218841552734375, 124.00509643554688, 89.58795166015625, 14.660066604614258, -25.845775604248047, 68.32303619384766, 77.40766906738281, 78.94601440429688, 65.866455078125, 104.79510498046875, 75.11102294921875, 74.30502319335938, -6.85040283203125, -4.882469177246094, -7.323448181152344, 65.17724609375, 59.401405334472656, 38.626495361328125, 10.14739990234375, -5.261695861816406, -42.068206787109375, 51.37440490722656, 9.557750701904297, 13.82342529296875, -7.474609375, 4.2725372314453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000196.npy"}
{"epoch": 0.41047120418848165, "step": 197, "batch_size": 128, "mean": 37.33830261230469, "std": 44.573490142822266, "min": -81.03427124023438, "p10": -13.954553222656244, "median": 41.83171081542969, "p90": 90.01852416992188, "max": 143.384765625, "pos_frac": 0.796875, "sample": [-66.87974548339844, 6.5876007080078125, -3.5255126953125, 81.66131591796875, 49.710723876953125, -10.9931640625, 66.12576293945312, 24.159561157226562, 23.739242553710938, 85.98992919921875, 97.15145874023438, 101.38343811035156, -11.2662353515625, 31.3548583984375, 37.93721008300781, -28.80449867248535, 124.23934936523438, 97.28057098388672, 61.13822937011719, 102.5277099609375, 1.8952064514160156, 56.21977996826172, 89.26400756835938, 28.233795166015625, 11.839248657226562, 29.83685302734375, 80.66351318359375, 89.9832763671875, 78.92578125, 80.07989501953125, 69.91307067871094, 52.25824737548828, -51.22590637207031, 52.913604736328125, 40.86175537109375, 32.82862091064453, 64.58392333984375, -0.08187103271484375, 28.218338012695312, 90.10076904296875, 44.69123077392578, 55.74639892578125, -75.78079223632812, 89.12396240234375, -81.03427124023438, 10.36151123046875, 99.01080322265625, 1.2547492980957031, -12.017593383789062, -11.345230102539062, 56.33076477050781, 56.908477783203125, 23.884735107421875, -17.311126708984375, 5.066444396972656, 2.224628448486328, 68.99923706054688, -12.516021728515625, 77.08087158203125, 68.692626953125, 61.22962951660156, 143.384765625, 4.661460876464844, 92.5859375, -32.951416015625, 5.898292541503906, 61.459503173828125, 75.40665435791016, 44.344886779785156, 52.814422607421875, 21.869300842285156, 40.4122314453125, -2.3661651611328125, 71.41015625, 3.7355804443359375, 79.04830932617188, -17.52667236328125, 35.416412353515625, -45.1356201171875, 3.37432861328125, -11.6099853515625, 10.201950073242188, 51.069549560546875, 55.19073486328125, -26.42254638671875, 72.42410278320312, 1.8939208984375, 56.172515869140625, 62.4849853515625, 42.717254638671875, 28.476318359375, -0.500091552734375, 83.9510498046875, 47.17730712890625, 55.332916259765625, 46.1685791015625, 27.522125244140625, -40.931304931640625, 18.382362365722656, -71.96224975585938, 11.089874267578125, 55.37132263183594, -5.0092315673828125, 72.92953491210938, 80.20574951171875, 55.594451904296875, 34.88092041015625, 33.72154235839844, 88.6307373046875, 40.9461669921875, 109.81143188476562, 101.25250244140625, 31.10845947265625, 4.8374481201171875, 69.82756042480469, 17.196929931640625, 91.79812622070312, 1.5407638549804688, 72.05073547363281, 15.836822509765625, 0.0, 126.1307373046875, 71.06866455078125, 67.21316528320312, 66.60148620605469, -41.2071533203125, 55.763092041015625, -2.9012908935546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000197.npy"}
{"epoch": 0.41256544502617803, "step": 198, "batch_size": 128, "mean": 28.690860748291016, "std": 49.173648834228516, "min": -106.67120361328125, "p10": -23.41417732238769, "median": 21.644058227539062, "p90": 91.09987182617188, "max": 127.10855102539062, "pos_frac": 0.75, "sample": [74.67498779296875, 21.397613525390625, 11.67071533203125, -9.660369873046875, 60.507568359375, -3.72332763671875, 75.58126831054688, 91.8787841796875, 34.64276123046875, 6.66180419921875, -13.806594848632812, -7.893768310546875, 43.2259521484375, 36.738128662109375, 9.62491226196289, -7.5437469482421875, 99.43220520019531, 58.24859619140625, 76.34271240234375, 9.67193603515625, 89.62213134765625, -87.33648681640625, -90.49203491210938, 90.76605224609375, 127.10855102539062, 113.0379638671875, -30.09619140625, 12.627655029296875, 17.50738525390625, 22.2430419921875, 57.232391357421875, 74.57893371582031, 77.215576171875, 33.006256103515625, 0.761474609375, 53.615196228027344, -34.9814453125, 124.7825927734375, -46.956634521484375, 116.443115234375, -11.10369873046875, 1.75616455078125, 86.0865478515625, 72.43486022949219, -26.690719604492188, 0.5237636566162109, 107.55450439453125, -22.009944915771484, -35.460426330566406, 5.171604156494141, 49.91133117675781, 46.617156982421875, -2.6884078979492188, 78.37077331542969, 34.867889404296875, 73.83682250976562, 98.80062866210938, 2.6012420654296875, 21.8905029296875, 45.515380859375, 11.944351196289062, 41.56195068359375, -2.932361602783203, 90.01644897460938, 67.31707763671875, 1.4508056640625, 1.9073619842529297, 98.90579223632812, 28.978363037109375, 34.37565612792969, 16.193016052246094, -76.49758911132812, -17.365081787109375, 54.35865783691406, -106.67120361328125, -20.467926025390625, 20.291351318359375, 80.75436401367188, -13.846214294433594, 48.702392578125, 39.57450866699219, 77.70101928710938, 51.49900817871094, -13.91845703125, 72.26448822021484, 0.38950347900390625, 56.22087097167969, 89.80184936523438, -73.48368835449219, 13.528076171875, 0.0, 86.38995361328125, 11.231414794921875, -18.656097412109375, -56.44639587402344, 13.985435485839844, -2.860532760620117, 63.579368591308594, 100.93426513671875, 37.202117919921875, 101.62457275390625, 2.8688125610351562, 74.62864685058594, 53.786712646484375, 29.041610717773438, 53.16700744628906, 3.832366943359375, 4.241237640380859, 1.6411895751953125, -104.72943115234375, 102.07122802734375, 10.30511474609375, 1.7201519012451172, 8.091911315917969, -4.618860244750977, -4.65093994140625, 108.990478515625, 31.158905029296875, 88.4010009765625, 12.92633056640625, 29.647857666015625, -27.806671142578125, 6.621917724609375, -11.174339294433594, 16.185897827148438, 83.36383056640625, 74.04608154296875, 2.78997802734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000198.npy"}
{"epoch": 0.41465968586387436, "step": 199, "batch_size": 128, "mean": 27.007282257080078, "std": 45.12615966796875, "min": -81.88357543945312, "p10": -31.38304443359375, "median": 22.725257873535156, "p90": 86.84094619750977, "max": 121.6884765625, "pos_frac": 0.734375, "sample": [17.420074462890625, 25.43731689453125, 83.50827026367188, 67.06242370605469, 93.51497650146484, 67.06645202636719, 67.53665161132812, 10.524085998535156, 86.83935546875, 22.9508056640625, 115.19769287109375, 75.81853485107422, 19.31512451171875, 59.333740234375, 105.89413452148438, -53.906982421875, 71.01126098632812, 58.27418518066406, 7.464937210083008, -3.082426071166992, -5.311805725097656, 2.5578842163085938, -1.5762462615966797, 2.587177276611328, -16.461700439453125, 37.76245880126953, 53.91191101074219, 2.6334152221679688, 15.126953125, 31.480010986328125, 40.592533111572266, 46.07365417480469, 61.68213653564453, 46.20147705078125, 29.559188842773438, 56.8084716796875, -44.385986328125, -13.989906311035156, 70.58161926269531, 11.9991455078125, 11.870101928710938, 75.320068359375, 11.216705322265625, 3.9963550567626953, 15.3624267578125, 76.9691162109375, -13.0595703125, -6.276588439941406, -16.261642456054688, -66.17037963867188, 64.5316162109375, 110.2239990234375, 16.897918701171875, 93.223876953125, -12.162826538085938, 27.258346557617188, 23.077117919921875, 22.499710083007812, 90.92420959472656, -38.35679626464844, -81.88357543945312, 71.80902099609375, 79.69342041015625, 25.142669677734375, -15.107551574707031, -45.30302429199219, 54.16230010986328, 12.672821044921875, -18.298568725585938, 30.284286499023438, 105.16427612304688, -0.39825439453125, 4.079378128051758, 108.37088012695312, 24.006301879882812, -3.664306640625, 48.38812255859375, 33.05570983886719, -1.3667545318603516, 47.1075439453125, 97.36090087890625, 7.3712005615234375, 64.77984619140625, -11.48016357421875, 71.71139526367188, 88.45626831054688, -66.93061065673828, 63.693145751953125, 76.44009399414062, 35.70539855957031, -31.289382934570312, -19.075469970703125, 56.13026428222656, 93.08270263671875, -6.5879974365234375, 50.408660888671875, 0.6444530487060547, 10.37896728515625, 21.9183349609375, 38.671051025390625, 34.189727783203125, -46.654659271240234, 18.038177490234375, 12.496490478515625, 65.73042297363281, -81.27163696289062, -61.01361083984375, 0.9662990570068359, 10.941917419433594, 54.346649169921875, 13.821449279785156, 13.4383544921875, 77.40707397460938, -2.9606781005859375, 67.30467224121094, 17.674468994140625, 86.84465789794922, 2.568389892578125, -6.178947448730469, 55.36687469482422, 9.647247314453125, -66.93341064453125, -31.314361572265625, -36.096656799316406, 121.6884765625, 80.0045166015625, 43.023040771484375, -31.543304443359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000199.npy"}
{"epoch": 0.4167539267015707, "step": 200, "batch_size": 128, "mean": 23.857168197631836, "std": 52.75448989868164, "min": -99.02027893066406, "p10": -47.886163330078126, "median": 14.706283569335938, "p90": 94.24550476074218, "max": 137.3956298828125, "pos_frac": 0.671875, "sample": [70.71688842773438, -60.6932373046875, 114.88754272460938, 17.090721130371094, -61.7784423828125, 13.991554260253906, -41.33271789550781, 59.9296875, -7.29669189453125, 6.1782989501953125, -2.1817779541015625, 78.27798461914062, 25.754119873046875, 2.989501953125, -81.544677734375, 78.59832763671875, 64.25482177734375, -13.418289184570312, 33.42193603515625, 119.2744140625, -81.968017578125, -37.268890380859375, 105.84855651855469, 52.71581268310547, 95.60968017578125, 68.54583740234375, -21.39801025390625, -14.186164855957031, 27.666839599609375, -5.75042724609375, -38.8265380859375, -91.17556762695312, 13.16733169555664, -46.95526123046875, 14.642303466796875, 31.34515380859375, -53.78693389892578, 93.66085815429688, 68.78678894042969, -12.8897705078125, 0.0, 52.218475341796875, 88.4443359375, 10.629714965820312, -7.98443603515625, -80.43827819824219, 82.40155029296875, 10.118896484375, 116.90145874023438, -8.469757080078125, 18.86273193359375, 14.770263671875, 92.8565673828125, 3.9842453002929688, -1.861114501953125, -19.221221923828125, 17.35523223876953, -73.58024597167969, 97.66571044921875, -3.7316970825195312, 41.42278289794922, 8.19384765625, 1.7264842987060547, 68.47463989257812, 74.40786743164062, 66.72782897949219, 10.774726867675781, 96.0601806640625, -47.943756103515625, 59.530517578125, 102.24275207519531, 50.89813232421875, -3.486236572265625, 47.21653747558594, 45.380615234375, 25.7098388671875, -6.426910400390625, -25.96576690673828, 8.869415283203125, 62.14910888671875, -82.48007202148438, 100.01422119140625, 137.3956298828125, -2.1873779296875, 14.781585693359375, 51.87522888183594, -8.598846435546875, 66.6043701171875, -99.02027893066406, 88.9053955078125, 10.75775146484375, 119.568115234375, -57.10809326171875, 88.54443359375, 90.10797119140625, 84.29461669921875, 5.843414306640625, 101.3309326171875, -28.6112060546875, -48.11073303222656, 14.5782470703125, 46.915069580078125, 72.96676635742188, 27.818199157714844, -16.289825439453125, 79.86607360839844, 12.11126708984375, -47.861480712890625, 3.877044677734375, 1.116455078125, 10.982093811035156, 68.36144256591797, 59.727783203125, 116.70718383789062, 56.453041076660156, 68.89450073242188, -5.8492279052734375, -21.497840881347656, 57.89404296875, 38.955379486083984, 9.2093505859375, -2.240398406982422, 56.387229919433594, 6.0223388671875, -32.41593933105469, 15.043670654296875, 37.525665283203125, 2.7637405395507812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000200.npy"}
{"epoch": 0.418848167539267, "step": 201, "batch_size": 128, "mean": 27.85923194885254, "std": 44.210655212402344, "min": -80.25091552734375, "p10": -25.491358184814448, "median": 21.294708251953125, "p90": 85.84371337890624, "max": 125.24954223632812, "pos_frac": 0.7109375, "sample": [-75.83026123046875, -14.35406494140625, 56.63738250732422, 74.81170654296875, 82.52615356445312, 47.91461181640625, -2.1760101318359375, 19.556610107421875, 25.130462646484375, 56.295745849609375, 78.49336242675781, 1.8727502822875977, -5.461875915527344, 24.905731201171875, -66.6651611328125, -17.126983642578125, 75.68692016601562, 95.46160888671875, 63.427955627441406, 43.05218505859375, 13.745468139648438, 93.58796691894531, 87.74645233154297, 0.0, 65.14810180664062, 38.19203186035156, 70.36366271972656, 46.90496826171875, 4.81219482421875, 89.17266845703125, 94.4853515625, 56.619049072265625, 21.794677734375, -11.228446960449219, 0.0, -61.95562744140625, 8.571563720703125, -0.6122207641601562, 9.242721557617188, 5.829559326171875, 30.392303466796875, 46.75372314453125, 79.10148620605469, 74.56026458740234, -43.225067138671875, 14.964492797851562, 61.35383605957031, 60.822837829589844, -17.707504272460938, 51.93555450439453, -2.6960906982421875, 20.661216735839844, 23.11102294921875, 62.9041748046875, -2.441225051879883, 13.866317749023438, 16.36095428466797, 82.30972290039062, 70.16163635253906, 77.18557739257812, -3.9362258911132812, 76.12130737304688, 101.15325927734375, 13.380767822265625, 89.11972045898438, 66.87763214111328, 30.72747802734375, 0.0, 9.911186218261719, -80.25091552734375, 18.5732421875, 10.023345947265625, -2.562347412109375, 77.582275390625, 45.9244384765625, -2.8358612060546875, 0.9572868347167969, 9.387351989746094, -10.0496826171875, 39.513763427734375, 29.99139404296875, 37.87396240234375, -14.228485107421875, -13.345314025878906, -1.5987548828125, 92.04327392578125, -8.038833618164062, 86.78036499023438, -28.147125244140625, 125.24954223632812, 48.53338623046875, 63.082275390625, 56.54571533203125, -5.25537109375, -33.9586181640625, -28.3577880859375, 9.985466003417969, 53.658538818359375, 6.146354675292969, 88.73831176757812, 74.715087890625, 69.33070373535156, -55.756591796875, 72.51840209960938, 18.206024169921875, -8.597335815429688, -62.93843078613281, 20.79473876953125, 85.002197265625, -44.099517822265625, 20.556503295898438, 35.8912353515625, 17.1915283203125, 1.577545166015625, -3.6442718505859375, 104.01763916015625, 29.710784912109375, 47.988792419433594, 51.63427734375, 85.44229125976562, 8.756694793701172, -56.9500732421875, 12.48687744140625, 95.1434326171875, 1.9794845581054688, 63.53533935546875, -24.353172302246094, -66.42901611328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000201.npy"}
{"epoch": 0.42094240837696334, "step": 202, "batch_size": 128, "mean": 27.344629287719727, "std": 51.046321868896484, "min": -107.5263671875, "p10": -30.79442596435546, "median": 18.92335319519043, "p90": 97.97391815185546, "max": 115.14779663085938, "pos_frac": 0.7109375, "sample": [102.60748291015625, 10.984886169433594, 1.871551513671875, 15.442245483398438, 90.67611694335938, -0.0861663818359375, -21.883888244628906, 101.349365234375, -63.08912658691406, -12.83642578125, -4.27996826171875, 97.472900390625, 23.676727294921875, 86.22210693359375, 67.444091796875, -2.2632808685302734, -22.19256591796875, 69.29153442382812, 5.309791564941406, 109.61886596679688, 18.896175384521484, 57.166046142578125, -18.343414306640625, 34.500518798828125, -20.32196807861328, 6.8853759765625, 105.0006103515625, 0.8428211212158203, 24.250625610351562, 8.147689819335938, -52.16516876220703, 29.328399658203125, -39.69483947753906, 99.95086669921875, 22.27399444580078, 75.19972229003906, 66.40814208984375, 106.88580322265625, 10.046333312988281, -68.08499145507812, -1.2133255004882812, 106.77484130859375, 62.72344207763672, 90.65753173828125, -0.6358795166015625, 21.41375732421875, 42.716583251953125, 37.017112731933594, 88.37008666992188, -4.7829437255859375, -71.97355651855469, 62.13987731933594, 84.55603790283203, -1.528228759765625, -77.4554443359375, 55.51512145996094, 50.690673828125, 18.950531005859375, 26.757659912109375, -43.03874969482422, 62.37602233886719, 67.71102905273438, 102.25625610351562, -25.588165283203125, -9.959716796875, -17.054046630859375, 103.21438598632812, 0.850494384765625, 7.666656494140625, 31.924224853515625, 1.8461570739746094, 84.30322265625, 111.52767944335938, 2.0358009338378906, 62.79901123046875, 94.82516479492188, 70.2956771850586, 63.37066650390625, 90.34364318847656, 69.99630737304688, 0.0, 115.14779663085938, 1.0210037231445312, 14.640716552734375, 16.326507568359375, 6.100677490234375, 0.724395751953125, -52.51905822753906, 3.5429840087890625, -24.350555419921875, 58.39056396484375, 76.60952758789062, 3.9634857177734375, 17.25592041015625, 95.52410888671875, 57.67359161376953, -36.16688537597656, -7.595184326171875, -17.72393798828125, -28.491943359375, -107.5263671875, 92.57893371582031, 57.866180419921875, 81.72224426269531, 38.58685302734375, 28.26348876953125, -13.1444091796875, 11.87255859375, 30.612272262573242, 0.8155422210693359, 63.3565673828125, 73.22901916503906, -54.627166748046875, 11.066680908203125, 39.1318359375, -100.423828125, 40.585845947265625, 69.2593994140625, -94.097900390625, -2.786407470703125, 3.756490707397461, 86.07101440429688, 0.8860244750976562, -2.3907203674316406, 2.3942718505859375, 99.14295959472656, -12.568138122558594, 111.5010986328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000202.npy"}
{"epoch": 0.42303664921465967, "step": 203, "batch_size": 128, "mean": 22.310562133789062, "std": 39.0410041809082, "min": -99.23831176757812, "p10": -20.34924621582031, "median": 13.191200256347656, "p90": 76.38929443359375, "max": 118.38113403320312, "pos_frac": 0.703125, "sample": [2.196554183959961, 8.149627685546875, 3.89654541015625, 66.27593994140625, 70.36297607421875, 16.34466552734375, 10.727792739868164, 6.870269775390625, 14.951034545898438, 68.84745788574219, 70.599853515625, 47.572296142578125, 4.474123001098633, 39.25933837890625, 19.40771484375, 34.32763671875, -13.024078369140625, -52.66327667236328, -4.707733154296875, 13.631179809570312, 16.047561645507812, 0.0, -44.0997314453125, 54.79060363769531, -24.662582397460938, 11.833045959472656, -24.230499267578125, 74.3323974609375, 3.65557861328125, -2.860931396484375, 29.415374755859375, 56.29750061035156, 29.526123046875, 68.531005859375, 55.98606872558594, -5.7598724365234375, 27.399993896484375, 109.01605224609375, -1.039764404296875, -13.03826904296875, 25.176712036132812, 84.36167907714844, 27.157470703125, 64.53767395019531, 2.011730194091797, -4.49615478515625, 12.081985473632812, 85.52302551269531, -5.07171630859375, 60.02519989013672, -0.09219169616699219, 54.23041534423828, 24.50396728515625, 39.723358154296875, 118.35269165039062, -18.68585205078125, 49.1251220703125, -1.906951904296875, 78.90423583984375, 118.38113403320312, 5.051475524902344, 49.08558654785156, 31.713714599609375, -2.4115257263183594, -53.62895202636719, 82.9124526977539, -43.38604736328125, 79.18255615234375, 11.032855987548828, -1.6711883544921875, 60.232269287109375, 2.5194320678710938, -99.23831176757812, 56.10675048828125, -6.08819580078125, -2.932096481323242, -26.22662353515625, 4.71099853515625, -31.49169921875, 2.5629119873046875, 30.652572631835938, 12.159622192382812, -46.99444580078125, 48.6334228515625, 76.912841796875, 20.168304443359375, 60.741790771484375, 25.609375, 92.8544921875, 76.1649169921875, 10.501350402832031, 20.682861328125, -8.995849609375, 58.99029541015625, 8.011760711669922, -8.507156372070312, 37.1239013671875, -10.106613159179688, 73.4953842163086, 10.597824096679688, 69.99299621582031, 2.8924407958984375, -2.12835693359375, 11.469482421875, 12.575241088867188, -71.64181518554688, -5.051300048828125, -17.833175659179688, 17.253082275390625, 101.20602416992188, 0.0, 44.57232666015625, 36.99497985839844, 44.83705139160156, 0.4797554016113281, -4.562744140625, 26.510848999023438, 81.32388305664062, 9.4490966796875, -1.448089599609375, 34.362640380859375, -31.36151123046875, 12.751220703125, 83.9078369140625, 18.65478515625, -24.643875122070312, 11.334693908691406, 24.6385440826416], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000203.npy"}
{"epoch": 0.42513089005235605, "step": 204, "batch_size": 128, "mean": 32.10176467895508, "std": 45.5564079284668, "min": -84.83932495117188, "p10": -22.567269134521478, "median": 26.43405532836914, "p90": 91.44520416259763, "max": 137.71505737304688, "pos_frac": 0.765625, "sample": [106.5238037109375, 71.73265075683594, 108.53594970703125, 28.510284423828125, 14.614753723144531, 87.2054443359375, 4.970073699951172, 40.841514587402344, 4.716777801513672, 99.52676391601562, 12.7288818359375, 11.6949462890625, 68.9456787109375, -21.185150146484375, -46.87188720703125, -84.83932495117188, 15.756660461425781, -28.420928955078125, 6.733100891113281, 24.4388427734375, 83.69302368164062, 3.5760574340820312, 40.5673828125, 66.41447448730469, 27.570281982421875, 44.3089599609375, 10.375007629394531, -60.64915466308594, 13.234451293945312, 20.53521728515625, 31.0927734375, 137.71505737304688, 1.7967681884765625, -30.779830932617188, 38.377532958984375, -13.93212890625, 11.21731948852539, 117.84454345703125, -2.5275421142578125, 77.471435546875, 97.95600891113281, 11.539039611816406, 58.09918212890625, -13.917827606201172, 73.61309814453125, 74.63226318359375, 13.738525390625, 113.55999755859375, 86.70361328125, 103.25469970703125, 73.77979278564453, 37.426055908203125, -3.0872802734375, 62.08659362792969, 40.4554443359375, -2.115509033203125, 74.86984252929688, 47.64604187011719, 1.9146347045898438, -28.740692138671875, 29.93988037109375, 72.44778442382812, 49.8570556640625, -64.24517822265625, -10.013916015625, 104.89279174804688, 68.01089477539062, 72.7811279296875, -70.6395263671875, 80.2672119140625, -7.812891006469727, 74.489013671875, -2.743011474609375, -11.958999633789062, 5.837742805480957, 5.1432037353515625, 29.790821075439453, 82.46184539794922, -3.206188201904297, 20.35186767578125, 8.810714721679688, -7.366546630859375, -47.77128601074219, 10.774749755859375, 68.60845947265625, -5.178955078125, 51.0240478515625, 13.099578857421875, 76.3592529296875, 69.1741943359375, 89.63827514648438, 40.79466247558594, -38.8040771484375, 95.987060546875, -11.204202651977539, 0.9471435546875, 80.28976440429688, -35.19989013671875, 33.226165771484375, 87.19378662109375, 9.458045959472656, -1.2716064453125, 13.90869140625, 21.7935791015625, 64.58110809326172, 100.3299560546875, 25.297828674316406, 24.712448120117188, 95.567626953125, 16.225341796875, 67.09257507324219, 0.2195281982421875, 55.031646728515625, 81.09524536132812, 20.666885375976562, 31.64490509033203, 71.375, 3.7893905639648438, -7.358734130859375, 34.2589111328125, -25.792213439941406, 33.75401306152344, 89.67845153808594, 51.52283477783203, -58.4093017578125, 113.58920288085938, 3.0456008911132812, -0.309326171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000204.npy"}
{"epoch": 0.4272251308900524, "step": 205, "batch_size": 128, "mean": 29.26951789855957, "std": 49.296875, "min": -96.10992431640625, "p10": -42.854255676269524, "median": 28.134124755859375, "p90": 89.80344238281249, "max": 130.5360107421875, "pos_frac": 0.75, "sample": [65.34219360351562, 0.0, -4.8447113037109375, 2.874225616455078, 10.686626434326172, 78.80291748046875, -41.70977783203125, -52.888031005859375, -45.52470397949219, 1.0210437774658203, 130.5360107421875, 65.70582580566406, 5.459236145019531, 10.672691345214844, -10.065765380859375, 10.69796371459961, 108.72587585449219, 99.69354248046875, 74.11282348632812, 10.93853759765625, 22.233367919921875, 38.543128967285156, -49.30328369140625, 89.51336669921875, 25.4298095703125, 8.4647216796875, 33.98590087890625, 15.69268798828125, 96.64093017578125, 45.535919189453125, 71.90365600585938, 51.50568389892578, 1.506927490234375, 6.899017333984375, -1.605010986328125, 6.885217666625977, 60.049644470214844, 11.436279296875, 51.100616455078125, 11.313591003417969, 28.678314208984375, 0.8266448974609375, -46.682373046875, 81.26290893554688, 12.8853759765625, 72.560546875, 0.0, 28.9686279296875, -67.51358032226562, 30.41937255859375, 3.0322837829589844, 21.845169067382812, 88.24114990234375, 90.48028564453125, -54.045074462890625, 4.58184814453125, 79.46771240234375, 57.201446533203125, -1.09466552734375, 55.158599853515625, -71.51358795166016, -0.93017578125, 27.589935302734375, 46.01512145996094, -29.6126708984375, 3.041259765625, 90.58944702148438, -7.2981719970703125, -61.90252685546875, 125.94281005859375, 77.81161499023438, 110.31201171875, -96.10992431640625, 50.687835693359375, -3.6528244018554688, 68.69035339355469, 5.884246826171875, 68.53742980957031, 66.90105438232422, -4.8240966796875, 30.754592895507812, 108.05836486816406, 63.059967041015625, 83.60125732421875, 67.42631530761719, 10.510931015014648, -93.37210083007812, 1.1840972900390625, 67.00990295410156, 70.18376159667969, 113.8599853515625, 73.66061401367188, 112.14599609375, -27.905029296875, 65.144775390625, 94.90727233886719, 80.56446838378906, -20.393104553222656, 15.19155502319336, 2.2106781005859375, 111.947021484375, 69.37457275390625, -11.498046875, -41.510406494140625, 51.950225830078125, 58.244384765625, 1.1115264892578125, 61.156707763671875, 64.07460021972656, 78.43438720703125, 23.653335571289062, 15.347412109375, -78.25204467773438, 6.928285598754883, 45.009490966796875, -53.06451416015625, -64.78278350830078, 83.54887390136719, -9.01446533203125, 65.82752990722656, 31.9886474609375, 57.44541931152344, -0.0533905029296875, 50.392120361328125, 34.49797058105469, 75.05059814453125, 0.0, 30.483909606933594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000205.npy"}
{"epoch": 0.4293193717277487, "step": 206, "batch_size": 128, "mean": 28.682573318481445, "std": 46.150630950927734, "min": -113.86175537109375, "p10": -27.736047363281244, "median": 27.03680419921875, "p90": 87.29183044433593, "max": 111.70144653320312, "pos_frac": 0.7578125, "sample": [95.72134399414062, 2.311176300048828, -71.63412475585938, 61.570556640625, 107.37646484375, -0.4609527587890625, -20.303558349609375, 9.284942626953125, 19.844512939453125, 84.54502868652344, 35.01139831542969, 86.18934631347656, 96.43792724609375, -16.665531158447266, -40.385345458984375, 58.749603271484375, 27.455322265625, 3.01727294921875, -3.755279541015625, 88.22573852539062, 0.7920856475830078, 75.48736572265625, -2.0453720092773438, -26.018051147460938, -3.437744140625, 9.439697265625, -56.016082763671875, 50.27093505859375, 57.303199768066406, 10.339736938476562, 1.89471435546875, -9.0670166015625, -56.39045715332031, 70.88880920410156, 71.19598388671875, -12.449478149414062, 105.43450927734375, 85.64244079589844, 105.52886962890625, 38.66303253173828, 32.9892578125, 32.21373748779297, -5.9958343505859375, 26.6182861328125, 53.584930419921875, 45.695556640625, -23.529449462890625, 86.79177856445312, 69.33599853515625, 78.07206726074219, 71.84228515625, 61.05638885498047, 64.36601257324219, 46.0771484375, 2.997537612915039, 42.531036376953125, 40.01637268066406, 101.98446655273438, -113.86175537109375, 21.520355224609375, 76.655517578125, -48.690391540527344, -15.687057495117188, 71.349853515625, 31.653297424316406, 11.604263305664062, 36.864410400390625, 27.61968994140625, -5.020183563232422, 43.560546875, 11.283462524414062, 72.7147216796875, 80.63119506835938, -73.04071044921875, 104.23590087890625, 3.0544891357421875, -11.786209106445312, 47.110687255859375, 10.286285400390625, 2.05157470703125, -33.96075439453125, -31.744705200195312, 93.70159912109375, 58.92364501953125, 70.11468505859375, 10.15252685546875, 52.58256530761719, -38.10457992553711, 9.758939743041992, 0.6182861328125, 63.6505126953125, 70.54965209960938, 64.81997680664062, 6.54962158203125, -23.927154541015625, 51.428985595703125, 13.6473388671875, -11.500167846679688, 6.897705078125, -2.806262969970703, 72.47032165527344, 18.1988525390625, 99.34146118164062, 88.1971435546875, 3.89825439453125, 76.3594970703125, -38.27728271484375, -19.30170440673828, -58.498992919921875, 17.112030029296875, 2.7183990478515625, 111.70144653320312, 73.82969665527344, 6.48492431640625, 2.1552963256835938, 99.8052978515625, 62.466346740722656, 6.432777404785156, 86.90383911132812, 38.50384521484375, 3.5682373046875, 11.8348388671875, 1.21826171875, 75.04029846191406, 49.27452087402344, 9.219284057617188, -56.46722412109375, 45.076812744140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000206.npy"}
{"epoch": 0.431413612565445, "step": 207, "batch_size": 128, "mean": 24.306991577148438, "std": 52.71219253540039, "min": -105.065673828125, "p10": -39.305363464355466, "median": 13.25955581665039, "p90": 93.25609130859375, "max": 134.98370361328125, "pos_frac": 0.6875, "sample": [92.256103515625, 46.870269775390625, 14.646942138671875, -9.179134368896484, 16.51727294921875, 3.4073944091796875, 25.36553955078125, 119.973876953125, 42.996856689453125, -4.475210189819336, 25.51535987854004, 130.1983642578125, -41.68780517578125, 4.293907165527344, -7.10528564453125, 84.54314422607422, 1.5245590209960938, 48.38322448730469, 127.75411987304688, -11.83489990234375, 81.33779907226562, -13.87286376953125, -95.12765502929688, -21.536163330078125, 69.80693054199219, 29.6107177734375, -12.618370056152344, 12.793136596679688, -76.30126953125, 63.71258544921875, 75.7808837890625, 15.605133056640625, 10.9066162109375, -30.4908447265625, 74.851806640625, 75.1722640991211, 15.861297607421875, -2.9662952423095703, 20.812244415283203, 98.53561401367188, -12.9525146484375, 9.536392211914062, 48.813873291015625, 5.688751220703125, 7.8916015625, -20.432586669921875, 3.6005516052246094, 4.022186279296875, -86.00003051757812, 3.58770751953125, -105.065673828125, -4.24700927734375, -50.064971923828125, 26.26611328125, 85.25201416015625, 0.0, 9.389060974121094, -29.74188232421875, 67.30255126953125, -23.445648193359375, 110.14389038085938, -10.8834228515625, -52.965293884277344, 27.61676025390625, 14.49261474609375, -90.79620361328125, 96.60578918457031, -68.96498107910156, 9.13018798828125, -2.3147506713867188, 35.73907470703125, -10.77545166015625, 45.11854553222656, 85.53683471679688, 101.09603881835938, 93.47071838378906, 64.32843780517578, -3.963897705078125, 3.6956787109375, 91.10627746582031, -0.03916168212890625, -67.7442626953125, -9.337158203125, 34.0777473449707, 9.491548538208008, 134.98370361328125, 15.710441589355469, 76.86304473876953, -53.612518310546875, 1.932098388671875, 11.293121337890625, 9.556777954101562, -30.320770263671875, 102.34231567382812, 21.269500732421875, 68.73469543457031, 48.768310546875, 3.53839111328125, 53.885292053222656, -63.54443359375, 102.4398193359375, 87.12875366210938, -13.42449951171875, 81.06204223632812, 6.310462951660156, 111.813232421875, 7.899505615234375, 3.933990478515625, 9.465864181518555, 79.57574462890625, 10.48104476928711, 85.3828125, -2.4529266357421875, 93.16410827636719, 78.689453125, -8.380882263183594, 87.45574951171875, 82.70684814453125, 113.46722412109375, -26.400390625, 13.725975036621094, -70.14178466796875, 79.914794921875, 82.23106384277344, -38.28431701660156, 26.06964111328125, 59.251243591308594, 21.70611572265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000207.npy"}
{"epoch": 0.43350785340314135, "step": 208, "batch_size": 128, "mean": 27.50531005859375, "std": 45.91183090209961, "min": -116.0706787109375, "p10": -13.93841781616211, "median": 18.82107925415039, "p90": 85.39790267944335, "max": 139.9373779296875, "pos_frac": 0.734375, "sample": [-4.438079833984375, -6.7873382568359375, -25.56231689453125, 17.896041870117188, -86.8919677734375, 5.895286560058594, -5.920677185058594, 23.30108642578125, 5.7714996337890625, 80.9632568359375, -11.117292404174805, 8.556640625, 56.967926025390625, 49.50715637207031, 56.32333755493164, 119.88333129882812, 15.627593994140625, 11.112590789794922, 18.13287353515625, 132.24163818359375, -79.24142456054688, 80.49505615234375, 4.25408935546875, 38.59088897705078, 14.5252685546875, 70.68960571289062, 113.2510986328125, 68.0869140625, 15.669548034667969, -11.163375854492188, 139.9373779296875, 60.177711486816406, 56.680908203125, 70.57095336914062, -13.1793212890625, 102.61848449707031, -116.0706787109375, 16.804946899414062, 15.265350341796875, 20.306976318359375, 22.744836807250977, -97.58383178710938, 78.02912902832031, -14.85067367553711, 81.75454711914062, 37.03717041015625, 55.18913269042969, 105.803955078125, 0.0, 127.634521484375, -32.274139404296875, 2.292722702026367, 20.960235595703125, -33.63288879394531, 3.3142547607421875, 126.21783447265625, 13.245574951171875, 3.7749786376953125, 22.745750427246094, 92.49713134765625, 44.54341506958008, 36.06883239746094, -10.082603454589844, 53.6492919921875, -2.1331634521484375, 4.503173828125, 3.462371826171875, -0.9501724243164062, 22.203857421875, 58.50566101074219, 126.080810546875, 0.0, 24.4364013671875, 53.22015380859375, 19.50928497314453, 83.75051879882812, -6.912895202636719, 83.79730224609375, 20.1422119140625, 63.093833923339844, -4.103538513183594, 40.0001220703125, 77.69212341308594, 91.90423583984375, -6.58544921875, 56.69224548339844, -9.236480712890625, -0.9520130157470703, -0.707275390625, 39.63006591796875, -13.794036865234375, 22.485626220703125, 45.9195556640625, 12.81219482421875, 71.60675048828125, 31.46978759765625, 73.36319732666016, 20.893157958984375, 96.177734375, -14.275306701660156, 1.4416351318359375, 3.7367115020751953, 6.119720458984375, 15.686080932617188, 18.10778045654297, 10.069061279296875, -1.2623558044433594, 75.93011474609375, 16.376602172851562, 24.48876953125, 21.723190307617188, -24.729248046875, 67.09630584716797, 14.124237060546875, 7.381591796875, -53.737548828125, 15.959083557128906, 0.5498428344726562, -26.355819702148438, 51.781158447265625, 61.340423583984375, 29.743896484375, -10.901325225830078, 89.13263702392578, 47.296546936035156, -38.7059326171875, -6.15728759765625, 51.92974853515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000208.npy"}
{"epoch": 0.4356020942408377, "step": 209, "batch_size": 128, "mean": 22.15717315673828, "std": 47.00916290283203, "min": -104.39739990234375, "p10": -31.108361816406248, "median": 12.968437194824219, "p90": 84.86518249511718, "max": 122.70669555664062, "pos_frac": 0.6328125, "sample": [116.87467956542969, -9.313720703125, 73.44662475585938, 73.20033264160156, 6.643463134765625, -4.440826416015625, 27.43804931640625, -33.218017578125, -35.98675537109375, 18.556060791015625, 80.89813232421875, -12.412467956542969, -2.469970703125, 7.200838088989258, 108.662353515625, 84.52298736572266, 18.804580688476562, 4.767738342285156, 7.853363037109375, 86.63673400878906, 89.49578857421875, -18.741363525390625, 6.7431793212890625, -66.45417785644531, 22.679107666015625, 7.582893371582031, 8.424436569213867, 9.536529541015625, -13.84423828125, 78.40725708007812, 79.43099975585938, -49.5673828125, 0.0645904541015625, -55.79750061035156, -45.84295654296875, -53.87281799316406, 114.88473510742188, 9.15228271484375, 0.0, 10.481048583984375, -30.2042236328125, -45.22810363769531, 58.869293212890625, -30.02880859375, 10.2210693359375, 57.467315673828125, 27.821826934814453, 4.377763748168945, -21.310760498046875, 100.47019958496094, 40.8441162109375, 29.722442626953125, 53.511627197265625, -25.8843994140625, 51.93182373046875, 0.7692527770996094, -14.94744873046875, 12.255111694335938, 69.55460357666016, -13.033758163452148, 20.006011962890625, -27.591171264648438, -6.056243896484375, 82.73306274414062, 84.7781982421875, 23.945335388183594, -10.662139892578125, -24.649658203125, 99.69491577148438, -23.14227294921875, 21.45123291015625, 86.35623168945312, 15.0648193359375, 19.95636749267578, -0.1344451904296875, 0.0, 60.71673583984375, 73.2127685546875, -22.65997314453125, 72.17971801757812, -15.8804931640625, -5.074462890625, 56.34161376953125, -1.88128662109375, 18.755767822265625, 85.06814575195312, 74.71630859375, -52.56636047363281, 13.6817626953125, 7.33538818359375, 23.893417358398438, -62.6875, -20.584747314453125, 63.31230545043945, -2.218740463256836, -3.04644775390625, 28.619384765625, 48.65618133544922, 122.70669555664062, -104.39739990234375, -9.641616821289062, 4.677341461181641, -40.94873046875, -7.647701263427734, -3.2035675048828125, -75.68846130371094, -6.27130126953125, 40.862274169921875, 71.43222045898438, 75.32037353515625, 31.495529174804688, 77.76972961425781, 71.84048461914062, 17.992843627929688, 59.00276184082031, 47.127685546875, 37.12617492675781, 80.951416015625, -1.20819091796875, 112.91717529296875, 33.16259765625, -27.9208984375, 56.859954833984375, 39.01849365234375, 99.33343505859375, 79.4549560546875, -20.79889678955078, 85.547607421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000209.npy"}
{"epoch": 0.437696335078534, "step": 210, "batch_size": 128, "mean": 28.505128860473633, "std": 47.08281326293945, "min": -102.18374633789062, "p10": -18.480375671386717, "median": 22.745415687561035, "p90": 86.60683898925781, "max": 167.42181396484375, "pos_frac": 0.6953125, "sample": [-5.393913269042969, -3.7589340209960938, 39.53411865234375, -7.93115234375, 70.27566528320312, 14.152481079101562, 106.23870849609375, 109.2589111328125, 86.96209716796875, 63.015350341796875, 81.87831115722656, 62.404972076416016, 4.99560546875, 64.35714721679688, 66.25079345703125, -1.2172927856445312, 63.713897705078125, -5.816154479980469, 61.130821228027344, 4.5944976806640625, 22.248409271240234, -1.2015838623046875, -59.60662841796875, 19.499893188476562, 63.42998504638672, 52.14031982421875, 9.10089111328125, 167.42181396484375, -51.4249267578125, -5.6877593994140625, 49.158355712890625, 26.903839111328125, 14.68853759765625, -1.6753997802734375, 73.79391479492188, -13.819732666015625, 5.635763168334961, 52.99041748046875, 0.8818931579589844, 27.28924560546875, 1.36920166015625, 74.50022888183594, 18.490591049194336, -7.1037139892578125, 12.988533020019531, 61.844696044921875, 30.151031494140625, -2.0897216796875, 72.77317810058594, 12.862428665161133, 27.79730224609375, -2.9063720703125, 67.9393310546875, 65.11724853515625, 43.158973693847656, 64.51226043701172, 14.6815185546875, -42.96327209472656, 96.9998779296875, -5.94525146484375, 55.519134521484375, 60.91412353515625, 80.63229370117188, 67.94989013671875, 23.38214111328125, 36.05213928222656, -85.39447021484375, -8.183197021484375, 10.74200439453125, 80.076904296875, -68.12159729003906, -28.971084594726562, 85.50076293945312, 79.75949096679688, 92.89505004882812, 86.95767211914062, 0.08780479431152344, 8.70086669921875, -8.625244140625, -102.18374633789062, 72.26483917236328, 10.787090301513672, -1.3792095184326172, 95.7047119140625, 41.290489196777344, 80.84970092773438, 115.90875244140625, 37.94622802734375, -3.70977783203125, -17.7423095703125, -20.261138916015625, 13.441390991210938, 3.2624473571777344, -26.914840698242188, 23.242422103881836, 59.8590087890625, -20.202529907226562, 113.26495361328125, -17.03985595703125, 52.900177001953125, 57.761199951171875, 119.160400390625, 6.953403472900391, -9.170242309570312, 100.5546875, 23.34154510498047, 92.72514343261719, -15.223068237304688, 12.259552001953125, -17.549774169921875, 59.65986633300781, 23.372833251953125, -3.595855712890625, 18.0867919921875, 16.538803100585938, -95.74795532226562, 14.399375915527344, 86.45648193359375, 28.98345947265625, -76.69699096679688, 70.1249771118164, -4.614654541015625, 73.34014892578125, 51.6829833984375, -27.694061279296875, -8.050048828125, -5.9842529296875, 45.826995849609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000210.npy"}
{"epoch": 0.4397905759162304, "step": 211, "batch_size": 128, "mean": 37.42865753173828, "std": 43.735557556152344, "min": -70.87509155273438, "p10": -10.728253173828124, "median": 40.959869384765625, "p90": 93.81802673339844, "max": 136.70266723632812, "pos_frac": 0.7890625, "sample": [57.80181121826172, 19.573623657226562, 107.772705078125, 57.95207214355469, 73.60758209228516, 90.65704345703125, 12.144134521484375, 6.599678039550781, -12.505996704101562, 61.8748779296875, 85.108154296875, -70.87509155273438, 40.223907470703125, 41.8961181640625, -10.54827880859375, 84.3773193359375, 0.6852569580078125, -0.8084068298339844, -7.031951904296875, -18.239761352539062, 9.264656066894531, 67.0569839477539, 15.356697082519531, 17.3590087890625, 70.84523010253906, 26.722503662109375, 9.016143798828125, -57.43896484375, 70.08523559570312, -11.148193359375, 5.863918304443359, 74.23509979248047, 0.47796630859375, 64.16427612304688, -3.6843643188476562, 53.898590087890625, 45.78028869628906, -70.68476104736328, 50.47943115234375, 78.2266845703125, 5.1263427734375, 56.82490539550781, 7.9500732421875, -1.381256103515625, -3.4176788330078125, 120.6302490234375, 90.37193298339844, 54.611541748046875, 23.035079956054688, 10.039676666259766, 66.77679443359375, -11.251190185546875, 8.3824462890625, 0.147430419921875, 136.70266723632812, 73.58938598632812, 93.86505126953125, -20.174781799316406, 21.554534912109375, 62.956787109375, -8.400390625, -18.119781494140625, 9.907821655273438, 126.6065673828125, -2.3697967529296875, 5.2429656982421875, -5.1221923828125, 72.7520751953125, 93.85321044921875, 15.557186126708984, -24.3525390625, -0.294189453125, -2.1928749084472656, 64.34346771240234, -2.0585384368896484, 11.372540473937988, 73.73155212402344, 63.2381591796875, 72.76873016357422, 63.55926513671875, 52.50921630859375, 56.3817138671875, -0.6776962280273438, 72.115478515625, 66.33786010742188, 124.69822692871094, 14.461090087890625, -0.0440826416015625, 67.72308349609375, 6.354215621948242, -13.89697265625, 122.08892822265625, 100.62701416015625, -32.260650634765625, 98.42803955078125, 18.90784454345703, 87.26254272460938, 51.44500732421875, 11.317218780517578, 2.340423583984375, 70.59197998046875, 105.4212646484375, 41.695831298828125, 6.593891143798828, -67.31703186035156, 3.847393035888672, 44.456939697265625, 42.93914794921875, 101.43914794921875, 55.241729736328125, 71.43060302734375, 12.58587646484375, 86.17041015625, 38.708709716796875, 2.9183998107910156, 33.849029541015625, 105.27947998046875, 93.80294799804688, 64.5362548828125, 67.94728088378906, 25.148956298828125, 54.504676818847656, 0.6701030731201172, 85.607177734375, 53.04138946533203, 54.61676025390625, 1.737762451171875, 84.7789306640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000211.npy"}
{"epoch": 0.4418848167539267, "step": 212, "batch_size": 128, "mean": 26.51364517211914, "std": 47.432289123535156, "min": -121.529541015625, "p10": -29.635101318359368, "median": 20.227951049804688, "p90": 89.29256439208984, "max": 124.30023193359375, "pos_frac": 0.71875, "sample": [3.5872650146484375, -3.681182861328125, 22.6927490234375, 51.86510467529297, -11.916549682617188, -42.410919189453125, -0.6374969482421875, 15.599594116210938, 3.496673583984375, -18.267425537109375, 99.0806884765625, -0.358642578125, 120.40667724609375, 2.9989395141601562, 44.05218505859375, -44.72906494140625, 9.24661636352539, 72.35549926757812, 76.01210021972656, 31.6480712890625, 89.13081359863281, 15.502838134765625, -27.599700927734375, 19.734130859375, 76.2608642578125, 15.91552734375, -7.6624298095703125, -72.5046615600586, 14.116901397705078, 55.30296325683594, -27.214412689208984, -52.869049072265625, 4.504241943359375, 35.810333251953125, 16.005149841308594, 58.0341796875, 23.636199951171875, 92.58747863769531, 46.305511474609375, -5.8916168212890625, 66.83602905273438, -88.47756958007812, -2.759796142578125, -2.4479026794433594, -9.833984375, 75.38888549804688, 73.60107421875, 72.6282958984375, -5.420257568359375, -54.65080261230469, 65.0476303100586, 18.281402587890625, -46.044464111328125, 101.18499755859375, 40.342926025390625, -4.88751220703125, -15.227096557617188, 69.2125244140625, 10.9600830078125, -3.0999412536621094, -7.555229187011719, 6.351728439331055, 74.5191421508789, 34.205902099609375, -9.873233795166016, 73.11775207519531, -11.377227783203125, 58.46351623535156, 48.38728332519531, 89.66998291015625, 41.857574462890625, 76.74832916259766, -56.194976806640625, 24.337615966796875, 81.84306335449219, 13.040069580078125, 52.683284759521484, 72.2713623046875, -1.5336971282958984, 99.96722412109375, 50.461883544921875, 104.5521240234375, 19.551712036132812, 55.1337890625, -107.1292724609375, 21.567657470703125, -2.361907958984375, -58.819732666015625, 21.648544311523438, 7.615970611572266, 76.544189453125, 47.034912109375, 9.883697509765625, 7.260349273681641, 118.8126220703125, 66.61221313476562, 24.1964111328125, 90.01207733154297, 58.81825256347656, 60.30802917480469, 31.2374267578125, 124.30023193359375, 62.30079650878906, 1.4360618591308594, 20.721771240234375, -121.529541015625, 70.75149536132812, 108.89547729492188, -34.384368896484375, -12.37164306640625, 8.706375122070312, -35.505889892578125, 33.29644775390625, 7.577030181884766, 48.134857177734375, -2.12713623046875, 34.28865051269531, 7.926841735839844, 64.87289428710938, 2.1792068481445312, 90.3172607421875, 10.340850830078125, 61.62010192871094, 83.83614349365234, 8.149627685546875, 0.54327392578125, 105.90252685546875, 18.916229248046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000212.npy"}
{"epoch": 0.44397905759162304, "step": 213, "batch_size": 128, "mean": 27.263103485107422, "std": 46.81389236450195, "min": -77.63597106933594, "p10": -25.269921874999998, "median": 14.088031768798828, "p90": 97.04084625244141, "max": 138.8521728515625, "pos_frac": 0.671875, "sample": [102.85360717773438, 72.27568054199219, -18.23450469970703, 84.62696838378906, -35.19219970703125, 68.4140853881836, 75.41195678710938, 71.03738403320312, -3.78955078125, 32.913116455078125, -6.036163330078125, 5.02227783203125, 21.136505126953125, -19.73117446899414, 4.62434196472168, 124.69491577148438, -9.651519775390625, 21.897003173828125, 22.5626220703125, 69.86260986328125, 24.84423828125, -7.79443359375, 134.73226928710938, -6.93243408203125, 97.99983215332031, 6.2462158203125, -5.3299102783203125, 14.947998046875, 17.915618896484375, 14.823249816894531, 6.3860931396484375, -28.90179443359375, -15.5189208984375, 13.3297119140625, 87.74702453613281, 37.30140686035156, 77.10769653320312, -0.663970947265625, 10.204818725585938, -24.23577880859375, 84.91983032226562, 44.22676086425781, 40.439697265625, 76.80758666992188, -23.25714111328125, 78.581298828125, 44.182647705078125, 6.078948974609375, -42.106475830078125, 16.685821533203125, 3.4835128784179688, 67.08804321289062, 97.0069580078125, 1.8565750122070312, 118.37335205078125, -37.26744842529297, -5.222480773925781, 69.5856704711914, -42.16789245605469, 98.23602294921875, 16.326385498046875, 17.993408203125, 25.719200134277344, -30.927139282226562, 9.569091796875, 39.57794189453125, 106.60992431640625, -31.802581787109375, -18.3797607421875, -7.4492340087890625, 72.80020141601562, -16.1485595703125, 25.425384521484375, -77.45880126953125, 24.6221923828125, 17.008895874023438, 51.64910888671875, 94.31488037109375, 8.951911926269531, 35.3653564453125, -18.68585205078125, 7.919412612915039, -2.4414138793945312, -17.35595703125, -2.2308273315429688, 93.4320068359375, -33.96199035644531, -5.49981689453125, 51.362640380859375, 119.21841430664062, 67.92556762695312, 96.314453125, 5.239830017089844, 6.769866943359375, 37.747833251953125, -3.153562545776367, 99.83795166015625, -15.832977294921875, -27.68292236328125, 138.8521728515625, 1.7797126770019531, 10.692977905273438, -4.4637451171875, 91.85712432861328, -33.715362548828125, 109.604736328125, 77.02107238769531, -5.3359375, 51.64463806152344, 0.0, 9.874387741088867, 53.515625, 13.352813720703125, 97.11991882324219, 10.614534378051758, -77.63597106933594, 83.867919921875, 39.572593688964844, -3.6525497436523438, 19.85406494140625, 1.9988975524902344, 76.7266845703125, -28.4169921875, 6.745979309082031, 108.69561767578125, -2.68096923828125, -18.25885009765625, 1.3174972534179688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000213.npy"}
{"epoch": 0.44607329842931936, "step": 214, "batch_size": 128, "mean": 24.430564880371094, "std": 47.3000602722168, "min": -92.77584838867188, "p10": -30.604860687255858, "median": 21.7548828125, "p90": 86.0413818359375, "max": 152.04513549804688, "pos_frac": 0.71875, "sample": [56.055686950683594, 66.29951477050781, -23.386268615722656, 3.90850830078125, 11.611343383789062, 85.95724487304688, 22.503097534179688, -15.568450927734375, -56.03765869140625, -15.325225830078125, 13.217529296875, 63.722137451171875, -30.91680908203125, 34.171600341796875, 47.6583251953125, 69.58119201660156, -92.77584838867188, 73.44245147705078, 34.2576904296875, 46.233306884765625, -4.611686706542969, -15.2169189453125, 19.649581909179688, 62.955413818359375, 37.14044189453125, 22.23480224609375, 64.8635025024414, -22.48321533203125, 9.8466796875, 6.889259338378906, 6.376426696777344, 14.032577514648438, 10.12156867980957, 2.882476806640625, 70.23187255859375, 22.928497314453125, 152.04513549804688, 45.91300964355469, 97.11611938476562, -74.97760009765625, 29.853622436523438, -27.59918212890625, 4.199735641479492, 40.45452880859375, 6.043281555175781, 2.338714599609375, -10.0472412109375, 86.23770141601562, 14.559173583984375, -6.900541305541992, 65.5594711303711, 22.82257080078125, -30.471168518066406, 51.84674072265625, 4.4312591552734375, 80.55360412597656, 29.62451171875, 48.318389892578125, -11.188690185546875, 83.1253662109375, 66.51649475097656, 109.11981201171875, 60.13404846191406, 10.911788940429688, 6.1253662109375, -21.572769165039062, -2.1721343994140625, -43.527488708496094, 1.992523193359375, 0.0, -43.161041259765625, 6.6891326904296875, -12.0067138671875, 12.8675537109375, 66.80538940429688, 0.19683837890625, 53.29644012451172, -57.314998626708984, 6.6282806396484375, 11.700654983520508, -17.738677978515625, -89.50909423828125, 91.989013671875, 3.06536865234375, 68.84078979492188, -32.53240966796875, 93.48233032226562, 54.06077575683594, 23.92840576171875, 14.5416259765625, -86.02645874023438, 43.363037109375, 27.080535888671875, 104.52786254882812, 84.12069702148438, 10.648544311523438, -8.279090881347656, 42.63130187988281, -5.448211669921875, 41.11137390136719, 82.41514587402344, 111.83575439453125, 34.566375732421875, 49.02622985839844, -8.71307373046875, 96.31442260742188, 94.74365234375, 29.19158935546875, -27.598281860351562, 41.06675720214844, 133.36407470703125, 26.879608154296875, -22.064102172851562, -40.12445068359375, -13.0654296875, 91.30477905273438, 27.002775192260742, -76.12617492675781, 15.07513427734375, -0.07916259765625, 21.27496337890625, 57.96757507324219, 25.251190185546875, 53.89515686035156, -34.489906311035156, 146.22186279296875, 50.09466552734375, 58.485107421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000214.npy"}
{"epoch": 0.4481675392670157, "step": 215, "batch_size": 128, "mean": 25.366071701049805, "std": 50.955265045166016, "min": -108.42512512207031, "p10": -40.40336303710937, "median": 21.19200897216797, "p90": 92.95189208984375, "max": 156.50225830078125, "pos_frac": 0.6796875, "sample": [15.475648880004883, 32.73980712890625, 37.24591064453125, 79.516845703125, 110.9862060546875, 72.11508178710938, 83.636474609375, 36.061180114746094, 66.7928466796875, -7.043006896972656, 105.00393676757812, 93.52252197265625, 99.79287719726562, -27.283935546875, 3.324981689453125, 31.5938720703125, -8.298553466796875, 130.72113037109375, 134.94818115234375, 72.13912963867188, 79.04437255859375, 0.656982421875, 0.0, -26.79595947265625, -3.2859115600585938, -9.811325073242188, 43.59002685546875, 70.27914428710938, 128.61355590820312, -10.029960632324219, 81.90869140625, -0.7244720458984375, -16.506866455078125, -46.72804260253906, 2.530071258544922, 10.2257080078125, 82.16938018798828, -0.929656982421875, 23.750289916992188, 124.24783325195312, -41.102935791015625, 28.3846435546875, -27.96038818359375, -33.692047119140625, -0.03265953063964844, -9.474746704101562, -68.31838989257812, 40.245445251464844, 76.95892333984375, 59.34638214111328, 33.576263427734375, 156.50225830078125, 18.018157958984375, 24.14398193359375, -52.570587158203125, 0.13018417358398438, 7.9814300537109375, -40.103546142578125, 0.0, 78.25784301757812, 1.931610107421875, -3.2225494384765625, 45.464599609375, -34.05598449707031, 51.506683349609375, 32.60914611816406, -65.49009704589844, 1.727691650390625, 93.57496643066406, 34.917877197265625, 49.06788635253906, 8.204401016235352, -3.608917236328125, 82.1131591796875, -59.404388427734375, 72.11923217773438, 5.887451171875, -108.42512512207031, 57.54200744628906, 42.8101806640625, 14.890487670898438, 38.921302795410156, 18.63372802734375, -15.539764404296875, -59.66899108886719, -1.302816390991211, 74.87590026855469, 13.533538818359375, 92.70733642578125, -37.686553955078125, -13.390350341796875, 27.20758056640625, 28.323646545410156, -7.287567138671875, 10.740215301513672, 13.025352478027344, 0.3939399719238281, 25.372268676757812, 110.61288452148438, 146.23199462890625, -33.35040283203125, 27.057029724121094, -69.50888061523438, 0.4413566589355469, -46.792694091796875, -52.70330810546875, 50.45338439941406, 85.90528106689453, -60.044097900390625, 80.35089111328125, 14.37152099609375, -7.820507049560547, 45.13705062866211, -1.650543212890625, 12.46258544921875, 28.307525634765625, 27.952957153320312, 82.26634216308594, -50.04498291015625, 26.12689208984375, 45.305870056152344, 3.18316650390625, 15.4422607421875, 95.01275634765625, 23.878387451171875, 65.53607940673828, 26.0260009765625, 82.2059326171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000215.npy"}
{"epoch": 0.450261780104712, "step": 216, "batch_size": 128, "mean": 28.846214294433594, "std": 54.10420608520508, "min": -134.0750732421875, "p10": -42.490188598632805, "median": 24.64447021484375, "p90": 100.94768981933593, "max": 140.04522705078125, "pos_frac": 0.734375, "sample": [36.859161376953125, 54.11323547363281, 62.72418975830078, 32.597694396972656, 16.0972900390625, -38.29527282714844, -134.0750732421875, 20.65753173828125, 67.75520324707031, -2.605022430419922, 76.30110168457031, 88.10079956054688, 55.593597412109375, -10.290992736816406, 20.442306518554688, 59.403717041015625, 140.04522705078125, 9.185619354248047, 101.35919189453125, 113.83833312988281, 23.122772216796875, -84.597412109375, 67.06640625, 0.9210357666015625, -12.152111053466797, 20.72393798828125, -91.95849609375, 13.885360717773438, 54.37467956542969, 82.0377197265625, -53.80049133300781, -47.57069396972656, -99.92327880859375, 1.8292617797851562, 66.68370056152344, 7.032295227050781, 83.88858032226562, -45.045654296875, 108.21145629882812, 74.24266815185547, -12.054302215576172, 85.46723937988281, 117.21087646484375, 77.71239471435547, 82.19485473632812, 60.329315185546875, 35.27215576171875, 37.08013916015625, 5.293128967285156, 0.9730682373046875, -46.313568115234375, -5.95428466796875, 127.8421630859375, 46.02400207519531, 6.614768981933594, 54.963226318359375, 12.185150146484375, 104.70516967773438, -9.735931396484375, 4.270416259765625, 0.0, 66.43276977539062, -22.338104248046875, 57.44140625, 102.90597534179688, 90.9700927734375, 95.579345703125, -8.606903076171875, -4.52392578125, 13.57196044921875, 58.20402526855469, 42.32427978515625, -111.09426879882812, 12.919952392578125, 109.12669372558594, 101.82589721679688, 6.369903564453125, 100.77133178710938, 59.3841552734375, 12.38763427734375, 20.946456909179688, 34.63731384277344, 87.92578125, 26.166168212890625, 108.7449951171875, 71.55766296386719, -52.98152160644531, -8.795783996582031, 106.24615478515625, 98.5986328125, -41.394989013671875, -64.63760375976562, 3.7467041015625, 58.346435546875, -3.606201171875, -17.2529296875, 19.67266845703125, -15.514801025390625, 30.878244400024414, 20.862701416015625, 118.5850830078125, 11.436965942382812, 0.0, 65.0849380493164, 75.53443908691406, -72.67291259765625, 2.2778091430664062, 69.16876983642578, 65.1482162475586, 49.21122741699219, -0.17741966247558594, 50.1614990234375, 79.649658203125, -8.8515625, 43.209205627441406, -6.583526611328125, 30.951568603515625, 48.817413330078125, 5.4722900390625, 50.904998779296875, 1.10150146484375, -102.55471801757812, 58.97613525390625, 78.07345581054688, 5.20733642578125, 2.580078125, -24.4122314453125, 5.3572998046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000216.npy"}
{"epoch": 0.4523560209424084, "step": 217, "batch_size": 128, "mean": 38.45703887939453, "std": 52.92842483520508, "min": -92.051513671875, "p10": -22.24420623779297, "median": 31.223796844482422, "p90": 104.52238616943357, "max": 197.39239501953125, "pos_frac": 0.7734375, "sample": [9.92901611328125, -1.9269790649414062, -4.35736083984375, 76.21270751953125, 102.96247863769531, -75.5416259765625, 2.436981201171875, 94.85458374023438, 6.2577667236328125, 27.81585693359375, 98.35362243652344, -3.013916015625, 2.084716796875, 38.38777160644531, -18.84100341796875, 97.98380279541016, 73.43626403808594, 81.32704162597656, 75.32044982910156, 30.02789306640625, 63.19035339355469, 71.59701538085938, 5.675878524780273, 72.54302978515625, 161.02099609375, 66.07296752929688, 81.35125732421875, 86.73479461669922, 85.18960571289062, 98.19657135009766, 30.91497802734375, 68.00473022460938, -12.800811767578125, 122.9879150390625, 56.58293151855469, 25.55303955078125, 197.39239501953125, 78.046142578125, 92.0391845703125, 45.23944091796875, -91.7552490234375, 89.21212768554688, 38.873443603515625, 145.11541748046875, -76.56031799316406, 50.23822021484375, 13.038169860839844, -25.465354919433594, 41.23065185546875, 19.9591064453125, 2.9770278930664062, 1.1473464965820312, 9.57684326171875, 39.43345642089844, 55.021331787109375, -46.96160888671875, -3.4243392944335938, 86.31010437011719, 62.711448669433594, -15.980911254882812, -2.5466766357421875, 42.90216064453125, -16.692214965820312, 118.82809448242188, 5.475830078125, 29.690223693847656, 12.125804901123047, 96.27938842773438, 71.60550689697266, 9.7232666015625, 73.92466735839844, 78.69652557373047, 31.532615661621094, -3.1474876403808594, 98.36883544921875, -8.070449829101562, -27.506210327148438, -22.199295043945312, 58.856048583984375, 153.97708129882812, 9.441238403320312, 109.35549926757812, 22.605812072753906, -11.318431854248047, -92.051513671875, 120.93296813964844, 108.16217041015625, 110.91778564453125, 14.857513427734375, -60.567962646484375, 15.72686767578125, 88.77908325195312, -12.703125, 81.12176513671875, 3.0630645751953125, 59.8458251953125, 0.228118896484375, 11.831226348876953, 8.313362121582031, 77.19839477539062, -23.846633911132812, 0.0, 17.16741943359375, 40.973175048828125, -28.311004638671875, 124.63864135742188, 77.222900390625, -29.355865478515625, 41.344085693359375, 116.71891784667969, 58.24546813964844, 23.939910888671875, 33.5867919921875, 129.28216552734375, 80.10073852539062, 10.258926391601562, 7.144683837890625, 21.94500732421875, -30.281982421875, 10.089935302734375, 62.58528137207031, 28.682220458984375, 49.35736083984375, 19.064300537109375, 36.5982666015625, -22.3489990234375, -7.7830810546875, 3.979228973388672], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000217.npy"}
{"epoch": 0.4544502617801047, "step": 218, "batch_size": 128, "mean": 30.345848083496094, "std": 47.42803955078125, "min": -93.71377563476562, "p10": -22.991146850585935, "median": 21.369644165039062, "p90": 89.13119354248046, "max": 144.53738403320312, "pos_frac": 0.765625, "sample": [41.41357421875, 14.589859008789062, 3.294036865234375, 94.00704956054688, -4.6619720458984375, -4.0895843505859375, -11.711883544921875, 106.37884521484375, 17.58531951904297, 7.5993804931640625, -67.37921142578125, 9.452072143554688, 41.24183654785156, 60.773468017578125, 46.145355224609375, -93.71377563476562, 8.204833984375, 77.666748046875, 76.53810119628906, 61.20011901855469, -13.164985656738281, 52.911285400390625, 82.1109619140625, 25.93463134765625, 34.742462158203125, 22.544960021972656, 8.745147705078125, -11.357421875, -24.710235595703125, 63.35198211669922, 67.63948822021484, 118.31988525390625, 5.6034088134765625, 86.4339828491211, -22.25439453125, 14.16656494140625, 75.50959777832031, -29.34442138671875, 74.35896301269531, 63.126495361328125, 136.79437255859375, 51.48463439941406, 21.674957275390625, 2.0162105560302734, -25.49688720703125, 90.59535217285156, 8.802448272705078, -62.662811279296875, -47.872467041015625, 86.7177734375, 40.3182373046875, -21.487686157226562, 76.14942169189453, 31.743789672851562, 93.45205688476562, -1.8429450988769531, 88.57598876953125, 21.0643310546875, 86.370361328125, -22.0587158203125, 12.369384765625, -5.192901611328125, -29.673046112060547, 10.984176635742188, 0.0, 60.03509521484375, 18.398590087890625, 0.16162109375, -66.81526184082031, 82.96823120117188, 16.1614990234375, 32.64886474609375, -59.709964752197266, 88.28457641601562, 29.763275146484375, 86.32672119140625, -0.5460739135742188, 13.562911987304688, 24.93731689453125, 1.6540985107421875, 144.53738403320312, 91.54824829101562, 113.4891357421875, 5.12152099609375, -6.454803466796875, 18.4202880859375, -1.09136962890625, 1.12579345703125, 108.63504028320312, 29.88787841796875, 14.842811584472656, 63.999725341796875, 18.35888671875, -67.81912231445312, 78.1958236694336, 73.72101593017578, 89.61946105957031, 88.92193603515625, -40.1820068359375, 44.1748046875, 5.558807373046875, 72.85855102539062, -8.203479766845703, 22.403167724609375, 49.018402099609375, -10.832748413085938, -88.58052825927734, 58.32209777832031, 12.14373779296875, 70.86129760742188, 11.547760009765625, 18.095489501953125, 44.43597412109375, 87.41629028320312, 0.84967041015625, 20.0323486328125, 114.31756591796875, 4.338226318359375, 2.718170166015625, -14.875244140625, 90.94482421875, 63.816993713378906, 62.825164794921875, 16.842926025390625, 44.991424560546875, 81.53082275390625, 22.989425659179688, 8.98876953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000218.npy"}
{"epoch": 0.45654450261780105, "step": 219, "batch_size": 128, "mean": 31.206573486328125, "std": 49.888954162597656, "min": -117.2388916015625, "p10": -17.05099258422851, "median": 23.946670532226562, "p90": 99.82547302246093, "max": 131.67254638671875, "pos_frac": 0.765625, "sample": [-117.2388916015625, 23.890350341796875, 50.32908630371094, 10.2469482421875, 88.828369140625, 31.274887084960938, -14.29522705078125, 60.792510986328125, 85.19522094726562, 27.79888916015625, 56.134490966796875, -1.2283172607421875, 41.0059814453125, -53.490478515625, -0.2706489562988281, 75.44808959960938, 57.85699462890625, -11.393646240234375, 19.397079467773438, 72.47999572753906, 62.27555847167969, 7.567995071411133, 43.825164794921875, 110.27243041992188, -2.6210708618164062, 22.054229736328125, 78.4617919921875, 4.195079803466797, 77.69247436523438, -23.764877319335938, -10.41921615600586, -8.844804763793945, -82.96034240722656, 40.914894104003906, 11.67724609375, 90.33674621582031, -5.6632080078125, 38.04701614379883, 21.8211669921875, 86.45744323730469, 104.84173583984375, 131.67254638671875, 7.660675048828125, 55.9161376953125, 8.019248962402344, -79.06658935546875, 8.994903564453125, 71.21905517578125, 10.083297729492188, 104.01644897460938, 0.0, 24.00299072265625, 18.262008666992188, 26.28302001953125, -52.66557312011719, 103.94491577148438, -24.052337646484375, 98.05999755859375, 107.71804809570312, 15.099899291992188, 54.16447448730469, 50.525634765625, 17.172760009765625, 5.89825439453125, 1.6757783889770508, -10.282867431640625, 5.770721435546875, 6.419849395751953, 2.457305908203125, 67.56254577636719, -13.038787841796875, 22.3310546875, 10.245033264160156, 112.22946166992188, 16.3568115234375, -1.7458953857421875, -82.18556213378906, 11.386932373046875, 58.6829833984375, 93.27212524414062, 2.663423538208008, 92.78855895996094, 24.345947265625, 26.03607177734375, -5.759897232055664, -36.569000244140625, 109.96554565429688, 4.028436660766602, 110.36279296875, -2.359100341796875, 112.5892333984375, 59.231292724609375, 44.66900634765625, 70.15838623046875, 74.21871185302734, 85.52432250976562, 37.601871490478516, 49.323150634765625, 27.75994873046875, 88.96867370605469, -15.316253662109375, 84.6270751953125, 79.63690185546875, -53.94419860839844, 93.33692932128906, 59.23899841308594, 54.249267578125, 3.0662078857421875, 23.57794189453125, 21.20745849609375, 41.497474670410156, -3.9090728759765625, 42.248870849609375, -106.64501953125, 127.98159790039062, 104.346435546875, 42.70689392089844, -21.098716735839844, 14.097259521484375, 12.840301513671875, 49.15618896484375, 18.335494995117188, 59.8135986328125, 12.283851623535156, 128.67730712890625, -10.583074569702148, 2.5287322998046875, -78.06085205078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000219.npy"}
{"epoch": 0.4586387434554974, "step": 220, "batch_size": 128, "mean": 33.31927490234375, "std": 51.616729736328125, "min": -122.30557250976562, "p10": -26.577874755859373, "median": 36.261940002441406, "p90": 103.37741088867188, "max": 137.5919647216797, "pos_frac": 0.7109375, "sample": [98.59927368164062, 110.49661254882812, 68.74266052246094, 62.79802703857422, 22.8942813873291, 68.269775390625, 4.086029052734375, 81.44857788085938, 22.555824279785156, 69.00592041015625, -11.484817504882812, 1.915863037109375, 73.00420379638672, -57.67205810546875, 9.785675048828125, 30.199493408203125, 26.66986083984375, 103.1279296875, -24.51531982421875, 12.865142822265625, 57.83338928222656, -8.98553466796875, 25.574325561523438, 2.296985626220703, 0.2940101623535156, 137.5919647216797, -7.388092041015625, 95.17201232910156, -68.24594116210938, 20.134124755859375, 59.900634765625, 4.044708251953125, -10.12408447265625, -30.40875244140625, 30.17779541015625, 7.485198974609375, -21.36383056640625, 59.503997802734375, 1.8233566284179688, 66.1292724609375, 54.309349060058594, 32.20330810546875, 88.21743774414062, 56.849700927734375, 109.46293640136719, 50.79644775390625, -7.609897613525391, -122.30557250976562, 65.69696044921875, 43.5411376953125, 57.0989990234375, 77.69157409667969, -9.570831298828125, 111.71206665039062, -33.24790954589844, 17.745590209960938, -9.8516845703125, -21.926979064941406, 5.928070068359375, 48.540077209472656, -28.602935791015625, 63.830360412597656, 48.440704345703125, 57.667755126953125, 10.139633178710938, -13.853782653808594, -8.083175659179688, 53.9920654296875, 2.2459259033203125, 84.055419921875, 98.89845275878906, 112.65371704101562, -3.498809814453125, 21.72454833984375, 73.19827270507812, 70.05734252929688, -22.12353515625, 65.5240478515625, 70.5085220336914, -9.872650146484375, -0.43292236328125, 15.904121398925781, 79.78689575195312, 47.2991943359375, -5.1503143310546875, 113.44026947021484, -86.01797485351562, 10.164031982421875, 122.54608154296875, 122.5413818359375, -35.25140380859375, 33.7919921875, -85.09564208984375, -0.7400398254394531, 61.47088623046875, 115.78286743164062, 103.95953369140625, 17.80987548828125, 73.48379516601562, -25.709991455078125, 67.47885131835938, 53.683807373046875, -41.1646728515625, -40.02825927734375, -0.9115486145019531, -15.429275512695312, 75.36932373046875, -33.43565368652344, 109.74615478515625, 104.75184631347656, 48.92216491699219, 43.514373779296875, 58.525390625, -25.561614990234375, 119.6385498046875, -20.4520263671875, 93.24334716796875, 68.75685119628906, 11.81679916381836, 81.88763427734375, -93.02662658691406, 48.9697265625, 69.1742172241211, 52.39671325683594, 81.682373046875, -12.852012634277344, 83.43522644042969, 38.73188781738281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000220.npy"}
{"epoch": 0.4607329842931937, "step": 221, "batch_size": 128, "mean": 25.303386688232422, "std": 50.34452438354492, "min": -111.79055786132812, "p10": -34.954721069335925, "median": 18.2664794921875, "p90": 91.0657958984375, "max": 124.60324096679688, "pos_frac": 0.7265625, "sample": [2.90472412109375, 61.57696533203125, -18.4805908203125, 20.892578125, 41.446533203125, 7.61285400390625, 20.05010986328125, 48.169647216796875, -13.68658447265625, -8.752477645874023, 62.880821228027344, 50.140869140625, 1.27984619140625, 56.5465087890625, -3.6003570556640625, 79.1600341796875, 11.55865478515625, -65.82948303222656, 7.773235321044922, -19.709453582763672, 67.52001953125, 39.44415283203125, 124.60324096679688, -30.748687744140625, 29.498809814453125, 85.88042449951172, 93.75149536132812, 67.648193359375, 86.05259704589844, 0.42529296875, 12.65045166015625, 12.222015380859375, 112.06414794921875, -111.79055786132812, -48.0516357421875, 12.5128173828125, 47.575347900390625, 80.341552734375, -10.870956420898438, 111.62664794921875, -44.768798828125, 9.91900634765625, 1.569976806640625, 28.35321044921875, -69.28861999511719, -6.244720458984375, 86.62918090820312, -63.35711669921875, 79.3062744140625, 33.88311767578125, -0.0054779052734375, 81.34721374511719, 19.3939208984375, -16.607757568359375, -5.8622894287109375, 61.83013916015625, 57.261077880859375, -61.107147216796875, 2.521942138671875, 53.98755645751953, 8.8084716796875, 15.674697875976562, 59.778594970703125, 10.05780029296875, 50.3756103515625, 0.09054374694824219, 62.542694091796875, 81.55258178710938, -2.3725433349609375, 74.8168716430664, 74.25061798095703, -13.2744140625, 54.010772705078125, 58.026817321777344, 100.64155578613281, 60.907470703125, 103.92047119140625, 24.730911254882812, -57.233116149902344, -78.17010498046875, 90.64154052734375, 92.05572509765625, -15.293731689453125, 6.823783874511719, -8.438751220703125, 88.61787414550781, 9.70489501953125, -16.250885009765625, 31.884201049804688, 70.71343994140625, 109.5899658203125, 98.56088256835938, 0.9563102722167969, 7.791120529174805, -50.50811767578125, 5.527919769287109, -15.535125732421875, 114.40321350097656, -98.06494140625, 41.035369873046875, 3.78814697265625, -73.35906982421875, -23.787765502929688, 111.65753173828125, 16.486785888671875, 17.1390380859375, 21.47528839111328, 93.8441162109375, 90.0872802734375, 10.462699890136719, 51.7288818359375, 28.213348388671875, -28.81585693359375, -19.08023452758789, 6.762092590332031, 3.5561752319335938, 46.933685302734375, 28.2506103515625, 77.84149169921875, -5.95196533203125, 49.60206604003906, 79.31793212890625, -23.3590087890625, 93.23605346679688, 28.498977661132812, 10.263677597045898, -93.59280395507812, 11.23291015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000221.npy"}
{"epoch": 0.46282722513089003, "step": 222, "batch_size": 128, "mean": 33.461727142333984, "std": 47.64698791503906, "min": -153.3973388671875, "p10": -18.397096252441404, "median": 24.10428524017334, "p90": 100.72139892578124, "max": 125.140625, "pos_frac": 0.78125, "sample": [60.626068115234375, 14.69158935546875, 71.00901794433594, 73.27552795410156, -23.004180908203125, -26.280105590820312, 41.581756591796875, 49.697227478027344, 16.969573974609375, 8.805351257324219, -28.957061767578125, 18.768310546875, 79.0732421875, 44.5609130859375, 22.960540771484375, 30.01941680908203, 75.54479217529297, 117.19390869140625, 62.554779052734375, 41.192169189453125, 8.287506103515625, 0.4472198486328125, -25.456466674804688, -45.93829345703125, 56.688140869140625, 122.0306396484375, -13.864013671875, 28.35223388671875, -72.35447692871094, 7.840574264526367, 10.914031982421875, 41.91344451904297, 124.87457275390625, 4.352699279785156, 18.18817901611328, 11.368106842041016, 8.740966796875, 61.28131103515625, 76.34870910644531, 33.944091796875, 0.1004791259765625, -57.90660095214844, 71.80191040039062, 3.2767333984375, 61.52235412597656, 7.914356231689453, -57.219093322753906, 58.45293426513672, -153.3973388671875, 44.8809814453125, 95.38848876953125, 67.54972839355469, 0.15074920654296875, 97.84347534179688, -6.196460723876953, 125.140625, 29.203102111816406, 99.96337890625, 1.4112014770507812, 5.123809814453125, 30.9830322265625, 23.270553588867188, 16.914459228515625, 104.91790771484375, -7.741790771484375, 17.64276123046875, 84.01627349853516, 123.65911865234375, 36.6591796875, 63.3526611328125, 92.30836486816406, 81.51223754882812, 1.8426971435546875, 83.67376708984375, -21.51556396484375, 23.647537231445312, 19.007286071777344, 2.033672332763672, 112.2010498046875, -7.864990234375, 7.657413482666016, 24.561033248901367, -7.251993179321289, 66.54202270507812, -18.027877807617188, -1.9704132080078125, 102.4901123046875, -21.82244873046875, -6.713134765625, 50.84690856933594, 85.50558471679688, 82.87002563476562, 27.07894515991211, 42.65662384033203, 5.7322845458984375, -13.213546752929688, 4.142608642578125, 61.89659118652344, 28.13873291015625, 105.990478515625, 112.7867431640625, -8.597763061523438, 2.77154541015625, 38.274261474609375, 105.0806884765625, -4.665962219238281, 89.77886962890625, 50.47857666015625, 17.76806640625, 88.07002258300781, 20.3021240234375, 38.591644287109375, -24.616302490234375, 12.236480712890625, -1.07763671875, -8.677780151367188, -14.97381591796875, 99.80160522460938, -19.25860595703125, 75.0081787109375, 54.64447021484375, 7.66314697265625, 113.10153198242188, 80.22512817382812, 21.580078125, -0.4799175262451172, 13.785964965820312, 112.62274169921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000222.npy"}
{"epoch": 0.4649214659685864, "step": 223, "batch_size": 128, "mean": 37.06852340698242, "std": 52.71433639526367, "min": -96.89083862304688, "p10": -32.72895812988281, "median": 39.416791915893555, "p90": 104.04619598388672, "max": 147.29763793945312, "pos_frac": 0.7578125, "sample": [9.529460906982422, -21.76605224609375, 6.3253326416015625, 81.30807495117188, 11.103012084960938, 18.31317901611328, 1.4256591796875, -12.179189682006836, 33.44086456298828, 95.63536071777344, 29.8167724609375, 97.10134887695312, 30.7330322265625, 7.00762939453125, 59.65155029296875, 136.35162353515625, 101.24745178222656, -95.57708740234375, 76.43377685546875, -36.843292236328125, 29.93878173828125, 117.7432861328125, -2.5843505859375, 47.4459228515625, 46.538299560546875, -34.86381530761719, 15.126815795898438, 10.8463134765625, 111.61126708984375, -16.62188720703125, 22.5113525390625, 65.39251708984375, 103.33998107910156, 96.94915771484375, 77.17672729492188, 90.01872253417969, 135.0784912109375, -96.89083862304688, 61.19245910644531, -60.03654479980469, 16.947738647460938, 59.35369873046875, 6.752349853515625, 70.94010925292969, 101.20086669921875, 84.48126220703125, 109.69256591796875, 17.44476318359375, 118.37176513671875, 22.510772705078125, -1.64825439453125, 26.242156982421875, 123.51376342773438, 122.0684814453125, 80.89094543457031, 49.387725830078125, -0.601470947265625, 106.41162109375, 23.181617736816406, 47.1395263671875, 42.548126220703125, 20.34107208251953, 82.37261962890625, 147.29763793945312, 46.56123352050781, 44.09043884277344, 25.1744384765625, 40.0380859375, 75.83264923095703, -5.1676177978515625, -15.1549072265625, 72.79840087890625, 4.6308441162109375, 93.924072265625, -36.98095703125, 38.79549789428711, -27.09046173095703, -31.603073120117188, 90.6334228515625, -14.568229675292969, 42.26832580566406, -7.191741943359375, 1.0991973876953125, 31.581680297851562, 87.12014770507812, 57.45469665527344, -64.08804321289062, -14.445968627929688, 69.44842529296875, -13.898185729980469, -79.40042877197266, 17.513626098632812, 68.04042053222656, 57.653076171875, 80.81561279296875, 119.0211181640625, 78.13140869140625, 76.11648559570312, 65.4220962524414, 9.82342529296875, -1.568674087524414, 60.042694091796875, 24.319580078125, 69.49856567382812, 53.316375732421875, -33.653411865234375, 84.676025390625, 131.00027465820312, 0.7189216613769531, 14.926620483398438, 105.69403076171875, -55.247520446777344, -1.2960205078125, 8.021202087402344, 94.017333984375, 57.61109924316406, -62.93308639526367, -2.9119873046875, 66.03303527832031, 6.167694091796875, -61.04890441894531, 87.08204650878906, 52.789398193359375, 42.160736083984375, 52.55717468261719, 17.544700622558594, -32.332763671875, -42.625885009765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000223.npy"}
{"epoch": 0.46701570680628274, "step": 224, "batch_size": 128, "mean": 37.76658630371094, "std": 48.398223876953125, "min": -99.55001831054688, "p10": -21.04018630981445, "median": 33.279266357421875, "p90": 101.70240783691405, "max": 141.1253662109375, "pos_frac": 0.796875, "sample": [121.58914184570312, 141.1253662109375, 74.271484375, 53.35430908203125, 4.349981307983398, 4.488864898681641, -21.02386474609375, 46.17436218261719, 80.2230453491211, 17.50720977783203, 5.222808837890625, 72.86609649658203, -2.51397705078125, 71.3111572265625, 27.29145050048828, 31.816558837890625, 9.795654296875, 78.403076171875, 24.634544372558594, -55.73341369628906, 74.07986450195312, 4.454193115234375, 112.68252563476562, 114.98898315429688, 17.2642822265625, 68.31776428222656, 71.73621368408203, 28.297805786132812, 17.742767333984375, 44.2666015625, 112.4730224609375, 83.98556518554688, -99.55001831054688, 46.05964660644531, 36.05915832519531, 20.445693969726562, 35.08580017089844, 73.27098846435547, 44.198516845703125, -4.5440673828125, 39.58958435058594, -19.104400634765625, 80.19302368164062, 66.44142150878906, 59.8092041015625, 33.245635986328125, 56.765899658203125, -49.04728698730469, 100.41213989257812, -0.3825511932373047, 68.94075012207031, 49.2047119140625, 0.0, 104.82135009765625, 84.82847595214844, 12.908649444580078, 8.3853759765625, -26.013748168945312, 5.2187957763671875, 31.497955322265625, 62.00604248046875, 13.526580810546875, 1.7066059112548828, 83.9483642578125, -1.456268310546875, 107.18447875976562, -21.66400146484375, -30.931121826171875, 49.796661376953125, -33.20466613769531, 14.902313232421875, 122.75762939453125, 33.312896728515625, 103.78277587890625, 18.16802978515625, -9.127593994140625, -33.66131591796875, 66.1875, 16.127670288085938, 109.70462036132812, -10.484848022460938, -13.874153137207031, -88.14376831054688, 10.250263214111328, 12.920936584472656, 90.36541748046875, 48.102264404296875, -19.142120361328125, 100.81082153320312, -78.24406433105469, 89.22816467285156, 84.34756469726562, 42.735382080078125, 97.44482421875, 29.39056396484375, 60.294952392578125, 100.68490600585938, 52.972076416015625, 36.482666015625, 6.058006286621094, -31.491737365722656, 25.580490112304688, 5.525482177734375, 17.4326171875, 26.494491577148438, 55.365325927734375, 140.92794799804688, -21.078269958496094, 106.95205688476562, 74.22299194335938, 18.644744873046875, -36.61439514160156, 10.177957534790039, 2.727783203125, 5.38604736328125, 30.580215454101562, 90.17889404296875, 8.897308349609375, 67.31869506835938, -17.77911376953125, 7.416133880615234, 95.24679565429688, 76.4833984375, 73.91409301757812, 95.03593444824219, 74.00334167480469, -2.20233154296875, 115.32479858398438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000224.npy"}
{"epoch": 0.46910994764397906, "step": 225, "batch_size": 128, "mean": 26.47385025024414, "std": 57.99826431274414, "min": -120.31640625, "p10": -48.08771133422851, "median": 21.821075439453125, "p90": 107.470654296875, "max": 141.23583984375, "pos_frac": 0.625, "sample": [107.83270263671875, -1.46148681640625, 35.483154296875, 105.22360229492188, 99.17640686035156, 6.1107635498046875, -71.4154052734375, -70.272705078125, 67.25341796875, 107.31549072265625, 26.741546630859375, 68.07489013671875, 9.175537109375, -8.080108642578125, 115.48013305664062, 36.71917724609375, 124.67086791992188, 74.82028198242188, 4.058982849121094, 21.739730834960938, -13.038360595703125, -13.53131103515625, 69.86395263671875, -33.811309814453125, 98.32522583007812, 44.579864501953125, 107.03134155273438, -9.608688354492188, 2.03814697265625, -33.587646484375, -10.947715759277344, -120.31640625, 81.22361755371094, 102.66485595703125, 13.9306640625, 10.112922668457031, 66.74093627929688, -1.3695869445800781, -0.83477783203125, -50.86772155761719, 67.96743774414062, 117.5589599609375, -11.352546691894531, -63.54890441894531, -87.99099731445312, -65.08063507080078, 45.000579833984375, 21.87310791015625, -9.426593780517578, 48.699920654296875, -77.5733642578125, 21.76904296875, 70.29574584960938, 88.2783203125, -12.0902099609375, 16.748779296875, -32.487579345703125, 34.1981201171875, 99.28482055664062, 76.75416564941406, 82.58143615722656, 114.52799987792969, 124.17474365234375, -22.36944580078125, 27.8778076171875, -10.516281127929688, -38.008636474609375, 75.703369140625, 48.374855041503906, 132.18069458007812, 0.674102783203125, 109.38629150390625, 28.72174072265625, 54.149085998535156, -6.144561767578125, 40.53361511230469, 96.1839828491211, 121.029296875, -23.772216796875, 83.52362060546875, -38.17364501953125, -56.48101806640625, -46.896278381347656, 39.56233215332031, -72.71287536621094, -9.16107177734375, 10.512811660766602, -0.21515274047851562, 43.58399963378906, -101.68588256835938, -4.548126220703125, -29.08660888671875, -12.7679443359375, -13.2418212890625, 24.46685791015625, -28.725341796875, -0.2708282470703125, 47.064476013183594, 141.23583984375, 63.10545349121094, 5.858146667480469, 58.20549774169922, 116.87094116210938, 104.18060302734375, -33.1685791015625, 13.361984252929688, -65.00666809082031, 56.38221740722656, 12.478912353515625, 127.9617919921875, -18.429039001464844, 52.563446044921875, 69.84823608398438, 129.43943786621094, 9.93658447265625, 65.28765869140625, 54.877315521240234, -31.984375, 61.165592193603516, -51.94065856933594, 23.513900756835938, 2.206634521484375, -7.800041198730469, -6.66119384765625, -23.324676513671875, 27.094131469726562, 34.28813171386719, 90.947265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000225.npy"}
{"epoch": 0.4712041884816754, "step": 226, "batch_size": 128, "mean": 24.867389678955078, "std": 50.215641021728516, "min": -95.48529052734375, "p10": -38.19666595458984, "median": 14.757316589355469, "p90": 93.76844482421875, "max": 129.048828125, "pos_frac": 0.7265625, "sample": [18.17413330078125, 129.048828125, -18.931884765625, 62.366241455078125, 90.983154296875, 104.72479248046875, 6.5767822265625, 0.2823009490966797, 68.44793701171875, 82.93521118164062, 108.25083923339844, 32.7681770324707, 5.5028076171875, 108.35231018066406, -5.183502197265625, 13.38192367553711, 0.7429962158203125, -65.65377807617188, 89.27125549316406, 99.69700622558594, 3.1335906982421875, 1.728912353515625, 83.73947143554688, 64.3285903930664, 4.855201721191406, 21.0010986328125, -9.70018196105957, -46.7901611328125, 77.89231872558594, 61.410400390625, 1.741159439086914, 47.0126953125, 55.48773193359375, 107.87115478515625, -0.4656829833984375, 15.24786376953125, 90.83773040771484, -42.61383056640625, -5.607757568359375, 55.05445098876953, -73.13531494140625, 57.29151153564453, -26.505874633789062, 0.2132854461669922, -13.30950927734375, 57.174560546875, 13.560699462890625, -10.732696533203125, -95.48529052734375, 48.865692138671875, 29.78515625, -49.518798828125, 89.76278686523438, 2.5739173889160156, 9.593315124511719, -69.37026977539062, 2.6005325317382812, 26.59625244140625, 1.6639137268066406, -0.4123382568359375, -7.4810791015625, 15.94244384765625, 5.2645416259765625, -88.91632080078125, 23.608108520507812, -8.912979125976562, 13.351882934570312, 94.76100158691406, 6.722591400146484, 10.83111572265625, -22.891693115234375, 6.459009170532227, 80.27566528320312, -23.08477783203125, 61.432029724121094, 3.02496337890625, 17.890884399414062, 91.94357299804688, 5.775543212890625, 109.23580932617188, -38.701446533203125, 91.9081039428711, 119.64486694335938, -62.44904327392578, 103.92697143554688, 8.39398193359375, 18.465179443359375, 3.5327377319335938, 35.24224853515625, 82.3594970703125, 64.85092163085938, 31.44183349609375, 86.62971496582031, 49.082794189453125, 80.45037841796875, 14.266769409179688, -18.096115112304688, 0.075592041015625, 71.73624420166016, 32.022186279296875, 111.51959228515625, 17.29205322265625, 104.16055297851562, 0.9241943359375, 6.16009521484375, 21.078857421875, 93.34306335449219, 35.763336181640625, 64.83634948730469, -22.67352294921875, 30.4532470703125, -21.03466796875, -1.3178482055664062, 68.615478515625, 41.266334533691406, -22.968544006347656, 41.68096923828125, -28.925460815429688, -24.258575439453125, 116.18603515625, 62.4769287109375, -43.756614685058594, -70.96675109863281, 17.090511322021484, -67.8253173828125, -37.98033142089844, 13.557098388671875, -10.800735473632812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000226.npy"}
{"epoch": 0.4732984293193717, "step": 227, "batch_size": 128, "mean": 37.63871765136719, "std": 52.62173080444336, "min": -115.4873046875, "p10": -28.418579101562496, "median": 35.254451751708984, "p90": 107.06188812255859, "max": 147.39547729492188, "pos_frac": 0.765625, "sample": [-62.27348327636719, 37.500946044921875, 19.421966552734375, 28.895156860351562, 64.78645324707031, 35.89556121826172, 115.33499145507812, 17.147201538085938, 8.107742309570312, 38.54723358154297, 102.73397827148438, 82.53730773925781, 13.178848266601562, -2.48956298828125, 26.155136108398438, 55.348907470703125, 0.0, 110.603515625, -101.5706787109375, -30.919498443603516, 66.5841064453125, -59.0367431640625, 83.52595520019531, 101.24150085449219, 137.77267456054688, 147.39547729492188, 106.6463623046875, -87.4947509765625, 92.5286636352539, -19.5408935546875, -7.1147308349609375, 91.34925842285156, 12.459014892578125, 13.470046997070312, 39.87005615234375, -11.033935546875, 39.510032653808594, 69.60466766357422, 37.134490966796875, 114.5025634765625, 116.8707275390625, 69.10052490234375, 14.354461669921875, 69.32669067382812, 36.298248291015625, -39.116485595703125, 18.658447265625, 50.16583251953125, -18.20013427734375, 34.31939697265625, 104.43350219726562, 61.44312286376953, -9.830322265625, -27.85552978515625, 59.99737548828125, 130.55841064453125, 6.49853515625, 72.58765411376953, 108.03144836425781, 27.667034149169922, 109.28184509277344, -44.09759521484375, 9.37548828125, 81.1998291015625, -115.4873046875, 23.329345703125, 12.122642517089844, 123.05538940429688, -44.798370361328125, 40.404296875, 73.45159912109375, 31.393096923828125, 3.846975326538086, 80.90583801269531, 105.32369995117188, 54.6932373046875, 76.6783447265625, 72.01580810546875, -35.7452392578125, 75.84130859375, 113.27777099609375, 7.02362060546875, 21.847259521484375, 103.91508483886719, 40.532012939453125, 28.405624389648438, 53.837127685546875, 30.958709716796875, 29.0858154296875, 75.0858154296875, -29.73236083984375, 65.54197692871094, 13.406524658203125, -59.80748748779297, 89.59197998046875, 42.824005126953125, 83.73422241210938, 14.258514404296875, -17.88396453857422, 96.33795928955078, -0.70880126953125, 120.14361572265625, -2.7046070098876953, 108.913818359375, -49.665672302246094, 95.26040649414062, -2.328805923461914, -1.677215576171875, 3.3692169189453125, 0.2429656982421875, 90.91790771484375, 23.909866333007812, 92.26902770996094, 91.91780090332031, 12.964202880859375, 30.847320556640625, 34.61334228515625, 52.9383544921875, 89.01007080078125, -4.32037353515625, -15.60211181640625, -23.209152221679688, 24.6973876953125, 11.238975524902344, -13.354324340820312, 5.398824691772461, 38.885902404785156, 55.1331787109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000227.npy"}
{"epoch": 0.47539267015706804, "step": 228, "batch_size": 128, "mean": 37.832244873046875, "std": 51.47672653198242, "min": -72.22135162353516, "p10": -23.601802825927734, "median": 29.167583465576172, "p90": 107.37861022949218, "max": 135.20626831054688, "pos_frac": 0.71875, "sample": [-63.21441650390625, -12.339988708496094, 62.1915283203125, 93.1219482421875, 46.98216247558594, -5.275382995605469, 123.44500732421875, 29.40362548828125, 61.511505126953125, 92.38934326171875, 3.985015869140625, -9.062934875488281, -10.707962036132812, 109.3369140625, 47.811676025390625, 66.17141723632812, -6.84326171875, 83.17048645019531, -14.087539672851562, 8.559249877929688, 81.94523620605469, 22.427505493164062, -72.22135162353516, -15.795745849609375, 15.830978393554688, 79.61456298828125, 25.203536987304688, 0.7720184326171875, -19.2535400390625, 60.54508972167969, 13.509796142578125, 24.939697265625, -39.04499816894531, 101.96673583984375, 35.726226806640625, 95.32106018066406, 62.50128173828125, 70.66525268554688, 12.18560791015625, -5.5472259521484375, 74.75344848632812, -6.82366943359375, -18.576004028320312, -24.89892578125, 76.97015380859375, 11.153594970703125, -5.659027099609375, 16.53374481201172, 45.29559326171875, 74.76321411132812, 41.34930419921875, 119.65899658203125, 61.54100036621094, 19.30170440673828, 102.46768188476562, 33.91522216796875, 78.5333251953125, 79.20452880859375, 113.53558349609375, 115.0753173828125, -1.0446319580078125, -61.48724365234375, -24.434104919433594, -0.0975341796875, 74.83268737792969, 17.940597534179688, 0.0, 23.720062255859375, 49.482452392578125, 18.407325744628906, 12.992240905761719, 133.70016479492188, 15.707267761230469, -6.38165283203125, -35.3614501953125, 90.03839111328125, -17.5263671875, 100.78280639648438, -14.146377563476562, 106.53933715820312, -16.115936279296875, -59.34358215332031, 18.26519775390625, 100.149658203125, 98.78947448730469, 112.77920532226562, -2.1505126953125, 3.839813232421875, 116.22212219238281, 77.015869140625, -23.245101928710938, 62.75347900390625, 16.11053466796875, 89.29116821289062, 43.985252380371094, 68.57511901855469, -67.06842041015625, 4.502281188964844, 21.299304962158203, 101.86149597167969, 102.57815551757812, -26.536087036132812, -30.877960205078125, 32.15296936035156, 80.092529296875, 59.2374267578125, 94.42636108398438, -52.35784912109375, -15.704925537109375, -0.397613525390625, 19.336013793945312, 114.36614990234375, 0.8742523193359375, 80.30265808105469, 94.10745239257812, 116.9925537109375, 126.47235107421875, 67.18830108642578, 26.42715835571289, -49.88653564453125, 76.67251586914062, 78.55499267578125, 118.40179443359375, 35.471435546875, 135.20626831054688, 6.100608825683594, 7.2792205810546875, 28.931541442871094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000228.npy"}
{"epoch": 0.4774869109947644, "step": 229, "batch_size": 128, "mean": 46.92871856689453, "std": 54.481414794921875, "min": -88.16377258300781, "p10": -18.571059799194334, "median": 49.355743408203125, "p90": 113.60673828124999, "max": 165.31192016601562, "pos_frac": 0.8125, "sample": [78.0325927734375, -11.55535888671875, 91.39044189453125, 71.87469482421875, -7.82769775390625, -87.78498840332031, 59.981346130371094, 88.95097351074219, 39.64727783203125, -57.358062744140625, 146.10223388671875, 116.24539184570312, 82.25424194335938, 1.4996757507324219, 127.98464965820312, 96.96690368652344, 14.921829223632812, 7.51202392578125, 20.86428451538086, 39.62158203125, 39.34315490722656, 22.2142333984375, 96.83749389648438, -10.08013916015625, 63.403236389160156, -66.43006896972656, 11.476455688476562, 15.716773986816406, 110.15872192382812, 76.87385559082031, 11.413970947265625, 48.831329345703125, 116.06964111328125, 68.22900390625, 20.285858154296875, 121.03436279296875, -1.389892578125, 19.346923828125, 139.56301879882812, -88.16377258300781, 6.47900390625, -55.4886474609375, 53.75297546386719, 93.80633544921875, 63.62642288208008, 134.30191040039062, 165.31192016601562, 104.20787048339844, 59.95652770996094, 90.09593200683594, 41.467681884765625, -1.018402099609375, 42.96617126464844, 5.329986572265625, 82.2710189819336, 105.33721923828125, 92.37167358398438, 17.948211669921875, 12.578567504882812, -21.60247802734375, 56.19513702392578, -68.84727478027344, 98.47998046875, 73.0069580078125, -9.661430358886719, -8.527313232421875, 98.5804443359375, 15.405876159667969, 74.50525665283203, -3.21649169921875, 2.194732666015625, 52.42949676513672, 1.1475391387939453, -18.915374755859375, 96.8038330078125, 48.369285583496094, 44.285850524902344, 91.23270416259766, 109.94378662109375, 43.6260986328125, 106.67744445800781, 24.163406372070312, 36.638763427734375, 94.87477111816406, 50.502685546875, 12.5223388671875, -32.50474548339844, 89.60211181640625, 123.1546630859375, 62.144996643066406, 133.74159240722656, 95.97355651855469, 12.544645309448242, 49.946044921875, 14.119171142578125, 102.52326965332031, 41.440093994140625, 111.83642578125, 45.618743896484375, 21.776809692382812, 85.35336303710938, 13.198204040527344, 43.3477783203125, 117.43026733398438, -5.354339599609375, 14.741180419921875, 51.77973937988281, -17.493057250976562, 10.132808685302734, 112.55120849609375, 59.90447998046875, -18.42349624633789, 30.833343505859375, 49.880157470703125, 77.14239501953125, 86.04046630859375, 88.5355224609375, -36.1392822265625, 86.69209289550781, -70.9979248046875, 126.41749572753906, 62.63946533203125, 73.51112365722656, 25.544326782226562, 93.3377685546875, -57.58962631225586, 135.35635375976562, -53.589622497558594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000229.npy"}
{"epoch": 0.47958115183246075, "step": 230, "batch_size": 128, "mean": 34.8034553527832, "std": 55.14002227783203, "min": -164.74765014648438, "p10": -20.444893646240235, "median": 27.281064987182617, "p90": 110.14968872070312, "max": 150.28884887695312, "pos_frac": 0.7421875, "sample": [110.0853271484375, 33.42547607421875, 31.0919189453125, 89.11962127685547, 149.43719482421875, 68.89242553710938, 114.15895080566406, -5.768585205078125, 110.29986572265625, 29.864974975585938, 15.915908813476562, 26.9849853515625, 18.927623748779297, -70.07662200927734, -164.74765014648438, 90.21397399902344, 93.66053771972656, -2.108133316040039, -3.317657470703125, -54.48663330078125, 6.16741943359375, 11.889240264892578, 115.14276123046875, -40.155364990234375, 0.0, 98.02969360351562, -2.083892822265625, 18.615516662597656, 64.96298217773438, 94.24629211425781, 46.66770935058594, 94.04135131835938, 5.66278076171875, 38.98822021484375, -0.18080902099609375, -71.54559326171875, 61.48687744140625, 86.78782653808594, 67.2528076171875, -21.622451782226562, -0.03764533996582031, -73.17572021484375, 15.219192504882812, 28.626312255859375, 27.577144622802734, 101.7918701171875, 65.06597900390625, -6.5106201171875, 89.73126220703125, -20.179397583007812, -4.926120758056641, 56.04937744140625, -34.335845947265625, 58.908721923828125, 96.35610961914062, -14.79815673828125, 37.69279479980469, 11.58514404296875, 97.10830688476562, 25.4617919921875, -70.72760772705078, 71.03225708007812, 124.38034057617188, 78.73920440673828, 15.876655578613281, 5.16943359375, -15.03192138671875, 90.41014099121094, -42.00404357910156, 71.30975341796875, 45.54986572265625, 6.776702880859375, -13.608985900878906, -12.722885131835938, 55.08575439453125, 17.059478759765625, 9.991943359375, 55.748409271240234, 125.17440795898438, -6.6802215576171875, 3.036996841430664, 70.01654052734375, 26.125823974609375, -21.715593338012695, 15.829879760742188, -121.50465393066406, 0.35671234130859375, -21.06438446044922, 125.89825439453125, 67.29763793945312, 133.879150390625, 60.051456451416016, 13.148841857910156, 2.4471588134765625, 110.85317993164062, 108.12704467773438, 1.5581321716308594, 29.82110595703125, 90.60131072998047, 70.64793395996094, 104.28228759765625, 5.281057357788086, 76.69480895996094, 9.402204513549805, -13.308746337890625, 84.28450012207031, -1.6651611328125, 118.0008544921875, 74.15284729003906, 6.6866912841796875, 76.14225769042969, 112.31785583496094, 67.35269165039062, 2.142131805419922, 41.754638671875, 33.049354553222656, 4.6515960693359375, 123.25297546386719, 150.28884887695312, -0.227783203125, 1.4727020263671875, 31.48968505859375, 2.0804443359375, 9.689910888671875, 26.498291015625, 36.533905029296875, -0.9918212890625, -16.545379638671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000230.npy"}
{"epoch": 0.4816753926701571, "step": 231, "batch_size": 128, "mean": 29.024816513061523, "std": 50.30856704711914, "min": -101.47288513183594, "p10": -37.899044799804685, "median": 24.30206298828125, "p90": 91.04814910888672, "max": 136.9248046875, "pos_frac": 0.7421875, "sample": [76.99369812011719, 18.349273681640625, -37.600555419921875, -59.9925537109375, 78.84183502197266, 75.65144348144531, 53.3782958984375, -5.5246429443359375, 70.65130615234375, -101.47288513183594, 98.43022155761719, -34.56085205078125, 61.151580810546875, 65.5670166015625, -5.065582275390625, 96.00308227539062, 9.595840454101562, 9.411548614501953, 67.49324035644531, -2.4694061279296875, -23.912811279296875, -63.94435119628906, -12.4415283203125, 47.98480224609375, 10.06298828125, 36.18231201171875, 12.130157470703125, 76.2835464477539, 1.9556884765625, -52.248199462890625, 27.6082763671875, 85.72068786621094, 90.73886108398438, 2.986967086791992, 44.750030517578125, -10.734130859375, 118.61495971679688, 45.31353759765625, 16.462875366210938, 41.64569091796875, -25.801071166992188, 36.9000244140625, -3.249420166015625, 0.8412094116210938, 102.75175476074219, 87.70816040039062, -6.67071533203125, 49.13690185546875, 48.11236572265625, 31.888946533203125, 88.00689697265625, -49.19549560546875, 21.06536865234375, 16.990570068359375, 3.711761474609375, 15.98565673828125, 85.282470703125, -19.06005859375, 29.967269897460938, 136.9248046875, -52.178985595703125, 4.760341644287109, 2.055572509765625, 49.34307861328125, 2.5949134826660156, 79.58981323242188, 51.9327392578125, 25.011505126953125, 19.59454345703125, 51.0146484375, 29.27886962890625, 51.698974609375, 128.169677734375, -27.001739501953125, -84.55189514160156, -0.76458740234375, 49.264137268066406, -8.461700439453125, 82.99853515625, -0.227020263671875, 18.188720703125, 17.884124755859375, 35.84898376464844, 62.610870361328125, 7.379371643066406, 21.20233154296875, 7.91619873046875, 101.08627319335938, 34.349945068359375, 9.509384155273438, 48.241241455078125, 69.34007263183594, 69.50331115722656, 23.592620849609375, -46.29029846191406, 68.25791931152344, 12.019977569580078, 66.4796142578125, 127.2210693359375, 0.6896286010742188, -72.6298828125, 22.70556640625, 56.63905334472656, 131.68780517578125, 79.44105529785156, -48.13623046875, -1.4447555541992188, 32.264129638671875, 17.219207763671875, 67.49332427978516, -4.870208740234375, 53.296142578125, 30.922225952148438, -85.59585571289062, 83.13827514648438, 128.84967041015625, 13.082672119140625, 107.06619262695312, 11.94158935546875, 54.24897766113281, -17.525955200195312, 4.701019287109375, 91.76982116699219, -38.59552001953125, 79.92926025390625, 0.0, -83.54209899902344, 110.67864990234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000231.npy"}
{"epoch": 0.4837696335078534, "step": 232, "batch_size": 128, "mean": 34.04143524169922, "std": 53.635231018066406, "min": -85.65847778320312, "p10": -29.389501190185538, "median": 23.39340591430664, "p90": 101.35267181396485, "max": 163.10336303710938, "pos_frac": 0.6796875, "sample": [-24.0836181640625, 98.05575561523438, -2.8292617797851562, 110.49234008789062, 30.69085693359375, 69.22761535644531, 87.7979736328125, 46.98895263671875, -2.196624755859375, -9.170181274414062, 78.44651794433594, 96.51019287109375, -0.6056976318359375, -23.841888427734375, -36.32911682128906, 27.613243103027344, 7.041423797607422, 3.9896621704101562, 23.26641845703125, 19.987777709960938, 25.7052001953125, 11.171852111816406, -85.65847778320312, 67.78005981445312, 95.55007934570312, 98.08203125, 22.60222625732422, 57.655426025390625, -0.8092422485351562, 129.2353515625, 38.55780029296875, -17.613121032714844, -38.460205078125, 96.72085571289062, 77.34506225585938, 94.22500610351562, -57.6181640625, 37.25640869140625, 17.9356689453125, 97.52658081054688, 65.09255981445312, -34.853492736816406, -0.6133079528808594, 23.52039337158203, 10.772331237792969, 0.9672698974609375, 77.36769104003906, 13.3310546875, 98.79704284667969, 0.738739013671875, 29.461883544921875, -24.377960205078125, 88.65106201171875, 27.6160888671875, 4.021697998046875, 76.36178588867188, 18.83953857421875, 93.08391571044922, -1.2698822021484375, 81.81468200683594, 19.030189514160156, 131.92523193359375, 102.02247619628906, -27.04779052734375, 87.3411865234375, 92.51799011230469, -4.29852294921875, 16.336395263671875, -5.92529296875, -10.600120544433594, 12.403411865234375, 110.94180297851562, 10.775909423828125, -35.616943359375, 121.97541809082031, 119.3310546875, -0.391326904296875, -24.415985107421875, 49.047637939453125, 29.87115478515625, 0.8385353088378906, 8.089385986328125, -20.047119140625, 87.0740737915039, 102.49688720703125, -18.576732635498047, -6.210235595703125, 72.36685180664062, -6.800506591796875, -35.33642578125, 85.9619140625, 66.13674926757812, 0.0, 111.756103515625, 49.98773956298828, 4.337257385253906, 53.68951416015625, 28.988388061523438, 93.2523422241211, 67.18511962890625, 86.71150970458984, 18.335800170898438, 127.19369506835938, -77.2060546875, 77.41798400878906, 163.10336303710938, 11.20086669921875, -20.32672119140625, 92.28263854980469, 64.17396545410156, -4.17742919921875, 112.85507202148438, -75.83590698242188, -13.023574829101562, -19.76952362060547, -50.13360595703125, 38.166297912597656, -36.152957916259766, 74.34201049804688, -40.37919616699219, 101.06561279296875, -3.47271728515625, -84.33100891113281, 123.75, 61.43208312988281, 74.60498809814453, 17.796417236328125, -20.2996826171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000232.npy"}
{"epoch": 0.48586387434554973, "step": 233, "batch_size": 128, "mean": 35.497745513916016, "std": 52.914878845214844, "min": -87.95472717285156, "p10": -27.459577941894526, "median": 26.057899475097656, "p90": 103.89000701904297, "max": 158.8201904296875, "pos_frac": 0.78125, "sample": [-43.38743591308594, 93.89290618896484, 1.010833740234375, -3.11614990234375, 10.731727600097656, 103.33578491210938, 37.43646240234375, 31.675750732421875, -2.8237152099609375, 13.323631286621094, 17.0133056640625, 27.565311431884766, 113.23123168945312, 78.47219848632812, 18.743892669677734, -2.577301025390625, 23.8282470703125, 101.2755126953125, 36.310821533203125, 151.82711791992188, 94.74807739257812, 88.71925354003906, 99.08576965332031, 90.08073425292969, 59.06853485107422, 103.90437316894531, 93.921630859375, 8.68572998046875, 88.20907592773438, -25.89520263671875, 30.97576904296875, -17.51958465576172, 86.29554748535156, 3.4898605346679688, 110.55235290527344, 43.840423583984375, 22.844131469726562, -72.25424194335938, 72.59967041015625, 121.62953186035156, -47.284950256347656, -4.078960418701172, 3.53021240234375, 23.287109375, -22.6312255859375, 139.10302734375, 62.42219543457031, 16.611724853515625, 18.321197509765625, 27.856895446777344, 110.3985595703125, 130.22352600097656, 17.626373291015625, -13.107391357421875, 36.370330810546875, -12.027252197265625, 142.030029296875, 31.2825927734375, 96.05390930175781, 9.79693603515625, 66.87139892578125, 92.43426513671875, -45.398040771484375, 22.343441009521484, 8.168777465820312, -45.01190185546875, -31.094833374023438, 57.67245101928711, -81.29129028320312, -73.93130493164062, 41.42626953125, 103.88385009765625, 117.78656005859375, 55.13017272949219, 137.47552490234375, 51.650054931640625, 18.700958251953125, 11.13934326171875, 14.826623916625977, 2.824981689453125, 2.6967697143554688, 7.0712738037109375, 16.165069580078125, 53.88932800292969, -40.648956298828125, 6.138330459594727, 158.8201904296875, 54.450927734375, -9.016876220703125, 40.4052734375, 24.674823760986328, 1.4508209228515625, -42.3319091796875, 127.96728515625, 22.71076202392578, -25.901611328125, 1.3436813354492188, 81.73397827148438, -2.4903564453125, 97.72593688964844, -14.284423828125, 6.123451232910156, -11.128067016601562, -87.95472717285156, 73.14251708984375, 101.73268127441406, 29.663436889648438, 76.08472442626953, 47.64776611328125, -72.77703857421875, 1.045684814453125, 86.2679443359375, 48.952667236328125, -3.1906280517578125, 33.84100341796875, 26.281494140625, -33.037933349609375, 94.25314331054688, 53.43763732910156, 46.256805419921875, 48.82973098754883, 4.362789154052734, 3.60247802734375, 78.68063354492188, 62.9068603515625, 17.361602783203125, 25.834304809570312, 20.77630615234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000233.npy"}
{"epoch": 0.48795811518324606, "step": 234, "batch_size": 128, "mean": 37.25542449951172, "std": 62.43062210083008, "min": -137.3458251953125, "p10": -43.043078613281246, "median": 33.13138198852539, "p90": 112.52980041503906, "max": 178.42947387695312, "pos_frac": 0.765625, "sample": [-47.949188232421875, -97.85150146484375, -50.70257568359375, 0.26506614685058594, 45.835693359375, 108.93807983398438, 113.96267700195312, 47.618927001953125, -26.72412109375, 132.04598999023438, 99.80418395996094, -94.21670532226562, 95.11557006835938, 88.97073364257812, -24.875518798828125, 97.22622680664062, 34.62799835205078, 43.30120849609375, 76.68608093261719, 119.07598876953125, 83.07110595703125, 62.6297607421875, 75.2164306640625, 127.2996826171875, 15.828399658203125, 5.0324554443359375, 88.581298828125, 82.22747802734375, 87.35575866699219, 52.3685302734375, 22.030471801757812, 55.21453857421875, 100.90289306640625, 129.37725830078125, 85.16986083984375, 21.09246826171875, 31.634765625, 7.594757080078125, 2.019866943359375, 51.06011962890625, 60.6070556640625, 20.9072265625, 6.379302978515625, -69.2701416015625, 22.561904907226562, -19.1046142578125, 82.29566955566406, 83.37625122070312, -4.090824127197266, 101.81514739990234, 8.598419189453125, 3.140625, 130.79653930664062, 60.349613189697266, -10.801841735839844, 152.167236328125, 87.8636474609375, 76.1847152709961, 92.1226806640625, 30.530670166015625, 11.505653381347656, 6.607765197753906, 117.07376098632812, 10.155303955078125, 8.170991897583008, -2.9886474609375, -32.29771423339844, 0.0, 25.375732421875, -12.861495971679688, 13.897811889648438, 44.09266662597656, -91.49177551269531, 1.9070663452148438, 97.0693359375, 139.16629028320312, 99.26541137695312, 27.914718627929688, -79.55307006835938, 53.872222900390625, -0.9750347137451172, 89.83204650878906, 110.54899597167969, -16.2437744140625, 47.90010070800781, 29.438705444335938, 80.74282836914062, 47.8348388671875, -137.3458251953125, 129.3743896484375, -70.58322143554688, 144.5882568359375, 18.180938720703125, 21.30633544921875, 86.5092544555664, 1.9999237060546875, 87.06132507324219, 34.77264404296875, 1.5278205871582031, 9.6015625, 85.27410888671875, 0.0, 21.869651794433594, 51.392303466796875, 178.42947387695312, 128.341064453125, -99.17060852050781, -30.902862548828125, 111.74957275390625, 111.91571044921875, 77.5359878540039, 3.077606201171875, -53.5509033203125, 21.711761474609375, 61.46636962890625, 102.40000915527344, -33.181640625, -30.095245361328125, -40.940460205078125, 0.23992919921875, 46.134307861328125, 4.6377716064453125, -66.8157958984375, 10.090446472167969, -86.00566101074219, 99.23989868164062, -31.732513427734375, 109.3380126953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000234.npy"}
{"epoch": 0.4900523560209424, "step": 235, "batch_size": 128, "mean": 34.838375091552734, "std": 59.50562286376953, "min": -119.11077880859375, "p10": -31.704262542724607, "median": 32.65230751037598, "p90": 103.64405822753905, "max": 147.60015869140625, "pos_frac": 0.703125, "sample": [23.416046142578125, 7.4652557373046875, -6.3497314453125, 1.1133193969726562, 8.265174865722656, -117.27658081054688, -61.414703369140625, -5.670448303222656, 92.8487548828125, 88.2108154296875, 10.980186462402344, 140.246826171875, 82.91937255859375, 50.401092529296875, 70.73356628417969, 32.6885871887207, 90.20155334472656, 95.64239501953125, -2.24713134765625, -7.7449951171875, 74.60340118408203, 24.401702880859375, 32.61602783203125, 106.40869140625, 129.3554229736328, 128.43963623046875, 80.571533203125, -19.365036010742188, 86.82778930664062, -6.745643615722656, 108.74505615234375, 1.54986572265625, 74.8916015625, 95.67005157470703, 63.057952880859375, 91.50077056884766, 121.06912231445312, 98.25186157226562, 7.019256591796875, 138.353759765625, -12.395469665527344, 94.91387939453125, -30.598609924316406, -11.532012939453125, 56.20765686035156, 130.20703125, -103.114990234375, -24.95084571838379, -93.18865966796875, 49.75581359863281, 54.447540283203125, -16.439300537109375, -1.9258193969726562, -19.938690185546875, 33.61016845703125, 86.41256713867188, -26.955307006835938, 74.04292297363281, -9.6021728515625, -61.32914733886719, 9.137580871582031, -3.57861328125, 101.76188659667969, -35.0299072265625, 138.3948974609375, 69.9287338256836, -9.76641845703125, -89.34354400634766, 6.9285125732421875, 38.586700439453125, -24.172508239746094, 46.9842529296875, -55.907440185546875, 85.86585998535156, 7.4829864501953125, 9.703948974609375, 4.67919921875, 111.16709899902344, 5.936767578125, 48.414215087890625, 56.350311279296875, 20.545059204101562, -119.11077880859375, 25.055564880371094, 70.53573608398438, -13.617431640625, 78.06465911865234, 95.03276062011719, 72.60357666015625, 3.47100830078125, 140.60699462890625, 102.12680053710938, 66.86654663085938, 147.60015869140625, 105.90376281738281, 69.30413818359375, 34.484710693359375, 19.517974853515625, 93.45370483398438, -9.494295120239258, -0.9855728149414062, 100.33013916015625, -65.049072265625, 57.807159423828125, -1.0394287109375, 82.68268585205078, -4.648139953613281, 54.3079833984375, 84.82888793945312, -7.26629638671875, -2.914764404296875, 29.00164794921875, 3.8982467651367188, -34.28411865234375, 33.129188537597656, 14.037715911865234, 96.81410217285156, -117.9317626953125, -42.965576171875, 4.01995849609375, 79.67311096191406, 78.4257583618164, 24.3662109375, 102.67561340332031, 79.56298828125, 57.53954315185547, 27.01739501953125, 0.5246372222900391], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000235.npy"}
{"epoch": 0.49214659685863876, "step": 236, "batch_size": 128, "mean": 38.50131607055664, "std": 54.43721389770508, "min": -117.0435791015625, "p10": -27.623977661132805, "median": 33.392547607421875, "p90": 105.31501770019531, "max": 153.48870849609375, "pos_frac": 0.7890625, "sample": [56.523529052734375, 29.1260986328125, -68.92623901367188, 60.164337158203125, 64.23544311523438, 83.45796203613281, 9.09033203125, 99.2034912109375, -12.28375244140625, -52.91239929199219, -80.29931640625, 27.676422119140625, -34.50775146484375, 35.07023620605469, 74.19696044921875, 64.62948608398438, -0.9184112548828125, 88.15675354003906, 33.6961669921875, 36.140380859375, 26.234893798828125, 82.7647476196289, 39.44482421875, -51.80312728881836, 46.15607452392578, 88.98860168457031, 85.55496215820312, 5.655765533447266, 66.30715942382812, 99.22386169433594, 18.532569885253906, 32.78387451171875, 58.568031311035156, 20.855178833007812, 11.561332702636719, 19.154380798339844, 7.93731689453125, 128.94021606445312, 95.5821533203125, 4.75982666015625, 7.5320587158203125, 95.63248443603516, 102.13641357421875, 36.04606628417969, 125.53463745117188, 1.3181400299072266, -43.984405517578125, 0.490814208984375, 23.099014282226562, 0.0, -4.363006591796875, 18.27899932861328, 89.27877807617188, 94.12439727783203, 10.315874099731445, -25.706817626953125, -97.08255004882812, 97.70831298828125, 10.41668701171875, -39.850372314453125, 74.97024536132812, 110.17477416992188, 50.68182373046875, 61.613685607910156, 1.87152099609375, -32.09735107421875, 77.17449951171875, -117.0435791015625, 65.520263671875, 5.281833648681641, -16.054168701171875, 153.48870849609375, 56.62305450439453, 136.0908203125, 1.53082275390625, -5.884086608886719, -46.81634521484375, -4.303955078125, 104.49888610839844, 67.26461029052734, 40.63092041015625, 64.0224609375, -42.5770263671875, 81.30635833740234, 95.01837158203125, 32.7342529296875, 109.48701477050781, 132.668212890625, -6.562675476074219, 75.93545532226562, 66.79339599609375, 104.68545532226562, 106.78399658203125, 19.94000244140625, 2.96502685546875, 104.15953063964844, 102.273681640625, 71.26140594482422, 85.7959976196289, 113.30841064453125, 96.32098388671875, 33.08892822265625, -16.100112915039062, 20.919631958007812, 0.0, 43.68121337890625, 7.9299163818359375, 29.3092041015625, -95.06988525390625, 98.17802429199219, 5.9955291748046875, 0.4038429260253906, -16.539161682128906, -18.696640014648438, 45.957763671875, 99.09038543701172, 115.86738586425781, 12.360260009765625, 8.648689270019531, -21.841354370117188, 111.26478576660156, 85.55975341796875, 109.71023559570312, 16.47113037109375, 0.280517578125, 109.8031005859375, 8.801980972290039, 31.907745361328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000236.npy"}
{"epoch": 0.4942408376963351, "step": 237, "batch_size": 128, "mean": 32.349430084228516, "std": 56.40996170043945, "min": -126.98880004882812, "p10": -24.35809478759765, "median": 22.499969482421875, "p90": 109.38943252563476, "max": 190.243408203125, "pos_frac": 0.7421875, "sample": [63.91912841796875, 12.33782958984375, 47.751014709472656, 113.41738891601562, -11.539108276367188, 31.16064453125, 69.3733139038086, 84.12276458740234, 40.241455078125, 14.54449462890625, -5.29058837890625, 36.16278076171875, 135.60342407226562, -62.424034118652344, -28.37444305419922, 1.8966350555419922, 3.91107177734375, 22.006317138671875, 190.243408203125, 86.5619888305664, 25.374496459960938, 31.464706420898438, 107.11993408203125, 64.15438079833984, 10.576431274414062, 16.658203125, 72.74026489257812, 0.0, 73.21343231201172, 84.7099609375, 1.1227493286132812, 13.865219116210938, -42.419342041015625, 24.058250427246094, -16.376922607421875, 36.00128173828125, 2.77362060546875, 2.9248199462890625, -126.98880004882812, -22.636802673339844, -3.49462890625, -1.257781982421875, 6.2822113037109375, 91.6241455078125, 7.6332550048828125, 19.2880859375, 106.91201782226562, 29.0303955078125, 35.73606872558594, 2.4632129669189453, 1.80499267578125, 138.8157958984375, 61.914031982421875, 21.21001434326172, 13.957870483398438, 79.84701538085938, 102.65228271484375, 77.60321044921875, 24.93377685546875, 54.675048828125, 119.94525146484375, 36.464820861816406, 22.993621826171875, 14.3759765625, 57.08203125, -7.6520538330078125, 119.27426147460938, 98.96409606933594, -47.63531494140625, -62.8338623046875, 1.4718017578125, 129.3973388671875, -82.69210815429688, 8.85198974609375, 101.4013671875, -2.9152793884277344, -59.15928649902344, -12.459564208984375, 140.74131774902344, -16.66119384765625, 56.461669921875, 26.642120361328125, 113.26992797851562, 38.547088623046875, 112.7840576171875, -65.0545654296875, -4.587821960449219, 3.493011474609375, 111.55770111083984, 25.496917724609375, 9.56640625, 137.24725341796875, -8.926513671875, -1.599822998046875, -35.57232666015625, 126.37518310546875, -45.0635986328125, 2.8843994140625, 104.66619110107422, 82.701171875, 10.748779296875, 9.262954711914062, 90.6524658203125, -2.2807846069335938, 4.656562805175781, 5.5052947998046875, 90.26795959472656, -78.61624145507812, -14.805450439453125, -13.162826538085938, 1.5567626953125, 108.46017456054688, -18.762649536132812, -1.238067626953125, 54.36632537841797, 61.081085205078125, -119.62677001953125, 38.43342590332031, 85.10736083984375, 102.7340087890625, 93.6734619140625, -15.67193603515625, 25.58612060546875, 81.84263610839844, 3.809671401977539, 27.738677978515625, 18.2572021484375, 65.7149658203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000237.npy"}
{"epoch": 0.4963350785340314, "step": 238, "batch_size": 128, "mean": 38.37005615234375, "std": 58.10742950439453, "min": -86.8326416015625, "p10": -26.083680725097654, "median": 26.827186584472656, "p90": 116.78504943847656, "max": 160.25033569335938, "pos_frac": 0.703125, "sample": [-35.50494384765625, 1.5093460083007812, 6.469429016113281, 13.115222930908203, -51.07213592529297, 142.13693237304688, 96.14730834960938, 45.926513671875, 93.0560302734375, -48.102081298828125, -4.2996826171875, 75.38493347167969, -4.0074462890625, 15.371734619140625, -16.39905548095703, -61.005550384521484, 119.85992431640625, 100.38848876953125, -2.3779850006103516, 53.87315368652344, 101.13850402832031, 75.7265625, 40.4080810546875, 109.08281707763672, 142.71400451660156, 24.19226837158203, -12.01947021484375, 2.2819137573242188, 25.2161865234375, -84.71815490722656, -82.42098999023438, -85.41161346435547, 126.2589111328125, -79.137451171875, -9.933990478515625, 127.02359008789062, 32.959625244140625, 97.06350708007812, 2.5903167724609375, 78.49932861328125, 139.1800537109375, 2.17889404296875, 46.26783752441406, 152.718505859375, 87.75469970703125, 109.17373657226562, 116.39663696289062, 108.43533325195312, -10.15386962890625, 36.6993408203125, 110.58976745605469, 147.82415771484375, 70.21109008789062, -7.878074645996094, 19.78307342529297, 14.54644775390625, -1.9598541259765625, 5.13422966003418, -9.612747192382812, 75.28023529052734, -5.781974792480469, 29.770294189453125, -17.151824951171875, -86.8326416015625, 91.88168334960938, 61.047821044921875, 81.15292358398438, 102.2655258178711, 79.45458984375, 105.83956146240234, -48.030853271484375, -25.81951904296875, 46.60984802246094, -4.769683837890625, -1.9380950927734375, 66.59274291992188, 120.76068115234375, -14.5963134765625, 81.35910034179688, 97.863037109375, 72.37255859375, 117.69134521484375, -38.2995491027832, -9.498714447021484, 3.40802001953125, 25.243408203125, 89.34917449951172, -3.629974365234375, 62.01662826538086, 35.325592041015625, 0.25163841247558594, -8.309249877929688, -26.700057983398438, 47.895111083984375, 160.25033569335938, -8.087066650390625, 36.127685546875, 67.19234466552734, 17.022361755371094, 3.419231414794922, 26.127456665039062, 118.93389892578125, 114.14826965332031, 27.52691650390625, 41.227481842041016, 4.797821044921875, -13.892791748046875, 103.63102722167969, 20.0814208984375, 10.872711181640625, 25.09454345703125, 94.06668090820312, 17.931236267089844, 13.334426879882812, 96.19464111328125, -13.186935424804688, 5.027587890625, 90.89208984375, -5.011932373046875, -9.634033203125, 105.64738464355469, 130.046630859375, 31.40484619140625, -28.17681884765625, 34.72987365722656, 6.4945068359375, 79.40129089355469, -3.61376953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000238.npy"}
{"epoch": 0.49842931937172774, "step": 239, "batch_size": 128, "mean": 31.86258316040039, "std": 60.209529876708984, "min": -141.39251708984375, "p10": -39.50549621582031, "median": 18.818687438964844, "p90": 118.91762008666991, "max": 159.59994506835938, "pos_frac": 0.7109375, "sample": [-61.27058410644531, -59.425048828125, 131.43463134765625, 39.125152587890625, -14.18603515625, 1.00616455078125, -141.39251708984375, 74.27803039550781, 127.62191772460938, 1.268951416015625, 68.45751953125, -140.00433349609375, -95.6451416015625, -6.085601806640625, 85.89713287353516, 78.8323974609375, 97.44515991210938, 120.44556427001953, 59.42903137207031, -9.89996337890625, -0.2834625244140625, 104.33197021484375, 3.4995079040527344, 23.797927856445312, -8.999244689941406, 80.44036865234375, 0.1562480926513672, 89.2390365600586, 19.775482177734375, -13.917098999023438, -43.85565185546875, 91.98782348632812, 70.97537994384766, -3.7144393920898438, 140.82620239257812, -14.3687744140625, 25.32806396484375, 95.67327880859375, 159.59994506835938, 110.81201171875, 80.9695816040039, -7.91571044921875, 24.796661376953125, 0.2611083984375, 143.918212890625, -10.853836059570312, -12.016845703125, 69.36981201171875, -73.19671630859375, 91.85281372070312, -70.052001953125, 64.71791076660156, 17.232688903808594, 58.0687255859375, 51.83533477783203, -19.17840576171875, 48.283477783203125, 8.978675842285156, 8.121246337890625, -37.641143798828125, 17.48492431640625, 115.04533386230469, 7.5096940994262695, 103.9921875, 24.66887664794922, 33.13067626953125, 95.49665069580078, 1.797332763671875, 8.480171203613281, 131.42767333984375, 0.6953296661376953, 12.869318008422852, 11.968017578125, 37.119171142578125, 94.85678100585938, -6.9595947265625, 85.523681640625, 16.149085998535156, -62.73188781738281, 10.2406005859375, -0.60870361328125, 33.95952606201172, 136.1036376953125, -18.812408447265625, -2.3070068359375, 124.85028076171875, -60.384063720703125, 29.380325317382812, 10.26458740234375, -44.22210693359375, 107.03944396972656, -3.8134117126464844, -9.485374450683594, 102.01897430419922, 3.5867347717285156, 44.51631546020508, 5.162361145019531, 3.32110595703125, 134.43719482421875, 48.629302978515625, 0.8100433349609375, 132.58203125, 15.311660766601562, -45.99650573730469, 50.2413330078125, -1.1601181030273438, -27.23834228515625, 28.650299072265625, 37.954376220703125, 124.514892578125, 7.2102813720703125, 94.89187622070312, 52.921051025390625, 135.71722412109375, 118.26278686523438, 0.0, 19.193313598632812, 1.0210723876953125, 104.78497314453125, 47.313323974609375, -18.852996826171875, 77.2752685546875, 18.444061279296875, 23.12060546875, 34.46734619140625, -77.30345153808594, -2.1199264526367188, 17.804779052734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000239.npy"}
{"epoch": 0.5005235602094241, "step": 240, "batch_size": 128, "mean": 41.25169372558594, "std": 54.73796844482422, "min": -106.89167785644531, "p10": -17.708326721191405, "median": 35.89036560058594, "p90": 110.10013732910156, "max": 156.96136474609375, "pos_frac": 0.7265625, "sample": [7.731473922729492, 109.24519348144531, 65.20880126953125, 103.24014282226562, -23.071533203125, -1.4931716918945312, 65.69929504394531, 41.163848876953125, 0.0, 130.4503173828125, 34.34844970703125, -23.15277099609375, -81.88032531738281, 36.712005615234375, -19.60736083984375, 96.94286346435547, 53.41376495361328, 73.01602172851562, 88.29808807373047, 54.627655029296875, 19.9117431640625, -0.27838134765625, 75.3673095703125, 102.05361938476562, 15.8607177734375, 99.77328491210938, 77.2305908203125, -20.642562866210938, 113.369384765625, 108.97567749023438, 38.625579833984375, 33.782928466796875, 125.85955810546875, 87.96862030029297, -2.1573486328125, -13.238677978515625, 120.84353637695312, 79.6663818359375, 91.33293151855469, 13.200111389160156, 24.509246826171875, 49.660308837890625, 116.63877868652344, -0.739044189453125, 80.08856201171875, 50.979766845703125, 87.50735473632812, 76.01150512695312, 78.4306411743164, 51.439361572265625, 106.23898315429688, 64.62679290771484, 0.0, 68.5072250366211, -106.89167785644531, 25.8258056640625, -10.48193359375, -5.523345947265625, -49.03424072265625, -3.04119873046875, 82.39228820800781, 93.37138366699219, -7.3316497802734375, -11.857955932617188, 111.172119140625, 101.10894775390625, 46.89306640625, 83.75273895263672, 106.47395324707031, 76.89059448242188, 131.43777465820312, -28.748138427734375, 0.0, 22.035079956054688, -16.894454956054688, -10.1065673828125, 137.77178955078125, -2.3524627685546875, 17.02448272705078, 28.28387451171875, -9.823074340820312, 35.536468505859375, 92.2586669921875, 26.510787963867188, 72.70796203613281, 156.96136474609375, -66.1605224609375, 153.778076171875, -97.49244689941406, 115.15670776367188, 6.47833251953125, 12.017425537109375, 22.036956787109375, 93.70558166503906, 94.15773010253906, 41.37553405761719, 6.133544921875, 79.19733428955078, 14.406370162963867, -12.773893356323242, -16.576339721679688, 7.241556167602539, 1.6956233978271484, 17.768539428710938, 97.06477355957031, 40.6748046875, 78.75311279296875, 109.64071655273438, 36.2442626953125, 9.514312744140625, 112.61344146728516, 32.06175231933594, 27.493408203125, -77.03852844238281, 29.628952026367188, 3.113384246826172, -41.7193603515625, 74.00888061523438, -11.5457763671875, 90.45295715332031, 0.0, 6.649742126464844, 65.42222595214844, 24.0107421875, 21.31341552734375, -62.951148986816406, -4.98486328125, 129.02951049804688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000240.npy"}
{"epoch": 0.5026178010471204, "step": 241, "batch_size": 128, "mean": 30.63422393798828, "std": 59.198211669921875, "min": -108.11778259277344, "p10": -45.33814544677735, "median": 17.219764709472656, "p90": 109.57598419189452, "max": 164.99298095703125, "pos_frac": 0.6875, "sample": [-5.4110107421875, 113.08721923828125, 79.26197814941406, 60.062103271484375, 27.496341705322266, -30.4873046875, 23.260894775390625, 0.6541309356689453, 63.38380432128906, 91.11907958984375, 98.47561645507812, 0.0, -8.21990966796875, 111.12591552734375, 86.6383056640625, 139.762451171875, 6.977996826171875, 17.387893676757812, 15.37158203125, 115.82220458984375, 70.34014892578125, 98.54222106933594, -13.22015380859375, -12.304473876953125, 90.49391174316406, 18.31512451171875, -39.2052001953125, 30.063446044921875, 95.92440795898438, -10.312591552734375, 116.10352325439453, -35.3411865234375, 99.34623718261719, 59.069976806640625, 11.594833374023438, 5.210784912109375, -9.282684326171875, -8.425613403320312, 4.816247940063477, 101.68112182617188, 9.866752624511719, 17.59283447265625, 88.77415466308594, 138.45294189453125, 12.064956665039062, 87.09185791015625, 58.349822998046875, -104.89157104492188, 12.0858154296875, 24.193374633789062, 20.23131561279297, 1.8082046508789062, 2.4713897705078125, 87.24514770507812, -23.993255615234375, -7.8596343994140625, 106.29898071289062, -0.853179931640625, 20.19378662109375, -45.14955139160156, 58.290679931640625, 69.62181091308594, 42.236785888671875, 99.4818115234375, -6.465766906738281, 17.0516357421875, -13.492446899414062, 23.282760620117188, 1.0932502746582031, 56.0924072265625, -59.159828186035156, 9.48382568359375, 122.45893859863281, 57.478759765625, -45.7781982421875, -48.771728515625, 118.66326904296875, 75.2472152709961, 4.118804931640625, -2.60369873046875, 3.0019989013671875, 1.5122413635253906, 139.72947692871094, 30.1971435546875, 31.057144165039062, 54.18206787109375, -48.794219970703125, 99.71920776367188, -8.361053466796875, 108.91172790527344, 6.5226593017578125, -69.90959167480469, -1.3430976867675781, -0.8716888427734375, 99.9637451171875, 78.55682373046875, 3.346982955932617, 14.418083190917969, 118.25827026367188, -72.8865966796875, 82.52159118652344, -79.773681640625, 21.0792236328125, -15.750411987304688, 105.21246337890625, 108.72512817382812, -76.51655578613281, 6.280113220214844, 127.774169921875, 59.5224609375, -25.54071044921875, -108.11778259277344, -29.54278564453125, -7.384044647216797, 85.43737030029297, -0.21729660034179688, 138.58737182617188, 51.35392761230469, -66.64813232421875, 70.68829345703125, 164.99298095703125, 4.995086669921875, 13.53118896484375, -95.09527587890625, 10.8448486328125, -20.436981201171875, 92.2509765625, -46.287933349609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000241.npy"}
{"epoch": 0.5047120418848168, "step": 242, "batch_size": 128, "mean": 38.21381378173828, "std": 53.698795318603516, "min": -116.84149169921875, "p10": -20.62102508544922, "median": 38.519195556640625, "p90": 108.98780212402343, "max": 140.39434814453125, "pos_frac": 0.7578125, "sample": [92.11637878417969, 11.319259643554688, 97.85321044921875, 10.339054107666016, 16.27606201171875, 52.97052001953125, 74.81948852539062, 70.64582824707031, -24.547393798828125, 11.549850463867188, 132.363525390625, 88.3953857421875, 53.86840057373047, -14.971923828125, -90.69994354248047, 33.47978973388672, 49.42301940917969, 97.40435791015625, -116.84149169921875, 42.780094146728516, 14.854990005493164, 127.61178588867188, 119.99163818359375, 45.50788116455078, 53.089111328125, 133.0963134765625, -1.488677978515625, 124.51947021484375, -0.76397705078125, 1.855813980102539, 65.56085205078125, 94.94259643554688, -0.3009033203125, 120.1015625, 37.147918701171875, 134.95150756835938, 118.07835388183594, 3.5733165740966797, -72.40145111083984, 1.755859375, 91.36422729492188, -20.23748779296875, -0.3228759765625, 56.162078857421875, -21.515945434570312, -5.868207931518555, 30.619232177734375, 64.25717163085938, 88.57000732421875, 3.1404342651367188, -8.252853393554688, -8.575550079345703, -6.04559326171875, 24.458084106445312, -7.905433654785156, -4.416748046875, 98.9678955078125, 14.469207763671875, -13.04364013671875, 92.38839721679688, 45.79351806640625, 0.2702922821044922, -27.22528076171875, 113.41641235351562, 10.122711181640625, -2.6811294555664062, 138.8604736328125, 62.387908935546875, 56.12193298339844, 21.455169677734375, 55.98309326171875, 65.16400146484375, -23.27557373046875, 13.620689392089844, 82.7679443359375, 21.50495147705078, 81.46837615966797, 17.8687744140625, 21.181396484375, -60.772857666015625, 54.31218719482422, 79.9559097290039, 24.1669921875, 49.36546325683594, -86.67974853515625, 15.406417846679688, 6.651237487792969, 43.15882110595703, 39.890472412109375, 84.65786743164062, 140.39434814453125, 6.3841552734375, 46.96330261230469, 28.414714813232422, -3.18145751953125, 19.0355224609375, 71.95318603515625, 97.63699340820312, 70.4794692993164, 20.391006469726562, 1.9557647705078125, 96.1539306640625, -29.919246673583984, -16.099960327148438, 87.28817749023438, 26.57946014404297, 56.14333724975586, -98.68887329101562, 60.011932373046875, 74.05908203125, 36.15057373046875, 8.061370849609375, 108.28402709960938, 123.09114074707031, 83.8245849609375, 78.7471923828125, 110.62994384765625, 41.60974884033203, -63.44816589355469, 84.16107177734375, -10.805267333984375, 53.86164093017578, 12.616073608398438, -47.493011474609375, 63.923377990722656, 91.86306762695312, -17.856109619140625, 94.83364868164062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000242.npy"}
{"epoch": 0.506806282722513, "step": 243, "batch_size": 128, "mean": 45.02009201049805, "std": 54.88347625732422, "min": -108.70747375488281, "p10": -15.661839294433591, "median": 36.11174011230469, "p90": 115.9686508178711, "max": 153.882080078125, "pos_frac": 0.7734375, "sample": [-56.489776611328125, -19.603347778320312, 31.270523071289062, 24.122039794921875, 99.84335327148438, 71.47137451171875, 13.0982666015625, 122.89152526855469, -72.44235229492188, 89.98814392089844, 26.1619873046875, 138.14602661132812, 21.234540939331055, -24.57854461669922, 54.07733154296875, -14.768692016601562, -64.28535461425781, 73.62289428710938, 126.3597412109375, -56.93836212158203, 8.7978515625, 16.81048583984375, 1.43560791015625, 0.05753326416015625, 60.969635009765625, 109.11114501953125, 153.882080078125, 120.78018188476562, -3.6242523193359375, 102.87075805664062, 6.798419952392578, 88.91868591308594, 24.618118286132812, 70.52294921875, 101.86016845703125, 9.591949462890625, 115.96656799316406, 17.352630615234375, 94.7781982421875, 117.3353271484375, -12.871749877929688, 93.99404907226562, 52.77569580078125, 83.1077880859375, 65.72587585449219, 90.32843017578125, 115.9735107421875, -42.621917724609375, 54.636474609375, 63.67973327636719, 27.20147705078125, -108.70747375488281, -1.654541015625, 42.04925537109375, 17.0086669921875, 4.8775634765625, 38.45142364501953, 74.252197265625, 12.577072143554688, -4.24908447265625, 138.65438842773438, 66.1446533203125, 85.58389282226562, -10.6881103515625, -6.0115966796875, 84.9375, 27.323944091796875, 98.36166381835938, 84.3122787475586, -6.548095703125, 59.275634765625, -48.625274658203125, -12.39306640625, 109.92851257324219, 81.66641235351562, 28.618423461914062, 16.623504638671875, 62.608642578125, -3.4133758544921875, 24.17981719970703, -8.53382682800293, 92.47893524169922, -25.88977813720703, 78.97915649414062, 4.541961669921875, 0.0, 3.4976119995117188, 82.42033386230469, 92.77155303955078, 102.20071411132812, 33.412391662597656, 75.52951049804688, 82.54742431640625, -5.913887023925781, 112.52084350585938, 115.60427856445312, 120.55291748046875, 9.395172119140625, 4.3778228759765625, 35.5162353515625, 21.841064453125, 0.7689208984375, -13.574951171875, 27.634613037109375, -17.745849609375, 114.89804077148438, 128.522705078125, 87.35240173339844, 4.142036437988281, 36.707244873046875, 10.650398254394531, 10.97015380859375, 105.31267547607422, -13.14556884765625, -37.01896667480469, 149.57418823242188, 19.284221649169922, 108.56101989746094, -9.51708984375, 58.11744689941406, 31.731536865234375, 113.71444702148438, 143.19705200195312, 75.68722534179688, 131.73333740234375, 87.46160888671875, 51.125213623046875, -20.5120849609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000243.npy"}
{"epoch": 0.5089005235602094, "step": 244, "batch_size": 128, "mean": 44.832069396972656, "std": 58.88654708862305, "min": -126.54171752929688, "p10": -16.38234252929687, "median": 41.33563995361328, "p90": 119.01919555664062, "max": 201.654296875, "pos_frac": 0.75, "sample": [-126.54171752929688, 11.86669921875, 95.66305541992188, -0.28472900390625, 29.119949340820312, 12.45401382446289, 86.44866943359375, -30.217559814453125, 137.5838623046875, 61.10071563720703, 98.21505737304688, 61.35235595703125, -1.5788421630859375, -6.2582550048828125, 10.198020935058594, 19.48345947265625, 82.633544921875, 84.71631622314453, 84.84834289550781, 109.3218994140625, 18.343841552734375, 96.96736145019531, -1.7363052368164062, -5.2191162109375, 103.25352478027344, -1.3592529296875, 5.6064453125, -61.077125549316406, 106.0583267211914, -22.93206787109375, 8.73876953125, -15.398040771484375, -3.38824462890625, -4.8546600341796875, -96.05307006835938, 8.476081848144531, 137.64569091796875, 3.9246826171875, 92.62765502929688, 102.20317840576172, 5.5703887939453125, 50.314361572265625, 125.15557861328125, 67.20861053466797, 128.452880859375, 70.03536987304688, 0.353057861328125, 68.83560180664062, 41.87931823730469, 73.98072814941406, 99.9571533203125, 88.63507080078125, 0.0, 88.59613037109375, -13.73394775390625, -44.1468505859375, -52.9974365234375, 94.5570068359375, -5.8724365234375, -3.538726806640625, 118.65957641601562, -33.09660339355469, 3.839324951171875, 78.86035919189453, 4.970619201660156, 117.14862060546875, 110.58497619628906, 125.92718505859375, 201.654296875, 14.562744140625, 46.38825988769531, 32.20936584472656, -36.12939453125, 100.71881103515625, 56.879730224609375, 6.61773681640625, 123.8634033203125, -8.26708984375, 127.5477294921875, 81.63543701171875, 3.0156707763671875, 107.15203857421875, 10.30487060546875, -75.35316467285156, 26.6639404296875, 76.060302734375, -5.303985595703125, 86.61199951171875, 94.50692749023438, 12.608245849609375, 4.3252105712890625, -1.6182861328125, 34.640411376953125, 68.8829345703125, 132.52676391601562, 119.85830688476562, 107.80226135253906, 4.443809509277344, 146.47409057617188, 74.81001281738281, 1.9761962890625, 86.43780517578125, 40.791961669921875, 81.91350555419922, -18.679046630859375, 83.62811279296875, 25.601669311523438, 23.08209228515625, 117.12108612060547, 64.09219360351562, -24.403564453125, 50.50927734375, 3.4626388549804688, 127.98779296875, 63.756683349609375, 143.09442138671875, -2.1687660217285156, 78.5299301147461, 81.10719299316406, -10.885726928710938, 13.602760314941406, 15.669876098632812, 98.0617904663086, -103.08258056640625, 27.376129150390625, 79.89833068847656, 114.02947998046875, -2.188201904296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000244.npy"}
{"epoch": 0.5109947643979058, "step": 245, "batch_size": 128, "mean": 46.07227325439453, "std": 58.40761947631836, "min": -124.34696960449219, "p10": -20.539882850646972, "median": 48.852783203125, "p90": 112.40016479492185, "max": 160.77691650390625, "pos_frac": 0.7890625, "sample": [15.137420654296875, -13.953208923339844, 32.92945861816406, 33.869773864746094, 1.688934326171875, 24.47760009765625, 83.96563720703125, 4.6202392578125, -8.018768310546875, 66.01187133789062, 50.13226318359375, -7.196441650390625, 29.45519256591797, 136.41326904296875, 77.71693420410156, 17.56182861328125, 145.25735473632812, 136.809814453125, 58.07807922363281, 160.77691650390625, -124.34696960449219, -28.014129638671875, -61.767974853515625, -27.75372314453125, 26.839859008789062, 99.33917236328125, 0.0, 34.29095458984375, 72.89336395263672, -7.887451171875, 88.67323303222656, 91.30403137207031, 89.96250915527344, 97.5379638671875, 2.46435546875, 139.78759765625, 2.996122360229492, 29.852996826171875, 88.10011291503906, 63.8524169921875, 126.98873901367188, 12.1529541015625, 44.63671875, 98.83233642578125, 11.1910400390625, 89.93893432617188, 28.2100830078125, 49.1875, 46.19170379638672, -1.3780975341796875, -24.344970703125, 52.45762634277344, 8.99737548828125, 107.38433837890625, 42.277069091796875, -6.73974609375, 80.394775390625, 88.90380859375, 101.34786987304688, 109.82768249511719, 141.23052978515625, 6.282384872436523, -3.3411407470703125, 66.9801254272461, 52.76414489746094, 97.46129608154297, 26.47962188720703, -38.52546691894531, 53.657318115234375, -82.95687866210938, 51.12715148925781, 95.05579376220703, 77.53706359863281, 82.59027099609375, 98.64542388916016, 27.2059326171875, -64.37869262695312, 33.55553436279297, 156.6064453125, 110.1541748046875, 6.5664215087890625, 101.07892608642578, -66.10647583007812, 7.422275543212891, 11.095458984375, -52.04942321777344, 5.53460693359375, 108.45513916015625, 95.62005615234375, 98.20187377929688, 83.23252868652344, 18.480560302734375, -2.3384323120117188, 140.1524658203125, 53.412750244140625, 151.93911743164062, 18.993789672851562, 54.9600830078125, 53.08690643310547, 82.70925903320312, -90.33016967773438, 107.31977844238281, 75.25313568115234, -16.854736328125, -5.201833724975586, 117.64080810546875, 31.363189697265625, 92.53790283203125, 78.33863830566406, 100.93193054199219, 12.097808837890625, -14.445816040039062, 41.049774169921875, 48.51806640625, 93.75059509277344, 17.9444580078125, 0.706573486328125, -20.517940521240234, 134.97921752929688, 0.7524871826171875, 102.44638061523438, 58.7841796875, 76.45745849609375, 157.6240234375, -100.80392456054688, -20.591081619262695, 0.0, 68.6046142578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000245.npy"}
{"epoch": 0.5130890052356021, "step": 246, "batch_size": 128, "mean": 27.94359588623047, "std": 62.50722122192383, "min": -121.05001831054688, "p10": -50.043435668945314, "median": 15.208694458007812, "p90": 119.64728393554688, "max": 157.04736328125, "pos_frac": 0.65625, "sample": [55.693939208984375, 88.85379028320312, 73.52098083496094, -2.9478759765625, 75.95526123046875, -6.494758605957031, 157.04736328125, 11.099655151367188, 19.073631286621094, 45.69563293457031, -34.75283432006836, 77.24124145507812, 9.92364501953125, 17.55524444580078, -54.38972473144531, -49.978424072265625, -13.851882934570312, 131.75497436523438, 130.83981323242188, -52.56947326660156, 112.2235107421875, 139.80606079101562, 18.443572998046875, 94.52495574951172, 39.789703369140625, -13.161651611328125, 9.970062255859375, -53.1522216796875, 24.571807861328125, -4.85638427734375, 42.3468017578125, -31.19036865234375, 117.0108642578125, 29.69689178466797, 17.876556396484375, 0.97711181640625, -112.98402404785156, 116.40093994140625, 3.699737548828125, 0.7908229827880859, 14.41632080078125, -110.82135009765625, -21.832977294921875, 117.33340454101562, -53.46653747558594, -21.8853759765625, -63.08759689331055, -21.2698974609375, 1.15167236328125, 77.45150756835938, 37.692291259765625, 138.2745361328125, -50.19512939453125, 36.858856201171875, 123.48037719726562, -15.99812126159668, 62.40106201171875, -22.717117309570312, -121.05001831054688, 33.67709732055664, 132.70648193359375, 70.16375732421875, 84.89289855957031, 56.21626281738281, -32.91571044921875, 63.17192840576172, 132.90582275390625, 76.60671997070312, -14.557510375976562, 134.63214111328125, 125.3330078125, 30.20257568359375, -29.495208740234375, -0.212982177734375, 21.920623779296875, 0.5516357421875, 102.64956665039062, 16.991134643554688, -28.373031616210938, -67.76060485839844, -71.75564575195312, 16.001068115234375, 94.54412841796875, 103.7918701171875, -27.148574829101562, -4.8289794921875, -112.5072021484375, -58.42449951171875, 12.51416015625, 7.5195465087890625, 52.33551025390625, 119.52566528320312, 74.83720397949219, -49.76902770996094, 22.127029418945312, 27.597190856933594, 1.4480972290039062, -9.462738037109375, 0.0, 108.67242431640625, 121.90147399902344, 97.68414306640625, -16.59320831298828, -22.289329528808594, 90.87405395507812, 2.0087738037109375, -0.687957763671875, 87.57833862304688, 7.239574432373047, 69.09373474121094, 42.07939147949219, 5.2640380859375, 111.48638916015625, 73.30291748046875, 14.110565185546875, 119.93106079101562, -7.715324401855469, -34.611114501953125, -22.15423583984375, -30.40447998046875, 124.48443603515625, 3.63153076171875, 0.32030487060546875, 56.34934997558594, 10.425491333007812, 11.571731567382812, -5.4436187744140625, 112.2276611328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000246.npy"}
{"epoch": 0.5151832460732985, "step": 247, "batch_size": 128, "mean": 38.52513885498047, "std": 57.240386962890625, "min": -122.94097900390625, "p10": -24.779543304443358, "median": 30.616104125976562, "p90": 116.303759765625, "max": 175.70257568359375, "pos_frac": 0.7421875, "sample": [104.57601928710938, -32.713134765625, 77.47218322753906, 69.51300811767578, -12.891799926757812, 54.909019470214844, 116.3377685546875, 115.76559448242188, 94.47242736816406, 59.451995849609375, 0.0, 90.09465789794922, 116.2891845703125, 18.894180297851562, 6.8169708251953125, 36.9384765625, -24.35956573486328, 5.789398193359375, -87.344970703125, 83.887451171875, -12.540435791015625, 116.38311767578125, -74.246337890625, -27.59069061279297, 75.458984375, 91.499267578125, 44.912750244140625, 0.947784423828125, 91.32485961914062, 113.911865234375, 112.13156127929688, 106.11563873291016, 18.191177368164062, 141.50921630859375, 60.3466796875, -25.759490966796875, 1.936859130859375, 131.2442626953125, -6.2686767578125, 10.107223510742188, -9.474601745605469, -19.509117126464844, 38.554534912109375, -9.583740234375, 14.244964599609375, 74.235595703125, -2.414583206176758, 37.9781494140625, 3.3503036499023438, -73.903564453125, 5.999082565307617, 90.84841918945312, 62.824981689453125, 66.09213256835938, -37.80024719238281, 105.61125946044922, 149.00787353515625, 19.758453369140625, 1.474517822265625, -13.602226257324219, -16.235313415527344, -11.623283386230469, -27.51824188232422, 126.44900512695312, -1.834930419921875, -10.77520751953125, 72.23603820800781, -76.25210571289062, 9.015838623046875, -23.274688720703125, 25.538665771484375, 155.30130004882812, -122.94097900390625, 0.5786285400390625, 112.75439453125, 19.25432586669922, 21.32623291015625, 11.159408569335938, 30.828338623046875, 80.99955749511719, 55.789642333984375, 81.90159606933594, 175.70257568359375, 35.462554931640625, -22.50811767578125, -6.665428161621094, 2.4159164428710938, 55.18678283691406, 92.37602233886719, 63.340240478515625, 29.120941162109375, 85.47325134277344, 50.82293701171875, 24.30462646484375, 15.7139892578125, 92.770263671875, 126.16653442382812, 116.63075256347656, 137.7344970703125, -17.465469360351562, 37.38597106933594, 13.322906494140625, 73.84054565429688, 42.970733642578125, 86.6987075805664, 47.79835510253906, 27.66839599609375, -28.7349853515625, 125.9007568359375, 8.91558837890625, -95.0797119140625, 92.10555267333984, 50.22087860107422, -28.51861572265625, 30.40386962890625, 7.210004806518555, 32.231231689453125, -9.321487426757812, 16.80987548828125, -5.14166259765625, 82.09188079833984, 44.916900634765625, 122.99714660644531, 4.210700988769531, 96.4251708984375, 26.944122314453125, 25.124481201171875, 65.35232543945312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000247.npy"}
{"epoch": 0.5172774869109947, "step": 248, "batch_size": 128, "mean": 31.642454147338867, "std": 59.19371795654297, "min": -159.84597778320312, "p10": -38.23471755981445, "median": 24.8519287109375, "p90": 113.33991394042968, "max": 172.00701904296875, "pos_frac": 0.7578125, "sample": [11.758583068847656, 52.1715087890625, -59.79803466796875, 14.858573913574219, 136.3082275390625, -22.899932861328125, 22.915756225585938, 5.04571533203125, 78.05535888671875, 11.935928344726562, 26.33428192138672, 67.99870300292969, 73.56646728515625, 25.176773071289062, 7.223060607910156, 32.748382568359375, 130.71463012695312, 24.527084350585938, 18.71240234375, 84.57160949707031, -9.710678100585938, 19.210311889648438, 83.75662994384766, 40.45915985107422, 130.81686401367188, 33.62579345703125, 45.015071868896484, 31.358856201171875, 2.14459228515625, 100.58883666992188, 20.02281951904297, -41.365806579589844, 139.12521362304688, -15.169570922851562, 6.799890518188477, 30.421600341796875, 116.77192687988281, -159.84597778320312, 8.494873046875, -2.8247909545898438, -42.8468017578125, 9.910125732421875, 30.942359924316406, -26.793731689453125, -8.323307037353516, 33.85699462890625, -3.6477794647216797, -113.90985107421875, 35.44520568847656, 11.615142822265625, 36.220916748046875, 152.0650177001953, 32.789520263671875, 22.47157859802246, 151.48858642578125, -29.539886474609375, 112.856689453125, 0.8793106079101562, 6.823272705078125, 125.53836059570312, 70.16986083984375, 94.28681182861328, -36.892822265625, -64.94058227539062, 16.166122436523438, 8.179473876953125, 172.00701904296875, 111.89569091796875, 75.37783813476562, 43.799072265625, 80.2697982788086, -18.040267944335938, 3.3596954345703125, 89.07164001464844, 0.33935546875, -95.24472045898438, 0.0, 30.576995849609375, 170.95501708984375, 114.46743774414062, 104.41494750976562, 53.006195068359375, 75.57711791992188, 38.37347412109375, 8.124765396118164, 57.626678466796875, -24.5643310546875, -42.78057861328125, -8.384735107421875, 11.125167846679688, -42.52705383300781, 19.041015625, 116.74362182617188, 95.81549072265625, 109.92520141601562, -80.8248291015625, 45.55560302734375, 28.032691955566406, 7.756439208984375, 98.45863342285156, 1.6474380493164062, -48.0853271484375, 35.90827941894531, -17.362762451171875, 90.52557373046875, 30.652008056640625, 79.95928955078125, 7.32427978515625, 12.340789794921875, -97.75901794433594, 25.266021728515625, 19.74786376953125, 51.198448181152344, -0.7531890869140625, 142.86099243164062, 91.53851318359375, 69.30935668945312, 52.77516174316406, 37.43212890625, -28.086166381835938, -11.702159881591797, -5.576637268066406, 21.565032958984375, 34.234039306640625, 87.80885314941406, -70.55078125, 24.23743438720703, 18.01751708984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000248.npy"}
{"epoch": 0.5193717277486911, "step": 249, "batch_size": 128, "mean": 36.84126281738281, "std": 59.848663330078125, "min": -148.90960693359375, "p10": -28.135809326171874, "median": 26.085654258728027, "p90": 116.46644897460938, "max": 160.59393310546875, "pos_frac": 0.71875, "sample": [-10.5235595703125, 5.549098968505859, -30.496536254882812, -80.03109741210938, 63.39408874511719, 5.316009521484375, -16.671615600585938, -12.488800048828125, -1.432403564453125, 26.342926025390625, -2.0945816040039062, -60.142906188964844, 91.4654541015625, 111.96505737304688, -9.532119750976562, 83.2529296875, 32.829315185546875, 117.31072998046875, 25.06500244140625, 118.11947631835938, 12.626266479492188, -126.98696899414062, 105.68018341064453, 16.665435791015625, -76.68946838378906, 140.84521484375, 119.22137451171875, -13.796844482421875, 50.98138427734375, 83.7265625, 17.862220764160156, -9.163726806640625, 49.0673828125, -12.816764831542969, 24.542816162109375, 54.2247314453125, 124.29415893554688, 50.480804443359375, 131.58108520507812, -43.976409912109375, 107.70048522949219, 100.82061767578125, -10.42425537109375, 107.95664978027344, 145.70858764648438, 64.09060668945312, -90.27137756347656, 91.80928039550781, 74.52798461914062, 9.831565856933594, -12.039337158203125, 22.841957092285156, 160.22804260253906, 76.61429595947266, 84.4420166015625, -2.307373046875, 13.174644470214844, 70.90535736083984, -29.934814453125, 87.5115966796875, 37.15071105957031, -1.17791748046875, 1.285552978515625, 2.593048095703125, 93.06085205078125, 9.559516906738281, 79.29519653320312, 2.0748291015625, 76.93789672851562, 86.70309448242188, -21.833892822265625, -66.62464141845703, 23.513168334960938, 73.061767578125, 52.20912170410156, 81.05328369140625, 160.59393310546875, 131.89808654785156, 102.59718322753906, 97.20509338378906, 140.8291015625, 76.34883880615234, 18.413787841796875, -8.927490234375, 93.0992431640625, 132.04454040527344, -2.8730335235595703, -21.396835327148438, 0.9398307800292969, 70.05715942382812, 15.68487548828125, -50.12152099609375, 78.52447509765625, 28.717041015625, 27.642303466796875, 90.53836059570312, 39.939178466796875, 27.510009765625, 4.192718505859375, -42.671600341796875, 116.1046142578125, 1.831939697265625, 20.303741455078125, 80.26437377929688, 12.821754455566406, 7.74322509765625, -5.686500549316406, -148.90960693359375, -35.177093505859375, -14.201465606689453, -27.36480712890625, 9.847640991210938, 52.385101318359375, 1.939697265625, 20.15673828125, -2.679412841796875, 77.81491088867188, 42.032379150390625, 100.42994689941406, 2.82220458984375, 106.73941040039062, 83.01736450195312, 71.09541320800781, -25.21051788330078, -3.5823898315429688, 48.84136962890625, 25.82838249206543, 128.0735321044922], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000249.npy"}
{"epoch": 0.5214659685863874, "step": 250, "batch_size": 128, "mean": 35.90658950805664, "std": 64.65668487548828, "min": -111.2828369140625, "p10": -50.16792106628418, "median": 28.523239135742188, "p90": 120.22024993896481, "max": 170.8261260986328, "pos_frac": 0.703125, "sample": [93.91757202148438, 14.586837768554688, -51.50166320800781, 115.0892333984375, 80.22991943359375, 155.19369506835938, 143.7191162109375, 84.42567443847656, 38.734100341796875, -89.8458480834961, 85.10752868652344, 21.064393997192383, 75.38882446289062, 28.245391845703125, 29.115509033203125, 130.8035430908203, 46.10467529296875, 9.928070068359375, 126.63226318359375, 99.632568359375, 117.47224426269531, 71.39898681640625, 9.909210205078125, 96.50135803222656, -30.038360595703125, 2.339111328125, -111.2828369140625, 20.857421875, -5.927703857421875, -1.47381591796875, 134.2774658203125, -101.13848876953125, 44.593780517578125, 154.70977783203125, -10.109420776367188, 7.67535400390625, 28.26922607421875, 1.470367431640625, 130.85055541992188, -102.8551025390625, -2.4349632263183594, -4.1264190673828125, 115.66513061523438, 25.91490936279297, 76.340087890625, 67.19822692871094, -88.15139770507812, -67.87530517578125, -1.6210556030273438, 68.61293029785156, 89.68504333496094, 9.140769958496094, 96.84365844726562, -38.41717529296875, 29.291717529296875, 11.5823974609375, -8.6143798828125, -31.46575927734375, 62.098304748535156, 9.377838134765625, 112.70637512207031, 127.34605407714844, 90.24174499511719, -0.0206451416015625, 24.07410430908203, 41.457275390625, 148.61593627929688, 97.89579772949219, -2.9628219604492188, 40.73854064941406, 82.8384780883789, 94.39912414550781, -21.30828857421875, 115.61127471923828, 35.51904296875, -16.168685913085938, 28.777252197265625, 106.25833129882812, 126.92807006835938, 38.257232666015625, 23.949792861938477, -23.150634765625, 48.805206298828125, -25.4588623046875, 108.29898071289062, 14.864360809326172, -2.0153350830078125, -102.56817626953125, 78.48464965820312, 135.3753662109375, -5.41265869140625, 170.8261260986328, -57.47929382324219, -49.596317291259766, -100.28778076171875, 4.3529510498046875, 19.024383544921875, -57.67027282714844, -89.871337890625, 98.01959228515625, 73.76158142089844, 110.62345123291016, 93.8115234375, 84.551025390625, 91.5388412475586, 46.173667907714844, -30.08917236328125, 38.010498046875, 16.421829223632812, -1.007425308227539, 9.88946533203125, 69.05574035644531, 144.6417694091797, -51.75773620605469, 40.326751708984375, 77.77830505371094, 98.49951171875, 5.60589599609375, -12.314178466796875, 1.9211883544921875, -23.93242645263672, 88.67047119140625, 44.249298095703125, 2.929534912109375, -14.990570068359375, 23.23193359375, 15.191619873046875, -19.559062957763672], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000250.npy"}
{"epoch": 0.5235602094240838, "step": 251, "batch_size": 128, "mean": 36.47743225097656, "std": 59.70838165283203, "min": -108.91720581054688, "p10": -30.686722564697266, "median": 25.08642578125, "p90": 112.94975357055664, "max": 223.3333740234375, "pos_frac": 0.734375, "sample": [84.77955627441406, 72.26541137695312, 28.820419311523438, -15.071136474609375, -38.466064453125, -42.76812744140625, -58.815338134765625, 41.380401611328125, 5.7862396240234375, 19.1025390625, 8.081449508666992, 75.45526123046875, 108.35623168945312, 81.6763916015625, 4.747367858886719, -3.985301971435547, 48.252105712890625, -12.606643676757812, 5.273235321044922, 141.9130859375, -1.1848297119140625, 5.8994140625, -26.3658447265625, 109.152099609375, 89.67172241210938, 121.62728881835938, -48.76800537109375, 100.288818359375, 127.483154296875, 22.209640502929688, 106.428466796875, 5.13134765625, -30.553085327148438, 41.77874755859375, 44.709259033203125, -55.38992691040039, 53.29443359375, -0.6818389892578125, 12.741241455078125, -20.613536834716797, 89.47988891601562, 63.49835205078125, 102.98739624023438, 122.69735717773438, 122.99710083007812, 18.739212036132812, -0.43414306640625, -69.91786193847656, 3.141704559326172, 7.4943695068359375, -30.99854278564453, 15.320556640625, 223.3333740234375, 26.10626220703125, 76.71770477294922, 95.25836944580078, 28.8941650390625, 108.75282287597656, 3.2181663513183594, 59.874366760253906, 80.92948913574219, -63.50616455078125, 20.480300903320312, 33.989288330078125, 68.62289428710938, -74.046630859375, 7.1701812744140625, -4.841716766357422, 24.06658935546875, -2.5878238677978516, 100.82687377929688, 28.75592041015625, 9.548187255859375, 4.07713508605957, 48.777313232421875, -13.109878540039062, 8.600296020507812, 27.5054931640625, -5.871219635009766, 79.24322509765625, 1.4698486328125, 42.0006103515625, 2.319830894470215, -4.8227386474609375, 79.73829650878906, 98.91265106201172, 111.0181884765625, -80.49531555175781, -0.539306640625, 8.43011474609375, 65.26678466796875, 114.83197021484375, 105.27700805664062, 14.21441650390625, 142.71060180664062, -5.567684173583984, 184.8419189453125, 115.84478759765625, 77.7863540649414, -5.130924224853516, 105.07229614257812, 32.661346435546875, 77.82855224609375, 53.938232421875, 54.299407958984375, 32.51812744140625, 112.1430892944336, 4.1263885498046875, 81.18360900878906, -108.91720581054688, 1.36712646484375, 28.169891357421875, 12.912872314453125, -5.255790710449219, 5.4064483642578125, -93.39723205566406, -26.34210205078125, -19.504043579101562, 1.935699462890625, 144.223388671875, 105.57745361328125, -46.69377136230469, 124.50987243652344, 132.17974853515625, -20.008453369140625, 4.222572326660156, 77.64486694335938, 72.37568664550781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000251.npy"}
{"epoch": 0.5256544502617801, "step": 252, "batch_size": 128, "mean": 38.24516677856445, "std": 66.86100006103516, "min": -141.94146728515625, "p10": -33.890795898437496, "median": 30.50216293334961, "p90": 125.14929046630859, "max": 200.79522705078125, "pos_frac": 0.71875, "sample": [181.86593627929688, 131.51641845703125, 127.64077758789062, 98.71025085449219, -74.64552307128906, 39.36601638793945, -74.38867950439453, 7.858001708984375, 7.910091400146484, -16.271392822265625, -6.325775146484375, -26.898635864257812, 16.212799072265625, 26.92596435546875, 200.79522705078125, 50.564849853515625, 121.61198425292969, 80.50617218017578, 107.91477966308594, 106.47824096679688, 5.422611236572266, 62.72483825683594, -26.546905517578125, 124.47874450683594, -141.94146728515625, 3.63446044921875, 5.830413818359375, 32.288475036621094, -32.669921875, 51.233001708984375, 1.3007049560546875, 121.50802612304688, 110.005126953125, 103.12405395507812, 79.62042999267578, 15.03900146484375, -52.131927490234375, -10.147665023803711, 9.2701416015625, 191.1624755859375, 40.703277587890625, 8.796119689941406, 81.02301025390625, 28.715850830078125, 15.564956665039062, 16.07647705078125, 25.541152954101562, -8.466339111328125, -1.76605224609375, 14.436538696289062, 101.3702392578125, 10.6103515625, 126.71389770507812, 85.44332122802734, -19.1832275390625, -5.0350341796875, 51.9998779296875, -88.6942138671875, 33.53712463378906, 22.984375, 86.7969970703125, -29.458648681640625, 39.016815185546875, 134.84002685546875, 51.794837951660156, 70.46177673339844, 115.90562438964844, 35.90869140625, 34.5137939453125, 123.17327880859375, -6.4447021484375, 100.00237274169922, 137.9583740234375, -6.793159484863281, -62.53235626220703, 108.42868041992188, -36.739501953125, 8.976814270019531, 67.97650146484375, 84.52943420410156, 82.317138671875, 5.38006591796875, 35.43798828125, 97.36006927490234, 16.643417358398438, -4.750480651855469, 0.0, 90.23060607910156, -102.48300170898438, 110.39995574951172, -0.055248260498046875, 100.46607971191406, -13.469833374023438, -19.60833740234375, 34.05950927734375, 133.22787475585938, -1.764801025390625, 155.62167358398438, 43.745269775390625, 113.60282897949219, -4.36456298828125, 80.55895233154297, 57.901554107666016, 2.31072998046875, 129.32046508789062, 1.067047119140625, 98.07040405273438, -116.36053466796875, 19.751220703125, 2.3603668212890625, -44.2860107421875, -30.76971435546875, 39.5030517578125, -108.71014404296875, -95.52420043945312, 1.924346923828125, 91.86904907226562, 127.8271484375, -52.163421630859375, 102.43795776367188, 17.14849853515625, 140.65625, 32.98423767089844, -31.01654052734375, -3.289764404296875, 93.81460571289062, 12.935443878173828, 93.82501220703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000252.npy"}
{"epoch": 0.5277486910994764, "step": 253, "batch_size": 128, "mean": 31.103050231933594, "std": 61.96000671386719, "min": -130.850830078125, "p10": -40.642808532714845, "median": 15.998039245605469, "p90": 117.12527770996094, "max": 152.1656494140625, "pos_frac": 0.734375, "sample": [142.11575317382812, -4.56353759765625, 11.31695556640625, 64.05838012695312, 17.5687255859375, 20.896759033203125, 2.2107086181640625, 88.92726135253906, 149.8837890625, 24.7659912109375, 54.75433349609375, 117.31678009033203, 5.06231689453125, -18.53948974609375, 88.27474975585938, -99.38568115234375, 61.982215881347656, -66.69517517089844, 103.71359252929688, -32.479095458984375, 9.3958740234375, -13.368438720703125, 20.601070404052734, 10.943021774291992, 0.82879638671875, -31.55126953125, 31.50360107421875, 131.64251708984375, 99.2881088256836, 87.05912017822266, -6.416160583496094, -13.374210357666016, -104.83563232421875, 41.45013427734375, 152.1656494140625, 8.059783935546875, -14.8125, 103.12921142578125, -130.850830078125, -70.313720703125, 116.6007080078125, 51.271636962890625, 123.05804443359375, 25.41590118408203, 17.91241455078125, -40.33818054199219, 105.99229431152344, 5.276885986328125, -7.6580810546875, 14.129425048828125, 117.20169067382812, 44.336700439453125, 116.03115844726562, 13.297409057617188, -64.72322082519531, 107.62615966796875, 10.476654052734375, 9.709541320800781, -41.353607177734375, 81.05647277832031, 30.340362548828125, -38.988800048828125, 5.474609375, 102.03792572021484, 28.733734130859375, 106.09315490722656, -53.537872314453125, -13.966876983642578, 83.14431762695312, -35.884063720703125, 25.46607208251953, -11.7618408203125, -103.10411834716797, 10.943695068359375, -105.81472778320312, 51.62377166748047, -72.84170532226562, 4.050516128540039, -4.4061279296875, -33.388458251953125, 71.16319274902344, 0.6307830810546875, 63.7960205078125, 83.84774780273438, 32.22789001464844, -1.965576171875, 25.628318786621094, 12.088134765625, -48.633697509765625, 135.99295043945312, -55.72303771972656, 7.888206481933594, 31.506683349609375, -0.13897705078125, 7.2393798828125, 106.77713012695312, -33.33992385864258, 144.25982666015625, 7.301795959472656, 75.84799194335938, 89.00370788574219, 12.089805603027344, -32.905517578125, 2.873046875, 33.27806091308594, 117.092529296875, 13.9261474609375, 118.49258422851562, 12.366592407226562, -3.8309593200683594, 29.39886474609375, 127.77890014648438, 91.79964447021484, 2.5811920166015625, 90.67328643798828, 14.427352905273438, 9.150665283203125, 14.305221557617188, 61.664306640625, 143.9879150390625, 8.715560913085938, 41.69476318359375, 131.20162963867188, 94.34602355957031, 11.1981201171875, 90.58726501464844, 67.1351089477539, 100.50076293945312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000253.npy"}
{"epoch": 0.5298429319371728, "step": 254, "batch_size": 128, "mean": 38.33012008666992, "std": 57.91658020019531, "min": -135.10205078125, "p10": -24.596510314941398, "median": 28.523917198181152, "p90": 114.77167358398437, "max": 151.96022033691406, "pos_frac": 0.7578125, "sample": [140.338623046875, 0.216156005859375, -78.60585021972656, 28.083656311035156, -13.454063415527344, -3.555736541748047, 16.48895263671875, 17.562721252441406, -0.8051567077636719, 20.16522216796875, -0.6312332153320312, -1.14373779296875, -38.734283447265625, -3.263641357421875, 7.233184814453125, 126.54385375976562, 3.609375, 97.49685668945312, 46.7938232421875, -32.65008544921875, -55.19122314453125, 124.24237060546875, 36.108367919921875, 14.62353515625, -5.658721923828125, -0.4207305908203125, 99.33309936523438, -6.482940673828125, 90.56048583984375, 58.737213134765625, 99.823486328125, 134.14288330078125, -63.101806640625, 74.2542724609375, 1.1012859344482422, 5.6202239990234375, -22.638916015625, 38.108154296875, 14.523567199707031, 11.262908935546875, 59.43412780761719, 52.93714904785156, 83.07244873046875, 40.33195495605469, -48.695701599121094, 19.323562622070312, 36.43280029296875, 44.59417724609375, 64.655029296875, 96.52113342285156, 37.46806335449219, 4.5142364501953125, 28.512372970581055, 113.80410766601562, 78.988037109375, 28.53546142578125, 85.4573974609375, -57.253326416015625, 74.06744384765625, 74.5115966796875, 5.186943054199219, 19.456626892089844, 131.9718017578125, 6.879669189453125, 36.13941955566406, 90.5715560913086, 58.34344482421875, 113.44754028320312, 116.24960327148438, 29.449325561523438, 20.088768005371094, 47.14569091796875, 126.759765625, 5.516777038574219, 43.670501708984375, -20.980056762695312, -4.403533935546875, 32.508758544921875, 2.314311981201172, 3.2272796630859375, -4.469329833984375, -10.326484680175781, 129.6014404296875, 112.6702880859375, 78.3263931274414, 138.093505859375, 100.091796875, 101.59815979003906, 17.08221435546875, 109.76602172851562, 20.511859893798828, 111.11402893066406, 112.37814331054688, 91.84588623046875, 104.5912094116211, 27.002960205078125, 0.4163818359375, 107.2531509399414, -6.47747802734375, -12.9451904296875, -46.22917175292969, 46.58966064453125, 47.52186584472656, -1.8266983032226562, 0.0, -29.164230346679688, 122.43124389648438, 73.89683532714844, 34.05712890625, 21.70269775390625, 11.92059326171875, 21.400291442871094, 23.56298065185547, 114.13827514648438, 83.2733154296875, 7.15704345703125, -135.10205078125, -65.12932586669922, 138.75762939453125, 22.022441864013672, -106.00299072265625, -86.62362670898438, 4.97930908203125, 151.96022033691406, 148.20838928222656, 63.5203857421875, 111.39505004882812, 38.32081604003906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000254.npy"}
{"epoch": 0.5319371727748691, "step": 255, "batch_size": 128, "mean": 44.33265686035156, "std": 66.7242431640625, "min": -113.08148193359375, "p10": -30.716670227050777, "median": 29.504470825195312, "p90": 128.72192077636717, "max": 177.86447143554688, "pos_frac": 0.7734375, "sample": [17.282814025878906, -113.08148193359375, 139.76141357421875, 39.78868103027344, 38.01603698730469, -95.3985595703125, 2.6470947265625, 121.26142883300781, -11.72601318359375, 101.6471939086914, -8.910232543945312, 9.210540771484375, 8.39144515991211, 10.12969970703125, 148.69967651367188, 123.7508544921875, 20.92444610595703, 14.127716064453125, 140.7630615234375, 11.019989013671875, 106.68017578125, -69.46395874023438, 0.80377197265625, 97.28732299804688, 40.93701171875, 126.70710754394531, 49.80543518066406, 107.74618530273438, -29.322113037109375, 25.57757568359375, 75.54110717773438, 7.680694580078125, 23.169998168945312, 12.77130126953125, 67.7334976196289, 32.737457275390625, 122.05303955078125, 105.79214477539062, -32.552001953125, -72.7840576171875, 177.86447143554688, 148.48135375976562, 129.46646118164062, 28.306045532226562, -83.38456726074219, -15.727821350097656, -93.05905151367188, 115.0491943359375, 91.37430572509766, 91.83805847167969, 115.46621704101562, 123.89773559570312, 83.4215087890625, 88.87698364257812, 145.62464904785156, -3.7386856079101562, 74.833984375, -81.0335693359375, -44.22587585449219, 111.09674072265625, 24.6654052734375, 115.70346069335938, 106.89370727539062, 99.58486938476562, 44.5194091796875, 100.79264831542969, 6.077781677246094, -99.28986358642578, 123.82110595703125, 18.124435424804688, 144.44656372070312, 52.164642333984375, 26.884231567382812, -6.611396789550781, 3.455841064453125, 9.396977424621582, 5.667026519775391, 17.404953002929688, 7.495452880859375, 94.71431732177734, -3.89276123046875, 88.71026611328125, -108.84867858886719, 140.66925048828125, 120.23464965820312, 95.51554870605469, -9.21795654296875, 46.384666442871094, 15.705703735351562, 14.522125244140625, 128.40283203125, 49.2413330078125, 73.19031524658203, 123.4477310180664, -16.1932373046875, 147.851806640625, 113.30615234375, 1.06787109375, 7.0877685546875, -16.062545776367188, 0.7237586975097656, -37.52587890625, 7.8643341064453125, 5.78399658203125, 134.75775146484375, 82.46466064453125, 8.087127685546875, 102.35086059570312, -15.232429504394531, 42.5689697265625, 9.6055908203125, 9.658851623535156, -12.047309875488281, 113.47073364257812, 19.886032104492188, 31.772933959960938, -11.273529052734375, -10.59857177734375, 104.90287780761719, 84.71115112304688, 26.75041961669922, 108.07430267333984, -7.31689453125, 142.189453125, -52.8409423828125, 156.34912109375, 30.702896118164062, -29.930099487304688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000255.npy"}
{"epoch": 0.5340314136125655, "step": 256, "batch_size": 128, "mean": 43.8728141784668, "std": 70.74752044677734, "min": -130.5968780517578, "p10": -57.043002700805665, "median": 54.75575256347656, "p90": 129.5247817993164, "max": 164.21786499023438, "pos_frac": 0.734375, "sample": [52.21919250488281, 99.7169189453125, 113.564453125, -38.4119873046875, 81.43769836425781, 82.01043701171875, 117.72232055664062, 109.2152099609375, 116.18087768554688, 0.0455322265625, 101.19789123535156, -27.180252075195312, -32.37908935546875, -99.32229614257812, 53.0892333984375, 89.91850280761719, 129.84005737304688, -17.950836181640625, 84.09236145019531, 61.21746826171875, 1.1884613037109375, 110.07579040527344, -71.9866943359375, 164.21786499023438, 18.8387451171875, 123.56544494628906, 107.09756469726562, -58.30389404296875, 138.90875244140625, 25.24658203125, 135.65380859375, 23.220401763916016, 32.37451171875, 48.403778076171875, 74.58683776855469, 14.2454833984375, -49.482025146484375, -84.52630615234375, 135.17465209960938, 57.93372344970703, 139.93994140625, 19.688690185546875, 22.21234130859375, 78.15077209472656, 56.65119934082031, 49.52716064453125, -66.60952758789062, -18.850051879882812, -36.39265441894531, 154.34613037109375, 4.800037384033203, -64.0572280883789, 131.58694458007812, 134.4862060546875, -19.592117309570312, 2.549896240234375, -130.5968780517578, -34.9083251953125, 111.17141723632812, 61.25721740722656, 97.38967895507812, 19.874298095703125, 104.20126342773438, 137.34539794921875, -1.6498336791992188, 89.52978515625, 104.2972412109375, -101.47660064697266, 93.09146118164062, -7.4405670166015625, -46.916656494140625, -94.03004455566406, 124.407958984375, 40.427764892578125, 40.973388671875, 110.33001708984375, -20.01666259765625, 115.789306640625, 26.78277587890625, 136.72463989257812, 146.82525634765625, 82.49858856201172, 89.55313873291016, -2.6990814208984375, 67.68704223632812, 64.0765380859375, -91.50508117675781, 19.583335876464844, 56.422271728515625, 129.945068359375, -42.00836181640625, 116.43775939941406, 98.3371353149414, 125.5069580078125, 13.005615234375, 102.18427276611328, 75.94845581054688, -70.73953247070312, 17.71117401123047, 25.4154052734375, 39.497039794921875, 42.42024612426758, 126.84048461914062, 81.0311279296875, 88.86309814453125, 59.38330078125, 11.450103759765625, 120.3209228515625, -56.502620697021484, -129.40792846679688, 5.69525146484375, 124.81759643554688, 129.38966369628906, 15.415626525878906, -10.405921936035156, 0.7186279296875, 125.68350219726562, -42.807464599609375, -30.28851318359375, -99.38079833984375, 95.61611938476562, -5.672088623046875, -6.275398254394531, 65.37910461425781, 19.76873779296875, 58.696903228759766, 66.40264892578125, 107.23190307617188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000256.npy"}
{"epoch": 0.5361256544502618, "step": 257, "batch_size": 128, "mean": 42.84898376464844, "std": 63.91341781616211, "min": -119.32997131347656, "p10": -28.571354103088378, "median": 28.728469848632812, "p90": 125.8055435180664, "max": 185.1641845703125, "pos_frac": 0.671875, "sample": [5.9432220458984375, 23.949615478515625, 174.37196350097656, -17.237590789794922, -2.5578155517578125, 185.1641845703125, 22.703338623046875, 123.51911926269531, 103.88909912109375, 107.61095428466797, 47.49481201171875, 140.87640380859375, 102.09603881835938, 79.88290405273438, -0.12914276123046875, 28.455604553222656, 43.064605712890625, -18.8017578125, -24.676658630371094, 129.53921508789062, 124.9664306640625, -36.565040588378906, 83.25371551513672, 158.26174926757812, -17.347564697265625, -46.69281005859375, 27.006103515625, 23.397048950195312, -4.16180419921875, -28.1602840423584, -34.84770965576172, -29.530517578125, 109.70098876953125, 23.804107666015625, 88.95635986328125, 122.293212890625, 178.92755126953125, 1.5428009033203125, 135.58120727539062, 92.67839813232422, -21.18829345703125, 7.140602111816406, -20.977012634277344, 106.501953125, 125.33297729492188, -37.273956298828125, -4.148506164550781, 28.940505981445312, 32.09796905517578, -88.61415100097656, 73.04261779785156, 154.635986328125, 61.95140838623047, 44.46900177001953, 94.71026611328125, 18.463226318359375, 86.85624694824219, -22.735519409179688, 73.093994140625, 67.3580322265625, 65.67535400390625, 127.92324829101562, 107.2913818359375, 18.620952606201172, -55.3392333984375, -16.846359252929688, 106.35006713867188, 2.898099899291992, 49.45024871826172, 107.04061889648438, -0.3543243408203125, 100.78091430664062, 75.64185333251953, 37.70159912109375, 20.526165008544922, 27.526519775390625, -119.32997131347656, -0.8410263061523438, 79.02215576171875, 122.83544921875, 4.9112396240234375, 3.735067367553711, 70.95596313476562, 28.516433715820312, 22.638042449951172, -15.473617553710938, -10.949462890625, 115.51278686523438, 80.56683349609375, -31.975067138671875, -16.290618896484375, 156.1009521484375, -30.676422119140625, 126.20156860351562, 56.21603012084961, 111.42929077148438, 145.54965209960938, 80.86508178710938, -18.588287353515625, 24.59607696533203, 40.002105712890625, 13.217071533203125, 106.446044921875, 33.2901611328125, -2.5812835693359375, -14.7613525390625, -5.550262451171875, -68.59356689453125, 27.279205322265625, 125.63581848144531, 25.880126953125, 113.81353759765625, -20.662811279296875, 165.01075744628906, -58.59906005859375, -17.986618041992188, 71.3995361328125, -24.992218017578125, 75.91464233398438, 33.955108642578125, 31.070510864257812, -5.882938385009766, -10.904747009277344, 61.0721435546875, -8.855056762695312, -11.624359130859375, -66.16842651367188, 109.55166625976562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000257.npy"}
{"epoch": 0.5382198952879581, "step": 258, "batch_size": 128, "mean": 47.604637145996094, "std": 59.57499694824219, "min": -141.2728271484375, "p10": -23.085211181640624, "median": 42.57379150390625, "p90": 122.13643493652343, "max": 156.18649291992188, "pos_frac": 0.7734375, "sample": [104.69313049316406, 22.626020431518555, 12.460929870605469, 6.650962829589844, 103.89918518066406, -32.34295654296875, 138.35720825195312, -23.268157958984375, 46.738311767578125, 106.4459228515625, -41.479644775390625, 36.06597900390625, 130.05584716796875, 74.9090576171875, 117.91989135742188, 74.92893981933594, -23.455062866210938, 42.247772216796875, 104.17816162109375, 94.27915954589844, 125.96063232421875, 89.07948303222656, -35.57289123535156, 11.94366455078125, 155.028076171875, -35.31037902832031, -141.2728271484375, 113.83384704589844, 129.2825927734375, 32.50947570800781, 104.376220703125, 84.44876098632812, -30.023468017578125, 70.16255187988281, 65.636474609375, 37.571746826171875, 3.7499122619628906, -23.508148193359375, 61.518218994140625, 103.99612426757812, 81.17355346679688, 88.54537963867188, 5.966651916503906, 38.716644287109375, 14.146804809570312, 64.05108642578125, 25.163055419921875, 15.42364501953125, 12.87225341796875, 13.742782592773438, 67.95911407470703, 6.680393218994141, 105.99038696289062, 12.581329345703125, -123.29989624023438, 75.02947998046875, 60.20953369140625, -6.591033935546875, -129.75042724609375, 6.299163818359375, -18.275863647460938, 23.455535888671875, 31.662384033203125, 39.0179443359375, 117.1148681640625, 2.5122451782226562, 148.64590454101562, 117.71307373046875, -7.819366455078125, 93.67337036132812, 116.83689880371094, 64.576171875, -6.2975006103515625, -8.80364990234375, -0.6270675659179688, 24.035308837890625, 128.43408203125, -51.68305969238281, 86.9013900756836, 48.16949462890625, -8.601083755493164, -23.006805419921875, 95.3309555053711, 104.06132507324219, -26.738510131835938, -0.36815643310546875, 96.037841796875, 93.46881103515625, 120.62188720703125, -0.6976318359375, 30.934894561767578, 31.595619201660156, 83.11557006835938, 19.360416412353516, -4.3540496826171875, 37.757476806640625, 99.06649780273438, 101.14320373535156, 22.144939422607422, -14.118316650390625, 42.899810791015625, -4.783935546875, 81.14067077636719, -16.98992919921875, 156.18649291992188, 1.0167388916015625, 110.47103881835938, 130.74105834960938, 125.67037963867188, 108.98536682128906, 45.823638916015625, -15.119964599609375, 149.20330810546875, 84.752685546875, 134.12872314453125, 85.93228149414062, 56.105438232421875, 17.257598876953125, 73.24176025390625, 92.69139099121094, -12.72344970703125, 78.45150756835938, 83.00579833984375, 144.88607788085938, 12.3773193359375, 27.771713256835938, 35.31298828125, 4.72918701171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000258.npy"}
{"epoch": 0.5403141361256545, "step": 259, "batch_size": 128, "mean": 40.74772644042969, "std": 63.80384826660156, "min": -119.581298828125, "p10": -30.925474548339835, "median": 38.00965881347656, "p90": 119.66470947265626, "max": 161.5618896484375, "pos_frac": 0.734375, "sample": [98.93313598632812, -43.388916015625, 126.41705322265625, 9.717269897460938, -28.32073974609375, 11.209075927734375, 23.5009765625, 139.8677215576172, 106.64227294921875, -9.5084228515625, 68.1102294921875, 14.820709228515625, 95.60514831542969, 103.07992553710938, 26.332504272460938, 57.15711975097656, 94.130615234375, -110.94509887695312, 81.23558044433594, 108.58883666992188, 38.577880859375, 147.3131103515625, 5.516571044921875, 34.00493240356445, 119.9339599609375, 26.393524169921875, 75.33479309082031, 78.3664321899414, -107.17742919921875, -5.6170806884765625, 8.523582458496094, 2.7446975708007812, -22.869873046875, 91.307373046875, 108.68441772460938, 40.0994873046875, 52.149169921875, 11.964950561523438, 16.885345458984375, 92.37197875976562, -39.11138916015625, 19.36767578125, 135.34307861328125, 21.37955665588379, 3.594482421875, -119.581298828125, 84.31083679199219, 131.17654418945312, -2.23193359375, 91.98887634277344, 89.44813537597656, 82.89559936523438, 110.6890869140625, 4.29913330078125, 19.73773956298828, 29.054908752441406, -9.266571044921875, 129.2003173828125, 34.364166259765625, 13.589263916015625, 161.5618896484375, 54.396461486816406, 80.91845703125, 2.1321678161621094, -6.52752685546875, -92.9471435546875, 47.18658447265625, -37.00318908691406, -89.61619567871094, 62.75439453125, 43.44073486328125, -11.97369384765625, -6.8349761962890625, 125.23629760742188, 127.43951416015625, -13.86163330078125, 83.61032104492188, 0.0, -11.83697509765625, 114.69699096679688, 93.5030517578125, 111.52618408203125, -64.76651000976562, 16.46795654296875, 51.7813720703125, 70.37294006347656, 128.4022216796875, -84.40973663330078, -0.8924636840820312, 48.839385986328125, -113.3643798828125, -21.664520263671875, 127.47468566894531, 128.3406982421875, 37.441436767578125, 25.385498046875, 86.85047149658203, -20.66473388671875, 106.84110260009766, -17.29522705078125, 60.50517272949219, 73.49005126953125, -19.901962280273438, 114.91845703125, 15.8870849609375, -4.20233154296875, 108.83842468261719, -101.539306640625, 101.02642822265625, 119.54931640625, 74.18118286132812, 109.85861206054688, 13.06402587890625, -17.27294921875, 13.336359024047852, 26.012611389160156, 117.05331420898438, 22.72528076171875, 39.89350891113281, -23.780853271484375, 97.13859558105469, -42.60931396484375, -23.25537109375, 114.82135009765625, 4.106292724609375, 95.42611694335938, 79.08897399902344, 86.4671630859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000259.npy"}
{"epoch": 0.5424083769633508, "step": 260, "batch_size": 128, "mean": 16.95346450805664, "std": 65.39718627929688, "min": -150.11260986328125, "p10": -60.017897796630855, "median": 8.985359191894531, "p90": 111.09094238281249, "max": 151.69366455078125, "pos_frac": 0.6015625, "sample": [123.01388549804688, -6.920158386230469, 96.86957550048828, 38.75822448730469, -120.431396484375, 107.44113159179688, 26.5489501953125, 129.23196411132812, 15.805389404296875, 93.18603515625, -20.283676147460938, 22.228759765625, 0.0, -17.5718994140625, 47.07426452636719, 4.481412887573242, 3.489776611328125, 151.69366455078125, 74.2079086303711, 40.2142219543457, 101.75079345703125, -150.11260986328125, 0.08003616333007812, -2.78985595703125, 147.1353759765625, -60.64009094238281, 15.660877227783203, 92.86961364746094, 35.87135314941406, 39.27484130859375, 30.095733642578125, -55.58338928222656, 15.958389282226562, 6.203695297241211, 4.476507186889648, 144.25302124023438, -82.88392639160156, -125.134521484375, -11.036575317382812, 27.967697143554688, 0.031524658203125, 15.1475830078125, 42.058349609375, -96.38800048828125, -26.639617919921875, 95.748291015625, -78.89983367919922, -25.910205841064453, -46.778167724609375, 13.147655487060547, -3.9439697265625, 4.1634521484375, 117.79644775390625, 38.9727783203125, -18.430347442626953, 14.373504638671875, 114.30575561523438, -125.51626586914062, 99.79109191894531, 87.14613342285156, -30.577362060546875, -3.3342819213867188, 91.61402893066406, -35.007904052734375, 6.1694488525390625, -90.32865905761719, 64.31707763671875, 110.0106201171875, -5.0006103515625, 124.28787231445312, 7.8524017333984375, 127.84262084960938, -49.61070251464844, 6.6068115234375, 43.56486511230469, -20.816970825195312, -18.14025115966797, 40.907684326171875, 13.86383056640625, -29.74664306640625, -87.73541259765625, 13.39862060546875, 138.56915283203125, -0.11529541015625, -0.287017822265625, 50.95684814453125, 24.424957275390625, -16.8927001953125, -27.843093872070312, -45.647308349609375, -29.27008056640625, 2.486724853515625, -109.78639221191406, 133.14508056640625, 11.099716186523438, -27.1337890625, -7.632785797119141, 52.004783630371094, -29.755706787109375, 113.6116943359375, 78.42141723632812, 68.64797973632812, 20.772735595703125, 62.77174377441406, -9.9307861328125, 5.53045654296875, -3.8865814208984375, 55.13964080810547, -59.751243591308594, 138.307373046875, -7.9693603515625, 100.66098022460938, 83.94244384765625, 10.118316650390625, -55.21295166015625, 4.5915374755859375, -55.718292236328125, -112.34677124023438, -12.235153198242188, -15.1668701171875, 44.2379150390625, -66.34649658203125, 16.84344482421875, 10.703960418701172, 91.04762268066406, 11.095718383789062, 98.15079498291016, -47.07916259765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000260.npy"}
{"epoch": 0.5445026178010471, "step": 261, "batch_size": 128, "mean": 34.819332122802734, "std": 67.92252349853516, "min": -122.50627136230469, "p10": -65.15066986083984, "median": 22.773025512695312, "p90": 121.3261001586914, "max": 188.04547119140625, "pos_frac": 0.6875, "sample": [-65.04853820800781, -116.3775634765625, 3.9598236083984375, -35.044166564941406, 105.37457275390625, 89.89315795898438, 85.94940185546875, -19.41412353515625, 61.03187561035156, 111.53453826904297, 122.78758239746094, 0.0, -12.093446731567383, -23.723876953125, 5.342937469482422, 66.75634765625, -23.54962921142578, 41.360137939453125, 121.47767639160156, 34.562744140625, 70.26649475097656, -66.18241882324219, 109.35205841064453, -81.7640380859375, 81.793701171875, -24.87720489501953, -49.817138671875, 76.29399108886719, -73.84815216064453, -96.76513671875, 90.35551452636719, 103.8076171875, 121.26113891601562, 80.34736633300781, 127.44454956054688, 87.84420776367188, -18.600635528564453, 15.418865203857422, -49.507476806640625, 66.56614685058594, 13.381057739257812, 16.371490478515625, 128.99749755859375, -24.8919677734375, -12.099578857421875, 19.343345642089844, -0.40499114990234375, 113.47246551513672, -89.25375366210938, 94.1202621459961, 97.80398559570312, -22.4111328125, 60.926239013671875, 137.4880828857422, 115.764404296875, 56.6610107421875, 79.66169738769531, -37.29643249511719, 88.37713623046875, 105.71293640136719, 114.22653198242188, 118.24319458007812, 61.283077239990234, 14.713607788085938, 90.68256378173828, -0.9619255065917969, -6.192771911621094, -83.75985717773438, 26.20270538330078, -1.2918167114257812, 112.73193359375, 188.04547119140625, 3.085968017578125, 148.54104614257812, 37.62348937988281, -17.291107177734375, 8.30670166015625, 3.249387741088867, 10.384368896484375, 12.381134033203125, 89.8187255859375, -6.496612548828125, 118.36775207519531, -65.38897705078125, 137.12857055664062, 84.4541015625, 78.76739501953125, 67.09911346435547, 15.033981323242188, -92.95797729492188, 11.677345275878906, 10.431427001953125, -93.95574951171875, 5.71881103515625, -52.1978759765625, 130.5969696044922, 66.233642578125, 110.54391479492188, 104.37385559082031, -11.49761962890625, -66.68109130859375, 44.66801452636719, 0.9512786865234375, 106.46511840820312, 16.703842163085938, 50.75, -67.34596252441406, 146.690673828125, 125.14678955078125, -24.23662567138672, 129.33480834960938, -6.464729309082031, 31.26038360595703, -122.50627136230469, 38.407623291015625, 10.278976440429688, 11.360565185546875, 15.8961181640625, 62.391357421875, 12.4310302734375, 83.36515808105469, 149.2835693359375, 11.772506713867188, -52.273193359375, -0.043292999267578125, 44.679656982421875, 2.2487640380859375, 88.49214935302734], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000261.npy"}
{"epoch": 0.5465968586387434, "step": 262, "batch_size": 128, "mean": 43.26238250732422, "std": 62.53471755981445, "min": -107.34458923339844, "p10": -25.946691894531245, "median": 28.817352294921875, "p90": 126.8262710571289, "max": 157.9285888671875, "pos_frac": 0.75, "sample": [72.70660400390625, 12.34356689453125, 129.24935913085938, 82.82588195800781, 98.54721069335938, 3.7527694702148438, 139.5860595703125, 54.100372314453125, 26.144378662109375, 141.38992309570312, 156.84585571289062, -19.7391357421875, 52.90087127685547, 16.350433349609375, 12.616653442382812, 4.265462875366211, 102.59928131103516, 32.74884033203125, 16.503494262695312, 83.06494140625, 122.9700927734375, 10.24591064453125, -53.94127655029297, 124.31597900390625, 19.873870849609375, 136.372314453125, 24.4234619140625, 33.438934326171875, 105.120849609375, 145.240234375, 4.18438720703125, -107.34458923339844, 35.233612060546875, -17.409378051757812, -1.5431365966796875, -40.89564514160156, -84.8228759765625, 102.5462646484375, -1.4684600830078125, 119.34423828125, 32.752105712890625, -6.659387588500977, -8.41943359375, -35.2879638671875, 33.64994812011719, 58.322288513183594, 11.318801879882812, -3.791168212890625, 61.357086181640625, -21.59002685546875, 119.88034057617188, 157.9285888671875, 87.48365783691406, 93.19692993164062, -85.4781494140625, -2.801593780517578, 24.228240966796875, 53.01416015625, -24.35382080078125, 16.580718994140625, 23.675140380859375, -15.254638671875, -7.1367950439453125, 73.73681640625, 0.9365768432617188, 13.748779296875, 138.71087646484375, 17.59228515625, 136.87286376953125, 68.17019653320312, -72.03196716308594, -1.35369873046875, 125.67791748046875, 1.572052001953125, 68.03164672851562, -84.61575317382812, -34.56220245361328, 118.44769287109375, 1.86651611328125, -20.552215576171875, -103.34840393066406, -12.203893661499023, 0.0, 123.15137481689453, 149.7608642578125, 40.2098388671875, 15.283332824707031, 93.90814971923828, 27.62871551513672, 4.3976593017578125, 106.46671295166016, 98.83993530273438, 108.49200439453125, 20.079010009765625, 126.80427551269531, 99.88778686523438, 15.843170166015625, 75.88177490234375, 33.201416015625, 132.21461486816406, 3.770111083984375, -18.0224609375, 117.86517333984375, 92.458740234375, -30.739532470703125, 13.460960388183594, 27.86456298828125, 29.7701416015625, 126.87759399414062, 75.7291259765625, -3.0881919860839844, 75.6352767944336, 141.90484619140625, 89.03611755371094, 116.76104736328125, 23.659423828125, 123.25105285644531, 74.40481567382812, 112.95821380615234, 109.60586547851562, 18.868194580078125, -72.74968719482422, 41.096923828125, 27.2371826171875, 11.014854431152344, 87.95167541503906, -15.383071899414062, -29.66339111328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000262.npy"}
{"epoch": 0.5486910994764398, "step": 263, "batch_size": 128, "mean": 36.09868621826172, "std": 68.98711395263672, "min": -143.47866821289062, "p10": -42.58305053710937, "median": 17.040199279785156, "p90": 133.02637939453126, "max": 187.38491821289062, "pos_frac": 0.6796875, "sample": [131.90884399414062, 74.10446166992188, 133.66778564453125, 19.81549072265625, -5.883003234863281, 59.92530822753906, -24.2293701171875, 9.387542724609375, 81.21988677978516, 56.47332763671875, -10.671636581420898, 25.418807983398438, 23.57659912109375, 24.9432373046875, -43.70794677734375, 5.375942230224609, 1.200876235961914, -84.0889892578125, 92.69924926757812, -7.28350830078125, 57.4586181640625, -65.84371948242188, -7.873046875, 86.03669738769531, -13.266510009765625, 122.45230102539062, 111.27192687988281, 58.345252990722656, -37.90966796875, -58.257080078125, -93.64917755126953, 117.03073120117188, 137.92181396484375, 138.90914916992188, 142.45294189453125, -23.26903533935547, -11.870086669921875, -15.016555786132812, -71.58102416992188, 121.16218566894531, 7.286956787109375, -5.23095703125, -9.772109985351562, 144.76699829101562, 59.197418212890625, 9.580564498901367, 95.35775756835938, -52.828582763671875, 117.47158813476562, 48.517311096191406, 132.0301513671875, -22.75970458984375, -19.377487182617188, 90.08883666992188, 140.09140014648438, 82.30764770507812, -33.63612365722656, 6.56793212890625, 111.53628540039062, 115.6287841796875, 0.0, 35.66485595703125, -143.47866821289062, 5.384464263916016, 54.5924072265625, 8.508056640625, 2.380706787109375, 1.7447662353515625, 11.886009216308594, 30.298980712890625, 138.7858428955078, 3.6747360229492188, -98.95159912109375, -28.864892959594727, 14.924446105957031, 3.7008056640625, 91.74137878417969, 117.82421875, 69.03564453125, 2.9571380615234375, -12.356613159179688, 8.93817138671875, -49.865966796875, 8.526611328125, 12.564804077148438, 65.33985900878906, -91.61114501953125, 138.85101318359375, 102.209228515625, 19.15595245361328, -21.55712890625, 140.7908935546875, 10.927864074707031, -122.326416015625, -5.820465087890625, -8.434173583984375, -83.1849365234375, 133.5789794921875, 19.467453002929688, -42.1009521484375, 98.59111022949219, 4.110931396484375, 0.0, 9.321578979492188, 93.51165771484375, 64.94515991210938, 187.38491821289062, 43.472991943359375, 171.95953369140625, -7.696220397949219, 113.19293212890625, 3.8909225463867188, 127.79139709472656, 0.0, 132.78955078125, 102.18743133544922, 19.341323852539062, 27.25493621826172, 106.99813079833984, 4.402427673339844, 117.6679458618164, 116.13627624511719, 48.12452697753906, -1.089630126953125, 170.33004760742188, -11.773639678955078, 78.1380615234375, -20.4378662109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000263.npy"}
{"epoch": 0.5507853403141362, "step": 264, "batch_size": 128, "mean": 46.656036376953125, "std": 59.088069915771484, "min": -118.79129028320312, "p10": -14.66858901977539, "median": 36.67869567871094, "p90": 132.46953125, "max": 190.08099365234375, "pos_frac": 0.7734375, "sample": [-4.993806838989258, -10.9381103515625, 155.55331420898438, 56.115386962890625, 138.92376708984375, 107.84671783447266, 94.18006134033203, 89.32479858398438, 2.985574722290039, 13.42694091796875, 28.94003677368164, 100.45413208007812, 88.07598876953125, 25.455291748046875, 36.51385498046875, 102.99019622802734, 89.57875061035156, 117.40806579589844, -15.27720832824707, 0.73858642578125, 71.3968505859375, 110.41000366210938, -9.761871337890625, 74.6551742553711, -15.515769958496094, 159.6375732421875, 118.7010498046875, 0.5666828155517578, 59.97967529296875, 87.20989990234375, -28.440078735351562, 100.57975006103516, 127.19155883789062, 119.11790466308594, 143.48727416992188, 0.0, -11.08404541015625, 154.60809326171875, 69.92938995361328, 28.097335815429688, -10.058868408203125, 133.37371826171875, -13.576217651367188, 61.98126220703125, 82.7776870727539, 113.000732421875, -35.50390625, 155.8914794921875, -60.696044921875, 35.55632019042969, 2.7451171875, 18.294464111328125, 89.80853271484375, 26.352981567382812, 109.16549682617188, -76.37225341796875, 121.91439819335938, 97.00360107421875, 21.571258544921875, 18.509986877441406, 50.759765625, 0.3820610046386719, 42.98521423339844, 40.600669860839844, 47.25987243652344, -23.22442626953125, -5.55438232421875, 21.579513549804688, 22.499298095703125, 118.128662109375, 0.170867919921875, 18.638595581054688, -118.79129028320312, -15.077629089355469, 34.43361282348633, -2.3897018432617188, 20.319034576416016, 10.479713439941406, 132.7667236328125, 131.71307373046875, 42.709808349609375, 1.2251052856445312, 0.04961967468261719, 64.70523071289062, -2.6330947875976562, 190.08099365234375, -7.090850830078125, 0.0, 97.15252685546875, 35.19943618774414, -43.406036376953125, 68.80410766601562, 109.81864166259766, 132.3421630859375, 19.52447509765625, 39.14442443847656, -4.234682083129883, 41.17816162109375, 89.12655639648438, 24.722488403320312, 133.11587524414062, 71.2579345703125, -56.80523681640625, -15.307586669921875, 50.6639404296875, 47.989410400390625, 30.42340087890625, 11.53814697265625, -14.4932861328125, 90.31993103027344, 0.67596435546875, 43.452911376953125, 42.931610107421875, 170.91717529296875, 20.27032470703125, 143.48614501953125, -5.019287109375, 21.56153106689453, 142.45159912109375, -8.111522674560547, -66.3604507446289, 36.843536376953125, 68.69486999511719, 6.2626495361328125, 56.90777587890625, 1.938018798828125, 39.3155517578125, 11.144866943359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000264.npy"}
{"epoch": 0.5528795811518324, "step": 265, "batch_size": 128, "mean": 42.461151123046875, "std": 66.54603576660156, "min": -116.88407897949219, "p10": -33.591776275634764, "median": 34.80339050292969, "p90": 133.57421264648437, "max": 163.40228271484375, "pos_frac": 0.7109375, "sample": [0.8008956909179688, 148.76353454589844, 118.80789184570312, -81.038818359375, 44.143035888671875, 7.65313720703125, 111.10086822509766, 76.12725830078125, 91.68301391601562, -78.77351379394531, 54.584442138671875, 42.69464111328125, 87.1524658203125, -14.7374267578125, 131.76910400390625, 60.99797058105469, 36.347381591796875, 138.41265869140625, 50.64312744140625, 83.16827392578125, 159.7493896484375, 140.924560546875, 144.97149658203125, 16.538604736328125, 155.81069946289062, 102.08918762207031, 126.0850830078125, 137.99630737304688, 59.975341796875, -6.3720855712890625, -32.958595275878906, 103.78097534179688, 150.75082397460938, 8.549407958984375, 121.09446716308594, 132.017578125, 78.23251342773438, 111.2381591796875, 40.18048095703125, 126.85882568359375, -30.01934814453125, 1.474700927734375, -21.923583984375, 41.2396240234375, -8.58343505859375, -13.5341796875, -18.78179931640625, -39.201995849609375, -3.4554595947265625, 95.60321044921875, 1.082763671875, -17.740493774414062, -35.06919860839844, 5.6847991943359375, -87.09786987304688, 154.88763427734375, 46.34375, 24.17010498046875, 92.614013671875, -116.88407897949219, 13.819316864013672, 104.70413208007812, 4.4056396484375, 102.51490783691406, 19.799163818359375, -88.31689453125, 126.29136657714844, 26.10809326171875, 30.787673950195312, -24.770957946777344, -42.51275634765625, 123.07075500488281, 143.02516174316406, 14.417518615722656, 0.0, -5.247732162475586, -12.969146728515625, -93.45095825195312, 113.50439453125, 114.43974304199219, 33.2593994140625, 5.1341552734375, 67.4118423461914, 1.131927490234375, 110.62230682373047, 107.9278564453125, 0.0, -6.688758850097656, -11.621063232421875, 89.13055419921875, 75.2220458984375, -26.14276123046875, -9.11871337890625, 1.7471694946289062, 48.60052490234375, 5.140129089355469, 52.57794189453125, 163.40228271484375, 10.72314453125, -98.6434326171875, 16.579490661621094, 111.11276245117188, 74.8394775390625, 75.63667297363281, -5.9180908203125, 23.260177612304688, -92.86236572265625, 139.939208984375, 109.814453125, -41.0499267578125, -5.600624084472656, 67.4990234375, 7.0799560546875, -10.950080871582031, 66.10256958007812, -38.153717041015625, 0.634979248046875, 137.20635986328125, 81.33576965332031, 89.37965393066406, 104.64234924316406, -12.6636962890625, 49.399940490722656, 29.38690185546875, 22.034517288208008, -25.614715576171875, 25.009262084960938, 88.88876342773438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000265.npy"}
{"epoch": 0.5549738219895288, "step": 266, "batch_size": 128, "mean": 45.52922058105469, "std": 62.2417106628418, "min": -133.0147705078125, "p10": -12.185767364501952, "median": 35.12992858886719, "p90": 122.9977813720703, "max": 185.1494140625, "pos_frac": 0.8046875, "sample": [40.43206787109375, 96.55980682373047, 167.16464233398438, 17.969696044921875, -11.371978759765625, -13.543502807617188, 97.55804443359375, -2.355682373046875, -13.433944702148438, 12.703788757324219, 30.37841796875, 4.5316162109375, -4.7018280029296875, 54.39501953125, 83.99574279785156, -8.62646484375, -8.4678955078125, 52.61244201660156, 27.641019821166992, -98.62393188476562, 117.18620300292969, 44.21953582763672, 4.212890625, 31.5509033203125, 127.61184692382812, 90.955322265625, -2.134246826171875, 43.38233947753906, 138.30523681640625, 128.68678283691406, 2.398143768310547, 155.46173095703125, -13.051223754882812, 59.508880615234375, 6.618457794189453, 106.93426513671875, -42.7469482421875, 9.216594696044922, 21.93719482421875, 122.17434692382812, 11.697410583496094, 26.1737060546875, 14.400604248046875, -8.627723693847656, 115.5873031616211, 102.04034423828125, 83.26937866210938, 118.7082748413086, 14.823745727539062, 27.830078125, 36.470489501953125, 102.87152099609375, -109.6124267578125, -110.05259704589844, -0.22879791259765625, -38.35777282714844, 5.715906143188477, 43.95082092285156, 47.834564208984375, 81.16421508789062, 124.91912841796875, 76.91162872314453, 15.37506103515625, 16.7322998046875, 21.358112335205078, 142.88540649414062, 98.40852355957031, 114.87257385253906, 11.6043701171875, 29.49444580078125, 33.78936767578125, 137.330078125, 64.788330078125, 54.73223876953125, 114.28229522705078, 160.32394409179688, 38.0377197265625, -4.840179443359375, -78.62448120117188, -97.91483306884766, -133.0147705078125, 108.05958557128906, 13.241119384765625, 118.10646057128906, -14.551345825195312, 44.191802978515625, 51.398223876953125, 21.87811279296875, 74.40377807617188, 26.58428955078125, -70.91179656982422, 4.66015625, -0.0889434814453125, 55.459808349609375, 110.09872436523438, 58.410064697265625, 95.15165710449219, 67.2425537109375, 15.74530029296875, 3.305755615234375, 18.85693359375, 53.451416015625, 27.668548583984375, 143.58792114257812, 18.079225540161133, 19.050537109375, -11.814857482910156, 0.8915786743164062, 40.03349304199219, -5.5658416748046875, 112.528564453125, 137.740234375, 143.2053985595703, 11.334251403808594, 117.95927429199219, 8.596954345703125, 119.98321533203125, 88.3295669555664, 11.119987487792969, 40.14697265625, 185.1494140625, 14.604415893554688, 116.97370910644531, 9.504066467285156, 113.5877685546875, 97.54967498779297, 106.5828857421875, 121.86361694335938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000266.npy"}
{"epoch": 0.5570680628272251, "step": 267, "batch_size": 128, "mean": 41.258426666259766, "std": 69.77111053466797, "min": -143.5370635986328, "p10": -33.81546554565429, "median": 32.80020332336426, "p90": 125.61571655273437, "max": 209.64990234375, "pos_frac": 0.71875, "sample": [3.9585933685302734, 113.05424499511719, 100.23054504394531, 90.36068725585938, 160.232421875, 14.93817138671875, 11.056503295898438, 29.369598388671875, -8.086669921875, -1.75244140625, -0.3234901428222656, 26.110198974609375, -4.33587646484375, 137.3268585205078, 36.15550994873047, -26.961517333984375, 9.103622436523438, 142.0423583984375, 141.56529235839844, 34.51434326171875, -57.50450134277344, 117.40863037109375, 115.18380737304688, 2.7315216064453125, 34.268741607666016, -25.16351318359375, 108.00753784179688, 106.10517883300781, 139.72210693359375, -122.482666015625, 106.53668212890625, 20.54827880859375, 27.1727294921875, -20.5426025390625, 13.799835205078125, 92.7320327758789, -9.175207138061523, 142.7183837890625, 123.9779052734375, 90.12462615966797, -84.84471130371094, 151.66311645507812, 2.3719253540039062, 95.71401977539062, 13.121223449707031, 179.37255859375, 1.75830078125, -23.727325439453125, 70.03134155273438, 125.21900939941406, -103.36893463134766, -61.501861572265625, 100.42352294921875, 105.91085052490234, 11.182937622070312, 171.60089111328125, 104.32012176513672, 92.9935302734375, 137.79434204101562, 9.652425765991211, 142.39437866210938, 80.0361557006836, -106.20196533203125, 11.515472412109375, 59.6259765625, 3.39044189453125, 120.6004638671875, -66.53855895996094, 47.569305419921875, 107.37600708007812, 11.392852783203125, 100.19863891601562, 81.28851318359375, -128.93246459960938, 42.27714538574219, 81.710693359375, 64.01101684570312, 73.94976806640625, 5.443393707275391, -0.487518310546875, -31.17375946044922, 94.76080322265625, -8.647483825683594, 42.380859375, -39.97944641113281, 91.07920837402344, -48.4420166015625, 50.995269775390625, 20.26287841796875, 88.96075439453125, -7.854026794433594, -25.577621459960938, 43.00917053222656, 42.37896728515625, 4.495635986328125, 104.21260833740234, 86.55779266357422, 37.652347564697266, 110.50538635253906, -11.591156005859375, 117.72169494628906, 36.23284912109375, -27.044189453125, 20.447357177734375, -11.746894836425781, 27.666229248046875, 24.823074340820312, 111.81282043457031, -6.026763916015625, 209.64990234375, -24.07220458984375, -43.520843505859375, 97.00460815429688, 31.3316650390625, 91.12138366699219, 1.0376224517822266, 124.77963256835938, 106.65284729003906, -10.1788330078125, -13.767822265625, 126.54136657714844, -143.5370635986328, -4.414794921875, 5.127254486083984, 50.33403778076172, -87.49334716796875, -28.110321044921875, 9.6861572265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000267.npy"}
{"epoch": 0.5591623036649215, "step": 268, "batch_size": 128, "mean": 40.72065353393555, "std": 63.45707702636719, "min": -107.80262756347656, "p10": -29.215927124023438, "median": 25.30242919921875, "p90": 127.11143341064452, "max": 175.183349609375, "pos_frac": 0.7578125, "sample": [33.19378662109375, 11.535789489746094, 88.31583404541016, 57.07073974609375, 111.091064453125, 5.121917724609375, -1.9056243896484375, -85.25713348388672, -6.132503509521484, -59.57695007324219, 20.189987182617188, 160.5911865234375, 19.18341064453125, 125.89435577392578, -0.0606689453125, 88.71011352539062, 175.183349609375, -4.510614395141602, -0.72991943359375, 100.99493408203125, 111.35684967041016, 118.98381042480469, 56.004608154296875, -101.77433776855469, 57.25054931640625, 28.788619995117188, 0.1410961151123047, 46.05572509765625, 21.0596923828125, 14.94613265991211, 7.918962478637695, 100.11813354492188, 99.0435791015625, 99.96153259277344, -29.685470581054688, 1.3912162780761719, -107.80262756347656, -44.897796630859375, -29.014694213867188, 170.78680419921875, 90.44033813476562, 28.72564697265625, 25.81207275390625, 128.877197265625, -3.57110595703125, 12.293472290039062, 73.73793029785156, -98.7164306640625, -42.7900390625, -6.185035705566406, 28.310089111328125, 17.06982421875, 53.166473388671875, 126.37655639648438, 4.051170349121094, 40.701141357421875, 149.80711364746094, -17.846054077148438, 67.16558837890625, 1.29425048828125, 162.20321655273438, 11.270065307617188, 117.15007781982422, -6.04168701171875, -55.905975341796875, 66.08953857421875, 128.39491271972656, 41.749603271484375, 1.747314453125, -9.79541015625, 97.45559692382812, 17.68109130859375, 111.56390380859375, 113.12689208984375, 16.53271484375, -9.827354431152344, 45.504638671875, 138.32464599609375, 126.55438232421875, 107.16510009765625, 48.2525634765625, 83.14993286132812, 65.3590087890625, -22.59984588623047, 75.28433990478516, 113.09835815429688, -77.40840148925781, 48.620697021484375, 43.78831481933594, 8.296356201171875, 30.957672119140625, 129.21701049804688, 149.87496948242188, 88.14688110351562, -0.3773193359375, 126.56137084960938, 17.934402465820312, 6.789703369140625, 65.92709350585938, 43.5528564453125, -10.8873291015625, 62.131065368652344, 51.09967041015625, 12.654983520507812, 5.7333221435546875, 57.52168273925781, 12.167633056640625, -55.507843017578125, 0.450469970703125, 24.79278564453125, 162.86572265625, 10.66122817993164, 98.32318115234375, 126.21615600585938, 0.1639404296875, 143.92831420898438, -79.14849853515625, -6.2102203369140625, 167.64901733398438, 12.569381713867188, 0.4169578552246094, -38.09613037109375, 19.200218200683594, 14.770980834960938, 9.64703369140625, 12.086326599121094, -6.58111572265625, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000268.npy"}
{"epoch": 0.5612565445026177, "step": 269, "batch_size": 128, "mean": 39.72681427001953, "std": 68.67146301269531, "min": -179.27255249023438, "p10": -27.033956146240232, "median": 30.62635040283203, "p90": 128.0298324584961, "max": 171.15951538085938, "pos_frac": 0.71875, "sample": [-103.35971069335938, 54.52271270751953, 107.74114990234375, -49.588714599609375, 24.189208984375, -7.231781005859375, 27.332054138183594, 118.272705078125, 8.135406494140625, -37.4993896484375, 96.14299774169922, -0.6364822387695312, -8.920173645019531, 100.52012634277344, -31.12384033203125, 113.03173828125, 5.15020751953125, -4.83184814453125, 120.33928680419922, 86.54898834228516, 171.15951538085938, 152.820068359375, 18.62823486328125, 138.03683471679688, 155.913330078125, -28.079795837402344, 34.124114990234375, 116.06507873535156, 29.5762939453125, 71.20999145507812, 3.5932369232177734, 140.35060119628906, 108.10968017578125, -89.69174194335938, 103.87060546875, 0.0, 32.5345458984375, -25.931262969970703, 90.13055419921875, 103.80813598632812, -62.250160217285156, 17.945465087890625, 134.17483520507812, 133.92913818359375, 5.039070129394531, 84.99160766601562, 139.8927764892578, 0.0, 54.956329345703125, 56.390899658203125, 105.7040023803711, -18.15142822265625, 125.63835144042969, -117.56961059570312, -64.1083984375, 123.04716491699219, 97.48981475830078, -18.42211151123047, -107.63186645507812, 0.34026527404785156, 121.71713256835938, 30.305892944335938, 72.29229736328125, 6.02197265625, 103.35029602050781, -22.5340576171875, 14.093017578125, 51.54571533203125, 116.94891357421875, 11.658088684082031, 54.515380859375, 110.4053955078125, 0.0, 130.73634338378906, 15.384536743164062, 20.410003662109375, -128.97052001953125, 2.661510467529297, 48.5289306640625, 4.432933807373047, -7.685546875, -5.56048583984375, 129.908935546875, 21.45233917236328, 96.4426040649414, 30.946807861328125, 39.607818603515625, 56.560028076171875, 46.592193603515625, 127.22450256347656, 40.17822265625, -22.155548095703125, 10.88360595703125, -17.93426513671875, 116.28411865234375, 46.917633056640625, -17.16168212890625, 109.23542785644531, 4.601081848144531, 96.77764892578125, 43.28240966796875, 40.608306884765625, 1.4171619415283203, -15.833686828613281, 3.3933982849121094, -112.58003234863281, 81.81582641601562, 131.15139770507812, 5.72125244140625, 94.38605499267578, -23.6475830078125, 10.596290588378906, 85.63912963867188, 1.950714111328125, 102.5809326171875, -25.630428314208984, -26.585739135742188, -5.460594177246094, 9.202590942382812, -2.7149429321289062, 95.30169677734375, 130.5230255126953, 61.76325225830078, -179.27255249023438, -16.25897216796875, 12.470237731933594, 125.87411499023438, 152.35104370117188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000269.npy"}
{"epoch": 0.5633507853403141, "step": 270, "batch_size": 128, "mean": 45.57858657836914, "std": 74.43566131591797, "min": -171.24676513671875, "p10": -37.90893249511718, "median": 45.0018367767334, "p90": 131.8198455810547, "max": 204.8924560546875, "pos_frac": 0.734375, "sample": [-140.9339599609375, 66.74569702148438, 98.426025390625, 113.2306900024414, -29.849594116210938, 127.4049072265625, -7.741619110107422, 29.10089111328125, 78.84559631347656, -47.314788818359375, 122.20367431640625, 17.912384033203125, 20.79803466796875, 45.107208251953125, 22.140213012695312, 111.42520141601562, -16.642608642578125, 137.2373809814453, 91.72490692138672, -127.72358703613281, -48.2044677734375, 75.27383422851562, 0.7472915649414062, 35.948333740234375, 3.758892059326172, -1.177886962890625, 114.21270751953125, 110.3552474975586, -83.395751953125, 0.8733158111572266, 43.422576904296875, 121.7939453125, 56.788726806640625, 2.797008514404297, 19.149948120117188, 10.369857788085938, -49.660430908203125, 54.76646423339844, -143.37603759765625, 90.59878540039062, -29.455963134765625, 133.8748779296875, 29.0283203125, 11.821090698242188, 118.93804931640625, 1.4028396606445312, -4.825273513793945, 132.26693725585938, 90.27496337890625, -117.63446044921875, -121.322021484375, 126.48165893554688, 130.84283447265625, 129.80299377441406, 34.85743713378906, 96.70368957519531, 141.41632080078125, -35.224853515625, 123.298095703125, 4.5349884033203125, 29.3221435546875, -18.530563354492188, 11.28070068359375, 13.08090591430664, 125.613525390625, 102.31233215332031, 139.5682373046875, 4.170295715332031, 124.2811279296875, 93.97261047363281, 92.13221740722656, -6.26708984375, -130.91912841796875, 116.04236602783203, -54.80279541015625, 134.12081909179688, 63.74267578125, 44.89646530151367, 100.43415832519531, -28.79364013671875, -14.77679443359375, 11.97100830078125, 80.19369506835938, 119.35812377929688, 20.713031768798828, -5.202171325683594, -1.837432861328125, -13.526107788085938, 1.1675186157226562, -3.7738723754882812, -35.0472412109375, -44.171783447265625, 101.9510498046875, 157.83428955078125, -8.022918701171875, 26.930503845214844, 95.37771606445312, 132.92063903808594, 55.357879638671875, -8.436279296875, 116.34506225585938, 153.984375, 26.588104248046875, 82.8742446899414, 59.546268463134766, -171.24676513671875, 43.930694580078125, -5.18328857421875, 146.47528076171875, 8.599578857421875, 75.49697875976562, 115.1131591796875, 58.15394592285156, 110.87069702148438, 79.61474609375, 141.70098876953125, 131.62823486328125, 54.36656951904297, 10.279876708984375, 145.25955200195312, -20.310699462890625, 204.8924560546875, 105.2607421875, 123.906005859375, 102.0530014038086, 103.3431396484375, 110.057373046875, -28.4010009765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000270.npy"}
{"epoch": 0.5654450261780105, "step": 271, "batch_size": 128, "mean": 44.94878005981445, "std": 62.62157440185547, "min": -101.29410552978516, "p10": -16.455429077148427, "median": 29.91608428955078, "p90": 123.9304946899414, "max": 205.23284912109375, "pos_frac": 0.765625, "sample": [103.97917938232422, -101.29410552978516, -8.557952880859375, 31.52655029296875, 14.505691528320312, -1.544769287109375, 6.348503112792969, -9.118122100830078, 33.65606689453125, -6.242523193359375, 59.138641357421875, 26.541946411132812, 119.06262969970703, 2.242401123046875, 34.90332794189453, 49.284027099609375, 85.0950927734375, 29.052703857421875, 30.367111206054688, 24.641128540039062, 56.61586380004883, 9.954627990722656, 117.49588012695312, 101.18335723876953, 3.3110504150390625, 4.5586700439453125, -42.4932861328125, 107.011962890625, 11.259140014648438, -11.122039794921875, 106.45639038085938, 137.34970092773438, -2.150390625, 39.3060302734375, 3.4905242919921875, 27.0240478515625, 30.705963134765625, 24.686721801757812, 92.18270874023438, -70.95260620117188, 114.7037353515625, 0.721588134765625, 1.54595947265625, 119.08816528320312, 104.12078857421875, 14.530364990234375, 45.56817626953125, 0.39107513427734375, 123.24168395996094, 11.8094482421875, -3.128875732421875, 78.05386352539062, 129.3424835205078, 58.801177978515625, 205.23284912109375, 152.31170654296875, 0.1263446807861328, 131.79031372070312, 28.545867919921875, 107.15542602539062, 96.16229248046875, 9.59613037109375, -5.651588439941406, 183.4029541015625, 107.9110107421875, -50.83447265625, -23.41302490234375, 125.5377197265625, -8.020416259765625, -0.1544189453125, 119.84547424316406, 165.37744140625, -29.97625732421875, 86.869873046875, -13.2315673828125, 21.12359619140625, -48.901084899902344, 30.584136962890625, 37.9388427734375, 89.05736541748047, -23.252593994140625, -67.85519409179688, 132.70346069335938, -2.0297012329101562, 159.1998291015625, 87.96319580078125, 61.763511657714844, -45.003387451171875, 27.178844451904297, 33.474822998046875, 6.840579986572266, 132.79388427734375, 107.22412109375, -13.5423583984375, 4.069061279296875, 103.48197937011719, 3.717435836791992, 100.22513580322266, 114.17742919921875, 76.61229705810547, 17.018539428710938, 23.802658081054688, -2.58746337890625, 100.34762573242188, 150.65887451171875, -7.9623565673828125, -88.11984252929688, 12.69805908203125, 18.401397705078125, 90.1441650390625, 34.7999267578125, 104.16656494140625, 109.76593017578125, 56.054931640625, 39.126953125, 16.878875732421875, 112.46725463867188, 122.29287719726562, -4.45819091796875, -8.761505126953125, 109.94374084472656, -90.96144104003906, 119.66952514648438, 150.5313720703125, 19.421607971191406, 0.1694488525390625, 29.465057373046875, -67.91091918945312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000271.npy"}
{"epoch": 0.5675392670157068, "step": 272, "batch_size": 128, "mean": 40.31340789794922, "std": 64.50022888183594, "min": -132.65615844726562, "p10": -22.202629089355465, "median": 24.95308494567871, "p90": 130.18654174804686, "max": 184.60113525390625, "pos_frac": 0.7578125, "sample": [90.42584228515625, 0.0, -3.267822265625, 32.0106201171875, -102.864501953125, 56.25982666015625, 157.596435546875, 84.340087890625, 111.77552795410156, 9.357009887695312, -24.688446044921875, 22.493816375732422, 48.202911376953125, 114.54345703125, 70.95745849609375, 145.44766235351562, 139.53094482421875, 105.525634765625, 47.631622314453125, 12.8673095703125, 110.33586120605469, 8.902397155761719, -8.963348388671875, -2.697967529296875, 110.16204833984375, -21.013442993164062, 4.447439193725586, 158.0206298828125, 0.992034912109375, 11.3729248046875, 111.2080078125, 62.044776916503906, 70.33346557617188, -0.9660530090332031, 130.00689697265625, 101.76348876953125, 151.34228515625, 130.605712890625, 19.665679931640625, -0.9621505737304688, 81.28123474121094, -43.69805908203125, -75.89303588867188, 62.46073913574219, -9.64483642578125, 31.757461547851562, 104.7369384765625, -45.91943359375, -2.3904342651367188, 1.398019790649414, 135.17022705078125, 109.51130676269531, -15.700883865356445, 19.284439086914062, 106.80426025390625, 149.0316162109375, 16.17169189453125, 44.28009033203125, 96.558837890625, 0.16031265258789062, 4.499734878540039, 8.79833984375, 8.006790161132812, 14.76806640625, 80.05973052978516, 33.29289245605469, 184.60113525390625, 30.540557861328125, 132.83782958984375, 11.900390625, 121.27978515625, -21.432693481445312, 69.73519897460938, 10.78558349609375, -52.27130126953125, -35.12895202636719, -6.598358154296875, 38.92744445800781, 4.843292236328125, 111.26901245117188, 95.13096618652344, 50.75146484375, -9.294525146484375, 156.50535583496094, 2.947357177734375, -132.65615844726562, 18.961008071899414, 13.822589874267578, 31.153778076171875, 3.2100257873535156, 0.2813720703125, 85.11781311035156, 5.5679168701171875, -112.04495239257812, 82.57243347167969, 36.662353515625, 50.67645263671875, 109.76305389404297, -10.544952392578125, 1.6408157348632812, 0.0, -23.9991455078125, 12.269842147827148, -105.3587646484375, -18.59490966796875, 0.94659423828125, 13.620330810546875, 3.3858795166015625, -6.176399230957031, 27.412353515625, 70.76927947998047, 181.569091796875, -56.22821044921875, 59.74280548095703, 122.342041015625, -87.64932250976562, 13.326744079589844, 3.482820510864258, 7.882713317871094, -4.45611572265625, 138.112548828125, 122.2975082397461, 27.76162338256836, 38.18951416015625, 41.40264892578125, 127.57818603515625, 105.11708068847656, 84.32431030273438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000272.npy"}
{"epoch": 0.5696335078534032, "step": 273, "batch_size": 128, "mean": 45.94484329223633, "std": 69.59053039550781, "min": -138.80938720703125, "p10": -35.01552543640136, "median": 34.87602996826172, "p90": 132.96815185546873, "max": 198.27020263671875, "pos_frac": 0.7109375, "sample": [11.89959716796875, 157.023193359375, 144.9228515625, 10.75965690612793, 155.515625, -5.927040100097656, 175.45587158203125, 20.3431396484375, 62.4244384765625, 8.302398681640625, -24.32763671875, 33.33110809326172, -3.003253936767578, 93.68815612792969, 198.27020263671875, -138.80938720703125, 71.2830810546875, 170.22061157226562, 16.1358642578125, 57.180694580078125, -66.21659851074219, 91.67450714111328, 119.36534118652344, 24.055818557739258, 137.42160034179688, 25.203857421875, -110.44058227539062, 33.83876037597656, 103.4774169921875, -14.96282958984375, 16.435806274414062, -39.650634765625, -21.55835723876953, 118.81040954589844, -22.650192260742188, 0.390838623046875, 7.828948974609375, 1.614837646484375, -6.320770263671875, 121.38888549804688, -5.6510009765625, -10.74853515625, 2.4096832275390625, 46.91094970703125, 103.65653991699219, 24.622283935546875, 103.18448638916016, -0.15814590454101562, -55.09319305419922, 3.8206634521484375, 111.87451171875, -22.423095703125, 141.63726806640625, -20.807891845703125, 84.39909362792969, -7.82537841796875, 1.742218017578125, 70.24948120117188, 6.2384796142578125, 53.209556579589844, -44.27191162109375, 106.6848373413086, 1.3204498291015625, 100.57261657714844, -19.88726806640625, 146.24923706054688, 95.17755126953125, 0.5383529663085938, 97.82205200195312, 3.172393798828125, 113.37649536132812, -40.82270812988281, -48.988800048828125, -1.26800537109375, 59.29046630859375, 135.21981811523438, 29.737442016601562, 121.69168090820312, 110.35816955566406, -66.94415283203125, 96.19831085205078, 109.59221649169922, 103.22846984863281, -31.84029769897461, 57.953125, 109.33382415771484, 30.108123779296875, 56.45311737060547, 134.44595336914062, 108.36248779296875, 109.3780517578125, -38.18930435180664, -3.7000579833984375, 111.22796630859375, -134.62982177734375, 35.913299560546875, -95.6817626953125, 117.43055725097656, 132.33480834960938, 14.924819946289062, 147.59771728515625, 47.326011657714844, 125.3201904296875, 123.58274841308594, 128.5654296875, -6.097650527954102, -7.268951416015625, 41.01031494140625, 117.96011352539062, 102.05519104003906, 17.12420654296875, 63.41961669921875, -2.1437835693359375, 42.246124267578125, 22.7708740234375, 61.422149658203125, -62.091766357421875, 137.81057739257812, 25.402618408203125, -7.0119781494140625, -9.526683807373047, 125.1700439453125, 47.00872802734375, 113.46971893310547, 131.73025512695312, -33.65533447265625, 121.15966796875, -20.933029174804688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000273.npy"}
{"epoch": 0.5717277486910994, "step": 274, "batch_size": 128, "mean": 47.84040069580078, "std": 68.06996154785156, "min": -137.23681640625, "p10": -17.514601135253905, "median": 36.27601623535156, "p90": 131.4772979736328, "max": 204.0531005859375, "pos_frac": 0.78125, "sample": [117.67415618896484, 117.37060546875, -44.22808837890625, 69.64151763916016, -0.85565185546875, 65.502685546875, 53.233154296875, 36.80140686035156, 98.5850601196289, 10.876157760620117, 26.429107666015625, 107.31897735595703, 162.21798706054688, 85.9384994506836, 11.1181640625, 104.02548217773438, 174.13656616210938, 146.29640197753906, -64.76593017578125, -16.91299057006836, 25.2054443359375, 143.61956787109375, 119.24984741210938, 39.600982666015625, -10.0711669921875, -4.58197021484375, -106.11762237548828, -30.646575927734375, 12.737213134765625, 33.922271728515625, 18.3800048828125, 31.962158203125, 103.50770568847656, 52.59765625, 116.00619506835938, 33.346771240234375, 41.6044921875, 108.00260925292969, 7.5829315185546875, 130.5477294921875, 145.67556762695312, -8.568208694458008, -8.067596435546875, 1.1267852783203125, -17.113265991210938, 120.83335876464844, 22.143508911132812, 9.704093933105469, 18.033905029296875, 148.556396484375, -9.981666564941406, 9.620697021484375, 111.24844360351562, 44.78916931152344, -121.08810424804688, 27.790420532226562, -44.874481201171875, 43.84405517578125, 1.6942901611328125, 34.058319091796875, 19.38748550415039, 4.830835342407227, -7.428077697753906, 83.79275512695312, -70.4803466796875, -1.693765640258789, 119.28327941894531, -0.46475791931152344, 87.23011779785156, 114.37144470214844, 8.089752197265625, -137.23681640625, 134.376220703125, 87.18409729003906, 60.17822265625, 52.986358642578125, 51.450103759765625, 35.75062561035156, 204.0531005859375, 8.103553771972656, 180.16299438476562, 100.94886779785156, 19.47412109375, 122.13162231445312, -3.4377593994140625, 56.617919921875, 111.49580383300781, 131.0328369140625, 73.70074462890625, 125.51299285888672, 110.31449890136719, 27.562973022460938, -68.49789428710938, 82.59996032714844, 129.17813110351562, 104.71267700195312, 12.03778076171875, 133.56631469726562, 148.63751220703125, -5.549713134765625, -86.429931640625, 16.77203369140625, 6.2244873046875, -15.023269653320312, 23.135055541992188, 110.20137786865234, 10.490425109863281, 125.96922302246094, 47.970977783203125, 52.4197998046875, 132.51437377929688, 104.9682846069336, -47.68145751953125, 64.64924621582031, 0.94970703125, -18.4510498046875, 1.830291748046875, 3.7166748046875, 14.935745239257812, 61.161895751953125, 12.757539749145508, -133.87167358398438, 117.25164794921875, -0.8762397766113281, 24.991809844970703, 102.28302001953125, 116.2489013671875, 172.21041870117188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000274.npy"}
{"epoch": 0.5738219895287958, "step": 275, "batch_size": 128, "mean": 55.19347381591797, "std": 64.35176086425781, "min": -107.98171997070312, "p10": -13.911972045898438, "median": 41.891666412353516, "p90": 135.2961669921875, "max": 170.32095336914062, "pos_frac": 0.78125, "sample": [167.82180786132812, -10.57708740234375, -2.667989730834961, -102.51739501953125, 129.95057678222656, 67.93911743164062, 129.1917724609375, 12.13385009765625, 33.65625, -103.243896484375, 58.30903625488281, 148.71282958984375, -36.5712890625, 135.08526611328125, 0.0, 9.036148071289062, 14.243515014648438, 90.08709716796875, 3.3969593048095703, 4.10125732421875, 0.0, 0.0, 111.93513488769531, 113.27826690673828, 95.41793060302734, 2.9478607177734375, 142.43191528320312, 101.12199401855469, 20.20257568359375, 33.61604309082031, 62.51458740234375, 6.322601318359375, 71.31243133544922, 126.45428466796875, 150.35198974609375, 142.61578369140625, 0.8479423522949219, -1.446502685546875, -107.98171997070312, 42.757606506347656, 130.43658447265625, 30.419036865234375, 81.01622772216797, 26.598419189453125, 16.791778564453125, -5.147621154785156, -28.903060913085938, 124.89311218261719, 135.78826904296875, 17.524444580078125, 131.11251831054688, -77.30999755859375, 2.2549591064453125, 26.874404907226562, 92.88247680664062, 17.556610107421875, 93.38441467285156, 27.501220703125, 131.1905517578125, 0.0, 22.260101318359375, 29.70124053955078, 25.65869140625, 132.2283477783203, -5.072288513183594, 155.69769287109375, 156.94216918945312, 79.64794921875, 59.21931457519531, 0.0, 9.242321014404297, 139.64671325683594, 85.28871154785156, 107.21170043945312, -3.600635528564453, 108.0164794921875, 132.67010498046875, 170.32095336914062, 75.60850524902344, 7.9688720703125, 110.70111846923828, 121.98829650878906, 79.93218994140625, 111.2135009765625, 86.96461486816406, 2.703248977661133, 22.438186645507812, 109.34172058105469, 92.87752532958984, 33.791656494140625, -16.69720458984375, 58.93671417236328, 9.661712646484375, 148.49755859375, -0.001312255859375, 123.5526123046875, -14.112457275390625, 4.7831878662109375, 164.08135986328125, 28.624252319335938, 50.61045837402344, 29.795486450195312, 124.10745239257812, -2.3202285766601562, 128.14974975585938, 117.07508087158203, 13.083267211914062, -44.356109619140625, -0.05377197265625, -14.44525146484375, 132.66497802734375, -52.4091796875, 108.86628723144531, 1.702423095703125, 41.025726318359375, 6.270090103149414, 18.90283203125, 117.63644409179688, 164.14993286132812, 112.48148345947266, -16.56072998046875, 112.2469482421875, 70.33741760253906, -16.75896453857422, 95.47079467773438, -13.8260498046875, 119.27188110351562, 56.0579833984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000275.npy"}
{"epoch": 0.5759162303664922, "step": 276, "batch_size": 128, "mean": 51.05894470214844, "std": 68.10711669921875, "min": -102.18966674804688, "p10": -25.942794799804688, "median": 31.747726440429688, "p90": 136.84339904785156, "max": 212.61282348632812, "pos_frac": 0.7734375, "sample": [212.61282348632812, 17.001087188720703, 110.86177825927734, -3.0076217651367188, 21.56353759765625, -0.77374267578125, 4.08209228515625, 70.63764953613281, 29.780242919921875, 73.82923889160156, 102.61590576171875, 4.573944091796875, 10.369903564453125, 1.3012275695800781, -3.050811767578125, 100.62336730957031, 40.99267578125, 27.067031860351562, -27.80609130859375, 54.110809326171875, 138.49713134765625, 72.73072814941406, 7.8730316162109375, 136.4576416015625, 87.75778198242188, -25.7281494140625, -19.22088623046875, -80.845703125, -7.7967529296875, 121.898681640625, 70.117919921875, 2.5258941650390625, 108.1292724609375, 118.0774917602539, 88.07525634765625, 14.34039306640625, 28.022369384765625, -1.4756183624267578, 114.12289428710938, 42.30909729003906, 20.24273681640625, 23.42864990234375, 124.51641845703125, 25.521697998046875, 6.3256378173828125, 155.1070556640625, 97.50997924804688, 135.84011840820312, 9.935546875, -8.201141357421875, 7.686042785644531, 118.2857666015625, 104.49545288085938, 32.804443359375, 128.068603515625, 11.462081909179688, 51.22639465332031, 23.347869873046875, 130.12884521484375, -40.593505859375, 0.0, -60.399200439453125, 34.51127624511719, -94.02243041992188, 9.793418884277344, -41.625335693359375, 65.22845458984375, 161.23648071289062, -2.8896713256835938, -26.443634033203125, 67.40663146972656, -17.34490966796875, 101.64596557617188, 126.39654541015625, 105.89708709716797, 134.1419677734375, 135.433837890625, -21.441741943359375, 40.219940185546875, 121.86424255371094, 132.5020294189453, 111.30697631835938, 109.43392944335938, 36.4981689453125, 11.928436279296875, 82.90937805175781, -82.08493041992188, 105.37015533447266, -2.9036712646484375, 29.392059326171875, 1.542123794555664, 186.84759521484375, 16.026123046875, 19.936553955078125, 10.350639343261719, 8.882766723632812, 0.2447052001953125, 148.37210083007812, 16.8743896484375, 128.75958251953125, 119.53042602539062, 37.11241149902344, 141.4637451171875, -0.0369873046875, -2.6260833740234375, 100.14410400390625, -16.575422286987305, -51.02153396606445, 30.691009521484375, 137.74349975585938, 6.1497650146484375, 116.66384887695312, 16.442886352539062, 158.89862060546875, 132.50753784179688, -88.15170288085938, 166.44595336914062, 17.9127197265625, 105.89485168457031, -38.080352783203125, 154.31468200683594, 11.054473876953125, 96.03472900390625, -102.18966674804688, -72.78599548339844, 137.9910888671875, 166.54196166992188, 121.28424072265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000276.npy"}
{"epoch": 0.5780104712041885, "step": 277, "batch_size": 128, "mean": 34.67694854736328, "std": 73.95387268066406, "min": -154.4578857421875, "p10": -60.510163879394526, "median": 27.94306182861328, "p90": 122.31078643798827, "max": 206.59796142578125, "pos_frac": 0.6953125, "sample": [0.589599609375, 6.5965576171875, -6.299049377441406, 125.11383056640625, 28.1826171875, -22.707107543945312, -107.14907836914062, -7.834503173828125, 98.49803161621094, 4.8748779296875, 22.8355712890625, 86.51853942871094, 117.62922668457031, 109.4202880859375, -8.33447265625, 112.805419921875, 88.06716918945312, -154.4578857421875, 97.39947509765625, 1.525146484375, 30.890625, 10.2471923828125, 102.00304412841797, 50.6060791015625, 96.45608520507812, 109.99868774414062, -0.9727602005004883, -5.4359283447265625, 128.6822052001953, 5.063163757324219, 84.01117706298828, -131.411376953125, -3.2187461853027344, -35.07263946533203, -23.78271484375, 23.030715942382812, 118.67623901367188, 6.8573455810546875, 90.81138610839844, 72.45306396484375, -153.08712768554688, 2.857147216796875, -21.196998596191406, -24.410308837890625, 9.684600830078125, 83.48406982421875, 28.455474853515625, 119.48736572265625, 3.03204345703125, 45.982635498046875, 66.81561279296875, -3.9516382217407227, 1.932516098022461, 206.59796142578125, 64.8197021484375, 0.0, 51.044281005859375, 88.43241119384766, 104.64604949951172, 28.07696533203125, 153.80523681640625, 84.3802490234375, -110.07577514648438, -5.233774185180664, -0.979766845703125, 2.33489990234375, 90.13977813720703, -97.04788208007812, 181.64935302734375, 177.21121215820312, -27.39205551147461, 40.173828125, 156.37551879882812, 52.64195251464844, 9.959259033203125, -154.0380096435547, 27.337127685546875, -66.26278686523438, -101.04959106445312, 26.5870361328125, 20.76214599609375, 101.1419906616211, 51.215240478515625, -74.96591186523438, 78.452392578125, 98.62535095214844, 103.88240814208984, 121.10948181152344, -69.80853271484375, 141.17904663085938, 2.8372325897216797, 14.131515502929688, -13.160018920898438, 44.787017822265625, 54.90509033203125, -20.48516845703125, 53.13389587402344, 101.63397216796875, 43.47425842285156, 118.77072143554688, 11.621063232421875, 15.1461181640625, 114.26547241210938, 131.66928100585938, 34.38426208496094, 118.18173217773438, 61.58514404296875, 130.79925537109375, -58.04475402832031, 0.0, 131.36273193359375, 164.5987548828125, -8.945777893066406, 60.66789245605469, -45.207305908203125, -10.21795654296875, -118.87835693359375, 127.43309020996094, 27.809158325195312, 98.07553100585938, 96.260498046875, -111.48397827148438, -0.8750476837158203, -2.3541259765625, 4.473041534423828, 21.989410400390625, 30.832069396972656, -24.505483627319336], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000277.npy"}
{"epoch": 0.5801047120418849, "step": 278, "batch_size": 128, "mean": 31.2699031829834, "std": 75.82532501220703, "min": -168.97628784179688, "p10": -85.30426788330078, "median": 17.687103271484375, "p90": 128.07281188964842, "max": 202.83114624023438, "pos_frac": 0.6875, "sample": [122.66278076171875, 96.65559387207031, -89.79953002929688, 59.78617858886719, 1.14569091796875, 23.809219360351562, 57.2391357421875, 17.60040283203125, 1.2647342681884766, 132.52926635742188, 39.693145751953125, 118.2738265991211, 91.916748046875, 31.095779418945312, 98.21717834472656, 110.26361083984375, -112.28952026367188, -17.721145629882812, 25.302017211914062, -5.0360107421875, -4.0269775390625, 76.47988891601562, -8.605255126953125, -24.470779418945312, 13.529144287109375, 172.020751953125, 146.90664672851562, 90.9876480102539, 10.283294677734375, -18.948486328125, -168.97628784179688, 181.83306884765625, -96.46800994873047, 4.9617919921875, 56.38323974609375, 91.17485046386719, 39.80207061767578, -61.888458251953125, 150.25869750976562, 23.725181579589844, 133.90234375, 151.1983642578125, 96.90045166015625, 12.88787841796875, -7.590797424316406, 66.79205322265625, -1.3673095703125, 150.01913452148438, 31.85883331298828, 75.4273452758789, 9.188289642333984, -0.80255126953125, 2.603466033935547, -83.96400451660156, -18.501358032226562, -162.58651733398438, 3.032228469848633, 156.00955200195312, -99.88955688476562, -100.2562484741211, 56.3924560546875, 69.4063720703125, 25.543975830078125, 7.881103515625, 24.406341552734375, 126.16290283203125, -38.63362121582031, 116.84010314941406, 117.51085662841797, 30.619308471679688, 159.2420654296875, 202.83114624023438, -14.616865158081055, -49.75376892089844, 75.40185546875, -101.56385803222656, 101.05422973632812, 37.4930419921875, 71.33822631835938, 41.43670654296875, -131.14137268066406, 5.391578674316406, 2.5108909606933594, 50.089599609375, 106.10734558105469, -6.3770294189453125, 4.126495361328125, 98.79425811767578, 7.07952880859375, 103.78817749023438, -0.0066070556640625, -5.0865478515625, 113.70088958740234, 172.7335205078125, -21.989288330078125, 122.68719482421875, -96.20600891113281, 17.7738037109375, 58.07029724121094, -1.1570281982421875, -11.299026489257812, 7.29669189453125, 22.90899658203125, -96.50118255615234, -17.356781005859375, 24.18792724609375, 137.16033935546875, -16.053993225097656, 11.394256591796875, 14.250225067138672, 111.14733123779297, 3.702821731567383, -12.330810546875, 109.7017822265625, 0.3649711608886719, -27.825416564941406, 12.528190612792969, -88.43154907226562, 90.02012634277344, -37.349517822265625, 4.541465759277344, 13.599288940429688, 12.6810302734375, 124.45204162597656, -118.61277770996094, 66.41177368164062, -7.9127349853515625, 117.55697631835938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000278.npy"}
{"epoch": 0.5821989528795811, "step": 279, "batch_size": 128, "mean": 52.50373840332031, "std": 68.4743423461914, "min": -120.20333099365234, "p10": -15.835596466064445, "median": 36.904029846191406, "p90": 141.49734191894532, "max": 219.228759765625, "pos_frac": 0.8046875, "sample": [43.0982666015625, 112.56985473632812, 43.03910827636719, -22.89794921875, 22.987396240234375, -4.322540283203125, 97.85160827636719, 133.3621826171875, 5.7194366455078125, 121.77047729492188, 81.92620849609375, 160.5867919921875, 38.024658203125, -0.562286376953125, -2.5050048828125, 36.71699523925781, 129.87393188476562, 104.05699920654297, 37.091064453125, 104.84051513671875, 17.568252563476562, 26.4537353515625, -6.6568603515625, 35.341339111328125, 26.545814514160156, 27.204742431640625, 18.113197326660156, 4.658294677734375, -115.13804626464844, 4.1705780029296875, 95.09713745117188, 162.78729248046875, 127.6844482421875, 134.60833740234375, 81.53201293945312, -41.003265380859375, 5.6202545166015625, 100.96990966796875, -27.297897338867188, 10.99041748046875, 106.1802749633789, 124.99755859375, 115.76655578613281, 147.20689392089844, 12.185646057128906, 141.78921508789062, 0.3441619873046875, 10.603240966796875, 6.32440185546875, 88.77973937988281, 23.53948974609375, 29.63201904296875, 11.417388916015625, 28.606891632080078, 43.57395935058594, -7.1690826416015625, 219.228759765625, -120.20333099365234, 105.64175415039062, 159.00723266601562, -13.404403686523438, -13.7071533203125, 15.09014892578125, 24.71990966796875, -43.191741943359375, -9.514236450195312, -22.365188598632812, 123.74409484863281, -2.7608299255371094, 183.87371826171875, -50.68524169921875, 186.24688720703125, 45.363739013671875, 89.62578582763672, 9.73708724975586, -114.52908325195312, 10.86724853515625, 134.54766845703125, 169.51171875, -103.96884155273438, 8.346343994140625, 24.103683471679688, 70.78152465820312, 92.180908203125, 106.50689697265625, 71.00430297851562, 50.390411376953125, 10.549701690673828, 138.54884338378906, 5.222446441650391, 100.45101928710938, -25.108016967773438, 107.67697143554688, 46.397216796875, 22.42207908630371, 19.337356567382812, 137.71563720703125, 2.030843734741211, 49.950164794921875, 134.6582489013672, 112.61215209960938, 141.83016967773438, -10.285171508789062, 126.091796875, 150.33819580078125, 137.82339477539062, 16.067657470703125, 10.3529052734375, 17.19073486328125, 141.37225341796875, -3.914154052734375, 34.79168701171875, 49.27764892578125, 169.40972900390625, 110.89299011230469, 53.784576416015625, 20.454315185546875, 40.05609130859375, -92.38395690917969, -20.801963806152344, 153.80404663085938, -7.756927490234375, 89.56671142578125, 1.35955810546875, 126.34713745117188, 66.37721252441406, 33.71075439453125, 79.8109130859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000279.npy"}
{"epoch": 0.5842931937172775, "step": 280, "batch_size": 128, "mean": 46.239990234375, "std": 72.11186218261719, "min": -146.89593505859375, "p10": -29.033093261718744, "median": 29.387439727783203, "p90": 144.01659545898437, "max": 179.15509033203125, "pos_frac": 0.7421875, "sample": [112.23629760742188, 115.04940795898438, 0.6538848876953125, 120.64590454101562, -4.48823356628418, -146.89593505859375, 143.0729217529297, 118.29931640625, 155.59976196289062, 56.81590270996094, -17.582656860351562, 31.553138732910156, 4.959358215332031, 98.8106689453125, 15.7677001953125, -26.217926025390625, 151.84710693359375, 98.82293701171875, 150.19384765625, -6.36309814453125, 2.695831298828125, 143.65701293945312, -75.02223205566406, 145.37435913085938, -24.6512451171875, 107.79042053222656, 26.840627670288086, 17.6671142578125, -7.856109619140625, 115.49874877929688, 0.0, 21.759719848632812, 75.00946044921875, -33.772918701171875, 45.72955322265625, 179.15509033203125, -24.065956115722656, 135.70428466796875, 12.154193878173828, 3.41986083984375, -6.959869384765625, 91.6788330078125, 9.268173217773438, 21.859710693359375, 101.28716278076172, 67.40742492675781, 107.96286010742188, -107.64199829101562, 144.85562133789062, 19.797271728515625, 44.86128234863281, 10.035911560058594, 6.438232421875, 5.752496719360352, 105.308349609375, 146.7891845703125, 70.172119140625, -27.001739501953125, 2.7602996826171875, 117.42083740234375, 133.60845947265625, 27.22174072265625, 38.188690185546875, -72.6345443725586, -9.078155517578125, 13.635570526123047, -76.32675170898438, 135.24044799804688, 7.1349029541015625, 92.52748107910156, 126.30084228515625, 20.20806121826172, 145.49859619140625, 145.82778930664062, 148.802001953125, -24.30712890625, 99.55148315429688, 84.43704223632812, 25.978118896484375, 50.7552490234375, 99.27426147460938, 129.97064208984375, -22.087005615234375, 112.1407470703125, 132.5555419921875, -8.149848937988281, 21.7103271484375, 10.450851440429688, -74.7535400390625, -0.341278076171875, 138.3984375, 40.1739501953125, 48.62385559082031, -64.47831726074219, 72.13873291015625, 155.96484375, -9.452438354492188, 100.26031494140625, 129.12420654296875, 41.376251220703125, 130.9360809326172, 10.299331665039062, 119.244140625, -18.018630981445312, 19.377792358398438, 16.654586791992188, -0.37732696533203125, -1.19403076171875, 152.16253662109375, 7.637092590332031, 40.25404357910156, 21.273330688476562, 5.513654708862305, 96.17005920410156, -60.569580078125, 91.8812026977539, 118.74272155761719, 170.8670654296875, -50.172332763671875, -139.5085906982422, -15.162109375, 10.437007904052734, -101.91787719726562, 37.93232345581055, 101.34109497070312, 1.3482246398925781, 93.29928588867188, -77.12298583984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000280.npy"}
{"epoch": 0.5863874345549738, "step": 281, "batch_size": 128, "mean": 47.14863204956055, "std": 66.93711853027344, "min": -138.93814086914062, "p10": -18.168560791015622, "median": 32.55536079406738, "p90": 139.39974975585938, "max": 243.139892578125, "pos_frac": 0.734375, "sample": [107.60211181640625, 47.80536651611328, 2.524871826171875, 6.187324523925781, 81.0233154296875, 10.353118896484375, 48.703948974609375, 51.06446075439453, 22.2008056640625, 144.67529296875, 135.078369140625, -102.65927124023438, -110.95933532714844, 151.02005004882812, 17.539379119873047, 26.075408935546875, 108.4328842163086, 109.52593994140625, 3.2383880615234375, 15.864692687988281, 143.2322540283203, 11.556159973144531, -6.6990814208984375, -13.2099609375, 121.27975463867188, -10.600326538085938, -33.127777099609375, 144.68136596679688, 19.71575927734375, 44.514862060546875, 0.0, 51.8267822265625, -2.471221923828125, 68.43386840820312, 18.98879623413086, 130.04905700683594, 40.8392333984375, -6.1995849609375, 151.32456970214844, 99.15054321289062, -138.93814086914062, 65.32986450195312, 124.5396728515625, -11.73301887512207, 0.0, 25.4576416015625, 44.9573974609375, 125.84274291992188, -23.061279296875, 29.559097290039062, 87.1700439453125, -43.73126220703125, 32.60568618774414, 38.69697570800781, 1.678375244140625, 166.9437255859375, -10.001800537109375, 34.685760498046875, 0.6680908203125, 86.61077880859375, 14.694183349609375, 39.63630676269531, -17.5745849609375, -1.5752544403076172, 97.3541259765625, 67.87045288085938, 24.761878967285156, -43.62687683105469, -4.654144287109375, 4.298980712890625, 18.020477294921875, 137.6611785888672, 0.15045166015625, 137.64486694335938, 126.16716003417969, 33.46923828125, 134.8502197265625, 4.437408447265625, 36.8531494140625, 49.98988342285156, 26.90601348876953, 102.61166381835938, 243.139892578125, 141.23648071289062, 131.05426025390625, -47.4139404296875, -9.385940551757812, 103.45098876953125, 56.28594970703125, 4.360282897949219, 173.25799560546875, 32.505035400390625, 101.55284118652344, -4.238880157470703, 20.115402221679688, -43.479248046875, -7.833465576171875, 151.42108154296875, 121.2125244140625, 43.74298095703125, -24.846038818359375, 38.936126708984375, 0.0, 150.88818359375, -3.6978225708007812, -19.55450439453125, 97.18212890625, -69.537841796875, 157.7083740234375, 148.50204467773438, 117.08135986328125, 137.45425415039062, 138.61257934570312, -3.2698974609375, 135.12838745117188, 14.650848388671875, -14.976425170898438, 86.01272583007812, 6.9320068359375, 18.560073852539062, 24.204605102539062, -36.977081298828125, -7.283866882324219, 133.88595581054688, 16.332733154296875, 76.99087524414062, 32.86505126953125, -3.55352783203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000281.npy"}
{"epoch": 0.5884816753926702, "step": 282, "batch_size": 128, "mean": 40.972740173339844, "std": 75.00157165527344, "min": -156.32003784179688, "p10": -43.7519733428955, "median": 22.66469955444336, "p90": 141.23509063720704, "max": 200.90771484375, "pos_frac": 0.734375, "sample": [135.8102569580078, -47.22187805175781, 57.303070068359375, 200.90771484375, 93.22763061523438, 6.440574645996094, -156.32003784179688, 142.5091552734375, 98.83927917480469, -3.4678192138671875, 12.296913146972656, 82.66433715820312, 22.735595703125, -45.677734375, 12.2200927734375, 9.963722229003906, 158.8663330078125, -38.9677734375, 124.24722290039062, -3.72955322265625, 14.39239501953125, 35.9061279296875, -92.49542236328125, 39.7225341796875, 11.82037353515625, 193.51373291015625, 126.99354553222656, 45.9066162109375, -2.990966796875, 14.764404296875, -40.3468017578125, 20.7158203125, 120.236572265625, 64.41258239746094, -6.532745361328125, 20.928359985351562, 97.62472534179688, 15.933788299560547, 45.391754150390625, 37.174560546875, 33.38084411621094, 87.70443725585938, 14.129425048828125, -102.58541870117188, 124.07171630859375, -29.09197998046875, -110.21841430664062, 38.42814636230469, 150.12615966796875, 5.897565841674805, 24.03350830078125, 15.7918701171875, -26.487518310546875, 12.361213684082031, -35.18096923828125, 171.92919921875, 45.90266418457031, 2.4116973876953125, 17.04007339477539, -75.83244323730469, 4.4093170166015625, 8.09124755859375, -19.9609375, 123.3353271484375, 135.11776733398438, 103.21736145019531, 97.7833251953125, -9.95977783203125, -5.8587799072265625, 131.88726806640625, -106.3431396484375, -0.579620361328125, 128.5128173828125, 53.4056396484375, 10.235382080078125, 163.80657958984375, 118.27734375, -42.9266471862793, 83.77490234375, -29.095001220703125, -80.36105346679688, 10.07724380493164, -7.53179931640625, 163.6533966064453, 0.292816162109375, -71.5079345703125, 19.2210693359375, 14.924697875976562, 149.48291015625, 135.17352294921875, 168.4796142578125, 35.643218994140625, 13.618927001953125, 76.34014892578125, -107.97027587890625, 42.0330810546875, 115.76800537109375, 141.6439971923828, 8.573270797729492, -3.8309173583984375, 128.50204467773438, 199.439208984375, 114.84422302246094, 5.678188323974609, 112.58795166015625, -0.304931640625, -141.55355834960938, 22.59380340576172, 38.818695068359375, 34.7629280090332, 129.55078125, 29.261016845703125, 55.045745849609375, 99.91027069091797, 7.902774810791016, 133.15667724609375, 148.17665100097656, 9.018783569335938, 141.05984497070312, 56.18206787109375, 43.2674560546875, -38.00528335571289, -3.855804443359375, -46.147705078125, 106.2259521484375, -4.4510498046875, 5.0607147216796875, 97.39862060546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000282.npy"}
{"epoch": 0.5905759162303665, "step": 283, "batch_size": 128, "mean": 45.28239822387695, "std": 72.11891174316406, "min": -121.66304016113281, "p10": -60.397763061523435, "median": 33.391456604003906, "p90": 132.42157592773438, "max": 194.00970458984375, "pos_frac": 0.734375, "sample": [-19.50347900390625, 28.061721801757812, 137.73876953125, -24.45208740234375, 11.677562713623047, -51.286712646484375, 126.68708801269531, -69.16139221191406, 64.7880859375, -0.776123046875, -0.150543212890625, 129.74462890625, 89.49430847167969, 91.38681030273438, -1.6870574951171875, -70.008544921875, -20.497283935546875, 110.90877532958984, 5.691741943359375, -35.68487548828125, 190.945556640625, 150.26638793945312, 33.595550537109375, -64.1875, 114.96170043945312, 117.48080444335938, 100.20590209960938, -6.3310546875, 73.14227294921875, 14.988067626953125, 87.60163879394531, 41.212188720703125, 18.308853149414062, 119.72297668457031, 34.000267028808594, 101.49456024169922, 6.7170867919921875, 0.6443710327148438, 56.602691650390625, 92.94635009765625, -8.22430419921875, 114.9495620727539, -61.570831298828125, -83.30584716796875, 111.80131530761719, 127.13134765625, 28.5428466796875, 17.0438232421875, -96.79949951171875, 165.97015380859375, 100.2742919921875, 132.33204650878906, 113.15533447265625, 94.0360107421875, 153.3588409423828, 2.0571823120117188, 2.658876419067383, 33.18736267089844, 27.50018310546875, 99.3924560546875, 101.741455078125, 74.426025390625, 139.14889526367188, 154.31549072265625, 132.84857177734375, -6.95281982421875, 132.63047790527344, -0.09283447265625, 45.773529052734375, 12.657058715820312, -86.54688262939453, 110.64158630371094, -79.31938171386719, 144.38442993164062, -4.374237060546875, 77.69342041015625, 126.98429870605469, 106.84932708740234, 0.0, 2.29620361328125, -72.1958999633789, 16.1644287109375, 105.00141906738281, -109.27537536621094, 126.91116333007812, 70.03692626953125, 99.37161254882812, 17.447128295898438, -59.89501953125, 39.39741516113281, 5.271186828613281, 0.785186767578125, 11.247238159179688, -19.5633544921875, 41.340850830078125, 114.07769775390625, 115.2806396484375, 54.72576904296875, -121.66304016113281, 119.45724487304688, 0.6130008697509766, 31.2969970703125, 192.67141723632812, -5.220762252807617, 31.186065673828125, 28.437637329101562, 115.68515014648438, -104.04446411132812, 19.95123291015625, 3.1473846435546875, 153.16458129882812, 18.748779296875, -0.0526275634765625, 2.8832664489746094, 0.0, 98.51715087890625, 114.60021209716797, 194.00970458984375, 12.239982604980469, 5.199167251586914, 68.96609497070312, 110.6600341796875, 63.96405029296875, 40.09556579589844, -15.196533203125, 126.38217163085938, -82.46926879882812, -29.09801483154297], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000283.npy"}
{"epoch": 0.5926701570680628, "step": 284, "batch_size": 128, "mean": 45.2121467590332, "std": 82.60773468017578, "min": -150.91958618164062, "p10": -77.5245620727539, "median": 36.45259475708008, "p90": 146.59117736816407, "max": 247.46841430664062, "pos_frac": 0.7421875, "sample": [-99.29673767089844, 87.45904541015625, -96.20574188232422, -150.91958618164062, 114.80364990234375, 34.4488525390625, 149.13351440429688, 140.8101806640625, -74.62065124511719, 36.34124755859375, 72.9548568725586, 133.31756591796875, 55.810386657714844, 36.563941955566406, 27.40667724609375, 107.26341247558594, 131.87399291992188, 69.9290771484375, 153.728759765625, 7.358978271484375, 126.64736938476562, 216.689697265625, -19.340972900390625, 139.432861328125, -6.796541213989258, 14.716995239257812, 17.501800537109375, -84.30035400390625, 16.412399291992188, -43.959747314453125, -97.36526489257812, 38.79962158203125, 94.15924072265625, -22.67913055419922, 5.5797882080078125, 38.80181884765625, 114.51950073242188, 94.7247314453125, 28.23175048828125, 8.972991943359375, -119.68279266357422, 10.652252197265625, 77.32582092285156, 34.95794677734375, -8.769203186035156, -2.579132080078125, -26.940872192382812, 44.788818359375, 14.158538818359375, 186.37612915039062, 139.81092834472656, -121.38815307617188, 74.32574462890625, 20.398468017578125, 142.13671875, 93.14639282226562, 12.858154296875, -24.16394805908203, 148.64877319335938, 70.83163452148438, 108.63119506835938, 159.02857971191406, 37.36952209472656, -6.339363098144531, 134.70664978027344, 50.72161865234375, 168.84130859375, 29.296142578125, 90.27816772460938, 10.815200805664062, 112.9561767578125, 22.556533813476562, -15.204034805297852, 33.94158935546875, 1.460653305053711, 247.46841430664062, 10.698760986328125, 13.138763427734375, 10.734832763671875, 2.228179931640625, 171.91111755371094, 132.07965087890625, 107.565185546875, 149.719482421875, -59.05303955078125, -132.15103149414062, 7.918243408203125, -20.84625244140625, 52.50669860839844, -52.23407745361328, 109.581298828125, 49.69403076171875, 97.17678833007812, 119.76576232910156, 1.1978569030761719, -2.2205810546875, 48.26348876953125, -114.97413635253906, -1.6331634521484375, 102.5093994140625, 145.7093505859375, 36.006866455078125, 106.67301940917969, 11.95501708984375, 0.8138713836669922, 133.58212280273438, -28.980087280273438, 101.08685302734375, 24.193115234375, 109.24935913085938, 132.18380737304688, 191.77288818359375, 76.47286987304688, -22.6094970703125, -120.4168701171875, -109.07257080078125, 101.74605560302734, -94.66571044921875, 1.995615005493164, -34.6558837890625, 176.22900390625, 62.876007080078125, 108.7891845703125, 188.27508544921875, 112.28742980957031, -2.828662872314453, -96.79351806640625, 99.3719482421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000284.npy"}
{"epoch": 0.5947643979057592, "step": 285, "batch_size": 128, "mean": 34.08090591430664, "std": 69.5055923461914, "min": -132.58749389648438, "p10": -45.1689453125, "median": 21.39985942840576, "p90": 128.24439239501953, "max": 178.60687255859375, "pos_frac": 0.6875, "sample": [16.112030029296875, 95.69418334960938, 113.27569580078125, 27.2314453125, 122.85394287109375, 130.130126953125, 115.8548583984375, -72.0521240234375, 107.46188354492188, 21.30694580078125, 38.13018798828125, -15.333602905273438, 9.622413635253906, -89.66029357910156, -24.7972412109375, 28.09478759765625, 17.533157348632812, -30.77227783203125, -18.10797119140625, -13.518096923828125, -6.26287841796875, 77.833251953125, -103.28961944580078, 17.294036865234375, 31.969078063964844, 29.93780517578125, 161.4166259765625, 45.689910888671875, 92.74232482910156, 38.037322998046875, 178.60687255859375, -107.71957397460938, 10.323272705078125, 38.038360595703125, 131.95135498046875, 133.0072479248047, 123.16217041015625, -38.3892822265625, -30.754653930664062, 59.27691650390625, -44.967987060546875, -31.046539306640625, -28.020965576171875, -6.1531982421875, 2.7916946411132812, 117.04669189453125, 67.294921875, 116.12667846679688, -1.464752197265625, 9.779464721679688, 130.8690948486328, 1.3229427337646484, 12.297195434570312, -78.38745880126953, 147.24679565429688, 2.2009105682373047, -6.06396484375, 0.6608486175537109, 117.51084899902344, 15.462257385253906, 19.281906127929688, 85.82061767578125, 127.66226959228516, 34.08256530761719, 44.86090087890625, 6.151611328125, -52.3582763671875, 6.812826156616211, -57.298736572265625, 49.8619384765625, -129.03408813476562, 128.7396240234375, 30.605636596679688, -4.665184020996094, 99.50469970703125, -9.37957763671875, 53.94708251953125, 21.492773056030273, 31.02667236328125, 16.2076416015625, 100.620361328125, 114.347900390625, 7.63238525390625, 1.8543853759765625, -16.041549682617188, 6.4888916015625, 33.45355224609375, 128.0321502685547, 146.45379638671875, 108.97856140136719, 104.97465515136719, 118.571044921875, -2.8157196044921875, -26.37274169921875, -27.968902587890625, 18.3856201171875, 36.41259765625, 12.399200439453125, -45.637847900390625, 18.28228759765625, 94.91427612304688, -41.726478576660156, 102.20474243164062, -36.092041015625, -39.46466064453125, -27.850738525390625, 36.533203125, -12.270454406738281, 142.28897094726562, 118.98712158203125, -132.58749389648438, -15.977981567382812, 29.351165771484375, -63.617462158203125, -9.7991943359375, -121.92193603515625, 122.51569366455078, 5.101051330566406, 123.4156494140625, 64.95932006835938, 42.58856201171875, 55.604248046875, 99.64254760742188, 134.19918823242188, -60.20086669921875, 101.841064453125, 156.70724487304688, 145.20144653320312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000285.npy"}
{"epoch": 0.5968586387434555, "step": 286, "batch_size": 128, "mean": 38.04480743408203, "std": 74.81420135498047, "min": -135.29180908203125, "p10": -53.390249633789054, "median": 28.49207305908203, "p90": 132.87897949218748, "max": 192.65692138671875, "pos_frac": 0.6796875, "sample": [-13.347930908203125, -25.48614501953125, 101.90975952148438, 15.622848510742188, 12.321739196777344, 35.02326965332031, 7.80389404296875, 8.451980590820312, 7.7774658203125, 85.76116943359375, 59.579261779785156, 18.59417724609375, -4.600547790527344, 84.37776184082031, -119.2222900390625, 152.8712921142578, 11.203048706054688, 148.43350219726562, 109.62191009521484, 145.68008422851562, -30.4857177734375, 133.80862426757812, 12.01873779296875, 25.356643676757812, 17.71971893310547, 26.00262451171875, -20.291900634765625, -6.353691101074219, 112.42025756835938, -10.640968322753906, -135.29180908203125, 10.5369873046875, 127.47400665283203, 120.79246520996094, -14.344085693359375, 32.93816375732422, 9.89727783203125, 87.86984252929688, 10.242794036865234, 80.84959411621094, -10.1859130859375, 132.48056030273438, 111.31729125976562, -93.5252456665039, 83.92109680175781, -45.292724609375, 37.67169189453125, 192.65692138671875, -16.153549194335938, -6.846099853515625, 128.75076293945312, -133.18539428710938, 131.70108032226562, 97.52059936523438, 47.2220458984375, 113.5638427734375, 5.817657470703125, 90.19036865234375, -47.32660675048828, 156.0018310546875, -11.371955871582031, 0.0, -7.048828125, -117.66543579101562, 33.06008529663086, 52.4620361328125, -5.575122833251953, -23.625167846679688, -8.887847900390625, 14.569580078125, -51.091461181640625, 105.27182006835938, 116.91816711425781, 40.824554443359375, -31.475460052490234, 33.494598388671875, 126.68154907226562, 46.147064208984375, -96.01414489746094, 52.029048919677734, 121.53753662109375, 43.983543395996094, 1.4634227752685547, 190.0762939453125, 127.49200439453125, 36.59468078613281, 24.4537353515625, -58.75408935546875, 107.58073425292969, -105.17071533203125, 81.71814727783203, 154.13534545898438, 29.3402099609375, 119.83856201171875, 139.1943359375, 43.08154296875, 90.66963195800781, 92.14930725097656, 99.81884002685547, 95.90545654296875, 159.450439453125, 33.4830322265625, 140.17904663085938, -15.589649200439453, 73.4508056640625, -47.9080810546875, 155.69320678710938, 53.920249938964844, -10.637054443359375, -63.495975494384766, -76.329833984375, 128.97128295898438, -6.3461761474609375, -100.3241195678711, -102.1916732788086, 180.150390625, -89.61477661132812, -3.5103607177734375, 18.342987060546875, 13.381988525390625, 27.643936157226562, 118.81256103515625, -0.12907028198242188, 16.744461059570312, 59.851226806640625, -23.370849609375, 109.84149169921875, 6.258018493652344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000286.npy"}
{"epoch": 0.5989528795811518, "step": 287, "batch_size": 128, "mean": 42.86847686767578, "std": 70.14762115478516, "min": -161.49258422851562, "p10": -35.566253662109375, "median": 41.92522048950195, "p90": 130.0568084716797, "max": 166.47122192382812, "pos_frac": 0.7109375, "sample": [108.09580993652344, -32.42955017089844, -61.841217041015625, -11.11880111694336, -23.22894287109375, -33.84370422363281, 129.18780517578125, 121.58807373046875, 78.3688735961914, 162.05270385742188, 115.09260559082031, 16.010818481445312, 9.564109802246094, 123.83684539794922, 3.8193359375, 18.523162841796875, -26.985382080078125, -74.88351440429688, -35.04545593261719, 42.03553009033203, -23.246170043945312, 166.47122192382812, 119.55245971679688, -2.7161865234375, 46.511138916015625, -22.755645751953125, -12.831398010253906, 40.14306640625, 89.54632568359375, 41.814910888671875, 11.62200927734375, 132.61093139648438, 89.20648193359375, 96.92373657226562, -6.403564453125, 3.724395751953125, 82.55337524414062, 112.51620483398438, -1.7384204864501953, 116.43400573730469, 120.10164642333984, 118.91911315917969, 19.683563232421875, 67.05230712890625, 134.13059997558594, 8.352106094360352, -1.0343170166015625, 123.12496948242188, 112.371826171875, 156.25311279296875, 57.45750427246094, 115.09820556640625, 121.5728759765625, 56.28628158569336, 25.751220703125, 41.621856689453125, 33.74346923828125, -12.218505859375, 101.6234130859375, -52.2919921875, 114.82431030273438, 87.49867248535156, 43.97615051269531, 25.67226791381836, 3.840618133544922, 28.836135864257812, 87.69638061523438, 119.0494384765625, 59.214599609375, 66.60171508789062, -125.38949584960938, -2.208009719848633, 139.8199462890625, 138.74264526367188, 47.98486328125, 94.8419189453125, 113.61410522460938, 131.203857421875, 18.9566650390625, 97.29763793945312, 163.7406005859375, -70.43511962890625, -101.25531005859375, -1.1509475708007812, -36.252838134765625, 59.22704315185547, -30.648696899414062, 85.03944396972656, -37.9716796875, 16.795730590820312, 103.24354553222656, 149.64297485351562, -161.49258422851562, -12.96990966796875, 62.562774658203125, -8.899566650390625, 92.84542846679688, 5.076601028442383, -119.76513671875, 43.22816467285156, 5.395263671875, 14.857757568359375, 7.883613586425781, 36.764892578125, -29.683135986328125, -34.366455078125, 113.13568878173828, -125.062744140625, 151.86993408203125, -12.833137512207031, 134.28768920898438, -50.08954620361328, 115.73805236816406, 49.608489990234375, 143.08726501464844, 7.8311767578125, 16.040786743164062, 63.071868896484375, 75.852294921875, 10.533863067626953, 51.78192138671875, 7.179119110107422, -35.272003173828125, -46.760711669921875, 58.125335693359375, 129.56521606445312, -21.945587158203125, 101.59567260742188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000287.npy"}
{"epoch": 0.6010471204188481, "step": 288, "batch_size": 128, "mean": 41.65445327758789, "std": 73.23181915283203, "min": -131.04217529296875, "p10": -35.35655517578124, "median": 26.994110107421875, "p90": 131.7331817626953, "max": 201.53329467773438, "pos_frac": 0.734375, "sample": [61.610687255859375, 4.570213317871094, -22.467803955078125, -19.499603271484375, 92.91392517089844, 2.354888916015625, -27.062728881835938, -17.547531127929688, 32.358062744140625, 103.19192504882812, 19.51318359375, 158.03543090820312, -3.777252197265625, 112.57522583007812, 103.24066162109375, -7.317535400390625, 8.168998718261719, 48.83500671386719, 22.5662841796875, 0.0, 17.69580078125, -10.47259521484375, 27.9429931640625, 8.979740142822266, 99.23965454101562, -8.612091064453125, -28.983673095703125, 5.77667236328125, -71.30548095703125, 17.998291015625, 165.14682006835938, 120.58584594726562, 32.60059356689453, 88.35334777832031, 7.8015899658203125, 0.0, 8.489921569824219, 42.44425964355469, 7.54644775390625, 20.02996826171875, -6.6050262451171875, 104.759033203125, -28.895263671875, -102.98918151855469, 13.614130020141602, 114.98484802246094, 158.7005157470703, 150.5026397705078, -131.04217529296875, 0.8616485595703125, 129.43539428710938, -7.1063385009765625, 126.916748046875, 3.0380020141601562, 8.35015869140625, -127.7164306640625, 5.049762725830078, 47.0, 119.80039978027344, -125.21878051757812, 200.41925048828125, -79.21928405761719, -43.259368896484375, 88.84396362304688, 175.7540283203125, 17.582550048828125, -13.6092529296875, -50.4722900390625, 124.49874877929688, 6.010887145996094, 126.273681640625, 32.1942138671875, -1.7433319091796875, 131.26998901367188, 11.142852783203125, 11.023910522460938, 20.4560546875, 52.67350769042969, -30.8458251953125, 66.11776733398438, 95.97084045410156, 36.97026062011719, -8.405426025390625, -96.55230712890625, -31.969635009765625, -87.0412826538086, 25.996662139892578, 151.1817626953125, 132.81396484375, 31.235382080078125, 78.72808837890625, 1.9465179443359375, 104.54541015625, -6.293512344360352, 88.50689697265625, 116.25398254394531, 97.42282104492188, -61.110931396484375, 136.53677368164062, 26.04522705078125, 49.332916259765625, 81.81010437011719, -106.90777587890625, 184.84921264648438, 8.728034973144531, 30.523773193359375, 14.885339736938477, 74.876708984375, 46.760833740234375, -11.94354248046875, -60.91949462890625, 91.34757995605469, 30.234634399414062, 129.906005859375, 100.7296142578125, 119.21261596679688, 28.829681396484375, 12.66375732421875, 15.323455810546875, 120.48619079589844, 60.8387336730957, 201.53329467773438, 112.64549255371094, 165.84390258789062, 131.17547607421875, 37.955535888671875, 194.528564453125, 111.66548156738281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000288.npy"}
{"epoch": 0.6031413612565445, "step": 289, "batch_size": 128, "mean": 30.809289932250977, "std": 73.70376586914062, "min": -148.30203247070312, "p10": -53.28834152221679, "median": 14.36767578125, "p90": 129.2782791137695, "max": 166.01034545898438, "pos_frac": 0.6484375, "sample": [-50.6519775390625, -21.709564208984375, 14.6473388671875, -59.439857482910156, -5.502838134765625, 1.650115966796875, 13.80157470703125, -30.2218017578125, 50.84686279296875, -75.00509643554688, 123.432861328125, -5.0218353271484375, 1.1650371551513672, 45.49298095703125, -0.9570159912109375, 66.18589782714844, -30.003890991210938, 137.430419921875, 17.042831420898438, -62.53147888183594, 165.89401245117188, 142.00253295898438, 151.89694213867188, 17.44060516357422, -4.32403564453125, -20.756820678710938, 2.322052001953125, 3.33685302734375, 8.935150146484375, 79.10102844238281, -3.4878082275390625, 118.88427734375, -0.920989990234375, 46.26080322265625, 0.95849609375, 0.0, 8.216354370117188, 87.56976318359375, -27.396331787109375, 14.0880126953125, 11.334081649780273, -24.23309326171875, -114.78390502929688, -127.18682861328125, 75.1585693359375, -27.896026611328125, 54.855224609375, 112.37548828125, 0.324462890625, -4.051300048828125, -8.418792724609375, 112.7841796875, -132.99664306640625, -135.27243041992188, 13.6571044921875, 13.203727722167969, -7.500215530395508, -2.0417861938476562, 47.006256103515625, 160.39523315429688, -93.896240234375, 101.71914672851562, 86.24769592285156, 117.68685913085938, 39.59539794921875, 114.18572235107422, -34.66070556640625, -111.85920715332031, -47.240928649902344, -0.5146484375, 130.96890258789062, -32.71490478515625, 1.520050048828125, 108.74871826171875, 16.613937377929688, 17.92132568359375, 40.374420166015625, 47.655029296875, 105.43276977539062, 6.508148193359375, 92.90823364257812, -21.84246826171875, -148.30203247070312, 138.1925048828125, -19.530746459960938, 104.17571258544922, -13.298431396484375, 128.55372619628906, -41.64952087402344, 135.2796630859375, 13.05670166015625, 153.71414184570312, 118.58668518066406, 10.54205322265625, 34.14741516113281, 100.923828125, 100.817626953125, 8.1295166015625, 93.34019470214844, 80.032470703125, 133.172119140625, 76.16852569580078, 155.29876708984375, 101.7244873046875, -20.868377685546875, 118.85971069335938, 16.67724609375, 42.79542541503906, 104.50631713867188, 166.01034545898438, -112.55902099609375, 118.60660552978516, 125.21327209472656, -34.2059326171875, 126.63188171386719, -62.51885986328125, -116.29043579101562, 52.431480407714844, -16.075164794921875, 126.74669647216797, 27.3592529296875, 27.28607177734375, 144.28453063964844, 0.82989501953125, 17.6290283203125, -6.702007293701172, -19.973297119140625, 31.09698486328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000289.npy"}
{"epoch": 0.6052356020942409, "step": 290, "batch_size": 128, "mean": 35.905067443847656, "std": 70.96113586425781, "min": -145.30697631835938, "p10": -45.137709045410155, "median": 22.015731811523438, "p90": 130.2757308959961, "max": 205.11956787109375, "pos_frac": 0.65625, "sample": [140.33045959472656, 7.98797607421875, -6.71649169921875, 101.79115295410156, 85.35250854492188, -145.30697631835938, -4.0214996337890625, 114.04913330078125, -11.8828125, -26.08306884765625, 161.6893310546875, 121.26092529296875, 45.44488525390625, -14.552192687988281, 119.81703186035156, 105.624267578125, 13.403739929199219, 31.718048095703125, 138.69989013671875, 119.20050048828125, 122.37335205078125, 1.6246414184570312, 2.562652587890625, 0.4210052490234375, 176.67568969726562, -8.1708984375, -89.86485290527344, -96.65643310546875, -19.229248046875, 0.829132080078125, 91.5677490234375, -0.8128471374511719, -5.010833740234375, 49.96063232421875, 142.86325073242188, 154.03387451171875, 130.88990783691406, 49.77173614501953, -4.1783447265625, -101.5142822265625, 105.39871215820312, 49.115333557128906, -0.2440185546875, 119.06021118164062, 58.212554931640625, 0.416595458984375, 0.825469970703125, -78.46688842773438, 49.2883186340332, 74.74819946289062, -15.692047119140625, 4.149200439453125, 53.861480712890625, 90.03333282470703, 18.10778045654297, -28.37139892578125, 41.002227783203125, 108.1439208984375, -9.7545166015625, 21.417083740234375, 13.29901123046875, 14.589031219482422, -13.39434814453125, 110.86669921875, 18.35675048828125, 31.2646484375, 18.11510467529297, 73.744873046875, 27.487960815429688, 93.68939208984375, -29.46759033203125, 33.21942138671875, 2.97003173828125, 26.491180419921875, -19.498214721679688, 161.2855224609375, -6.72956657409668, -45.7977294921875, -3.684173583984375, 29.035614013671875, -2.933868408203125, -13.379867553710938, -44.85484313964844, 129.45895385742188, 116.24862670898438, 124.49945068359375, 129.78335571289062, -11.94305419921875, -132.82998657226562, 7.431671142578125, 86.92337799072266, -30.920455932617188, 107.7083740234375, -22.959747314453125, -91.190673828125, 5.054252624511719, -69.64199829101562, -10.745269775390625, -76.37574768066406, 106.20668029785156, 8.50308609008789, -88.30786895751953, 42.17431640625, 205.11956787109375, 85.9119873046875, 65.56011962890625, 130.01251220703125, 135.16326904296875, 137.27108764648438, 30.725784301757812, -59.87443542480469, 22.6143798828125, 106.45213317871094, 44.837890625, 80.45590209960938, -58.77400207519531, -12.106925964355469, 9.06951904296875, -35.72308349609375, 112.13509368896484, -13.7481689453125, -0.070465087890625, 23.473587036132812, 109.23328399658203, -29.770111083984375, 61.244720458984375, 148.469482421875, 137.21893310546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000290.npy"}
{"epoch": 0.6073298429319371, "step": 291, "batch_size": 128, "mean": 52.391605377197266, "std": 74.37333679199219, "min": -161.89581298828125, "p10": -35.099848937988284, "median": 41.60742950439453, "p90": 146.88441772460936, "max": 188.312255859375, "pos_frac": 0.71875, "sample": [153.39422607421875, 0.15932846069335938, 177.12423706054688, 16.84173583984375, 170.1020965576172, 65.89393615722656, 146.39376831054688, 125.11080932617188, 59.00591278076172, -35.22203063964844, -2.497344970703125, 102.37954711914062, 39.8145751953125, -4.80877685546875, 148.02926635742188, 11.927688598632812, 153.80914306640625, -15.816482543945312, 112.31982421875, 32.84949493408203, 82.56884765625, -2.8826904296875, 95.35830688476562, 17.278411865234375, -10.769317626953125, 129.06832885742188, 43.40028381347656, 7.47784423828125, 105.84939575195312, -27.477935791015625, -133.5941619873047, -16.17211151123047, -38.87257385253906, 17.59048080444336, -161.89581298828125, 27.6922607421875, 108.55450439453125, 86.6466064453125, 49.37457275390625, 7.26727294921875, 144.88522338867188, -31.663070678710938, -91.99530029296875, 72.95976257324219, 140.6139678955078, 32.181732177734375, 4.1951904296875, 95.60990905761719, 90.8626708984375, 31.227981567382812, 127.17074584960938, 107.23844909667969, 95.40702056884766, -18.0919189453125, 188.312255859375, 110.97434997558594, 27.781707763671875, -35.0474853515625, -42.709747314453125, 115.24069213867188, 154.24127197265625, 15.19329833984375, -8.7081298828125, -8.949249267578125, 133.18841552734375, -57.49188232421875, 8.331748962402344, -0.764312744140625, -52.99188232421875, 143.66639709472656, -94.90376281738281, 14.581451416015625, 70.13311767578125, 184.02630615234375, 61.289764404296875, 25.884429931640625, 67.57666778564453, 122.9276123046875, 140.05767822265625, -4.68621826171875, 75.72872924804688, 0.55853271484375, -7.43798828125, -23.61883544921875, 95.94795227050781, 151.193359375, 89.8095932006836, -51.072265625, 0.0, -7.847217559814453, -37.01564025878906, 120.97091674804688, 22.147674560546875, -109.03304290771484, 180.70474243164062, 113.51651000976562, 137.6171875, 46.003082275390625, 148.12841796875, -29.2252197265625, 69.55880737304688, 128.59303283691406, 31.18133544921875, 124.4609375, -1.407623291015625, 63.65545654296875, 18.859344482421875, 32.0394287109375, 175.82147216796875, 32.69354248046875, -80.26776123046875, 117.95291137695312, -8.172309875488281, 90.35698699951172, 118.0498046875, 22.3929443359375, 120.0838623046875, -1.0139942169189453, 154.7314453125, 135.96847534179688, 86.03372192382812, 19.330184936523438, 3.054779052734375, -11.055334091186523, 141.37924194335938, 146.087890625, 110.94450378417969, 26.705184936523438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000291.npy"}
{"epoch": 0.6094240837696335, "step": 292, "batch_size": 128, "mean": 44.80021286010742, "std": 69.73712158203125, "min": -176.486572265625, "p10": -27.567138671875, "median": 31.10580062866211, "p90": 136.78173370361327, "max": 205.3251953125, "pos_frac": 0.765625, "sample": [16.066741943359375, 4.8516998291015625, 110.87097930908203, -61.33551025390625, 84.80936431884766, 114.22379302978516, 127.66520690917969, 186.9599609375, 20.876388549804688, 136.48941040039062, 193.94976806640625, 81.88972473144531, -27.970458984375, -4.2432098388671875, 52.858428955078125, 26.4061279296875, -19.25004768371582, -8.503019332885742, 28.576385498046875, -3.47906494140625, 17.832427978515625, 28.782470703125, 36.961212158203125, 65.66146850585938, 117.50293731689453, 104.63168334960938, 120.02076721191406, 29.018585205078125, 25.02996063232422, -5.397733688354492, 140.7237548828125, 107.55865478515625, 1.5618858337402344, 39.7025146484375, 106.81707763671875, 23.49935531616211, 36.291229248046875, 9.84765625, -173.47393798828125, 10.51983642578125, 62.36671447753906, 59.42479705810547, 119.75830078125, 33.193016052246094, -24.726348876953125, 8.414886474609375, 64.37410736083984, -95.73239135742188, 0.08794403076171875, 137.77850341796875, -54.54718017578125, 151.022705078125, -24.61292266845703, -12.940719604492188, -4.693267822265625, 84.005126953125, 0.83026123046875, -27.394287109375, 80.04437255859375, 13.186172485351562, 111.60923767089844, 22.07977294921875, 124.589599609375, -38.111053466796875, -13.389495849609375, 13.286449432373047, 13.826217651367188, 25.645599365234375, 205.3251953125, 54.5596923828125, 7.47076416015625, 67.05474853515625, 88.36117553710938, 37.53619384765625, 111.47589111328125, 167.214111328125, 18.524906158447266, 137.4638214111328, 27.846851348876953, -13.266563415527344, -51.685760498046875, -176.486572265625, 164.93057250976562, 75.67875671386719, 131.5902099609375, 94.96113586425781, 90.0313720703125, 39.62660217285156, 7.241039276123047, 138.53814697265625, -35.89312744140625, 2.3207473754882812, -25.242156982421875, 17.4461669921875, 76.4518814086914, -40.308631896972656, -10.09675407409668, -25.432205200195312, 18.324663162231445, 144.75228881835938, 93.54977416992188, 105.66503143310547, 18.448776245117188, 41.855186462402344, 27.85138702392578, 76.67990112304688, 63.243682861328125, -0.037933349609375, -19.128311157226562, 11.544158935546875, 10.925689697265625, 47.13995361328125, 41.0234375, 0.470855712890625, 128.43380737304688, 123.55276489257812, 177.53744506835938, -49.76849365234375, 87.64889526367188, 81.74429321289062, 144.77218627929688, -122.79559326171875, -51.54204559326172, 106.61483764648438, 50.3623046875, 124.56808471679688, 106.20437622070312, 27.367141723632812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000292.npy"}
{"epoch": 0.6115183246073298, "step": 293, "batch_size": 128, "mean": 45.958099365234375, "std": 70.71823120117188, "min": -137.8367919921875, "p10": -26.859486389160153, "median": 29.145797729492188, "p90": 145.12611236572263, "max": 192.91552734375, "pos_frac": 0.734375, "sample": [138.46189880371094, -11.495010375976562, 127.87358093261719, 154.97537231445312, 42.669403076171875, -57.8668212890625, 14.755584716796875, 102.19403076171875, 2.4787673950195312, 58.861167907714844, -6.6258087158203125, 115.76329803466797, 38.950836181640625, 23.106185913085938, 47.35144805908203, 14.238601684570312, -75.96541595458984, -1.467376708984375, -40.895050048828125, 12.880218505859375, 3.487346649169922, 1.329315185546875, 151.39785766601562, 117.19076538085938, 17.7886962890625, 0.0, 150.7176513671875, 39.86662292480469, 149.98468017578125, 17.594070434570312, 115.0162353515625, -29.09698486328125, -12.436660766601562, 2.214599609375, 15.6944580078125, -19.204666137695312, 27.56524658203125, 115.0008544921875, -1.719573974609375, 58.16606140136719, 140.06007385253906, -2.442138671875, 154.08676147460938, 129.18768310546875, -17.83892822265625, -117.054443359375, 89.71517944335938, 133.88868713378906, 174.77462768554688, 51.49149703979492, 68.01322937011719, 41.0045166015625, 118.80055236816406, 123.80926513671875, 2.5021514892578125, 121.13789367675781, 43.32769775390625, -9.3016357421875, 6.837860107421875, 20.251190185546875, -2.579071044921875, 143.0438690185547, 129.28561401367188, -20.200803756713867, -120.61884307861328, 46.47563934326172, 14.257526397705078, -43.07501220703125, 91.56929016113281, -3.3736724853515625, -137.8367919921875, -35.451873779296875, 105.10176086425781, 192.91552734375, -67.9190673828125, 11.518463134765625, -10.77813720703125, 116.53396606445312, -17.867462158203125, 39.37255859375, 18.71600341796875, 22.357513427734375, 48.251953125, 15.911338806152344, 138.71905517578125, 29.2994384765625, 51.77081298828125, 124.15533447265625, 105.18609619140625, -25.900558471679688, -18.71368408203125, 154.93798828125, 20.667022705078125, 126.92559814453125, 47.662689208984375, 42.03752136230469, 38.52510070800781, -104.07077026367188, 28.992156982421875, 26.29181480407715, 82.23916625976562, 165.99166870117188, 186.51251220703125, 159.51510620117188, 67.904052734375, 123.29019165039062, 11.730445861816406, 35.525421142578125, -44.83744812011719, 1.4188117980957031, -3.5518341064453125, 20.616729736328125, -2.088592529296875, 76.0784912109375, 191.08749389648438, 188.17535400390625, 20.09295654296875, 28.056564331054688, 115.07546997070312, -55.16717529296875, 102.52182006835938, -9.505226135253906, 8.288803100585938, 97.55158996582031, 43.618202209472656, -8.616180419921875, 39.38189697265625, 26.581573486328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000293.npy"}
{"epoch": 0.6136125654450262, "step": 294, "batch_size": 128, "mean": 41.11857604980469, "std": 72.34480285644531, "min": -127.16244506835938, "p10": -38.63311843872069, "median": 18.816112518310547, "p90": 136.33898468017577, "max": 202.28045654296875, "pos_frac": 0.734375, "sample": [133.080322265625, 111.9576416015625, 129.60336303710938, 0.9317779541015625, -6.84954833984375, 8.950050354003906, 0.9436588287353516, 5.210113525390625, 15.556709289550781, 134.26895141601562, 44.63385009765625, 142.72286987304688, 113.86041259765625, 17.624374389648438, -4.236785888671875, 7.342462539672852, 127.5404052734375, 160.52857971191406, 135.3121337890625, 23.78314208984375, 118.40786743164062, 129.77523803710938, 104.87742614746094, -43.80998992919922, 16.696029663085938, 128.2734375, 11.058258056640625, -21.495513916015625, 12.30560302734375, 17.610877990722656, 160.4390869140625, 16.919921875, 11.412689208984375, 91.6185302734375, -6.226615905761719, 18.4957275390625, -30.34906005859375, -120.77679443359375, 40.4368896484375, 11.56097412109375, 75.78143310546875, 157.2197265625, -46.68194580078125, -105.3094711303711, 143.71405029296875, 58.008094787597656, 104.2022933959961, -5.170166015625, -6.141471862792969, 37.6416015625, 138.82562255859375, -4.71038818359375, 147.18710327148438, 132.6273193359375, 42.60893630981445, 138.73497009277344, 141.45668029785156, -27.561309814453125, 18.87225341796875, 12.624687194824219, 104.9097900390625, 28.138397216796875, 30.771270751953125, 122.73873901367188, -119.70660400390625, 100.01386260986328, 107.19822692871094, 45.804840087890625, 11.153518676757812, 9.50677490234375, 15.81951904296875, 160.01466369628906, 147.3406982421875, 0.0, 129.78216552734375, 202.28045654296875, -106.83171081542969, -6.2123565673828125, -11.993438720703125, 18.509002685546875, -6.848991394042969, 134.92813110351562, 6.363269805908203, -122.70927429199219, -15.798431396484375, 40.7095947265625, -36.414459228515625, 3.923828125, 107.18081665039062, -45.36396789550781, 25.77850341796875, 6.789268493652344, 1.2330398559570312, -121.7880859375, 18.759971618652344, 85.99595642089844, 36.11590576171875, 2.4545822143554688, 4.86383056640625, -14.001296997070312, 37.64398956298828, 115.27657318115234, 12.898345947265625, 19.173789978027344, 93.57072448730469, 117.78236389160156, -77.29541015625, -61.619384765625, 33.2957763671875, -0.3551006317138672, -44.933006286621094, 97.55073547363281, 90.71044921875, 22.63226318359375, 88.71054077148438, 95.38862609863281, -0.336090087890625, -27.70366668701172, 96.72816467285156, -12.994682312011719, 15.4403076171875, -24.406280517578125, 90.3439712524414, 2.5877761840820312, 140.17376708984375, -127.16244506835938, 110.87676239013672, 133.86386108398438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000294.npy"}
{"epoch": 0.6157068062827226, "step": 295, "batch_size": 128, "mean": 44.789886474609375, "std": 77.08121490478516, "min": -128.67071533203125, "p10": -33.98861083984375, "median": 21.985485076904297, "p90": 140.76630554199218, "max": 220.5477294921875, "pos_frac": 0.7265625, "sample": [154.803955078125, 18.049652099609375, -34.579193115234375, -112.90237426757812, -11.006935119628906, 35.523406982421875, -17.0010986328125, 131.5106201171875, -108.35305786132812, -12.9442138671875, -15.849647521972656, 4.507759094238281, 136.26173400878906, 19.75933837890625, -14.352020263671875, -20.344482421875, -117.33535766601562, 107.42349243164062, -106.75784301757812, 33.94461441040039, 107.44924926757812, 176.64044189453125, -11.53271484375, -128.67071533203125, -17.661117553710938, 108.1358871459961, 14.51513671875, -94.79063415527344, 104.68350219726562, -2.0381927490234375, 7.448577880859375, 23.922500610351562, 33.710693359375, 124.21481323242188, 123.11628723144531, 8.301300048828125, 68.96832275390625, 8.996429443359375, 80.0478515625, 89.50372314453125, 179.57933044433594, -10.607330322265625, 113.296630859375, 134.29641723632812, -2.034393310546875, 38.17742919921875, 64.08116912841797, 163.49827575683594, 131.56884765625, 117.49830627441406, -117.31582641601562, -85.48087310791016, 138.94203186035156, 220.5477294921875, 134.2040557861328, 186.248291015625, -26.092010498046875, 98.3716049194336, 51.273193359375, 16.561744689941406, -18.132965087890625, 4.658660888671875, 19.705551147460938, 2.5051345825195312, 114.7176513671875, 76.52913665771484, 18.245323181152344, 65.17485046386719, 117.7088623046875, 18.723587036132812, 4.9148406982421875, 8.7161865234375, 0.36727142333984375, 139.86276245117188, -31.720184326171875, 101.16728210449219, 135.5452117919922, 122.60282897949219, 13.155868530273438, 95.48900604248047, 8.835502624511719, 20.432022094726562, 64.19114685058594, 46.029090881347656, 7.35040283203125, 22.179351806640625, 0.0, 91.646484375, 8.717910766601562, 15.7047119140625, 139.26095581054688, 167.93185424804688, 25.728836059570312, 44.821868896484375, 122.44253540039062, 0.2513923645019531, -91.76251220703125, 146.20233154296875, 66.40435791015625, 139.4893035888672, 98.00529479980469, -18.353538513183594, 9.363998413085938, 20.94476318359375, 27.81121826171875, -108.65667724609375, 125.82027435302734, 15.418363571166992, -0.1706390380859375, 96.3720703125, -32.69318771362305, 131.22923278808594, 189.62225341796875, 143.80291748046875, -0.5314788818359375, 164.45022583007812, 1.092306137084961, -5.86895751953125, 21.79161834716797, -46.690673828125, -37.66046142578125, 6.04949951171875, 142.87457275390625, -18.17694091796875, 118.58120727539062, 149.82518005371094, 74.8594970703125, -33.735504150390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000295.npy"}
{"epoch": 0.6178010471204188, "step": 296, "batch_size": 128, "mean": 61.38551330566406, "std": 76.07234191894531, "min": -147.55560302734375, "p10": -26.62777862548828, "median": 62.92547607421875, "p90": 152.91697387695314, "max": 183.0479736328125, "pos_frac": 0.796875, "sample": [111.57946014404297, 39.507076263427734, 20.3094482421875, 114.70579528808594, 116.94425964355469, -23.37213897705078, 2.2108917236328125, 0.5791530609130859, 83.28524780273438, 5.118541717529297, 106.5321044921875, -83.900390625, 44.255889892578125, 43.016937255859375, 9.840850830078125, 126.50222778320312, 134.25848388671875, 99.2279052734375, 117.35186004638672, -62.8682861328125, 129.09413146972656, 82.11724853515625, -9.4921875, 121.21949768066406, 61.538330078125, 151.94049072265625, -8.491924285888672, 17.884845733642578, 142.3157958984375, -41.738861083984375, 13.194686889648438, 90.60445404052734, 166.61517333984375, 5.157440185546875, -121.67265319824219, 164.17477416992188, 50.09912109375, 129.3896484375, 138.02239990234375, 37.183616638183594, -16.801223754882812, 25.931976318359375, 159.9222412109375, 127.84098052978516, 42.032920837402344, -28.62255859375, 30.938934326171875, -25.772872924804688, 177.2135009765625, 15.616653442382812, 157.9735107421875, 76.59010314941406, 30.154541015625, 134.30413818359375, 152.73663330078125, 177.46759033203125, 11.372764587402344, 161.730224609375, 64.3126220703125, 116.99659729003906, -3.992034912109375, -14.755271911621094, 118.06172180175781, -42.9747314453125, 49.41741943359375, 103.42939758300781, 27.458343505859375, 142.6727294921875, -121.76518249511719, 150.1748046875, 133.6055908203125, 49.21571350097656, -112.3428955078125, 125.48773956298828, 81.64781188964844, 40.875732421875, 165.13250732421875, 178.76760864257812, -2.857341766357422, 113.42022705078125, 92.405517578125, 128.30975341796875, 54.678138732910156, -147.55560302734375, 149.44754028320312, 108.11357116699219, 1.5645904541015625, -87.26907348632812, 153.3377685546875, -8.222076416015625, 116.35372924804688, 6.582977294921875, 103.23397827148438, -41.460052490234375, 38.272216796875, 0.0, 126.57821655273438, 122.26446533203125, 123.88726806640625, 106.85894775390625, 24.223663330078125, 140.8734130859375, 64.39305114746094, 125.03713989257812, 16.365631103515625, 38.052520751953125, 11.931427001953125, 11.79205322265625, 42.0478515625, 122.5582275390625, 164.30413818359375, 87.24128723144531, 8.731630325317383, -32.43011474609375, -5.765602111816406, 132.35980224609375, 127.45977783203125, 0.88299560546875, -11.904693603515625, 153.8179473876953, 3.591033935546875, 96.0469970703125, 22.3134765625, 150.31851196289062, 74.49125671386719, 183.0479736328125, -131.3685302734375, -7.3093109130859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000296.npy"}
{"epoch": 0.6198952879581152, "step": 297, "batch_size": 128, "mean": 47.55979919433594, "std": 72.87791442871094, "min": -142.09518432617188, "p10": -24.107655334472657, "median": 29.5452880859375, "p90": 145.32775268554687, "max": 179.30142211914062, "pos_frac": 0.6875, "sample": [170.92221069335938, 20.242965698242188, 127.88600158691406, 145.707275390625, -7.081573486328125, -15.482044219970703, 130.04611206054688, 51.23419189453125, 144.912109375, -22.35919189453125, -1.410125732421875, -2.59765625, -22.811721801757812, 4.1724700927734375, 154.19757080078125, -85.47940063476562, -9.905644416809082, 32.455413818359375, 132.39820861816406, 96.09819030761719, 12.085975646972656, 32.035888671875, -3.5238723754882812, 49.736083984375, -28.79833984375, -3.240264892578125, -21.13714599609375, -142.09518432617188, 17.5350341796875, -16.068119049072266, -51.86463928222656, 87.13117980957031, 29.125457763671875, 83.50647735595703, 16.32666015625, 7.647613525390625, 140.31805419921875, 0.2125244140625, 176.74520874023438, 21.590576171875, 4.4174041748046875, 20.464725494384766, 56.706207275390625, -78.72549438476562, 2.156768798828125, -16.85479736328125, -1.11407470703125, 39.07977294921875, -111.38262939453125, 140.04888916015625, 23.166229248046875, -23.857208251953125, -0.80322265625, -34.6287841796875, 0.55364990234375, 179.30142211914062, 132.13009643554688, 107.86787414550781, -3.7606449127197266, -23.497344970703125, 88.71533203125, 126.47189331054688, -116.164794921875, 25.4088134765625, 162.0637969970703, 43.61151123046875, 14.491851806640625, -65.59580993652344, 158.248046875, -18.222381591796875, 27.345428466796875, 106.30596923828125, 154.97564697265625, 148.2714385986328, 167.78759765625, -59.53363037109375, 48.082088470458984, 5.5783233642578125, 105.98907470703125, -7.060558319091797, -88.55886840820312, 129.82765197753906, 74.88275146484375, -6.841941833496094, 1.9233341217041016, 28.134597778320312, 34.48915100097656, -9.788078308105469, 140.84104919433594, 6.14886474609375, 119.3017578125, -21.68475341796875, 71.1842041015625, 119.71839904785156, 98.44171142578125, 97.56028747558594, 6.6046600341796875, 144.0821990966797, -2.3617401123046875, 158.763916015625, 106.48977661132812, 121.27691650390625, 71.3238525390625, 134.39031982421875, -8.082527160644531, 101.55783081054688, 91.63272094726562, 6.65283203125, 102.4448471069336, 117.97866821289062, 161.0125732421875, 122.40446472167969, 6.49072265625, 93.29998779296875, -0.072540283203125, -24.692031860351562, -11.952056884765625, 71.4423828125, 121.13444519042969, 31.327728271484375, 29.965118408203125, 127.84945678710938, 159.47064208984375, -46.126953125, -3.315399169921875, 59.9688720703125, 117.49425506591797, 145.16510009765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000297.npy"}
{"epoch": 0.6219895287958115, "step": 298, "batch_size": 128, "mean": 53.278541564941406, "std": 79.76042938232422, "min": -162.34716796875, "p10": -43.128515625, "median": 36.870635986328125, "p90": 151.82868652343748, "max": 287.58868408203125, "pos_frac": 0.7421875, "sample": [198.04673767089844, 133.28927612304688, -19.120391845703125, 51.44677734375, -43.2435302734375, -48.446929931640625, 4.72601318359375, -14.760650634765625, -24.757080078125, 154.07147216796875, 90.03413391113281, -11.1295166015625, 9.152938842773438, 14.52117919921875, 117.09059143066406, -56.3858642578125, 140.54916381835938, 78.30350494384766, 57.47251892089844, 110.28653717041016, 141.401123046875, 128.08663940429688, 39.353790283203125, 194.1973876953125, 4.4969329833984375, -34.562255859375, 126.49261474609375, -108.35905456542969, 130.87860107421875, 72.06916809082031, 140.5179443359375, -35.895721435546875, 123.40357971191406, 73.78349304199219, 33.46710205078125, -56.64385986328125, 7.812904357910156, 70.47854614257812, 135.20135498046875, -64.30024719238281, 119.24171447753906, 0.741180419921875, 120.40966796875, -19.159515380859375, 37.71881103515625, 161.475341796875, 40.3841552734375, 38.01936340332031, -95.51011657714844, -30.987335205078125, 30.195098876953125, 131.78302001953125, 21.3316650390625, 142.52841186523438, -60.96551513671875, 64.27239990234375, 123.14820861816406, 155.72103881835938, 13.846588134765625, 119.4534912109375, 181.20718383789062, -162.34716796875, 129.79803466796875, -14.545730590820312, -17.352798461914062, 186.65957641601562, 150.26901245117188, 44.99822998046875, -66.5732421875, 0.0, 6.4482421875, 8.463333129882812, 173.927001953125, 72.26889038085938, 24.8466796875, 12.392803192138672, 118.971435546875, 39.239776611328125, 6.008552551269531, 133.1322021484375, -48.24571228027344, -20.21279525756836, 15.101825714111328, 73.17877197265625, 123.86028289794922, 6.303630828857422, 143.7599334716797, 134.45565795898438, 130.78062438964844, 8.649749755859375, 202.76025390625, 36.0224609375, -26.916770935058594, -20.2825927734375, 15.20916748046875, 31.0269775390625, 35.3319091796875, 13.540771484375, -5.601612091064453, -8.7587890625, -7.389678955078125, 131.29678344726562, 46.49967956542969, 135.90606689453125, 191.53582763671875, 166.97134399414062, -21.32476806640625, 103.9539794921875, 287.58868408203125, 6.6985626220703125, 139.7078857421875, 33.3494873046875, 3.5378360748291016, 34.31913757324219, 150.86749267578125, 22.78271484375, -11.1636962890625, 128.877197265625, 41.72161865234375, 168.02215576171875, 4.72453498840332, 83.79672241210938, 17.38641357421875, 26.989044189453125, -67.45928955078125, -43.0792236328125, 77.63261413574219, -78.54579162597656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000298.npy"}
{"epoch": 0.6240837696335079, "step": 299, "batch_size": 128, "mean": 42.58831787109375, "std": 66.69744873046875, "min": -140.26089477539062, "p10": -30.66375732421875, "median": 21.99187469482422, "p90": 134.39334106445312, "max": 178.10464477539062, "pos_frac": 0.71875, "sample": [100.15989685058594, 132.20547485351562, 17.53145980834961, 96.69261169433594, 21.03570556640625, 3.384490966796875, 133.93017578125, -103.28302001953125, 99.14923095703125, -140.26089477539062, 153.72500610351562, 40.90533447265625, 20.309776306152344, 105.10458374023438, -52.97283935546875, 130.77511596679688, -2.415496826171875, 37.91692352294922, 12.601715087890625, -125.33470153808594, 125.5595932006836, -1.283050537109375, 56.916595458984375, -88.50212097167969, 141.590576171875, -4.7250518798828125, -52.55924987792969, -30.564056396484375, 116.65119934082031, 19.604156494140625, 99.83331298828125, -6.955078125, 137.61961364746094, -2.12982177734375, -48.15074157714844, 93.46124267578125, 109.0548095703125, -11.822998046875, -57.76747131347656, 102.34576416015625, 10.300506591796875, 162.57432556152344, 22.948043823242188, 11.119098663330078, -8.280609130859375, 0.63995361328125, 141.34683227539062, -4.68341064453125, -18.508438110351562, -66.35006713867188, -33.3470458984375, 13.413543701171875, 8.860816955566406, 16.249801635742188, 112.3297119140625, -81.31930541992188, 159.768798828125, -6.444976806640625, 171.99456787109375, 88.52339935302734, 113.08494567871094, 49.147125244140625, 17.608108520507812, 76.61126708984375, -3.3936119079589844, 178.10464477539062, -4.19012451171875, 10.861799240112305, 107.39776611328125, 20.733978271484375, 110.9764404296875, 8.729461669921875, 58.39306640625, 5.921875, 110.78175354003906, 106.3154296875, -7.725288391113281, -8.87533950805664, -6.5822906494140625, 126.11663818359375, -6.207672119140625, 80.300048828125, 56.51617431640625, 14.26348876953125, 35.699066162109375, -6.0091094970703125, 135.52471923828125, 44.167266845703125, 58.92152404785156, 94.91896057128906, 14.950958251953125, 105.3226318359375, 140.48634338378906, 17.274200439453125, 11.342395782470703, -11.121383666992188, 39.58625793457031, 5.393836975097656, 119.47296905517578, -8.588645935058594, -30.896392822265625, 50.388275146484375, 31.722442626953125, 11.179962158203125, 124.46327209472656, 173.24002075195312, 13.886520385742188, 135.47406005859375, 28.445281982421875, -1.40130615234375, 12.780975341796875, 2.317577362060547, 121.5964584350586, -32.95442199707031, 7.42938232421875, 11.423004150390625, -3.379150390625, 32.89593505859375, 69.82099914550781, 38.09857177734375, -13.1087646484375, 105.56927490234375, 146.68812561035156, 59.784332275390625, 36.5133056640625, 126.76235961914062, 23.7332763671875, 76.12625122070312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000299.npy"}
{"epoch": 0.6261780104712041, "step": 300, "batch_size": 128, "mean": 45.97510528564453, "std": 77.542236328125, "min": -125.85604858398438, "p10": -52.57850189208984, "median": 39.651268005371094, "p90": 146.4629699707031, "max": 222.5797119140625, "pos_frac": 0.6796875, "sample": [148.94638061523438, 111.02192687988281, 21.218246459960938, -77.02836608886719, 22.331172943115234, -26.953521728515625, 105.00387573242188, 91.55934143066406, 21.01167106628418, -48.1090087890625, 159.2738037109375, 127.16952514648438, 144.05874633789062, 22.242481231689453, -4.158668518066406, 135.70620727539062, -111.95364379882812, 23.74578857421875, 76.22416687011719, -6.4512939453125, 151.48782348632812, 115.0791015625, 50.292144775390625, 83.17935180664062, 72.88655090332031, 7.492279052734375, 37.80841064453125, -13.8873291015625, 83.69110107421875, -46.12248992919922, 74.02719116210938, 63.316314697265625, 132.18023681640625, 118.27935791015625, 158.3768310546875, -40.270751953125, -125.85604858398438, 16.06146240234375, 3.7444849014282227, 185.23934936523438, 56.993499755859375, -21.040313720703125, -36.990234375, 0.21728515625, 100.09841918945312, 130.17727661132812, -9.01153564453125, 134.27716064453125, 84.20088195800781, 170.21322631835938, 12.095760345458984, -117.68025207519531, -80.85879516601562, 25.0286865234375, 34.17173385620117, -4.417110443115234, 51.17156982421875, 109.81846618652344, 145.39865112304688, -8.623886108398438, 123.6801528930664, 127.9464111328125, -72.35116577148438, 155.51844787597656, 47.25628662109375, -5.760345458984375, 2.9278564453125, 101.64974975585938, 106.20736694335938, 128.28651428222656, 58.71307373046875, 101.84974670410156, -0.6724681854248047, 16.1258544921875, -18.742202758789062, -95.35218811035156, 163.96446228027344, 56.085693359375, -51.72972106933594, 132.92227172851562, 140.40780639648438, 12.790531158447266, 47.74798583984375, -54.558990478515625, 116.34180450439453, 19.328338623046875, -2.30731201171875, 0.0, -13.467758178710938, 41.49412536621094, 170.74365234375, -66.82037353515625, -120.95947265625, 108.42144775390625, 3.8053741455078125, 136.26870727539062, 70.81379699707031, 20.01190185546875, 175.362548828125, 122.997802734375, 6.2330322265625, 222.5797119140625, 6.3799285888671875, 70.53091430664062, -1.635009765625, 101.02847290039062, 115.47216796875, -59.35528564453125, -32.31207275390625, -15.61663818359375, 79.56723022460938, -9.167402267456055, 83.43121337890625, -24.196914672851562, 63.92974853515625, 115.98158264160156, 66.56283569335938, -0.3233642578125, -97.91558837890625, -1.402618408203125, 175.4622802734375, 1.375732421875, -77.67550659179688, -28.200439453125, 130.0758056640625, 148.94644165039062, -6.38330078125, 7.388092041015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000300.npy"}
{"epoch": 0.6282722513089005, "step": 301, "batch_size": 128, "mean": 42.174774169921875, "std": 74.6893081665039, "min": -178.79461669921875, "p10": -31.221505737304685, "median": 24.131187438964844, "p90": 147.0352828979492, "max": 209.58197021484375, "pos_frac": 0.7109375, "sample": [11.965438842773438, 169.1898193359375, 129.63816833496094, 84.5521240234375, 19.43439483642578, 137.55987548828125, 40.70265579223633, -34.65965270996094, 100.41534423828125, 8.880716323852539, 3.7498130798339844, -1.4764404296875, 133.94070434570312, 30.072357177734375, 3.078369140625, 4.383766174316406, -19.432769775390625, -42.1956787109375, 118.56396484375, 161.8616943359375, 15.145832061767578, 18.819412231445312, 114.4232177734375, 22.793609619140625, 46.106048583984375, 18.20440673828125, -0.506072998046875, 49.474365234375, 8.178558349609375, 117.3951416015625, 114.95700073242188, -10.088653564453125, 0.0, 156.5045623779297, 125.879638671875, 86.0894546508789, -14.25909423828125, 31.77703857421875, -20.933807373046875, 123.50802612304688, -24.911474227905273, 97.88331604003906, 156.1164093017578, 32.69841766357422, -149.24951171875, 5.885162353515625, -50.2994384765625, 113.04559326171875, -30.448211669921875, 22.850494384765625, 74.87286376953125, -12.052490234375, -59.4127197265625, 145.44129943847656, 171.6990966796875, 16.31317138671875, 36.5745849609375, 157.90542602539062, -0.2125396728515625, 209.58197021484375, 64.41593933105469, 117.36947631835938, 21.613616943359375, 29.847671508789062, -64.80740356445312, -75.21302795410156, 11.60321044921875, -178.79461669921875, -123.27520751953125, 85.18234252929688, 94.82124328613281, 33.39617919921875, -16.9700927734375, 7.6905517578125, 10.902847290039062, 26.683837890625, 138.95181274414062, 22.8240966796875, 151.0086669921875, -76.11032104492188, 5.6964263916015625, -14.1756591796875, 172.06459045410156, 40.988250732421875, 150.75457763671875, 25.8006591796875, -5.368865966796875, -23.438919067382812, 114.71630859375, -109.50679016113281, 133.73504638671875, 36.1102294921875, -17.681488037109375, 2.0016937255859375, 113.71905517578125, -30.588043212890625, 90.04165649414062, 22.608917236328125, 76.5440444946289, 113.17974853515625, 121.80476379394531, 135.14100646972656, 154.4456787109375, -9.524307250976562, 28.96234893798828, 142.02108764648438, 25.339111328125, -20.984407424926758, 156.37660217285156, -15.450553894042969, 18.637542724609375, -3.34307861328125, -11.28753662109375, 2.5425567626953125, 3.8420333862304688, 107.35017395019531, -104.976318359375, 31.170825958251953, 3.7158737182617188, 134.5177764892578, 166.78579711914062, -32.6995849609375, -21.227142333984375, 77.01097869873047, 22.923263549804688, 32.331268310546875, 108.26719665527344, -11.6387939453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000301.npy"}
{"epoch": 0.6303664921465969, "step": 302, "batch_size": 128, "mean": 51.955570220947266, "std": 78.35281372070312, "min": -149.93862915039062, "p10": -44.61460189819334, "median": 47.14912414550781, "p90": 146.37249450683592, "max": 184.84298706054688, "pos_frac": 0.7421875, "sample": [150.43983459472656, 53.661834716796875, 92.86885070800781, 151.14898681640625, 22.088104248046875, 111.15418243408203, 0.0, -9.150070190429688, -0.624176025390625, -61.57099914550781, -133.44015502929688, 64.24880981445312, 41.38574981689453, 135.03936767578125, -6.73291015625, 54.427215576171875, 67.30792236328125, 46.871185302734375, -15.515579223632812, -87.07758331298828, 156.44778442382812, 138.55609130859375, 164.51812744140625, -10.048828125, 102.2701416015625, 125.29147338867188, 178.78228759765625, 16.99737548828125, -104.7677001953125, 108.39080047607422, 43.13298034667969, 136.41519165039062, 122.7847900390625, 78.11404418945312, 73.8232421875, 97.98941040039062, 168.90335083007812, 72.34056091308594, 46.75306701660156, 47.42706298828125, 122.54385375976562, -1.4504623413085938, -149.93862915039062, -3.869415283203125, 90.74466705322266, 57.7669677734375, 174.0677490234375, 25.002334594726562, 161.10302734375, 141.59678649902344, -10.451629638671875, 139.95530700683594, 14.472602844238281, 3.8619537353515625, -8.823204040527344, -77.83413696289062, 116.05189514160156, 6.032684326171875, 118.39910888671875, 39.20680236816406, 104.1588134765625, 98.90978240966797, 44.6953125, 4.323822021484375, 60.670257568359375, 9.6114501953125, 21.3826904296875, 12.809310913085938, -5.7778778076171875, -92.02469635009766, 38.82909393310547, 150.01205444335938, -23.853973388671875, -107.8544921875, 135.79803466796875, 116.90945434570312, -34.077674865722656, 153.87698364257812, 122.05618286132812, 102.48778533935547, 23.990890502929688, 45.9130859375, 133.42953491210938, -110.30584716796875, 144.01742553710938, -6.469696044921875, -130.82437133789062, 76.01303100585938, -40.78509521484375, -105.32431030273438, 15.232772827148438, -100.50506591796875, -2.76727294921875, 117.01422119140625, 7.79193115234375, 44.906707763671875, 105.73712921142578, 141.5982666015625, 142.18505859375, -53.55011749267578, 2.136199951171875, 25.858001708984375, 55.597023010253906, -20.215679168701172, 127.16516876220703, 0.7463283538818359, 30.4752197265625, 184.84298706054688, 82.93765258789062, 109.86550903320312, 22.51422119140625, 7.968223571777344, 49.77545166015625, -16.345550537109375, 133.75436401367188, 100.41688537597656, 177.26077270507812, 3.4883689880371094, 124.10513305664062, 36.326171875, 18.94207763671875, 179.51617431640625, 101.55262756347656, 138.23138427734375, 96.53237915039062, -7.789031982421875, 144.81268310546875, -19.4884033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000302.npy"}
{"epoch": 0.6324607329842932, "step": 303, "batch_size": 128, "mean": 47.601318359375, "std": 78.38227844238281, "min": -158.75970458984375, "p10": -40.186215209960935, "median": 42.737754821777344, "p90": 136.43376770019532, "max": 195.67486572265625, "pos_frac": 0.7578125, "sample": [14.32269287109375, 24.481788635253906, -72.23886108398438, -97.57305908203125, 114.34014892578125, -7.5343170166015625, 110.46157836914062, 52.1781005859375, 128.90292358398438, 110.26146697998047, -49.3597412109375, 127.1115951538086, 143.82855224609375, 134.72406005859375, 38.077728271484375, -0.769439697265625, -39.765594482421875, 117.32148742675781, -11.53532600402832, 153.0677490234375, 108.24644470214844, 20.572723388671875, 116.5345687866211, 33.557861328125, 103.07498168945312, -41.16766357421875, 97.79112243652344, -10.83892822265625, 156.44775390625, 14.667640686035156, 135.26943969726562, 131.5762481689453, -39.05433654785156, 130.8787078857422, -9.443618774414062, 13.1807861328125, 1.4776763916015625, 44.22265625, 0.0, 7.9756317138671875, -107.92626190185547, 54.92170715332031, -12.964744567871094, 25.269790649414062, 9.763816833496094, 1.452911376953125, 134.73081970214844, 45.839599609375, 83.0102310180664, 114.27789306640625, 57.414520263671875, -111.84803771972656, 119.83480834960938, -158.75970458984375, 125.40957641601562, 136.0702667236328, 149.11282348632812, 88.91790771484375, 42.781402587890625, 119.75042724609375, 0.08949661254882812, -2.8727264404296875, 25.1468505859375, -8.308837890625, -64.9949951171875, 7.368865966796875, 88.22784423828125, 116.96836853027344, 137.2819366455078, 44.43571472167969, 156.33132934570312, -124.07757568359375, 19.577484130859375, -1.4447174072265625, 129.40423583984375, 110.2679443359375, -1.3544158935546875, 12.386871337890625, 135.5574951171875, 189.328857421875, 78.22985076904297, -157.6607666015625, 114.47862243652344, -125.52786254882812, 7.480571746826172, 90.599365234375, 116.004638671875, 35.13111877441406, 80.54025268554688, 80.2945556640625, 11.9083251953125, 0.8380126953125, 144.64051818847656, 0.939208984375, 151.02987670898438, 2.7087860107421875, 2.8556900024414062, 82.44023132324219, -30.304931640625, 2.078948974609375, -29.16400146484375, -144.06033325195312, 54.302734375, -3.3603057861328125, 6.547996520996094, 98.65513610839844, 131.38095092773438, 24.998016357421875, 130.5497589111328, 106.3656997680664, 54.208526611328125, 92.31475830078125, 90.40282440185547, 120.1116943359375, -22.5360107421875, 195.67486572265625, 120.88972473144531, -108.51004028320312, 161.91363525390625, -26.2000732421875, 153.72637939453125, 19.75830078125, 180.10498046875, 17.074996948242188, 31.330596923828125, 6.08868408203125, 7.34698486328125, 42.69410705566406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000303.npy"}
{"epoch": 0.6345549738219896, "step": 304, "batch_size": 128, "mean": 41.68301010131836, "std": 75.1611099243164, "min": -148.45513916015625, "p10": -50.89035339355468, "median": 31.161170959472656, "p90": 138.44171142578125, "max": 178.9747314453125, "pos_frac": 0.7109375, "sample": [0.0, 29.78265380859375, -14.6734619140625, 111.72190856933594, -39.50715637207031, 0.908447265625, -133.9454803466797, 138.29705810546875, 126.8133544921875, -26.1806640625, 85.348876953125, 136.81353759765625, 31.674270629882812, 89.88917541503906, -71.83718872070312, 40.43658447265625, 18.71551513671875, 146.54287719726562, 43.24407958984375, 13.99481201171875, 10.84423828125, 18.623886108398438, -3.5464019775390625, -12.318695068359375, 111.55731201171875, 134.19235229492188, 119.08389282226562, 60.61546325683594, 8.22845458984375, 1.27587890625, -12.92803955078125, 12.544118881225586, -1.4341583251953125, -2.6669769287109375, 31.329055786132812, 22.17572021484375, 37.38121795654297, 20.953651428222656, 48.07562255859375, 114.12982177734375, 45.693206787109375, 101.51258850097656, 112.56404113769531, -54.71830749511719, 169.65194702148438, 20.084396362304688, -7.625282287597656, 115.397216796875, -78.90066528320312, 39.3179931640625, 36.512725830078125, -50.571044921875, 95.43400573730469, -148.45513916015625, 0.0, 141.5042266845703, -87.49348449707031, 98.38452911376953, 7.098663330078125, -37.22344970703125, -51.635406494140625, 23.364715576171875, -111.17710876464844, 96.9691162109375, 20.111328125, 86.47532653808594, -33.986846923828125, 48.6451416015625, 178.9747314453125, 123.89337158203125, -87.13296508789062, 21.131744384765625, 40.97499084472656, -109.50885009765625, -42.6077880859375, 157.43170166015625, 148.01959228515625, 2.5015640258789062, 149.172607421875, -108.71966552734375, 130.16323852539062, 112.07667541503906, -22.923240661621094, -5.248004913330078, 115.51376342773438, 18.925262451171875, 141.73721313476562, -6.1522216796875, 134.14175415039062, -129.7835693359375, -7.15814208984375, 0.3268604278564453, 22.951385498046875, 141.54913330078125, 148.77587890625, 10.211053848266602, -46.157379150390625, 122.10813903808594, 134.07171630859375, 125.93780517578125, 30.9932861328125, 32.37725830078125, 72.47496032714844, 9.282958984375, 94.65082550048828, 135.080078125, 52.88861083984375, 161.49826049804688, 112.00959777832031, 138.77923583984375, 103.91989135742188, 47.73602294921875, -58.1395263671875, 7.018707275390625, 98.95118713378906, 21.059661865234375, -45.6785888671875, -12.808486938476562, 120.89341735839844, -10.802871704101562, 40.400535583496094, 106.07544708251953, 1.107177734375, 122.31341552734375, -17.429946899414062, 132.56607055664062, 22.317962646484375, 157.60186767578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000304.npy"}
{"epoch": 0.6366492146596858, "step": 305, "batch_size": 128, "mean": 59.03711700439453, "std": 76.71515655517578, "min": -182.86749267578125, "p10": -18.1877056121826, "median": 48.996028900146484, "p90": 153.31039123535157, "max": 276.37255859375, "pos_frac": 0.8125, "sample": [7.613349914550781, 13.764019012451172, -13.52663803100586, 195.5084228515625, 37.8675537109375, 5.06884765625, 111.6170654296875, -52.443817138671875, 72.14862060546875, 19.455284118652344, 48.93743133544922, 39.40740966796875, -6.58428955078125, 117.98171997070312, 18.95220947265625, -63.63908386230469, -30.56036376953125, 25.552383422851562, 68.38485717773438, 0.9828262329101562, 66.47579956054688, 141.5137939453125, 4.273895263671875, 84.30540466308594, 6.173126220703125, 16.5621337890625, -112.5087890625, -4.8839263916015625, 31.475921630859375, 122.09037780761719, 22.69769287109375, 62.663818359375, 124.4764404296875, 103.60986328125, 145.96307373046875, 18.761459350585938, -0.2358856201171875, -3.256103515625, 130.43832397460938, -4.19390869140625, -111.1038818359375, 72.19725036621094, -5.750423431396484, 112.22640228271484, 144.08587646484375, 131.3980255126953, 49.05462646484375, 157.40005493164062, 137.99264526367188, 54.47523498535156, -5.199062347412109, -111.41828918457031, 163.21176147460938, 153.371337890625, 38.22528076171875, 73.39877319335938, 276.37255859375, 108.92608642578125, 178.1278076171875, 158.50921630859375, 24.081253051757812, 0.4966468811035156, -9.55963134765625, 147.84564208984375, 37.72967529296875, 112.29781341552734, 151.19424438476562, 113.18475341796875, 82.4404296875, 99.00961303710938, -29.06352996826172, 119.2538833618164, -65.5815200805664, 36.42433166503906, 115.8914794921875, -182.86749267578125, 31.764068603515625, 124.31361389160156, 1.218902587890625, 49.199432373046875, 7.442535400390625, 151.435546875, 156.59317016601562, 10.39727783203125, 114.2813720703125, 96.41842651367188, -30.318206787109375, 195.50167846679688, 1.5477828979492188, 12.664962768554688, 155.593505859375, 83.1319580078125, 136.372802734375, 37.51762390136719, 69.7589111328125, 126.86807250976562, 50.643463134765625, 135.94580078125, 174.8267822265625, 62.87635040283203, 32.379547119140625, -2.4991455078125, 123.82522583007812, 126.49966430664062, 190.93008422851562, 25.081100463867188, 33.17681884765625, 42.499267578125, 42.426025390625, 20.37646484375, 151.64306640625, 0.3162269592285156, 20.413284301757812, 6.10546875, 6.032440185546875, -85.7458724975586, 153.28427124023438, 4.849998474121094, 10.042312622070312, -52.8690185546875, 115.58047485351562, 51.4501953125, 176.4765625, 135.89112854003906, 106.88323974609375, -76.38410949707031, -4.686065673828125, 145.6072235107422], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000305.npy"}
{"epoch": 0.6387434554973822, "step": 306, "batch_size": 128, "mean": 53.597633361816406, "std": 76.00166320800781, "min": -143.60617065429688, "p10": -29.99454727172851, "median": 51.341278076171875, "p90": 147.1607406616211, "max": 184.08316040039062, "pos_frac": 0.7421875, "sample": [89.6934814453125, -111.01318359375, -98.45760345458984, 181.94467163085938, -143.60617065429688, 177.07470703125, 1.177988052368164, 48.70332336425781, -12.225616455078125, 135.6204833984375, 43.53760528564453, 23.686065673828125, 139.57911682128906, 103.90623474121094, 73.94880676269531, 6.186859130859375, -37.070709228515625, 124.00767517089844, -3.0451889038085938, 22.8809814453125, 48.084930419921875, 102.02527618408203, -39.184906005859375, 75.34478759765625, -139.27120971679688, 164.506103515625, 174.22503662109375, 125.55264282226562, 114.977783203125, -1.4432373046875, 60.5767822265625, 81.08782958984375, 107.42337036132812, 128.36294555664062, 147.27183532714844, 125.93709564208984, 169.01373291015625, 125.53826904296875, 115.11099243164062, 105.71371459960938, 6.794731140136719, 112.30391693115234, -5.930328369140625, 8.495071411132812, 97.66708374023438, 95.6083984375, 47.6873779296875, 114.5546875, 56.716766357421875, -88.3548583984375, -9.942718505859375, 142.2975616455078, -12.768798828125, 180.69451904296875, 43.1749267578125, -137.3350830078125, 93.63666534423828, 68.662353515625, -28.331375122070312, 175.560302734375, -3.11895751953125, 64.64703369140625, 103.19766235351562, 44.50933837890625, 120.8882827758789, 3.6183013916015625, 5.194976806640625, 3.0113086700439453, 106.69259643554688, -40.78080749511719, 110.6239013671875, -119.88943481445312, 112.12687683105469, -16.7432861328125, 177.1602783203125, -0.962646484375, -38.728363037109375, 52.020233154296875, 19.214859008789062, 29.585845947265625, 16.023391723632812, 58.027069091796875, 118.27435302734375, 122.47061157226562, 152.41583251953125, -17.521713256835938, -26.5950927734375, 60.998046875, 85.45904541015625, 184.08316040039062, 64.00564575195312, 27.3760986328125, 22.048843383789062, -23.62078857421875, 55.88946533203125, 114.98388671875, 114.32688903808594, -10.157058715820312, 148.57699584960938, -33.875282287597656, 142.87155151367188, 50.662322998046875, 132.13868713378906, 90.22140502929688, 25.625228881835938, 43.14916229248047, 4.727897644042969, -11.70599365234375, -5.568389892578125, -3.738494873046875, -7.584644317626953, 4.26910400390625, 140.26844787597656, 68.61325073242188, 16.11322021484375, 173.37469482421875, 29.788732528686523, -2.7052154541015625, -21.427978515625, 4.178497314453125, 46.337310791015625, 117.32185363769531, 0.372589111328125, -96.88238525390625, 106.47599792480469, 42.626129150390625, 139.82928466796875, 147.11312866210938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000306.npy"}
{"epoch": 0.6408376963350786, "step": 307, "batch_size": 128, "mean": 55.59379959106445, "std": 71.9971694946289, "min": -183.27725219726562, "p10": -15.286740493774412, "median": 52.74219512939453, "p90": 139.4512908935547, "max": 192.52239990234375, "pos_frac": 0.7734375, "sample": [140.68173217773438, 134.8133544921875, 16.489913940429688, 170.8760986328125, -127.06597900390625, -5.831993103027344, 83.98442840576172, 104.45956420898438, 100.48236083984375, 134.37063598632812, -162.29965209960938, -8.66779899597168, 92.23001098632812, -1.833465576171875, -9.0228271484375, 117.97805786132812, 135.40963745117188, 119.14273071289062, 9.983428955078125, 19.882049560546875, 56.8695068359375, -40.48785400390625, 75.45245361328125, 9.359878540039062, 172.40505981445312, -27.24566650390625, 77.797607421875, 10.853271484375, 126.357666015625, 160.9743194580078, 101.17742919921875, 24.31439208984375, 29.686676025390625, 93.04305267333984, 118.35809326171875, -12.216552734375, 136.44821166992188, 67.17153930664062, 20.1737060546875, -7.626686096191406, 24.10858154296875, 98.45829772949219, 124.09120178222656, -14.482177734375, 133.404541015625, 97.71844482421875, 139.4659881591797, -10.143463134765625, 1.0754547119140625, 24.804595947265625, 83.60881042480469, 6.736703872680664, 139.4449920654297, 8.397241592407227, 65.54676818847656, 123.84565734863281, 111.951171875, 147.49676513671875, 16.416297912597656, 3.948638916015625, 16.802001953125, 60.091392517089844, -68.37374877929688, 109.3543701171875, -0.37036895751953125, -3.165416717529297, 128.2346649169922, 65.1532211303711, -7.420963287353516, 5.583503723144531, -4.544380187988281, 133.71714782714844, -58.98663330078125, -14.61285400390625, 30.39117431640625, -22.704025268554688, 59.40887451171875, 130.5684356689453, 33.05657958984375, 25.6357421875, 142.43955993652344, -17.56353759765625, 71.37225341796875, 103.767822265625, 164.30535888671875, 106.07846069335938, -16.859142303466797, 120.91453552246094, 6.126237869262695, 47.99493408203125, 16.20258331298828, -38.396240234375, 48.61488342285156, 166.3062744140625, 102.69151306152344, 20.2945556640625, 134.1772003173828, -125.74311828613281, -14.363922119140625, 124.1053466796875, 2.867523193359375, 175.47906494140625, 32.930450439453125, 154.554931640625, -183.27725219726562, 7.254669189453125, 2.1005859375, 87.28036499023438, 39.7135009765625, 93.4444580078125, -18.0699462890625, 25.073486328125, 192.52239990234375, 122.63382720947266, 33.30317687988281, 103.40084838867188, -3.1940155029296875, -8.7550048828125, 47.84840393066406, 130.93997192382812, 94.17929077148438, 11.140533447265625, 118.96145629882812, 135.91845703125, 96.03781127929688, 3.1965713500976562, 142.78558349609375, 110.62800598144531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000307.npy"}
{"epoch": 0.6429319371727749, "step": 308, "batch_size": 128, "mean": 39.008174896240234, "std": 75.89021301269531, "min": -176.4332275390625, "p10": -44.44987297058105, "median": 18.786338806152344, "p90": 138.56063537597655, "max": 198.21890258789062, "pos_frac": 0.7265625, "sample": [145.83387756347656, 32.99695587158203, -34.3785400390625, 148.40011596679688, 61.06536865234375, -91.94868469238281, -47.24639892578125, 52.67626953125, 9.625579833984375, 86.23811340332031, 4.6080322265625, 156.6414794921875, -0.15163421630859375, -76.0089111328125, -5.689849853515625, 9.858245849609375, 117.96513366699219, 0.502105712890625, 119.79132080078125, 124.7958984375, 127.76482391357422, 29.688323974609375, 1.120941162109375, -86.3554458618164, 11.504486083984375, 101.10419464111328, -151.4041748046875, -11.903114318847656, 16.982772827148438, 144.01608276367188, 117.14839172363281, 42.07421875, 53.95880126953125, 106.60630798339844, -19.11090087890625, -2.87188720703125, 127.33702087402344, 5.170036315917969, 106.12739562988281, 8.700057983398438, -1.4462890625, 41.090919494628906, 183.21258544921875, 1.30987548828125, -35.87681579589844, -10.352142333984375, 1.1094970703125, 136.2935791015625, -4.863332748413086, 118.75810241699219, 5.352851867675781, 37.05510711669922, 2.88641357421875, 114.36011505126953, -8.194282531738281, 28.264434814453125, 126.1109619140625, 67.23051452636719, -176.4332275390625, 156.8521728515625, 83.27525329589844, 17.720687866210938, -6.88751220703125, -16.658905029296875, 198.21890258789062, 118.94352722167969, -2.4932498931884766, -122.04730224609375, 12.715164184570312, -3.5306854248046875, 67.036865234375, -94.94271087646484, 144.43930053710938, 74.38777923583984, 11.49884033203125, 20.7327880859375, 159.55087280273438, 182.49542236328125, 143.85043334960938, -0.9197196960449219, 23.025924682617188, 7.86480712890625, 2.2718772888183594, 4.72161865234375, 70.22564697265625, -125.93801879882812, 47.22740936279297, 7.601959228515625, -56.29669189453125, 111.72267150878906, 19.85198974609375, 13.157569885253906, 34.628173828125, 9.690689086914062, 119.96276092529297, 116.21907806396484, 156.06790161132812, 131.144287109375, 31.349380493164062, 128.18902587890625, 24.768081665039062, 74.489013671875, 176.1368408203125, -94.46891021728516, 36.488365173339844, 10.002716064453125, 16.6002197265625, -43.25136184692383, 120.3360595703125, 118.06015014648438, -80.18731689453125, 13.751556396484375, -25.573272705078125, 2.5050735473632812, 3.6364669799804688, 99.36483764648438, -29.3565673828125, 12.585601806640625, -5.370086669921875, 7.051239013671875, -42.912506103515625, 97.64912414550781, 106.55532836914062, 129.23904418945312, -89.333740234375, -12.0875244140625, 128.31192016601562, 72.0283203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000308.npy"}
{"epoch": 0.6450261780104712, "step": 309, "batch_size": 128, "mean": 56.044254302978516, "std": 82.29788208007812, "min": -131.90139770507812, "p10": -41.6039180755615, "median": 46.02863693237305, "p90": 159.52929077148437, "max": 283.0987548828125, "pos_frac": 0.734375, "sample": [38.86940002441406, 130.0660400390625, 136.01239013671875, 121.88296508789062, -0.99310302734375, -32.525543212890625, -18.92218017578125, -104.0929183959961, 136.46804809570312, 44.66778564453125, 161.46783447265625, 11.957695007324219, -10.48773193359375, 6.2482757568359375, -99.24765014648438, 111.03107452392578, 14.21484375, 20.34272003173828, -9.38555908203125, 39.613800048828125, -10.222183227539062, 114.50714874267578, -1.37115478515625, -67.03160095214844, 10.20574951171875, 28.780364990234375, 3.4189453125, 283.0987548828125, 116.82365417480469, 69.2406005859375, -8.708663940429688, 167.65850830078125, 94.66159057617188, 2.7080841064453125, -2.326751708984375, -10.723968505859375, 84.45679473876953, -78.72735595703125, 199.25900268554688, -75.680908203125, 183.58230590820312, -75.01150512695312, -4.398349761962891, 68.03372192382812, 125.27716064453125, -55.197593688964844, 1.802001953125, -32.902862548828125, -94.46829223632812, 14.299667358398438, 132.45497131347656, -10.986856460571289, 16.189987182617188, -100.8856201171875, 143.66046142578125, 122.79733276367188, 131.4622802734375, 17.46844482421875, 106.53326416015625, 16.8284912109375, 158.698486328125, -5.11737060546875, 60.759803771972656, 225.9052734375, -12.639572143554688, 168.28631591796875, 9.873062133789062, 113.42794036865234, 26.38640594482422, 47.663177490234375, 33.68824768066406, 35.19294738769531, 0.6144485473632812, 175.845947265625, 2.637298583984375, 153.6595458984375, 139.93084716796875, 98.04646301269531, 11.085113525390625, 163.2696533203125, 129.75143432617188, 132.01312255859375, 6.258506774902344, 90.93174743652344, 123.54546356201172, 143.58331298828125, -3.9587173461914062, 98.1105728149414, 37.24713134765625, -35.77805709838867, 133.23419189453125, -131.90139770507812, 11.894744873046875, -16.707914352416992, 163.50714111328125, 123.65859985351562, 18.613922119140625, 132.2045135498047, 57.27703857421875, 9.439765930175781, -110.44589233398438, 60.85856628417969, 132.92709350585938, 156.60809326171875, 143.13339233398438, -2.04718017578125, 188.07052612304688, 31.272216796875, 67.57379150390625, -109.24501037597656, 154.05157470703125, 52.21527099609375, 67.10205078125, 145.2889404296875, 56.95489501953125, 138.1246337890625, 53.4368896484375, 166.07672119140625, -12.821624755859375, 145.4635467529297, 163.90325927734375, 47.389488220214844, 113.41944885253906, -72.3995361328125, 108.18560791015625, 40.499298095703125, -2.334218978881836, 96.51004028320312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000309.npy"}
{"epoch": 0.6471204188481675, "step": 310, "batch_size": 128, "mean": 45.1485710144043, "std": 71.18560791015625, "min": -131.15960693359375, "p10": -28.49957237243652, "median": 24.027175903320312, "p90": 138.1463623046875, "max": 217.2867431640625, "pos_frac": 0.7578125, "sample": [217.2867431640625, 16.122467041015625, 5.092540740966797, 183.118896484375, 33.49761962890625, 34.458984375, 45.1561279296875, 11.01980972290039, 123.29222106933594, 74.28610229492188, 113.71693420410156, 169.54833984375, 78.80693054199219, 119.2127685546875, 23.96722412109375, 35.90997314453125, 6.559661865234375, -5.088878631591797, 39.88189697265625, -63.87678527832031, 4.1171417236328125, 2.49078369140625, 19.881362915039062, 57.43634033203125, 82.1858139038086, -94.58201599121094, 126.36235046386719, -40.462890625, 120.75794982910156, -0.193603515625, -5.1677703857421875, 1.1458663940429688, -10.88824462890625, 128.5093994140625, -0.470428466796875, -3.019256591796875, 55.59088134765625, 163.40869140625, -47.571990966796875, -100.93035888671875, 115.94833374023438, -1.1119537353515625, 7.969940185546875, 22.390228271484375, 106.84880065917969, 37.2242431640625, 95.96328735351562, 30.03820037841797, 33.8360595703125, 8.422393798828125, -9.100433349609375, 24.087127685546875, 149.8372802734375, 105.10293579101562, 100.57131958007812, 134.64727783203125, 176.30145263671875, 16.795562744140625, -4.2916259765625, -131.15960693359375, -26.15369415283203, 159.30398559570312, 27.979347229003906, 22.9095458984375, -3.988037109375, -114.92130279541016, 8.218307495117188, 18.791168212890625, 104.25372314453125, 4.434549331665039, -19.026893615722656, 56.44584655761719, 18.09698486328125, 30.366729736328125, 29.707733154296875, 118.30914306640625, 129.24462890625, 163.85809326171875, 58.7598876953125, -5.5216827392578125, -28.187305450439453, 103.48237609863281, 110.62307739257812, 0.3804168701171875, -43.724517822265625, 97.91761779785156, 135.833740234375, 3.74298095703125, 115.37040710449219, -8.842529296875, 22.806053161621094, -51.573974609375, 132.48495483398438, -85.04888916015625, 38.8486213684082, 2.6921310424804688, 9.6417236328125, 138.01168823242188, 2.87646484375, 21.5426025390625, 50.10436248779297, 152.91465759277344, 3.0469970703125, -29.228195190429688, 138.46060180664062, -12.374176025390625, 0.995513916015625, -45.42437744140625, -103.08474731445312, 105.15458679199219, -14.70526123046875, 136.19366455078125, 20.736282348632812, 118.55087280273438, 6.84271240234375, -13.2120361328125, 17.78533935546875, 5.536495208740234, 10.8966064453125, 102.56855773925781, 41.70477294921875, 5.9781951904296875, 133.61776733398438, 150.51690673828125, 198.27804565429688, 100.63128662109375, 143.0302734375, 112.66523742675781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000310.npy"}
{"epoch": 0.6492146596858639, "step": 311, "batch_size": 128, "mean": 49.199867248535156, "std": 84.72686004638672, "min": -178.16851806640625, "p10": -62.19524040222165, "median": 35.30696105957031, "p90": 151.69811706542967, "max": 209.07662963867188, "pos_frac": 0.7265625, "sample": [5.759803771972656, -6.675384521484375, 123.04898071289062, 120.6587142944336, 144.77398681640625, 18.805788040161133, 29.242202758789062, -1.5717926025390625, 144.52285766601562, 116.2408447265625, 137.33811950683594, 6.223178863525391, -51.09181594848633, -23.56951904296875, 35.46784973144531, -1.8770751953125, 142.58343505859375, 62.24894714355469, -82.47298431396484, 61.92071533203125, 173.87405395507812, 137.77532958984375, 2.9248199462890625, 118.77191162109375, -0.8933639526367188, -16.4058837890625, 158.10147094726562, 117.8212890625, 6.035503387451172, -0.806182861328125, -94.3360824584961, -169.312255859375, -178.16851806640625, 85.63427734375, 35.29034423828125, 157.06512451171875, -126.35220336914062, -90.52783203125, 131.83535766601562, 51.348480224609375, -80.72999572753906, 104.15235900878906, 24.677642822265625, 2.7564697265625, -2.4387664794921875, 51.88018798828125, -14.752754211425781, 124.92472839355469, -107.66416931152344, 30.360511779785156, 113.60091400146484, 24.609619140625, 3.4423828125, 26.145477294921875, 121.27558898925781, -25.080825805664062, 74.34832763671875, -27.03021240234375, -21.9473876953125, 158.33929443359375, 157.83392333984375, 136.53237915039062, 144.21493530273438, 122.49141693115234, 125.183349609375, -93.29072570800781, 35.323577880859375, 12.02374267578125, 19.220169067382812, 33.07105255126953, -96.30232238769531, -12.35555648803711, 173.18057250976562, 15.70452880859375, 21.25616455078125, 71.3480453491211, 159.3865966796875, 113.72821044921875, 40.183349609375, 151.02151489257812, 209.07662963867188, 2.09716796875, -103.9039306640625, 77.21149444580078, 0.4444427490234375, -34.4266357421875, 164.8072509765625, 153.27685546875, 136.90155029296875, 24.206298828125, 118.96952819824219, 100.18672180175781, 109.03523254394531, 61.2635498046875, 171.7210693359375, -130.7089385986328, 164.87237548828125, -5.154388427734375, -122.95669555664062, 117.44438171386719, 7.741626739501953, 148.90499877929688, -14.719772338867188, 24.89306640625, 10.635162353515625, -5.392726898193359, 147.46632385253906, 144.90597534179688, 9.473810195922852, 41.6451416015625, 131.6233673095703, 108.22503662109375, 106.96585083007812, 27.222503662109375, -36.18736267089844, 5.847312927246094, 108.75775146484375, 138.91661071777344, 141.4157257080078, 1.9738502502441406, -54.251773834228516, -3.672271728515625, 5.8456878662109375, 187.6090087890625, 109.36700439453125, 131.43307495117188, 81.639892578125, -16.942489624023438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000311.npy"}
{"epoch": 0.6513089005235602, "step": 312, "batch_size": 128, "mean": 48.54539489746094, "std": 76.9007797241211, "min": -165.1595458984375, "p10": -38.584120941162105, "median": 37.81989288330078, "p90": 142.54758300781248, "max": 195.23086547851562, "pos_frac": 0.7421875, "sample": [30.10791015625, -3.47900390625, -18.660186767578125, 129.86947631835938, 133.89845275878906, -165.1595458984375, 135.41433715820312, 120.8697509765625, -37.19617462158203, -85.71490478515625, -22.339996337890625, 10.680915832519531, 37.9241943359375, 112.37472534179688, 72.7420654296875, 156.33999633789062, 97.88519287109375, 130.74557495117188, -16.800674438476562, -107.78323364257812, 6.96929931640625, 120.06092834472656, 127.235595703125, -2.350006103515625, 7.914543151855469, 50.70685577392578, 56.75482177734375, 195.23086547851562, -41.822662353515625, -13.502044677734375, 151.5762939453125, -5.230140686035156, 4.936279296875, 156.21429443359375, 75.83929443359375, 34.3345947265625, 119.97247314453125, 0.3612518310546875, 96.12600708007812, 113.524658203125, 169.009521484375, 149.2289276123047, 59.624473571777344, 98.0506591796875, 112.38235473632812, 115.9129638671875, 57.230438232421875, 109.34426879882812, 125.80508422851562, 32.086273193359375, 133.23599243164062, 136.9893341064453, 17.902069091796875, 2.97369384765625, -33.6956787109375, 9.75933837890625, 39.66120910644531, 18.568252563476562, 0.3567352294921875, 174.09405517578125, 8.496009826660156, -28.66259765625, 109.16586303710938, -112.66612243652344, 51.554840087890625, -34.051910400390625, 166.5643310546875, 130.60586547851562, 3.397796630859375, 111.67341613769531, 27.487579345703125, 128.34197998046875, 25.39794158935547, -5.67265510559082, 68.73019409179688, 82.51031494140625, 2.4754714965820312, 146.13760375976562, -2.4773712158203125, 110.76840209960938, 141.00900268554688, 147.43438720703125, -57.098358154296875, 122.72267150878906, -1.355447769165039, 0.0, 55.61090087890625, -84.55252838134766, -111.1700439453125, -89.240234375, 14.506103515625, 74.41818237304688, 10.753860473632812, 37.71559143066406, 92.2943115234375, 1.6347885131835938, -110.88008117675781, 17.875568389892578, 106.948974609375, 1.1579437255859375, 40.509857177734375, 8.304214477539062, 128.29763793945312, 68.16488647460938, -13.999176025390625, 109.81991577148438, 122.93450927734375, -2.28216552734375, 172.58627319335938, 34.54937744140625, 136.21319580078125, 19.076690673828125, 37.334686279296875, -5.267023086547852, -79.0714111328125, 79.55158996582031, 113.19219970703125, 3.695709228515625, -25.516693115234375, 170.52752685546875, -111.68217468261719, 154.32977294921875, -28.81646728515625, 132.6959686279297, 116.59229278564453, 23.613677978515625, 21.944812774658203, -68.14311218261719], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000312.npy"}
{"epoch": 0.6534031413612565, "step": 313, "batch_size": 128, "mean": 47.11670684814453, "std": 81.33078002929688, "min": -151.935546875, "p10": -54.9728042602539, "median": 31.792984008789062, "p90": 154.16501007080078, "max": 221.81982421875, "pos_frac": 0.71875, "sample": [153.85415649414062, 140.29000854492188, -21.8333740234375, -27.300811767578125, -24.08978271484375, -2.1207504272460938, 66.40567016601562, 132.4500274658203, 158.26007080078125, -15.504730224609375, -8.672439575195312, 136.17843627929688, -89.9232177734375, 141.38279724121094, 98.1017837524414, 154.8903350830078, 140.07745361328125, 40.433746337890625, -17.630584716796875, -27.38214111328125, 4.2931365966796875, 14.80694580078125, 3.1469383239746094, 141.1497039794922, 117.8238525390625, -8.484344482421875, -5.110595703125, 23.102066040039062, 110.7325439453125, 177.67816162109375, 166.2685546875, 203.5341796875, 23.845672607421875, 85.7010498046875, -40.45550537109375, 62.8973388671875, 101.41981506347656, 28.697784423828125, 176.3009033203125, 40.50800323486328, -63.721649169921875, 179.35260009765625, 141.1265106201172, 28.030624389648438, -74.276611328125, 68.46546936035156, 19.825469970703125, -57.28814697265625, 113.92692565917969, 136.22869873046875, 36.15086364746094, 92.31005859375, -107.16189575195312, 18.292232513427734, 88.44436645507812, 0.5220298767089844, 152.32321166992188, 56.83674621582031, 25.00738525390625, 2.581298828125, 121.2679672241211, 5.3534088134765625, 123.4148941040039, 20.554244995117188, 9.572380065917969, 41.962371826171875, -53.98051452636719, 111.07675170898438, -108.83207702636719, -43.3663330078125, 161.31968688964844, -15.70281982421875, 22.911216735839844, 156.0933837890625, 19.59454345703125, -5.19219970703125, 33.26329040527344, 137.58270263671875, -109.60653686523438, 105.61480712890625, 12.843338012695312, -107.69012451171875, 100.93606567382812, -45.43896484375, 37.595703125, 128.54293823242188, 90.06805419921875, -10.362213134765625, 7.992799758911133, 5.609565734863281, 26.82373046875, 221.81982421875, 114.09463500976562, 54.5076904296875, 127.8082275390625, 149.6490020751953, 25.97753143310547, 163.4267578125, -40.567352294921875, -148.28024291992188, 62.25799560546875, 43.94348907470703, 7.589788436889648, 158.59710693359375, 52.267059326171875, -83.38957214355469, -0.34504127502441406, -2.5392990112304688, 142.021728515625, 112.26296997070312, 25.926483154296875, 88.60221862792969, 50.257171630859375, 3.5697021484375, -1.4536361694335938, -9.689414978027344, 118.59933471679688, 30.322677612304688, -104.04257202148438, -104.31184387207031, -1.1269607543945312, -151.935546875, 94.00584411621094, 87.11040496826172, 16.4990234375, 151.485595703125, 23.960174560546875, 159.53721618652344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000313.npy"}
{"epoch": 0.6554973821989529, "step": 314, "batch_size": 128, "mean": 48.260616302490234, "std": 81.43144226074219, "min": -169.311767578125, "p10": -57.1707794189453, "median": 35.5875358581543, "p90": 143.06304626464842, "max": 205.78250122070312, "pos_frac": 0.703125, "sample": [73.03107452392578, 108.40695190429688, 117.10841369628906, 44.22479248046875, -11.558761596679688, 19.810943603515625, -70.07069396972656, 105.96047973632812, -80.72000885009766, -125.04031372070312, -11.10369873046875, 68.49504089355469, 141.39553833007812, 157.9249725341797, 156.43890380859375, -62.968017578125, -82.25222778320312, 177.91180419921875, 131.6724853515625, 147.51553344726562, 91.49826049804688, -9.436737060546875, 37.31377410888672, 114.70143127441406, 2.775115966796875, 14.942207336425781, 11.597465515136719, 168.489013671875, 7.108551025390625, 171.20303344726562, 5.625446319580078, -15.978515625, 9.285919189453125, -112.64950561523438, 111.32023620605469, 51.851104736328125, 17.87286376953125, 143.84432983398438, 37.97607421875, 141.898193359375, -15.5115966796875, 118.13544464111328, -75.41995239257812, 114.96781921386719, 5.2012176513671875, 29.978530883789062, 179.63873291015625, 117.94219970703125, 27.0291748046875, -0.70794677734375, 129.5894775390625, -0.24323272705078125, 16.340408325195312, 80.995361328125, -0.4934196472167969, 106.28121948242188, 31.851593017578125, 19.34979248046875, -13.932693481445312, -97.06314086914062, 137.07305908203125, 140.94796752929688, 120.28996276855469, 0.0, 0.4638671875, -66.65394592285156, -161.2408447265625, 123.93182373046875, 22.31939697265625, -19.884796142578125, -54.686248779296875, 18.538665771484375, 109.68048095703125, 139.89129638671875, 157.9477996826172, -42.54845428466797, -28.076637268066406, 160.25210571289062, -32.09869384765625, -169.311767578125, 134.52163696289062, 49.703948974609375, 133.2960205078125, 85.97567749023438, 0.0, 138.05975341796875, -3.375518798828125, 60.123626708984375, 132.655029296875, 26.29949951171875, 20.018295288085938, 124.76080322265625, 132.0848846435547, 33.861297607421875, 103.99807739257812, 142.72821044921875, 112.86962127685547, 24.036605834960938, 42.95343017578125, 7.823699951171875, 8.765972137451172, -26.18524932861328, -4.47991943359375, -39.5621337890625, 114.78701782226562, -0.97003173828125, 90.41266632080078, 122.73475646972656, 44.22479248046875, 104.8143539428711, 167.32534790039062, 11.59283447265625, -42.294677734375, -3.6335296630859375, 3.518463134765625, 205.78250122070312, 135.57485961914062, -12.029182434082031, 19.56806182861328, 91.18344116210938, 130.75169372558594, 99.86206817626953, -73.82763671875, 86.22259521484375, -161.53622436523438, 167.040283203125, 101.83885192871094, -4.7033538818359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000314.npy"}
{"epoch": 0.6575916230366492, "step": 315, "batch_size": 128, "mean": 55.791709899902344, "std": 78.36016845703125, "min": -146.35643005371094, "p10": -26.124822998046863, "median": 29.34756088256836, "p90": 157.75324707031248, "max": 244.25177001953125, "pos_frac": 0.796875, "sample": [1.199737548828125, 15.1019287109375, 2.920564651489258, 5.4803619384765625, 13.243507385253906, -34.7349853515625, -19.817138671875, 13.040390014648438, 165.99761962890625, -11.650238037109375, 35.33345031738281, -14.401092529296875, 122.00127410888672, 20.54541015625, 124.14480590820312, -35.833465576171875, 50.38117980957031, 156.0574951171875, 50.315185546875, -22.43475341796875, -71.63356018066406, -20.445953369140625, 7.7059783935546875, -2.9058761596679688, 175.80511474609375, 11.33380126953125, 116.93563842773438, 167.75660705566406, 192.27093505859375, 0.8592681884765625, 24.285572052001953, 39.203704833984375, 150.49151611328125, 153.09375, 29.137481689453125, 175.63958740234375, 155.73388671875, 142.4954833984375, 103.566650390625, -3.058837890625, 0.798919677734375, -36.293212890625, 18.99359130859375, 147.16094970703125, -43.41987609863281, 41.83290100097656, 8.118682861328125, 6.047210693359375, 13.5806884765625, 9.5439453125, 138.63677978515625, -7.051769256591797, 140.49636840820312, 159.9169921875, 139.720458984375, 141.07933044433594, 15.146942138671875, 29.546104431152344, 39.3785400390625, 71.55608367919922, 12.95526123046875, 4.480560302734375, -8.63949966430664, 106.57284545898438, 131.87167358398438, 143.36273193359375, 99.62139892578125, 113.61400604248047, 27.011093139648438, 244.25177001953125, -84.7911376953125, 32.30372619628906, 120.89471435546875, 28.009063720703125, 142.59585571289062, -41.270416259765625, -143.07276916503906, 169.88897705078125, 2.9619827270507812, 163.28140258789062, 138.51327514648438, 11.37591552734375, 1.2427825927734375, 134.36871337890625, 184.26007080078125, 14.231864929199219, 39.205291748046875, 111.0704345703125, 21.656280517578125, 174.53741455078125, 16.297332763671875, 76.26937866210938, -140.9722900390625, -41.21319580078125, 148.54769897460938, 165.68820190429688, 71.847900390625, 76.80216979980469, -146.35643005371094, 150.5667724609375, 16.89575958251953, -8.112213134765625, -3.72222900390625, 100.9201431274414, 9.567169189453125, -69.6085205078125, -58.40521240234375, 47.3077392578125, 20.63443374633789, 18.672119140625, 152.53692626953125, 13.192863464355469, 125.58154296875, 195.5780487060547, 15.768798828125, 107.58407592773438, 156.825927734375, 8.524063110351562, -8.0540771484375, 14.542755126953125, 113.79540252685547, -2.5615234375, 134.0758819580078, 32.08934020996094, 132.053466796875, 85.4244384765625, 67.28863525390625, 29.149017333984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000315.npy"}
{"epoch": 0.6596858638743456, "step": 316, "batch_size": 128, "mean": 57.464054107666016, "std": 79.87915802001953, "min": -144.00184631347656, "p10": -29.595469665527343, "median": 39.39453125, "p90": 163.54231872558591, "max": 250.01107788085938, "pos_frac": 0.734375, "sample": [123.73370361328125, 83.35391235351562, 63.6048583984375, 36.609344482421875, -18.2789306640625, -35.61659240722656, -9.4219970703125, 166.32870483398438, 83.67510986328125, 168.42633056640625, 25.924758911132812, -36.080810546875, 141.6465606689453, 43.497894287109375, -62.27787780761719, 12.970212936401367, 122.08525085449219, -29.713165283203125, 135.12255859375, 11.327728271484375, 132.13162231445312, 152.8690185546875, -24.0499267578125, 120.92192077636719, 132.21839904785156, 231.6339111328125, 40.1319580078125, 129.8031005859375, 41.9166259765625, 180.9911346435547, -10.084388732910156, 139.79872131347656, -24.388458251953125, 120.71015930175781, 167.2105712890625, 12.444061279296875, 99.25833892822266, 147.276123046875, 161.16050720214844, 31.37860107421875, -1.863372802734375, 93.8631591796875, 5.64752197265625, 55.84027099609375, 10.649269104003906, 84.1070327758789, -29.545028686523438, 153.5318603515625, 28.5009765625, 0.25372314453125, 84.9234619140625, 4.855987548828125, -144.00184631347656, -85.4200439453125, -27.563308715820312, 189.29925537109375, 112.23625183105469, 91.02435302734375, -4.4002685546875, 0.2046356201171875, -25.530841827392578, 53.71173095703125, 176.2821044921875, 139.50750732421875, 20.781402587890625, 24.908935546875, -65.10498046875, 250.01107788085938, 37.54389953613281, 113.30206298828125, -2.6143341064453125, -0.8994903564453125, -32.234130859375, -11.623992919921875, 2.598114013671875, 138.49465942382812, 85.911376953125, 150.21243286132812, 5.080085754394531, 137.71926879882812, 124.5955581665039, 129.55206298828125, 162.388671875, -5.929466247558594, -6.064260482788086, 19.24822998046875, 34.659515380859375, 182.95208740234375, -79.61844635009766, 150.78424072265625, -85.79351806640625, 66.6361083984375, 0.0, -40.491363525390625, 15.49139404296875, 168.09927368164062, 166.23416137695312, 14.641769409179688, -0.2324676513671875, 178.98220825195312, 98.71047973632812, 13.40435791015625, -125.87869262695312, 10.092208862304688, -2.58746337890625, 52.149017333984375, 148.90811157226562, 7.034881591796875, 100.10698699951172, 117.90432739257812, 173.9254150390625, 22.5313720703125, 154.07513427734375, -25.92645263671875, 21.161163330078125, 91.207275390625, -17.837215423583984, -126.42729187011719, 38.6571044921875, 7.596893310546875, 91.02133178710938, 116.57632446289062, 114.66149139404297, 38.417266845703125, 69.82818603515625, 17.464584350585938, 133.02749633789062, -10.9912109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000316.npy"}
{"epoch": 0.6617801047120419, "step": 317, "batch_size": 128, "mean": 50.863075256347656, "std": 76.81656646728516, "min": -147.39990234375, "p10": -27.042291259765626, "median": 33.19486999511719, "p90": 149.3403778076172, "max": 239.70733642578125, "pos_frac": 0.7109375, "sample": [131.2971649169922, -120.919189453125, 0.0, 6.557441711425781, 189.1727294921875, -27.429283142089844, 159.82760620117188, 67.73550415039062, 10.540325164794922, 167.8155517578125, 15.254852294921875, 148.88174438476562, 39.73455810546875, -12.712921142578125, 145.99221801757812, -60.628387451171875, -27.741867065429688, 140.48428344726562, 70.306884765625, 7.478706359863281, 55.083740234375, 10.4835205078125, 138.62530517578125, 116.09111022949219, -1.00567626953125, 134.5650634765625, 13.679115295410156, 1.0032672882080078, -9.165740966796875, 25.2635498046875, -5.045196533203125, -51.405731201171875, 33.798004150390625, -6.74591064453125, -23.486541748046875, 101.89955139160156, 118.68791961669922, 167.61729431152344, 109.94749450683594, -26.9605712890625, -66.09512329101562, 34.34358215332031, 160.79998779296875, -138.92462158203125, -127.43118286132812, 147.16473388671875, 169.99517822265625, 147.92840576171875, 16.66082000732422, 0.0, 73.05023193359375, 78.22232055664062, 38.0301513671875, 179.869873046875, -118.65750122070312, -10.875785827636719, 26.28778076171875, 40.33976745605469, 2.5586395263671875, -9.14971923828125, 17.23236083984375, 152.26263427734375, 141.86666870117188, 110.98734283447266, 0.0, 11.42431640625, 15.650909423828125, -1.187286376953125, 30.731197357177734, 100.08087158203125, 147.62985229492188, -6.278007507324219, -10.556304931640625, -147.39990234375, 165.35696411132812, 80.05616760253906, 150.4105224609375, -83.779296875, 29.69230842590332, -14.752548217773438, 75.43986511230469, 121.52083587646484, 112.77229309082031, -11.36212158203125, 32.59173583984375, 96.75387573242188, 46.39744567871094, -27.23297119140625, 109.67779541015625, 145.96450805664062, -2.412872314453125, 18.534515380859375, 157.6539764404297, 12.151420593261719, 6.1896514892578125, -0.8913173675537109, -2.0431365966796875, 131.35415649414062, 114.04360961914062, 239.70733642578125, 5.361236572265625, 113.14863586425781, 131.21356201171875, -9.1607666015625, -32.71968078613281, 34.80230712890625, 21.897979736328125, 36.436370849609375, 0.39968299865722656, 66.09536743164062, 106.00860595703125, 5.538179397583008, 133.51580810546875, 6.489795684814453, 98.92153930664062, 102.58200073242188, 118.08455657958984, 32.58792495727539, 157.7100067138672, 14.823013305664062, -7.829181671142578, 73.94296264648438, -12.87091064453125, 148.2952117919922, 52.98632049560547, 121.77430725097656, -4.190773010253906, 129.69537353515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000317.npy"}
{"epoch": 0.6638743455497382, "step": 318, "batch_size": 128, "mean": 54.223602294921875, "std": 74.49516296386719, "min": -158.27340698242188, "p10": -29.90585632324218, "median": 56.28582763671875, "p90": 146.12671508789063, "max": 188.50457763671875, "pos_frac": 0.7421875, "sample": [84.63169860839844, 22.033798217773438, 137.4739990234375, 131.67068481445312, 23.47564697265625, -8.13409423828125, 32.110015869140625, 25.997833251953125, 143.86495971679688, 42.900726318359375, 8.863525390625, -19.52996063232422, 145.24667358398438, 33.62052917480469, 167.90478515625, 9.65606689453125, 128.09371948242188, -15.08282470703125, 112.45140075683594, 87.60360717773438, -21.575286865234375, -17.244171142578125, 128.54531860351562, 148.3278350830078, -17.404388427734375, 40.248809814453125, 121.03465270996094, 139.87522888183594, -115.1841049194336, 30.370849609375, 1.68634033203125, -44.36894989013672, 72.7982177734375, 80.17820739746094, 54.04911804199219, -27.520294189453125, 9.146455764770508, 60.34112548828125, 137.98873901367188, 172.92425537109375, 91.16378784179688, 130.30020141601562, 105.17276000976562, 59.09942626953125, 10.787002563476562, 7.484344482421875, -136.90855407714844, 111.59587860107422, -3.60418701171875, 109.18795776367188, -147.43670654296875, 13.93264389038086, 92.03317260742188, 161.86740112304688, 119.59873962402344, 45.171905517578125, -16.745887756347656, 118.72775268554688, 143.03518676757812, -17.83660888671875, 97.55126953125, 116.11553955078125, 169.87527465820312, 56.76484680175781, -96.942138671875, 63.63409423828125, -42.4190673828125, 5.084259033203125, 153.4083251953125, -3.5065155029296875, 128.32672119140625, 13.734542846679688, 5.47796630859375, 88.78008270263672, 117.40310668945312, 131.44229125976562, 62.080657958984375, 126.40383911132812, 85.88723754882812, -10.105743408203125, 119.92984771728516, 61.622711181640625, 33.81121826171875, 134.80722045898438, 77.29212951660156, -39.50299072265625, 89.9005126953125, 162.77984619140625, -40.97577667236328, -7.1089324951171875, -7.381744384765625, 62.7784423828125, 24.669921875, 120.55416870117188, 51.469970703125, 49.668182373046875, 92.57781982421875, -58.63075256347656, 80.13211059570312, -16.542022705078125, 116.34673309326172, 188.50457763671875, 161.89166259765625, 42.207855224609375, -35.47216796875, 121.63702392578125, 70.10296630859375, 10.886367797851562, -3.1094818115234375, -35.4931640625, 151.498046875, 34.61981201171875, 55.80680847167969, 152.6376953125, 26.844680786132812, 148.18014526367188, 50.25398635864258, 142.1986083984375, -25.373321533203125, 107.07705688476562, -128.57855224609375, 68.06439208984375, -1.9310684204101562, 29.493759155273438, 152.61415100097656, -8.559867858886719, -158.27340698242188, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000318.npy"}
{"epoch": 0.6659685863874345, "step": 319, "batch_size": 128, "mean": 57.31303024291992, "std": 69.11637115478516, "min": -157.0742950439453, "p10": -15.597204589843743, "median": 41.512657165527344, "p90": 146.47157287597656, "max": 195.30535888671875, "pos_frac": 0.8203125, "sample": [108.97816467285156, 2.995302200317383, 26.49444580078125, 54.57075500488281, 30.557571411132812, 195.30535888671875, 119.77691650390625, -38.32354736328125, 42.74263000488281, 24.728118896484375, -11.554153442382812, 12.617549896240234, 106.64624786376953, 161.56765747070312, 72.25958251953125, 155.77096557617188, 8.67071533203125, 0.0, 58.834197998046875, 55.826995849609375, 10.062007904052734, 20.371963500976562, 82.54464721679688, 11.561019897460938, 7.86614990234375, 126.28805541992188, 7.37274169921875, 83.68846130371094, -6.667236328125, 104.56463623046875, -13.522125244140625, 9.339675903320312, 158.0936279296875, 5.126468658447266, 117.98291778564453, 44.775787353515625, 6.149078369140625, 81.15823364257812, 138.38897705078125, 65.05807495117188, 3.888032913208008, 138.18197631835938, 43.92095947265625, -108.52944946289062, -2.3470382690429688, 113.55667114257812, 156.1627197265625, 8.925994873046875, -48.988433837890625, 142.295166015625, 129.186279296875, 145.92587280273438, 11.905563354492188, 30.9046630859375, 140.384521484375, 26.89190673828125, 28.677276611328125, -76.23104095458984, 63.60852813720703, 133.50376892089844, -157.0742950439453, 59.318138122558594, 107.3058090209961, -71.7950439453125, 83.02632141113281, -20.439056396484375, 17.175689697265625, 24.829927444458008, 60.6326904296875, 121.35427856445312, 2.3233871459960938, 112.9301986694336, 108.7073974609375, 141.87451171875, -2.453033447265625, -11.35711669921875, 116.5718994140625, 113.52824401855469, 144.35403442382812, -22.495803833007812, 15.616096496582031, 142.94680786132812, 127.06546783447266, 22.63604736328125, 36.04876708984375, 0.14736175537109375, 187.00100708007812, 193.9595947265625, 95.62893676757812, 18.1241455078125, -12.369903564453125, 40.282684326171875, 95.70956420898438, -33.17083740234375, 128.3319549560547, 25.277923583984375, 29.113983154296875, 156.7197265625, -41.720184326171875, 158.9814453125, 30.813720703125, 14.553482055664062, 164.93780517578125, 32.9898681640625, -55.353004455566406, 9.459564208984375, 125.33975219726562, 70.13760375976562, 139.0240020751953, -6.2075958251953125, 33.7109375, 15.340232849121094, -7.691118240356445, 134.4188232421875, -24.127197265625, 117.19798278808594, 118.22544860839844, 173.36083984375, -26.78179168701172, 21.567184448242188, 148.9359130859375, 13.326202392578125, 71.34782409667969, 1.0443267822265625, 143.07968139648438, 51.20172119140625, 33.329933166503906, 147.744873046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000319.npy"}
{"epoch": 0.6680628272251309, "step": 320, "batch_size": 128, "mean": 53.739280700683594, "std": 83.19242095947266, "min": -154.244140625, "p10": -38.6030662536621, "median": 59.886009216308594, "p90": 152.9424057006836, "max": 193.97567749023438, "pos_frac": 0.734375, "sample": [113.69346618652344, 84.34213256835938, -18.39288330078125, -141.078369140625, 106.20913696289062, 147.84423828125, -79.16827392578125, 7.3596954345703125, 145.68765258789062, 148.99575805664062, 18.85345458984375, 117.00009155273438, 18.4832763671875, 10.823486328125, 160.56243896484375, 82.41485595703125, -17.94842529296875, 0.17516326904296875, 15.10894775390625, 34.85204315185547, 16.9403076171875, 137.22482299804688, 40.93634033203125, 2.1506195068359375, 1.7821197509765625, -47.5570068359375, 167.5390625, 60.88215637207031, 79.5279769897461, 72.4222640991211, 101.2178726196289, 119.2725830078125, 153.44671630859375, 20.832427978515625, 48.492156982421875, 136.6519775390625, -11.5980224609375, 154.48861694335938, 6.891357421875, -0.9722023010253906, 5.879262924194336, 91.67353820800781, 114.70004272460938, -24.921783447265625, 161.00328063964844, 172.82781982421875, 107.72918701171875, 42.111083984375, 117.53240966796875, 91.40003967285156, -74.65115356445312, 9.38067626953125, -132.4879150390625, -31.07867431640625, 157.51190185546875, -16.150390625, 90.6387939453125, -12.832550048828125, 96.3299560546875, 149.5093994140625, 134.51007080078125, 146.9244384765625, 58.889862060546875, 139.29339599609375, 152.12673950195312, -26.39508056640625, 144.48672485351562, -15.64019775390625, 159.93814086914062, 137.85357666015625, 14.879570007324219, -6.055389404296875, 25.327606201171875, -13.1817626953125, -154.244140625, 98.3177490234375, 11.298095703125, 150.469482421875, 8.818084716796875, 129.2048797607422, -127.06124877929688, 34.421051025390625, 4.519763946533203, -4.02081298828125, -17.17034912109375, 65.18819427490234, 106.87359619140625, 141.28399658203125, 92.68270874023438, -1.8807220458984375, -93.13542175292969, 49.696685791015625, 184.80194091796875, -8.06512451171875, 123.4273681640625, -25.397823333740234, 10.653976440429688, 103.63585662841797, -26.468124389648438, 158.44818115234375, 84.44940185546875, -47.1083984375, 85.67315673828125, 152.7262725830078, 135.46075439453125, 193.97567749023438, 130.9123077392578, -18.100692749023438, 103.36280059814453, -116.89373779296875, 70.47412109375, 76.57559204101562, -133.10147094726562, -4.824310302734375, 159.47076416015625, 33.3128662109375, 144.11312866210938, 66.00395202636719, 113.91606140136719, 110.26170349121094, -34.957923889160156, 58.57025146484375, 24.41455078125, -123.83882904052734, 176.67831420898438, 58.24967956542969, -145.97642517089844, 123.07839965820312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000320.npy"}
{"epoch": 0.6701570680628273, "step": 321, "batch_size": 128, "mean": 44.80250549316406, "std": 76.02672576904297, "min": -156.7117919921875, "p10": -32.70491943359375, "median": 25.960060119628906, "p90": 149.74390869140626, "max": 184.0040283203125, "pos_frac": 0.7421875, "sample": [136.40249633789062, -16.172286987304688, 105.287109375, -146.783447265625, 25.951736450195312, 165.02426147460938, 25.9683837890625, 13.561553955078125, -57.22462463378906, 169.261474609375, 125.12620544433594, 19.726272583007812, -78.66374206542969, 60.05046844482422, 85.82914733886719, 3.6170654296875, -4.874992370605469, 139.39097595214844, 159.29940795898438, 2.1988983154296875, 83.03373718261719, -73.789794921875, 2.4810562133789062, -7.4620361328125, 153.11166381835938, 112.04499816894531, -1.9078826904296875, 4.679046630859375, 107.81646728515625, 7.392608642578125, 122.64109802246094, 93.44999694824219, 104.34414672851562, 15.388900756835938, 41.11236572265625, -51.339111328125, -30.28802490234375, 77.66477966308594, 6.2903289794921875, 131.857666015625, 66.22439575195312, 29.79315185546875, -7.8074951171875, 51.7554931640625, 10.1009521484375, 159.6919708251953, -32.45277404785156, 141.38941955566406, 184.0040283203125, 111.42770385742188, 12.253959655761719, 19.692428588867188, -28.46548843383789, 40.18603515625, 0.04000091552734375, 9.641067504882812, 147.38226318359375, -9.149368286132812, 145.82362365722656, -21.997100830078125, 50.708984375, 111.62218475341797, 29.82671356201172, 132.94622802734375, 66.01651000976562, 151.88214111328125, -32.3271484375, 48.054107666015625, -77.93474578857422, -5.7274627685546875, -10.120819091796875, 54.8148193359375, -33.29325866699219, 162.51666259765625, -149.31988525390625, 116.60197448730469, 164.08021545410156, 178.75796508789062, 22.651031494140625, 119.00311279296875, 20.486358642578125, -29.907958984375, 112.386962890625, 53.65025329589844, 7.8147430419921875, 149.66432189941406, -21.302703857421875, 149.9296112060547, -93.03317260742188, 125.30099487304688, 12.86386489868164, 15.480247497558594, 109.13258361816406, 0.25514984130859375, 17.87942123413086, 177.29986572265625, 3.7359619140625, 79.79519653320312, 26.050933837890625, 68.4481201171875, 123.5101547241211, -4.8048095703125, 101.65725708007812, 24.249557495117188, -10.208160400390625, 119.12554931640625, 5.647365570068359, 4.617982864379883, 1.1078910827636719, 1.304534912109375, -36.314056396484375, -69.7293701171875, 9.394575119018555, -21.203994750976562, 31.338096618652344, 167.37680053710938, -156.7117919921875, 143.58685302734375, 36.4310302734375, 74.3393325805664, 17.91668701171875, 49.92388916015625, 3.2078094482421875, -1.3288650512695312, 139.8251953125, 141.1505126953125, 0.0, -108.41201782226562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000321.npy"}
{"epoch": 0.6722513089005235, "step": 322, "batch_size": 128, "mean": 57.22248077392578, "std": 74.75283813476562, "min": -236.3870849609375, "p10": -29.053378295898433, "median": 51.599388122558594, "p90": 152.19632263183593, "max": 191.91290283203125, "pos_frac": 0.75, "sample": [-54.141021728515625, -9.65960693359375, -43.56683349609375, 101.62397766113281, 27.493392944335938, 13.067520141601562, 49.242767333984375, -40.26611328125, 136.1998291015625, 63.12841796875, -14.6798095703125, 17.464385986328125, 130.91448974609375, 12.30816650390625, 142.77322387695312, 117.01300048828125, 55.813140869140625, -0.45538330078125, 112.01580810546875, 81.11688232421875, 87.03564453125, -17.366241455078125, 163.21395874023438, 2.92218017578125, 0.44390869140625, 51.85322570800781, -27.934677124023438, 99.53385925292969, 6.914825439453125, 136.31985473632812, 85.15451049804688, 132.65402221679688, 1.5176887512207031, -10.692596435546875, 125.17893981933594, 34.234588623046875, 50.22692108154297, 136.7882080078125, -39.13262939453125, 122.9444351196289, 136.91632080078125, 115.69784545898438, -33.78556823730469, 15.67022705078125, -3.52655029296875, 65.62947082519531, 96.22895050048828, -31.663681030273438, 122.05709838867188, -126.60676574707031, 123.92611694335938, 3.807952880859375, 139.35128784179688, 183.1414794921875, -10.16632080078125, 130.12388610839844, 114.10787963867188, -15.299041748046875, 52.312286376953125, 119.649169921875, 157.7861328125, 85.21253967285156, 185.111572265625, 191.91290283203125, -63.210601806640625, 120.70647430419922, 91.16896057128906, 49.881866455078125, 110.22171020507812, 13.357223510742188, 168.67999267578125, 151.19570922851562, 104.84564208984375, 58.274261474609375, 10.869171142578125, 0.0, -59.70989990234375, 46.14306640625, 51.345550537109375, -33.91835021972656, 153.826416015625, 46.843421936035156, 135.517578125, 151.49771118164062, -20.715484619140625, 60.823455810546875, 51.14599609375, 5.86688232421875, -0.85321044921875, 122.50965881347656, 166.59927368164062, -2.009918212890625, 8.483795166015625, -41.129425048828125, -14.668701171875, 0.4752922058105469, 95.02316284179688, 3.9576568603515625, -236.3870849609375, 67.22076416015625, 134.9603271484375, 72.62547302246094, 34.634674072265625, 168.0447998046875, 143.4366455078125, -39.31244659423828, -1.8521270751953125, 186.4611053466797, 183.01644897460938, 6.353549957275391, -12.9215087890625, 132.49986267089844, 19.183982849121094, 43.63456726074219, -8.201889038085938, 11.240848541259766, -21.0927734375, 18.22589111328125, 21.07244873046875, 109.01931762695312, 75.68734741210938, 173.73330688476562, 72.47418212890625, -6.52740478515625, 18.5330810546875, 141.89065551757812, 52.28363037109375, 164.68392944335938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000322.npy"}
{"epoch": 0.6743455497382199, "step": 323, "batch_size": 128, "mean": 35.42110061645508, "std": 85.5790786743164, "min": -207.78656005859375, "p10": -71.17796020507811, "median": 20.636810302734375, "p90": 145.92193450927735, "max": 185.31805419921875, "pos_frac": 0.671875, "sample": [13.94268798828125, 14.895492553710938, -132.67872619628906, -114.65509033203125, 124.79385375976562, 136.34352111816406, -84.51835632324219, 85.70545959472656, 20.7386474609375, 178.64451599121094, -97.94822692871094, 135.28753662109375, 27.00726318359375, -18.735595703125, 5.4747314453125, 108.69390106201172, 146.4683074951172, -10.815963745117188, 66.79991149902344, -30.79803466796875, 179.45513916015625, 104.86709594726562, 21.746826171875, 135.49484252929688, 22.871124267578125, 162.84336853027344, 13.6416015625, 129.99313354492188, -88.96189880371094, 143.29617309570312, -14.837188720703125, -181.46893310546875, -6.337760925292969, 13.5618896484375, -0.4279308319091797, 105.26750183105469, 20.53497314453125, 6.000347137451172, 97.19268798828125, 17.942354202270508, 29.080535888671875, 131.99591064453125, -30.4169921875, 69.32110595703125, 131.62820434570312, -23.52203369140625, -19.960067749023438, -207.78656005859375, 150.71029663085938, 24.75293731689453, 7.946752548217773, 0.5656166076660156, -56.60557556152344, 9.021522521972656, 55.81813049316406, 65.10556030273438, 49.14434814453125, 136.14471435546875, 180.45774841308594, -25.61249542236328, 80.58477783203125, 74.0496826171875, 124.5831527709961, -19.696533203125, 155.49932861328125, 23.093292236328125, 126.143798828125, -14.723037719726562, -47.92242431640625, 94.49163818359375, 29.71636962890625, 2.1426753997802734, -65.46064758300781, -0.6583251953125, 123.77958679199219, 77.32713317871094, 171.17901611328125, -63.22041320800781, 159.53018188476562, 3.58013916015625, 121.54630279541016, -143.03187561035156, 3.8701744079589844, 30.455291748046875, 19.653045654296875, 170.85025024414062, -46.20245361328125, 144.2030029296875, -123.91531372070312, 1.1776084899902344, 136.2615966796875, 26.181167602539062, -43.71546936035156, 112.17337799072266, -7.767852783203125, -8.912857055664062, -90.58251190185547, -12.434661865234375, -37.2298583984375, -7.231292724609375, 19.311004638671875, -102.07565307617188, -11.833526611328125, 5.302215576171875, 26.28314208984375, 50.741546630859375, 19.728580474853516, -9.234359741210938, 124.83404541015625, 170.27313232421875, -2.6817798614501953, 41.30787658691406, 148.9280242919922, 143.65513610839844, -128.05857849121094, -140.65933227539062, 78.71955871582031, 24.161041259765625, 100.84855651855469, 18.819908142089844, 17.109025955200195, -5.16119384765625, 103.97654724121094, 145.68777465820312, 130.0838623046875, -36.159423828125, 0.19752883911132812, 185.31805419921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000323.npy"}
{"epoch": 0.6764397905759162, "step": 324, "batch_size": 128, "mean": 61.765869140625, "std": 71.78398132324219, "min": -104.28104400634766, "p10": -12.930718231201169, "median": 37.52788543701172, "p90": 155.78222045898437, "max": 197.53973388671875, "pos_frac": 0.796875, "sample": [65.60990905761719, 155.5648193359375, 15.52801513671875, 125.78460693359375, 23.823974609375, 116.4456787109375, 2.8665313720703125, 8.591135025024414, 133.18826293945312, 32.23167419433594, 35.1829833984375, 2.3385238647460938, -68.91558837890625, 159.2496337890625, 5.134979248046875, 164.53488159179688, 123.9178237915039, 14.398822784423828, -9.02691650390625, 0.0, 8.57635498046875, 6.385902404785156, 127.70489501953125, -8.346614837646484, 4.515453338623047, 32.266082763671875, 139.244384765625, -39.853233337402344, 127.73135375976562, -32.66844177246094, 5.9723663330078125, 139.75949096679688, 143.59535217285156, 154.86911010742188, 128.9176483154297, 124.69005584716797, -14.8634033203125, 52.167755126953125, 77.8819580078125, 0.0, -6.691291809082031, 143.71978759765625, 157.24951171875, 114.14952087402344, 53.932647705078125, 129.44168090820312, 0.0, -1.203460693359375, 108.54220581054688, -21.91571044921875, -85.92521667480469, 39.87278747558594, 30.021926879882812, 22.4561767578125, 103.4354248046875, 92.5648193359375, 10.680068969726562, 138.71981811523438, 159.15264892578125, 8.963134765625, -46.32086944580078, 163.83837890625, 10.553878784179688, 130.67083740234375, 108.17157745361328, -1.8884124755859375, 118.3683090209961, 126.462158203125, 16.697845458984375, -104.28104400634766, 154.7745819091797, 99.13678741455078, -82.61194610595703, 127.12370300292969, 15.777847290039062, -2.57391357421875, 21.288223266601562, 75.2496337890625, 10.508010864257812, 197.53973388671875, 181.38885498046875, 7.156028747558594, -18.18524932861328, 80.848388671875, 178.040771484375, 148.3511962890625, 16.605987548828125, 29.4144287109375, -17.435028076171875, 5.550323486328125, 135.16476440429688, 44.891693115234375, 24.344093322753906, 1.846710205078125, -11.491043090820312, 55.389251708984375, 29.972217559814453, 158.61981201171875, 131.14682006835938, 61.1588134765625, -42.821205139160156, -4.729570388793945, 7.37115478515625, 41.5806884765625, -48.80279541015625, 189.85791015625, 177.1505126953125, 161.44866943359375, 28.244216918945312, 5.243499755859375, 148.34262084960938, 104.17066955566406, -12.102424621582031, 127.88819885253906, 26.131500244140625, 128.9706573486328, 156.28948974609375, 128.07315063476562, -0.7375926971435547, 146.5336456298828, 130.8173828125, 1.3816070556640625, 131.833251953125, 139.4288330078125, 137.84365844726562, 9.730789184570312, 2.38330078125, 21.082199096679688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000324.npy"}
{"epoch": 0.6785340314136126, "step": 325, "batch_size": 128, "mean": 49.687217712402344, "std": 89.3533935546875, "min": -142.89804077148438, "p10": -43.51147766113279, "median": 31.535362243652344, "p90": 168.8535369873047, "max": 246.93606567382812, "pos_frac": 0.671875, "sample": [64.22317504882812, 143.33837890625, -115.62771606445312, 16.97161865234375, 103.9327392578125, 23.85272216796875, 35.356689453125, 146.11465454101562, 32.40660095214844, -2.27911376953125, 11.623779296875, 12.9307861328125, 139.84727478027344, 140.48468017578125, -10.169830322265625, 4.428264617919922, 156.34681701660156, 83.8397216796875, 59.71075439453125, 173.05313110351562, 139.10968017578125, 171.72862243652344, 113.17562866210938, -3.3524246215820312, 129.4053955078125, 3.8230361938476562, 13.324569702148438, 77.53877258300781, 127.05397033691406, 29.17413330078125, -104.87266540527344, 7.60125732421875, 82.31349182128906, 161.0377655029297, 157.88275146484375, 48.98216247558594, 21.514141082763672, 162.35154724121094, 30.692489624023438, -76.76559448242188, 136.31182861328125, -142.37533569335938, 152.91709899902344, -1.6983184814453125, -80.98933410644531, -1.30731201171875, 246.93606567382812, -125.40232849121094, 182.05393981933594, -3.19854736328125, -1.38665771484375, 32.37823486328125, 159.48251342773438, 13.171783447265625, 8.568572998046875, -119.0962142944336, -8.529083251953125, -115.92484283447266, -3.866668701171875, -58.628021240234375, 69.19448852539062, 134.0174560546875, -7.443473815917969, -2.689300537109375, 80.89645385742188, 35.96076965332031, -14.784408569335938, -117.48129272460938, 191.21395874023438, 54.413604736328125, 104.05715942382812, 20.296218872070312, 238.570068359375, -12.605499267578125, -19.45477294921875, 0.2537994384765625, -12.060333251953125, -9.195137023925781, 101.8309326171875, 144.36492919921875, -133.73782348632812, -11.82757568359375, -31.654022216796875, 171.68023681640625, -106.69204711914062, 72.1851806640625, 125.367431640625, 40.0634765625, 45.13148498535156, -29.0238037109375, -21.32098388671875, 115.02760314941406, 168.49786376953125, -28.913223266601562, 2.416658401489258, -142.89804077148438, -14.771713256835938, 175.0152587890625, 36.858245849609375, 134.55224609375, 169.68344116210938, 48.27894592285156, 138.7512969970703, 196.17431640625, 158.0635986328125, -37.032958984375, 5.568164825439453, 22.16576385498047, 20.423660278320312, -2.17889404296875, -20.678695678710938, 150.71600341796875, 50.17326354980469, 135.86593627929688, 8.320472717285156, 125.36771392822266, 116.349853515625, 185.14239501953125, 13.467620849609375, 0.34807586669921875, -6.36724853515625, -30.306060791015625, 202.58297729492188, -22.15509033203125, -15.05035400390625, 199.58773803710938, 71.471435546875, 148.39651489257812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000325.npy"}
{"epoch": 0.680628272251309, "step": 326, "batch_size": 128, "mean": 52.13031005859375, "std": 83.63520050048828, "min": -189.9520263671875, "p10": -46.51870117187499, "median": 53.094852447509766, "p90": 148.7386291503906, "max": 178.63340759277344, "pos_frac": 0.71875, "sample": [141.14431762695312, 133.80105590820312, 127.01754760742188, 172.92613220214844, 88.63099670410156, 62.5518798828125, -42.87200927734375, -78.18032836914062, 131.91128540039062, -71.02592468261719, 128.93035888671875, -11.92401123046875, 135.9615936279297, 137.6383819580078, 22.12749481201172, 3.5350685119628906, 5.540985107421875, 150.36782836914062, 178.63340759277344, 20.81243896484375, 45.5025634765625, 138.48316955566406, 4.777099609375, -20.27850341796875, 61.89862060546875, 106.31913757324219, 120.0596923828125, 136.6976318359375, 77.24783325195312, 17.639434814453125, -119.07311248779297, 148.099609375, 0.26611328125, 72.85499572753906, -127.61279296875, -93.27679443359375, 15.091190338134766, 96.22652435302734, -31.8023681640625, 137.16107177734375, 81.64703369140625, 112.33688354492188, -23.9625244140625, 144.77841186523438, 61.189117431640625, 118.64683532714844, -41.52436828613281, 136.7757110595703, -22.043182373046875, 163.5149383544922, 126.34332275390625, 102.23002624511719, 143.60574340820312, 17.21656036376953, 145.07199096679688, -122.88567352294922, -36.70777893066406, 10.812591552734375, 127.426513671875, -11.46664810180664, 125.29635620117188, 112.97771453857422, 24.56982421875, 88.65582275390625, 0.0, -189.9520263671875, 14.60275650024414, -1.67205810546875, 128.16793823242188, -33.180877685546875, 18.222427368164062, 160.06082153320312, 173.63934326171875, 58.7117919921875, 55.44751739501953, -55.02764892578125, 34.43431854248047, -72.10868835449219, 31.57452392578125, 116.60965728759766, -1.824310302734375, 157.364501953125, -5.1986083984375, 25.743621826171875, 13.513694763183594, 135.82083129882812, -31.33929443359375, -90.922119140625, 109.66390991210938, 10.294952392578125, 135.70492553710938, 17.818145751953125, 160.06793212890625, -9.664031982421875, 73.66790008544922, 28.33050537109375, 97.66068267822266, 150.22967529296875, 26.603515625, 132.17916870117188, 157.17642211914062, 0.0, 131.40512084960938, 50.7421875, 60.9256591796875, 131.7964630126953, -22.58367919921875, 142.88868713378906, 109.10194396972656, 0.0, 132.73129272460938, -146.53048706054688, -32.545013427734375, -120.04570770263672, 17.585968017578125, -119.98672485351562, 155.83090209960938, -11.91461181640625, 145.86956787109375, 20.426944732666016, 35.318824768066406, 25.901763916015625, -16.710205078125, -39.23503875732422, 154.65701293945312, 150.68612670898438, 40.299835205078125, 135.32647705078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000326.npy"}
{"epoch": 0.6827225130890052, "step": 327, "batch_size": 128, "mean": 58.944976806640625, "std": 81.74337768554688, "min": -155.58212280273438, "p10": -23.911407470703125, "median": 39.29943084716797, "p90": 161.41968688964843, "max": 215.80569458007812, "pos_frac": 0.78125, "sample": [36.24463653564453, 134.8755340576172, 120.60929870605469, 1.350738525390625, 192.68780517578125, -143.26596069335938, 102.45913696289062, 140.80374145507812, 6.6888885498046875, 113.15116882324219, 1.0198497772216797, 179.26177978515625, -7.15582275390625, -131.59185791015625, 94.5800552368164, -19.04107666015625, -60.89642333984375, 10.806015014648438, 174.4327392578125, 168.26712036132812, 113.47252655029297, 103.2213363647461, 20.048187255859375, 64.98248291015625, 115.09396362304688, -57.44305419921875, 15.575729370117188, 109.06614685058594, 12.121978759765625, 125.09305572509766, 39.664764404296875, 173.71405029296875, -4.740364074707031, -9.764762878417969, 73.73101806640625, 112.38497924804688, 167.111083984375, 8.56341552734375, 215.80569458007812, 120.53045654296875, -24.111572265625, -6.411628723144531, -9.203781127929688, 113.22805786132812, 33.78118133544922, -12.689788818359375, 159.53549194335938, 38.93409729003906, 114.05180358886719, 10.622421264648438, 149.9345703125, 5.1246337890625, -5.58807373046875, -155.58212280273438, 118.14188385009766, 31.180084228515625, 137.130615234375, 129.9106903076172, 174.05712890625, 34.441680908203125, 66.37371826171875, 133.513671875, 111.06272888183594, 43.684783935546875, 160.88327026367188, 147.18487548828125, 169.04849243164062, 32.836029052734375, 108.15708923339844, -20.199066162109375, 31.194305419921875, 4.991058349609375, -10.36236572265625, -70.05622100830078, 108.72317504882812, 10.5892333984375, 30.948623657226562, 52.87054443359375, 168.83999633789062, 11.480712890625, -23.82562255859375, 4.915618896484375, 23.0814208984375, 117.10491943359375, 22.6058349609375, 162.67132568359375, 12.959442138671875, 0.0, 149.86746215820312, 20.139503479003906, 25.834136962890625, 157.65771484375, 184.18121337890625, 20.90380859375, 128.7822265625, 20.115318298339844, 138.06838989257812, 34.13194274902344, 139.7253875732422, 141.13148498535156, 61.14019775390625, 4.7524261474609375, -145.04656982421875, -12.715171813964844, 106.341552734375, 115.05684661865234, -97.67987823486328, 15.820556640625, 61.6893310546875, 131.8069610595703, -132.95159912109375, 60.295013427734375, 166.10443115234375, 8.288818359375, -39.109893798828125, 149.39276123046875, -4.353355407714844, 160.24041748046875, 1.0451297760009766, -6.1298370361328125, 155.36041259765625, 147.40261840820312, 136.58413696289062, 5.4201202392578125, -66.07015991210938, 143.90167236328125, 19.221511840820312, -42.6751708984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000327.npy"}
{"epoch": 0.6848167539267016, "step": 328, "batch_size": 128, "mean": 51.67345428466797, "std": 76.16577911376953, "min": -142.0694580078125, "p10": -24.138575744628906, "median": 35.295188903808594, "p90": 146.57872009277344, "max": 201.672119140625, "pos_frac": 0.765625, "sample": [23.15654754638672, 81.35516357421875, 144.9525146484375, 36.309814453125, 101.62796020507812, 23.14117431640625, 3.0032958984375, 124.94963073730469, 126.51170349121094, 122.72367858886719, 118.01643371582031, 82.82437133789062, 4.986667633056641, 115.3603515625, 63.583290100097656, 0.42511940002441406, 11.36328125, -5.178680419921875, 45.7640380859375, 15.308303833007812, 32.01774597167969, 103.33332824707031, 96.7022705078125, -25.817378997802734, 180.06170654296875, 40.915740966796875, -2.1916351318359375, 41.755126953125, 106.19024658203125, 118.87313079833984, -3.163116455078125, 122.77904510498047, 35.58320617675781, 156.78408813476562, 193.74798583984375, 99.53360748291016, 168.31202697753906, 5.031150817871094, 172.128662109375, -0.5546894073486328, 4.4090576171875, 136.2086639404297, 87.77896118164062, 0.459014892578125, 7.982208251953125, -7.53831672668457, -21.7266845703125, 21.940704345703125, -24.304443359375, 138.17930603027344, 27.79705810546875, -2.4906005859375, -24.6409912109375, 164.488037109375, 9.172882080078125, 33.090576171875, 10.99591064453125, 10.872268676757812, 39.21746063232422, -24.067489624023438, 16.880088806152344, 44.41931915283203, 143.60122680664062, 149.13262939453125, 64.46774291992188, 157.465576171875, 121.4795913696289, -70.00604248046875, 135.64541625976562, -102.86341857910156, 13.981796264648438, 121.79229736328125, 122.2842788696289, 125.47804260253906, 49.475067138671875, 191.8994140625, 178.40679931640625, 17.353118896484375, 4.6077880859375, 134.0826416015625, -132.60818481445312, 8.11993408203125, 140.73614501953125, 38.167327880859375, -113.12667846679688, 7.033935546875, 122.38731384277344, 123.36847686767578, 121.45010375976562, -21.4776611328125, -15.035797119140625, 165.33953857421875, 5.3351287841796875, -1.6691741943359375, 5.9056396484375, -20.745773315429688, -3.237213134765625, 22.7806396484375, -114.23532104492188, 30.550722122192383, 77.35734558105469, 34.11883544921875, 116.62517547607422, 4.0899658203125, -1.06842041015625, 142.03762817382812, 142.27880859375, -142.0694580078125, -2.907745361328125, -26.03167724609375, 35.007171630859375, 19.441070556640625, 141.05694580078125, -43.958984375, 39.5047607421875, 22.235824584960938, 1.3414154052734375, 145.69326782226562, -5.074186325073242, -135.43553161621094, -3.5568695068359375, 148.644775390625, 101.96934509277344, -92.39598083496094, 66.47381591796875, 201.672119140625, 124.57455444335938, 43.9249267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000328.npy"}
{"epoch": 0.6869109947643979, "step": 329, "batch_size": 128, "mean": 44.138092041015625, "std": 87.91250610351562, "min": -172.56048583984375, "p10": -48.623979187011706, "median": 27.241580963134766, "p90": 157.7315734863281, "max": 210.2095947265625, "pos_frac": 0.671875, "sample": [-11.031646728515625, 6.832122802734375, -5.9827117919921875, 42.899261474609375, 2.413909912109375, -16.632213592529297, 6.3098602294921875, 25.985946655273438, 3.7657470703125, 42.24676513671875, 142.19772338867188, -17.356521606445312, -142.26416015625, 119.25820922851562, 107.48556518554688, 175.81304931640625, -12.32696533203125, -37.46435546875, 143.8973846435547, 120.46200561523438, 56.11688232421875, 16.52569580078125, 153.69595336914062, -8.878326416015625, 125.07987976074219, 152.88099670410156, -129.0158233642578, 117.08135986328125, 173.0193634033203, 48.304718017578125, 1.30169677734375, 21.520263671875, -1.0303955078125, 154.90960693359375, -41.2772216796875, 42.306640625, 102.11984252929688, 103.05438232421875, 18.800979614257812, 118.64404296875, 42.65216064453125, 168.4962158203125, 2.09820556640625, -33.9853515625, 59.096771240234375, -88.026123046875, -9.326583862304688, 151.23388671875, 15.232421875, -172.56048583984375, -4.849285125732422, 167.40484619140625, -57.24993896484375, -21.1842041015625, 136.41375732421875, 33.90864562988281, 92.30497741699219, 29.794296264648438, 27.942245483398438, -123.66217041015625, 7.817329406738281, 141.28738403320312, -0.11383056640625, -44.92713928222656, -0.6822586059570312, -4.985603332519531, 0.00841522216796875, -29.782806396484375, 197.4464111328125, 26.540916442871094, 35.04447937011719, 0.08758544921875, 128.873046875, 206.92724609375, 14.926183700561523, 121.82208251953125, 62.026458740234375, -22.467697143554688, -106.23273468017578, 149.48928833007812, 135.77671813964844, -14.364715576171875, -163.37762451171875, 119.18443298339844, -15.241485595703125, 170.9147491455078, 51.56036376953125, 3.624725341796875, 137.27584838867188, 0.5175933837890625, 186.0228271484375, -1.3359756469726562, -99.70582580566406, 171.539306640625, 129.98899841308594, -126.79046630859375, 204.83470153808594, 164.316162109375, -1.8494110107421875, 127.5850830078125, -5.2958526611328125, 13.626739501953125, 105.31791687011719, 34.979156494140625, -0.9643325805664062, 1.9561386108398438, 210.2095947265625, 59.11962890625, 140.73353576660156, 210.1114501953125, -17.292694091796875, 124.21163940429688, 128.8897705078125, -7.291351318359375, 36.34283447265625, -106.68622589111328, 25.36309814453125, -23.986541748046875, 31.129440307617188, -0.9481201171875, 78.77606964111328, 130.6329345703125, 1.9048290252685547, 117.85889434814453, 136.6637420654297, 44.54499816894531, -79.009765625, -142.20632934570312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000329.npy"}
{"epoch": 0.6890052356020943, "step": 330, "batch_size": 128, "mean": 37.060279846191406, "std": 80.6889877319336, "min": -131.01502990722656, "p10": -71.37902679443359, "median": 22.478763580322266, "p90": 152.73948974609374, "max": 249.4842529296875, "pos_frac": 0.7109375, "sample": [-10.542938232421875, 135.0107421875, 18.683685302734375, -76.81961059570312, 65.17032623291016, 112.81051635742188, 56.464874267578125, 175.41494750976562, 145.57566833496094, 30.8929443359375, -10.93994140625, 14.78240966796875, 19.776214599609375, -125.548095703125, 42.123779296875, -69.04734802246094, 154.41583251953125, 28.50616455078125, 0.8157958984375, -3.21136474609375, 157.94078063964844, -103.70600891113281, -33.115997314453125, 9.0638427734375, -1.096343994140625, 116.3170166015625, 24.9149169921875, -19.829315185546875, 12.099594116210938, -0.5998687744140625, 59.066314697265625, 68.90138244628906, 171.260009765625, 106.01704406738281, 8.620956420898438, -6.062103271484375, -1.76373291015625, -96.28652954101562, -13.96917724609375, 17.422210693359375, 45.71539306640625, 167.9595947265625, 34.5897216796875, 41.202728271484375, 53.0338134765625, 26.393951416015625, 17.09390640258789, 188.2626953125, 12.96917724609375, 104.36370849609375, 2.377443313598633, 150.75009155273438, 54.61920166015625, -129.81045532226562, 7.709625244140625, 22.495101928710938, -120.91960906982422, 33.30218505859375, 41.78936767578125, 54.632347106933594, 15.074844360351562, 169.33070373535156, 13.201126098632812, 108.430419921875, -131.01502990722656, 1.989593505859375, 130.53076171875, -109.70074462890625, 5.681697845458984, 9.659652709960938, 42.533966064453125, -60.08473205566406, 173.90341186523438, 39.6787109375, -2.6214599609375, 90.12515258789062, -128.6832733154297, 172.7955322265625, -33.2452392578125, 1.549530029296875, 21.9197998046875, 91.61561584472656, -6.2729339599609375, -11.878082275390625, 15.891265869140625, 2.2625579833984375, 32.440673828125, 152.02105712890625, 149.791748046875, 90.1449203491211, 171.79307556152344, 10.129638671875, 106.42345428466797, 103.04336547851562, 15.80462646484375, -36.449371337890625, 16.0330810546875, 13.933923721313477, 93.01776885986328, 136.25344848632812, 172.35546875, 34.861724853515625, -118.25506591796875, 89.76912689208984, -116.13363647460938, 147.10855102539062, -16.06109619140625, 22.462425231933594, -22.187541961669922, 37.53004455566406, 141.2947998046875, -104.06729888916016, -0.8604660034179688, -34.572235107421875, 0.599365234375, 183.1827392578125, 41.905609130859375, 43.77561950683594, -1.654327392578125, 134.4305877685547, -25.07879638671875, 249.4842529296875, 75.29876708984375, -84.7586669921875, -25.504364013671875, 134.86526489257812, 89.80530548095703, 29.005767822265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000330.npy"}
{"epoch": 0.6910994764397905, "step": 331, "batch_size": 128, "mean": 46.4903450012207, "std": 75.94818878173828, "min": -166.99136352539062, "p10": -21.95782775878906, "median": 26.457321166992188, "p90": 157.25103454589842, "max": 193.1134033203125, "pos_frac": 0.7578125, "sample": [0.08280181884765625, 185.335693359375, 74.44482421875, -94.54251098632812, 193.1134033203125, 113.45950317382812, 55.357666015625, -4.141265869140625, 44.08905029296875, 60.156951904296875, -2.56756591796875, -73.57899475097656, 46.314239501953125, -23.86883544921875, 179.9921875, -7.022705078125, 4.5428619384765625, 4.035451889038086, -9.260955810546875, -2.73052978515625, 13.72784423828125, 43.133819580078125, -31.02545166015625, 19.526565551757812, 10.297523498535156, 154.99465942382812, 138.45806884765625, 41.78741455078125, 3.928802490234375, 127.48464965820312, 31.97088623046875, 41.25177001953125, -14.66888427734375, 29.417877197265625, 27.148193359375, 158.41021728515625, 5.47686767578125, 57.879547119140625, 167.044677734375, 125.7883071899414, -28.114173889160156, 13.488922119140625, -3.498931884765625, -159.12596130371094, 4.497138977050781, 6.612804412841797, 155.81455993652344, 25.284423828125, 34.73602294921875, 0.0, 27.218170166015625, -0.6994476318359375, 115.48776245117188, -12.64166259765625, 10.131805419921875, 84.54747009277344, 7.161865234375, 143.74600219726562, 40.292877197265625, 56.0482177734375, 24.759918212890625, 50.6180419921875, -29.4466552734375, -21.138824462890625, 176.00033569335938, 28.612348556518555, 158.81509399414062, 120.64433288574219, -11.77203369140625, 38.240509033203125, 4.561614990234375, 25.766448974609375, 149.46685791015625, 7.03558349609375, 60.707462310791016, 18.493289947509766, -7.506439208984375, 7.690032958984375, 9.718269348144531, -36.823814392089844, 21.621337890625, 29.522979736328125, 122.37332153320312, 35.03297424316406, -166.99136352539062, 185.3275146484375, 5.59393310546875, 156.75424194335938, -16.467605590820312, 28.559112548828125, 176.9976806640625, 122.81939697265625, 108.92452239990234, 179.04241943359375, 51.340240478515625, 164.20782470703125, 114.35633850097656, 4.246971130371094, -14.84039306640625, -7.90509033203125, -40.04168701171875, 15.04150390625, 150.32728576660156, 118.9005126953125, -138.43258666992188, 14.651206970214844, 189.42510986328125, 1.9691581726074219, 131.01580810546875, -2.119649887084961, 1.5317535400390625, 14.696975708007812, 137.85781860351562, 13.1893310546875, 114.48403930664062, 127.72601318359375, -4.745849609375, 69.8841552734375, 17.53750991821289, 126.13599395751953, -145.95510864257812, 23.236305236816406, 175.72628784179688, 137.89990234375, 93.79537963867188, -32.322265625, 133.69308471679688, 14.464759826660156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000331.npy"}
{"epoch": 0.6931937172774869, "step": 332, "batch_size": 128, "mean": 58.09014892578125, "std": 84.01102447509766, "min": -178.16542053222656, "p10": -34.76749725341796, "median": 50.69932556152344, "p90": 156.5968536376953, "max": 225.5726318359375, "pos_frac": 0.765625, "sample": [80.32681274414062, 18.731201171875, 11.999519348144531, 184.73724365234375, 123.87141418457031, 132.28512573242188, 110.92362976074219, 54.756019592285156, 188.291259765625, 130.60867309570312, -1.7666778564453125, 10.175058364868164, 8.874465942382812, -4.347541809082031, 54.80766296386719, 109.4072265625, -31.569290161132812, 9.396163940429688, 142.2269287109375, -60.4571533203125, 147.86349487304688, -6.6100006103515625, 44.52862548828125, 22.524471282958984, 0.11133575439453125, 7.69659423828125, -10.47808837890625, 5.878021240234375, 13.30841064453125, 10.068801879882812, 145.45880126953125, 164.46600341796875, 4.138580322265625, 3.899566650390625, 152.49453735351562, 111.41130828857422, 136.05657958984375, 167.772216796875, 44.36798095703125, 124.55705261230469, 121.20706176757812, 7.931671142578125, 131.438232421875, 0.531402587890625, 140.76629638671875, 46.962860107421875, 91.69326782226562, 148.68910217285156, -5.34088134765625, 25.91790771484375, 147.114501953125, 179.484130859375, -106.91372680664062, -86.2603759765625, 183.02716064453125, 128.56201171875, -86.74139404296875, 102.49418640136719, 144.63018798828125, -162.58059692382812, 32.004173278808594, 153.90255737304688, 141.4686279296875, -2.149106979370117, 13.051750183105469, 7.1393585205078125, 99.41449737548828, 21.83990478515625, 12.678680419921875, -42.22998046875, 4.45404052734375, 159.25772094726562, 111.04261779785156, 19.898696899414062, -1.7660369873046875, 168.36529541015625, 54.435791015625, 121.9611587524414, -89.14016723632812, 89.45965576171875, -15.278762817382812, 111.16714477539062, -24.229019165039062, 2.6519622802734375, 92.27230834960938, -107.97308349609375, 119.838623046875, 19.963592529296875, 5.7125244140625, 114.92295837402344, 146.44924926757812, 111.27130126953125, -95.70218658447266, 146.83255004882812, 69.14825439453125, -10.1810302734375, 141.6590576171875, -21.830596923828125, -15.122207641601562, 71.65214538574219, 162.10385131835938, 154.3004150390625, -128.69573974609375, -80.43060302734375, 225.5726318359375, 187.18023681640625, 126.80613708496094, 20.489593505859375, 46.91908264160156, 14.822601318359375, 155.45648193359375, 73.01481628417969, 123.1167221069336, 179.63632202148438, 55.32562255859375, 132.6309814453125, -25.526031494140625, 146.38482666015625, -9.480754852294922, 118.35846710205078, 117.5675048828125, 172.1888427734375, -1.0652084350585938, -62.50877380371094, 0.0, 11.544647216796875, -178.16542053222656, 44.27301025390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000332.npy"}
{"epoch": 0.6952879581151833, "step": 333, "batch_size": 128, "mean": 43.63203430175781, "std": 74.04560089111328, "min": -136.04660034179688, "p10": -44.432504653930664, "median": 29.578073501586914, "p90": 140.59473266601563, "max": 172.79522705078125, "pos_frac": 0.71875, "sample": [121.05968475341797, -121.317138671875, -43.860042572021484, 164.07553100585938, 159.10867309570312, -43.35333251953125, 74.07345581054688, 50.694976806640625, 58.142852783203125, 76.37799072265625, -51.39161682128906, 8.154541015625, -17.466583251953125, 154.14096069335938, 140.61810302734375, 133.53961181640625, 117.63814544677734, 6.698394775390625, 114.45084381103516, 60.2464599609375, -21.89019775390625, 34.66968536376953, 7.235687255859375, 142.3870849609375, 137.7642364501953, 93.64186096191406, 55.666748046875, 98.14056396484375, 132.89035034179688, -7.5821990966796875, 70.341064453125, 164.7760467529297, -9.4976806640625, 75.8671875, -136.04660034179688, 3.5926742553710938, 51.187225341796875, -11.85675048828125, 132.29421997070312, 121.72251892089844, 41.056671142578125, 12.003429412841797, 12.11358642578125, 13.3172607421875, 81.0833740234375, 138.01022338867188, 4.508678436279297, 161.10140991210938, 24.1729736328125, -71.56324768066406, -65.111328125, 66.28591918945312, 0.0, -6.003204345703125, 134.2098388671875, 17.471466064453125, 128.09906005859375, 160.45233154296875, 140.584716796875, 71.20362854003906, -1.99432373046875, 34.1356201171875, -57.62164306640625, 135.08270263671875, 84.52099609375, -29.404327392578125, 143.40463256835938, 172.79522705078125, 69.49336242675781, 32.5643310546875, -107.76380920410156, 80.08383178710938, 32.37200927734375, -65.97012329101562, 8.396858215332031, 132.67572021484375, 0.1220703125, -2.43194580078125, 29.400115966796875, 119.34395599365234, 26.74315643310547, 21.289817810058594, 28.455734252929688, 146.76522827148438, -25.106201171875, 137.0165252685547, 134.07952880859375, -7.25592041015625, -45.76824951171875, 146.38194274902344, 29.756031036376953, 69.33602905273438, -8.970939636230469, 100.16796875, -26.459564208984375, -18.434326171875, 20.231964111328125, 21.50799560546875, -123.54057312011719, 109.86943054199219, -8.576019287109375, 116.9112319946289, -128.48529052734375, -6.76385498046875, 117.51739501953125, 131.82049560546875, 1.730255126953125, 47.41015625, 143.0560302734375, 8.3251953125, 125.97997283935547, 3.8444557189941406, 123.87393188476562, 7.396820068359375, 28.14605140686035, -6.98663330078125, -94.28559875488281, 85.6013412475586, -21.47369384765625, 28.326019287109375, -8.2672119140625, 6.72357177734375, -110.260986328125, 12.785850524902344, -1.325286865234375, 56.37995910644531, 18.070220947265625, 2.2231101989746094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000333.npy"}
{"epoch": 0.6973821989528796, "step": 334, "batch_size": 128, "mean": 51.579376220703125, "std": 84.8083267211914, "min": -183.7168426513672, "p10": -28.00796661376953, "median": 34.73442077636719, "p90": 163.9738296508789, "max": 268.26611328125, "pos_frac": 0.7265625, "sample": [121.0105972290039, 105.6143798828125, -27.5380859375, -9.73895263671875, 0.0, -7.967864990234375, 2.832244873046875, -10.606826782226562, 31.873720169067383, 1.4311408996582031, 26.629119873046875, 24.991912841796875, 50.313385009765625, 129.67446899414062, -122.5556640625, 130.01986694335938, -2.10321044921875, 165.19692993164062, 32.13975524902344, -15.1524658203125, 116.69132995605469, 63.01104736328125, 200.86793518066406, 163.59474182128906, -94.4970703125, 35.567291259765625, 19.4561767578125, 47.206787109375, -2.1499404907226562, 17.622787475585938, 54.51318359375, -9.506828308105469, -0.6097774505615234, 87.384033203125, 130.4581298828125, 33.84918212890625, -2.4018096923828125, 121.68792724609375, -161.65936279296875, 150.79888916015625, -0.6115932464599609, 8.15625, 19.59423828125, 17.4346923828125, 30.36377716064453, 47.9605712890625, 11.8173828125, -23.8270263671875, 128.17257690429688, -0.752655029296875, 76.73284912109375, -123.411376953125, 152.09715270996094, 85.7515869140625, 188.27691650390625, 129.8812255859375, -4.675626754760742, 111.45742797851562, 9.686050415039062, 38.77619934082031, 12.838005065917969, 7.475696563720703, 189.9718017578125, 57.241180419921875, -29.104354858398438, -119.61131286621094, 45.993072509765625, -122.82806396484375, 6.6265106201171875, -33.541229248046875, 168.63421630859375, 14.043121337890625, -16.356857299804688, 175.68670654296875, 130.88311767578125, 126.88882446289062, 99.74116516113281, 20.064186096191406, 10.12054443359375, 130.72848510742188, 33.90155029296875, 52.0888786315918, -4.573028564453125, 126.8409423828125, 119.72691345214844, 172.69691467285156, 62.841064453125, 51.64453125, 124.8363037109375, -4.209930419921875, 140.50985717773438, 132.29107666015625, 138.40211486816406, -17.032867431640625, 2.0357627868652344, 26.748306274414062, -9.464157104492188, 14.545318603515625, 153.15203857421875, 175.05938720703125, 164.85836791992188, 18.342315673828125, 11.818561553955078, -114.80020141601562, 145.71755981445312, -43.1229248046875, 107.33148956298828, 124.60269165039062, 167.0987548828125, 139.66122436523438, -183.7168426513672, 194.51605224609375, 152.68130493164062, 0.4564476013183594, 123.53091430664062, -151.48013305664062, 10.510345458984375, 94.2064208984375, -19.090835571289062, 65.82803344726562, 36.124481201171875, 268.26611328125, 144.37744140625, 89.21591186523438, 62.69886779785156, -11.764617919921875, -55.16650390625, 191.093994140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000334.npy"}
{"epoch": 0.6994764397905759, "step": 335, "batch_size": 128, "mean": 45.605194091796875, "std": 87.5495834350586, "min": -170.72918701171875, "p10": -63.42250671386718, "median": 24.590184211730957, "p90": 154.84687194824218, "max": 205.07864379882812, "pos_frac": 0.6796875, "sample": [-93.28399658203125, -143.50082397460938, 23.537750244140625, -43.15867614746094, 114.19024658203125, -115.93978881835938, 179.330322265625, -11.0638427734375, 20.313751220703125, 136.63543701171875, 116.24270629882812, -143.50436401367188, 41.26513671875, -4.98121452331543, -2.6187515258789062, 21.5870361328125, -1.3280105590820312, -2.212289810180664, -27.61322021484375, -47.345001220703125, -16.15729522705078, 172.09527587890625, -2.85784912109375, 8.504470825195312, 14.93212890625, -1.7352294921875, 51.99261474609375, 50.839927673339844, -35.177398681640625, 132.9930419921875, -170.72918701171875, 187.89773559570312, -22.422828674316406, 122.0447998046875, 47.56884765625, 186.63211059570312, 6.930625915527344, 130.15585327148438, 151.16848754882812, 7.600009918212891, 154.59683227539062, 143.29737854003906, 155.4302978515625, 120.92289733886719, 162.93939208984375, 154.38131713867188, 80.94322204589844, 126.87911224365234, 140.96798706054688, 161.21588134765625, -6.919219970703125, 129.4180145263672, 16.937461853027344, 118.68057250976562, 130.119140625, 135.88134765625, 190.28167724609375, 104.16082763671875, 1.9339675903320312, 36.47882080078125, 23.615478515625, -138.99447631835938, -67.22187805175781, -19.460113525390625, -126.44155883789062, 146.15740966796875, 82.80552673339844, 146.578125, 103.80665588378906, 11.803176879882812, 35.5721435546875, 0.0733489990234375, -2.5678253173828125, 129.41357421875, 193.57559204101562, 133.384521484375, 5.114967346191406, -61.79420471191406, 11.5089111328125, 25.564889907836914, 79.33290100097656, 74.57408142089844, -12.684600830078125, 108.93099975585938, 161.4945831298828, -15.189033508300781, -68.2725830078125, 6.20660400390625, 69.03408813476562, 58.12123107910156, -38.997222900390625, -1.31732177734375, 127.70700073242188, 4.301197052001953, 20.607200622558594, -15.331329345703125, 12.9503173828125, 21.256431579589844, 81.10482788085938, -5.0280914306640625, 176.17752075195312, -131.88259887695312, 14.18560791015625, 83.4134750366211, 140.24462890625, 48.28822326660156, 0.0, 153.41159057617188, 157.489501953125, 129.9102783203125, -100.982421875, 138.28907775878906, 138.7359161376953, -119.5657958984375, -130.4533233642578, 205.07864379882812, 7.935081481933594, -2.7632522583007812, 8.22126579284668, -12.158050537109375, 12.439605712890625, 107.9052734375, 45.128936767578125, 103.98765563964844, -53.43836212158203, -14.084991455078125, 81.970703125, 51.309844970703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000335.npy"}
{"epoch": 0.7015706806282722, "step": 336, "batch_size": 128, "mean": 43.372886657714844, "std": 76.79298400878906, "min": -148.55239868164062, "p10": -33.47864685058593, "median": 22.68341064453125, "p90": 150.92288818359376, "max": 244.2222900390625, "pos_frac": 0.6953125, "sample": [59.550384521484375, 19.2100830078125, 60.72613525390625, -12.05133056640625, 137.0765380859375, 29.68914794921875, -20.395965576171875, 124.00395202636719, 124.72747802734375, 147.66387939453125, -13.4283447265625, 23.63275146484375, 39.843101501464844, 80.73224639892578, 3.4360008239746094, 70.26864624023438, -1.87274169921875, 176.04942321777344, 7.903076171875, 151.57943725585938, 178.17665100097656, 6.7587738037109375, 108.33003234863281, 131.12548828125, -52.7880859375, -16.119735717773438, 8.597183227539062, 17.57868194580078, 14.984115600585938, 32.836578369140625, 143.80770874023438, 13.10403823852539, -3.14178466796875, 137.53311157226562, 136.0479736328125, 155.88194274902344, 29.3424072265625, 203.20529174804688, 17.63616180419922, 7.72882080078125, -91.40928649902344, 129.41688537597656, -42.289398193359375, 142.54452514648438, 64.47821044921875, -0.8102493286132812, -0.6036872863769531, 138.63412475585938, -36.22991943359375, 21.32272720336914, 93.6282730102539, 0.58062744140625, 19.147705078125, 112.74822998046875, 37.21443176269531, -51.277252197265625, -13.989501953125, -13.735809326171875, 59.19297790527344, -14.105476379394531, 92.05184936523438, 32.29315185546875, -93.28826904296875, 76.87188720703125, -13.50640869140625, 100.96688842773438, 46.314544677734375, -41.100067138671875, 104.217041015625, 133.9478759765625, 4.689788818359375, -3.92401123046875, 141.73451232910156, 111.96257019042969, -4.923664093017578, 16.24444580078125, -5.709686279296875, 13.468582153320312, 28.659530639648438, -62.29689025878906, 3.082956314086914, 39.80108642578125, 45.665008544921875, -124.541748046875, 126.68223571777344, 36.01832580566406, 114.11679077148438, 43.036376953125, 20.40245819091797, 166.17926025390625, -20.391571044921875, -115.63980865478516, 164.39198303222656, 162.36349487304688, 67.323974609375, 131.61376953125, 173.9483642578125, 33.83472442626953, 23.740020751953125, 99.96267700195312, -5.228292465209961, 158.07176208496094, 12.603515625, -18.792388916015625, -137.74453735351562, 244.2222900390625, 16.885345458984375, 130.5134735107422, 186.78680419921875, -148.55239868164062, 5.541637420654297, -11.548240661621094, -0.176055908203125, 150.64151000976562, 25.039154052734375, 9.509208679199219, 15.739990234375, -18.66693115234375, -90.1590576171875, 0.0, -19.3406982421875, 25.4937744140625, 7.4560394287109375, 157.71453857421875, -32.299530029296875, -22.773712158203125, -14.633132934570312, 21.73406982421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000336.npy"}
{"epoch": 0.7036649214659686, "step": 337, "batch_size": 128, "mean": 50.98927307128906, "std": 80.71178436279297, "min": -130.0100860595703, "p10": -58.05637817382811, "median": 43.97196960449219, "p90": 153.8021499633789, "max": 193.49169921875, "pos_frac": 0.71875, "sample": [37.883758544921875, -25.300933837890625, 13.334014892578125, 1.1575660705566406, 14.426185607910156, -4.525909423828125, -16.21912384033203, 177.02813720703125, 57.468170166015625, 193.49169921875, -0.9020843505859375, 101.04252624511719, 21.92822265625, 51.1636962890625, 174.25323486328125, 159.94485473632812, -54.056549072265625, 22.138397216796875, -2.6441650390625, 16.807022094726562, -108.54660034179688, -97.99878692626953, -6.4530029296875, -44.6968994140625, 36.258033752441406, 3.2335052490234375, 6.28497314453125, 153.65159606933594, 0.2520484924316406, 143.98834228515625, 54.457733154296875, 167.12069702148438, 79.25253295898438, 131.16168212890625, 152.72671508789062, 128.2555389404297, 28.278564453125, 10.374565124511719, 100.87872314453125, 110.61727905273438, -10.9984130859375, -25.486000061035156, -5.2518310546875, 142.95184326171875, 57.408477783203125, 143.19052124023438, -10.592910766601562, -115.98115539550781, -16.582275390625, 31.5377197265625, 9.06524658203125, 131.93441772460938, -14.241989135742188, 124.41763305664062, 84.4649658203125, -72.1146240234375, 99.06837463378906, 104.79306030273438, 93.58316040039062, 21.06793212890625, 0.5466156005859375, -7.226226806640625, 64.18592834472656, 169.64239501953125, 150.55218505859375, -16.38494873046875, 135.63702392578125, 129.5424041748047, -130.0100860595703, 71.87408447265625, 128.68115234375, 166.6263427734375, 38.310943603515625, 28.391326904296875, 121.85389709472656, 3.589630126953125, -5.385345458984375, -25.420326232910156, 160.52906799316406, 121.32845306396484, 13.627761840820312, 87.46136474609375, 104.67121124267578, 152.45819091796875, 127.99638366699219, 116.13775634765625, -16.876354217529297, 113.98614501953125, -80.00848388671875, -90.01634979248047, 26.751785278320312, 130.2826385498047, 68.29721069335938, 184.47894287109375, -81.04586791992188, 121.155517578125, 7.721588134765625, -7.565277099609375, 114.70226287841797, 84.9732666015625, 88.89595794677734, -14.8370361328125, -109.20701599121094, 163.41281127929688, -40.003173828125, -90.09564971923828, 176.57037353515625, 76.59835815429688, 2.9462928771972656, 2.665315628051758, -24.02068328857422, 62.03764343261719, 147.53399658203125, 103.88449096679688, -67.38931274414062, 159.45697021484375, 3.1571044921875, 3.0185165405273438, -91.068115234375, 131.17959594726562, 133.31277465820312, 111.60456848144531, -104.88473510742188, 49.63299560546875, 16.16265869140625, 154.1534423828125, 147.27024841308594, 88.8338623046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000337.npy"}
{"epoch": 0.7057591623036649, "step": 338, "batch_size": 128, "mean": 58.84113311767578, "std": 79.3184585571289, "min": -159.70823669433594, "p10": -19.86340751647949, "median": 38.90697479248047, "p90": 167.80442810058594, "max": 214.69793701171875, "pos_frac": 0.765625, "sample": [167.51058959960938, -1.4574775695800781, 4.84783935546875, 125.84590148925781, 155.89007568359375, 172.24771118164062, 122.88215637207031, 123.37826538085938, 3.6420974731445312, -37.13218688964844, 105.024658203125, 140.07691955566406, 38.14544677734375, 18.31134605407715, 151.63241577148438, 25.325347900390625, 138.34378051757812, 34.203948974609375, -8.437026977539062, -19.296722412109375, 161.43765258789062, 41.678009033203125, -1.3156013488769531, 178.36663818359375, 5.165863037109375, 72.33151245117188, 171.77825927734375, 114.8338623046875, -10.563873291015625, 1.751810073852539, 41.53814697265625, 11.941650390625, -3.281890869140625, 16.301116943359375, 46.28215026855469, 122.5268783569336, 68.29437255859375, 156.69268798828125, 24.891952514648438, 37.38484191894531, 83.534912109375, 33.217071533203125, 56.63883972167969, 151.66192626953125, 168.49005126953125, 15.632064819335938, -29.395050048828125, 144.45559692382812, 102.83709716796875, 40.569427490234375, 76.47041320800781, -18.14410400390625, 161.63265991210938, -122.06693267822266, 29.534088134765625, -2.8310394287109375, -9.447128295898438, 171.43991088867188, -82.49258422851562, 161.4898681640625, 133.76014709472656, 214.69793701171875, 140.93746948242188, 44.981964111328125, 74.39837646484375, 47.40171813964844, -21.185672760009766, -11.963211059570312, -135.95465087890625, -4.81781005859375, 7.746845245361328, 128.5753936767578, 177.30718994140625, 22.466339111328125, -27.734222412109375, 2.6131763458251953, 16.91065216064453, 155.16075134277344, 153.26661682128906, -0.8959083557128906, 118.74456787109375, 1.0448455810546875, 130.16578674316406, -4.526985168457031, 124.81521606445312, 23.74658203125, -11.513725280761719, 10.229522705078125, 106.03726196289062, 137.45431518554688, 119.96932983398438, 177.31378173828125, 18.099502563476562, -5.189056396484375, -120.09336853027344, 155.96206665039062, 22.035614013671875, 190.3641357421875, -78.71083068847656, 30.963951110839844, 21.298797607421875, -37.563995361328125, 16.225605010986328, 41.835968017578125, 39.66850280761719, 164.56686401367188, 118.86555480957031, 21.742233276367188, -30.666183471679688, 184.2794189453125, 44.936676025390625, 0.249786376953125, 0.48892974853515625, 33.3599853515625, -6.52734375, 168.506591796875, 41.25775146484375, 102.82853698730469, 121.836181640625, 25.391357421875, -159.70823669433594, 1.340667724609375, 29.3538818359375, -14.062980651855469, 186.40576171875, 183.44705200195312, -34.944427490234375, 150.41883850097656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000338.npy"}
{"epoch": 0.7078534031413612, "step": 339, "batch_size": 128, "mean": 59.64130401611328, "std": 83.59053802490234, "min": -179.2012939453125, "p10": -32.455203247070315, "median": 48.164398193359375, "p90": 151.9620849609375, "max": 219.60446166992188, "pos_frac": 0.7578125, "sample": [109.20895385742188, -46.55792236328125, 68.57987213134766, 127.8936767578125, 64.05127716064453, -9.398056030273438, 32.234405517578125, -85.87396240234375, 122.06491088867188, -4.66033935546875, 182.04605102539062, 22.20758056640625, 7.30584716796875, 5.416938781738281, 144.00662231445312, 0.778533935546875, 108.47024536132812, 106.57736206054688, 96.80237579345703, -58.321868896484375, 121.37954711914062, 14.492340087890625, 63.552978515625, 118.35733032226562, 0.0, 175.8653564453125, -32.2601318359375, 61.069183349609375, 0.0, 36.1438102722168, 58.311363220214844, -32.910369873046875, 184.08135986328125, 136.75462341308594, 7.886810302734375, -145.16578674316406, 153.4324951171875, 161.17913818359375, 35.55558776855469, -22.446090698242188, 111.40560913085938, 142.97308349609375, -31.702392578125, 136.6190185546875, 82.4652099609375, 89.45669555664062, 128.1336669921875, 0.21212005615234375, -5.4604949951171875, 147.06277465820312, 44.496978759765625, 8.63848876953125, 215.46987915039062, 112.36399841308594, 146.07366943359375, 137.41641235351562, 14.565574645996094, 26.69361114501953, 145.2940673828125, 168.35861206054688, 143.0045166015625, -1.16876220703125, -7.42242431640625, -52.099456787109375, -1.1980018615722656, 171.62811279296875, 7.942047119140625, 119.88311767578125, 124.1346435546875, 32.4820556640625, 41.320587158203125, 15.652069091796875, -111.43418884277344, 118.9627685546875, 134.57229614257812, 17.37091064453125, 109.3170166015625, 146.73828125, -49.094207763671875, 128.12965393066406, 3.7551040649414062, 142.43429565429688, 156.21435546875, -12.116241455078125, 149.89389038085938, 219.60446166992188, 56.89947509765625, 0.0, 47.385223388671875, 48.943572998046875, 128.62701416015625, 139.54257202148438, 139.24838256835938, 151.3319091796875, 148.58425903320312, -10.785400390625, 30.349197387695312, -8.3297119140625, 27.92034912109375, 77.88943481445312, 135.6524658203125, -80.18780517578125, 27.970458984375, 7.3944091796875, 11.004913330078125, 136.37643432617188, 132.05665588378906, 15.0048828125, -20.452011108398438, 6.336463928222656, 170.25619506835938, 144.5614013671875, -179.2012939453125, 146.9021453857422, 25.0750732421875, 10.702644348144531, 10.774131774902344, -5.17510986328125, 158.39788818359375, -108.343994140625, -163.1144256591797, 185.93426513671875, 44.12841796875, 24.867706298828125, 147.42965698242188, -5.1840362548828125, -124.01555633544922, 144.173095703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000339.npy"}
{"epoch": 0.7099476439790576, "step": 340, "batch_size": 128, "mean": 50.84252166748047, "std": 82.04447937011719, "min": -141.97317504882812, "p10": -40.46957702636718, "median": 35.485877990722656, "p90": 151.7249267578125, "max": 213.04632568359375, "pos_frac": 0.7265625, "sample": [197.71617126464844, 94.28026580810547, -33.93043518066406, 139.9688720703125, 2.2119827270507812, 149.0861358642578, -41.144134521484375, -19.18377685546875, -54.5, 148.22740173339844, -126.7796630859375, 131.76988220214844, -72.04547119140625, 1.4020767211914062, 0.0, 156.21270751953125, 19.507659912109375, -47.42066955566406, 115.11672973632812, 152.85714721679688, 143.94320678710938, 16.146209716796875, -32.55070495605469, 88.160400390625, -40.18048095703125, -3.250579833984375, -26.73931884765625, 134.628173828125, 10.959121704101562, 151.89300537109375, 189.98406982421875, 4.376440048217773, 64.9246826171875, 151.65289306640625, 27.20526123046875, -12.493865966796875, -0.11278343200683594, -113.45391845703125, 39.75117492675781, 15.558059692382812, 146.27804565429688, 132.17755126953125, -106.95947265625, 44.110443115234375, 66.5049057006836, 153.18516540527344, -12.311172485351562, 50.63433837890625, 8.715576171875, -17.401473999023438, 110.74411010742188, 182.37844848632812, -2.225208282470703, -92.496337890625, 8.309225082397461, -11.245269775390625, 136.67758178710938, 122.5693359375, 38.985137939453125, 43.33949279785156, 8.347015380859375, -141.97317504882812, 160.30613708496094, 0.0, 189.84698486328125, 12.412628173828125, 49.9259033203125, 17.82806396484375, 146.04583740234375, 35.582763671875, 15.526763916015625, -8.79718017578125, 132.2965087890625, -101.35212707519531, 163.40113830566406, 31.770263671875, 77.6558837890625, -28.17108154296875, 151.00811767578125, 10.148101806640625, 149.81365966796875, 213.04632568359375, -80.51904296875, 35.6439208984375, -12.79150390625, 127.83712768554688, 109.43557739257812, 27.3388671875, 135.84039306640625, 24.349945068359375, 57.6685791015625, -14.324838638305664, 105.55447387695312, -14.03216552734375, 135.40716552734375, 141.9840087890625, -87.01715087890625, 3.901611328125, 50.253631591796875, 53.63134765625, 212.41900634765625, 9.41253662109375, 142.24700927734375, 4.9366607666015625, 145.83990478515625, -33.208526611328125, 13.368797302246094, -133.46945190429688, 138.68710327148438, 119.94253540039062, 133.00018310546875, 37.99021911621094, 13.075611114501953, 108.60426330566406, 155.32156372070312, 150.3933563232422, 27.875823974609375, 16.811737060546875, 13.033447265625, 127.27835083007812, 35.38899230957031, 66.72360229492188, -20.90283203125, 20.506046295166016, 51.76991271972656, 17.693695068359375, 132.60781860351562, -10.059444427490234], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000340.npy"}
{"epoch": 0.7120418848167539, "step": 341, "batch_size": 128, "mean": 40.0311279296875, "std": 85.71984100341797, "min": -150.9200439453125, "p10": -66.82073974609374, "median": 17.732315063476562, "p90": 147.40600433349607, "max": 212.8740234375, "pos_frac": 0.6796875, "sample": [6.803504943847656, -1.748748779296875, 106.24508666992188, 145.4454803466797, 117.4945068359375, -38.016090393066406, 121.78204345703125, 16.92938232421875, 21.656631469726562, 4.5384063720703125, 134.70294189453125, 39.292572021484375, 141.16714477539062, 10.828937530517578, 187.4024658203125, 178.12655639648438, -41.068206787109375, 75.870849609375, 10.0692138671875, 15.289474487304688, 140.46591186523438, 14.809600830078125, 129.29791259765625, -79.64930725097656, 180.65545654296875, 53.828399658203125, 10.743328094482422, 7.76336669921875, 8.386337280273438, -1.36785888671875, 0.115478515625, 37.0909423828125, -14.763595581054688, -5.987548828125, -121.97021484375, 10.023284912109375, -63.21044921875, -118.61138916015625, -0.4399871826171875, 8.06243896484375, 2.80487060546875, 44.434452056884766, 151.98056030273438, 109.16461181640625, 81.31071472167969, 5.5941162109375, 144.43336486816406, 95.69656372070312, -2.3814945220947266, 116.25416564941406, 104.5264892578125, 144.59788513183594, 182.427490234375, -24.0286865234375, 115.16793823242188, 114.20944213867188, 111.15072631835938, 98.73709106445312, 11.799591064453125, -26.0211181640625, 28.38311767578125, -115.27642822265625, 212.8740234375, -10.103559494018555, 92.06979370117188, -122.94880676269531, -100.60211181640625, 145.30325317382812, 160.74598693847656, 2.94921875, 76.0743408203125, 142.46063232421875, 107.30831909179688, -20.880844116210938, 128.43948364257812, 142.68292236328125, 140.08285522460938, 12.571113586425781, 20.167434692382812, -145.8960723876953, 63.46553039550781, 2.785980224609375, -12.280105590820312, -59.1429443359375, 193.54833984375, 29.699447631835938, -18.672637939453125, 39.09967041015625, 62.729339599609375, 153.20071411132812, -111.30941772460938, 82.8646240234375, 176.78524780273438, 135.97496032714844, -126.01377868652344, -13.7349853515625, -39.282066345214844, -1.5091056823730469, -11.032196044921875, 28.736083984375, -16.790771484375, 142.4796142578125, 16.7274169921875, -15.16217041015625, -124.9730224609375, 38.270843505859375, 35.17828369140625, -18.713241577148438, -150.9200439453125, -29.07025146484375, 18.535247802734375, -59.676971435546875, -3.35614013671875, 155.62094116210938, -140.8389434814453, -1.6667022705078125, 130.34567260742188, 152.97756958007812, 132.1409912109375, 64.92774963378906, 133.7645263671875, 7.5857086181640625, 30.71770477294922, -75.2447509765625, 185.509033203125, 13.19622802734375, -12.599151611328125, 12.788185119628906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000341.npy"}
{"epoch": 0.7141361256544503, "step": 342, "batch_size": 128, "mean": 50.36888885498047, "std": 76.34138488769531, "min": -168.6802978515625, "p10": -23.233412170410155, "median": 23.873451232910156, "p90": 152.5736831665039, "max": 212.81353759765625, "pos_frac": 0.734375, "sample": [-6.609405517578125, 23.37237548828125, 16.483123779296875, 21.0885009765625, 93.14837646484375, 203.0579833984375, 18.711090087890625, 95.74990844726562, 6.144401550292969, 38.158660888671875, -23.706512451171875, 5.59722900390625, -1.68890380859375, 131.21063232421875, 106.59684753417969, 131.504638671875, -7.344184875488281, 64.978759765625, -64.77136993408203, 3.6944427490234375, 46.67521667480469, 7.344085693359375, 10.23809814453125, 29.12969207763672, -49.748992919921875, 9.399269104003906, 3.20513916015625, -42.637451171875, 39.210052490234375, 129.04421997070312, 12.725967407226562, 140.69580078125, 108.44364929199219, 63.6656494140625, -8.182449340820312, 14.261444091796875, 18.064483642578125, -12.724042892456055, -23.030654907226562, 32.205108642578125, 48.57099151611328, 170.9835205078125, 21.10797119140625, -35.6754150390625, -18.822372436523438, -22.968040466308594, -47.716583251953125, 56.93927001953125, 53.6004638671875, 10.57391357421875, 122.74612426757812, -3.987945556640625, -30.54644775390625, 156.40997314453125, 8.930618286132812, 181.23501586914062, 149.159423828125, -2.5594482421875, 58.04205322265625, 102.97393798828125, 151.843994140625, 158.95562744140625, 115.49871826171875, 126.45671844482422, 124.19407653808594, -121.13493347167969, 5.3513336181640625, 0.0, -14.75006103515625, 30.18731689453125, 171.04052734375, 10.434829711914062, 166.81903076171875, 3.624669075012207, 26.989971160888672, 128.4808349609375, 148.59161376953125, 41.92432403564453, 176.24900817871094, 3.546205520629883, -15.4844970703125, 139.03402709960938, 35.891693115234375, 111.1845703125, -65.32675170898438, -11.571319580078125, -168.6802978515625, -15.896759033203125, 118.40374755859375, 107.93424224853516, 150.25875854492188, 114.34933471679688, -1.5281524658203125, 9.080078125, 11.699203491210938, 88.25352478027344, 114.7728271484375, 19.654678344726562, 20.608428955078125, -104.57229614257812, 19.743255615234375, 132.7071533203125, -12.429214477539062, 212.81353759765625, 14.187835693359375, 154.2762908935547, -78.01141357421875, 12.657257080078125, 85.66668701171875, -99.80081939697266, 157.52655029296875, 127.6075439453125, 54.059783935546875, 147.882080078125, 136.35635375976562, 151.48870849609375, 24.374526977539062, 14.495086669921875, 124.72462463378906, 141.57000732421875, -14.787055969238281, -12.184051513671875, -19.194931030273438, 187.7393798828125, 122.37757873535156, -7.576629638671875, 180.1751708984375, 14.04541015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000342.npy"}
{"epoch": 0.7162303664921466, "step": 343, "batch_size": 128, "mean": 43.637855529785156, "std": 82.65492248535156, "min": -196.96600341796875, "p10": -40.329978942871094, "median": 24.49264907836914, "p90": 152.01590576171876, "max": 207.17767333984375, "pos_frac": 0.6953125, "sample": [-30.686920166015625, 127.57986450195312, 140.16004943847656, 4.563875198364258, 41.37579345703125, 35.37689208984375, 42.047088623046875, 22.3543701171875, -80.03489685058594, 23.35009765625, 147.54400634765625, -40.832366943359375, -11.987060546875, -26.76123046875, 104.95132446289062, 171.70175170898438, 158.51165771484375, 42.839630126953125, -10.768455505371094, -6.885866165161133, 195.4505615234375, 22.6461181640625, 154.7769775390625, 147.648681640625, 100.35053253173828, 52.81312561035156, -0.5185317993164062, 84.77349853515625, 9.731201171875, 130.41818237304688, 43.382171630859375, -135.76800537109375, 58.97064208984375, -9.595077514648438, -116.45085144042969, 13.074317932128906, 13.07861328125, 140.56802368164062, -25.85479736328125, 8.87115478515625, -40.11466979980469, 175.90069580078125, 39.911590576171875, 105.4722900390625, 0.0, -113.37799072265625, 19.32281494140625, 140.2288360595703, 40.43976593017578, 200.5357666015625, -127.17471313476562, -0.47171783447265625, 28.361692428588867, 11.748542785644531, 147.24163818359375, -35.425201416015625, 119.02900695800781, 67.99444580078125, -0.11170196533203125, 113.03263854980469, 127.205322265625, 8.571914672851562, 4.735645294189453, 3.3344573974609375, 135.5525665283203, 142.73318481445312, -23.726654052734375, 146.13809204101562, 20.21527099609375, 85.74232482910156, 52.42811584472656, -20.550750732421875, 52.206817626953125, -16.131393432617188, -0.2153778076171875, 46.41149139404297, 9.7999267578125, 207.17767333984375, 123.12550354003906, 150.73011779785156, 33.89531707763672, 49.147064208984375, -87.51065063476562, 6.37188720703125, 125.04762268066406, -8.467010498046875, 4.3719482421875, 33.85002136230469, 12.946868896484375, -28.1480712890625, 4.7393341064453125, 141.804443359375, 144.40658569335938, 198.8187255859375, 10.0634765625, -14.720054626464844, 124.97926330566406, 66.3841552734375, 138.44732666015625, 25.63520050048828, 160.38951110839844, 155.83592224121094, -2.849029541015625, 124.3905029296875, -196.96600341796875, 151.35211181640625, -18.363067626953125, -4.031909942626953, -91.67022705078125, -66.91595458984375, 11.304798126220703, 129.76596069335938, 153.56475830078125, 1.5091552734375, -97.24252319335938, -5.44775390625, -7.6731109619140625, -48.988006591796875, 27.282684326171875, 183.22219848632812, 91.68983459472656, 1.0511817932128906, 28.857666015625, 9.154335021972656, 8.14874267578125, 163.98587036132812, -13.31610107421875, -135.25387573242188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000343.npy"}
{"epoch": 0.7183246073298429, "step": 344, "batch_size": 128, "mean": 53.541358947753906, "std": 80.36581420898438, "min": -160.174072265625, "p10": -32.46097106933593, "median": 31.696575164794922, "p90": 163.52763061523436, "max": 202.93673706054688, "pos_frac": 0.7265625, "sample": [13.868560791015625, -3.453155517578125, 125.06440734863281, 100.92355346679688, -20.3651123046875, 153.08758544921875, 142.3816375732422, 19.475860595703125, 138.78912353515625, -32.166748046875, 121.18789672851562, 55.44996643066406, 121.94508361816406, -15.008060455322266, 54.85626220703125, 174.81890869140625, 16.65460205078125, 31.730453491210938, -150.88693237304688, 83.40988159179688, -104.109375, 18.137191772460938, -0.43505859375, -34.433837890625, 183.3358154296875, 99.51690673828125, -14.880525588989258, 170.83798217773438, 161.53175354003906, 9.879150390625, 169.39373779296875, -38.75469970703125, -5.30950927734375, 135.08993530273438, -160.174072265625, 30.676666259765625, 141.07843017578125, 132.17347717285156, 13.546295166015625, -11.789703369140625, -146.8148956298828, 19.73895263671875, 20.406494140625, -15.229293823242188, -6.091499328613281, 19.501876831054688, -59.2557373046875, 134.9125213623047, 175.70382690429688, 38.576568603515625, -33.147491455078125, -52.417938232421875, -0.010711669921875, -49.022621154785156, 112.25869750976562, 108.29219055175781, 173.2094268798828, 20.855621337890625, 185.24996948242188, 167.1395263671875, 146.1441192626953, 143.58203125, 202.93673706054688, 112.847412109375, 151.07200622558594, -3.19476318359375, 19.722137451171875, 180.0384521484375, -12.763412475585938, -6.872772216796875, -53.981353759765625, 15.395172119140625, 24.561614990234375, 195.02239990234375, 53.353271484375, 39.431640625, 19.78399658203125, 96.849609375, 25.3114013671875, 91.90931701660156, 31.662696838378906, 22.502471923828125, 60.131011962890625, 137.27232360839844, -6.1448822021484375, 83.96246337890625, 1.1566619873046875, 145.316162109375, 21.827438354492188, 42.640838623046875, 182.3197021484375, 121.29122161865234, 23.15643310546875, -7.6891021728515625, -6.472450256347656, 159.70376586914062, 53.4990234375, -1.6368904113769531, 36.09428405761719, -17.399063110351562, -7.5192108154296875, 34.154815673828125, 15.846221923828125, 130.3409423828125, -13.67388916015625, -80.29058837890625, 146.95603942871094, 37.29327392578125, 20.45728302001953, 5.771392822265625, 9.18402099609375, 113.47633361816406, 11.657241821289062, 162.13577270507812, 100.96493530273438, 30.822647094726562, 49.50775146484375, 82.67706298828125, 27.928192138671875, 166.77529907226562, 139.90484619140625, 31.352081298828125, 64.9324951171875, 150.69488525390625, 47.84649658203125, -18.473007202148438, 161.20315551757812, -133.9052734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000344.npy"}
{"epoch": 0.7204188481675393, "step": 345, "batch_size": 128, "mean": 43.388099670410156, "std": 72.5413818359375, "min": -171.5938720703125, "p10": -26.873913955688472, "median": 20.524900436401367, "p90": 150.5388214111328, "max": 185.9898681640625, "pos_frac": 0.7578125, "sample": [35.01617431640625, 127.30602264404297, 72.49981689453125, -126.82505798339844, 174.84649658203125, 53.72723388671875, -9.663970947265625, -39.4019775390625, 15.517608642578125, 0.21508407592773438, 126.75540161132812, 12.282966613769531, -0.5151596069335938, 3.9316482543945312, 78.3583984375, 1.08392333984375, 160.77078247070312, 14.339630126953125, 13.1478271484375, -4.531402587890625, 11.6983642578125, -25.15673828125, 124.1840591430664, 148.38233947753906, 3.1588668823242188, 40.897705078125, 132.83535766601562, 30.801116943359375, -98.0845947265625, 0.7605476379394531, -9.775794982910156, 27.504852294921875, -24.91622543334961, 26.039215087890625, 0.074066162109375, 55.958038330078125, 56.2542724609375, 4.543609619140625, 17.2767333984375, 28.42047119140625, 29.3282470703125, -7.739938735961914, -30.880657196044922, 131.93174743652344, 13.8172607421875, 116.22264099121094, 15.540435791015625, 1.3336143493652344, 44.24128723144531, 183.644775390625, 125.9710693359375, 133.750732421875, 48.57720947265625, 157.7168426513672, 165.04002380371094, 63.754730224609375, 21.235595703125, -52.181060791015625, 15.420272827148438, -7.615364074707031, 154.5650634765625, 28.127471923828125, 80.38496398925781, 19.255874633789062, 19.834747314453125, -74.85725402832031, -171.5938720703125, 14.660747528076172, -114.35308837890625, -22.398895263671875, 161.41339111328125, 169.0335693359375, 144.9344940185547, 22.549911499023438, 25.761276245117188, -15.980804443359375, 185.9898681640625, -53.800270080566406, 6.35223388671875, -3.42919921875, 150.01040649414062, 154.09292602539062, 79.69024658203125, -85.77523803710938, 74.45367431640625, 9.506332397460938, -37.6527099609375, 181.01141357421875, 120.94140625, 60.072105407714844, 163.91905212402344, 136.0282745361328, 21.99359130859375, -17.042694091796875, 90.63386535644531, 15.011703491210938, 69.38501739501953, 138.76321411132812, 144.35845947265625, -7.34613037109375, 143.427734375, -33.11090087890625, 128.29168701171875, -38.24220275878906, 4.404449462890625, 26.4239501953125, 3.7582626342773438, 21.950164794921875, -6.684295654296875, -10.093597412109375, -11.354446411132812, 119.91582489013672, 18.38909912109375, 21.014175415039062, 151.77178955078125, 130.061767578125, 13.263092041015625, 20.035625457763672, 0.0, 10.062095642089844, 2.7730941772460938, 17.484893798828125, 8.602119445800781, -0.008258819580078125, 140.60711669921875, 6.2862548828125, 100.49600219726562, 96.8193359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000345.npy"}
{"epoch": 0.7225130890052356, "step": 346, "batch_size": 128, "mean": 51.16846466064453, "std": 81.28536224365234, "min": -181.65899658203125, "p10": -41.33550186157226, "median": 33.18538284301758, "p90": 159.76622619628904, "max": 187.41513061523438, "pos_frac": 0.71875, "sample": [-47.299041748046875, -27.58583641052246, 0.9956989288330078, -50.90638732910156, 84.19955444335938, 45.7735595703125, 168.2752685546875, 170.45114135742188, 128.21371459960938, -3.887725830078125, 136.7639923095703, -20.872604370117188, -13.065887451171875, 24.651535034179688, 139.91693115234375, -133.92453002929688, -74.55325317382812, 4.46295166015625, 23.57501220703125, 177.66244506835938, 22.596237182617188, -8.965835571289062, 14.966705322265625, -43.872772216796875, 27.984210968017578, 130.9810791015625, 11.699386596679688, 185.8873291015625, -14.14215087890625, -0.494354248046875, 40.390960693359375, 107.18321990966797, -107.68436431884766, 27.011138916015625, -3.4997100830078125, 154.719970703125, 103.37602233886719, 166.2381591796875, 10.589143753051758, 17.3265380859375, 77.79962158203125, 114.35598754882812, -101.5054931640625, 22.44140625, -54.46781921386719, 73.56045532226562, -10.4920654296875, -17.131439208984375, -2.25714111328125, 73.62651062011719, 138.588623046875, 15.09185791015625, 126.71128845214844, -43.38672637939453, 19.836090087890625, 89.69612884521484, 179.56283569335938, 11.823410034179688, -4.244209289550781, 126.14126586914062, 164.88619995117188, 185.4034423828125, 28.78228759765625, 11.14877700805664, 29.09161376953125, 36.23351287841797, 168.512451171875, -16.188133239746094, 74.21957397460938, 85.09329223632812, 69.51565551757812, 38.87603759765625, 130.96629333496094, 179.77679443359375, 158.42462158203125, 157.69802856445312, -111.54299926757812, 162.89663696289062, 58.15684509277344, 33.14520263671875, 124.3643798828125, -170.80496215820312, 156.17868041992188, 4.2080078125, -7.15911865234375, -16.154369354248047, -9.607009887695312, 5.602775573730469, -27.821197509765625, 126.2403564453125, 122.7462158203125, 39.45556640625, -14.271759033203125, 150.75213623046875, -1.8841552734375, -13.870414733886719, 123.19818115234375, -181.65899658203125, 2.3187408447265625, 151.09292602539062, 1.5660400390625, 8.68780517578125, 123.596435546875, 119.47843933105469, 174.76527404785156, 57.286216735839844, 125.42378234863281, 114.04522705078125, 10.192672729492188, 9.884613037109375, 0.0, 57.64421081542969, -4.56365966796875, 41.45475769042969, 131.00595092773438, 117.85807800292969, 45.765716552734375, 133.43515014648438, 33.225563049316406, -77.47210693359375, 128.16412353515625, 28.443603515625, -40.45640563964844, 121.82928466796875, 149.65664672851562, 126.93086242675781, 5.389699935913086, 187.41513061523438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000346.npy"}
{"epoch": 0.724607329842932, "step": 347, "batch_size": 128, "mean": 44.56103515625, "std": 78.2769546508789, "min": -201.01144409179688, "p10": -27.98829956054687, "median": 25.426799774169922, "p90": 148.79680633544922, "max": 181.18084716796875, "pos_frac": 0.734375, "sample": [8.928237915039062, 76.19735717773438, 138.78042602539062, 2.582489013671875, 61.77130126953125, 105.07888793945312, 1.4460830688476562, 150.5433349609375, 15.041252136230469, 148.0482940673828, -201.01144409179688, 84.080078125, 146.77525329589844, 32.9727783203125, 163.92025756835938, -65.49327087402344, -39.24505615234375, -0.4667816162109375, 139.21884155273438, 164.67335510253906, 135.09884643554688, -30.01861572265625, 122.886474609375, 11.46316909790039, 16.469100952148438, 36.7955322265625, 144.30438232421875, 16.96575927734375, 10.701541900634766, -3.7453765869140625, 0.22039794921875, 65.09237670898438, 176.1918487548828, 34.98651123046875, 136.13421630859375, 21.27113151550293, 155.0184326171875, -9.246109008789062, 41.778778076171875, 24.872947692871094, 15.984001159667969, 146.64404296875, 176.73501586914062, 25.98065185546875, 26.787384033203125, 100.24516296386719, -15.527801513671875, 11.2423095703125, 18.075180053710938, -118.77532958984375, -3.306793212890625, -3.5074920654296875, 72.35430908203125, -75.94606018066406, 77.155029296875, 112.09278869628906, -0.5021209716796875, 143.05795288085938, 111.93377685546875, -98.87373352050781, 126.51632690429688, 14.629852294921875, -1.023345947265625, -16.361894607543945, 0.115966796875, 103.34956359863281, 60.48858642578125, 41.95037841796875, 178.56338500976562, 38.03944396972656, 31.253814697265625, -137.38772583007812, -18.388259887695312, 8.642562866210938, 102.37557983398438, 0.625946044921875, -46.88203430175781, -14.465068817138672, -44.17582702636719, 3.717864990234375, 6.674873352050781, 108.32891845703125, 159.23495483398438, -14.43084716796875, 11.378772735595703, -146.51846313476562, 116.77930450439453, 181.18084716796875, 22.3565673828125, 81.57379150390625, -11.829010009765625, 59.4564208984375, -55.8912353515625, 143.03799438476562, 5.008552551269531, -12.576126098632812, 43.10417175292969, -2.391376495361328, 122.47695922851562, -7.199981689453125, -170.50582885742188, -18.786865234375, 175.90621948242188, 146.53860473632812, 0.0, 24.468658447265625, 20.740814208984375, 34.37384033203125, 133.44174194335938, 160.18408203125, 30.618988037109375, 1.117889404296875, 144.50485229492188, 128.23330688476562, -11.837158203125, 1.7776947021484375, 40.2525634765625, -4.7848663330078125, 133.1030731201172, 33.599273681640625, 23.3597412109375, 161.86444091796875, 173.9406280517578, 82.31875610351562, 11.674293518066406, -27.1181640625, 67.32037353515625, 3.2338027954101562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000347.npy"}
{"epoch": 0.7267015706806282, "step": 348, "batch_size": 128, "mean": 42.57643127441406, "std": 75.40428924560547, "min": -191.06875610351562, "p10": -23.932785034179684, "median": 13.262725830078125, "p90": 144.9831771850586, "max": 194.11300659179688, "pos_frac": 0.6875, "sample": [20.337493896484375, 9.933197021484375, 3.506967544555664, 2.52044677734375, -6.595794677734375, 11.892669677734375, 138.48095703125, -7.792629241943359, 108.96255493164062, 126.12793731689453, -12.552947998046875, 144.73431396484375, 56.12939453125, 132.49853515625, 23.50042724609375, 12.3868408203125, 181.28109741210938, 119.76254272460938, -85.58528900146484, 0.0, 15.793212890625, 49.789154052734375, 110.22669219970703, -1.4883270263671875, -55.44598388671875, 175.5257110595703, -5.1670989990234375, 3.640899658203125, -10.206939697265625, 107.69143676757812, 112.57307434082031, 110.80572509765625, -9.686843872070312, -2.0581207275390625, 82.49755859375, 3.08709716796875, 147.48123168945312, -4.908477783203125, 126.89339447021484, 156.821533203125, 6.050565719604492, -18.846435546875, -75.65509033203125, 28.406280517578125, -2.3037872314453125, 142.85342407226562, 35.778167724609375, 14.13861083984375, 18.721221923828125, 102.13231658935547, 70.89354705810547, 115.03204345703125, 6.0633544921875, -0.07925796508789062, -11.5313720703125, 67.20500183105469, 154.07192993164062, 182.09994506835938, 156.58837890625, -100.79568481445312, -7.39898681640625, -18.15753173828125, -18.906005859375, 115.77154541015625, 1.0820846557617188, 102.89524841308594, 10.323974609375, -2.5047035217285156, 6.722900390625, 41.9534912109375, 142.635498046875, -121.94149780273438, -138.10305786132812, -4.6688079833984375, -95.74114990234375, 44.280181884765625, -17.652679443359375, 7.7164764404296875, 98.32562255859375, 6.89727783203125, 184.3038330078125, 7.3892822265625, 63.212799072265625, 191.37413024902344, 2.966796875, 166.99765014648438, 30.520614624023438, -25.173187255859375, -31.7451171875, 18.6773681640625, -8.788793563842773, 145.56385803222656, 132.2607421875, -7.099601745605469, 194.11300659179688, -23.40118408203125, 8.306785583496094, 6.1428375244140625, -28.459823608398438, 7.8706817626953125, 140.8114013671875, 10.949432373046875, 129.98097229003906, -11.179454803466797, 72.2160415649414, 0.49560546875, 142.20751953125, 46.260467529296875, -12.25201416015625, 143.53402709960938, 119.90864562988281, -17.076690673828125, 81.53152465820312, -38.821990966796875, 8.018356323242188, 54.91606140136719, 160.85813903808594, 35.41387939453125, 140.94281005859375, 32.18023681640625, 127.3760986328125, 2.7627639770507812, -4.844539642333984, 12.272750854492188, -25.364013671875, 34.86177062988281, -191.06875610351562, 80.1126708984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000348.npy"}
{"epoch": 0.7287958115183246, "step": 349, "batch_size": 128, "mean": 56.66209030151367, "std": 83.27786254882812, "min": -191.25961303710938, "p10": -23.887457275390624, "median": 33.419464111328125, "p90": 171.05358276367187, "max": 231.530517578125, "pos_frac": 0.7578125, "sample": [127.01646423339844, 158.4593048095703, 1.67327880859375, -191.25961303710938, -89.85025024414062, -0.3168792724609375, 231.530517578125, 126.98388671875, -111.30056762695312, 54.78289794921875, 44.538116455078125, -59.536895751953125, 128.6807403564453, 173.28564453125, 100.78237915039062, 137.56838989257812, 185.40631103515625, 12.856857299804688, 109.57760620117188, 176.56494140625, 73.88873291015625, 3.53277587890625, -24.30816650390625, -23.7071533203125, -3.3344573974609375, 6.400970458984375, -37.59002685546875, -0.1673603057861328, 176.22247314453125, 170.09698486328125, 69.81315612792969, 130.5723876953125, 141.07620239257812, -46.11474609375, 32.454620361328125, 43.72137451171875, 186.2250213623047, -4.8905029296875, 18.408218383789062, 48.503173828125, 7.201225280761719, 0.9773540496826172, 145.55145263671875, -0.522247314453125, 11.432723999023438, 47.1295166015625, -9.9970703125, 204.48910522460938, 26.603912353515625, 17.772293090820312, -46.179168701171875, -0.7284488677978516, -2.146209716796875, 159.90097045898438, 153.61557006835938, 21.074905395507812, 7.2983856201171875, 6.0403900146484375, -109.07809448242188, 83.72929382324219, 0.0, 200.34231567382812, 4.28564453125, 35.05865478515625, 134.28866577148438, 28.05352783203125, 149.27072143554688, 159.80712890625, 149.4305419921875, 8.741998672485352, 157.8984832763672, -23.19775390625, -106.96499633789062, 208.50018310546875, 2.880889892578125, 48.6295166015625, -6.88128662109375, 118.28227233886719, 5.1074981689453125, 5.385528564453125, 19.427337646484375, 115.17694854736328, 193.27789306640625, 95.9403076171875, 173.3011474609375, 7.93865966796875, 65.12413024902344, 13.335376739501953, 149.56597900390625, 47.68902587890625, 56.415283203125, -2.051513671875, 147.13165283203125, 9.53891372680664, -21.534805297851562, 6.253326416015625, 145.2377471923828, 71.36253356933594, 27.03448486328125, -122.79708862304688, 37.95880126953125, 30.458839416503906, 63.61640930175781, -0.8162841796875, -96.95269775390625, 24.81268310546875, -4.730072021484375, 6.1470794677734375, 13.987937927246094, 142.9520263671875, 34.384307861328125, 142.61337280273438, 29.2686767578125, 128.3751678466797, 152.16712951660156, 175.16087341308594, 146.68475341796875, -21.389404296875, 109.98883056640625, 187.82177734375, 9.844329833984375, 98.24296569824219, 137.66213989257812, 128.32876586914062, -48.248260498046875, -2.4211273193359375, 13.444549560546875, 134.6844940185547], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000349.npy"}
{"epoch": 0.7308900523560209, "step": 350, "batch_size": 128, "mean": 49.50843048095703, "std": 86.11686706542969, "min": -154.97021484375, "p10": -52.38122444152831, "median": 34.177764892578125, "p90": 162.2425994873047, "max": 208.15911865234375, "pos_frac": 0.7890625, "sample": [163.0205078125, 56.466575622558594, 0.6056556701660156, -4.9325408935546875, 10.572505950927734, 13.17816162109375, 14.867721557617188, 179.74298095703125, 1.7385787963867188, 127.78794860839844, 29.6077880859375, 15.247848510742188, 157.78240966796875, -11.068115234375, 27.376861572265625, 159.66018676757812, 143.9230499267578, 161.90921020507812, 22.97856903076172, -56.97048568725586, 147.38621520996094, 15.06396484375, 2.429229736328125, 120.12344360351562, 107.6661605834961, 134.81248474121094, 44.85517883300781, 114.47491455078125, 191.99264526367188, 180.44168090820312, 57.57940673828125, 57.46136474609375, 15.514400482177734, 12.532390594482422, -45.656219482421875, 100.02590942382812, 14.61761474609375, 20.79894256591797, 55.752349853515625, 31.86767578125, 13.37835693359375, 116.74748229980469, 126.41189575195312, -140.27188110351562, 155.69369506835938, -13.468658447265625, 7.4675750732421875, 49.997772216796875, -154.97021484375, 5.6062469482421875, 41.12281799316406, 142.95806884765625, 203.7552490234375, 110.57337951660156, 27.784576416015625, 56.145233154296875, 170.38601684570312, 138.27987670898438, 60.654937744140625, 189.8419189453125, 118.96273803710938, -44.1171875, 159.59678649902344, 9.630767822265625, -15.103179931640625, -40.56922912597656, 40.258880615234375, 34.617645263671875, -138.347412109375, 8.778997421264648, 85.57415771484375, 140.720703125, 2.841400146484375, -97.68490600585938, -13.893524169921875, 142.210205078125, -7.713623046875, 21.834075927734375, 29.803466796875, 64.0157470703125, 178.26202392578125, 149.36837768554688, 120.03343963623047, 44.850799560546875, 32.67413330078125, 143.23587036132812, 20.5733642578125, 166.971923828125, -108.31997680664062, 7.840667724609375, 178.08079528808594, 34.11912536621094, 24.23992919921875, 36.262359619140625, 148.54296875, 93.69145202636719, 30.8310546875, 137.91696166992188, 208.15911865234375, 157.96182250976562, 152.97235107421875, -70.13201904296875, 57.12261962890625, 6.265155792236328, -147.970703125, -93.97077941894531, -101.08074951171875, 12.002143859863281, -147.5875244140625, 47.8143310546875, -145.7582244873047, -81.42884826660156, 12.911285400390625, 176.10781860351562, 73.28727722167969, 46.87786865234375, 8.209228515625, 34.23640441894531, 164.25143432617188, -29.253021240234375, -12.087890625, 23.64459991455078, -24.64556884765625, -7.7373046875, 157.98260498046875, -50.414398193359375, 47.884735107421875, 19.533935546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000350.npy"}
{"epoch": 0.7329842931937173, "step": 351, "batch_size": 128, "mean": 34.0787239074707, "std": 87.52424621582031, "min": -200.73294067382812, "p10": -50.535343933105466, "median": 14.922119140625, "p90": 157.7618255615234, "max": 225.9617919921875, "pos_frac": 0.6171875, "sample": [0.0, 140.4058837890625, 14.159271240234375, 194.38375854492188, -2.7061004638671875, 180.46334838867188, 11.884025573730469, -40.42829132080078, 121.60971069335938, -6.503875732421875, 31.523635864257812, -49.90472412109375, 15.15740966796875, 20.71917724609375, 156.24795532226562, -133.09910583496094, 24.581363677978516, 108.81404876708984, -32.060791015625, 20.05535888671875, 21.09027099609375, 136.0883331298828, -14.625900268554688, 198.0745849609375, 91.61582946777344, 154.60455322265625, 8.457025527954102, -17.018585205078125, -6.18804931640625, 1.841278076171875, 168.98297119140625, 174.0568084716797, 63.892578125, 38.9150505065918, -28.91314697265625, -54.4071044921875, -2.9997406005859375, 136.81752014160156, -34.148826599121094, -83.15104675292969, -98.47552490234375, 141.262939453125, 12.57891845703125, 45.60052490234375, -76.4912109375, 144.04566955566406, 8.27130126953125, 16.047576904296875, -24.563629150390625, 18.75537109375, -67.45797729492188, -27.41925048828125, 16.68011474609375, -10.519256591796875, 15.15869140625, -2.542572021484375, 153.78695678710938, -35.164306640625, -11.24737548828125, 11.67608642578125, 174.42193603515625, 222.50454711914062, 35.050010681152344, 194.2208251953125, 151.06460571289062, -19.151519775390625, 37.13958740234375, 14.68682861328125, 125.55999755859375, -98.35844421386719, -4.57159423828125, 119.18275451660156, -32.21478271484375, 11.5224609375, 161.294189453125, -155.77679443359375, 30.3238525390625, -13.001922607421875, 24.499847412109375, 23.64691162109375, 0.0, -2.8614501953125, 24.4178466796875, 155.90176391601562, 140.16477966308594, -0.7951240539550781, 0.2147064208984375, 19.507972717285156, 121.806640625, -29.040130615234375, 31.158782958984375, 6.4169158935546875, 98.12832641601562, -14.14394760131836, 0.0, 178.25665283203125, 163.19232177734375, 0.9620361328125, 12.46938705444336, -12.519287109375, 3.1008758544921875, -200.73294067382812, 110.5072021484375, 155.10748291015625, 25.069442749023438, -4.2481231689453125, 225.9617919921875, 19.4222412109375, 154.5046844482422, -3.272216796875, -0.4124584197998047, 8.009674072265625, -13.061279296875, 33.49913024902344, -0.7273769378662109, 147.13851928710938, 20.301353454589844, -45.390472412109375, -147.360595703125, -148.54718017578125, 25.394989013671875, 71.22189331054688, -175.69244384765625, 83.71368408203125, -42.283477783203125, -52.00679016113281, 214.63372802734375, 44.674407958984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000351.npy"}
{"epoch": 0.7350785340314137, "step": 352, "batch_size": 128, "mean": 58.40721130371094, "std": 80.99685668945312, "min": -133.8724365234375, "p10": -40.160055541992186, "median": 47.59492492675781, "p90": 166.90416259765624, "max": 203.03585815429688, "pos_frac": 0.7421875, "sample": [116.2451171875, 15.019775390625, 157.376220703125, -49.774810791015625, 9.454414367675781, 189.44244384765625, 121.06946563720703, 110.10050201416016, 112.55461120605469, 3.5033035278320312, -48.57891845703125, 133.505859375, -133.8724365234375, 18.1785888671875, 143.01124572753906, 4.611469268798828, 32.699188232421875, 175.06130981445312, 141.89083862304688, 115.08840942382812, 195.1793212890625, -37.02610778808594, -40.312774658203125, 176.23995971679688, 35.98291015625, -94.45391845703125, -16.52691650390625, 97.47477722167969, 17.63214111328125, 119.90851593017578, -1.4306411743164062, 57.345733642578125, 171.114501953125, 16.100982666015625, 131.3177490234375, -95.11798858642578, 13.80902099609375, 99.61572265625, -16.881942749023438, 14.06658935546875, 75.9130859375, -120.48773193359375, -23.538497924804688, 91.58963012695312, -27.100189208984375, 23.694427490234375, 0.0, 74.18118286132812, 18.428146362304688, 167.286376953125, -2.113433837890625, 179.90382385253906, 126.37081909179688, 30.94610595703125, 145.06808471679688, 71.09982299804688, 22.335586547851562, 0.0, 67.79351806640625, 95.0623779296875, 109.45428466796875, 2.7030181884765625, 17.820037841796875, 107.4947509765625, -64.97512817382812, -22.380401611328125, 11.942855834960938, 53.790313720703125, -100.03216552734375, 5.885009765625, 109.86396026611328, 189.61050415039062, 69.41705322265625, 36.26240539550781, 7.96710205078125, -54.308685302734375, 170.83905029296875, 4.048152923583984, -70.83172607421875, -86.57646179199219, 203.03585815429688, 170.98548889160156, 166.5605926513672, 22.30750274658203, -7.930877685546875, 142.10635375976562, 10.4659423828125, -40.0946044921875, -0.1402740478515625, 138.00363159179688, 60.721923828125, -9.358955383300781, 54.107421875, 32.660099029541016, 38.84422302246094, 45.4581298828125, 153.25698852539062, 132.58297729492188, -7.6392059326171875, 177.7716827392578, 49.731719970703125, 108.84571838378906, 134.0559539794922, 166.7403564453125, 149.61619567871094, -30.045166015625, 140.40093994140625, 152.77020263671875, 100.66697692871094, -5.759574890136719, 23.295013427734375, 130.21328735351562, 162.85089111328125, 3.547576904296875, -83.35310363769531, -9.050003051757812, 29.532398223876953, 162.973876953125, 130.08172607421875, 174.19786071777344, -23.108963012695312, -33.729156494140625, 131.3508758544922, 132.73687744140625, 88.91427612304688, 116.64647674560547, 20.471935272216797, 142.77194213867188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000352.npy"}
{"epoch": 0.7371727748691099, "step": 353, "batch_size": 128, "mean": 55.71308517456055, "std": 82.15934753417969, "min": -162.9925537109375, "p10": -25.4486572265625, "median": 46.217037200927734, "p90": 153.53270111083984, "max": 239.93243408203125, "pos_frac": 0.765625, "sample": [42.28059387207031, 138.22348022460938, 61.06318283081055, -67.36001586914062, 65.93862915039062, 165.4078369140625, -116.32355499267578, 149.16796875, 153.42367553710938, 14.7073974609375, 73.17877197265625, 29.365737915039062, 6.997802734375, 68.45352172851562, -7.950531005859375, 17.883636474609375, -17.158233642578125, -26.13507080078125, 41.713958740234375, -17.205902099609375, 104.31715393066406, 85.65846252441406, 0.7022476196289062, -162.9925537109375, 135.98252868652344, 29.82373046875, 71.12016296386719, 127.53842163085938, 221.84869384765625, -0.6496658325195312, 176.501220703125, 134.8638916015625, 152.67120361328125, -25.15447998046875, 160.8050537109375, -144.63380432128906, 180.85870361328125, 95.6339111328125, 78.16357421875, 239.93243408203125, 42.11674499511719, -94.80349731445312, 37.46092224121094, 153.41253662109375, 36.512237548828125, -9.37353515625, 152.29327392578125, 226.25613403320312, 196.90847778320312, 129.38818359375, 135.3691864013672, 62.688331604003906, -90.89575958251953, 124.70188903808594, 15.01788330078125, 206.99935913085938, 2.616537094116211, -17.446006774902344, 124.7667007446289, 72.50834655761719, 179.20840454101562, 86.97296142578125, -32.441558837890625, -101.73236083984375, 1.1902923583984375, 98.67733764648438, -123.29734802246094, -14.54144287109375, 76.76068115234375, 26.997894287109375, 32.86334228515625, 74.53425598144531, 133.6691131591797, 146.63796997070312, 143.28216552734375, 124.07489013671875, 13.141410827636719, 138.16677856445312, 68.53956604003906, 9.670433044433594, 139.188720703125, 50.153480529785156, 56.939300537109375, 25.400634765625, 27.678924560546875, 97.64373779296875, 32.63676452636719, 127.33174133300781, 119.35440063476562, 4.394355773925781, -6.662309646606445, -22.880706787109375, 130.61273193359375, 75.78396606445312, 15.384605407714844, -21.84979248046875, -3.6662120819091797, 11.668533325195312, 11.476905822753906, -13.51409912109375, 70.64013671875, 9.86712646484375, 26.41314697265625, 3.7927417755126953, 156.8617401123047, 38.09356689453125, 82.70635986328125, 52.01019287109375, 146.7398223876953, -63.266937255859375, 27.25232696533203, -0.6523208618164062, 27.537357330322266, 63.479705810546875, 5.816373825073242, 0.0, 200.00656127929688, 74.86724853515625, -29.4705810546875, -17.123931884765625, 2.65875244140625, -147.20089721679688, -13.19482421875, 19.36737060546875, 153.78709411621094, 114.7515869140625, 138.21307373046875, 102.70965576171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000353.npy"}
{"epoch": 0.7392670157068063, "step": 354, "batch_size": 128, "mean": 34.746826171875, "std": 88.47724914550781, "min": -169.17059326171875, "p10": -79.6040756225586, "median": 13.474225997924805, "p90": 149.1128967285156, "max": 221.87362670898438, "pos_frac": 0.640625, "sample": [8.029243469238281, -48.99775695800781, -136.58287048339844, 135.50465393066406, 46.53851318359375, 200.60650634765625, 100.5096435546875, 7.127647399902344, -9.557003021240234, 3.8987770080566406, 56.89598083496094, 38.92999267578125, -69.2230224609375, 14.957504272460938, 148.568115234375, -12.06842041015625, -13.428001403808594, -0.2769775390625, 7.053863525390625, -2.36541748046875, 111.70146179199219, 132.05564880371094, 89.84457397460938, -75.66876220703125, 27.120201110839844, 13.329933166503906, -10.18691635131836, 8.928939819335938, 139.406494140625, 198.96066284179688, -82.18217468261719, -2.139495849609375, 0.15329933166503906, 56.172760009765625, 174.672607421875, 24.685546875, 13.618518829345703, -1.15460205078125, 145.8001251220703, 140.12692260742188, 6.967155456542969, -55.14056396484375, 111.97987365722656, -17.1234130859375, 148.80804443359375, 147.4376220703125, -1.948394775390625, 8.066978454589844, -163.83963012695312, 4.9603118896484375, -11.3458251953125, 197.09506225585938, 56.88134765625, 171.57217407226562, 174.48403930664062, 173.58026123046875, 127.45870208740234, 58.90924072265625, 35.70556640625, 90.97659301757812, 69.44834899902344, -26.700668334960938, -16.11034393310547, -125.37112426757812, 1.4898681640625, 133.86448669433594, 9.726009368896484, 221.87362670898438, -1.2749481201171875, -4.080940246582031, 149.82421875, -43.50269317626953, 125.59844970703125, 7.638574600219727, 194.49066162109375, 58.52162170410156, 110.11966705322266, -117.59278869628906, 3.679962158203125, 1.1682872772216797, -2.85308837890625, -74.0057373046875, 182.5958251953125, 41.179779052734375, -18.521774291992188, 34.29071044921875, -4.29705810546875, 140.5708770751953, -18.468597412109375, -78.49917602539062, 18.575439453125, -2.7929000854492188, 146.509033203125, 8.926223754882812, 118.86878967285156, 26.277420043945312, -9.621978759765625, -27.22344970703125, -106.771728515625, -142.06607055664062, 74.29405212402344, 4.40411376953125, -3.88055419921875, 125.81011962890625, 125.73904418945312, 53.062225341796875, 4.62939453125, -4.4527587890625, 19.511734008789062, 181.93463134765625, 144.43191528320312, 82.69404602050781, -21.89295196533203, 153.6854248046875, -133.4311065673828, 73.15673828125, -169.17059326171875, 126.93217468261719, -110.30181884765625, -101.99176025390625, 119.22247314453125, 24.9630069732666, -106.14645385742188, 16.361328125, -54.10614013671875, -110.28251647949219, 82.64727783203125, 17.436080932617188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000354.npy"}
{"epoch": 0.7413612565445026, "step": 355, "batch_size": 128, "mean": 55.332244873046875, "std": 83.3764877319336, "min": -194.78839111328125, "p10": -33.363513183593746, "median": 38.204917907714844, "p90": 166.1258361816406, "max": 206.1087646484375, "pos_frac": 0.78125, "sample": [-2.1215057373046875, -194.78839111328125, 160.3050537109375, 107.94256591796875, 1.04827880859375, 11.434516906738281, 53.5762939453125, 2.86041259765625, 48.06976318359375, 64.50848388671875, 154.0384063720703, 7.869499206542969, -10.311546325683594, 129.68646240234375, 4.124237060546875, 151.70228576660156, 178.25213623046875, 147.607421875, -5.4149932861328125, 0.6976318359375, -63.38078308105469, 16.82342529296875, 8.671173095703125, 177.4263916015625, 13.75543212890625, 180.6593475341797, -14.326622009277344, -9.98779296875, 38.58229064941406, 16.83544921875, 28.616546630859375, 24.018829345703125, 58.099090576171875, 181.376220703125, 176.4093017578125, 49.31159973144531, 185.13003540039062, 37.06184387207031, 8.81671142578125, -33.131500244140625, 82.20206451416016, 147.6553497314453, 73.16445922851562, 93.55265808105469, -55.64787292480469, 31.49285888671875, -8.235504150390625, 81.9296875, 11.125831604003906, 14.0743408203125, 149.10568237304688, 24.205764770507812, 127.0216064453125, 11.810226440429688, 2.017730712890625, 75.54644775390625, 2.7207679748535156, 39.23529815673828, 39.62127685546875, -14.057125091552734, 168.16888427734375, 9.636539459228516, -33.904876708984375, -11.795928955078125, 116.54924774169922, -150.49859619140625, 24.26116943359375, 206.1087646484375, 89.6268310546875, 150.9718475341797, -0.2717323303222656, 170.2938232421875, -155.6466064453125, 153.51303100585938, 0.9076919555664062, 50.25335693359375, 139.44839477539062, 91.61688232421875, 184.65975952148438, 78.458251953125, 103.11444091796875, 144.63992309570312, 2.892120361328125, -2.7674102783203125, 31.5537109375, -53.3978271484375, 32.64897155761719, -4.76165771484375, 31.526580810546875, 35.144378662109375, 137.22137451171875, -75.01815795898438, -140.35653686523438, 3.669281005859375, -8.9276123046875, 41.31451416015625, 19.490097045898438, 165.66790771484375, -38.5003662109375, 151.15878295898438, 37.827545166015625, -87.32482147216797, 198.447509765625, 120.49066162109375, 175.64849853515625, 152.49935913085938, 155.11793518066406, 94.55120849609375, -64.03413391113281, -14.971435546875, 11.975616455078125, 34.31016540527344, 14.664169311523438, 162.02285766601562, 132.916259765625, 137.59994506835938, 146.60653686523438, 125.39196014404297, 143.61473083496094, 118.26121520996094, 39.417144775390625, -3.79840087890625, 89.42288208007812, 167.1943359375, 56.57110595703125, -131.83786010742188, 158.0340118408203, 32.843231201171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000355.npy"}
{"epoch": 0.743455497382199, "step": 356, "batch_size": 128, "mean": 46.64875030517578, "std": 77.45056915283203, "min": -169.86416625976562, "p10": -31.921499252319332, "median": 31.796131134033203, "p90": 143.85018768310547, "max": 179.4369659423828, "pos_frac": 0.71875, "sample": [0.221282958984375, -8.708084106445312, 143.61241149902344, -3.271087646484375, 112.35476684570312, -73.61431884765625, 18.2822265625, -46.14093017578125, 8.746450424194336, 1.956817626953125, -13.048271179199219, 10.18218994140625, 22.85650634765625, 106.49472045898438, -7.740800857543945, 3.0938491821289062, 24.973976135253906, -20.27972412109375, 4.0577392578125, 101.99188232421875, -47.583404541015625, 139.24072265625, 29.623687744140625, 35.76824951171875, -1.6277828216552734, 171.89349365234375, -111.38261413574219, 137.7871856689453, 3.1086578369140625, 122.28460693359375, 36.055023193359375, 160.84991455078125, 5.2852783203125, 101.66542053222656, 141.0000762939453, 152.34703063964844, 122.67214965820312, 58.91485595703125, 169.61148071289062, 112.96321105957031, -137.6739044189453, 142.98526000976562, 11.032981872558594, 9.048759460449219, 71.20780944824219, 33.8577880859375, -7.068023681640625, 171.95538330078125, 90.58648681640625, -6.310390472412109, 40.74781799316406, 100.41549682617188, 149.0740203857422, 168.14874267578125, 0.06727790832519531, 38.301300048828125, -93.94906616210938, 136.69000244140625, 116.41024780273438, 39.137298583984375, 151.38497924804688, -130.20376586914062, 139.1675262451172, 114.96664428710938, 139.30125427246094, 135.8653564453125, 126.67208862304688, -9.625442504882812, 2.0011367797851562, 4.3050537109375, -11.841842651367188, 22.9613037109375, 126.43206787109375, 20.72582244873047, -12.71685791015625, 37.522300720214844, 120.55342102050781, 0.2840728759765625, 157.43028259277344, -56.55863952636719, 31.700111389160156, -30.486759185791016, 177.0557861328125, 30.8863525390625, 2.128692626953125, 84.86930847167969, 132.59896850585938, 105.09657287597656, 32.377994537353516, -0.9614486694335938, -85.78346252441406, -4.183055877685547, -10.571090698242188, -169.86416625976562, 56.78987121582031, 16.564117431640625, 47.5064697265625, 0.0, 80.4130859375, 31.89215087890625, 120.2684326171875, 25.302154541015625, -8.768671035766602, -19.278228759765625, 0.3216419219970703, 16.485321044921875, 36.99667739868164, 127.09440612792969, -2.19879150390625, -88.85533142089844, 144.40499877929688, 179.4369659423828, 142.85247802734375, 134.69439697265625, -24.184226989746094, 154.27960205078125, 133.90216064453125, -20.513534545898438, 43.684967041015625, 136.490478515625, -35.26922607421875, 83.90957641601562, -0.8430938720703125, -120.41780853271484, 52.489654541015625, -14.570415496826172, 23.147323608398438, 142.35739135742188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000356.npy"}
{"epoch": 0.7455497382198953, "step": 357, "batch_size": 128, "mean": 52.0751838684082, "std": 84.22610473632812, "min": -174.42750549316406, "p10": -34.536810302734374, "median": 30.43604278564453, "p90": 167.4090133666992, "max": 238.472412109375, "pos_frac": 0.7265625, "sample": [65.364013671875, 147.57614135742188, -15.778005599975586, 0.8822021484375, 153.53143310546875, 41.859344482421875, 0.0, 2.857269287109375, 103.60218811035156, 5.6616058349609375, 148.43881225585938, -140.18026733398438, 5.740966796875, 238.472412109375, 196.14483642578125, 203.9482421875, 10.832714080810547, -0.921600341796875, 34.39405822753906, 80.42388916015625, 16.173675537109375, 15.89422607421875, 100.54225158691406, 149.78501892089844, 166.8189239501953, -85.19119262695312, 219.01950073242188, 21.93212127685547, 25.89113998413086, 175.49246215820312, 0.853302001953125, 13.231231689453125, -20.42886734008789, 7.771175384521484, 204.82913208007812, 49.959556579589844, -18.1470947265625, -57.37560272216797, 7.694244384765625, 116.44886016845703, 109.34698486328125, 199.350341796875, 23.959808349609375, 120.13606262207031, 127.85328674316406, -31.05657958984375, 4.19073486328125, -110.58181762695312, 159.6251220703125, 69.73590087890625, -31.801971435546875, 36.25609588623047, 42.76811218261719, 9.5452880859375, 110.62063598632812, 134.0709686279297, 177.95947265625, 165.0538330078125, 33.28082275390625, -1.606719970703125, 143.22940063476562, 61.543182373046875, 16.99237060546875, 3.86322021484375, 52.77606201171875, -8.71893310546875, 143.66786193847656, 146.96881103515625, 57.99346923828125, 34.131195068359375, -11.61077880859375, -86.43368530273438, -27.275436401367188, 4.0236663818359375, 2.150909423828125, 145.74185180664062, 88.98265075683594, -60.01690673828125, 19.87060546875, -34.38232421875, -53.29803466796875, 4.8623046875, 13.771896362304688, 27.591262817382812, 178.7139892578125, 43.168243408203125, 155.44973754882812, -34.89727783203125, -40.114105224609375, -1.3247222900390625, -9.043937683105469, 42.884765625, 82.79986572265625, -38.48719787597656, 50.516632080078125, 140.20118713378906, -0.52642822265625, -108.80780029296875, 0.0, 168.785888671875, -134.671142578125, 0.0, 39.2965087890625, -1.998931884765625, 160.95126342773438, 14.101997375488281, 15.826805114746094, 114.31478881835938, 157.21200561523438, 133.74261474609375, 173.1929931640625, -174.42750549316406, 152.72207641601562, 83.59786987304688, 2.320333480834961, 109.97091674804688, -3.809783935546875, 120.374755859375, 22.553239822387695, 184.13140869140625, 185.81985473632812, 34.90819549560547, -7.090354919433594, -23.363265991210938, 131.5329132080078, -2.4752273559570312, 2.775238037109375, 115.61991882324219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000357.npy"}
{"epoch": 0.7476439790575916, "step": 358, "batch_size": 128, "mean": 56.87163543701172, "std": 88.29068756103516, "min": -173.96456909179688, "p10": -49.54200439453124, "median": 38.05805587768555, "p90": 167.03619079589842, "max": 215.63104248046875, "pos_frac": 0.7578125, "sample": [-0.27349853515625, 3.0733566284179688, 104.12184143066406, -55.188873291015625, 7.179473876953125, 95.7597885131836, -34.983642578125, 115.87371826171875, 181.2371826171875, 215.63104248046875, 10.2403564453125, 5.158580780029297, 23.609237670898438, 151.7660369873047, 166.5755615234375, 153.8082733154297, 18.5006103515625, 179.65036010742188, 168.11099243164062, 35.39849853515625, -119.41139221191406, 139.19427490234375, -27.935958862304688, -119.89865112304688, 57.804595947265625, 6.7467498779296875, 127.92622375488281, -1.9090576171875, 173.83096313476562, -8.60845947265625, 13.108505249023438, 109.83460998535156, 38.593780517578125, 127.95887756347656, 95.4046630859375, 38.229217529296875, 197.5068359375, -73.42105102539062, 33.3887939453125, 127.25204467773438, -117.61280822753906, 165.99343872070312, 159.20440673828125, 169.98223876953125, 30.03064727783203, 154.11416625976562, 123.2917709350586, 149.9078826904297, 154.48648071289062, 4.47467041015625, -31.00152587890625, -6.575328826904297, -7.0521697998046875, 25.448211669921875, 146.46958923339844, -10.663131713867188, 157.76324462890625, 160.45870971679688, -67.22831726074219, 202.595703125, -73.36532592773438, -173.96456909179688, 165.25262451171875, -15.573036193847656, 6.616310119628906, 179.493408203125, 81.2108154296875, 30.660602569580078, 64.83355712890625, 11.212005615234375, 37.88689422607422, 11.489540100097656, 19.718894958496094, 25.030990600585938, 6.243316650390625, -118.34820556640625, 144.41873168945312, 99.62818908691406, -13.532707214355469, 179.2137451171875, 76.46842193603516, -85.57516479492188, 129.68719482421875, 50.47808837890625, 168.25228881835938, 37.53814697265625, 115.739990234375, 60.20367431640625, 51.14677047729492, 163.334716796875, 8.140693664550781, 184.21713256835938, 142.161376953125, 98.68486022949219, 123.097900390625, 127.52130126953125, -33.27099609375, 10.91741943359375, 12.05328369140625, 137.50924682617188, 185.41748046875, 8.4779052734375, 120.74029541015625, 166.52230834960938, 95.77801513671875, 164.43167114257812, 154.45855712890625, -164.73300170898438, -16.22443389892578, -101.29910278320312, -118.23587036132812, 6.033496856689453, 131.65748596191406, 5.074249267578125, 28.57459259033203, -33.00634002685547, 8.150886535644531, 13.651792526245117, -2.113983154296875, 29.470428466796875, 0.0, 69.84698486328125, 65.93365478515625, 45.547332763671875, 141.91708374023438, -47.121917724609375, 32.95098876953125, -3.6957244873046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000358.npy"}
{"epoch": 0.749738219895288, "step": 359, "batch_size": 128, "mean": 53.29281997680664, "std": 81.1598892211914, "min": -186.10400390625, "p10": -30.780877685546876, "median": 34.76361083984375, "p90": 157.4881408691406, "max": 217.22543334960938, "pos_frac": 0.765625, "sample": [14.395027160644531, -54.5341796875, 133.96595764160156, 175.50323486328125, 104.58248901367188, -101.22119140625, 87.29457092285156, 15.469955444335938, 34.743408203125, 58.457000732421875, 29.31855583190918, 108.02784729003906, 39.963558197021484, 7.354381561279297, -13.076751708984375, -16.4324951171875, 111.96411895751953, -5.9710540771484375, 20.0419921875, 22.221649169921875, 124.79840850830078, -11.809890747070312, 142.59646606445312, 167.522216796875, -33.5050048828125, 143.47610473632812, 13.936676025390625, 32.72161865234375, 73.52044677734375, -180.544921875, 0.97198486328125, 198.70635986328125, 5.46746826171875, 9.49481201171875, 2.708353042602539, 59.751739501953125, -8.070016860961914, 140.36019897460938, 157.11297607421875, 166.40179443359375, -24.156509399414062, -118.6756591796875, 137.54571533203125, 3.268890380859375, 92.37040710449219, 105.622314453125, 15.698974609375, 101.71759033203125, 132.57672119140625, 125.58318328857422, 143.9271240234375, 3.1212844848632812, 210.44821166992188, -40.79948425292969, 134.59579467773438, 1.3884735107421875, 55.03057861328125, -164.9522705078125, 125.02625274658203, 43.361328125, 191.8402099609375, 123.28652954101562, -186.10400390625, 60.321929931640625, 15.577667236328125, -6.6071319580078125, -10.8232421875, 47.11639404296875, 129.12252807617188, 90.954833984375, -2.940093994140625, 48.35931396484375, -30.37957000732422, 34.7838134765625, 98.15935516357422, 136.204833984375, -13.47088623046875, 131.9891357421875, 13.266021728515625, 141.3509979248047, 1.38702392578125, 0.337432861328125, 6.144927978515625, 114.77371215820312, 24.789840698242188, 17.52165985107422, 18.276336669921875, 145.07601928710938, 37.2392578125, 9.889404296875, 163.58409118652344, 14.047500610351562, 158.363525390625, 129.17892456054688, 21.64227294921875, -31.0318603515625, -103.77603149414062, 123.63900756835938, -34.930809020996094, -0.5437297821044922, 29.60589599609375, 175.00997924804688, 32.80804443359375, -57.125244140625, 14.982070922851562, -26.255828857421875, 167.06219482421875, -30.94720458984375, 53.539581298828125, 149.7861328125, 135.26577758789062, 114.97467803955078, 8.91802978515625, 111.07442474365234, -9.652679443359375, 174.5997314453125, 151.99896240234375, 97.43313598632812, 97.53561401367188, -1.0412673950195312, -6.6759033203125, 217.22543334960938, 28.856903076171875, 9.892240524291992, 160.8760986328125, 37.28959655761719, 127.15225219726562, -30.7095947265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000359.npy"}
{"epoch": 0.7518324607329843, "step": 360, "batch_size": 128, "mean": 47.82188415527344, "std": 74.77693176269531, "min": -201.950927734375, "p10": -20.318609619140624, "median": 27.163433074951172, "p90": 155.58990173339842, "max": 211.26593017578125, "pos_frac": 0.7265625, "sample": [88.63204956054688, 121.08207702636719, -67.59536743164062, 157.88427734375, -137.15359497070312, -8.534543991088867, -56.78550720214844, 55.5440673828125, 211.26593017578125, -46.16505432128906, 32.78790283203125, 104.78678894042969, -20.747390747070312, 120.63470458984375, 45.76513671875, 99.57148742675781, 15.256591796875, 141.344970703125, -20.29888916015625, 54.459259033203125, 171.55548095703125, 60.415771484375, 12.74444580078125, 110.34805297851562, 124.23822784423828, -18.793243408203125, 4.198600769042969, 20.1456298828125, -15.4268798828125, 163.26480102539062, 34.075836181640625, 178.67262268066406, 95.81265258789062, 15.931060791015625, 62.04058837890625, 6.58355712890625, -20.3646240234375, 132.38323974609375, -30.69976806640625, 60.64244079589844, 94.53045654296875, 99.71026611328125, 140.36907958984375, 24.71600341796875, 72.08880615234375, 22.458377838134766, 18.334991455078125, 138.07667541503906, -8.73004150390625, 39.200042724609375, -10.331268310546875, 146.89793395996094, 122.90017700195312, 100.50262451171875, 190.05636596679688, -1.6547698974609375, 169.40003967285156, -37.976348876953125, 145.26242065429688, 11.992744445800781, 53.302734375, 39.9620361328125, 151.020751953125, 114.10486602783203, 17.787689208984375, 1.3804512023925781, 166.27169799804688, 18.5975341796875, 52.77833557128906, -129.291015625, 10.822479248046875, 127.65802001953125, 1.4516925811767578, 156.71116638183594, -17.384796142578125, 19.789241790771484, 85.21598815917969, 164.85726928710938, 142.5972900390625, 133.96405029296875, -9.388748168945312, 115.1982421875, 6.257940292358398, -6.681922912597656, -201.950927734375, 105.29325866699219, 39.598602294921875, 9.119964599609375, -18.65460205078125, 10.202217102050781, 148.03753662109375, -47.861114501953125, 178.61257934570312, 83.0743637084961, -0.9039459228515625, -5.011474609375, 4.3101806640625, -126.44039916992188, 10.075286865234375, -0.9471015930175781, 53.04804992675781, 0.0, 22.03521728515625, 20.398956298828125, 155.10935974121094, 2.3993263244628906, 64.2025146484375, 44.9365234375, -1.6961212158203125, 4.008661270141602, 7.1898193359375, -5.412261962890625, -4.356636047363281, 120.78482055664062, 14.232643127441406, 96.00167846679688, -23.688880920410156, -15.66180419921875, 0.0, 10.340583801269531, 42.07280731201172, 29.610862731933594, 35.816650390625, -14.731498718261719, -14.693489074707031, 17.88916015625, 159.05166625976562, 161.466796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000360.npy"}
{"epoch": 0.7539267015706806, "step": 361, "batch_size": 128, "mean": 63.165191650390625, "std": 88.21361541748047, "min": -169.97906494140625, "p10": -32.081784057617185, "median": 57.917327880859375, "p90": 166.04994201660153, "max": 233.69537353515625, "pos_frac": 0.7265625, "sample": [129.51626586914062, -8.264251708984375, -119.45281982421875, 33.868072509765625, 31.511581420898438, -6.14862060546875, 0.8544425964355469, -27.582672119140625, 156.73623657226562, 44.975189208984375, 124.5804214477539, 14.021728515625, 0.3448333740234375, 199.35540771484375, -6.207605361938477, 155.24256896972656, -3.6048011779785156, 14.603919982910156, -28.167816162109375, 0.8837814331054688, 155.05184936523438, 184.78244018554688, -96.14839172363281, 152.88973999023438, -0.19953155517578125, 184.1148681640625, 127.83804321289062, 135.2318115234375, 53.38652801513672, 186.60186767578125, 9.307342529296875, -5.1089324951171875, 108.09314727783203, 81.28319549560547, 14.6978759765625, 141.9321746826172, -147.560791015625, 92.60986328125, 125.19430541992188, 96.46246337890625, 186.09176635742188, 170.22488403320312, 24.744369506835938, 90.86161804199219, 200.3232421875, 27.6531982421875, -113.19476318359375, -2.5110015869140625, 58.246978759765625, -11.2884521484375, -0.5009059906005859, 52.74763488769531, 109.15699768066406, 0.0, 57.565704345703125, 12.67529296875, 145.85194396972656, 65.9029541015625, 3.436370849609375, -20.687713623046875, 36.58085632324219, 138.23097229003906, -6.3780364990234375, 164.26068115234375, 118.1888427734375, -49.336181640625, 33.82122802734375, 99.90869903564453, 23.55413818359375, -71.56948852539062, 78.9417724609375, -4.40545654296875, -119.33212280273438, -34.742950439453125, 114.98028564453125, 0.0, 98.5562744140625, -40.21305847167969, 57.587677001953125, -10.9112548828125, 202.29855346679688, 152.88449096679688, 176.9473114013672, 69.97689819335938, -169.97906494140625, 139.77923583984375, 2.097320556640625, -30.9412841796875, 233.69537353515625, 134.61334228515625, -66.89041137695312, 63.27452087402344, -125.5421142578125, 156.83462524414062, 171.4385986328125, 115.89109802246094, 104.69129943847656, 28.615020751953125, 150.78736877441406, 128.65902709960938, 159.800537109375, -7.216564178466797, 203.0228729248047, 146.66827392578125, 159.45166015625, 150.35516357421875, 133.58892822265625, 41.0169677734375, 24.516937255859375, 146.16168212890625, 160.29840087890625, 48.541221618652344, -15.279876708984375, 108.61827087402344, 170.37777709960938, 138.48199462890625, -3.790802001953125, 154.46209716796875, 163.73226928710938, 103.98443603515625, 10.803787231445312, 125.99259948730469, -6.3704376220703125, 100.42842102050781, 34.992164611816406, 18.5430908203125, 148.09710693359375, -165.81472778320312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000361.npy"}
{"epoch": 0.7560209424083769, "step": 362, "batch_size": 128, "mean": 55.9232177734375, "std": 82.28237915039062, "min": -175.26004028320312, "p10": -33.32489013671875, "median": 46.5399284362793, "p90": 158.95775604248044, "max": 247.40496826171875, "pos_frac": 0.7578125, "sample": [21.048912048339844, 91.1173095703125, -8.828876495361328, 69.12614440917969, 142.99310302734375, 37.94606018066406, 135.29974365234375, -8.64251708984375, 79.6998291015625, 12.231292724609375, 125.59298706054688, -175.26004028320312, -140.3959197998047, -104.74634552001953, 205.70779418945312, 6.772470474243164, 165.7215118408203, -11.861572265625, 48.296875, 123.67788696289062, 94.59130859375, 140.49899291992188, 247.40496826171875, 47.36492919921875, 66.865234375, 163.1531524658203, 126.19400024414062, -18.7142333984375, -157.26223754882812, 100.42195129394531, -25.85272216796875, -50.52734375, 90.78456115722656, -6.372711181640625, -37.08519744873047, 131.247802734375, 46.58424377441406, -12.600784301757812, 157.15972900390625, 14.633262634277344, 195.992431640625, 99.7305908203125, 180.7769775390625, 145.22850036621094, 163.44003295898438, 154.08712768554688, 131.17337036132812, 0.646240234375, -12.89404296875, 9.170433044433594, -36.08062744140625, 132.9906463623047, 88.27728271484375, -34.10675048828125, 2.469219207763672, 4.66180419921875, -32.98980712890625, 177.96279907226562, -5.54913330078125, 143.30609130859375, 143.40072631835938, -19.64898681640625, 31.261810302734375, 119.23260498046875, 114.97599792480469, 114.99679565429688, 28.778091430664062, 19.677032470703125, 117.54435729980469, 83.49465942382812, 168.56936645507812, 149.5596466064453, 136.69302368164062, 57.1505126953125, -4.395408630371094, 191.98724365234375, 114.09123229980469, 4.304679870605469, 117.00784301757812, 129.59295654296875, -43.15399169921875, 138.7802734375, -23.84035873413086, 10.279281616210938, -26.249267578125, -70.86447143554688, -36.0421257019043, 182.0281219482422, 9.360382080078125, 14.426460266113281, 25.000885009765625, 2.933319091796875, -7.1855316162109375, 13.089263916015625, 24.947265625, 26.902725219726562, 122.90774536132812, 39.21380615234375, 12.230560302734375, 6.156097412109375, -28.961334228515625, 144.89947509765625, 140.6400604248047, 20.69964599609375, 3.2259674072265625, 101.55032348632812, -142.87454223632812, 13.174591064453125, 28.965576171875, 19.63597869873047, 176.15248107910156, 11.59457015991211, 121.32745361328125, 88.3717041015625, 28.439361572265625, -10.299531936645508, 80.47100830078125, 173.45916748046875, 26.93609619140625, 46.49561309814453, -26.9490966796875, 97.6456527709961, 131.6092071533203, 151.15435791015625, 55.4881591796875, -99.9537353515625, 75.5843505859375, 48.21641540527344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000362.npy"}
{"epoch": 0.7581151832460733, "step": 363, "batch_size": 128, "mean": 61.279449462890625, "std": 84.3860092163086, "min": -190.5465087890625, "p10": -16.128372192382805, "median": 44.48280334472656, "p90": 168.36647338867186, "max": 268.804931640625, "pos_frac": 0.7890625, "sample": [147.318115234375, -3.1200485229492188, -13.693939208984375, 169.04669189453125, 147.91534423828125, 57.2777099609375, -0.5196666717529297, -118.14303588867188, 56.63726806640625, 64.7571792602539, -14.522674560546875, 7.605682373046875, 147.7252197265625, 68.06983947753906, 6.180763244628906, 94.98858642578125, 99.32028198242188, 189.8761749267578, 0.98248291015625, -26.23392677307129, 22.32830810546875, 134.021484375, 188.10488891601562, 83.4718017578125, 21.8507080078125, 178.053955078125, 187.43199157714844, 222.10443115234375, 12.695533752441406, 44.79878234863281, 149.4521942138672, 1.5669403076171875, -7.425846099853516, -4.3943939208984375, 109.4802474975586, 1.435659408569336, -3.488067626953125, 78.001220703125, 150.7627716064453, -0.16400146484375, 7.229469299316406, 23.879661560058594, 100.38827514648438, 126.2702865600586, 69.71473693847656, 68.2060546875, -13.044891357421875, 112.37118530273438, 172.86851501464844, 44.16682434082031, 129.64608764648438, 19.31236457824707, 268.804931640625, 155.92169189453125, -128.74954223632812, 78.20404052734375, 4.128173828125, 163.02008056640625, 34.25726318359375, 116.3160400390625, 25.121063232421875, 35.95478820800781, 167.04124450683594, 130.7655029296875, 177.86642456054688, 0.8609237670898438, 180.303955078125, 72.99786376953125, 39.14080810546875, 6.243206024169922, 145.88856506347656, 2.14715576171875, 176.86846923828125, 222.4334716796875, 7.1447296142578125, 156.50894165039062, -81.44265747070312, -10.388717651367188, 59.09614944458008, 59.1754150390625, 5.587059020996094, 11.943191528320312, -8.24530029296875, 8.824249267578125, 57.01365661621094, -13.183441162109375, -1.67120361328125, -6.54833984375, -54.73760986328125, 15.397125244140625, 96.7991943359375, -127.516357421875, 141.55914306640625, 7.2832489013671875, 41.49189758300781, 168.074951171875, 147.66473388671875, -19.875, 160.19198608398438, 89.4735107421875, 23.442733764648438, 113.40415954589844, -146.4619903564453, 163.172119140625, 169.1929931640625, 142.14175415039062, 25.857398986816406, 49.9349365234375, 161.5924530029297, 58.89079284667969, -190.5465087890625, -30.375320434570312, -36.886260986328125, 0.40793609619140625, 19.09149169921875, 41.91693115234375, 7.334564208984375, 10.613677978515625, 132.06312561035156, 166.931640625, 164.9791717529297, 160.75579833984375, 25.14447021484375, 166.14447021484375, -52.505126953125, 1.632720947265625, 23.595794677734375, -25.418243408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000363.npy"}
{"epoch": 0.7602094240837697, "step": 364, "batch_size": 128, "mean": 44.97412109375, "std": 85.19692993164062, "min": -203.133056640625, "p10": -38.952359008789045, "median": 24.919265747070312, "p90": 152.66011047363278, "max": 238.50341796875, "pos_frac": 0.6796875, "sample": [63.21379089355469, -2.3499183654785156, 111.5048828125, 44.52883529663086, 113.05941772460938, 14.444198608398438, 126.2259521484375, 123.13861083984375, 121.13632202148438, 16.88363265991211, 37.18089294433594, 112.89178466796875, -10.476531982421875, -8.498077392578125, -14.9385986328125, 110.30807495117188, 53.781097412109375, 22.025146484375, 3.751800537109375, 120.45419311523438, -150.28466796875, -13.09918212890625, 107.47810363769531, -155.774658203125, 28.69500732421875, -78.82208251953125, 182.4620361328125, 137.45046997070312, -10.752174377441406, 138.82989501953125, 6.1849365234375, 16.64362335205078, 54.818267822265625, 6.744129180908203, -24.8309326171875, -3.931640625, -13.405685424804688, 135.8707733154297, 4.990264892578125, 143.99044799804688, 114.373779296875, 8.395538330078125, -178.52777099609375, 57.73158264160156, 131.36959838867188, 175.30691528320312, -69.69636535644531, 104.15679931640625, 16.04925537109375, -25.142013549804688, 150.64608764648438, 157.3594970703125, -20.5819091796875, 58.30242919921875, -11.41632080078125, -0.273590087890625, 173.4923095703125, -1.073446273803711, -203.133056640625, 215.6326446533203, 128.38812255859375, -13.752197265625, -9.42926025390625, -61.25276184082031, 34.73454284667969, 126.617919921875, -27.765716552734375, 171.75057983398438, -16.65130615234375, 6.1094970703125, 34.3218994140625, 147.35264587402344, 81.4769287109375, -14.427536010742188, 124.66270446777344, 9.300384521484375, 178.17730712890625, 55.98345184326172, 18.420822143554688, 138.83181762695312, 127.8006591796875, -6.967010498046875, 27.813385009765625, -6.0028076171875, -6.08642578125, 36.373046875, 21.961685180664062, -21.404296875, 51.37284851074219, 207.69512939453125, 127.9154052734375, 13.221488952636719, -34.99468994140625, 9.919418334960938, 141.72781372070312, 126.46697998046875, 2.125934600830078, -1.2802391052246094, -50.739501953125, 238.50341796875, 34.458709716796875, 162.06272888183594, -106.11105346679688, 204.63140869140625, 119.63687133789062, -53.366485595703125, 73.79116821289062, 45.7459716796875, 1.59124755859375, -18.704254150390625, 132.04083251953125, 0.97760009765625, -15.070358276367188, 178.9113311767578, 12.277778625488281, -1.8865966796875, 5.26373291015625, -109.22328186035156, 90.201171875, -102.49746704101562, 4.839168548583984, 57.63844299316406, 43.6395263671875, 9.563018798828125, -48.186920166015625, 107.68798828125, 214.8396453857422, 69.16935729980469], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000364.npy"}
{"epoch": 0.762303664921466, "step": 365, "batch_size": 128, "mean": 60.351539611816406, "std": 81.5783920288086, "min": -166.78103637695312, "p10": -22.894773864746092, "median": 46.6994743347168, "p90": 170.55685119628907, "max": 205.18881225585938, "pos_frac": 0.75, "sample": [0.0, 3.6729698181152344, 119.8450927734375, 4.070747375488281, 165.113525390625, 185.16253662109375, 13.893035888671875, 63.375579833984375, 0.0, 15.9327392578125, 24.51904296875, 4.839141845703125, 14.03375244140625, 175.3388671875, 13.309028625488281, 12.201553344726562, 18.4324951171875, 77.76197052001953, 13.863204956054688, 160.7211151123047, 43.5213623046875, 163.2432861328125, -5.349884033203125, 18.899871826171875, 56.82810974121094, 92.01011657714844, -22.860061645507812, 154.18783569335938, 136.41793823242188, 150.89352416992188, 155.1060791015625, -48.554542541503906, -25.07916259765625, 125.62137603759766, -120.13623046875, -59.36944580078125, 138.20050048828125, 47.960426330566406, 135.8640899658203, 74.80859375, -19.746597290039062, 140.5906982421875, 179.08160400390625, 135.66395568847656, -30.259750366210938, 182.31338500976562, 1.35369873046875, 129.2795867919922, 55.582401275634766, 156.73928833007812, 136.96173095703125, -22.97576904296875, 55.118896484375, -1.3106536865234375, 16.409454345703125, 9.171966552734375, 113.90998840332031, 162.7053680419922, 154.08706665039062, -13.994232177734375, 25.48248291015625, 147.61834716796875, -37.974822998046875, -9.23394775390625, 4.855617523193359, 93.742431640625, 128.24778747558594, 138.971923828125, 70.48117065429688, 101.91357421875, 187.021484375, -7.3976593017578125, 179.35833740234375, -0.1956634521484375, 9.28289794921875, 187.57757568359375, 101.41299438476562, -135.79916381835938, 170.5335693359375, 3.1470413208007812, 177.78060913085938, 3.5548019409179688, 196.96853637695312, 7.17108154296875, 23.891916275024414, -14.830276489257812, 143.80413818359375, 53.91310119628906, 2.3771934509277344, -16.279884338378906, 78.56143188476562, 45.43852233886719, 76.03219604492188, 170.61117553710938, -6.502235412597656, 5.7937469482421875, 3.4481201171875, -17.984710693359375, 137.400146484375, -3.3385009765625, 23.6728515625, -118.22285461425781, -18.1519775390625, -4.6649169921875, 138.03030395507812, 0.0, 29.37567138671875, 7.083038330078125, 136.4250030517578, 181.19552612304688, 79.84310913085938, 109.83421325683594, 6.554805755615234, -22.385498046875, -51.6844482421875, -166.78103637695312, 139.19656372070312, 177.47286987304688, -37.701934814453125, 136.11199951171875, 32.73272705078125, -87.92034912109375, 205.18881225585938, 92.46414184570312, 101.75543212890625, 135.6938934326172, 145.95980834960938, 88.0799560546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000365.npy"}
{"epoch": 0.7643979057591623, "step": 366, "batch_size": 128, "mean": 58.61199951171875, "std": 83.55686950683594, "min": -194.53167724609375, "p10": -31.878826904296872, "median": 46.96324157714844, "p90": 160.69696960449218, "max": 214.61993408203125, "pos_frac": 0.75, "sample": [43.426788330078125, 33.06004333496094, 214.61993408203125, 143.5445098876953, 124.43602752685547, 138.15203857421875, 164.54306030273438, 129.45059204101562, 107.82194519042969, 182.6656494140625, 165.5302734375, 131.64157104492188, 115.30284881591797, -40.997100830078125, 37.555599212646484, 122.02363586425781, 121.49549865722656, 92.34617614746094, -139.60159301757812, 36.24345397949219, -194.53167724609375, 198.2462158203125, -132.23410034179688, 23.528186798095703, 50.49969482421875, 127.68490600585938, 134.73875427246094, 78.1094970703125, 147.32333374023438, 81.98056030273438, -15.988800048828125, 153.08349609375, 27.362075805664062, -34.3277587890625, 159.04864501953125, -177.99911499023438, -3.676513671875, 116.0399169921875, 31.37762451171875, 27.259918212890625, 3.2319812774658203, -46.88751220703125, -15.2838134765625, -18.323875427246094, 0.662506103515625, 104.43580627441406, 143.57240295410156, -26.662445068359375, 55.656463623046875, -13.888916015625, 79.71670532226562, 34.27398681640625, 11.921051025390625, 101.27861022949219, 118.35050964355469, -19.190277099609375, 29.21075439453125, 18.073440551757812, 29.405853271484375, 18.151145935058594, -1.65460205078125, 142.69873046875, 108.47874450683594, 3.5275917053222656, 2.3733291625976562, 170.358642578125, 93.33804321289062, 66.16204833984375, 103.72876739501953, -67.96475219726562, 135.80780029296875, 171.68426513671875, -64.39836120605469, 0.0, -27.04620361328125, 168.8595733642578, 56.32421875, 140.17742919921875, 124.51632690429688, -4.950111389160156, 14.8028564453125, 42.5277099609375, 182.726806640625, 81.31349182128906, -5.1395263671875, 52.0748291015625, 26.513641357421875, 191.73472595214844, -16.697463989257812, -9.928802490234375, -4.55908203125, -30.82928466796875, 104.27145385742188, 13.080352783203125, 100.23420715332031, 135.01426696777344, 100.005615234375, -5.291015625, 57.364013671875, 156.6980743408203, 143.66510009765625, 16.886032104492188, 4.4557647705078125, 191.35888671875, 145.79502868652344, 28.981033325195312, 135.8332977294922, -18.714534759521484, 195.13525390625, 2.1636829376220703, 3.0645751953125, 126.9015884399414, 164.67022705078125, 2.9351806640625, 28.50806427001953, -39.519866943359375, 154.1146240234375, -139.10462951660156, 131.63571166992188, 34.04327392578125, 133.37452697753906, 8.94158935546875, 0.047637939453125, 148.69696044921875, -48.27911376953125, -10.933029174804688, 158.482177734375, -37.231689453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000366.npy"}
{"epoch": 0.7664921465968586, "step": 367, "batch_size": 128, "mean": 59.733154296875, "std": 79.16960144042969, "min": -127.83840942382812, "p10": -27.20403137207031, "median": 42.47123718261719, "p90": 167.46692810058593, "max": 196.59573364257812, "pos_frac": 0.7578125, "sample": [133.67947387695312, 29.66961669921875, -15.574859619140625, 180.7982177734375, -2.1601505279541016, 16.833511352539062, 48.053260803222656, -127.83840942382812, -6.85296630859375, 102.239990234375, 160.662109375, 24.324491500854492, 10.760536193847656, -0.1745452880859375, 90.2271728515625, 185.67098999023438, 25.050399780273438, 166.80764770507812, -0.43994140625, 9.81646728515625, 10.341064453125, 193.08197021484375, 91.50457763671875, -41.564910888671875, 15.367668151855469, 182.17300415039062, -93.77349853515625, -12.1669921875, 50.87889099121094, -18.719955444335938, -24.268051147460938, 81.47976684570312, 3.065164566040039, -37.27458953857422, 144.09213256835938, -13.30572509765625, 100.71707153320312, 134.75840759277344, 127.56991577148438, -94.3397216796875, 155.8155517578125, 96.10916137695312, 6.611644744873047, 118.99844360351562, 159.5015869140625, 4.403099060058594, -70.75021362304688, 7.0875244140625, -12.781219482421875, -13.315780639648438, 47.309417724609375, 9.784439086914062, 99.01420593261719, 77.53216552734375, 142.41856384277344, 153.04208374023438, 132.3782958984375, 117.51380157470703, 76.66343688964844, -28.81304931640625, -17.133544921875, -29.1251220703125, 190.45529174804688, 122.32785034179688, 17.27996826171875, 171.6217041015625, -27.1661376953125, 131.4163818359375, 196.59573364257812, 36.94742202758789, 60.097625732421875, 17.182647705078125, 38.98478698730469, 148.77243041992188, 174.52691650390625, 156.14324951171875, 10.09869384765625, -27.292449951171875, 16.694427490234375, -52.08343505859375, 40.52728271484375, 171.0374298095703, 153.26824951171875, 162.0694580078125, 18.6571044921875, 27.428192138671875, 162.40875244140625, 23.462371826171875, -53.121551513671875, 102.39431762695312, 154.98150634765625, 120.62509155273438, 21.65472412109375, -11.79034423828125, 7.668924331665039, -7.277915954589844, -90.21521759033203, 171.06320190429688, 81.47691345214844, 166.80459594726562, -2.0711421966552734, 177.4722900390625, 151.97752380371094, 5.9027099609375, -13.152511596679688, 89.73377990722656, 67.71124267578125, 164.9459228515625, 7.598846435546875, 169.0052490234375, 75.79714965820312, 112.87262725830078, 27.755279541015625, -123.54186248779297, 47.97669982910156, 5.495880126953125, 2.69500732421875, 135.66751098632812, 95.66476440429688, 148.38369750976562, 18.550840377807617, 44.415191650390625, 18.084228515625, 5.4036865234375, 193.55209350585938, 114.97027587890625, -19.950714111328125, 53.76757049560547], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000367.npy"}
{"epoch": 0.768586387434555, "step": 368, "batch_size": 128, "mean": 65.99723815917969, "std": 82.71160125732422, "min": -151.56866455078125, "p10": -24.3270523071289, "median": 60.48529815673828, "p90": 164.4991912841797, "max": 253.57839965820312, "pos_frac": 0.796875, "sample": [10.40460205078125, 178.03787231445312, 121.30157470703125, 151.0093994140625, 19.665130615234375, 158.8848876953125, 59.69508361816406, -3.483154296875, 6.697845458984375, -11.891983032226562, 136.849853515625, -28.789306640625, 92.20635986328125, 189.9923095703125, 117.883056640625, 56.3594970703125, 208.44613647460938, 18.51800537109375, 69.55001831054688, 106.9792709350586, -19.67449951171875, -10.2547607421875, 47.83349609375, -17.97724151611328, 5.81402587890625, 23.310089111328125, -22.414657592773438, 5.735755920410156, 62.43524169921875, 160.68035888671875, 1.6218032836914062, 49.80133056640625, -75.60192108154297, 71.33760070800781, 11.824462890625, -151.56866455078125, 2.252410888671875, 253.57839965820312, 140.91262817382812, 128.13320922851562, 32.137359619140625, 163.41766357421875, 152.15582275390625, 134.8094482421875, 26.57275390625, 22.724029541015625, 135.49282836914062, -9.164566040039062, 22.17658042907715, -149.0264892578125, 6.34051513671875, 142.6756591796875, 7.178388595581055, 122.78378295898438, -37.812583923339844, -7.402374267578125, 180.02923583984375, 131.83383178710938, -61.181304931640625, 117.89488220214844, 45.313812255859375, 61.2755126953125, 95.6466064453125, 148.7326202392578, 169.87582397460938, 131.32579040527344, -61.701904296875, 75.94020080566406, 30.104217529296875, -7.4904327392578125, -2.924896240234375, 180.81195068359375, 177.0760498046875, 31.99346923828125, 29.792434692382812, 131.34596252441406, 148.6412353515625, 80.23101806640625, 161.46319580078125, 138.16854858398438, 2.9731903076171875, 121.28369140625, 153.73841857910156, 55.36529541015625, -11.069984436035156, 152.39706420898438, 10.575714111328125, 73.9581527709961, -29.406661987304688, 124.32846069335938, -0.7431735992431641, 9.714988708496094, 162.04623413085938, 21.48265838623047, 113.97333526611328, 38.16499328613281, 105.56841278076172, 121.59588623046875, 0.10692596435546875, 122.99246978759766, 164.49758911132812, -84.39722442626953, 93.47273254394531, 179.86212158203125, -68.337890625, 28.405517578125, -9.07293701171875, 98.55999755859375, 204.1356201171875, 172.91952514648438, 27.3021240234375, 148.4126434326172, 29.967498779296875, 10.030288696289062, 22.426353454589844, 142.2686767578125, 164.5029296875, 153.85858154296875, 137.49160766601562, 118.3494873046875, 34.92559814453125, -141.9597625732422, -87.51483154296875, 4.804901123046875, 94.98851776123047, 166.36033630371094, -85.88070678710938, 152.86886596679688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000368.npy"}
{"epoch": 0.7706806282722513, "step": 369, "batch_size": 128, "mean": 46.73614501953125, "std": 86.60578155517578, "min": -176.7765655517578, "p10": -45.89872741699219, "median": 23.668296813964844, "p90": 158.44545288085936, "max": 225.75466918945312, "pos_frac": 0.71875, "sample": [144.6083526611328, 29.46033477783203, 139.07191467285156, -71.62350463867188, 36.86171340942383, -9.71820068359375, 79.50578308105469, 4.172943115234375, 137.91262817382812, 2.029510498046875, 133.11956787109375, 65.3282470703125, -42.181434631347656, 72.61029052734375, 179.82937622070312, 177.672607421875, 120.17288208007812, 135.10121154785156, 21.64227294921875, 128.94400024414062, 186.25503540039062, 3.0127792358398438, 108.83039855957031, 121.16154479980469, -38.05755615234375, -155.23773193359375, 146.68511962890625, -12.37396240234375, 16.157562255859375, 5.07025146484375, 130.2911834716797, 24.8759765625, 158.45681762695312, 107.45831298828125, -3.1947021484375, 138.62127685546875, -41.938262939453125, 46.71527099609375, 120.32742309570312, 34.39045715332031, -2.8715286254882812, 7.381927490234375, 152.10086059570312, -37.193817138671875, 6.15496826171875, -112.89593505859375, -5.470340728759766, 116.0855712890625, -112.87770080566406, 181.35821533203125, -114.75306701660156, -125.3851547241211, 72.18594360351562, -76.680908203125, 173.438720703125, 121.1612777709961, -111.70127868652344, -4.878517150878906, 137.93048095703125, 6.965667724609375, -129.89706420898438, -0.921905517578125, -125.97150421142578, 155.32972717285156, 166.46603393554688, 99.5110092163086, 44.36973571777344, -2.4135360717773438, 159.3201904296875, -32.34663391113281, 225.75466918945312, -14.349639892578125, 34.09156799316406, 17.524139404296875, 2.7535400390625, 199.96987915039062, 65.77398681640625, 0.175750732421875, -40.68018341064453, 129.882568359375, -5.113395690917969, -176.7765655517578, 8.586357116699219, 52.7716064453125, 18.46990966796875, 120.43779754638672, 162.5072021484375, 148.94264221191406, -45.515716552734375, 0.0, 15.58428955078125, 138.38302612304688, 8.922149658203125, 101.82969665527344, -5.4708404541015625, 0.609283447265625, 115.35237121582031, -21.929397583007812, 45.837982177734375, 150.18032836914062, 17.91192626953125, 145.95587158203125, 16.82421875, -46.79241943359375, 65.3271484375, -39.65552520751953, 140.98715209960938, 14.77252197265625, 156.3480224609375, -73.38119506835938, 40.24559020996094, 41.30448913574219, 15.3726806640625, 140.68960571289062, 158.44058227539062, 2.855804443359375, 10.32794189453125, 172.696533203125, -7.349700927734375, 108.74649047851562, 5.646432876586914, 171.97225952148438, 13.29571533203125, 22.460617065429688, 20.621604919433594, -18.064437866210938, 41.34033203125, 3.29205322265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000369.npy"}
{"epoch": 0.7727748691099476, "step": 370, "batch_size": 128, "mean": 60.467498779296875, "std": 88.14250946044922, "min": -192.426025390625, "p10": -21.39812316894531, "median": 35.762123107910156, "p90": 165.80758666992188, "max": 317.1313781738281, "pos_frac": 0.7890625, "sample": [112.26751708984375, -154.58401489257812, 149.63119506835938, 35.109161376953125, 36.65838623046875, 51.679656982421875, -25.609268188476562, 3.62835693359375, 122.54396057128906, 14.872024536132812, 55.07564163208008, 114.91215515136719, -10.566581726074219, 18.4661865234375, 129.10150146484375, 10.64898681640625, -0.620697021484375, 222.060546875, 2.6179885864257812, 30.014251708984375, 133.66790771484375, 29.727331161499023, 130.25836181640625, 141.9176025390625, -17.06927490234375, -3.2976150512695312, -17.569881439208984, 144.96536254882812, 6.447872161865234, 29.50787353515625, 27.624267578125, -22.130615234375, 138.10601806640625, 32.726985931396484, 28.132675170898438, 0.6684722900390625, 221.4696044921875, 92.31199645996094, 3.9237709045410156, 172.79261779785156, 122.54708099365234, -156.58909606933594, 165.94815063476562, 9.317855834960938, 122.94989013671875, 101.50294494628906, 100.88168334960938, 35.447296142578125, 38.6099853515625, 0.5000495910644531, 118.11488342285156, 15.946441650390625, 9.85107421875, 131.77125549316406, 90.45654296875, -17.2760009765625, 137.80953979492188, 1.5244140625, 158.32931518554688, -82.42975616455078, -22.395225524902344, 157.88473510742188, -11.156539916992188, -105.97038269042969, 7.661960601806641, 129.0829620361328, 209.48455810546875, 61.433624267578125, 124.7850341796875, -44.020111083984375, 2.985870361328125, -8.6915283203125, 60.458465576171875, 159.0697021484375, 165.74734497070312, -21.084197998046875, 176.49038696289062, 171.01373291015625, 119.35684204101562, 154.83753967285156, 167.99688720703125, 147.97349548339844, 52.75885009765625, 205.53631591796875, 142.04953002929688, 149.15719604492188, -24.891380310058594, 10.786865234375, -149.4351806640625, 149.16891479492188, 131.227783203125, 35.932533264160156, -102.34457397460938, 24.653350830078125, 317.1313781738281, -5.160099029541016, 16.705291748046875, 33.18231201171875, 149.07391357421875, 154.29931640625, 158.6181182861328, 170.27359008789062, 17.4930419921875, 166.44390869140625, -9.620231628417969, 88.20594787597656, 81.10321044921875, 54.920650482177734, 35.079986572265625, 62.91748046875, 185.47650146484375, -192.426025390625, 8.982383728027344, -7.538360595703125, 2.6771316528320312, 144.47067260742188, 14.5089111328125, 8.434486389160156, 156.29576110839844, 15.035903930664062, 15.650009155273438, -131.04074096679688, 35.591712951660156, 150.39840698242188, -9.346076965332031, 10.834671020507812, 125.1043930053711, -8.78521728515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000370.npy"}
{"epoch": 0.774869109947644, "step": 371, "batch_size": 128, "mean": 51.375701904296875, "std": 82.31646728515625, "min": -206.78408813476562, "p10": -35.93103332519531, "median": 33.267791748046875, "p90": 160.6220520019531, "max": 198.08267211914062, "pos_frac": 0.734375, "sample": [-0.5560569763183594, 112.62307739257812, 50.28271484375, 142.58035278320312, -12.258636474609375, 0.29925537109375, 181.58172607421875, 5.07257080078125, 37.99409484863281, 19.12847900390625, 1.22235107421875, -190.80877685546875, 8.62799072265625, 12.51104736328125, 28.439590454101562, -10.87158203125, -50.3369140625, 96.68751525878906, 8.667449951171875, -21.00140380859375, -25.09130859375, 113.92355346679688, -9.654472351074219, 164.82815551757812, 78.23921203613281, 108.84469604492188, -0.3132343292236328, 42.78662109375, -23.922622680664062, 39.070953369140625, 132.12722778320312, 162.5062713623047, -72.13481140136719, 19.672073364257812, -206.78408813476562, 69.55010986328125, -38.3795166015625, 0.512298583984375, 124.725341796875, 23.7496337890625, 178.87205505371094, 12.642135620117188, -90.86842346191406, -1.99908447265625, 40.1553955078125, 8.614837646484375, 14.5611572265625, 0.0, -34.881683349609375, 8.234855651855469, 172.15957641601562, 148.08641052246094, 111.8826904296875, -72.6319580078125, -128.23980712890625, -5.0919036865234375, 30.886672973632812, 174.65484619140625, 148.62008666992188, 12.001483917236328, 15.463844299316406, 105.63446044921875, 73.18418884277344, 190.87142944335938, 37.77496337890625, 88.8203125, 109.63955688476562, 142.25592041015625, 161.59954833984375, -21.043746948242188, 26.744441986083984, 149.24822998046875, 124.85187530517578, -89.77557373046875, 115.87603759765625, 133.615234375, 123.61827087402344, 33.37370300292969, 161.8255615234375, 19.120254516601562, -43.47662353515625, -138.78720092773438, 18.248779296875, 197.47384643554688, 111.42607116699219, -19.24493408203125, -105.5837631225586, 134.4434814453125, 136.32220458984375, 45.154266357421875, 0.882904052734375, -18.52520751953125, 74.82711791992188, 108.15187072753906, 33.16188049316406, -2.1064529418945312, 141.61624145507812, 67.38214111328125, 22.542701721191406, 197.13705444335938, 27.524267196655273, -1.8746166229248047, 87.83981323242188, 80.61550903320312, 139.6038055419922, 138.6251220703125, -0.0367279052734375, 32.07659912109375, 32.188499450683594, 131.49447631835938, 138.19039916992188, -40.1561279296875, 47.725830078125, 61.36188507080078, 12.313690185546875, 145.98822021484375, 160.203125, 25.368154525756836, 198.08267211914062, 145.80157470703125, -0.2720947265625, -24.69415283203125, 31.881752014160156, -3.2572174072265625, 70.38304901123047, 139.412109375, 112.80052947998047, 171.3540802001953], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000371.npy"}
{"epoch": 0.7769633507853403, "step": 372, "batch_size": 128, "mean": 66.03838348388672, "std": 84.65963745117188, "min": -158.71051025390625, "p10": -17.58841400146484, "median": 53.74704170227051, "p90": 172.3906036376953, "max": 236.05072021484375, "pos_frac": 0.7890625, "sample": [176.517578125, -5.907989501953125, 133.70794677734375, 53.2534065246582, -1.8802604675292969, 135.60987854003906, 121.81198120117188, -6.4459228515625, 121.68842315673828, 165.3649444580078, 42.505062103271484, 60.03233337402344, 130.0353240966797, 132.17208862304688, -23.253570556640625, 122.15338134765625, 14.98678970336914, 140.4130859375, -9.989473342895508, 18.609703063964844, 75.79629516601562, 41.68580627441406, 113.56222534179688, 0.8402099609375, -11.6820068359375, 117.28924560546875, 197.93524169921875, 92.88563537597656, 148.01524353027344, 129.82620239257812, -28.71868896484375, -19.197113037109375, 177.82626342773438, -10.7783203125, 82.67208862304688, 119.63917541503906, 185.022216796875, 5.6601104736328125, 93.77112579345703, 56.51805877685547, -16.898971557617188, 175.8624267578125, 95.19214630126953, 128.2767333984375, 186.89105224609375, -157.00421142578125, 194.853271484375, 84.79367065429688, 11.887130737304688, 33.80364990234375, 16.953880310058594, 6.3858489990234375, 125.53968811035156, 1.05596923828125, 20.295684814453125, 30.31231689453125, 109.16107177734375, 151.50961303710938, 122.7113037109375, 36.626617431640625, 113.15760803222656, 31.867355346679688, 127.43649291992188, 18.943756103515625, 5.447605133056641, 54.24067687988281, 150.56430053710938, 24.0321044921875, 165.82302856445312, -11.895729064941406, 1.60076904296875, 8.48370361328125, 15.419647216796875, 153.94873046875, 171.43679809570312, 188.9788818359375, -111.31640625, 179.48724365234375, 22.20953369140625, -147.33396911621094, 148.65293884277344, 90.82958984375, 38.33404541015625, 55.757568359375, -136.33724975585938, 50.66023254394531, 52.996429443359375, 15.251022338867188, -6.133991241455078, 163.46421813964844, -11.921295166015625, -4.22552490234375, 152.32272338867188, 104.97634887695312, 39.9893798828125, -158.71051025390625, -60.4224853515625, 36.10406494140625, 124.77804565429688, -62.29737091064453, 159.39593505859375, -5.5831146240234375, -51.27252197265625, 4.9286041259765625, 169.83456420898438, 139.33792114257812, -154.4228515625, 174.61614990234375, 236.05072021484375, 184.341796875, 50.154052734375, 202.76763916015625, 164.00885009765625, 3.0397109985351562, 52.2972412109375, 166.8636474609375, -1.774301528930664, 119.1025390625, -38.2071533203125, 144.0843505859375, 109.93988037109375, 139.31304931640625, 41.78896713256836, 2.5214385986328125, -2.8740806579589844, 117.88632202148438, 47.52362060546875, 32.487335205078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000372.npy"}
{"epoch": 0.7790575916230367, "step": 373, "batch_size": 128, "mean": 61.36882781982422, "std": 85.51822662353516, "min": -201.78082275390625, "p10": -33.81963920593261, "median": 45.468048095703125, "p90": 176.8507568359375, "max": 200.59146118164062, "pos_frac": 0.7890625, "sample": [11.36175537109375, 41.839332580566406, 161.93637084960938, 29.074478149414062, 0.3709869384765625, -19.21649169921875, 17.4962158203125, 189.52713012695312, 15.7252197265625, 181.150146484375, 147.91543579101562, -15.8311767578125, 17.633285522460938, 157.2958221435547, 105.810302734375, 91.02584838867188, 1.447113037109375, 59.6962890625, -36.28216552734375, 154.7659912109375, 63.781097412109375, 1.0896644592285156, 111.98652648925781, 54.751953125, -42.34423828125, 44.699798583984375, -5.344718933105469, -37.35871887207031, 36.301849365234375, 189.67437744140625, 3.2768173217773438, -24.3236083984375, 26.886962890625, 47.682342529296875, 54.8048095703125, 123.6929931640625, 152.54840087890625, -18.79644775390625, 119.2381591796875, 77.117919921875, 85.80303955078125, 143.39364624023438, 170.74942016601562, 97.04044342041016, 13.93035888671875, 173.47305297851562, 33.416015625, -81.505615234375, 46.236297607421875, 179.46664428710938, -32.7642707824707, 156.57861328125, -92.68561553955078, -39.80804443359375, 77.12811279296875, -53.731536865234375, 191.2621612548828, -9.9691162109375, 28.001434326171875, 49.19903564453125, 5.2007293701171875, -201.78082275390625, 192.92715454101562, 31.556243896484375, 159.27999877929688, 15.98712158203125, -117.88272094726562, 164.6021270751953, 13.422286987304688, 128.8666534423828, 118.71685791015625, 161.084716796875, 11.608154296875, 84.20938110351562, 23.972503662109375, 56.26268005371094, 58.7376708984375, 1.1923828125, 89.17886352539062, 5.480133056640625, -145.28237915039062, -15.175247192382812, 186.73959350585938, 43.653785705566406, 40.411712646484375, 163.29104614257812, 160.58853149414062, 189.30535888671875, 197.35943603515625, 107.00546264648438, 4.9239501953125, 170.70765686035156, -2.09649658203125, 20.651702880859375, 164.9300079345703, 162.0765380859375, 21.27734375, 186.4053955078125, -79.43621826171875, -138.00770568847656, 82.69528198242188, 153.61465454101562, 200.59146118164062, 155.55194091796875, 57.902061462402344, 13.1256103515625, 129.8672332763672, -11.7757568359375, 197.10128784179688, 17.125640869140625, 13.967639923095703, 14.637985229492188, 22.5655517578125, 98.81494140625, -20.543170928955078, -5.63665771484375, 137.55909729003906, 6.629486083984375, -40.59637451171875, 116.34722900390625, -3.229341506958008, 176.50421142578125, -15.82675552368164, 7.935791015625, 165.14779663085938, 177.65936279296875, 125.34882354736328, 12.851104736328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000373.npy"}
{"epoch": 0.7811518324607329, "step": 374, "batch_size": 128, "mean": 41.45066833496094, "std": 86.9827651977539, "min": -179.98162841796875, "p10": -72.98711547851562, "median": 20.527206420898438, "p90": 162.7979766845703, "max": 228.64111328125, "pos_frac": 0.71875, "sample": [216.96820068359375, 0.0, 6.00457763671875, 211.84844970703125, 6.9356689453125, -134.74185180664062, -9.032936096191406, 23.709442138671875, 158.2231903076172, 121.08328247070312, 18.916353225708008, 200.66622924804688, 18.299530029296875, -79.28729248046875, 76.31797790527344, 35.0421142578125, 79.72103118896484, 129.114990234375, 17.35174560546875, 12.000625610351562, -125.62690734863281, -52.526214599609375, 100.27283477783203, -109.7364501953125, 158.20179748535156, 27.58643341064453, 18.447296142578125, 10.74505615234375, 7.069053649902344, -21.413925170898438, 4.8157958984375, 8.376724243164062, -85.43283081054688, 63.79777526855469, -8.50616455078125, 18.953125, 20.0609130859375, -14.128875732421875, 180.82308959960938, 138.37782287597656, 51.79132080078125, 10.995574951171875, 228.64111328125, 1.0687484741210938, 62.151031494140625, 0.312591552734375, 10.027114868164062, 66.8128662109375, -179.98162841796875, -17.32671356201172, 39.047271728515625, -45.1571044921875, 166.02191162109375, 95.1443862915039, 25.39935302734375, 48.188934326171875, -2.3790283203125, 65.45947265625, 81.71907043457031, -0.1398468017578125, 19.919151306152344, 182.3804931640625, 184.91607666015625, -39.865570068359375, -0.8658905029296875, 68.10930633544922, 52.126220703125, 38.66455078125, -32.03570556640625, 120.17866516113281, 19.4537353515625, 55.9967041015625, 132.24606323242188, 119.43769836425781, -164.39785766601562, 119.44120788574219, -71.52679443359375, 10.8255615234375, 22.9796142578125, 144.7975616455078, 117.06446075439453, 9.336334228515625, 55.32403564453125, 147.81268310546875, 0.0, 148.10421752929688, 7.695011138916016, -76.39453125, -37.136962890625, -112.47705078125, 5.169944763183594, 136.8880615234375, 137.9400634765625, 169.4689178466797, 20.48858642578125, -17.6650390625, 198.25180053710938, 20.565826416015625, 182.21466064453125, 57.73309326171875, 17.680419921875, -61.177425384521484, -99.0242919921875, 27.89569091796875, 18.578001022338867, 3.7139244079589844, 33.07452392578125, 185.11819458007812, -16.134620666503906, -118.65562438964844, -0.70758056640625, 105.67926025390625, 50.623138427734375, -3.2549591064453125, 161.41629028320312, -2.57794189453125, 143.36968994140625, 87.73851013183594, -79.03826904296875, 107.17343139648438, 211.63217163085938, 142.95437622070312, 147.49337768554688, 22.938705444335938, 33.17724609375, -3.044147491455078, 11.667831420898438, -132.88369750976562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000374.npy"}
{"epoch": 0.7832460732984293, "step": 375, "batch_size": 128, "mean": 51.64662170410156, "std": 86.63146209716797, "min": -178.96600341796875, "p10": -32.80259857177734, "median": 24.516189575195312, "p90": 163.8564208984375, "max": 249.80987548828125, "pos_frac": 0.75, "sample": [8.368545532226562, 73.99232482910156, 102.96719360351562, 8.845428466796875, 62.7474365234375, 4.2249603271484375, 21.766021728515625, -116.29438781738281, 6.255401611328125, 36.171722412109375, 178.3797607421875, 0.26220703125, 158.62277221679688, 163.06985473632812, 87.79428100585938, -10.1043701171875, 72.55911254882812, 112.20893859863281, 6.758970260620117, 7.990592956542969, 154.09371948242188, 163.582275390625, 11.879989624023438, 59.7659912109375, 56.78606414794922, 154.3938446044922, -14.922080993652344, 9.09002685546875, -59.47881317138672, 115.80561065673828, 16.19896697998047, 19.21542739868164, -4.322147369384766, -97.32110595703125, 13.008811950683594, 162.29205322265625, 111.13284301757812, 164.49609375, 21.82585906982422, -21.4156494140625, 23.799667358398438, 11.465476989746094, -123.02801513671875, 1.173675537109375, -178.96600341796875, 124.72523498535156, 148.60955810546875, 147.75775146484375, 17.41858673095703, -140.40457153320312, -133.5293426513672, -26.3321533203125, 118.801025390625, -0.3340740203857422, 27.386993408203125, -3.7437744140625, 136.68710327148438, -10.87125015258789, -31.731903076171875, -121.8116455078125, 174.53466796875, 161.27090454101562, -11.68414306640625, 211.11544799804688, 55.62468338012695, 22.3946533203125, -65.99800109863281, 2.2516860961914062, 249.80987548828125, -7.862558364868164, 86.55279541015625, 12.888919830322266, 25.232711791992188, 78.15193939208984, 29.581695556640625, 180.61370849609375, 68.27182006835938, 8.244232177734375, 44.382965087890625, 10.299942016601562, 124.71804809570312, 116.29214477539062, -6.77557373046875, -35.30088806152344, -2.3537139892578125, 119.08932495117188, -11.391693115234375, 126.25006103515625, 119.63113403320312, 151.8826904296875, -143.36904907226562, 168.6052703857422, 201.8419189453125, -12.468490600585938, 205.486328125, 216.415771484375, -45.349639892578125, 163.09506225585938, 93.58842468261719, 167.314697265625, -3.1978721618652344, -56.86748504638672, 162.60720825195312, -1.8774871826171875, 143.7374267578125, 2.238100051879883, 118.6298828125, 168.682861328125, 47.217529296875, 3.39007568359375, -3.7137908935546875, 174.21066284179688, 34.634918212890625, 41.63972473144531, 156.70050048828125, -29.424087524414062, 3.130147933959961, 161.0670928955078, 1.642486572265625, 19.15484619140625, 6.77581787109375, 17.70374298095703, 46.997222900390625, 67.24078369140625, 14.562530517578125, 139.74400329589844, 8.754791259765625, 102.7393798828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000375.npy"}
{"epoch": 0.7853403141361257, "step": 376, "batch_size": 128, "mean": 50.59193420410156, "std": 89.4982681274414, "min": -156.7343292236328, "p10": -56.499812316894534, "median": 37.986724853515625, "p90": 164.74982910156248, "max": 255.296630859375, "pos_frac": 0.71875, "sample": [89.69544982910156, 50.839385986328125, 166.56396484375, 255.296630859375, 122.82197570800781, 220.3778076171875, 106.39706420898438, 27.231014251708984, 51.08424377441406, -120.46527099609375, 144.5703125, -24.337112426757812, -135.54791259765625, 36.687347412109375, -20.772857666015625, 69.86854553222656, 3.6893539428710938, 44.35955810546875, 126.18878173828125, 167.17437744140625, 102.44821166992188, -147.72976684570312, 0.7408065795898438, 12.105072021484375, 69.55203247070312, 26.236446380615234, -27.041854858398438, 177.78900146484375, -3.3099212646484375, 148.86212158203125, 147.05758666992188, 48.78175354003906, 166.10137939453125, 45.432586669921875, -13.159576416015625, -17.55194091796875, 164.17059326171875, 94.06678009033203, 163.50445556640625, 10.640594482421875, 0.0, 19.156631469726562, 18.406265258789062, 118.87678527832031, -56.37892150878906, 186.9927978515625, -89.5833740234375, 141.66262817382812, 105.67807006835938, 130.29208374023438, -141.24874877929688, 110.146484375, 44.28550720214844, 8.5001220703125, 151.08102416992188, 9.296600341796875, 25.232513427734375, 103.79719543457031, -52.09843444824219, -17.93682861328125, 113.49032592773438, -3.192962646484375, 209.29067993164062, 199.28170776367188, 23.785888671875, 183.971435546875, 159.3243408203125, 54.14848709106445, -17.945526123046875, -28.11444091796875, 80.4346923828125, -56.781890869140625, 27.260101318359375, -25.873313903808594, -3.5128211975097656, 10.388671875, -120.3253173828125, 4.949226379394531, -79.97293090820312, 170.37423706054688, 124.66139221191406, 56.6031494140625, 160.20913696289062, 123.01422119140625, 155.15460205078125, 127.79856872558594, 7.240678787231445, -156.7343292236328, 11.689117431640625, -98.00190734863281, 49.642822265625, 3.0511131286621094, 144.52294921875, -22.025245666503906, 8.003776550292969, 132.5164031982422, 29.528472900390625, -13.806884765625, -116.52496337890625, 158.39508056640625, 4.268096923828125, 150.323974609375, -95.32611083984375, 138.70831298828125, 15.88818359375, 23.572542190551758, 103.29856872558594, 34.91523742675781, 57.36628723144531, 141.4625244140625, -16.25799560546875, -42.2105712890625, -7.237743377685547, 38.44134521484375, 101.00473022460938, -5.42706298828125, 34.5894775390625, -36.86792755126953, 52.209869384765625, -1.9068679809570312, 5.5033111572265625, 72.55760192871094, 119.08462524414062, 37.5321044921875, 176.37692260742188, -130.26675415039062, 190.04287719726562, 161.6260986328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000376.npy"}
{"epoch": 0.787434554973822, "step": 377, "batch_size": 128, "mean": 47.01289749145508, "std": 81.90528106689453, "min": -153.95379638671875, "p10": -36.06289672851562, "median": 24.674407958984375, "p90": 167.47928771972653, "max": 213.4410400390625, "pos_frac": 0.6953125, "sample": [178.255859375, 77.39152526855469, 130.2260284423828, -14.137313842773438, 95.97723388671875, 12.248794555664062, 10.332855224609375, 115.83502197265625, 141.01475524902344, 9.82177734375, 21.588455200195312, 37.7073974609375, 132.37991333007812, -52.53485107421875, 116.58938598632812, 124.30088806152344, 173.8687744140625, -29.5599365234375, 1.0947914123535156, -39.2708740234375, 135.59609985351562, 1.643402099609375, -0.38062286376953125, 29.727977752685547, -13.928647994995117, -14.617851257324219, 65.53829956054688, -12.780715942382812, 165.28819274902344, 40.30574035644531, 112.43805694580078, 21.04107666015625, 112.92237854003906, -102.72906494140625, -9.217391967773438, 150.26144409179688, -2.893321990966797, -5.99658203125, -33.70440673828125, 41.798614501953125, 16.90753173828125, 63.59321212768555, 12.827049255371094, -65.2896728515625, 132.25601196289062, 8.76983642578125, 187.53759765625, 62.00531005859375, 8.377914428710938, -101.40869140625, 8.28277587890625, 197.52786254882812, -6.0318450927734375, 0.19378662109375, 175.2081298828125, 5.519842147827148, -77.34764099121094, -1.7930335998535156, 138.68894958496094, 134.89291381835938, 156.5885009765625, 28.40673828125, 74.85446166992188, 90.9686279296875, -24.7509765625, 144.461181640625, 83.27424621582031, 0.0, -34.68804931640625, 69.65119171142578, 139.01812744140625, -5.4550628662109375, 120.11460876464844, 139.82911682128906, -6.597099304199219, 46.013092041015625, 9.083454132080078, -33.737213134765625, -0.8622150421142578, 136.4091339111328, 18.35015869140625, 206.44314575195312, 9.1458740234375, -26.226486206054688, -15.6168212890625, -92.39188385009766, 25.423812866210938, 12.93426513671875, 163.44842529296875, 40.095428466796875, 56.077362060546875, 29.469390869140625, 8.779029846191406, 153.34957885742188, -48.261962890625, 175.78289794921875, 11.656082153320312, 23.925003051757812, 45.217803955078125, 132.54385375976562, 34.881439208984375, -111.93150329589844, -11.7486572265625, 91.8038330078125, 181.53848266601562, -72.7096176147461, 195.46380615234375, 35.0751953125, 69.54705810546875, 146.27163696289062, 20.7347412109375, 10.60293960571289, 25.976699829101562, 6.329833984375, 179.47247314453125, 137.926025390625, 172.5918426513672, -130.363037109375, -13.56866455078125, 213.4410400390625, 23.251304626464844, -10.478271484375, -117.48164367675781, 177.814453125, -12.8309326171875, -10.119110107421875, -153.95379638671875, 53.2254638671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000377.npy"}
{"epoch": 0.7895287958115184, "step": 378, "batch_size": 128, "mean": 47.39453887939453, "std": 83.03558349609375, "min": -157.04095458984375, "p10": -41.53191223144531, "median": 27.139877319335938, "p90": 166.42931060791014, "max": 204.90255737304688, "pos_frac": 0.703125, "sample": [132.06832885742188, 5.48956298828125, 26.00677490234375, 125.47525024414062, -23.562957763671875, 5.844053268432617, -27.7568359375, 7.7119140625, 174.2041015625, 51.63543701171875, -20.561233520507812, 7.6015472412109375, 28.272979736328125, 130.94540405273438, 131.4456787109375, -46.927154541015625, -157.04095458984375, 174.30484008789062, 189.05072021484375, -27.8089599609375, 38.999908447265625, 42.2381591796875, 12.962966918945312, -42.754669189453125, 95.33566284179688, 179.68231201171875, 204.90255737304688, 144.65176391601562, 22.8001708984375, 121.59878540039062, 3.2271270751953125, 137.63902282714844, 114.76057434082031, 24.1204833984375, 11.895263671875, -138.673828125, 2.2534122467041016, 2.6856155395507812, 81.80938720703125, 89.8748779296875, 0.0, 155.50546264648438, -5.988006591796875, 130.8597412109375, 181.96929931640625, -48.0623779296875, -0.19527816772460938, 198.94903564453125, -15.224491119384766, 175.996337890625, 103.65213012695312, 113.98226928710938, -17.382659912109375, 5.7950439453125, 7.259796142578125, 3.2723236083984375, -32.37249755859375, 135.40554809570312, -18.920684814453125, 140.93569946289062, 68.31744384765625, -1.1974315643310547, 131.82785034179688, 142.41998291015625, 37.799530029296875, 121.64816284179688, -41.00787353515625, -59.795074462890625, 168.24661254882812, 84.29362487792969, 90.14755249023438, 34.079620361328125, 0.254638671875, -28.579116821289062, -117.1751708984375, 164.30780029296875, -56.65299987792969, -84.42803955078125, 45.01300048828125, -1.2935028076171875, 31.866989135742188, 6.0759735107421875, 91.3951416015625, 101.77243041992188, 8.71893310546875, 100.368408203125, 178.478759765625, 8.270599365234375, 17.9703369140625, 75.95175170898438, 151.55447387695312, 151.43544006347656, -22.6365966796875, -0.720184326171875, 52.30499267578125, -25.20849609375, -72.47357177734375, 8.146011352539062, 12.56707763671875, -20.247299194335938, -94.3623046875, 174.6046142578125, -7.27447509765625, 56.72377014160156, -139.5474395751953, -3.857147216796875, 25.95306396484375, 121.35295867919922, -6.86114501953125, 33.22027587890625, 42.564910888671875, 45.342926025390625, 1.128692626953125, -12.368789672851562, 193.6541748046875, 152.1673583984375, 164.36541748046875, 165.6504669189453, 175.95098876953125, 20.448379516601562, -142.03936767578125, 108.8818588256836, 129.54800415039062, 49.380615234375, 31.338699340820312, -38.582122802734375, 16.353668212890625, -2.89971923828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000378.npy"}
{"epoch": 0.7916230366492146, "step": 379, "batch_size": 128, "mean": 50.26569747924805, "std": 88.74102783203125, "min": -172.78863525390625, "p10": -54.479177093505854, "median": 32.60772705078125, "p90": 167.2153060913086, "max": 202.59149169921875, "pos_frac": 0.7421875, "sample": [-20.002029418945312, 10.965900421142578, 3.65777587890625, 7.8524627685546875, -1.8851699829101562, -61.375946044921875, -2.9121322631835938, 147.68646240234375, 63.63694763183594, 1.3973922729492188, 135.70333862304688, 132.33578491210938, 38.94658660888672, 36.26837158203125, 13.813385009765625, 4.7391815185546875, 105.0770034790039, -118.74311828613281, 180.23587036132812, 30.77337646484375, 96.67465209960938, 91.17059326171875, 155.10031127929688, -35.0528564453125, 10.742225646972656, -53.638885498046875, 151.8188018798828, 2.32666015625, 68.61553955078125, -9.2493896484375, 32.81365966796875, 167.13514709472656, 169.14743041992188, 51.844818115234375, 58.17706298828125, 9.300933837890625, 52.7896728515625, 178.73980712890625, 121.50245666503906, 8.362815856933594, -7.8341064453125, -78.07696533203125, 35.731201171875, 17.47234344482422, 7.494903564453125, 194.49752807617188, 9.673385620117188, -66.79916381835938, -16.770904541015625, 7.16748046875, 135.88897705078125, 6.426207542419434, 32.40179443359375, -146.7244873046875, -4.2679443359375, 10.47259521484375, 161.97808837890625, 147.24444580078125, 125.8160400390625, 130.15283203125, 48.55908203125, -56.439857482910156, 119.39254760742188, 88.93952941894531, 136.12435913085938, 202.59149169921875, 16.9171142578125, 168.631103515625, 11.329132080078125, -24.933944702148438, -41.586090087890625, 0.0, -7.7357177734375, 79.14637756347656, 18.092864990234375, 176.73287963867188, 56.66546630859375, 153.4873046875, 169.40902709960938, 158.88648986816406, 10.80621337890625, 124.16214752197266, 18.760238647460938, 10.976753234863281, 146.5357666015625, 126.54727935791016, -26.127418518066406, 121.11639404296875, -22.898880004882812, 134.7944793701172, 93.74398803710938, -18.00853729248047, 177.7113037109375, 2.729856491088867, 116.7733154296875, -11.51373291015625, 2.3939590454101562, 126.94412231445312, -125.59075164794922, 198.38177490234375, 197.89553833007812, -172.78863525390625, 19.868377685546875, 54.22792053222656, 162.35528564453125, 79.03372192382812, 61.658660888671875, -47.6444091796875, 157.14242553710938, 159.57525634765625, 178.27157592773438, 106.35113525390625, 161.90634155273438, 72.78691101074219, -117.39029693603516, -136.7511444091797, 32.377479553222656, 4.830371856689453, 3.374847412109375, 167.40234375, -44.13671875, 119.31389617919922, -145.51922607421875, -95.22653198242188, -146.53143310546875, 23.76447868347168, -9.751129150390625, 166.7279052734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000379.npy"}
{"epoch": 0.793717277486911, "step": 380, "batch_size": 128, "mean": 47.89772033691406, "std": 85.95724487304688, "min": -136.9357147216797, "p10": -51.12885665893553, "median": 27.123371124267578, "p90": 169.2028839111328, "max": 253.03338623046875, "pos_frac": 0.71875, "sample": [157.22482299804688, 46.58403015136719, 96.525634765625, 19.040237426757812, 131.2474365234375, 18.890960693359375, -129.5422821044922, 3.074554443359375, 1.0493316650390625, -78.14486694335938, 110.29949951171875, 52.30169677734375, 145.9786376953125, -116.21595001220703, 25.808349609375, 14.108131408691406, 107.54718017578125, 44.7745361328125, 141.10372924804688, -3.8979034423828125, 0.0, 47.81121826171875, 8.411941528320312, 25.3994140625, 26.46990966796875, 153.55804443359375, 6.404632568359375, 31.23114013671875, -0.9675140380859375, 8.541107177734375, 49.93463134765625, 206.368896484375, 12.918472290039062, 130.93115234375, 0.201263427734375, -134.2734375, 124.18161010742188, -4.394886016845703, 26.83252716064453, -20.396820068359375, 45.29290771484375, 60.64360046386719, 166.29165649414062, -41.792633056640625, -17.513526916503906, 14.146697998046875, 15.5537109375, 175.7630615234375, 175.04669189453125, -12.751129150390625, -73.2972412109375, 109.25253295898438, 47.018287658691406, -3.510528564453125, 22.523193359375, -81.20603942871094, -27.125431060791016, 14.78692626953125, 175.94500732421875, 139.73358154296875, -62.3966064453125, 157.08921813964844, 74.987548828125, 30.549407958984375, 176.858642578125, 162.21701049804688, -114.39713287353516, -3.8402862548828125, 17.58319091796875, -1.9496040344238281, 136.07366943359375, 181.90249633789062, 2.3404388427734375, 114.14209747314453, 155.13917541503906, -3.7063751220703125, 45.990966796875, -40.21484375, 34.0914306640625, -136.9357147216797, 116.63092041015625, 253.03338623046875, -9.048027038574219, 132.6238555908203, 155.42889404296875, 30.16876220703125, 146.88327026367188, 166.69839477539062, 163.210205078125, 56.24494171142578, 35.19073486328125, 215.46157836914062, -68.57806396484375, 52.12811279296875, 3.3796844482421875, 66.05815887451172, 3.15130615234375, -41.00970458984375, 179.72100830078125, 238.21542358398438, 34.165016174316406, 8.958221435546875, 14.276657104492188, 159.1055908203125, 55.969818115234375, 16.99602508544922, 27.414215087890625, 45.845062255859375, -33.54759216308594, -47.021080017089844, -89.566650390625, -60.71366882324219, 9.8829345703125, 12.270259857177734, 25.816879272460938, 116.810302734375, -21.76776123046875, 57.52044677734375, 102.39599609375, 224.8319091796875, 29.134033203125, -16.59881591796875, -35.11181640625, 178.07183837890625, -0.3426513671875, -89.99409484863281, 198.43734741210938, -5.1704254150390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000380.npy"}
{"epoch": 0.7958115183246073, "step": 381, "batch_size": 128, "mean": 56.14626693725586, "std": 80.62554168701172, "min": -113.50221252441406, "p10": -41.23912353515624, "median": 43.763328552246094, "p90": 164.34361572265624, "max": 186.60049438476562, "pos_frac": 0.7109375, "sample": [133.81741333007812, 184.321533203125, 167.02188110351562, 53.52129364013672, 183.1750946044922, -46.915985107421875, 166.2925262451172, 179.60397338867188, 119.73153686523438, 5.730781555175781, 181.20848083496094, 19.318862915039062, 14.611663818359375, 143.29302978515625, 135.73062133789062, 140.4256591796875, 0.0, 12.775146484375, 47.510406494140625, 95.13613891601562, 109.82389831542969, 58.5478515625, 6.7119903564453125, 186.60049438476562, 47.80352783203125, 59.55091857910156, 136.38888549804688, -113.50221252441406, 0.0, -27.8233642578125, 74.6455078125, 123.16197967529297, -10.27880859375, 144.9821014404297, -8.183504104614258, 25.005096435546875, 16.511505126953125, 73.76904296875, 12.12286376953125, 113.9088134765625, 184.9700927734375, 81.14183044433594, -17.799407958984375, 132.18031311035156, 129.6497802734375, 6.48358154296875, 164.075439453125, -11.542236328125, -84.87419128417969, 147.54881286621094, 147.58583068847656, 159.56341552734375, 43.78153991699219, 1.0194854736328125, 179.34176635742188, 162.1339111328125, 11.2564697265625, 125.19601440429688, 164.9693603515625, 88.83712768554688, -47.196197509765625, 43.11305236816406, 77.60255432128906, 11.175628662109375, -7.594329833984375, 37.69276428222656, 152.4432373046875, -2.860107421875, -26.412841796875, -70.0135498046875, 16.442161560058594, -13.46238899230957, -4.067047119140625, 149.846923828125, -20.32318115234375, -38.806182861328125, -0.8404388427734375, 104.50074768066406, 55.03547668457031, -93.23760986328125, 159.916259765625, -28.821029663085938, 144.83445739746094, 43.7451171875, 178.17947387695312, -3.24261474609375, -25.72747802734375, -8.101051330566406, 157.62779235839844, 13.954010009765625, -4.05194091796875, -53.025970458984375, 137.20401000976562, 124.26988220214844, 113.82138061523438, -49.741790771484375, 3.4082298278808594, 7.6910400390625, 36.52516174316406, -106.59283447265625, 108.8381576538086, 9.749137878417969, -97.10342407226562, 82.2373275756836, -111.12509155273438, 119.47163391113281, -22.0760498046875, 7.554494857788086, -18.375076293945312, 133.29617309570312, -8.248748779296875, 177.43431091308594, -65.22943115234375, 185.41253662109375, 7.9501953125, 48.04755401611328, 61.628379821777344, 8.561408996582031, 131.515869140625, 18.64947509765625, 152.060546875, 155.004150390625, 34.736106872558594, -26.83544921875, -62.572235107421875, 115.20584106445312, 23.3155517578125, 71.13247680664062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000381.npy"}
{"epoch": 0.7979057591623037, "step": 382, "batch_size": 128, "mean": 68.8196792602539, "std": 85.23173522949219, "min": -129.58688354492188, "p10": -34.466943359375, "median": 67.9473876953125, "p90": 170.35150451660155, "max": 211.32275390625, "pos_frac": 0.7734375, "sample": [-54.05558776855469, 22.980812072753906, 59.196502685546875, 55.9736328125, 99.56439208984375, 154.02462768554688, 157.2021484375, 5.313232421875, 16.199819564819336, 121.24070739746094, 130.56149291992188, -128.90863037109375, 111.54460144042969, 15.095405578613281, 175.39349365234375, 112.34091186523438, 52.83026123046875, 69.27033996582031, 137.28521728515625, -13.019599914550781, 49.702880859375, -127.62852478027344, 157.53858947753906, 5.6883544921875, 3.8869171142578125, 179.24185180664062, 40.7001953125, -33.9068603515625, 135.32553100585938, 109.47730255126953, -11.2113037109375, 130.12806701660156, 117.22866821289062, 33.886695861816406, 154.21188354492188, 96.03527069091797, -13.212104797363281, 63.468894958496094, -8.668493270874023, 162.13858032226562, 40.00796127319336, 14.26611328125, 164.79953002929688, 23.51165771484375, 160.83041381835938, 166.33486938476562, 175.9191436767578, 48.082733154296875, 26.1031494140625, 10.254051208496094, 168.47579956054688, 137.60255432128906, 8.906051635742188, 65.96942138671875, -8.857666015625, 114.54891967773438, 211.32275390625, 179.0506591796875, -81.25802612304688, -10.134765625, -35.7738037109375, -129.58688354492188, 209.69992065429688, 38.44285583496094, 2.8374786376953125, 104.21687316894531, 184.6519775390625, 68.71127319335938, 183.53216552734375, 29.0924072265625, 190.82833862304688, 23.829666137695312, 41.83952331542969, -12.1845703125, 164.16537475585938, 106.27415466308594, 134.74969482421875, 19.386886596679688, 123.43060302734375, 163.37396240234375, 37.13128662109375, -32.281036376953125, 56.2991943359375, 143.11424255371094, -107.77156066894531, -3.741729736328125, -54.65496826171875, 159.7374267578125, -20.210037231445312, 159.3365478515625, 174.7281494140625, -19.924224853515625, 11.272689819335938, 175.1313934326172, 164.60134887695312, 67.18350219726562, 124.15203094482422, -115.75749206542969, 7.304107666015625, -11.617462158203125, 37.68162536621094, 72.087158203125, 79.10026550292969, 153.7317657470703, 124.00650024414062, -68.40451049804688, 39.87324523925781, 144.6639404296875, -111.27462768554688, 136.5620574951172, 148.1602783203125, -8.831932067871094, 139.036376953125, 52.9317626953125, 81.21420288085938, 127.20646667480469, -55.819732666015625, 87.17654418945312, -18.610572814941406, 123.43873596191406, 164.13961791992188, -20.259075164794922, 201.02395629882812, 135.03541564941406, 176.1888427734375, 106.82757568359375, 164.9799041748047, -88.30010986328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000382.npy"}
{"epoch": 0.8, "step": 383, "batch_size": 128, "mean": 53.536170959472656, "std": 82.7784423828125, "min": -146.037353515625, "p10": -45.66500549316406, "median": 30.726585388183594, "p90": 163.7781005859375, "max": 209.42312622070312, "pos_frac": 0.7421875, "sample": [-3.4078369140625, 44.6512451171875, 144.01715087890625, 132.28994750976562, -32.27140808105469, 16.776519775390625, -52.383331298828125, 25.099456787109375, -1.5103759765625, 92.65292358398438, 134.321044921875, -128.85032653808594, 135.82937622070312, 31.3349609375, 0.0, -11.542671203613281, 27.68060302734375, -57.412139892578125, 51.218414306640625, -48.180267333984375, 19.06341552734375, 173.47882080078125, 145.3523712158203, 115.6295166015625, -120.48101806640625, 174.79820251464844, -96.94439697265625, -5.002216339111328, -55.029052734375, 9.15380859375, 73.156005859375, 8.269775390625, -15.004600524902344, -36.324073791503906, 150.88082885742188, 1.2176513671875, 188.71981811523438, 165.226806640625, 33.66484069824219, 142.5240936279297, 94.38345336914062, 6.207466125488281, 141.85305786132812, 209.42312622070312, -106.46183776855469, -2.6890411376953125, 44.83909606933594, 158.54180908203125, 107.4639892578125, 7.7247314453125, 163.1572265625, 3.71307373046875, 2.494993209838867, 143.17898559570312, 88.42929077148438, 20.811614990234375, 4.554180145263672, -12.772125244140625, 0.08508110046386719, -121.35147094726562, 131.8428497314453, 146.56298828125, 11.471162796020508, 168.12908935546875, -25.7415771484375, 11.876602172851562, 161.224365234375, 7.403350830078125, 9.798385620117188, 24.023468017578125, 104.15100860595703, 172.8349151611328, 130.40322875976562, 135.16778564453125, 116.50106811523438, 37.921443939208984, -2.24261474609375, 3.1759796142578125, 36.24192810058594, 135.92112731933594, 2.7742767333984375, 118.14095306396484, 120.73820495605469, 50.55027770996094, 107.0164794921875, 29.597824096679688, 29.16668701171875, 22.39834976196289, 30.118209838867188, -146.037353515625, 171.2174072265625, 27.43521499633789, 55.82051086425781, 51.27471160888672, 45.2362060546875, 5.34495735168457, -17.5396728515625, 187.6122589111328, 137.112060546875, 131.13223266601562, 140.43118286132812, -44.5870361328125, 209.2296142578125, -104.04780578613281, -69.10986328125, 23.566864013671875, 109.34793090820312, 20.762725830078125, 0.0, 127.27669525146484, 149.537353515625, -2.8557472229003906, 58.27642822265625, 118.1617431640625, 98.55148315429688, -25.739456176757812, -42.948577880859375, 16.31783676147461, 174.14813232421875, 114.319580078125, 8.813125610351562, 139.40280151367188, 193.67080688476562, -50.85131072998047, -20.422607421875, -4.1414642333984375, 166.99627685546875, 146.496826171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000383.npy"}
{"epoch": 0.8020942408376963, "step": 384, "batch_size": 128, "mean": 37.36378479003906, "std": 88.40168762207031, "min": -167.82647705078125, "p10": -66.73925018310547, "median": 22.51373291015625, "p90": 157.42357177734374, "max": 211.94537353515625, "pos_frac": 0.6484375, "sample": [189.90216064453125, -167.82647705078125, 107.568115234375, -21.8638916015625, -23.7685546875, 143.29702758789062, 44.63731384277344, 211.94537353515625, 22.955184936523438, -5.744239807128906, 146.08570861816406, -19.208385467529297, -56.96748352050781, -74.40621948242188, 34.687835693359375, -116.15145874023438, 149.97120666503906, -3.7431678771972656, 178.66769409179688, -20.048553466796875, 6.334405899047852, 166.20742797851562, 135.79925537109375, 134.30795288085938, -30.296493530273438, 152.04046630859375, 23.940841674804688, 131.23760986328125, 19.736083984375, -66.93743896484375, 3.2791881561279297, -59.256103515625, -2.043975830078125, 52.760284423828125, -27.016937255859375, 118.93184661865234, 22.491790771484375, 32.475341796875, 143.83822631835938, -15.312538146972656, 7.166259765625, -120.982666015625, -1.678314208984375, 16.50983428955078, 133.36044311523438, 147.46031188964844, 147.47100830078125, -158.67715454101562, 3.612823486328125, 20.85057830810547, 119.44194793701172, 182.22930908203125, -66.65431213378906, -2.40142822265625, -142.07669067382812, 41.83373260498047, 90.69894409179688, 79.68548583984375, -118.1546630859375, -24.97076416015625, 12.015029907226562, -120.9005126953125, 123.4315185546875, 132.056884765625, 169.96051025390625, 48.55511474609375, 45.0557861328125, 52.94903564453125, 70.24191284179688, -4.039424896240234, -10.63705062866211, 105.1851577758789, 157.82437133789062, 11.926254272460938, 123.10476684570312, 162.01455688476562, -11.876800537109375, 16.570083618164062, -11.98006820678711, 82.4107666015625, 81.53065490722656, -119.44551086425781, 130.46954345703125, 81.96377563476562, 145.06787109375, 201.86929321289062, 24.69091796875, 8.421356201171875, 81.52337646484375, 38.49493408203125, 173.90438842773438, -11.966392517089844, 144.36346435546875, 61.47625732421875, -5.8190765380859375, 3.3682403564453125, -39.71832275390625, 169.7776336669922, 12.439300537109375, -9.196769714355469, -6.41363525390625, -49.025421142578125, 39.43756103515625, -117.91796875, -135.3436279296875, 26.378293991088867, 0.3064422607421875, 29.097246170043945, 11.643320083618164, 38.298126220703125, 183.50067138671875, -136.07244873046875, 51.80047607421875, -4.602973937988281, -4.370058059692383, -11.690673828125, 169.73484802246094, 157.25180053710938, -66.41050720214844, 126.32136535644531, 137.29693603515625, -64.67013549804688, 2.4166412353515625, 15.419708251953125, 22.535675048828125, 68.20632934570312, 13.401153564453125, -58.278533935546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000384.npy"}
{"epoch": 0.8041884816753927, "step": 385, "batch_size": 128, "mean": 49.15149688720703, "std": 83.68626403808594, "min": -164.45355224609375, "p10": -47.22144775390625, "median": 30.75634765625, "p90": 163.86874694824218, "max": 224.53424072265625, "pos_frac": 0.734375, "sample": [-40.5804443359375, 64.44092559814453, 162.49209594726562, 131.72979736328125, 46.66796875, 28.74566650390625, 153.2017822265625, 31.703582763671875, 161.83245849609375, 140.5093994140625, 29.809112548828125, 1.32745361328125, -51.774078369140625, 21.069351196289062, 26.879287719726562, 75.90232849121094, 58.29302978515625, 124.34410858154297, -81.84606170654297, -8.59326171875, 167.0809326171875, -46.900665283203125, 141.62124633789062, -77.07315063476562, 10.72308349609375, 157.79542541503906, 133.27334594726562, 10.149726867675781, 128.86669921875, 23.217742919921875, 5.548736572265625, 195.75296020507812, 131.92205810546875, 34.7110595703125, 173.05349731445312, 224.53424072265625, 148.549560546875, 32.84661865234375, 108.74757385253906, 212.4743194580078, -5.7987060546875, -33.03575134277344, 53.66749572753906, 70.36634826660156, 57.32865905761719, -121.68695068359375, 135.0469207763672, 134.2006072998047, -131.4671630859375, 18.4000244140625, 209.59249877929688, 170.97317504882812, 214.18743896484375, -9.286949157714844, 125.08119201660156, 24.915725708007812, 52.004669189453125, 1.1569347381591797, -159.03695678710938, 67.1386947631836, 19.449920654296875, 182.0438232421875, -11.573089599609375, 157.57635498046875, 4.63470458984375, 2.888671875, -28.1380615234375, 134.25942993164062, -3.939208984375, 67.8543701171875, 107.81069946289062, 4.219490051269531, -26.056976318359375, 10.714698791503906, 20.657272338867188, -22.385574340820312, 104.47129821777344, 137.07290649414062, 129.26422119140625, -93.74044799804688, 12.7403564453125, 106.7757568359375, 138.71218872070312, 70.79026794433594, 121.5673828125, 13.226348876953125, -21.38690185546875, 26.704757690429688, -164.45355224609375, 4.167758941650391, 52.708038330078125, -5.317493438720703, -5.8971405029296875, 27.20301055908203, 170.9033203125, 50.24969482421875, 0.0, 38.70222473144531, 69.93304443359375, 26.36072540283203, 2.7907562255859375, 12.62640380859375, 47.646759033203125, -33.93975830078125, 184.16876220703125, -6.264183044433594, -104.05950927734375, 113.52873229980469, -33.58119201660156, 186.51373291015625, 44.62225341796875, 0.0, -47.969940185546875, -93.33991241455078, 127.59564208984375, 3.220458984375, 64.93588256835938, -64.95868682861328, -0.648712158203125, -9.446624755859375, -64.04551696777344, 13.733016967773438, 82.12032318115234, 36.888427734375, 126.26962280273438, 22.70208740234375, 167.89273071289062, 14.817962646484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000385.npy"}
{"epoch": 0.806282722513089, "step": 386, "batch_size": 128, "mean": 61.612159729003906, "std": 89.58456420898438, "min": -227.45330810546875, "p10": -22.28418807983398, "median": 47.46815490722656, "p90": 175.62186584472656, "max": 234.315673828125, "pos_frac": 0.7421875, "sample": [118.17153930664062, -3.517242431640625, 4.93707275390625, 128.44448852539062, 64.00384521484375, 26.0194091796875, 66.799072265625, 168.94464111328125, 121.10049438476562, 141.03086853027344, 140.13931274414062, 234.315673828125, -43.63836669921875, 35.6324462890625, 189.5423583984375, 40.08351135253906, 18.8360595703125, 31.33306884765625, -0.7947158813476562, 117.60104370117188, 110.697021484375, 166.24453735351562, 24.381439208984375, 57.5040283203125, -12.7841796875, -141.69094848632812, 39.905364990234375, -14.21551513671875, 144.06983947753906, 81.08529663085938, 33.76318359375, -117.70785522460938, 187.3519287109375, 33.397491455078125, -19.386444091796875, 16.333450317382812, 14.764175415039062, -27.617874145507812, 49.932037353515625, 167.0247802734375, 223.41851806640625, 170.51248168945312, 67.89430236816406, 36.776100158691406, 101.39779663085938, 125.84353637695312, 187.13565063476562, -73.92568969726562, 58.87849426269531, 202.28314208984375, 97.569580078125, -24.339385986328125, -109.68359375, -36.586875915527344, 192.353271484375, 135.28025817871094, -9.894233703613281, 0.0, -80.9241943359375, -21.40338897705078, -144.52163696289062, -93.1654052734375, 118.6893310546875, 29.236099243164062, 1.4370307922363281, -10.898956298828125, 51.61712646484375, -1.8927745819091797, 154.08506774902344, -177.05364990234375, 180.07635498046875, -21.01959228515625, 143.09109497070312, 197.0960693359375, 197.165771484375, 131.77438354492188, 6.381649017333984, -8.34515380859375, 160.28125, 143.8809051513672, 132.2606964111328, 10.98553466796875, -227.45330810546875, 49.87982177734375, -2.293792724609375, 167.79165649414062, 147.39468383789062, 15.7550048828125, 134.50225830078125, 1.4060821533203125, 48.752471923828125, 18.432830810546875, -1.17510986328125, 0.0, 193.50070190429688, 132.9757843017578, -8.94140625, 7.33526611328125, 62.81292724609375, 142.33450317382812, 156.88986206054688, 165.83335876464844, 34.044036865234375, 157.68565368652344, 13.086135864257812, 173.71279907226562, 29.888702392578125, -21.24810791015625, -3.21185302734375, 151.82452392578125, 180.54452514648438, 46.183837890625, 181.8452606201172, 68.23866271972656, 4.53204345703125, 28.516021728515625, 32.71568298339844, 15.591796875, 81.19015502929688, 142.19766235351562, 33.12416458129883, 172.13980102539062, 13.274356842041016, -4.469596862792969, 121.43817138671875, 0.0, 139.10000610351562, 52.895233154296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000386.npy"}
{"epoch": 0.8083769633507853, "step": 387, "batch_size": 128, "mean": 58.80384063720703, "std": 88.25665283203125, "min": -188.28070068359375, "p10": -39.05187988281249, "median": 45.35185241699219, "p90": 172.33487243652343, "max": 210.3709716796875, "pos_frac": 0.7109375, "sample": [54.38201904296875, 37.226715087890625, 136.8829345703125, 20.47006607055664, 7.23077392578125, 173.12820434570312, 192.4074249267578, -74.18792724609375, -143.70635986328125, -35.41326904296875, 0.0, 180.04049682617188, -11.700958251953125, 199.2060546875, 134.50656127929688, 141.00555419921875, 4.781074523925781, -14.215362548828125, -26.9364013671875, 97.48306274414062, -46.32746887207031, 33.44306182861328, 155.60655212402344, -44.02813720703125, 42.015419006347656, 163.26441955566406, -99.3966064453125, -22.7706298828125, 161.54571533203125, 0.7545204162597656, 152.22056579589844, -72.1734619140625, 160.80218505859375, 34.35673522949219, 147.01031494140625, 145.31875610351562, -15.72821044921875, 56.48857116699219, 59.995262145996094, -8.42156982421875, -37.465087890625, 134.45423889160156, 10.372650146484375, -11.059173583984375, 179.34304809570312, 30.466705322265625, 191.04052734375, 59.824798583984375, -5.990278244018555, 1.5412750244140625, -7.880165100097656, 177.88360595703125, -42.75439453125, -14.3970947265625, 63.15093994140625, 23.342010498046875, 11.640228271484375, -115.15571594238281, 143.00421142578125, 156.6186981201172, 116.42579650878906, 37.64227294921875, 65.0377197265625, 25.258453369140625, 137.033447265625, -122.83773803710938, 46.814971923828125, 36.340362548828125, 7.885467529296875, -12.12054443359375, -19.773696899414062, -20.216796875, 49.6800537109375, 181.43490600585938, 146.5519561767578, 124.9053955078125, 137.56809997558594, 38.753440856933594, -7.4345245361328125, 109.39816284179688, 13.025848388671875, -6.164283752441406, 166.84658813476562, 21.093963623046875, 86.7723388671875, 171.994873046875, 127.05921936035156, 6.2464599609375, 116.728271484375, 140.07720947265625, 1.5460205078125, 163.88839721679688, 68.16754913330078, 21.987228393554688, 198.5245361328125, 83.35113525390625, 147.03121948242188, -56.0994873046875, 148.97027587890625, 160.25440979003906, 7.504520416259766, 53.2684326171875, -25.96600341796875, 143.1632080078125, -5.788848876953125, -62.9600830078125, -188.28070068359375, -20.329204559326172, 140.72509765625, 210.3709716796875, 62.812774658203125, -3.1570968627929688, 61.871620178222656, 104.54248046875, 190.7138671875, 140.4785919189453, 149.7458953857422, 43.88873291015625, 203.29742431640625, 13.022674560546875, -16.793060302734375, -142.18392944335938, 133.63299560546875, 182.555908203125, 2.6071853637695312, 152.80697631835938, -24.507896423339844, 135.651611328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000387.npy"}
{"epoch": 0.8104712041884817, "step": 388, "batch_size": 128, "mean": 43.73088073730469, "std": 90.14434051513672, "min": -176.37371826171875, "p10": -56.974441528320305, "median": 20.424992561340332, "p90": 161.75985107421874, "max": 255.355224609375, "pos_frac": 0.6796875, "sample": [102.59307861328125, -109.13742065429688, 138.10940551757812, 0.0, 255.355224609375, 151.9761505126953, -4.824066162109375, 178.34564208984375, 39.664581298828125, 159.16162109375, -144.3468017578125, 141.16787719726562, -138.592529296875, 4.342813491821289, 5.8379669189453125, 27.383636474609375, 69.08829498291016, 120.427734375, 198.55902099609375, -23.113037109375, 103.23477172851562, 200.709228515625, -141.01812744140625, -20.717193603515625, -10.929351806640625, -42.91911315917969, 98.03980255126953, 125.69265747070312, 162.96267700195312, 169.11529541015625, 62.051971435546875, 17.27814483642578, 0.0, 12.1661376953125, 19.942214965820312, -87.706787109375, -16.3262939453125, 13.316131591796875, 110.71456909179688, 84.01480102539062, -68.33488464355469, 23.23131561279297, 118.97294616699219, 139.1697235107422, -27.9451904296875, -41.873321533203125, 166.64337158203125, -20.723793029785156, -7.459102630615234, 4.402496337890625, 107.92622375488281, -10.067764282226562, 168.68780517578125, -39.660430908203125, 13.339324951171875, 125.60938262939453, -3.17486572265625, 175.30252075195312, 82.4210205078125, -29.037078857421875, -7.7661895751953125, 52.20365905761719, 140.2683868408203, 7.21392822265625, 20.543649673461914, -54.08343505859375, 76.39202880859375, -167.23928833007812, 180.75192260742188, -104.48477172851562, 58.863037109375, 155.96673583984375, 127.43807983398438, 104.36449432373047, 25.178863525390625, 0.071014404296875, -1.1849365234375, 158.57305908203125, 12.037437438964844, -135.00840759277344, -11.757232666015625, 177.125, 152.15248107910156, -17.293258666992188, 20.005481719970703, 9.455581665039062, 30.700942993164062, 132.9817352294922, 71.67079162597656, 18.713024139404297, -122.91104125976562, 24.7822265625, -1.345123291015625, -176.37371826171875, -0.3104286193847656, 178.7043914794922, 156.4429931640625, 138.9111328125, 8.246095657348633, 150.754638671875, 3.9845123291015625, 121.24685668945312, 161.18124389648438, 188.16973876953125, -5.035308837890625, 11.024215698242188, 31.8402099609375, 22.841751098632812, -29.12957763671875, 1.590475082397461, 161.24435424804688, 12.405670166015625, -31.968978881835938, 20.30633544921875, 15.121147155761719, -63.720123291015625, 133.29193115234375, 13.781951904296875, 142.8729248046875, 46.0552978515625, -103.58650970458984, -48.62620544433594, 32.084716796875, 7.289398193359375, 28.116744995117188, 128.20848083496094, 120.73831939697266, -23.581947326660156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000388.npy"}
{"epoch": 0.812565445026178, "step": 389, "batch_size": 128, "mean": 59.917171478271484, "std": 86.55052185058594, "min": -160.09814453125, "p10": -28.042704772949218, "median": 56.194175720214844, "p90": 161.69738006591797, "max": 194.3822021484375, "pos_frac": 0.7421875, "sample": [1.5489959716796875, -111.55799865722656, -17.48603057861328, 99.535400390625, 144.68740844726562, 143.8125, 28.891998291015625, 26.58892822265625, -9.353164672851562, 176.58502197265625, 27.1768798828125, 162.48399353027344, 31.22283935546875, 185.93698120117188, 44.32365417480469, 141.43746948242188, -9.971649169921875, 145.0957489013672, 100.88845825195312, 186.3759765625, -74.24107360839844, 129.4292755126953, 168.90142822265625, 112.35794067382812, 5.186271667480469, 154.30482482910156, -155.71450805664062, 181.32647705078125, 34.31980895996094, 164.62603759765625, 33.720703125, -155.63107299804688, 127.90338134765625, -29.60107421875, -20.864105224609375, -4.6727294921875, -47.647865295410156, 131.65066528320312, 178.52493286132812, 180.28567504882812, 139.52410888671875, 0.0, -112.08827209472656, -15.958106994628906, 142.72781372070312, -86.95513916015625, 160.9651336669922, 135.29629516601562, 147.04856872558594, 20.79644775390625, -12.5067138671875, 143.9913330078125, 38.185760498046875, 10.204858779907227, 123.50050354003906, 61.556488037109375, 32.07041931152344, 43.80914306640625, -21.52874755859375, -3.7535171508789062, 37.33564758300781, 1.6866683959960938, 96.70759582519531, 145.9612579345703, -76.60968017578125, 135.62127685546875, 82.94906616210938, 138.60169982910156, 122.79115295410156, 109.26829528808594, 137.66592407226562, -24.89727783203125, 43.049072265625, 153.26524353027344, 1.73114013671875, 194.3822021484375, 83.70046997070312, 67.48175048828125, -30.443832397460938, -6.13641357421875, 58.03973388671875, 128.03085327148438, 22.81305694580078, 63.467010498046875, 16.717849731445312, 151.69854736328125, -152.96751403808594, 133.30345153808594, 123.83480834960938, 1.992095947265625, 179.55224609375, 24.87335205078125, 4.0726318359375, 6.303470611572266, -0.6973342895507812, 97.99864959716797, 8.82806396484375, -160.09814453125, -27.2213134765625, 161.36026000976562, 1.822998046875, 150.9620361328125, -15.23846435546875, -25.439441680908203, 84.791748046875, 174.12155151367188, 0.0, 68.31462097167969, 136.55517578125, 185.44395446777344, 10.246343612670898, 54.34861755371094, -3.2836761474609375, 88.33441162109375, 85.32554626464844, 158.66285705566406, 148.36138916015625, 9.251884460449219, -27.374832153320312, 19.014625549316406, 103.36405944824219, 145.58489990234375, -136.36074829101562, 16.2745361328125, 138.62387084960938, -1.099609375, 133.24874877929688, 150.25341796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000389.npy"}
{"epoch": 0.8146596858638744, "step": 390, "batch_size": 128, "mean": 55.4965705871582, "std": 94.30902862548828, "min": -175.66632080078125, "p10": -69.53649749755857, "median": 56.076168060302734, "p90": 168.3517822265625, "max": 195.71783447265625, "pos_frac": 0.75, "sample": [158.2108154296875, 1.6609954833984375, 40.77667236328125, 18.055694580078125, 73.86248779296875, 137.37228393554688, 0.5871334075927734, 167.98773193359375, 125.924560546875, 66.19935607910156, -36.739013671875, 7.3797760009765625, 2.4149398803710938, -61.15704345703125, 33.03208923339844, 134.55844116210938, 0.033966064453125, 173.79318237304688, 154.25607299804688, 136.69085693359375, 76.33401489257812, -31.922698974609375, 158.9955291748047, 10.558242797851562, 65.02935791015625, 79.2117919921875, -140.03366088867188, 145.494140625, 145.2418212890625, 103.78204345703125, 157.947998046875, 128.3874053955078, -111.09489440917969, 42.134605407714844, 143.3729248046875, 35.10479736328125, 95.65370178222656, 20.607711791992188, -2.5909271240234375, 62.47209167480469, -162.452880859375, 144.14773559570312, -61.65223693847656, 127.5120849609375, 81.3501968383789, -92.26535034179688, -20.788040161132812, -162.82308959960938, 140.01718139648438, -18.272918701171875, 54.42070007324219, -59.82049560546875, -141.84242248535156, 25.10943603515625, 191.83871459960938, 15.017017364501953, 45.626190185546875, 164.83111572265625, 131.06396484375, -143.85440063476562, -109.29684448242188, 91.79617309570312, 162.92359924316406, 68.545166015625, 187.14358520507812, -142.85992431640625, 193.42086791992188, 79.91171264648438, -16.251251220703125, 145.27444458007812, 49.82499694824219, 187.75775146484375, 69.3245849609375, 0.0, -14.134918212890625, -9.530181884765625, 169.20123291015625, -25.309524536132812, 139.18020629882812, 144.8865509033203, -19.21935272216797, 155.66732788085938, 183.86444091796875, -175.66632080078125, -18.21044921875, 158.59561157226562, 195.71783447265625, 154.0435028076172, 1.8326416015625, 8.923091888427734, 68.3750228881836, 178.84521484375, 81.24993896484375, 50.880706787109375, 28.317794799804688, 122.56952667236328, -12.980873107910156, 167.0252685546875, 40.93183898925781, -52.36224365234375, 0.7373943328857422, -100.60134887695312, 178.70248413085938, 149.439208984375, -19.22528076171875, 36.926666259765625, 186.30908203125, 6.04986572265625, 1.2702217102050781, 10.89630126953125, -87.93310546875, 3.8003082275390625, -8.230903625488281, 124.034423828125, 57.73163604736328, 27.012374877929688, 159.52801513671875, 152.47393798828125, 66.39627075195312, 166.83355712890625, 137.2313232421875, 27.152435302734375, 8.82513427734375, 170.20135498046875, 169.50018310546875, -109.86370849609375, 120.99901580810547, 0.40578460693359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000390.npy"}
{"epoch": 0.8167539267015707, "step": 391, "batch_size": 128, "mean": 54.02735900878906, "std": 78.54264068603516, "min": -146.93646240234375, "p10": -29.516752624511714, "median": 38.913055419921875, "p90": 155.84612121582032, "max": 317.624267578125, "pos_frac": 0.7578125, "sample": [-6.746498107910156, -21.82709503173828, 4.525991439819336, 51.756126403808594, 69.00312042236328, 10.147979736328125, -2.10357666015625, 5.619358062744141, 12.923583984375, 108.1198959350586, 3.8792495727539062, 15.92645263671875, -14.224174499511719, 95.77769470214844, 178.7012939453125, 66.689208984375, 48.180419921875, 101.9522476196289, -56.32066345214844, -18.21424102783203, -18.181884765625, 149.38458251953125, 73.21701049804688, 63.840789794921875, -46.043670654296875, 44.83689880371094, 155.52926635742188, 61.179534912109375, 2.133045196533203, 38.76702880859375, 3.5518798828125, -59.67584228515625, 146.195556640625, -7.9239959716796875, 47.64501953125, 109.4635009765625, 124.09223937988281, 9.688028335571289, -12.690948486328125, 75.637939453125, 11.08443832397461, 132.6896514892578, 146.8573455810547, 205.5361328125, 153.70901489257812, -4.068878173828125, 156.58544921875, 13.251602172851562, -11.482818603515625, -4.676361083984375, 196.80892944335938, 101.99079895019531, -65.57947540283203, 39.05908203125, 0.9259796142578125, 105.0335693359375, 127.98814392089844, 89.91328430175781, 138.867919921875, -32.44354248046875, 36.965301513671875, -95.32902526855469, 133.68417358398438, 46.449493408203125, 32.938232421875, 12.403717041015625, 116.985107421875, 7.746067047119141, 61.85516357421875, 5.9857330322265625, 1.187103271484375, 165.97732543945312, 23.78863525390625, 85.4225082397461, 120.80718231201172, -9.71575927734375, 125.7073974609375, 46.2271728515625, -7.076019287109375, 148.8236083984375, -4.688995361328125, 79.24679565429688, 153.6939697265625, 317.624267578125, 174.3118896484375, 9.308990478515625, -8.105461120605469, -43.196197509765625, 148.16275024414062, 10.921905517578125, 13.83935546875, 0.05886268615722656, -145.39181518554688, 153.9332275390625, 40.324615478515625, 52.29693603515625, -31.923782348632812, 123.23844909667969, 175.7874755859375, 172.40386962890625, -32.7664794921875, -44.80877685546875, 35.86592102050781, -98.69216918945312, 6.9333953857421875, -2.4262428283691406, 33.87568664550781, 126.54887390136719, 132.54638671875, 160.7049560546875, 13.013275146484375, 131.16867065429688, 129.885009765625, 17.080596923828125, 74.29579162597656, 66.083984375, 174.61773681640625, 143.52859497070312, 204.885498046875, 8.55963134765625, -28.48516845703125, 8.614501953125, 31.5477294921875, 7.223030090332031, -7.704093933105469, 42.595245361328125, -146.93646240234375, 158.60391235351562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000391.npy"}
{"epoch": 0.818848167539267, "step": 392, "batch_size": 128, "mean": 69.51496887207031, "std": 81.76326751708984, "min": -148.68377685546875, "p10": -10.532078552246093, "median": 48.232513427734375, "p90": 184.44281311035155, "max": 219.33383178710938, "pos_frac": 0.7578125, "sample": [155.98190307617188, 137.837890625, 44.09577941894531, 41.451507568359375, 4.194732666015625, 2.3837814331054688, 49.774749755859375, 123.16574096679688, 78.23974609375, 92.38851928710938, 21.5557861328125, 115.85415649414062, 119.63591003417969, 30.43218994140625, -10.4580078125, -2.4972076416015625, -148.68377685546875, 211.58038330078125, 103.16007995605469, -44.41645812988281, 185.00286865234375, -6.429691314697266, 8.29296875, 131.45913696289062, 177.33209228515625, 197.02713012695312, 0.0, -32.8846435546875, 147.48660278320312, 141.99862670898438, 7.299369812011719, 27.17529296875, 187.16336059570312, 43.186279296875, 56.69047927856445, 189.46038818359375, 184.08160400390625, 163.25831604003906, 136.1818084716797, 74.63406372070312, 141.15890502929688, 155.901611328125, 14.69598388671875, 22.9798583984375, -7.484405517578125, 20.553390502929688, -7.088890075683594, 4.07916259765625, 110.54408264160156, 159.95465087890625, 46.690277099609375, 168.89637756347656, 192.04217529296875, 69.26557922363281, 182.74343872070312, 132.2283935546875, 9.7928466796875, 120.73809814453125, 198.55731201171875, -4.628692626953125, -11.806953430175781, 113.74207305908203, -61.244842529296875, 183.428955078125, 28.840885162353516, 31.732872009277344, -25.706695556640625, 127.52728271484375, 196.65240478515625, -0.10479736328125, 54.74418640136719, 175.93307495117188, 191.1444091796875, 133.73202514648438, 56.663299560546875, 186.88858032226562, 39.093475341796875, -108.96905517578125, -1.5137481689453125, -0.034694671630859375, 8.98370361328125, 154.89474487304688, 2.5255126953125, 99.158935546875, 15.174919128417969, -5.743377685546875, -7.90106201171875, 5.757598876953125, 21.708831787109375, -27.354934692382812, -2.9993896484375, 0.0, 137.06658935546875, 22.51824951171875, 152.06655883789062, 194.00265502929688, 55.8699951171875, 144.01742553710938, 184.20278930664062, -1.15313720703125, 171.68814086914062, -0.3133869171142578, 27.05194091796875, 2.025300979614258, 21.99658203125, 219.33383178710938, -3.2777481079101562, 191.33734130859375, 21.39190673828125, 66.83556365966797, -114.381103515625, 162.44375610351562, -10.704910278320312, 10.249923706054688, 147.66009521484375, 28.6724853515625, 65.95114135742188, -67.98873901367188, -6.189483642578125, 112.06794738769531, 63.88800048828125, 142.02386474609375, -30.146743774414062, 149.09317016601562, -24.008102416992188, 24.23725128173828, 155.4115753173828, 32.31390380859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000392.npy"}
{"epoch": 0.8209424083769633, "step": 393, "batch_size": 128, "mean": 50.68855667114258, "std": 87.08293914794922, "min": -192.72622680664062, "p10": -39.043539428710936, "median": 30.020523071289062, "p90": 154.4952865600586, "max": 294.25653076171875, "pos_frac": 0.734375, "sample": [20.126670837402344, -6.48846435546875, 5.436859130859375, 101.17633056640625, 110.13941192626953, -63.57220458984375, 125.19510650634766, 102.58697509765625, 145.999267578125, 6.72711181640625, 12.8917236328125, -3.66424560546875, 83.69549560546875, -14.309867858886719, 128.38803100585938, 25.7724609375, -127.87109375, 104.73699951171875, 49.87507629394531, 92.55914306640625, 162.0924072265625, 48.520721435546875, 12.8719482421875, 82.61715698242188, -43.84564208984375, 171.03433227539062, 155.8109588623047, 150.39947509765625, 61.085968017578125, 27.190689086914062, 152.40032958984375, 2.8255157470703125, 50.55718994140625, 142.7962646484375, 32.597076416015625, 86.15016174316406, -4.3012847900390625, 153.0179443359375, 25.52191162109375, 136.8137969970703, 15.8818359375, -37.823394775390625, 92.72811889648438, 161.52310180664062, -97.0018310546875, 159.53338623046875, -17.656288146972656, -23.65350341796875, -91.23600769042969, 64.68853759765625, 247.39202880859375, 50.914119720458984, 9.518264770507812, 19.59246826171875, 13.915443420410156, 34.78472900390625, 1.0302085876464844, -119.07441711425781, 143.38465881347656, 116.55998229980469, -192.72622680664062, 26.78741455078125, 14.6119384765625, -16.058349609375, 22.59332275390625, 148.11944580078125, 108.86622619628906, 156.02203369140625, -162.3160400390625, -42.29156494140625, -38.4344482421875, 27.4439697265625, 22.407272338867188, 52.38603210449219, 16.150897979736328, 60.1219482421875, 194.0787353515625, -16.950485229492188, 56.490814208984375, -8.527175903320312, 9.483112335205078, -14.018875122070312, 141.66531372070312, -155.8694305419922, 22.75135040283203, 109.55610656738281, 147.19058227539062, 135.60215759277344, 139.1039276123047, 150.3111572265625, -26.36139678955078, 173.81130981445312, 145.94325256347656, 158.1041717529297, 159.7059783935547, 26.7816162109375, 21.10791015625, -40.464752197265625, 153.68756103515625, -186.3292236328125, 5.979545593261719, 36.60212707519531, -0.7333984375, -7.823905944824219, 294.25653076171875, 69.98123168945312, 41.677215576171875, 0.0, 9.424575805664062, 56.4180908203125, 204.17498779296875, -42.3914794921875, 3.1753082275390625, 9.353500366210938, -11.992462158203125, 109.92117309570312, 142.13662719726562, -11.179328918457031, -22.318511962890625, -3.3859710693359375, 0.47121620178222656, -27.45965576171875, 149.44091796875, 153.93142700195312, 1.287353515625, 106.0216064453125, 125.21044921875, 144.85931396484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000393.npy"}
{"epoch": 0.8230366492146597, "step": 394, "batch_size": 128, "mean": 49.70208740234375, "std": 80.59403228759766, "min": -147.5179901123047, "p10": -37.18258438110351, "median": 31.904830932617188, "p90": 163.43333435058594, "max": 207.2000732421875, "pos_frac": 0.6875, "sample": [164.935302734375, 16.96826171875, 166.71578979492188, 105.30023193359375, 53.5792236328125, 150.70547485351562, -11.83038330078125, -4.0906524658203125, 24.1241455078125, 168.45452880859375, 127.10113525390625, 11.045486450195312, 33.55792236328125, 71.55172729492188, -38.6729736328125, -2.595226287841797, 158.23291015625, -35.659767150878906, 141.0628662109375, -22.83810043334961, 5.681610107421875, 151.67538452148438, 202.67398071289062, -17.08636474609375, 1.8086509704589844, -58.71214294433594, 31.537506103515625, 132.76397705078125, 106.51220703125, 34.3956298828125, -34.31451416015625, 0.0, -36.543846130371094, -52.6759033203125, 162.822021484375, 11.813674926757812, 15.96539306640625, 110.57306671142578, 5.802490234375, 51.29755401611328, 61.4521484375, 90.24296569824219, 44.38348388671875, -147.5179901123047, 17.03028106689453, 90.637451171875, 159.24755859375, 50.721378326416016, 2.933563232421875, 16.57171630859375, -16.365325927734375, -41.404937744140625, 47.064605712890625, -14.73309326171875, -22.00213623046875, 176.32107543945312, 12.373397827148438, 58.96441650390625, 9.979080200195312, 54.037261962890625, 68.07559204101562, -53.347442626953125, -0.9905624389648438, -23.829086303710938, 160.8002471923828, 3.167713165283203, -4.864097595214844, 100.45294189453125, 21.588478088378906, 103.61117553710938, 142.55999755859375, 159.86228942871094, 173.21640014648438, 4.007654190063477, 198.59967041015625, -2.88446044921875, -62.65447998046875, -30.061553955078125, -6.3607330322265625, 205.30084228515625, 45.22439193725586, -122.45538330078125, 132.60140991210938, -12.94097900390625, -51.493499755859375, 20.274368286132812, 39.18853759765625, 28.25054931640625, 8.793350219726562, 90.0938720703125, -2.4814605712890625, -5.684146881103516, 108.62387084960938, 16.585403442382812, -24.4434814453125, -98.27008056640625, 66.9481201171875, 74.02020263671875, 83.04916381835938, 65.4906005859375, -132.9346466064453, 145.94940185546875, 32.27215576171875, -29.295120239257812, 182.76422119140625, 207.2000732421875, -27.222824096679688, 49.4547119140625, 109.07240295410156, 169.37109375, 118.17089080810547, 131.744140625, 156.61495971679688, 143.3350830078125, 163.0343017578125, -49.11016845703125, 186.976806640625, 19.1602783203125, 4.098358154296875, 164.36441040039062, 12.450469970703125, 102.21823120117188, -73.65447998046875, 132.52880859375, -22.087913513183594, -14.544357299804688, -35.799285888671875, 144.53463745117188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000394.npy"}
{"epoch": 0.8251308900523561, "step": 395, "batch_size": 128, "mean": 52.457149505615234, "std": 81.06787109375, "min": -145.07467651367188, "p10": -23.00142211914062, "median": 28.64014434814453, "p90": 165.93792419433595, "max": 206.932861328125, "pos_frac": 0.75, "sample": [56.28985595703125, 9.5028076171875, -8.059440612792969, -7.273124694824219, -5.5960693359375, -14.160858154296875, 1.60870361328125, 201.12457275390625, 189.31103515625, 19.018722534179688, 32.747093200683594, 49.433349609375, 26.5072021484375, -19.453887939453125, 0.6089210510253906, 37.88856506347656, 12.213050842285156, -77.8531494140625, 62.460906982421875, -21.153411865234375, 29.0927734375, 153.71340942382812, -18.39714813232422, -59.93609619140625, 177.09286499023438, 121.03469848632812, 23.257949829101562, 1.6742324829101562, 138.461181640625, 38.31501770019531, 130.625, 137.94927978515625, 108.90884399414062, 104.42408752441406, 43.03485107421875, 164.51486206054688, 92.2958984375, 48.74224853515625, 0.3573875427246094, -14.913833618164062, 100.10475158691406, 12.840965270996094, -14.150482177734375, 14.036407470703125, 160.13800048828125, 125.02902221679688, -12.308761596679688, -92.5944595336914, 130.08285522460938, 169.6211395263672, 184.30218505859375, -70.9349365234375, 54.130859375, 149.708251953125, 163.61965942382812, 115.25692749023438, 40.289581298828125, 164.48077392578125, 15.1539306640625, 109.21092224121094, -15.342971801757812, -87.89697265625, 23.639739990234375, -7.21368408203125, 152.63577270507812, -26.793701171875, 20.33684539794922, 107.34680938720703, -13.468994140625, 8.701812744140625, -17.560638427734375, 23.052490234375, -49.46620178222656, -85.62371826171875, 34.4442138671875, 75.30320739746094, 140.1260223388672, -1.9500503540039062, 17.18304443359375, 21.46484375, 24.899810791015625, 20.46826934814453, 206.932861328125, 165.85394287109375, 159.07313537597656, -21.37615966796875, 5.6543426513671875, -53.462493896484375, 13.208663940429688, 84.10783386230469, 9.611808776855469, 87.04391479492188, -125.21820068359375, 116.82433319091797, -14.399322509765625, 12.677528381347656, 178.69500732421875, 185.2211151123047, 165.1988983154297, 166.13388061523438, -7.543449401855469, 83.6822509765625, 32.87969970703125, -9.960372924804688, 127.9937744140625, 29.743759155273438, 157.21849060058594, 9.524078369140625, 150.97898864746094, -132.5441436767578, 4.3189239501953125, 141.69749450683594, 176.36654663085938, -65.963134765625, 11.848968505859375, -145.07467651367188, 34.732177734375, 6.546468734741211, 176.2244873046875, 173.74887084960938, 160.882568359375, 167.53529357910156, 19.949462890625, 8.820068359375, 28.187515258789062, 115.8630142211914, 7.7490692138671875, 59.6119384765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000395.npy"}
{"epoch": 0.8272251308900523, "step": 396, "batch_size": 128, "mean": 52.02455520629883, "std": 88.20069122314453, "min": -201.04074096679688, "p10": -47.61538391113281, "median": 46.52484130859375, "p90": 162.54114990234373, "max": 216.10287475585938, "pos_frac": 0.6875, "sample": [130.12759399414062, 18.10797119140625, -11.05255126953125, 165.28451538085938, 142.9683837890625, 14.951873779296875, -5.223560333251953, 16.680740356445312, -40.16558837890625, 178.6595458984375, 98.36087036132812, -41.67095947265625, 104.60899353027344, 44.953216552734375, 1.18463134765625, 48.312530517578125, 178.5987548828125, -65.11456298828125, 7.5750732421875, 137.36233520507812, -32.84727478027344, -11.488456726074219, -7.86199951171875, 216.10287475585938, -5.3688507080078125, 5.47113037109375, 85.2464599609375, 135.17852783203125, 77.27592468261719, 114.31856536865234, 65.32525634765625, -117.26644897460938, 69.15509033203125, 157.67782592773438, -124.07171630859375, 165.46102905273438, 10.851470947265625, -9.973114013671875, -5.573825836181641, 114.30299377441406, 39.16365051269531, 138.03958129882812, 75.42559814453125, 153.45188903808594, 181.41802978515625, -21.30291748046875, 14.515600204467773, 16.427734375, -2.45904541015625, 138.94186401367188, 188.22097778320312, 180.81948852539062, 214.28131103515625, 4.903739929199219, 157.42694091796875, 82.76278686523438, 147.7740478515625, -0.013092041015625, 16.2010498046875, -3.6613197326660156, 7.178749084472656, -22.257354736328125, -46.2728271484375, -42.73237609863281, 110.95962524414062, -88.02313232421875, 51.468170166015625, -140.15234375, 14.186264038085938, -2.1816864013671875, 140.9532470703125, -6.84832763671875, -67.05404663085938, -5.197380065917969, 145.2086181640625, 164.69464111328125, 62.010101318359375, 141.06472778320312, 23.037109375, -148.9491729736328, 49.788330078125, 112.5763168334961, 66.96944427490234, 135.619873046875, 51.78619384765625, 161.61822509765625, 22.551929473876953, 184.2669219970703, -105.02849578857422, 149.65911865234375, -29.55535888671875, 92.38568115234375, -50.748016357421875, 107.69001007080078, 86.39277648925781, 108.39080047607422, 142.2999267578125, 2.129638671875, 17.972686767578125, 44.9832763671875, 160.21141052246094, 80.24348449707031, -0.0753173828125, -201.04074096679688, -36.161376953125, 158.49305725097656, -10.979934692382812, 181.63735961914062, 160.04449462890625, 167.34625244140625, 0.0, 19.16595458984375, 155.83493041992188, 37.877166748046875, 14.1351318359375, 9.754829406738281, 0.0, 48.06640625, 75.48165893554688, -15.717697143554688, 62.25096130371094, -131.79180908203125, 75.60989379882812, 146.67967224121094, 145.65682983398438, -77.64871215820312, -102.91590881347656, 145.37989807128906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000396.npy"}
{"epoch": 0.8293193717277487, "step": 397, "batch_size": 128, "mean": 50.12998580932617, "std": 87.19039154052734, "min": -172.7880859375, "p10": -52.56491088867186, "median": 39.079132080078125, "p90": 164.52744140625, "max": 201.39990234375, "pos_frac": 0.7109375, "sample": [-6.501068115234375, 7.0614013671875, -12.779754638671875, 17.53618621826172, 170.186767578125, 39.923980712890625, -100.09209442138672, 48.6644287109375, -4.634006500244141, 17.28729248046875, 8.849563598632812, 14.8004150390625, 117.48785400390625, 3.37493896484375, 27.954818725585938, 165.8506622314453, 22.455123901367188, 117.40986633300781, 193.5059814453125, 134.35333251953125, -133.693115234375, 25.24151611328125, -18.258880615234375, 81.62911987304688, 135.93463134765625, 148.67739868164062, 0.0, -1.93206787109375, 136.19573974609375, 41.696502685546875, 0.0, -37.63555908203125, 128.07257080078125, -17.094467163085938, -4.798828125, 2.3591690063476562, 9.17999267578125, 148.04273986816406, -72.4970703125, 137.3291778564453, 173.45449829101562, 45.457244873046875, 147.2906494140625, 164.01812744140625, -29.33441162109375, 0.5079116821289062, 139.107421875, 150.943115234375, 201.39990234375, -41.7181396484375, 197.32107543945312, -64.71597290039062, 182.26394653320312, -172.7880859375, -103.16159057617188, -38.90374755859375, 45.23779296875, -125.05845642089844, 0.2983207702636719, -147.13607788085938, -34.420166015625, 81.81571960449219, 165.3157958984375, 136.98670959472656, 85.50120544433594, 28.719215393066406, 140.56790161132812, -10.70166015625, 47.20365905761719, -1.404348373413086, 57.42279052734375, 187.00750732421875, 75.53945922851562, 91.36277770996094, 146.299072265625, 130.2899169921875, 138.11990356445312, -80.98358154296875, 96.52568054199219, 51.857757568359375, 150.0423126220703, 27.8125, 171.37255859375, 167.02517700195312, 38.234283447265625, 0.9098453521728516, 142.87979125976562, -5.63824462890625, 11.450126647949219, 18.97308349609375, 137.64373779296875, 16.453170776367188, 35.34065246582031, 9.744705200195312, 152.944580078125, 142.93551635742188, 11.584627151489258, 16.453933715820312, 169.18563842773438, 60.13067626953125, 20.46875, -91.27239990234375, 39.9599609375, -16.093158721923828, 180.0686798095703, 164.1895751953125, 126.91658020019531, 4.035491943359375, -14.981098175048828, -19.01727294921875, -24.364517211914062, 160.8358154296875, 129.65087890625, -63.625244140625, 43.6041259765625, 41.513916015625, -124.23489379882812, -13.405487060546875, 140.19837951660156, 9.723403930664062, -41.69189453125, -47.82476806640625, 114.61691284179688, 116.43289947509766, 68.40679931640625, -135.22225952148438, 119.76129150390625, 103.85578918457031], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000397.npy"}
{"epoch": 0.831413612565445, "step": 398, "batch_size": 128, "mean": 40.003379821777344, "std": 93.42713928222656, "min": -170.64816284179688, "p10": -74.05093917846679, "median": 20.829315185546875, "p90": 164.15410766601562, "max": 242.47210693359375, "pos_frac": 0.6640625, "sample": [13.869873046875, 18.647918701171875, 147.40582275390625, -8.5380859375, 242.47210693359375, -92.56924438476562, 59.26206970214844, -70.47775268554688, 52.01667785644531, 165.3494873046875, -138.794189453125, -59.180908203125, -160.96713256835938, 14.378753662109375, 7.977439880371094, 3.3614501953125, 32.222747802734375, -4.213470458984375, 158.96084594726562, 170.18321228027344, 161.0511932373047, -133.89361572265625, -110.9598388671875, 133.39755249023438, 39.22216796875, 117.41073608398438, 19.3582763671875, 139.6024169921875, 151.60708618164062, 36.6788330078125, -74.175048828125, 6.308139801025391, 35.244537353515625, -170.64816284179688, -62.0184326171875, 135.94332885742188, 30.106689453125, 144.96307373046875, 0.53240966796875, -108.36589813232422, 1.834381103515625, -85.518798828125, 16.837425231933594, 150.51785278320312, -0.5197944641113281, -48.81903076171875, 21.5201416015625, 135.75831604003906, 20.13848876953125, -13.742828369140625, 32.06268310546875, -15.96746826171875, 105.57138061523438, -8.885452270507812, 136.8091278076172, 52.148193359375, 2.934131622314453, -106.87808227539062, 23.299240112304688, -16.115516662597656, 30.92608642578125, 46.0810546875, 164.93075561523438, -26.020172119140625, 183.14764404296875, 6.2963104248046875, 23.056175231933594, -142.48089599609375, 129.60336303710938, 171.0918426513672, 29.486526489257812, 196.031494140625, 12.399818420410156, 203.39474487304688, -11.72064208984375, -17.421920776367188, 10.072822570800781, 161.5528564453125, -30.98394775390625, -12.48834228515625, 172.88595581054688, 174.82373046875, 7.2456817626953125, 36.52045440673828, 33.008453369140625, 6.8011627197265625, -58.87199401855469, 141.7738800048828, 41.339813232421875, -49.362457275390625, 116.83087158203125, 155.60421752929688, -12.453399658203125, -145.24029541015625, 78.647216796875, -7.73797607421875, 143.86163330078125, -24.133636474609375, -12.746673583984375, 108.34603881835938, -16.704566955566406, -11.299896240234375, 139.82171630859375, 120.30455017089844, 12.456939697265625, 14.429206848144531, 94.23037719726562, 165.409423828125, 159.54806518554688, -30.567840576171875, -73.99774932861328, 180.73733520507812, 163.01507568359375, -49.98168182373047, 161.0596923828125, -3.1779327392578125, -31.23046875, -165.94454956054688, 12.538818359375, 163.82125854492188, 129.9132080078125, -7.742889404296875, 6.106807708740234, 212.28472900390625, 39.306732177734375, 29.595123291015625, 128.2027130126953, 98.48233032226562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000398.npy"}
{"epoch": 0.8335078534031414, "step": 399, "batch_size": 128, "mean": 49.108726501464844, "std": 80.38562774658203, "min": -168.93069458007812, "p10": -36.688604736328124, "median": 23.90326690673828, "p90": 163.65821228027343, "max": 208.14678955078125, "pos_frac": 0.6953125, "sample": [-5.9738311767578125, 138.52708435058594, 142.31069946289062, 145.25823974609375, 12.59698486328125, -11.774223327636719, -0.246795654296875, 40.424072265625, 31.00335693359375, 171.92050170898438, 47.566139221191406, -48.42280578613281, -56.394439697265625, 151.71351623535156, -2.8583602905273438, 68.0947265625, 91.88839721679688, 6.422454833984375, 175.63389587402344, 30.897354125976562, -3.627716064453125, -72.67202758789062, -23.18231201171875, 159.6590576171875, 8.7491455078125, -1.3192138671875, 11.861038208007812, -131.05856323242188, -168.93069458007812, 131.55848693847656, -4.566802978515625, 163.65280151367188, 22.578887939453125, 172.95672607421875, 18.178245544433594, -14.28057861328125, 146.05487060546875, -9.364891052246094, -0.8013153076171875, 184.27871704101562, 34.963623046875, 151.43226623535156, 208.14678955078125, 150.81423950195312, -4.59088134765625, -11.625694274902344, 158.6575927734375, -53.442291259765625, 4.325967788696289, 144.87567138671875, -29.935028076171875, 41.84606170654297, 128.5122833251953, 105.16119384765625, 184.418212890625, -8.22601318359375, 4.237663269042969, 9.72882080078125, 15.894386291503906, -9.69833755493164, 19.91632080078125, 164.0517578125, -1.52069091796875, 22.149505615234375, 148.18917846679688, 14.316715240478516, 149.09234619140625, 20.707725524902344, 103.928955078125, 42.50482177734375, 68.19438171386719, 121.23086547851562, -11.09381103515625, -128.75852966308594, -132.629150390625, -81.34445190429688, -11.15411376953125, 34.18367004394531, 163.67083740234375, 167.98455810546875, 23.7305908203125, 143.8504638671875, 91.0601806640625, 145.00257873535156, 138.99832153320312, 62.584381103515625, 19.156158447265625, -2.0247955322265625, 45.29046630859375, 133.42947387695312, 10.233304977416992, -39.137237548828125, 18.352272033691406, 78.400634765625, 12.17376708984375, 19.5255126953125, -47.067047119140625, 0.7442131042480469, 170.38137817382812, 9.130355834960938, 74.7960205078125, 20.693572998046875, 47.92266845703125, -48.64019775390625, 12.94580078125, 167.48715209960938, -19.274368286132812, 195.61178588867188, 117.24978637695312, 59.0997314453125, 173.3603973388672, -89.81195068359375, -35.639190673828125, 38.15110778808594, 24.405643463134766, 24.075942993164062, 105.76692199707031, 115.67868041992188, -9.182235717773438, 140.86483764648438, -33.934051513671875, 32.20237731933594, 122.06500244140625, 87.42369079589844, -17.2940673828125, -1.067138671875, 112.93046569824219, 16.784988403320312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000399.npy"}
{"epoch": 0.8356020942408376, "step": 400, "batch_size": 128, "mean": 55.662322998046875, "std": 82.91958618164062, "min": -177.99258422851562, "p10": -35.99599914550781, "median": 36.369998931884766, "p90": 161.4956268310547, "max": 204.94931030273438, "pos_frac": 0.75, "sample": [6.0204620361328125, -57.58079528808594, 0.773681640625, 35.0550537109375, 114.81365966796875, 22.44812774658203, 114.9377670288086, 120.41549682617188, 49.49664306640625, -99.14885711669922, 6.248268127441406, 86.900390625, -177.99258422851562, 21.65341567993164, 169.2727508544922, 15.668487548828125, 35.32361602783203, 151.59898376464844, 123.30282592773438, 140.53334045410156, 194.20880126953125, 13.574295043945312, 24.006423950195312, 83.98834228515625, 157.78701782226562, 118.93646240234375, -42.28057861328125, 104.03176879882812, -23.00091552734375, 105.4161376953125, 42.37066650390625, 8.302902221679688, 22.887596130371094, 126.79811096191406, 44.799591064453125, 165.58013916015625, 159.31219482421875, 108.339111328125, 23.739486694335938, 63.183868408203125, -143.19686889648438, 138.22381591796875, 127.35392761230469, 161.33010864257812, 20.29888916015625, 156.49722290039062, 0.52386474609375, 79.67962646484375, 8.548828125, -4.050933837890625, 37.4163818359375, 135.752685546875, 172.10308837890625, -120.46881103515625, 2.71563720703125, -31.48188018798828, 4.489205360412598, 12.691551208496094, 158.32113647460938, -2.3742809295654297, -1.3730907440185547, 51.584930419921875, 165.01400756835938, 132.95233154296875, 29.704437255859375, -35.394287109375, 20.623260498046875, 151.00421142578125, 107.43435668945312, 172.1875762939453, -5.382049560546875, 2.9760971069335938, -23.619415283203125, 27.896087646484375, -44.2545166015625, -18.38421630859375, 160.4278564453125, 171.86949157714844, 1.157012939453125, -44.318878173828125, 90.27130126953125, 29.437652587890625, 127.9664306640625, 114.49856567382812, -3.5048885345458984, 161.8818359375, 132.54273986816406, -37.399993896484375, 48.08592224121094, 45.2890625, 71.9660415649414, -21.08697509765625, -8.256210327148438, -3.2573585510253906, 2.70074462890625, 151.45541381835938, -27.846847534179688, 125.01173400878906, 2.9511756896972656, -110.97940063476562, 48.519287109375, 171.32034301757812, 132.57827758789062, 25.734222412109375, -28.30560302734375, 25.1893310546875, 130.7581024169922, -6.587158203125, 13.225448608398438, 160.93333435058594, 143.59278869628906, 7.056510925292969, 154.434814453125, -13.306854248046875, 155.55343627929688, 163.6693572998047, 152.0716552734375, -15.996612548828125, 204.94931030273438, 180.094482421875, 164.0611114501953, 156.51730346679688, 0.0, -70.74560546875, -39.245758056640625, 2.41131591796875, -143.72311401367188, 40.09007263183594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000400.npy"}
{"epoch": 0.418848167539267, "step": 201, "batch_size": 128, "mean": 27.85923194885254, "std": 44.210655212402344, "min": -80.25091552734375, "p10": -25.491358184814448, "median": 21.294708251953125, "p90": 85.84371337890624, "max": 125.24954223632812, "pos_frac": 0.7109375, "sample": [-75.83026123046875, -14.35406494140625, 56.63738250732422, 74.81170654296875, 82.52615356445312, 47.91461181640625, -2.1760101318359375, 19.556610107421875, 25.130462646484375, 56.295745849609375, 78.49336242675781, 1.8727502822875977, -5.461875915527344, 24.905731201171875, -66.6651611328125, -17.126983642578125, 75.68692016601562, 95.46160888671875, 63.427955627441406, 43.05218505859375, 13.745468139648438, 93.58796691894531, 87.74645233154297, 0.0, 65.14810180664062, 38.19203186035156, 70.36366271972656, 46.90496826171875, 4.81219482421875, 89.17266845703125, 94.4853515625, 56.619049072265625, 21.794677734375, -11.228446960449219, 0.0, -61.95562744140625, 8.571563720703125, -0.6122207641601562, 9.242721557617188, 5.829559326171875, 30.392303466796875, 46.75372314453125, 79.10148620605469, 74.56026458740234, -43.225067138671875, 14.964492797851562, 61.35383605957031, 60.822837829589844, -17.707504272460938, 51.93555450439453, -2.6960906982421875, 20.661216735839844, 23.11102294921875, 62.9041748046875, -2.441225051879883, 13.866317749023438, 16.36095428466797, 82.30972290039062, 70.16163635253906, 77.18557739257812, -3.9362258911132812, 76.12130737304688, 101.15325927734375, 13.380767822265625, 89.11972045898438, 66.87763214111328, 30.72747802734375, 0.0, 9.911186218261719, -80.25091552734375, 18.5732421875, 10.023345947265625, -2.562347412109375, 77.582275390625, 45.9244384765625, -2.8358612060546875, 0.9572868347167969, 9.387351989746094, -10.0496826171875, 39.513763427734375, 29.99139404296875, 37.87396240234375, -14.228485107421875, -13.345314025878906, -1.5987548828125, 92.04327392578125, -8.038833618164062, 86.78036499023438, -28.147125244140625, 125.24954223632812, 48.53338623046875, 63.082275390625, 56.54571533203125, -5.25537109375, -33.9586181640625, -28.3577880859375, 9.985466003417969, 53.658538818359375, 6.146354675292969, 88.73831176757812, 74.715087890625, 69.33070373535156, -55.756591796875, 72.51840209960938, 18.206024169921875, -8.597335815429688, -62.93843078613281, 20.79473876953125, 85.002197265625, -44.099517822265625, 20.556503295898438, 35.8912353515625, 17.1915283203125, 1.577545166015625, -3.6442718505859375, 104.01763916015625, 29.710784912109375, 47.988792419433594, 51.63427734375, 85.44229125976562, 8.756694793701172, -56.9500732421875, 12.48687744140625, 95.1434326171875, 1.9794845581054688, 63.53533935546875, -24.353172302246094, -66.42901611328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000201.npy"}
{"epoch": 0.42094240837696334, "step": 202, "batch_size": 128, "mean": 27.30573081970215, "std": 51.115657806396484, "min": -105.95184326171875, "p10": -32.24271697998046, "median": 18.760168075561523, "p90": 98.1092315673828, "max": 114.128662109375, "pos_frac": 0.71875, "sample": [103.3828125, 9.796394348144531, 0.510284423828125, 16.283050537109375, 89.38453674316406, -0.098175048828125, -18.86767578125, 101.27581787109375, -63.339874267578125, -11.33905029296875, -6.34716796875, 98.13702392578125, 26.8602294921875, 86.1187744140625, 67.81183624267578, -1.9787006378173828, -22.1551513671875, 69.16422271728516, 6.608863830566406, 108.9912109375, 19.708126068115234, 55.830718994140625, -29.671234130859375, 37.13957214355469, -20.575294494628906, 5.850738525390625, 102.95172119140625, 0.9451513290405273, 26.656478881835938, 7.572967529296875, -52.54547882080078, 28.98828125, -40.62562561035156, 100.5631103515625, 20.535301208496094, 76.03622436523438, 67.0452880859375, 106.94454956054688, 9.298828125, -68.77156066894531, 0.5828781127929688, 107.14285278320312, 62.6016845703125, 90.45147705078125, -0.6100921630859375, 20.412872314453125, 43.336578369140625, 36.66722869873047, 89.07330322265625, -5.4639739990234375, -74.80274963378906, 56.322021484375, 84.27183532714844, -0.8849449157714844, -77.32461547851562, 54.90008544921875, 49.2266845703125, 17.167633056640625, 26.680801391601562, -45.43468475341797, 64.15594482421875, 67.96934509277344, 100.896240234375, -24.991546630859375, -9.01959228515625, -15.754791259765625, 102.49029541015625, 0.27538299560546875, 9.513229370117188, 34.826263427734375, 1.6814289093017578, 83.40472412109375, 110.8905029296875, 1.6900920867919922, 64.51708984375, 96.96932983398438, 70.00135040283203, 63.762847900390625, 89.84315490722656, 70.54043579101562, 0.0, 114.128662109375, 0.7836380004882812, 14.080657958984375, 15.901336669921875, 5.7757720947265625, 0.9443283081054688, -51.67387390136719, 3.8729782104492188, -27.252044677734375, 58.22918701171875, 76.91036987304688, 2.9075164794921875, 17.812210083007812, 96.13943481445312, 58.02814483642578, -38.24284362792969, -7.200469970703125, -14.282958984375, -27.7269287109375, -105.95184326171875, 92.770263671875, 58.549827575683594, 81.81074523925781, 40.892425537109375, 28.26629638671875, -8.5374755859375, 14.3460693359375, 30.040178298950195, 0.20819854736328125, 63.53363037109375, 73.79481506347656, -53.91920471191406, 12.224395751953125, 38.24308776855469, -100.07565307617188, 40.016693115234375, 69.37171936035156, -95.3953857421875, -3.0895919799804688, 3.5181407928466797, 85.34352111816406, 1.0187530517578125, -2.308511734008789, 1.7390785217285156, 98.09732055664062, -12.83819580078125, 112.27359008789062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000202.npy"}
{"epoch": 0.42303664921465967, "step": 203, "batch_size": 128, "mean": 22.19290542602539, "std": 39.1568489074707, "min": -99.94331359863281, "p10": -21.528671264648434, "median": 12.281874656677246, "p90": 76.34958190917969, "max": 118.95687866210938, "pos_frac": 0.7109375, "sample": [1.7682476043701172, 6.882316589355469, 5.876434326171875, 67.14752197265625, 70.76220703125, 16.933349609375, 10.427209854125977, 8.63909912109375, 10.63739013671875, 68.1107177734375, 68.120361328125, 44.675201416015625, 4.394620895385742, 39.92535400390625, 18.117095947265625, 38.409423828125, -12.759521484375, -51.06635284423828, -4.364265441894531, 12.720901489257812, 14.582626342773438, 0.0, -43.336181640625, 55.65302276611328, -24.833084106445312, 11.682586669921875, -22.87652587890625, 75.51556396484375, 2.6708984375, -3.383403778076172, 31.857421875, 56.19740295410156, 29.850067138671875, 64.67474365234375, 56.41722106933594, -6.664699554443359, 28.076324462890625, 110.13687133789062, -6.219268798828125, -13.6234130859375, 26.267166137695312, 84.70036315917969, 26.4736328125, 64.60308837890625, 1.7298202514648438, -4.69384765625, 9.744781494140625, 88.92640686035156, -5.845733642578125, 60.030052185058594, 0.3168144226074219, 53.91362762451172, 24.408538818359375, 40.13090515136719, 118.95687866210938, -20.951019287109375, 49.1224365234375, -2.0480384826660156, 79.19021606445312, 115.53976440429688, 5.776649475097656, 49.37566375732422, 37.312591552734375, -2.4530792236328125, -56.33648681640625, 82.56143951416016, -43.42364501953125, 78.4638671875, 11.35770034790039, -2.100921630859375, 60.083770751953125, 2.1737747192382812, -99.94331359863281, 55.3321533203125, -5.639862060546875, -2.850870132446289, -26.520263671875, 3.595794677734375, -32.87017822265625, 3.3611984252929688, 31.455459594726562, 12.056365966796875, -47.800811767578125, 47.575714111328125, 77.27365112304688, 19.873687744140625, 61.022003173828125, 26.42706298828125, 92.32794189453125, 75.95355224609375, 11.705780029296875, 21.619384765625, -7.431060791015625, 63.06219482421875, 8.119701385498047, -8.52972412109375, 36.6044921875, -10.405166625976562, 73.1829833984375, 10.979568481445312, 69.84773254394531, 2.57269287109375, -1.358306884765625, 11.14031982421875, 12.197843551635742, -75.53201293945312, -5.358428955078125, -14.560531616210938, 15.62347412109375, 99.53085327148438, 0.0, 35.053558349609375, 35.50318145751953, 52.553985595703125, 0.37709808349609375, -3.521759033203125, 25.007369995117188, 79.72463989257812, 9.431938171386719, -1.3819427490234375, 33.80859375, -33.2266845703125, 12.36590576171875, 82.1802978515625, 21.942398071289062, -23.678955078125, 11.56869125366211, 24.300064086914062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000203.npy"}
{"epoch": 0.42513089005235605, "step": 204, "batch_size": 128, "mean": 32.24190139770508, "std": 45.48811340332031, "min": -84.88919067382812, "p10": -20.744973754882807, "median": 26.285911560058594, "p90": 91.50243225097655, "max": 137.86502075195312, "pos_frac": 0.765625, "sample": [107.69253540039062, 72.49713134765625, 108.180908203125, 33.97998046875, 13.683631896972656, 86.75689697265625, 5.447330474853516, 40.895408630371094, 4.583110809326172, 100.15472412109375, 12.383056640625, 16.1778564453125, 70.1436767578125, -19.13153076171875, -45.38189697265625, -84.88919067382812, 15.63494873046875, -29.052825927734375, 4.673942565917969, 22.778472900390625, 85.56451416015625, 3.5593185424804688, 40.9176025390625, 66.38163757324219, 26.921142578125, 47.16253662109375, 9.473106384277344, -59.97492980957031, 14.418472290039062, 19.63458251953125, 32.757598876953125, 137.86502075195312, 2.389678955078125, -29.171127319335938, 38.866668701171875, -13.57415771484375, 12.945001602172852, 117.6610107421875, -3.4705810546875, 77.88613891601562, 97.87889099121094, 11.153907775878906, 57.116119384765625, -13.801109313964844, 72.4791259765625, 73.54867553710938, 13.67694091796875, 115.4525146484375, 86.125732421875, 104.78237915039062, 74.86022186279297, 36.17071533203125, -3.1138916015625, 62.719146728515625, 40.26292419433594, -2.3301467895507812, 72.95407104492188, 44.474220275878906, 1.1070404052734375, -28.314666748046875, 30.510528564453125, 71.22146606445312, 54.756103515625, -63.462371826171875, -9.7742919921875, 104.6494140625, 69.85311889648438, 71.421142578125, -71.00970458984375, 80.11099243164062, -7.825315475463867, 74.3675537109375, -2.8460235595703125, -11.811622619628906, 5.862323760986328, 5.9358673095703125, 29.82614517211914, 82.7630844116211, -3.505199432373047, 20.00787353515625, 7.776786804199219, -7.057525634765625, -45.49629211425781, 11.3472900390625, 67.385498046875, -4.34613037109375, 49.23411560058594, 12.171798706054688, 73.75885009765625, 68.03643798828125, 90.142333984375, 44.47941589355469, -39.94525146484375, 96.42816162109375, -8.501829147338867, 1.4632568359375, 81.24896240234375, -37.54998779296875, 33.905426025390625, 88.39013671875, 10.05206298828125, -0.19341278076171875, 13.0013427734375, 21.240264892578125, 65.3341293334961, 98.94732666015625, 25.650680541992188, 24.361099243164062, 94.67599487304688, 14.568901062011719, 67.62654113769531, 0.59613037109375, 54.09716796875, 74.74954223632812, 23.47931671142578, 33.011375427246094, 71.44599914550781, 3.000030517578125, -7.792964935302734, 40.5936279296875, -24.509674072265625, 33.93574523925781, 89.77064514160156, 51.67042541503906, -59.383544921875, 112.43228149414062, 1.5169143676757812, -3.4610595703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000204.npy"}
{"epoch": 0.4272251308900524, "step": 205, "batch_size": 128, "mean": 29.363374710083008, "std": 49.29885482788086, "min": -96.6861572265625, "p10": -44.06981048583984, "median": 27.639007568359375, "p90": 90.44454040527343, "max": 131.13543701171875, "pos_frac": 0.7578125, "sample": [66.89230346679688, 0.0, -4.251495361328125, 3.62249755859375, 10.783105850219727, 77.57321166992188, -43.36871337890625, -53.859161376953125, -45.70570373535156, 1.0400257110595703, 131.13543701171875, 65.52772521972656, 5.6991119384765625, 11.457649230957031, -9.74542236328125, 11.408771514892578, 107.69749450683594, 99.609130859375, 73.45449829101562, 12.071258544921875, 23.289276123046875, 38.580299377441406, -52.181640625, 90.87710571289062, 19.113525390625, 9.412353515625, 35.203765869140625, 16.06580352783203, 93.1185302734375, 46.415252685546875, 73.74966430664062, 51.73419189453125, 3.1911163330078125, 7.89013671875, -0.805938720703125, 7.009548187255859, 62.283241271972656, 9.97705078125, 52.59529113769531, 10.269462585449219, 27.624664306640625, 0.4725799560546875, -46.32830810546875, 83.18601989746094, 12.32220458984375, 74.36529541015625, 0.0, 30.453369140625, -67.49574279785156, 27.653350830078125, 2.8443832397460938, 21.903194427490234, 87.899169921875, 91.04722595214844, -54.46527099609375, 3.6228485107421875, 79.86396789550781, 57.257843017578125, 1.00714111328125, 56.007049560546875, -71.72379302978516, -2.79461669921875, 27.920257568359375, 46.00041198730469, -32.581268310546875, 2.5679931640625, 90.2591552734375, -7.295745849609375, -62.13433837890625, 124.304931640625, 77.8138427734375, 109.81219482421875, -96.6861572265625, 53.074066162109375, -4.778404235839844, 65.51847839355469, 5.73370361328125, 68.75509643554688, 66.4862289428711, -5.11126708984375, 29.653488159179688, 108.94175720214844, 62.900054931640625, 83.09587097167969, 66.16854858398438, 11.895116806030273, -93.84600830078125, 3.0418243408203125, 68.15605926513672, 70.19375610351562, 110.412841796875, 75.27581787109375, 112.608154296875, -22.53173828125, 67.7421875, 95.1324462890625, 81.45722961425781, -16.86833953857422, 15.728168487548828, 2.96478271484375, 110.646728515625, 70.28668212890625, -8.893341064453125, -41.93756103515625, 50.580841064453125, 57.940673828125, 1.149688720703125, 61.6767578125, 64.48799133300781, 77.65440368652344, 22.679641723632812, 14.1715087890625, -77.08795166015625, 7.574485778808594, 47.403564453125, -54.52470397949219, -64.59680938720703, 83.75920104980469, -9.148536682128906, 65.6722412109375, 33.62322998046875, 58.57666015625, -0.1415252685546875, 50.13232421875, 35.66682434082031, 73.87542724609375, 0.0, 31.949928283691406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000205.npy"}
{"epoch": 0.4293193717277487, "step": 206, "batch_size": 128, "mean": 28.832061767578125, "std": 46.818267822265625, "min": -113.27508544921875, "p10": -28.749368286132807, "median": 26.442901611328125, "p90": 88.13540344238281, "max": 111.64273071289062, "pos_frac": 0.765625, "sample": [95.88714599609375, 4.332454681396484, -72.57722473144531, 62.2874755859375, 106.8316650390625, -0.32592010498046875, -19.69140625, 10.321502685546875, 23.89337158203125, 84.9104232788086, 51.7288818359375, 89.60700988769531, 96.77688598632812, -16.773048400878906, -46.3814697265625, 63.17315673828125, 25.9385986328125, 1.5991973876953125, -2.5438919067382812, 88.46829223632812, 1.174875259399414, 75.64370727539062, -0.17136383056640625, -26.629608154296875, 0.3968505859375, 8.012313842773438, -56.497650146484375, 51.442962646484375, 57.29730987548828, 10.040336608886719, 3.51055908203125, -9.503173828125, -64.38807678222656, 70.21940612792969, 70.04544067382812, -12.951274871826172, 104.85197448730469, 86.97491455078125, 109.40081787109375, 37.415138244628906, 45.765716552734375, 33.814605712890625, -5.751007080078125, 26.94720458984375, 57.11724853515625, 43.5517578125, -27.53857421875, 87.40151977539062, 73.59695434570312, 78.70783996582031, 72.71029663085938, 61.450439453125, 64.3787841796875, 46.72998046875, 2.6182632446289062, 40.21583557128906, 34.47145080566406, 103.93548583984375, -113.27508544921875, 20.9615478515625, 78.02114868164062, -53.97826385498047, -15.236923217773438, 71.92134094238281, 38.40409851074219, 12.265899658203125, 33.558135986328125, 27.529510498046875, -4.924421310424805, 43.731719970703125, 9.740875244140625, 71.6165771484375, 80.31484985351562, -72.371337890625, 107.04229736328125, 1.79364013671875, -11.359321594238281, 47.09962463378906, 9.71295166015625, 1.5392684936523438, -37.740325927734375, -31.574554443359375, 93.80325317382812, 51.827178955078125, 68.74699401855469, 10.412841796875, 53.19120788574219, -39.493408203125, 9.861886978149414, 0.03326416015625, 63.079437255859375, 71.25146484375, 64.4998779296875, 6.870655059814453, -26.168121337890625, 55.255859375, 20.56787109375, -11.828834533691406, 2.15771484375, -2.9745559692382812, 71.54896545410156, 16.8800048828125, 99.791015625, 87.99273681640625, 7.51025390625, 76.12646484375, -37.752716064453125, -24.39080810546875, -59.85380554199219, 16.928131103515625, 3.1586456298828125, 111.64273071289062, 74.15837097167969, 4.83349609375, 1.8542747497558594, 99.1328125, 62.392120361328125, 6.471382141113281, 87.50822448730469, 36.486328125, 4.44378662109375, 10.8690185546875, 0.803741455078125, 74.13594055175781, 49.62469482421875, 10.514450073242188, -57.1690673828125, 45.1024169921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000206.npy"}
{"epoch": 0.431413612565445, "step": 207, "batch_size": 128, "mean": 24.45763397216797, "std": 53.04088592529297, "min": -103.22216796875, "p10": -38.80601348876953, "median": 14.71078109741211, "p90": 94.2525146484375, "max": 138.8612060546875, "pos_frac": 0.6953125, "sample": [100.5657958984375, 47.71434020996094, 16.30487060546875, -9.425361633300781, 16.37908935546875, 2.326995849609375, 20.0870361328125, 118.39828491210938, 41.626251220703125, -4.448883056640625, 26.902082443237305, 131.04840087890625, -31.8809814453125, 4.555397033691406, -8.71441650390625, 83.57117462158203, 1.5832138061523438, 52.083038330078125, 125.9805908203125, -11.57745361328125, 81.13024139404297, -12.853652954101562, -92.23184204101562, -27.78583526611328, 70.00895690917969, 29.789932250976562, -13.667747497558594, 13.96630859375, -86.7557373046875, 62.73728942871094, 76.67507934570312, 15.544593811035156, 11.625213623046875, -31.51104736328125, 74.84246826171875, 75.17257690429688, 16.076904296875, -3.4994888305664062, 19.72830581665039, 98.41555786132812, -12.63336181640625, 17.147125244140625, 47.527496337890625, 7.79266357421875, 6.67926025390625, -22.857177734375, 4.128145217895508, 4.7865142822265625, -82.579345703125, -1.4062652587890625, -103.22216796875, -1.080322265625, -50.07977294921875, 25.2042236328125, 85.169677734375, 0.0, 9.845455169677734, -31.15032958984375, 70.63623046875, -25.722213745117188, 108.74420166015625, -10.892929077148438, -43.7960205078125, 25.760833740234375, 14.683364868164062, -90.08584594726562, 95.34931945800781, -69.736328125, 13.401504516601562, -2.216827392578125, 25.15911865234375, -10.2589111328125, 51.33839416503906, 85.65011596679688, 100.646240234375, 92.45082092285156, 65.71900177001953, 4.6298828125, 2.5452117919921875, 91.82884216308594, 0.20908164978027344, -69.24261474609375, -8.5489501953125, 33.865665435791016, 9.523204803466797, 138.8612060546875, 14.738197326660156, 78.58863067626953, -63.005218505859375, 1.3301467895507812, 11.94598388671875, 7.2484588623046875, -35.17333984375, 102.35481262207031, 23.23541259765625, 68.75520324707031, 46.406463623046875, 7.20001220703125, 56.19709014892578, -61.862030029296875, 101.8189697265625, 86.93046569824219, -15.52154541015625, 81.09794616699219, 6.466300964355469, 111.09832763671875, 9.946136474609375, 4.8017578125, 9.890188217163086, 80.71426391601562, 10.153789520263672, 86.20889282226562, -1.86871337890625, 93.78245544433594, 74.11553955078125, -8.303901672363281, 88.98233032226562, 83.38829040527344, 114.36102294921875, -39.965057373046875, 17.469024658203125, -71.17617797851562, 80.67205810546875, 83.78218078613281, -38.30928039550781, 29.160064697265625, 60.99077606201172, 27.6700439453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000207.npy"}
{"epoch": 0.43350785340314135, "step": 208, "batch_size": 128, "mean": 28.37810516357422, "std": 46.038700103759766, "min": -114.13482666015625, "p10": -14.461370468139647, "median": 17.9659423828125, "p90": 85.81981964111327, "max": 141.4459228515625, "pos_frac": 0.734375, "sample": [-3.21527099609375, -8.484207153320312, -26.794403076171875, 17.71179962158203, -86.464599609375, 5.3822174072265625, -6.112464904785156, 23.304901123046875, 5.155853271484375, 81.6614990234375, -10.468063354492188, 8.684051513671875, 56.36347961425781, 47.9964599609375, 62.534297943115234, 120.42254638671875, 16.06304931640625, 11.319488525390625, 17.37274169921875, 131.27554321289062, -67.09536743164062, 80.61575317382812, 15.465179443359375, 40.05552673339844, 15.34979248046875, 69.12158203125, 112.4127197265625, 69.23722839355469, 15.6890869140625, -17.117919921875, 141.4459228515625, 57.579322814941406, 57.269287109375, 75.56634521484375, -13.904159545898438, 104.42010498046875, -114.13482666015625, 15.3870849609375, 17.493621826171875, 27.683074951171875, 43.58683776855469, -98.22756958007812, 77.70841979980469, -15.465679168701172, 81.08404541015625, 50.0279541015625, 65.12855529785156, 105.97494506835938, 0.0, 127.35504150390625, -32.981048583984375, 1.9275712966918945, 30.65106201171875, -26.818435668945312, 2.7283706665039062, 126.847412109375, 12.437667846679688, 2.6593170166015625, 27.24456024169922, 94.45907592773438, 40.6282844543457, 35.193115234375, -10.768058776855469, 51.2259521484375, -4.126312255859375, 5.44305419921875, 6.643768310546875, -0.8147354125976562, 21.2109375, 57.5123291015625, 124.720947265625, 0.0, 27.271469116210938, 56.526519775390625, 27.752193450927734, 84.069091796875, -7.099235534667969, 84.34060668945312, 29.5963134765625, 63.27808380126953, -1.453369140625, 44.57421875, 78.24214172363281, 92.15115356445312, -6.876678466796875, 57.576446533203125, -10.720428466796875, -1.1994609832763672, -0.135040283203125, 40.721832275390625, -11.113555908203125, 24.98480224609375, 51.526580810546875, 12.852691650390625, 66.23565673828125, 29.7491455078125, 73.28482055664062, 32.515625, 98.09866333007812, -14.030952453613281, 2.5579681396484375, 6.448850631713867, 5.344390869140625, 15.50579833984375, 18.22008514404297, 10.310443878173828, -1.3579750061035156, 73.84747314453125, 15.987983703613281, 25.876480102539062, 21.812530517578125, -25.025634765625, 66.77165985107422, 5.090911865234375, 9.87139892578125, -66.87734985351562, 16.20728302001953, 0.46101951599121094, -26.678756713867188, 53.90576171875, 60.703887939453125, 11.531951904296875, -11.989540100097656, 89.27131652832031, 54.459312438964844, -37.8988037109375, -3.598876953125, 49.472900390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000208.npy"}
{"epoch": 0.4356020942408377, "step": 209, "batch_size": 128, "mean": 22.460460662841797, "std": 47.309200286865234, "min": -102.5308837890625, "p10": -32.37318115234375, "median": 12.945934295654297, "p90": 85.40804748535156, "max": 122.51394653320312, "pos_frac": 0.6328125, "sample": [115.66569519042969, -9.62158203125, 73.70052337646484, 73.87519073486328, 19.226837158203125, -4.443267822265625, 26.702972412109375, -32.43829345703125, -38.2711181640625, 18.500823974609375, 80.75006103515625, -12.3111572265625, -9.172454833984375, 6.547767639160156, 108.96923828125, 85.68537139892578, 12.875625610351562, 5.239145278930664, 7.99969482421875, 93.66595458984375, 90.6177978515625, -13.349365234375, 6.17828369140625, -67.26193237304688, 22.544403076171875, 7.6854095458984375, 10.323583602905273, 9.44488525390625, -10.39898681640625, 74.06156158447266, 77.74017333984375, -51.18208312988281, -0.6646881103515625, -56.97657775878906, -45.170135498046875, -53.044891357421875, 116.95919799804688, 4.713348388671875, 0.0, 11.059555053710938, -32.34527587890625, -46.41871643066406, 59.270172119140625, -30.211669921875, 3.86566162109375, 56.58404541015625, 29.078937530517578, 4.584817886352539, -21.6888427734375, 99.89326477050781, 39.90167236328125, 31.805572509765625, 60.7928466796875, -30.928131103515625, 47.319244384765625, 0.9402332305908203, -20.285491943359375, 13.016242980957031, 69.97793579101562, -14.16091537475586, 17.897918701171875, -25.051666259765625, -5.27545166015625, 84.6485595703125, 83.58514404296875, 24.362274169921875, -11.066162109375, -22.4305419921875, 99.4176025390625, -24.2650146484375, 23.3416748046875, 94.14797973632812, 15.974945068359375, 20.065458297729492, -0.372344970703125, 0.0, 59.32012939453125, 74.45094299316406, -21.78900909423828, 74.79840087890625, -14.580841064453125, -6.2905120849609375, 57.63279724121094, -1.420989990234375, 37.26985168457031, 85.36361694335938, 75.89649963378906, -51.47282409667969, 4.78436279296875, 8.69097900390625, 24.853988647460938, -63.172088623046875, -19.480682373046875, 61.99778747558594, -2.155691146850586, -1.58990478515625, 28.552978515625, 49.252174377441406, 122.51394653320312, -102.5308837890625, -11.255447387695312, 5.548343658447266, -41.71722412109375, -6.461204528808594, -2.9136810302734375, -76.0096206665039, -5.815910339355469, 42.63878631591797, 71.44613647460938, 75.346435546875, 27.992050170898438, 79.13920593261719, 72.68730163574219, 19.992660522460938, 59.436981201171875, 45.001678466796875, 37.50361633300781, 79.76260375976562, 1.990478515625, 115.842529296875, 33.512237548828125, -30.50982666015625, 57.57319641113281, 39.71600341796875, 98.05679321289062, 78.9852294921875, -11.35552978515625, 85.51171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000209.npy"}
{"epoch": 0.437696335078534, "step": 210, "batch_size": 128, "mean": 28.687538146972656, "std": 47.5285758972168, "min": -102.57470703125, "p10": -22.995544433593746, "median": 23.148680686950684, "p90": 88.27127532958984, "max": 166.36572265625, "pos_frac": 0.6953125, "sample": [-5.311790466308594, -4.239711761474609, 38.88787841796875, -9.52874755859375, 72.13636779785156, 17.772048950195312, 106.6961669921875, 110.85134887695312, 88.60101318359375, 65.56489562988281, 82.71231079101562, 64.03515625, 5.5551910400390625, 63.200111389160156, 68.2232666015625, -0.851318359375, 65.96771240234375, -5.635585784912109, 61.66798400878906, 3.1993560791015625, 20.625307083129883, -0.088134765625, -60.748504638671875, 29.338409423828125, 62.62476348876953, 52.7490234375, 7.46160888671875, 166.36572265625, -53.216949462890625, -6.8779296875, 52.597412109375, 26.8812255859375, 17.153656005859375, -0.8639068603515625, 75.14251708984375, -14.076995849609375, 5.15593147277832, 52.74072265625, 0.8827362060546875, 31.185302734375, 0.420318603515625, 75.97322082519531, 17.503725051879883, -7.572906494140625, 13.076820373535156, 62.3087158203125, 28.6953125, -1.4174652099609375, 73.49784851074219, 12.573541641235352, 24.948974609375, -2.1781005859375, 77.25839233398438, 57.4647216796875, 41.566505432128906, 63.48424530029297, 5.178680419921875, -51.76593017578125, 99.06561279296875, -6.912384033203125, 55.0185546875, 63.173004150390625, 80.86517333984375, 68.46368408203125, 24.413360595703125, 38.12092590332031, -84.68359375, -7.7559661865234375, 11.098388671875, 82.34808349609375, -68.15812683105469, -29.416290283203125, 87.6492919921875, 82.23977661132812, 94.43075561523438, 88.12995910644531, -0.05240440368652344, 8.332870483398438, -10.913421630859375, -102.57470703125, 72.2625503540039, 10.286148071289062, -1.5926685333251953, 97.07666015625, 46.66630554199219, 80.78652954101562, 93.41571044921875, 45.486724853515625, -2.963623046875, -22.338104248046875, -20.823486328125, 13.422355651855469, 2.6769065856933594, -28.897476196289062, 23.704710006713867, 67.1634521484375, -24.529571533203125, 110.872314453125, -17.34222412109375, 54.577117919921875, 58.617645263671875, 120.29473876953125, 7.15754508972168, -11.619659423828125, 99.49969482421875, 22.5926513671875, 92.84492492675781, -28.557830810546875, 10.617828369140625, -16.50885009765625, 59.169281005859375, 23.8875732421875, -3.159423828125, 17.425445556640625, 17.401824951171875, -93.26303100585938, 14.281326293945312, 89.88482666015625, 30.30322265625, -74.69776916503906, 70.64276123046875, 1.36212158203125, 74.6859130859375, 54.6361083984375, -25.539215087890625, -8.2593994140625, -5.97802734375, 55.93775939941406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000210.npy"}
{"epoch": 0.4397905759162304, "step": 211, "batch_size": 128, "mean": 37.755889892578125, "std": 43.63819122314453, "min": -69.31611633300781, "p10": -10.410014343261718, "median": 41.67901611328125, "p90": 94.22625122070312, "max": 139.02426147460938, "pos_frac": 0.78125, "sample": [61.42510986328125, 26.868667602539062, 107.09814453125, 46.77008056640625, 74.01363372802734, 91.42913818359375, 13.013708114624023, 6.501930236816406, -12.952590942382812, 62.2000732421875, 85.46755981445312, -62.836700439453125, 40.6614990234375, 42.843719482421875, -11.091567993164062, 84.70895385742188, 0.0351715087890625, -1.206186294555664, -6.9400634765625, -17.860397338867188, 10.013206481933594, 68.13762664794922, 16.027488708496094, 27.03900146484375, 71.98258972167969, 26.979583740234375, 7.9626312255859375, -56.3612060546875, 70.77381896972656, -10.117919921875, 3.8197669982910156, 74.06927490234375, -0.2739715576171875, 66.40702056884766, -4.394783020019531, 56.118377685546875, 45.504859924316406, -69.31611633300781, 49.23841857910156, 75.035888671875, 4.80657958984375, 50.27146911621094, 9.82281494140625, -1.461883544921875, -3.5388641357421875, 118.7945556640625, 90.6795654296875, 56.9888916015625, 32.89839172363281, 9.605045318603516, 65.17971801757812, -13.42169189453125, 7.582916259765625, -0.9968414306640625, 139.02426147460938, 74.76223754882812, 94.31802368164062, -21.899749755859375, 20.974761962890625, 61.8045654296875, -7.85174560546875, -19.999481201171875, 12.869735717773438, 126.78131103515625, 5.0759735107421875, 5.1410369873046875, -4.37530517578125, 71.02180480957031, 94.39462280273438, 14.908027648925781, -24.52325439453125, -0.45269775390625, -2.641254425048828, 65.74164581298828, -1.6971015930175781, 11.545310020446777, 74.50552368164062, 63.678863525390625, 72.9411392211914, 46.343017578125, 54.018463134765625, 57.50677490234375, -0.180419921875, 79.456298828125, 67.50796508789062, 124.57235717773438, 20.24749755859375, -0.9537353515625, 67.75757598876953, 7.062232971191406, -12.889434814453125, 120.54129028320312, 101.36538696289062, -32.705352783203125, 98.34375, 19.60467529296875, 87.62083435058594, 50.7855224609375, 11.075021743774414, 4.415191650390625, 73.13128662109375, 107.302978515625, 42.696533203125, 6.889026641845703, -67.07987976074219, 3.9004669189453125, 44.592742919921875, 43.61114501953125, 100.43353271484375, 49.97535705566406, 92.2452392578125, 11.956720352172852, 85.80322265625, 37.3482666015625, 4.78605842590332, 34.122406005859375, 105.99227905273438, 94.18692016601562, 65.68637084960938, 70.88526916503906, 24.649658203125, 47.707122802734375, 0.12746620178222656, 85.638427734375, 53.89002227783203, 42.93896484375, 1.3744659423828125, 84.41265106201172], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000211.npy"}
{"epoch": 0.4418848167539267, "step": 212, "batch_size": 128, "mean": 27.284339904785156, "std": 47.77200698852539, "min": -121.6640625, "p10": -31.768344116210937, "median": 22.434097290039062, "p90": 88.63667602539061, "max": 126.316162109375, "pos_frac": 0.71875, "sample": [3.7533416748046875, -4.575469970703125, 31.08917236328125, 57.562217712402344, -13.939697265625, -41.979400634765625, 1.07000732421875, 15.35845947265625, 3.6229705810546875, -0.00067138671875, 99.68878936767578, -0.30670166015625, 121.31210327148438, 2.8914852142333984, 33.47265625, -46.35986328125, 12.602182388305664, 71.16720581054688, 74.36256408691406, 32.54022216796875, 87.98974609375, 18.4368896484375, -35.7069091796875, 20.6973876953125, 78.85787963867188, 15.87060546875, -14.205535888671875, -73.1810531616211, 25.67755126953125, 57.24726867675781, -22.519023895263672, -55.55921173095703, 4.1013336181640625, 47.3604736328125, 17.071693420410156, 59.610107421875, 24.395355224609375, 92.75808715820312, 50.91746520996094, -5.4527740478515625, 70.81048583984375, -90.31356811523438, -5.718475341796875, -2.2021255493164062, -8.37322998046875, 75.01902770996094, 74.04873657226562, 73.89761352539062, -3.257965087890625, -55.489776611328125, 65.71858978271484, 19.856842041015625, -32.146820068359375, 104.15005493164062, 40.3831787109375, -6.79913330078125, -13.604782104492188, 69.7564697265625, 12.078582763671875, -3.1280479431152344, -7.924224853515625, 6.4599761962890625, 74.13609313964844, 31.71514892578125, -9.788246154785156, 74.46871948242188, -11.542030334472656, 58.004730224609375, 52.6090087890625, 90.34480285644531, 44.565277099609375, 77.34441375732422, -58.09991455078125, 28.044189453125, 81.30793762207031, 13.727790832519531, 53.92041015625, 71.70443725585938, -1.2380962371826172, 101.49929809570312, 47.83154296875, 105.169677734375, 18.701934814453125, 56.1258544921875, -108.48521423339844, 23.4609375, -3.3672943115234375, -62.99371337890625, 32.94541931152344, 7.712223052978516, 77.22860717773438, 42.7615966796875, 11.780853271484375, 6.329730987548828, 117.853759765625, 65.77302551269531, 21.407257080078125, 91.81543731689453, 59.45409393310547, 54.218719482421875, 32.63887023925781, 126.316162109375, 52.15766906738281, 0.8858489990234375, 20.91375732421875, -121.6640625, 82.131103515625, 108.710693359375, -31.60614013671875, -13.039260864257812, 7.982330322265625, -37.077880859375, 34.904266357421875, 7.536455154418945, 58.95330810546875, -2.2852020263671875, 33.713165283203125, 9.017929077148438, 65.01741027832031, 2.3067398071289062, 90.14617919921875, 11.06817626953125, 62.62035369873047, 84.04036712646484, 6.572723388671875, -1.6828765869140625, 106.38265991210938, 26.3662109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000212.npy"}
{"epoch": 0.44397905759162304, "step": 213, "batch_size": 128, "mean": 27.24388885498047, "std": 47.391849517822266, "min": -77.64007568359375, "p10": -24.201354980468743, "median": 11.854785919189453, "p90": 98.19454345703124, "max": 138.130859375, "pos_frac": 0.671875, "sample": [102.50555419921875, 73.45603942871094, -19.44910430908203, 85.30599975585938, -34.4788818359375, 67.07745361328125, 75.58274841308594, 73.00286865234375, -4.732421875, 31.048370361328125, -2.06878662109375, 9.992706298828125, 21.0430908203125, -21.297523498535156, 6.280405044555664, 125.57708740234375, -9.412017822265625, 20.153045654296875, 23.17724609375, 72.20159912109375, 40.9700927734375, -7.368255615234375, 133.98843383789062, -4.740234375, 98.63078308105469, 5.44287109375, -5.87750244140625, 15.713623046875, 18.27393341064453, 14.748336791992188, 7.492523193359375, -29.66229248046875, -16.47296142578125, 7.231170654296875, 87.19549560546875, 37.92976379394531, 77.52922058105469, -0.3486328125, 8.044540405273438, -22.09454345703125, 86.21806335449219, 45.22715759277344, 38.40765380859375, 72.13479614257812, -22.03131103515625, 78.67852783203125, 39.56842041015625, 8.0440673828125, -41.502838134765625, 20.56292724609375, 6.449516296386719, 79.32257080078125, 98.21005249023438, 2.2441024780273438, 121.10406494140625, -42.097412109375, -3.8253402709960938, 70.15756225585938, -36.21617126464844, 98.18789672851562, 15.8502197265625, 15.571044921875, 25.374832153320312, -37.07792663574219, 8.70867919921875, 33.99456787109375, 106.4027099609375, -45.515045166015625, -18.455718994140625, -8.859832763671875, 74.02018737792969, -14.95166015625, 26.880126953125, -77.64007568359375, 27.43157958984375, 13.020095825195312, 52.922706604003906, 92.62841796875, 8.222759246826172, 25.54998779296875, -19.851654052734375, 7.78703498840332, -2.7067089080810547, -12.09765625, -2.1458358764648438, 105.71405029296875, -36.99296569824219, -5.3023681640625, 40.26873779296875, 121.2093505859375, 69.1766586303711, 97.67892456054688, 5.342508316040039, 6.38836669921875, 39.96636962890625, -3.596639633178711, 99.33279418945312, -17.963058471679688, -34.27130126953125, 138.130859375, 1.0621261596679688, 9.676490783691406, -3.6927490234375, 91.66989135742188, -34.895660400390625, 111.6688232421875, 80.69473266601562, -5.566459655761719, 51.621246337890625, 0.0, 9.849958419799805, 37.668121337890625, 9.865859985351562, 97.36906433105469, 10.689476013183594, -73.2574462890625, 85.40536499023438, 38.79111862182617, -4.020683288574219, 21.74163818359375, 2.6194076538085938, 77.722900390625, -29.11724853515625, 7.051116943359375, 111.6845703125, -2.4221954345703125, -19.882537841796875, 2.6395797729492188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000213.npy"}
{"epoch": 0.44607329842931936, "step": 214, "batch_size": 128, "mean": 24.260433197021484, "std": 47.25223922729492, "min": -95.11557006835938, "p10": -30.533837890624998, "median": 20.16242218017578, "p90": 86.61103515625, "max": 154.48062133789062, "pos_frac": 0.703125, "sample": [57.03082275390625, 66.432861328125, -24.993736267089844, 5.1319580078125, 11.999847412109375, 87.8626708984375, 21.567031860351562, -14.814468383789062, -55.13238525390625, -16.049102783203125, -2.91339111328125, 63.79850769042969, -31.742431640625, 32.268524169921875, 45.4505615234375, 69.62876892089844, -95.11557006835938, 73.92375946044922, 32.007781982421875, 36.45501708984375, -5.458930969238281, -16.106719970703125, 20.232406616210938, 63.9637451171875, 37.11737060546875, 19.356689453125, 65.9862289428711, -22.59417724609375, 7.9559326171875, 6.755943298339844, 5.8044891357421875, 12.303863525390625, 10.173067092895508, 2.961456298828125, 68.01499938964844, 20.092437744140625, 154.48062133789062, 52.57984924316406, 96.46243286132812, -74.590087890625, 28.655044555664062, -26.5977783203125, 3.579710006713867, 46.346435546875, 5.6793365478515625, 2.0155487060546875, -16.288330078125, 86.08544921875, 20.61419677734375, -6.975406646728516, 65.53734588623047, 23.25616455078125, -32.959228515625, 48.13385009765625, 5.07220458984375, 81.65382385253906, 36.100341796875, 49.2542724609375, -12.905364990234375, 81.04898071289062, 67.28633117675781, 109.49996948242188, 59.96946716308594, 10.191558837890625, 6.3923492431640625, -23.4383544921875, -4.3425750732421875, -36.95545959472656, 0.41845703125, 0.0, -38.916900634765625, 8.395797729492188, -12.015838623046875, 9.97906494140625, 66.54971313476562, -0.0529327392578125, 53.01918029785156, -57.18977737426758, 6.841758728027344, 11.98472785949707, -17.306121826171875, -85.01849365234375, 87.83740234375, 5.259490966796875, 71.25762939453125, -30.015869140625, 92.51698303222656, 55.70716857910156, 20.49310302734375, 15.571929931640625, -83.48135375976562, 42.960906982421875, 27.398956298828125, 106.23733520507812, 85.22317504882812, 11.764312744140625, -8.229949951171875, 43.3575439453125, -3.676544189453125, 38.18031311035156, 83.46690368652344, 113.47772216796875, 31.087615966796875, 49.47383117675781, -9.430511474609375, 95.75119018554688, 88.3006591796875, 29.430999755859375, -25.032058715820312, 45.91973876953125, 135.364990234375, 24.167999267578125, -22.109588623046875, -40.619903564453125, -5.996612548828125, 90.56681823730469, 20.989946365356445, -73.4249267578125, 10.0262451171875, -2.500274658203125, 19.40478515625, 56.46495056152344, 22.2891845703125, 53.08845520019531, -32.539710998535156, 149.8956298828125, 49.709510803222656, 58.8623046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000214.npy"}
{"epoch": 0.4481675392670157, "step": 215, "batch_size": 128, "mean": 25.35620880126953, "std": 50.963993072509766, "min": -109.31614685058594, "p10": -38.86772003173828, "median": 22.75360107421875, "p90": 87.65583114624022, "max": 154.49285888671875, "pos_frac": 0.671875, "sample": [14.245061874389648, 33.833160400390625, 42.6502685546875, 79.90196228027344, 112.900634765625, 71.771728515625, 83.51605224609375, 43.15386199951172, 66.43280029296875, -7.7667236328125, 104.55618286132812, 77.76651000976562, 97.35916137695312, -26.620941162109375, 3.25079345703125, 30.72381591796875, -9.822052001953125, 131.529052734375, 133.89617919921875, 72.67533874511719, 78.6309814453125, -1.048126220703125, 0.0, -27.5396728515625, -3.9775619506835938, -10.490249633789062, 42.577392578125, 69.89166259765625, 128.54461669921875, -10.201431274414062, 81.44239807128906, -1.35943603515625, -14.811065673828125, -46.75556945800781, 2.297117233276367, 10.447601318359375, 82.56171417236328, -0.80694580078125, 22.38909912109375, 122.58926391601562, -30.670166015625, 26.25006103515625, -29.92340087890625, -34.0384521484375, -0.005680084228515625, -10.012588500976562, -71.26400756835938, 43.942222595214844, 74.28839111328125, 58.83985137939453, 45.22816467285156, 154.49285888671875, 20.25750732421875, 25.1536865234375, -53.227630615234375, 0.6566085815429688, 8.473678588867188, -41.960693359375, 0.0, 78.37533569335938, 1.8483734130859375, -3.3540115356445312, 45.518890380859375, -37.54216003417969, 51.652618408203125, 32.92809295654297, -65.15711975097656, 1.9672927856445312, 92.85833740234375, 34.53924560546875, 49.827178955078125, 8.498350143432617, -5.6099853515625, 82.122314453125, -59.749969482421875, 72.81002807617188, 8.128662109375, -109.31614685058594, 58.28251647949219, 42.31268310546875, 15.261001586914062, 35.90034484863281, 19.611724853515625, -18.0823974609375, -58.56724548339844, -0.7050189971923828, 74.35476684570312, 14.68780517578125, 91.65884399414062, -36.619415283203125, -13.560447692871094, 31.744476318359375, 25.05921173095703, -7.84027099609375, 6.4152679443359375, 12.632965087890625, 0.32756614685058594, 26.826690673828125, 112.42591857910156, 148.11160278320312, -32.940673828125, 26.292747497558594, -69.26604461669922, 0.49837303161621094, -46.980010986328125, -52.38385009765625, 55.175010681152344, 85.94025421142578, -60.83100891113281, 83.45074462890625, 26.208770751953125, -7.716226577758789, 42.61483383178711, -6.637542724609375, 13.396347045898438, 30.915435791015625, 27.53265380859375, 82.20040893554688, -53.41845703125, 28.276611328125, 37.54689025878906, 0.800811767578125, 15.16412353515625, 95.0545654296875, 23.11810302734375, 65.91023254394531, 25.509033203125, 82.76341247558594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000215.npy"}
{"epoch": 0.450261780104712, "step": 216, "batch_size": 128, "mean": 28.61859130859375, "std": 54.347652435302734, "min": -136.233642578125, "p10": -42.18970031738281, "median": 26.142791748046875, "p90": 101.00572509765625, "max": 139.31903076171875, "pos_frac": 0.7421875, "sample": [37.1290283203125, 55.9361572265625, 63.33061218261719, 32.705589294433594, 16.11962890625, -40.33930969238281, -136.233642578125, 22.791397094726562, 68.01528930664062, -3.3007659912109375, 76.50223541259766, 85.26980590820312, 65.255615234375, -10.33416748046875, 18.591079711914062, 60.05570983886719, 139.31903076171875, 9.278446197509766, 103.29653930664062, 114.97116088867188, 24.11651611328125, -86.29827880859375, 69.21908569335938, 0.1211395263671875, -11.56869125366211, 19.1070556640625, -92.50823974609375, 13.635848999023438, 54.0411376953125, 82.63641357421875, -52.11155700683594, -50.68934631347656, -105.35894775390625, 2.7205886840820312, 67.31973266601562, 6.498146057128906, 82.11764526367188, -45.8421630859375, 107.91436767578125, 75.98419952392578, -9.08319091796875, 86.39019775390625, 118.86236572265625, 78.04108428955078, 81.45787048339844, 60.525299072265625, 35.2950439453125, 36.28509521484375, 4.935089111328125, 0.32904815673828125, -46.00274658203125, -5.522041320800781, 128.02597045898438, 46.827972412109375, 6.101371765136719, 56.659271240234375, 11.163299560546875, 104.01239013671875, -10.940887451171875, 3.9942893981933594, 0.0, 64.5135498046875, -22.772064208984375, 60.556640625, 103.60177612304688, 90.05889892578125, 96.92434692382812, -7.753814697265625, -2.9178466796875, 14.0860595703125, 54.78077697753906, 39.8216552734375, -114.08966064453125, 16.59735107421875, 109.68623352050781, 101.75387573242188, 7.139007568359375, 100.68508911132812, 54.31536865234375, 11.649566650390625, 25.62347412109375, 32.498199462890625, 92.05496215820312, 26.662109375, 107.30154418945312, 73.27626037597656, -49.834014892578125, -9.4114990234375, 107.8074951171875, 85.09246826171875, -40.624359130859375, -70.1944580078125, 2.356689453125, 53.576904296875, -3.12353515625, -17.6990966796875, 21.334228515625, -16.829376220703125, 29.568344116210938, 20.57025146484375, 117.11154174804688, 11.966682434082031, 0.0, 64.49701690673828, 76.671630859375, -73.14064025878906, 2.5421905517578125, 72.08299255371094, 65.87615966796875, 47.08100128173828, 0.2183246612548828, 49.215118408203125, 80.351318359375, -9.71209716796875, 42.850135803222656, -4.877471923828125, 32.819366455078125, 39.01898193359375, 5.55242919921875, 49.954437255859375, 2.04541015625, -101.10415649414062, 50.48918151855469, 79.35871887207031, 7.51824951171875, 0.71099853515625, -24.76959228515625, 3.4375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000216.npy"}
{"epoch": 0.4523560209424084, "step": 217, "batch_size": 128, "mean": 38.26862335205078, "std": 52.967472076416016, "min": -92.99932861328125, "p10": -21.779080963134763, "median": 32.15419006347656, "p90": 104.08720550537109, "max": 196.7806396484375, "pos_frac": 0.765625, "sample": [10.412620544433594, -1.9331207275390625, -5.252685546875, 77.87796020507812, 102.74647521972656, -73.06816101074219, 3.0142974853515625, 96.59552001953125, 7.0316619873046875, 35.58355712890625, 98.19209289550781, -6.44549560546875, 1.9972305297851562, 32.708465576171875, -17.91357421875, 96.78567504882812, 72.90910339355469, 81.11004638671875, 75.86199951171875, 22.62725830078125, 64.70167541503906, 74.52178955078125, 5.65655517578125, 69.3204345703125, 162.2940673828125, 66.25689697265625, 82.22055053710938, 86.9234848022461, 86.27548217773438, 98.1324691772461, 31.59991455078125, 62.644012451171875, -9.745101928710938, 123.530029296875, 57.81907653808594, 26.169525146484375, 196.7806396484375, 78.34246826171875, 90.31365966796875, 44.99432373046875, -92.5279541015625, 89.6656494140625, 39.629150390625, 145.46078491210938, -76.81529235839844, 51.845497131347656, 12.223297119140625, -24.993446350097656, 41.55755615234375, 20.49420166015625, 4.306571960449219, 2.0733642578125, 8.429473876953125, 39.61798095703125, 54.159881591796875, -48.057403564453125, -3.3881988525390625, 83.74215698242188, 62.287818908691406, -17.45306396484375, -2.8680953979492188, 43.3046875, -8.826705932617188, 119.85195922851562, 4.9362640380859375, 30.061599731445312, 12.278522491455078, 95.98115539550781, 71.89228057861328, 13.533981323242188, 73.47836303710938, 79.0604476928711, 35.783172607421875, -3.507068634033203, 97.53680419921875, -8.100311279296875, -27.144790649414062, -22.34571075439453, 57.02191162109375, 154.44247436523438, 9.02789306640625, 107.215576171875, 21.83538055419922, -11.628475189208984, -92.99932861328125, 120.35284423828125, 108.1015625, 109.2032470703125, 16.99212646484375, -64.15399169921875, 16.103851318359375, 88.70767211914062, -12.6915283203125, 80.61282348632812, 1.560333251953125, 57.56689453125, -1.1632614135742188, 11.259780883789062, 7.401317596435547, 76.97006225585938, -31.480117797851562, 0.0, 15.635101318359375, 38.16447448730469, -28.158905029296875, 123.06990051269531, 76.97946166992188, -28.782470703125, 32.797943115234375, 116.93403625488281, 59.133087158203125, 26.070053100585938, 28.0428466796875, 130.10858154296875, 80.33139038085938, 11.178756713867188, 7.060157775878906, 21.51446533203125, -30.08526611328125, 10.039375305175781, 65.4024658203125, 28.292762756347656, 51.15252685546875, 17.668533325195312, 33.88246154785156, -21.536239624023438, -7.98675537109375, 4.458835601806641], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000217.npy"}
{"epoch": 0.4544502617801047, "step": 218, "batch_size": 128, "mean": 30.555227279663086, "std": 47.933837890625, "min": -100.19482421875, "p10": -25.268212890624998, "median": 20.717750549316406, "p90": 89.74112701416016, "max": 146.38076782226562, "pos_frac": 0.7421875, "sample": [42.685699462890625, 15.741950988769531, 3.3870849609375, 95.4732666015625, -3.584259033203125, -5.61541748046875, -11.45831298828125, 107.06451416015625, 18.44048309326172, 8.564582824707031, -67.21954345703125, 9.389175415039062, 40.02607727050781, 58.11627197265625, 45.852783203125, -100.19482421875, 11.3486328125, 77.96642303466797, 77.4044189453125, 61.914703369140625, -12.4625244140625, 58.924072265625, 83.72528076171875, 27.69158935546875, 34.250579833984375, 26.31866455078125, 11.550994873046875, -10.3624267578125, -26.514404296875, 63.48841857910156, 67.2386245727539, 119.11358642578125, 5.6659698486328125, 86.77254486083984, -24.734130859375, 12.77105712890625, 74.92732238769531, -27.5242919921875, 76.74745178222656, 63.988739013671875, 140.2802734375, 51.817405700683594, 20.899917602539062, 1.7176895141601562, -29.982101440429688, 90.53996276855469, 8.084583282470703, -62.268646240234375, -44.000213623046875, 87.15423583984375, 42.16108703613281, -22.02649688720703, 75.92788696289062, 28.288223266601562, 94.06402587890625, -2.1002063751220703, 88.90362548828125, 18.149810791015625, 87.73822021484375, -20.1746826171875, 11.22021484375, -5.580474853515625, -31.764129638671875, 11.443634033203125, 0.0, 60.91400146484375, 18.631561279296875, 2.63519287109375, -68.35189819335938, 83.015869140625, 16.097412109375, 33.84477233886719, -60.00801467895508, 88.91317749023438, 32.76580810546875, 86.0089111328125, -0.19951629638671875, 13.194412231445312, 25.22259521484375, 1.421478271484375, 146.38076782226562, 89.65737915039062, 113.25579833984375, 5.4952392578125, -4.852813720703125, 18.746795654296875, -3.6461181640625, -0.20854949951171875, 106.7613525390625, 32.34759521484375, 15.510116577148438, 64.49139404296875, 19.11492919921875, -70.76580810546875, 79.67285919189453, 75.31461334228516, 89.93653869628906, 91.51593017578125, -41.99151611328125, 45.065673828125, 5.48846435546875, 74.86715698242188, -7.695598602294922, 25.578765869140625, 51.11163330078125, -10.587142944335938, -89.1805191040039, 59.15596008300781, 10.765625, 70.53445434570312, 10.08978271484375, 17.6580810546875, 45.418701171875, 85.43380737304688, -0.14300537109375, 20.53558349609375, 115.39352416992188, 4.635284423828125, -2.205810546875, -14.130401611328125, 91.55953979492188, 64.18270111083984, 63.028533935546875, 20.46331787109375, 45.69744873046875, 79.62347412109375, 21.808547973632812, 8.698532104492188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000218.npy"}
{"epoch": 0.45654450261780105, "step": 219, "batch_size": 128, "mean": 31.367931365966797, "std": 49.9495735168457, "min": -116.15863037109375, "p10": -16.95995483398437, "median": 23.629241943359375, "p90": 97.44316558837889, "max": 131.0931396484375, "pos_frac": 0.7578125, "sample": [-116.15863037109375, 21.329437255859375, 49.333831787109375, 10.51910400390625, 89.59771728515625, 29.952903747558594, -14.524566650390625, 55.6512451171875, 83.89984130859375, 26.712158203125, 58.297271728515625, -0.5101032257080078, 40.2841796875, -52.777313232421875, -0.17410659790039062, 74.45040130615234, 56.793609619140625, -12.076904296875, 18.370285034179688, 72.53398132324219, 72.41302490234375, 7.554462432861328, 42.22357177734375, 114.90814208984375, -2.1340408325195312, 23.531402587890625, 79.87760925292969, 3.9093399047851562, 76.26631164550781, -23.333541870117188, -8.931442260742188, -4.564167022705078, -89.06809997558594, 40.62773895263672, 10.672271728515625, 90.33502197265625, -5.172882080078125, 46.223140716552734, 22.061004638671875, 87.13200378417969, 104.52587890625, 131.0931396484375, 7.556110382080078, 53.996612548828125, 6.335784912109375, -80.26382446289062, 11.583221435546875, 68.96710205078125, 11.469329833984375, 100.19175720214844, 0.0, 23.868942260742188, 17.394638061523438, 26.33782958984375, -38.97700500488281, 103.60597229003906, -23.5673828125, 96.26519775390625, 107.88674926757812, 16.108352661132812, 58.24058532714844, 54.31640625, 14.581985473632812, 5.402099609375, 1.6950578689575195, -10.00665283203125, 3.281829833984375, 5.613130569458008, 1.3390350341796875, 63.16429138183594, -12.158966064453125, 20.999786376953125, 10.071517944335938, 115.56869506835938, 15.4146728515625, -0.9245758056640625, -81.47328186035156, 16.089111328125, 58.717864990234375, 93.22634887695312, -2.4443912506103516, 94.03656005859375, 23.62579345703125, 27.10980224609375, -5.241510391235352, -36.36274719238281, 111.23165893554688, 4.124210357666016, 110.35189819335938, -1.9451751708984375, 115.38128662109375, 61.4222412109375, 45.02592468261719, 70.3157958984375, 73.1954574584961, 84.98725891113281, 33.903968811035156, 49.884124755859375, 27.3927001953125, 89.30880737304688, -15.676193237304688, 85.16741943359375, 80.15383911132812, -56.18132019042969, 94.66348266601562, 59.823699951171875, 53.962066650390625, 4.0419158935546875, 23.6326904296875, 21.189422607421875, 37.9234733581543, -5.4725341796875, 44.709747314453125, -107.871826171875, 129.24264526367188, 103.7691650390625, 46.54264831542969, -19.955398559570312, 14.435592651367188, 14.020492553710938, 49.9998779296875, 18.31878662109375, 53.2413330078125, 11.44793701171875, 129.046875, -10.557487487792969, 2.4507064819335938, -75.850341796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000219.npy"}
{"epoch": 0.4586387434554974, "step": 220, "batch_size": 128, "mean": 33.91455841064453, "std": 51.678924560546875, "min": -122.96578979492188, "p10": -27.493283081054685, "median": 35.17585754394531, "p90": 103.17625122070312, "max": 139.94140625, "pos_frac": 0.71875, "sample": [96.66802978515625, 113.10134887695312, 62.114715576171875, 63.020164489746094, 22.341995239257812, 68.127685546875, 3.801849365234375, 81.7738037109375, 19.90154266357422, 68.7442626953125, -11.277122497558594, 2.941558837890625, 74.8158187866211, -57.60235595703125, 10.888702392578125, 33.605133056640625, 34.085296630859375, 103.75213623046875, -26.3670654296875, 13.681182861328125, 57.72772216796875, -7.689239501953125, 24.648582458496094, 1.967458724975586, 0.4381141662597656, 139.94140625, -6.417938232421875, 94.20130920410156, -69.54876708984375, 21.056533813476562, 53.419769287109375, 5.243804931640625, -8.705078125, -30.996826171875, 31.264007568359375, 8.128692626953125, -23.58282470703125, 61.847015380859375, 1.977243423461914, 63.9696044921875, 51.371543884277344, 31.251708984375, 90.526123046875, 50.294830322265625, 108.10377502441406, 62.95948791503906, -5.329536437988281, -122.96578979492188, 64.5618896484375, 42.3697509765625, 55.32220458984375, 77.547607421875, -10.214599609375, 110.48284912109375, -36.29601287841797, 16.969093322753906, -10.945098876953125, -22.627098083496094, 7.3284912109375, 55.280067443847656, -24.029632568359375, 65.1708984375, 58.606781005859375, 50.61968231201172, 8.903144836425781, -13.545982360839844, -8.690216064453125, 54.27197265625, 2.2093048095703125, 84.66311645507812, 99.39775085449219, 112.80128479003906, -1.7938995361328125, 24.873931884765625, 72.07968139648438, 70.07843017578125, -21.737945556640625, 68.25489807128906, 72.10042572021484, -10.854156494140625, 2.2677459716796875, 15.360748291015625, 79.15899658203125, 35.4715576171875, -4.7169036865234375, 113.70995330810547, -85.88485717773438, 9.170013427734375, 122.94168090820312, 125.19232177734375, -31.0164794921875, 34.880157470703125, -81.277587890625, -1.1804695129394531, 60.59332275390625, 116.50872802734375, 102.929443359375, 17.533615112304688, 72.05612182617188, -17.107933044433594, 70.143798828125, 59.868255615234375, -38.708770751953125, -43.450225830078125, -0.7881698608398438, -13.423965454101562, 76.06887817382812, -30.121124267578125, 110.20516967773438, 106.09376525878906, 54.62571716308594, 46.310394287109375, 59.87127685546875, -34.17878723144531, 121.9119873046875, -14.3438720703125, 94.32135009765625, 68.30984497070312, 11.6622314453125, 81.3646240234375, -90.84614562988281, 55.82928466796875, 75.13883209228516, 54.56536865234375, 80.48699951171875, -11.0772705078125, 89.49186706542969, 38.761871337890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000220.npy"}
{"epoch": 0.4607329842931937, "step": 221, "batch_size": 128, "mean": 25.404136657714844, "std": 51.029258728027344, "min": -111.25885009765625, "p10": -35.579679107665996, "median": 17.454761505126953, "p90": 92.968701171875, "max": 124.42898559570312, "pos_frac": 0.71875, "sample": [2.7484130859375, 55.244415283203125, -23.6622314453125, 21.130950927734375, 44.16465759277344, 9.3511962890625, 19.498809814453125, 48.52946472167969, -13.79803466796875, -9.146982192993164, 62.99274444580078, 51.541961669921875, 3.06182861328125, 59.7254638671875, -4.2886199951171875, 79.28466796875, 14.409637451171875, -64.52354431152344, 7.514457702636719, -12.893600463867188, 67.9183349609375, 38.45135498046875, 124.42898559570312, -24.647354125976562, 29.8963623046875, 89.41753387451172, 95.4658203125, 67.53585815429688, 84.45458984375, 0.701080322265625, 15.717529296875, 12.893218994140625, 112.45330810546875, -111.25885009765625, -50.859893798828125, 12.946807861328125, 46.558929443359375, 76.28076171875, -11.090591430664062, 112.4024658203125, -49.62816619873047, 10.808013916015625, -7.309478759765625, 29.5264892578125, -70.09727478027344, -5.77227783203125, 88.10527038574219, -62.537139892578125, 77.65333557128906, 33.229827880859375, 0.09389495849609375, 82.46614074707031, 21.26702880859375, -17.735992431640625, -5.889049530029297, 63.33294677734375, 58.439788818359375, -60.21160888671875, 1.447723388671875, 53.81785583496094, 8.6451416015625, 15.719619750976562, 58.616485595703125, 10.397317886352539, 53.1405029296875, -0.23977279663085938, 70.91650390625, 85.92460632324219, -2.585540771484375, 72.86198425292969, 75.22140502929688, -15.631195068359375, 54.149322509765625, 52.316673278808594, 100.02775573730469, 60.84527587890625, 102.9598388671875, 23.673324584960938, -60.21344757080078, -79.99765014648438, 92.5811767578125, 93.8729248046875, -17.11962890625, 5.431060791015625, -8.57171630859375, 89.18057250976562, 10.311370849609375, -15.38623046875, 30.368438720703125, 70.17160034179688, 111.81866455078125, 98.75091552734375, 0.488677978515625, 6.89344596862793, -57.6925048828125, 5.418636322021484, -15.738418579101562, 114.65609741210938, -99.1151123046875, 39.821258544921875, 4.1087493896484375, -74.55303955078125, -28.44127655029297, 112.83975219726562, 15.694244384765625, 13.24169921875, 19.189903259277344, 97.14361572265625, 89.85212707519531, 9.530250549316406, 51.044708251953125, 27.7720947265625, -28.000244140625, -18.511425018310547, 6.987617492675781, 3.0144729614257812, 73.0882568359375, 29.251678466796875, 76.636962890625, -6.035186767578125, 52.34431457519531, 78.69216918945312, -29.55889892578125, 95.73574829101562, 29.170318603515625, 9.200263977050781, -92.80340576171875, 10.645416259765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000221.npy"}
{"epoch": 0.46282722513089003, "step": 222, "batch_size": 128, "mean": 33.421592712402344, "std": 47.97218322753906, "min": -152.6822509765625, "p10": -18.008334350585937, "median": 22.956839561462402, "p90": 100.51138610839843, "max": 127.08523559570312, "pos_frac": 0.7734375, "sample": [61.60565185546875, 14.858306884765625, 74.73097229003906, 72.73529052734375, -24.259063720703125, -26.828903198242188, 40.2933349609375, 49.472129821777344, 17.804534912109375, 8.562850952148438, -30.70367431640625, 20.6419677734375, 75.82998657226562, 41.56842041015625, 22.251953125, 26.72193145751953, 75.39322662353516, 116.22174072265625, 61.56585693359375, 35.13013458251953, 7.1910552978515625, -0.179168701171875, -26.757583618164062, -42.786865234375, 57.847442626953125, 119.78387451171875, -13.85150146484375, 23.353057861328125, -74.88113403320312, 7.894989013671875, 11.101577758789062, 41.492942810058594, 127.08523559570312, 4.7162322998046875, 17.967758178710938, 11.771373748779297, 8.68890380859375, 60.18206787109375, 75.59471130371094, 35.265625, 0.59979248046875, -54.72688293457031, 70.51519775390625, 1.8112564086914062, 77.15150451660156, 8.544788360595703, -56.102012634277344, 56.68310546875, -152.6822509765625, 44.8072509765625, 95.70741271972656, 67.43333435058594, 0.129638671875, 99.80604553222656, -16.34622573852539, 126.259033203125, 24.4493408203125, 98.53106689453125, 1.042083740234375, 4.2899169921875, 33.952484130859375, 22.915809631347656, 15.18353271484375, 104.89996337890625, -6.820587158203125, 16.0374755859375, 82.96640014648438, 124.36898803710938, 40.1988525390625, 63.162261962890625, 94.51559448242188, 82.36358642578125, 2.0766754150390625, 85.60568237304688, -21.754425048828125, 29.56939697265625, 18.0045166015625, 1.9207725524902344, 113.21646118164062, -8.67938232421875, 7.021095275878906, 22.99786949157715, -7.917016983032227, 66.55329895019531, -17.565521240234375, -2.0661773681640625, 102.15718078613281, -20.0286865234375, -7.2200164794921875, 54.10545349121094, 86.0536880493164, 80.80499267578125, 15.035072326660156, 47.89262390136719, 5.927595138549805, -14.968276977539062, 4.502998352050781, 70.72303771972656, 27.27215576171875, 107.145751953125, 114.58489990234375, -9.186172485351562, 5.23065185546875, 39.21479797363281, 106.5137939453125, -4.454063415527344, 89.85617065429688, 36.18017578125, 16.892059326171875, 88.66738891601562, 19.74957275390625, 43.8463134765625, -24.9964599609375, 14.5279541015625, -1.8066253662109375, -9.950637817382812, -13.87188720703125, 99.43661499023438, -19.04156494140625, 76.768310546875, 55.390869140625, 10.231109619140625, 112.15185546875, 81.19216918945312, 20.245529174804688, -0.39862632751464844, 12.876510620117188, 113.00173950195312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000222.npy"}
{"epoch": 0.4649214659685864, "step": 223, "batch_size": 128, "mean": 37.01826477050781, "std": 53.11519241333008, "min": -95.7275390625, "p10": -34.32427520751953, "median": 32.77213668823242, "p90": 103.31368255615233, "max": 146.98272705078125, "pos_frac": 0.765625, "sample": [10.189376831054688, -19.002304077148438, 6.543693542480469, 82.17677307128906, 10.665206909179688, 14.811027526855469, 1.7805328369140625, -11.905982971191406, 31.95697021484375, 97.21234130859375, 29.9876708984375, 99.48225402832031, 32.428924560546875, 7.6361083984375, 57.8367919921875, 136.649658203125, 102.14494323730469, -94.94717407226562, 78.89625549316406, -37.104095458984375, 27.326416015625, 118.12045288085938, -2.7615814208984375, 45.91888427734375, 42.74601745605469, -35.57289123535156, 16.5740966796875, 13.681243896484375, 112.48504638671875, -15.899078369140625, 21.343841552734375, 66.11721801757812, 101.29684448242188, 97.18408203125, 82.05416870117188, 91.83602905273438, 136.93994140625, -95.7275390625, 61.95561218261719, -66.08158874511719, 18.389205932617188, 62.68536376953125, 9.488006591796875, 71.05557250976562, 101.2838134765625, 84.20565795898438, 111.05319213867188, 16.643646240234375, 117.8328857421875, 23.394989013671875, 0.28973388671875, 25.61065673828125, 124.86651611328125, 124.59625244140625, 80.98051452636719, 48.5438232421875, 1.33367919921875, 107.80145263671875, 22.00860595703125, 47.8878173828125, 40.464447021484375, 20.68077850341797, 83.3814697265625, 146.98272705078125, 43.99916076660156, 45.74354553222656, 25.325286865234375, 35.9840087890625, 76.75286102294922, -4.9068756103515625, -13.517486572265625, 64.91940307617188, 7.830718994140625, 94.65882873535156, -36.59955596923828, 29.37438201904297, -36.81397247314453, -29.905197143554688, 92.85971069335938, -13.492790222167969, 34.84748840332031, -8.094696044921875, 0.502349853515625, 26.61443328857422, 87.54608154296875, 58.9537353515625, -65.39913940429688, -13.686729431152344, 69.55960083007812, -12.700881958007812, -79.7538070678711, 27.760894775390625, 67.19979858398438, 63.3162841796875, 84.32373046875, 120.26962280273438, 76.56939697265625, 75.34050750732422, 65.27241516113281, 11.629669189453125, -1.7805347442626953, 61.427001953125, 21.90679931640625, 69.83724975585938, 55.95263671875, -32.817138671875, 82.87557983398438, 131.96078491210938, -0.9358901977539062, 17.7735595703125, 106.04074096679688, -59.87456512451172, -2.1251220703125, 7.715217590332031, 97.28369140625, 57.85638427734375, -61.86572265625, -2.62237548828125, 66.0567626953125, 5.679168701171875, -59.39103698730469, 87.92007446289062, 53.0579833984375, 43.855194091796875, 33.11534881591797, 17.34764862060547, -33.789154052734375, -44.842132568359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000223.npy"}
{"epoch": 0.46701570680628274, "step": 224, "batch_size": 128, "mean": 37.87016296386719, "std": 48.424198150634766, "min": -98.43426513671875, "p10": -19.06900939941406, "median": 33.66119384765625, "p90": 102.36678771972655, "max": 143.06201171875, "pos_frac": 0.7890625, "sample": [120.52133178710938, 143.06201171875, 76.67239379882812, 54.88800811767578, 3.983684539794922, 2.8188858032226562, -21.92877197265625, 48.4810791015625, 80.91323852539062, 16.730857849121094, 5.57220458984375, 72.90198516845703, -4.2310028076171875, 71.44537353515625, 27.335220336914062, 32.686767578125, 10.35894775390625, 75.86724853515625, 23.94818878173828, -54.853485107421875, 72.67413330078125, -0.064117431640625, 112.4678955078125, 113.98980712890625, 20.148727416992188, 68.46381378173828, 71.44828796386719, 34.6356201171875, 18.079803466796875, 45.40838623046875, 113.24554443359375, 84.73419189453125, -98.43426513671875, 51.08995056152344, 35.45793151855469, 20.53887939453125, 35.64070129394531, 72.95278930664062, 44.595306396484375, -6.429351806640625, 39.1661376953125, -17.53564453125, 81.42634582519531, 66.4671630859375, 59.43133544921875, 31.363243103027344, 60.40283203125, -41.40545654296875, 99.31195068359375, -1.0589046478271484, 69.1549072265625, 47.769195556640625, 0.0, 105.16082763671875, 83.935791015625, 13.45547866821289, 8.55194091796875, -34.68699645996094, 4.88580322265625, 29.7713623046875, 61.63203430175781, 15.2099609375, 1.5281009674072266, 86.69192504882812, -2.19561767578125, 108.16033935546875, -20.990570068359375, -31.551605224609375, 49.29705810546875, -32.58198547363281, 10.2010498046875, 123.61956787109375, 31.32867431640625, 105.2537841796875, 17.8468017578125, -10.214111328125, -15.09466552734375, 67.15362548828125, 13.787094116210938, 108.7513427734375, -10.050582885742188, -14.006248474121094, -86.4039306640625, 10.471302032470703, 12.793502807617188, 92.72433471679688, 50.61419677734375, -18.2454833984375, 101.16934204101562, -79.5522232055664, 89.27052307128906, 85.33636474609375, 37.595062255859375, 96.70135498046875, 29.46661376953125, 62.140838623046875, 98.25015258789062, 52.547088623046875, 36.7503662109375, 6.018318176269531, -31.60131072998047, 24.012054443359375, 6.881103515625, 16.34320068359375, 24.325546264648438, 56.12346649169922, 141.87200927734375, -22.889083862304688, 107.66708374023438, 72.94717407226562, 19.315200805664062, -43.22257995605469, 10.36703872680664, 2.6689453125, 4.6392822265625, 30.291183471679688, 90.386474609375, 8.497734069824219, 69.879638671875, -17.2177734375, 7.106861114501953, 94.00161743164062, 72.4608154296875, 73.94766235351562, 95.04539489746094, 73.58069610595703, -1.1785774230957031, 116.321044921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000224.npy"}
{"epoch": 0.46910994764397906, "step": 225, "batch_size": 128, "mean": 26.509456634521484, "std": 58.29530715942383, "min": -119.32571411132812, "p10": -47.369441986083984, "median": 22.978431701660156, "p90": 107.83284454345701, "max": 141.45901489257812, "pos_frac": 0.625, "sample": [110.55633544921875, -1.8519287109375, 33.310302734375, 104.88180541992188, 98.16871643066406, 6.0822296142578125, -72.03227233886719, -67.935546875, 71.87966918945312, 106.66563415527344, 27.292022705078125, 66.64691162109375, 6.90203857421875, -8.218505859375, 114.187744140625, 35.76739501953125, 126.24688720703125, 74.07608032226562, 3.75115966796875, 23.657333374023438, -12.766265869140625, -13.87548828125, 66.91648864746094, -34.5526123046875, 98.83779907226562, 45.373809814453125, 103.93991088867188, -9.37255859375, 1.1383056640625, -31.62139892578125, -15.175704956054688, -119.32571411132812, 81.29454803466797, 100.864990234375, 14.531875610351562, 9.421440124511719, 74.0982666015625, -1.9891471862792969, -1.1312255859375, -49.37864685058594, 67.16119384765625, 118.12893676757812, -15.90545654296875, -65.27389526367188, -90.27914428710938, -65.80609130859375, 44.079345703125, 21.60540771484375, -7.505466461181641, 47.7833251953125, -76.29095458984375, 23.81463623046875, 71.49234008789062, 87.87933349609375, -11.530364990234375, 15.48858642578125, -32.624664306640625, 34.695556640625, 100.37649536132812, 76.07884216308594, 83.90896606445312, 115.95927429199219, 124.9683837890625, -20.178436279296875, 29.34893798828125, -11.187149047851562, -34.884613037109375, 83.86801147460938, 48.4769287109375, 131.7197265625, 2.291748046875, 111.11553955078125, 28.90447998046875, 61.85685729980469, -3.281524658203125, 43.64344787597656, 95.8666763305664, 122.8956298828125, -24.9149169921875, 83.471923828125, -36.564910888671875, -56.714019775390625, -46.50835418701172, 42.556121826171875, -78.30810546875, -9.650146484375, 8.50312614440918, -0.6886100769042969, 42.42723083496094, -100.95611572265625, -5.2469482421875, -30.344451904296875, -12.6226806640625, -13.5426025390625, 26.02191162109375, -29.53668212890625, -1.0716552734375, 47.55944061279297, 141.45901489257812, 62.33367919921875, 5.0705718994140625, 55.774864196777344, 115.84432983398438, 104.97125244140625, -45.531646728515625, 11.448471069335938, -65.32596588134766, 58.248870849609375, 11.39971923828125, 129.17974853515625, -17.322463989257812, 45.91676330566406, 68.14614868164062, 129.26771545410156, 14.46441650390625, 67.05889892578125, 53.870792388916016, -30.360382080078125, 59.322208404541016, -53.35820007324219, 22.299530029296875, 1.9492301940917969, -8.386802673339844, -5.55853271484375, -20.9898681640625, 29.58905029296875, 35.210601806640625, 87.425048828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000225.npy"}
{"epoch": 0.4712041884816754, "step": 226, "batch_size": 128, "mean": 25.17685890197754, "std": 49.94892883300781, "min": -90.75164794921875, "p10": -37.733924102783206, "median": 15.515853881835938, "p90": 95.39923248291015, "max": 127.171630859375, "pos_frac": 0.734375, "sample": [19.59765625, 127.171630859375, -20.3848876953125, 63.02845764160156, 88.04519653320312, 107.86611938476562, 7.158233642578125, 0.6035499572753906, 67.75811767578125, 83.71490478515625, 108.39277648925781, 35.69619369506836, 28.22210693359375, 108.40188598632812, -4.98638916015625, 12.790630340576172, -0.03564453125, -65.2742919921875, 89.54176330566406, 100.100830078125, 3.78692626953125, 1.5337944030761719, 82.09274291992188, 64.59416961669922, 5.824493408203125, 22.330215454101562, -8.854219436645508, -44.279327392578125, 79.03577423095703, 59.55584716796875, 1.177389144897461, 42.19383239746094, 58.93218994140625, 109.68975830078125, 4.231201171875, 14.03167724609375, 90.07302856445312, -40.994964599609375, -8.409759521484375, 36.919898986816406, -64.32887268066406, 58.09309387207031, -27.508163452148438, 0.7319202423095703, -14.03076171875, 55.249542236328125, 14.409744262695312, -10.358673095703125, -88.92282104492188, 32.31829833984375, 30.836700439453125, -49.279754638671875, 89.91851806640625, 1.7140960693359375, 9.363754272460938, -68.57577514648438, 3.8887405395507812, 23.91644287109375, 1.755208969116211, 0.087371826171875, -7.097320556640625, 16.621963500976562, 2.338714599609375, -90.75164794921875, 22.35125732421875, -9.512580871582031, 13.786956787109375, 94.21968078613281, 7.548585891723633, 8.63824462890625, -22.726638793945312, 6.809370040893555, 81.50082397460938, -22.534530639648438, 62.658363342285156, 1.83013916015625, 26.102447509765625, 98.15151977539062, 5.708290100097656, 108.6552734375, -37.7894287109375, 92.12808990478516, 117.79241943359375, -64.60999298095703, 107.42208862304688, 11.182769775390625, 18.11138916015625, 2.8979721069335938, 36.833984375, 74.09979248046875, 66.14678955078125, 31.83319091796875, 86.66236114501953, 49.464508056640625, 88.50192260742188, 13.995513916015625, -18.179718017578125, 0.437744140625, 70.01476287841797, 35.59588623046875, 112.49801635742188, 18.62603759765625, 105.11776733398438, 2.6549072265625, 9.79986572265625, 21.461517333984375, 93.37355041503906, 34.140289306640625, 64.86991882324219, -24.139450073242188, 40.74674987792969, -22.6737060546875, -1.610809326171875, 69.31733703613281, 33.34845733642578, -21.562599182128906, 41.30455017089844, -30.5622615814209, -24.62493896484375, 118.8997802734375, 61.99359130859375, -37.71013641357422, -71.34815979003906, 17.153186798095703, -64.04258728027344, -41.65596008300781, 10.707977294921875, -10.438026428222656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000226.npy"}
{"epoch": 0.4732984293193717, "step": 227, "batch_size": 128, "mean": 37.84011459350586, "std": 52.9171257019043, "min": -119.8310546875, "p10": -27.832194519042964, "median": 33.77899169921875, "p90": 106.24830780029296, "max": 147.21417236328125, "pos_frac": 0.765625, "sample": [-58.33335876464844, 35.97694396972656, 19.75726318359375, 27.69183349609375, 66.38554382324219, 40.69892120361328, 114.90380859375, 18.08582305908203, 7.751556396484375, 34.55244445800781, 102.307861328125, 86.42112731933594, 12.427108764648438, -1.325897216796875, 28.51513671875, 52.268402099609375, 0.0, 114.676025390625, -97.24456787109375, -41.981998443603516, 67.44772338867188, -59.95269775390625, 84.228515625, 101.69674682617188, 137.43206787109375, 147.21417236328125, 105.26797485351562, -88.45220947265625, 92.82913970947266, -18.8192138671875, -6.89923095703125, 91.7196044921875, 12.624153137207031, 13.88995361328125, 44.083404541015625, -13.96337890625, 40.129737854003906, 70.66219329833984, 30.440032958984375, 114.2000732421875, 117.82330322265625, 68.56375122070312, 13.25537109375, 67.75714111328125, 34.01324462890625, -31.044754028320312, 17.2186279296875, 49.713043212890625, -18.090057373046875, 33.54473876953125, 105.03805541992188, 60.81891632080078, -9.708251953125, -26.45538330078125, 65.08663940429688, 132.64761352539062, 5.32794189453125, 72.93385314941406, 108.70466613769531, 28.207775115966797, 108.86408996582031, -42.45460510253906, 14.39630126953125, 81.3756103515625, -119.8310546875, 24.34393310546875, 13.453987121582031, 124.65289306640625, -45.010833740234375, 38.85760498046875, 73.43267822265625, 26.424591064453125, 3.1983280181884766, 80.56613159179688, 104.56254577636719, 55.582977294921875, 88.9527587890625, 75.20974731445312, -37.28375244140625, 73.89993286132812, 115.22323608398438, 7.9358062744140625, 19.986251831054688, 104.9019775390625, 41.0699462890625, 28.608261108398438, 54.272064208984375, 31.19647216796875, 29.814239501953125, 75.71109008789062, -32.435089111328125, 69.81483459472656, 9.679229736328125, -59.679893493652344, 89.63595581054688, 42.16351318359375, 84.54322814941406, 14.771141052246094, -16.519744873046875, 96.9067153930664, -0.7554931640625, 119.65582275390625, -2.155710220336914, 108.53575134277344, -50.44641876220703, 94.4127197265625, -3.5787887573242188, -1.1761474609375, 3.0346755981445312, 0.5100555419921875, 92.16107177734375, 26.430801391601562, 91.22238159179688, 92.64045715332031, 14.57427978515625, 30.19390869140625, 34.8348388671875, 55.12384033203125, 84.66229248046875, -3.88916015625, -15.9395751953125, -23.016998291015625, 26.62750244140625, 11.880638122558594, -15.713516235351562, 5.448888778686523, 40.94402313232422, 51.850555419921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000227.npy"}
{"epoch": 0.47539267015706804, "step": 228, "batch_size": 128, "mean": 37.237396240234375, "std": 51.37420654296875, "min": -72.06715393066406, "p10": -22.586080932617186, "median": 29.3785343170166, "p90": 107.73797607421875, "max": 133.65411376953125, "pos_frac": 0.7265625, "sample": [-63.190887451171875, -11.080940246582031, 61.81011962890625, 92.515869140625, 42.837127685546875, -5.682952880859375, 122.7027587890625, 29.60906982421875, 60.8775634765625, 90.87858581542969, 4.8190460205078125, -10.948158264160156, -10.21832275390625, 109.20364379882812, 46.9676513671875, 66.70640563964844, -5.550933837890625, 82.91139221191406, -11.80072021484375, 8.820236206054688, 83.92515563964844, 23.293994903564453, -72.06715393066406, -15.1251220703125, 16.196258544921875, 81.40225219726562, 25.358551025390625, 2.915679931640625, -21.692657470703125, 61.00434875488281, 12.757598876953125, 25.603164672851562, -33.33509826660156, 104.5555419921875, 31.934539794921875, 95.33943176269531, 63.109405517578125, 71.77984619140625, 13.94287109375, -4.070648193359375, 63.32281494140625, -9.038330078125, -18.831756591796875, -24.659912109375, 78.35498046875, 7.807342529296875, -5.62109375, 15.949333190917969, 47.536590576171875, 71.10214233398438, 39.68035888671875, 120.15060424804688, 60.06669616699219, 20.078563690185547, 87.41390991210938, 35.50923156738281, 79.267333984375, 76.69561767578125, 110.50421142578125, 115.9927978515625, -1.9039459228515625, -61.36918640136719, -31.68964385986328, 0.47290802001953125, 69.72703552246094, 20.940948486328125, 0.0, 23.914886474609375, 57.5255126953125, 17.439979553222656, 12.567710876464844, 133.51657104492188, 15.318992614746094, -8.152587890625, -35.111328125, 90.87828063964844, -19.76123046875, 100.01885986328125, -13.282211303710938, 107.10983276367188, -20.01751708984375, -58.96131896972656, 19.01812744140625, 103.31634521484375, 97.67338562011719, 112.99557495117188, -1.6309814453125, 3.66973876953125, 114.38948059082031, 76.4156494140625, -21.697296142578125, 55.75738525390625, 15.480712890625, 87.65606689453125, 40.12371063232422, 70.61869812011719, -67.51876831054688, 4.5498199462890625, 19.035663604736328, 102.72779846191406, 102.93222045898438, -25.128433227539062, -43.54368591308594, 27.554075241088867, 78.76792907714844, 58.713623046875, 94.49369812011719, -52.41668701171875, -17.0322265625, 3.6094970703125, 16.686676025390625, 114.80783081054688, -3.7889251708984375, 80.93559265136719, 91.83865356445312, 116.55508422851562, 126.28472900390625, 66.97335815429688, 32.1356086730957, -50.578857421875, 76.13970947265625, 77.0687255859375, 116.59237670898438, 30.139633178710938, 133.65411376953125, 6.257232666015625, 5.52752685546875, 29.147998809814453], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000228.npy"}
{"epoch": 0.4774869109947644, "step": 229, "batch_size": 128, "mean": 46.503013610839844, "std": 54.431583404541016, "min": -89.43460083007812, "p10": -19.060986328124997, "median": 47.55708694458008, "p90": 113.6120132446289, "max": 165.80657958984375, "pos_frac": 0.8125, "sample": [76.17498779296875, -5.204345703125, 91.38201904296875, 71.69342041015625, -2.74859619140625, -87.77885437011719, 59.11250305175781, 86.98606872558594, 40.06353759765625, -59.22721862792969, 148.17572021484375, 114.25868225097656, 83.13372802734375, 0.5393409729003906, 128.53955078125, 96.33645629882812, 15.598175048828125, 10.64813232421875, 22.366592407226562, 44.01922607421875, 37.93098449707031, 22.683013916015625, 97.50137329101562, -10.16412353515625, 51.55841064453125, -65.71051025390625, 10.532035827636719, 17.10417938232422, 108.86386108398438, 76.62393188476562, 5.4321136474609375, 45.20556640625, 116.63494873046875, 62.066650390625, 20.333267211914062, 119.2764892578125, -2.504150390625, 17.60333251953125, 137.56851196289062, -89.43460083007812, 8.36322021484375, -56.580291748046875, 52.47889709472656, 92.97341918945312, 64.45315551757812, 133.32052612304688, 165.80657958984375, 105.67984008789062, 55.67231750488281, 91.19009399414062, 39.33366394042969, -1.5204811096191406, 41.39385986328125, 3.242523193359375, 83.32697296142578, 105.50143432617188, 90.8760986328125, 17.325225830078125, 11.1641845703125, -20.731689453125, 62.233612060546875, -68.71897888183594, 99.08554077148438, 58.5367431640625, -9.084518432617188, -7.2114105224609375, 98.367431640625, 14.994544982910156, 74.4195785522461, -1.0968017578125, 2.65130615234375, 52.46974182128906, 1.0569019317626953, -18.344970703125, 96.150634765625, 47.29157257080078, 39.284767150878906, 91.08258819580078, 108.63916015625, 45.25506591796875, 105.61341857910156, 21.177017211914062, 37.880706787109375, 96.28245544433594, 51.58639144897461, 12.91693115234375, -32.47724914550781, 84.66407775878906, 122.784912109375, 64.40606689453125, 133.62216186523438, 97.45018005371094, 12.618610382080078, 42.62608337402344, 12.78619384765625, 102.89863586425781, 41.20845031738281, 112.7977294921875, 48.99755859375, 21.176162719726562, 83.72340393066406, 17.25688934326172, 42.101531982421875, 118.25961303710938, -5.391571044921875, 14.216827392578125, 55.52919006347656, -17.408538818359375, 9.902957916259766, 113.33486938476562, 61.69611358642578, -23.285114288330078, 24.8060302734375, 47.822601318359375, 77.95693969726562, 86.68014526367188, 87.12374877929688, -52.867340087890625, 87.32646179199219, -72.8956298828125, 126.4951171875, 60.6185302734375, 74.141845703125, 24.467880249023438, 91.3978271484375, -56.28036880493164, 138.00521850585938, -36.8724365234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000229.npy"}
{"epoch": 0.47958115183246075, "step": 230, "batch_size": 128, "mean": 34.64807891845703, "std": 54.92072677612305, "min": -164.90545654296875, "p10": -20.173059082031248, "median": 26.98760223388672, "p90": 109.29421997070312, "max": 149.6514892578125, "pos_frac": 0.7421875, "sample": [109.3551025390625, 28.459442138671875, 30.39727783203125, 89.7425308227539, 148.08535766601562, 68.37188720703125, 114.90745544433594, -6.966705322265625, 109.26812744140625, 27.315872192382812, 14.659378051757812, 27.46734619140625, 18.630531311035156, -70.4693374633789, -164.90545654296875, 90.70455932617188, 95.18046569824219, -2.182981491088867, -5.136749267578125, -27.50537109375, 8.27386474609375, 11.892192840576172, 115.47952270507812, -36.38958740234375, 0.0, 98.90682983398438, -3.713531494140625, 16.232852935791016, 63.51019287109375, 94.73933410644531, 44.445526123046875, 93.66453552246094, 5.523468017578125, 39.037689208984375, -0.7695465087890625, -76.67605590820312, 58.04071044921875, 87.65754699707031, 64.4805908203125, -21.241836547851562, 0.04778480529785156, -77.44091796875, 5.1346588134765625, 27.5537109375, 34.64494323730469, 101.33544921875, 67.0922622680664, -8.445709228515625, 89.71759033203125, -19.715011596679688, -6.102643966674805, 59.58412170410156, -34.71820068359375, 54.9027099609375, 98.191162109375, -11.407958984375, 38.10272216796875, 17.038543701171875, 97.31683349609375, 26.08465576171875, -69.62076568603516, 68.49969482421875, 123.486328125, 78.77600860595703, 14.703208923339844, 5.00335693359375, -17.010528564453125, 90.36444091796875, -32.7996826171875, 65.50984191894531, 44.53033447265625, 6.361541748046875, -12.4427490234375, -13.034126281738281, 58.0582275390625, 17.48248291015625, 9.60394287109375, 53.9019889831543, 124.766845703125, -7.0571441650390625, 3.635141372680664, 62.211578369140625, 26.659332275390625, -25.108861923217773, 16.245193481445312, -120.04673767089844, 1.9058380126953125, -21.72966766357422, 127.11041259765625, 56.42230224609375, 137.90139770507812, 80.3245849609375, 12.36151123046875, -0.775360107421875, 109.93359375, 107.01312255859375, 1.9494667053222656, 28.966819763183594, 90.80938720703125, 65.03532409667969, 105.52156066894531, 5.777219772338867, 76.5069580078125, 16.182519912719727, -12.909568786621094, 80.655517578125, -0.5467529296875, 118.530029296875, 74.05203247070312, 6.5798797607421875, 76.69859313964844, 113.58110046386719, 66.15898132324219, 2.5496463775634766, 36.93310546875, 33.205101013183594, 5.214235305786133, 124.47068786621094, 149.6514892578125, -1.3960723876953125, 0.29294586181640625, 10.13818359375, 2.7356719970703125, 10.525482177734375, 25.69097900390625, 36.57322692871094, -1.7146759033203125, -14.001373291015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000230.npy"}
{"epoch": 0.4816753926701571, "step": 231, "batch_size": 128, "mean": 29.11216926574707, "std": 49.40181350708008, "min": -101.89533996582031, "p10": -34.4888671875, "median": 26.16863250732422, "p90": 92.87080078125, "max": 139.14224243164062, "pos_frac": 0.7421875, "sample": [78.23416137695312, 16.02581787109375, -37.18463134765625, -61.23712921142578, 78.54485321044922, 72.10981750488281, 54.39447021484375, -1.2175445556640625, 57.96435546875, -101.89533996582031, 97.33116149902344, -33.630615234375, 68.11306762695312, 57.48138427734375, -3.3101959228515625, 92.75601196289062, 8.843338012695312, 9.537334442138672, 69.294677734375, 0.409515380859375, -23.932037353515625, -64.98104858398438, -13.585601806640625, 51.448455810546875, 9.324981689453125, 35.69556427001953, 9.850173950195312, 66.2773666381836, -6.878692626953125, -56.921844482421875, 28.8807373046875, 83.71293640136719, 93.13864135742188, 3.2242679595947266, 43.26495361328125, -11.246566772460938, 117.09085083007812, 43.80450439453125, 15.855545043945312, 42.19036865234375, -17.589447021484375, 37.57757568359375, -2.05816650390625, 1.0584869384765625, 103.55184936523438, 84.84521484375, -6.8638916015625, 49.49864196777344, 46.69488525390625, 31.431045532226562, 86.23187255859375, -52.23956298828125, 19.564720153808594, 16.7080078125, 6.0472412109375, 16.24224853515625, 84.22955322265625, -20.781463623046875, 31.4681396484375, 139.14224243164062, -36.491455078125, 4.176609039306641, 11.029800415039062, 40.291839599609375, 2.476154327392578, 76.69241333007812, 75.49630737304688, 24.40411376953125, 19.95233154296875, 47.53022766113281, 26.717819213867188, 41.954742431640625, 130.1578369140625, -26.61895751953125, -81.63218688964844, -1.860748291015625, 56.329505920410156, -8.821441650390625, 82.28167724609375, -4.402191162109375, 19.0460205078125, 21.403472900390625, 33.682098388671875, 62.48712158203125, 10.317245483398438, 29.466766357421875, 8.72662353515625, 102.059326171875, 20.23028564453125, 9.595260620117188, 45.953155517578125, 66.22494506835938, 63.71583557128906, 21.857986450195312, -28.407211303710938, 60.3084716796875, 8.77835464477539, 64.16348266601562, 114.6807861328125, 0.6897468566894531, -56.308685302734375, 25.61944580078125, 56.74525451660156, 132.23768615722656, 80.24351501464844, -52.153778076171875, -1.8436965942382812, 33.64399719238281, 27.040557861328125, 68.7676010131836, -4.992584228515625, 53.03314971923828, 31.539520263671875, -79.62722778320312, 104.93710327148438, 129.46835327148438, 11.093597412109375, 105.07730102539062, 12.040817260742188, 52.39337158203125, -16.683998107910156, 5.8983154296875, 92.33848571777344, -38.6898193359375, 70.44110107421875, 0.0, -83.33079528808594, 109.24758911132812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000231.npy"}
{"epoch": 0.4837696335078534, "step": 232, "batch_size": 128, "mean": 34.02851867675781, "std": 53.6618537902832, "min": -86.33306884765625, "p10": -26.115703964233397, "median": 21.004501342773438, "p90": 103.11344757080077, "max": 163.06829833984375, "pos_frac": 0.6875, "sample": [-22.02044677734375, 100.02328491210938, -2.0102157592773438, 111.08059692382812, 29.875396728515625, 69.1171875, 90.86416625976562, 46.393768310546875, -2.76544189453125, -8.403182983398438, 72.97291564941406, 94.7332763671875, 5.9513092041015625, -25.365463256835938, -39.68421936035156, 31.9029541015625, 7.913806915283203, 3.4205856323242188, 19.92431640625, 19.269699096679688, 25.60418701171875, 10.319099426269531, -86.33306884765625, 67.12271118164062, 95.81756591796875, 100.43804931640625, 22.66222381591797, 66.970703125, -0.9630126953125, 130.50775146484375, 21.24041748046875, -18.850723266601562, -40.087371826171875, 105.62295532226562, 75.61920166015625, 94.27536010742188, -52.613555908203125, 37.1533203125, 18.90045166015625, 99.55917358398438, 63.88511657714844, -30.355979919433594, -0.6138916015625, 23.524574279785156, 11.831718444824219, 1.387399673461914, 74.81100463867188, 14.069580078125, 101.17787170410156, 0.20201873779296875, 29.53521728515625, -25.76226806640625, 89.18000030517578, 26.390243530273438, 4.4899444580078125, 58.925048828125, 19.167083740234375, 92.63695526123047, -2.102630615234375, 81.90350341796875, 20.768585205078125, 130.1307373046875, 102.73106384277344, -25.21502685546875, 86.00071716308594, 91.05680847167969, -5.46282958984375, 4.883056640625, -6.5902099609375, -12.709724426269531, 12.804985046386719, 106.73971557617188, 10.349090576171875, -36.68035888671875, 122.64892578125, 120.65876770019531, -2.2903900146484375, -15.495086669921875, 51.794586181640625, 17.2767333984375, 0.6019039154052734, 10.122390747070312, -19.58349609375, 88.48033905029297, 104.00567626953125, -18.55636978149414, -6.211830139160156, 72.88874816894531, -6.182952880859375, -37.00897216796875, 86.51935577392578, 65.95693969726562, 0.0, 112.54013061523438, 42.07398986816406, 6.259479522705078, 51.011016845703125, 28.17987823486328, 98.52621459960938, 70.10498046875, 87.62735748291016, 15.674453735351562, 125.90106201171875, -72.90287780761719, 75.98635864257812, 163.06829833984375, 11.707763671875, -20.1514892578125, 92.39054870605469, 65.0584716796875, -5.88775634765625, 114.01708984375, -76.92892456054688, -12.625503540039062, -19.815330505371094, -50.482177734375, 40.723716735839844, -26.940387725830078, 89.12472534179688, -41.47607421875, 96.26362609863281, -4.83746337890625, -82.89491271972656, 124.93478393554688, 61.49430847167969, 74.96725463867188, 18.685760498046875, -20.60198974609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000232.npy"}
{"epoch": 0.48586387434554973, "step": 233, "batch_size": 128, "mean": 35.032310485839844, "std": 53.888980865478516, "min": -107.98272705078125, "p10": -27.705067443847653, "median": 26.52196502685547, "p90": 103.89886627197265, "max": 160.9566650390625, "pos_frac": 0.7734375, "sample": [-43.23158264160156, 93.2020492553711, 0.33740234375, -4.65887451171875, 11.030677795410156, 103.28045654296875, 34.87115478515625, 30.103485107421875, -12.098983764648438, 13.601402282714844, 15.13629150390625, 28.187358856201172, 113.35771179199219, 78.3055419921875, 26.228103637695312, -1.4364013671875, 23.050384521484375, 101.8421630859375, 39.71958923339844, 155.08987426757812, 95.09005737304688, 88.58775329589844, 101.0269775390625, 90.55979919433594, 58.70367431640625, 103.86170959472656, 93.61831665039062, 8.546905517578125, 89.35380554199219, -25.93206024169922, 26.781768798828125, -18.38349151611328, 83.37608337402344, 2.467498779296875, 109.50489807128906, 60.302093505859375, 24.0926513671875, -73.1358871459961, 73.22129821777344, 121.46360778808594, -49.28705596923828, -4.184131622314453, 3.9773025512695312, 23.376708984375, -27.099807739257812, 139.76116943359375, 59.95916748046875, 17.446624755859375, -0.836639404296875, 27.57073974609375, 103.58392333984375, 131.40628051757812, 13.370635986328125, -10.305511474609375, 37.15069580078125, -12.12176513671875, 143.732177734375, 30.1845703125, 96.18037414550781, 9.59332275390625, 68.3187255859375, 90.29571533203125, -45.1612548828125, 17.284343719482422, 8.380569458007812, -53.089691162109375, -29.117340087890625, 56.8421516418457, -107.98272705078125, -74.085693359375, 36.95123291015625, 103.98556518554688, 117.934814453125, 53.038604736328125, 137.913818359375, 51.588897705078125, 18.846939086914062, 11.606216430664062, 14.016817092895508, 3.8460235595703125, 2.2855796813964844, 7.3764190673828125, 17.4163818359375, 39.7052001953125, -46.81427001953125, 5.24412727355957, 160.9566650390625, 53.818939208984375, -8.711029052734375, 39.8751220703125, 33.58987808227539, 4.5191192626953125, -39.5777587890625, 128.8431396484375, 23.110557556152344, -23.2425537109375, 0.9655838012695312, 80.09756469726562, -5.55963134765625, 97.82695007324219, -14.1890869140625, 4.636528015136719, -9.607498168945312, -87.75294494628906, 74.02142333984375, 108.81869506835938, 28.180267333984375, 76.40656280517578, 73.89541625976562, -72.57270812988281, 0.702728271484375, 83.82095336914062, 47.529693603515625, -2.65887451171875, 33.59165954589844, 25.9630126953125, -32.29524230957031, 79.44210815429688, 55.35344696044922, 37.671630859375, 51.458160400390625, 4.347324371337891, 2.6639404296875, 89.8956298828125, 54.6068115234375, 17.67706298828125, 26.262161254882812, 20.64306640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000233.npy"}
{"epoch": 0.48795811518324606, "step": 234, "batch_size": 128, "mean": 37.645164489746094, "std": 62.01789474487305, "min": -140.44058227539062, "p10": -34.1975830078125, "median": 34.34495162963867, "p90": 115.3827423095703, "max": 179.62255859375, "pos_frac": 0.7578125, "sample": [-42.89945983886719, -97.12261962890625, -33.613494873046875, 0.4074382781982422, 48.065185546875, 108.60179138183594, 115.01251220703125, 49.64068603515625, -27.85504150390625, 132.26651000976562, 100.54859924316406, -87.93411254882812, 94.93341064453125, 86.47616577148438, -24.07318115234375, 97.70144653320312, 34.186912536621094, 52.053375244140625, 78.50022888183594, 119.9989013671875, 81.59953308105469, 61.8983154296875, 73.04690551757812, 127.77212524414062, 18.453033447265625, 5.086143493652344, 89.77034759521484, 82.33282470703125, 87.80642700195312, 34.83642578125, 24.241455078125, 58.53369140625, 103.15005493164062, 130.23126220703125, 85.66213989257812, 19.96619415283203, 34.04754638671875, 11.898345947265625, 1.5545806884765625, 48.62738037109375, 59.5291748046875, 21.079376220703125, 5.999267578125, -70.05087280273438, 22.558929443359375, -20.76312255859375, 83.17634582519531, 83.385986328125, -4.705295562744141, 102.13030242919922, 13.961410522460938, 2.34906005859375, 129.01300048828125, 61.09824752807617, -6.0960693359375, 148.8961181640625, 88.3409423828125, 76.9000473022461, 93.22685241699219, 31.163543701171875, 12.613962173461914, 6.337442398071289, 116.24661254882812, 11.36151123046875, 8.148143768310547, -2.74993896484375, -32.52250671386719, 0.0, 24.598876953125, -12.479408264160156, 14.826171875, 36.81791687011719, -90.25282287597656, -6.356269836425781, 95.58685302734375, 137.13504028320312, 99.38980102539062, 27.370498657226562, -80.75180053710938, 53.14958190917969, -0.5260868072509766, 88.75593566894531, 109.77691650390625, -19.63946533203125, 41.27251434326172, 34.50299072265625, 80.4571533203125, 46.40570068359375, -140.44058227539062, 128.8414306640625, -65.47457885742188, 147.84429931640625, 18.233184814453125, 20.844390869140625, 86.49215698242188, 1.36090087890625, 86.62445068359375, 36.8331298828125, 1.5530014038085938, 8.926025390625, 80.91055297851562, 0.0, 20.506515502929688, 51.73454284667969, 179.62255859375, 130.00225830078125, -99.62303161621094, -32.54144287109375, 112.21258544921875, 118.33499145507812, 78.26464080810547, 2.918487548828125, -50.808868408203125, 25.24664306640625, 62.05682373046875, 102.48857116699219, -31.096588134765625, -32.37225341796875, -35.560455322265625, 1.58721923828125, 46.197998046875, 4.7332305908203125, -66.958251953125, 10.020454406738281, -85.71469116210938, 100.31298828125, -20.400619506835938, 108.790283203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000234.npy"}
{"epoch": 0.4900523560209424, "step": 235, "batch_size": 128, "mean": 34.99993133544922, "std": 59.73373794555664, "min": -120.06381225585938, "p10": -32.46454086303711, "median": 32.83515930175781, "p90": 103.06622314453125, "max": 146.61972045898438, "pos_frac": 0.703125, "sample": [23.13006591796875, 7.952247619628906, -7.0924072265625, 1.49786376953125, 8.55887222290039, -120.06381225585938, -61.547874450683594, -6.0933685302734375, 92.14280700683594, 88.78105163574219, 10.977165222167969, 140.175048828125, 82.056396484375, 50.458465576171875, 69.42333984375, 37.28206253051758, 90.03827667236328, 94.43560791015625, -2.316150665283203, -5.439056396484375, 73.9476318359375, 23.0462646484375, 32.854095458984375, 104.58761596679688, 129.88531494140625, 132.47158813476562, 80.25787353515625, -19.467971801757812, 87.13162231445312, -6.7482147216796875, 107.816162109375, 2.658905029296875, 77.05520629882812, 96.51598358154297, 66.2796630859375, 91.72769927978516, 120.39315795898438, 98.3978271484375, 7.5027923583984375, 139.30218505859375, -11.391716003417969, 102.73965454101562, -31.946876525878906, -1.527984619140625, 59.237884521484375, 131.2786865234375, -104.0938720703125, -24.664705276489258, -94.38671875, 50.32148742675781, 54.455963134765625, -15.258941650390625, -1.9573440551757812, -19.886749267578125, 32.81622314453125, 89.15574645996094, -25.801620483398438, 66.29280090332031, -9.062652587890625, -61.21733856201172, 9.423988342285156, -5.374725341796875, 102.2969970703125, -34.919158935546875, 138.3680419921875, 69.65670013427734, -9.17596435546875, -89.73458862304688, 8.334747314453125, 38.38189697265625, -27.50786590576172, 48.01129150390625, -62.83869934082031, 87.28562927246094, 8.059074401855469, 11.65277099609375, 4.84783935546875, 110.70518493652344, 7.74853515625, 48.688079833984375, 55.31207275390625, 19.996597290039062, -119.77151489257812, 24.816993713378906, 71.92620849609375, -16.766414642333984, 78.6324234008789, 94.14044189453125, 73.68606567382812, 3.6179962158203125, 140.96560668945312, 100.710693359375, 67.0306396484375, 146.61972045898438, 103.82821655273438, 69.10745239257812, 39.948272705078125, 19.2154541015625, 93.6783218383789, -10.02910041809082, -1.8765411376953125, 101.16567993164062, -67.99835205078125, 59.237091064453125, -2.2962646484375, 83.46236419677734, -4.569049835205078, 54.222991943359375, 82.62692260742188, -7.12744140625, -2.891021728515625, 29.32513427734375, 3.4844818115234375, -33.67242431640625, 42.313514709472656, 14.207759857177734, 97.10333251953125, -118.40911865234375, -41.354209899902344, 5.4401702880859375, 80.23277282714844, 78.49992370605469, 24.115234375, 97.19181823730469, 79.21786499023438, 54.20800018310547, 28.07830810546875, 0.40224266052246094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000235.npy"}
{"epoch": 0.49214659685863876, "step": 236, "batch_size": 128, "mean": 38.42571258544922, "std": 54.428382873535156, "min": -118.1593017578125, "p10": -29.004348754882805, "median": 33.62249755859375, "p90": 106.44448852539061, "max": 149.97564697265625, "pos_frac": 0.7890625, "sample": [55.7982177734375, 33.63397216796875, -70.1602783203125, 60.9925537109375, 63.936126708984375, 83.1566162109375, 9.25225830078125, 98.48294067382812, -12.985382080078125, -49.729156494140625, -79.98687744140625, 30.845367431640625, -35.86286163330078, 29.116378784179688, 76.648193359375, 59.240753173828125, -0.6742057800292969, 88.55191040039062, 27.52362060546875, 43.683319091796875, 25.93560791015625, 83.66576385498047, 40.704315185546875, -50.337257385253906, 50.99755859375, 88.77548217773438, 88.31893920898438, 5.929006576538086, 67.27459716796875, 100.98115539550781, 18.911300659179688, 32.824981689453125, 53.969398498535156, 19.536697387695312, 10.236930847167969, 18.741241455078125, 8.77056884765625, 127.96759033203125, 95.9490966796875, 6.459197998046875, 7.2953948974609375, 96.64433288574219, 102.51165771484375, 37.23231506347656, 122.47518920898438, 1.3214397430419922, -41.77534484863281, 0.877899169921875, 20.836135864257812, 0.0, -4.650543212890625, 20.196563720703125, 91.86996459960938, 93.56450653076172, 9.688730239868164, -26.8551025390625, -97.57479858398438, 98.65036010742188, 10.50286865234375, -39.51043701171875, 74.72433471679688, 110.12200927734375, 50.92138671875, 61.330894470214844, 2.38079833984375, -34.019256591796875, 75.16485595703125, -118.1593017578125, 65.84371948242188, 5.0302276611328125, -16.33013153076172, 149.97564697265625, 56.563438415527344, 135.207763671875, 1.3099746704101562, -5.841583251953125, -49.503509521484375, -5.2425537109375, 104.37010192871094, 68.29956817626953, 37.556396484375, 64.48263549804688, -39.40960693359375, 81.79781341552734, 94.88836669921875, 33.61102294921875, 109.26997375488281, 132.471923828125, -6.1654205322265625, 75.96734619140625, 61.839111328125, 102.91497802734375, 109.77548217773438, 20.367141723632812, 3.872406005859375, 105.23356628417969, 102.36843872070312, 72.68022155761719, 86.36698150634766, 112.45022583007812, 95.61456298828125, 34.1126708984375, -15.913726806640625, 12.360298156738281, 0.0, 41.5146484375, 11.032852172851562, 30.581756591796875, -94.47560119628906, 98.20561218261719, 5.2607574462890625, 0.14560317993164062, -21.601097106933594, -18.280914306640625, 45.43475341796875, 99.2165756225586, 115.40721130371094, 11.235763549804688, 9.009979248046875, -22.2327880859375, 110.08882141113281, 86.6556396484375, 111.08560180664062, 17.07861328125, 2.56622314453125, 110.21969604492188, 7.900392532348633, 31.403030395507812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000236.npy"}
{"epoch": 0.4942408376963351, "step": 237, "batch_size": 128, "mean": 32.5600471496582, "std": 56.536293029785156, "min": -127.25445556640625, "p10": -22.817348480224606, "median": 22.96011734008789, "p90": 108.87336196899413, "max": 189.38870239257812, "pos_frac": 0.734375, "sample": [66.50006103515625, 12.5367431640625, 44.014163970947266, 114.1893310546875, -13.698989868164062, 30.483993530273438, 64.65631866455078, 83.42398834228516, 41.31903076171875, 18.670501708984375, -7.2136077880859375, 42.753173828125, 135.84286499023438, -66.79285430908203, -22.132110595703125, 4.833864212036133, 5.19415283203125, 11.667922973632812, 189.38870239257812, 85.83460235595703, 27.851486206054688, 31.138259887695312, 107.77053833007812, 65.25936126708984, 10.629371643066406, 24.4266357421875, 74.88531494140625, 0.0, 73.29850006103516, 85.64895629882812, -0.35379791259765625, 13.6029052734375, -29.35882568359375, 22.54230499267578, -14.810455322265625, 36.37359619140625, 2.42718505859375, 0.7824859619140625, -127.25445556640625, -24.416236877441406, -3.7361526489257812, -1.30938720703125, 3.2810516357421875, 89.81314086914062, 7.5098876953125, 18.159027099609375, 106.71389770507812, 29.91595458984375, 33.07176208496094, 2.877155303955078, 2.025421142578125, 141.43255615234375, 61.73126220703125, 20.98882293701172, 14.641189575195312, 79.99948120117188, 103.6094970703125, 82.85592651367188, 32.793853759765625, 56.70037841796875, 120.00457763671875, 36.07733917236328, 23.3779296875, 14.02215576171875, 57.12263488769531, -7.5078125, 119.93121337890625, 100.49105834960938, -46.9259033203125, -64.69476318359375, 1.016510009765625, 128.79879760742188, -83.12098693847656, 10.20819091796875, 102.09346008300781, -3.1762542724609375, -58.67645263671875, -15.747268676757812, 139.5541534423828, -20.90625, 49.951751708984375, 25.16217041015625, 113.58238220214844, 42.522857666015625, 112.976318359375, -64.31869506835938, -3.9440536499023438, 4.144744873046875, 111.44661712646484, 24.571319580078125, 8.626884460449219, 136.96600341796875, -7.323635101318359, -1.89862060546875, -38.725341796875, 126.36566162109375, -38.6461181640625, 3.601531982421875, 105.91032409667969, 82.65544128417969, 8.98675537109375, 9.965827941894531, 92.25570678710938, -2.0146713256835938, 5.226646423339844, 5.03679084777832, 89.67730712890625, -82.57624816894531, -14.141143798828125, -12.638824462890625, 0.8355712890625, 107.42449951171875, -19.591888427734375, -2.1863632202148438, 59.36235046386719, 57.089569091796875, -121.14984130859375, 46.143890380859375, 83.53656005859375, 103.12655639648438, 94.52243041992188, -17.362548828125, 25.375701904296875, 81.89744567871094, 5.228797912597656, 28.920623779296875, 19.570068359375, 64.60498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000237.npy"}
{"epoch": 0.4963350785340314, "step": 238, "batch_size": 128, "mean": 38.608009338378906, "std": 58.09569549560547, "min": -85.92000579833984, "p10": -22.348983001708984, "median": 31.601409912109375, "p90": 119.36507568359374, "max": 162.19912719726562, "pos_frac": 0.6953125, "sample": [-31.640716552734375, 1.2880477905273438, 6.255363464355469, 12.904983520507812, -50.773345947265625, 142.51943969726562, 95.58566284179688, 44.238037109375, 94.2003173828125, -48.272857666015625, -3.216796875, 76.68740844726562, -4.3641510009765625, 15.316055297851562, -21.93712615966797, -59.58945083618164, 121.54763793945312, 100.94970703125, -2.613199234008789, 53.67425537109375, 100.67698669433594, 74.7076416015625, 38.89697265625, 108.59554290771484, 143.59396362304688, 34.56597900390625, -12.5386962890625, 2.6788253784179688, 22.0765380859375, -85.16696166992188, -79.3118896484375, -85.92000579833984, 126.5086669921875, -80.1246337890625, -8.581085205078125, 126.13742065429688, 32.78228759765625, 94.58108520507812, 3.4235687255859375, 77.22100830078125, 138.9716796875, -0.84039306640625, 52.230316162109375, 151.654541015625, 86.915771484375, 109.65145874023438, 122.36068725585938, 112.38070678710938, -13.73870849609375, 35.02349853515625, 111.37846374511719, 148.505615234375, 68.47909545898438, -7.016876220703125, 18.564300537109375, 16.845855712890625, -3.546722412109375, 5.371084213256836, -10.002151489257812, 76.71997833251953, -6.816619873046875, 31.232879638671875, -15.7357177734375, -85.817138671875, 93.90328979492188, 52.86741638183594, 79.39950561523438, 103.4839096069336, 80.453857421875, 105.76117706298828, -46.92919921875, -21.18450927734375, 44.76307678222656, -5.071685791015625, -2.1036949157714844, 66.30593872070312, 120.7979736328125, -15.32513427734375, 81.89505004882812, 97.197509765625, 72.54463195800781, 118.7509765625, -35.502262115478516, -9.540264129638672, 5.94671630859375, 26.98614501953125, 89.04173278808594, -3.7717323303222656, 65.3532943725586, 34.47259521484375, 0.5033702850341797, -7.8540802001953125, -23.309982299804688, 53.8946533203125, 162.19912719726562, -9.245361328125, 38.0078125, 67.37442016601562, 16.710304260253906, 4.238441467285156, 24.525985717773438, 122.17181396484375, 114.62999725341797, 32.00811767578125, 37.02217483520508, 5.2064361572265625, -14.373733520507812, 102.41462707519531, 20.623809814453125, 12.49945068359375, 24.759552001953125, 94.28317260742188, 17.748870849609375, 12.131011962890625, 92.67315673828125, -14.085891723632812, 3.7334442138671875, 91.38323974609375, -6.0042266845703125, -9.8895263671875, 105.1192626953125, 126.70419311523438, 31.969940185546875, -27.664459228515625, 33.47599792480469, 8.434234619140625, 79.50408935546875, -2.52886962890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000238.npy"}
{"epoch": 0.49842931937172774, "step": 239, "batch_size": 128, "mean": 31.900774002075195, "std": 59.874149322509766, "min": -141.46026611328125, "p10": -38.18903198242187, "median": 18.815338134765625, "p90": 119.91144180297852, "max": 159.38641357421875, "pos_frac": 0.703125, "sample": [-61.887062072753906, -49.4825439453125, 131.6126708984375, 33.299285888671875, -12.37310791015625, 0.05426025390625, -141.46026611328125, 75.14422607421875, 127.36465454101562, 1.6485824584960938, 64.33816528320312, -130.32354736328125, -95.28543090820312, -5.33380126953125, 86.32254791259766, 76.203857421875, 96.6455078125, 119.9100570678711, 55.18359375, -11.559234619140625, -0.458648681640625, 103.4005126953125, 4.979572296142578, 24.332870483398438, -8.194190979003906, 77.6702880859375, 0.028665542602539062, 88.9413070678711, 19.352081298828125, -13.686065673828125, -42.75189208984375, 92.3153076171875, 70.00481414794922, -4.365257263183594, 139.327880859375, -15.049835205078125, 25.916351318359375, 93.22210693359375, 159.38641357421875, 109.608154296875, 82.7698745727539, -6.3359375, 25.080612182617188, -0.641754150390625, 145.0924072265625, -11.647018432617188, -14.057373046875, 71.38522338867188, -73.24870300292969, 91.8770751953125, -69.85348510742188, 69.38044738769531, 17.288795471191406, 58.9495849609375, 53.11384582519531, -19.12384033203125, 47.1790771484375, 8.325759887695312, 15.210205078125, -36.2335205078125, 18.562454223632812, 114.20960998535156, 7.401386260986328, 104.91110229492188, 29.35529327392578, 33.15435791015625, 95.12076568603516, 0.9749374389648438, 7.4132232666015625, 131.340576171875, 0.7006473541259766, 14.921232223510742, 12.0172119140625, 34.754669189453125, 95.9222412109375, -6.2066650390625, 84.64297485351562, 17.48627471923828, -64.45939636230469, 11.256240844726562, -1.5211181640625, 28.585891723632812, 136.64385986328125, -19.12493896484375, -3.8095703125, 125.587158203125, -56.384857177734375, 29.232948303222656, 10.195083618164062, -46.6585693359375, 107.84039306640625, -3.309040069580078, -10.706748962402344, 102.2605972290039, 3.5223464965820312, 37.593990325927734, 5.484870910644531, 1.9822998046875, 135.6729736328125, 48.0384521484375, 0.20079803466796875, 133.81393432617188, 15.722747802734375, -46.41326904296875, 49.31146240234375, -2.0471649169921875, -28.66748046875, 29.830230712890625, 39.700958251953125, 124.03982543945312, 6.8528900146484375, 93.990966796875, 50.92915344238281, 138.9453125, 119.9146728515625, 0.0, 18.632949829101562, 0.9586877822875977, 104.74787902832031, 49.331878662109375, -18.178176879882812, 75.83069610595703, 17.147857666015625, 20.813323974609375, 33.947998046875, -76.11033630371094, -2.0655136108398438, 18.997726440429688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000239.npy"}
{"epoch": 0.5005235602094241, "step": 240, "batch_size": 128, "mean": 41.457183837890625, "std": 54.206111907958984, "min": -106.01121520996094, "p10": -19.03155651092529, "median": 37.731781005859375, "p90": 109.83014068603515, "max": 157.22286987304688, "pos_frac": 0.75, "sample": [8.155698776245117, 108.98773193359375, 64.07638549804688, 103.53887939453125, -25.236473083496094, 1.2806396484375, 65.37974548339844, 42.69929504394531, 0.0, 129.8018035888672, 33.303466796875, -22.03790283203125, -82.33064270019531, 39.128448486328125, -19.890586853027344, 96.1952896118164, 50.10993194580078, 71.97381591796875, 87.81533813476562, 57.83860778808594, 20.0028076171875, -0.936981201171875, 76.57089233398438, 101.29666137695312, 17.871826171875, 99.44224548339844, 78.03349304199219, -22.7047119140625, 113.64572143554688, 109.44279479980469, 38.8082275390625, 35.27764892578125, 123.4007568359375, 90.23817443847656, 2.49755859375, -13.59185791015625, 119.8690185546875, 77.7017593383789, 91.49422454833984, 13.429542541503906, 24.772979736328125, 50.225494384765625, 115.27235412597656, -2.040874481201172, 77.88800048828125, 49.595611572265625, 88.21426391601562, 76.15036010742188, 79.76124572753906, 43.51838684082031, 102.552001953125, 66.3914566040039, 0.0, 68.13121795654297, -106.01121520996094, 25.447998046875, -12.13623046875, -5.4579925537109375, -47.270660400390625, -2.368927001953125, 81.99180603027344, 94.08091735839844, -7.6853485107421875, -11.7919921875, 110.73394775390625, 101.16267395019531, 45.072509765625, 83.67804718017578, 105.75871276855469, 76.34466552734375, 131.10488891601562, -29.17645263671875, 0.0, 15.28778076171875, -16.989585876464844, -11.53326416015625, 138.7388916015625, -1.6898345947265625, 16.42803192138672, 28.549957275390625, -10.186637878417969, 34.942291259765625, 91.19000244140625, 26.046615600585938, 70.99983215332031, 157.22286987304688, -63.99725341796875, 152.74691772460938, -97.68978881835938, 114.79730224609375, 5.792045593261719, 11.47027587890625, 16.979171752929688, 94.35945129394531, 94.21849060058594, 39.729034423828125, 28.785995483398438, 79.05274963378906, 15.90327262878418, -18.663400650024414, -17.956878662109375, 7.333414077758789, 1.4401588439941406, 18.161636352539062, 97.06752014160156, 40.393402099609375, 78.80410766601562, 109.21920776367188, 36.65533447265625, 9.826812744140625, 112.74855041503906, 42.2855224609375, 28.57373046875, -60.76213836669922, 29.976547241210938, 4.167682647705078, -42.6124267578125, 73.64828491210938, -9.70867919921875, 90.47018432617188, 0.0, 5.157142639160156, 64.09749603271484, 25.8648681640625, 21.71649169921875, -61.09941482543945, 4.0006103515625, 128.07009887695312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000240.npy"}
{"epoch": 0.5026178010471204, "step": 241, "batch_size": 128, "mean": 30.531747817993164, "std": 59.13029479980469, "min": -108.45832824707031, "p10": -46.910552978515625, "median": 17.79827117919922, "p90": 109.66310577392578, "max": 164.68710327148438, "pos_frac": 0.6953125, "sample": [-2.716796875, 110.93707275390625, 78.41728210449219, 63.0643310546875, 21.829269409179688, -30.490570068359375, 22.386871337890625, 0.29769325256347656, 65.18798828125, 89.93524169921875, 98.0494384765625, 0.0, -10.543701171875, 110.6947021484375, 87.77339172363281, 141.79736328125, 7.88214111328125, 15.45855712890625, 13.6595458984375, 115.88201904296875, 61.62063980102539, 98.51090240478516, -13.84820556640625, -15.013763427734375, 91.43815612792969, 18.368331909179688, -39.44775390625, 32.13201904296875, 95.90095520019531, -9.600265502929688, 116.45552825927734, -36.08221435546875, 99.33999633789062, 48.20549011230469, 10.287155151367188, 6.395050048828125, -8.34716796875, -9.563552856445312, 5.078361511230469, 101.17050170898438, 10.137825012207031, 18.3446044921875, 90.97600555419922, 137.8404541015625, 11.112823486328125, 85.625732421875, 59.647216796875, -106.97201538085938, 12.782958984375, 25.205642700195312, 21.284912109375, 1.9920368194580078, 2.4667816162109375, 85.87005615234375, -23.03564453125, -7.6352996826171875, 106.54173278808594, 14.64019775390625, 18.617584228515625, -51.368804931640625, 57.783477783203125, 79.45899963378906, 45.2611083984375, 99.55941772460938, -6.397308349609375, 18.356658935546875, -13.598236083984375, 33.88902282714844, 2.113006591796875, 63.33416748046875, -55.83715057373047, 8.218414306640625, 122.28831481933594, 32.75872802734375, -58.994873046875, -48.6805419921875, 118.67611694335938, 76.37353515625, 3.835235595703125, -0.82464599609375, 3.79522705078125, 1.378173828125, 137.7249755859375, 27.485992431640625, 32.21199035644531, 55.896026611328125, -46.43731689453125, 98.754638671875, -8.796707153320312, 109.22099304199219, 7.3613739013671875, -70.20469665527344, -2.3376541137695312, -0.1071624755859375, 100.55010986328125, 80.12057495117188, 3.5223941802978516, 17.251937866210938, 120.40777587890625, -76.18063354492188, 80.77720642089844, -66.4012451171875, 20.49700927734375, -18.955780029296875, 105.66314697265625, 107.42288208007812, -76.13165283203125, 6.4906158447265625, 128.3720703125, 53.210540771484375, -25.29693603515625, -108.45832824707031, -29.924530029296875, -7.002532958984375, 85.30377197265625, -0.38814353942871094, 136.97860717773438, 51.23097229003906, -64.81427001953125, 68.73489379882812, 164.68710327148438, 3.9972686767578125, 15.714202880859375, -92.79638671875, 10.55877685546875, -22.417678833007812, 93.2586669921875, -48.0147705078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000241.npy"}
{"epoch": 0.5047120418848168, "step": 242, "batch_size": 128, "mean": 38.03529357910156, "std": 53.62942123413086, "min": -115.60841369628906, "p10": -19.95923919677734, "median": 39.95212173461914, "p90": 107.9588409423828, "max": 140.57916259765625, "pos_frac": 0.7578125, "sample": [93.25863647460938, 11.703926086425781, 101.76214599609375, 10.76297378540039, 14.893280029296875, 49.905792236328125, 75.52542114257812, 73.19912719726562, -25.447509765625, 11.370697021484375, 132.94967651367188, 88.5352783203125, 43.655372619628906, -13.659942626953125, -91.6048355102539, 33.68059539794922, 57.09761047363281, 97.00749969482422, -115.60841369628906, 39.823856353759766, 12.044731140136719, 126.99085998535156, 120.61883544921875, 44.16179656982422, 52.2978515625, 132.3564453125, -1.235565185546875, 125.97430419921875, -1.822845458984375, 2.1952972412109375, 69.61825561523438, 94.29461669921875, -0.8682861328125, 121.2774658203125, 38.85089111328125, 136.47622680664062, 117.82728576660156, 2.4282455444335938, -77.14977264404297, 2.1590347290039062, 89.2220458984375, -19.11663818359375, -1.38323974609375, 58.41192626953125, -21.925308227539062, -4.610721588134766, 30.810791015625, 68.26654052734375, 66.59086608886719, 3.1204986572265625, -8.600921630859375, -13.863574981689453, -4.679256439208984, 26.8369140625, -6.704795837402344, -3.42620849609375, 97.97184753417969, 41.50581359863281, -10.857177734375, 90.70269775390625, 46.2249755859375, 0.09438705444335938, -28.54803466796875, 115.4124755859375, 8.075775146484375, -3.020477294921875, 139.32177734375, 60.548553466796875, 56.64250183105469, 20.043106079101562, 43.68206787109375, 64.4056396484375, -23.259170532226562, 13.241493225097656, 83.0902099609375, 12.030967712402344, 80.99945831298828, 17.141056060791016, 21.27752685546875, -61.35877990722656, 54.51091003417969, 79.06201171875, 23.8472900390625, 40.080387115478516, -81.49185180664062, 13.412887573242188, 6.27216911315918, 42.55377960205078, 39.182403564453125, 84.77420043945312, 140.57916259765625, 5.75323486328125, 48.66571044921875, 28.862804412841797, -2.6415328979492188, 18.68365478515625, 67.17813110351562, 95.70050048828125, 70.0637435913086, 20.142425537109375, 1.751983642578125, 92.7921142578125, -28.656131744384766, -16.173782348632812, 87.32980346679688, 40.78523254394531, 55.650108337402344, -103.94268798828125, 62.917236328125, 67.91900634765625, 36.285614013671875, 7.338600158691406, 107.056396484375, 124.17665100097656, 83.7332763671875, 78.95114135742188, 110.06454467773438, 44.371009826660156, -60.88734436035156, 84.66397094726562, -9.970733642578125, 54.04463195800781, 12.505416870117188, -45.746063232421875, 63.69654846191406, 93.59512329101562, -17.373291015625, 94.82293701171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000242.npy"}
{"epoch": 0.506806282722513, "step": 243, "batch_size": 128, "mean": 43.82208251953125, "std": 54.54512405395508, "min": -108.61051940917969, "p10": -16.799455261230467, "median": 36.540069580078125, "p90": 116.02401428222656, "max": 154.001953125, "pos_frac": 0.7734375, "sample": [-53.989471435546875, -22.655288696289062, 31.396011352539062, 24.44000244140625, 100.11215209960938, 70.672607421875, 10.37957763671875, 122.92724609375, -72.43862915039062, 90.21788024902344, 24.2061767578125, 138.1917724609375, 21.90609359741211, -24.98095703125, 56.3087158203125, -10.584854125976562, -54.820831298828125, 76.80404663085938, 126.53778076171875, -49.389869689941406, 5.6091156005859375, 17.55938720703125, 1.037689208984375, 0.2710723876953125, 61.142730712890625, 109.39788818359375, 154.001953125, 121.43673706054688, -3.3299407958984375, 101.70355224609375, 2.456930160522461, 87.23934936523438, 17.321319580078125, 38.542205810546875, 103.28070068359375, 8.09417724609375, 115.02447509765625, 14.569290161132812, 91.22547912597656, 120.69424438476562, -14.122299194335938, 93.7685546875, 53.9990234375, 69.75729370117188, 67.92376708984375, 91.1727294921875, 118.345947265625, -40.06964111328125, 53.550689697265625, 61.410980224609375, 39.34857177734375, -108.61051940917969, 0.851226806640625, 38.0947265625, 14.646484375, 4.2596435546875, 24.159202575683594, 74.10708618164062, 14.54412841796875, -5.4698486328125, 139.05526733398438, 61.757415771484375, 84.9249267578125, -12.0672607421875, -8.505096435546875, 83.93696594238281, 11.565017700195312, 95.90545654296875, 83.88135528564453, -4.92877197265625, 47.157012939453125, -55.486602783203125, -14.0025634765625, 108.84489440917969, 80.93002319335938, 17.975601196289062, 4.214324951171875, 61.911865234375, -1.2230300903320312, 20.224746704101562, -8.944046020507812, 88.29645538330078, -26.10291290283203, 76.66812133789062, 7.2366943359375, 0.0, 2.86395263671875, 80.13356018066406, 92.29077911376953, 100.77284240722656, 28.186744689941406, 76.38427734375, 76.73321533203125, -5.36297607421875, 112.51705932617188, 115.02890014648438, 118.45901489257812, 11.97418212890625, 4.115837097167969, 34.98541259765625, 23.12628173828125, -1.326171875, -12.811203002929688, 29.268463134765625, -17.14111328125, 113.34661865234375, 128.38214111328125, 88.79789733886719, 4.090812683105469, 31.52862548828125, 11.020389556884766, 9.199394226074219, 103.78764343261719, -16.653030395507812, -35.717491149902344, 147.41848754882812, 18.344036102294922, 108.44151306152344, -7.008331298828125, 45.28401184082031, 47.86891174316406, 101.88287353515625, 142.60366821289062, 81.13976287841797, 131.7982177734375, 85.64471435546875, 51.983489990234375, -21.569122314453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000243.npy"}
{"epoch": 0.5089005235602094, "step": 244, "batch_size": 128, "mean": 44.69786071777344, "std": 58.20124816894531, "min": -126.3485107421875, "p10": -15.341183471679686, "median": 39.23213195800781, "p90": 118.1205352783203, "max": 200.21380615234375, "pos_frac": 0.75, "sample": [-126.3485107421875, 27.86260986328125, 90.30123901367188, -0.42034912109375, 28.468765258789062, 12.25112533569336, 105.0994873046875, -30.951995849609375, 138.08856201171875, 23.361190795898438, 97.26080322265625, 58.112396240234375, 18.637954711914062, -5.9883575439453125, 10.596443176269531, 18.911972045898438, 82.98504638671875, 83.9816665649414, 84.40653991699219, 105.802490234375, 18.476791381835938, 97.08507537841797, -2.2358169555664062, -7.750885009765625, 102.72627258300781, -0.95648193359375, -16.459991455078125, -59.01213073730469, 104.19879150390625, -23.920501708984375, 5.20306396484375, -14.8616943359375, -4.1367340087890625, -5.5159149169921875, -96.42025756835938, 11.926933288574219, 136.77325439453125, 15.067611694335938, 91.263916015625, 101.2005615234375, 7.591400146484375, 51.783203125, 124.48342895507812, 73.51164245605469, 126.7947998046875, 54.750213623046875, -0.333526611328125, 69.20924377441406, 40.32508850097656, 72.90560913085938, 97.17205810546875, 92.9117431640625, 0.0, 89.94827270507812, -14.669219970703125, -44.19134521484375, -50.98876953125, 93.41928100585938, -2.0751953125, -5.921897888183594, 119.0128173828125, -27.383392333984375, 4.314361572265625, 78.76861572265625, 4.775325775146484, 117.3387451171875, 111.45287322998047, 125.00820922851562, 200.21380615234375, 14.616790771484375, 48.831939697265625, 52.69548034667969, -35.23480224609375, 83.9951171875, 41.37284851074219, 3.0363311767578125, 123.9906005859375, -11.0570068359375, 128.702880859375, 82.3935546875, 0.325958251953125, 86.38528442382812, 10.763031005859375, -56.49195861816406, 21.0032958984375, 76.301025390625, -4.445404052734375, 94.94699096679688, 94.246337890625, 9.234466552734375, 4.4020538330078125, -3.83160400390625, 36.916961669921875, 68.7960205078125, 134.29693603515625, 117.73812866210938, 108.21151733398438, 4.261482238769531, 146.80905151367188, 84.26683044433594, 1.3364791870117188, 84.66427612304688, 38.13917541503906, 83.23283386230469, -10.76104736328125, 86.04959106445312, 25.575942993164062, 18.555740356445312, 117.73424530029297, 66.22726440429688, -25.82293701171875, 57.868194580078125, 3.6556472778320312, 123.31939697265625, 63.585296630859375, 140.3814697265625, -1.2596721649169922, 78.57656860351562, 81.475341796875, -10.278945922851562, 11.526054382324219, 15.60601806640625, 96.13355255126953, -100.90447998046875, 29.59423828125, 81.29762268066406, 112.6207275390625, 0.519287109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000244.npy"}
{"epoch": 0.5109947643979058, "step": 245, "batch_size": 128, "mean": 45.11431121826172, "std": 57.4915885925293, "min": -126.00376892089844, "p10": -18.673140335083005, "median": 46.81548309326172, "p90": 110.1755142211914, "max": 158.59686279296875, "pos_frac": 0.8125, "sample": [10.26922607421875, -12.423286437988281, 33.62144470214844, 54.577980041503906, 0.924835205078125, 24.418060302734375, 84.002197265625, 7.98583984375, 4.8580322265625, 81.78341674804688, 40.99298095703125, -18.25982666015625, 18.036880493164062, 131.44827270507812, 79.25186157226562, 16.19610595703125, 145.0479278564453, 137.13720703125, 56.65321350097656, 157.76031494140625, -126.00376892089844, -27.29888916015625, -58.99542236328125, -26.9034423828125, 48.491851806640625, 75.3759765625, 0.0, 37.19525146484375, 74.45671844482422, -7.529083251953125, 87.791748046875, 91.32080078125, 83.25654602050781, 101.11239624023438, 3.463623046875, 137.024169921875, 3.8057823181152344, 37.195831298828125, 87.67770385742188, 49.596343994140625, 127.21939086914062, 11.8294677734375, 58.749298095703125, 97.08377075195312, 11.669342041015625, 88.14117431640625, 30.12432861328125, 45.7486572265625, 44.75862121582031, -1.474334716796875, -25.10772705078125, 36.57664489746094, 6.910247802734375, 107.2730712890625, 23.527191162109375, -7.3136138916015625, 80.45436096191406, 58.806915283203125, 76.65725708007812, 110.66252136230469, 139.49761962890625, 6.886232376098633, 4.046409606933594, 66.93553161621094, 47.88230895996094, 93.69342803955078, 17.720664978027344, -39.02125549316406, 52.163726806640625, -83.83798217773438, 58.18424987792969, 92.0212631225586, 72.22869110107422, 82.90151977539062, 94.7051010131836, 8.5379638671875, -62.8502197265625, 33.13142395019531, 154.79150390625, 109.966796875, 6.142766952514648, 101.01825714111328, -69.88790893554688, 7.293922424316406, 10.8841552734375, -53.31077575683594, 6.755889892578125, 107.34933471679688, 91.08786010742188, 97.52650451660156, 79.12947082519531, 16.80999755859375, 8.742622375488281, 139.94972229003906, 55.478607177734375, 151.38888549804688, 26.950286865234375, 50.45782470703125, 56.10106658935547, 79.68659973144531, -84.60498046875, 106.34217834472656, 74.1360855102539, -16.8865966796875, -4.896844863891602, 95.076171875, 30.57830810546875, 98.07159423828125, 80.99238586425781, 99.51225280761719, 12.342010498046875, -16.788238525390625, 39.748809814453125, 45.747772216796875, 93.052490234375, 19.02459716796875, -0.690032958984375, -19.63753890991211, 133.8647003173828, 1.435445785522461, 105.87973022460938, 59.11505126953125, 74.62631225585938, 158.59686279296875, -109.36715698242188, 4.74498176574707, 0.0, 67.86178588867188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000245.npy"}
{"epoch": 0.5130890052356021, "step": 246, "batch_size": 128, "mean": 27.499004364013672, "std": 61.48928451538086, "min": -120.14694213867188, "p10": -49.59604721069336, "median": 14.306793212890625, "p90": 117.67052612304687, "max": 153.89202880859375, "pos_frac": 0.65625, "sample": [70.79886627197266, 88.55593872070312, 80.22676086425781, -2.244598388671875, 71.63735961914062, -7.069007873535156, 153.89202880859375, 11.00665283203125, 19.01821517944336, 52.42597961425781, -25.033580780029297, 74.80471801757812, 14.258758544921875, 18.681137084960938, -62.329063415527344, -56.83746337890625, -13.572555541992188, 131.27891540527344, 130.92935180664062, -48.87812042236328, 110.47761535644531, 139.60330200195312, 18.938812255859375, 93.32788848876953, 39.214141845703125, -18.58624267578125, 13.23358154296875, -52.78202819824219, 25.121994018554688, -5.3666229248046875, 38.8719482421875, -31.642181396484375, 116.1329345703125, 42.44115447998047, 10.56671142578125, -7.7049560546875, -110.31318664550781, 115.72906494140625, 3.95703125, 1.107269287109375, 14.64105224609375, -104.21112060546875, -21.7786865234375, 118.54949951171875, -43.92631530761719, -22.00250244140625, -60.03341293334961, -14.106353759765625, 3.74169921875, 80.42721557617188, 29.96392822265625, 136.72113037109375, -21.684326171875, 24.3331298828125, 122.5888671875, -15.925323486328125, 46.5877685546875, -22.418853759765625, -120.14694213867188, 16.318565368652344, 131.88665771484375, 69.05670166015625, 55.61128234863281, 52.94612121582031, -37.305450439453125, 69.44197845458984, 131.94818115234375, 72.87335205078125, -14.163665771484375, 133.66046142578125, 134.95584106445312, 33.9482421875, -51.271209716796875, 0.981781005859375, 21.26544189453125, 6.2362060546875, 101.41575622558594, 16.326705932617188, -31.01123046875, -58.08319091796875, -62.5052490234375, -0.624237060546875, 93.4583740234375, 72.86865234375, -27.0145263671875, -1.959625244140625, -108.29501342773438, -64.5537109375, 14.354827880859375, 7.8885345458984375, 49.872100830078125, 117.11639404296875, 80.44850158691406, -62.022491455078125, 31.703269958496094, 32.469444274902344, 0.5430145263671875, -10.106414794921875, 0.0, 107.521728515625, 121.60246276855469, 99.9381103515625, -17.264869689941406, -13.01824951171875, 92.41293334960938, 1.2860565185546875, 3.08074951171875, 83.66128540039062, 4.384151458740234, 68.27129364013672, 26.513763427734375, 6.148811340332031, 111.00140380859375, 75.709716796875, 13.165313720703125, 117.2938232421875, -3.9076805114746094, -27.638404846191406, -21.237701416015625, -22.101593017578125, 123.04815673828125, 4.549163818359375, 1.8587112426757812, 46.81132507324219, 5.4582672119140625, 11.151779174804688, -5.7307586669921875, 110.02345275878906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000246.npy"}
{"epoch": 0.5151832460732985, "step": 247, "batch_size": 128, "mean": 37.84230041503906, "std": 56.45536804199219, "min": -124.69412231445312, "p10": -24.41413803100586, "median": 30.1531982421875, "p90": 115.46800231933594, "max": 171.39398193359375, "pos_frac": 0.7265625, "sample": [106.74894714355469, -30.75872802734375, 73.43620300292969, 70.69390869140625, -13.368057250976562, 58.2476806640625, 115.0638427734375, 115.92196655273438, 96.209716796875, 63.276458740234375, 0.0, 89.24576568603516, 115.59457397460938, 16.665496826171875, 7.441474914550781, 37.39306640625, -23.954078674316406, 7.807861328125, -78.25332641601562, 87.40815734863281, -12.582565307617188, 116.66165161132812, -35.1507568359375, -58.63690948486328, 59.7171630859375, 87.62367248535156, 42.737548828125, -3.1826171875, 92.21517944335938, 115.41375732421875, 94.9859619140625, 105.24117279052734, 13.532577514648438, 141.62005615234375, 73.3200454711914, -26.489715576171875, -1.726531982421875, 129.00311279296875, -6.544921875, 9.859100341796875, -9.575370788574219, -1.5691909790039062, 32.043304443359375, -44.38627624511719, 8.710296630859375, 71.61810302734375, -1.93743896484375, 37.04229736328125, 3.6825942993164062, -68.33920288085938, 5.786079406738281, 81.74099731445312, 58.624046325683594, 63.976531982421875, -32.052154541015625, 103.48682403564453, 148.06146240234375, 15.881668090820312, 1.525115966796875, -14.56048583984375, -21.669212341308594, -8.69854736328125, -5.873191833496094, 127.21701049804688, 0.501922607421875, -0.42645263671875, 60.39012145996094, -79.94647216796875, 12.544464111328125, -23.79736328125, 29.963623046875, 154.39730834960938, -124.69412231445312, 1.3374290466308594, 105.06549072265625, 12.3243408203125, 19.032943725585938, 5.7024993896484375, 31.543182373046875, 93.49040222167969, 43.4986572265625, 65.95645141601562, 171.39398193359375, 21.024810791015625, -25.48760986328125, -6.372039794921875, 1.7249832153320312, 40.657501220703125, 90.909912109375, 54.713592529296875, 28.45159912109375, 84.722412109375, 71.56291198730469, 25.38433837890625, 15.068695068359375, 93.64889526367188, 122.2476806640625, 109.61906433105469, 134.42373657226562, -15.189910888671875, 30.3427734375, 11.8887939453125, 65.26164245605469, 39.30706787109375, 86.45469665527344, 79.67259216308594, 28.4495849609375, -30.09619140625, 125.63916015625, 13.624176025390625, -93.11868286132812, 91.1735610961914, 73.96720123291016, -23.143768310546875, 24.886751174926758, 35.2177734375, 37.149261474609375, -10.400978088378906, -5.30023193359375, -2.94024658203125, 80.61314392089844, 39.359039306640625, 119.73030090332031, 3.7474899291992188, 97.42457580566406, 8.175247192382812, 22.97430419921875, 65.1927490234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000247.npy"}
{"epoch": 0.5172774869109947, "step": 248, "batch_size": 128, "mean": 28.934532165527344, "std": 57.06718063354492, "min": -128.64285278320312, "p10": -47.80657958984375, "median": 21.871320724487305, "p90": 112.79960632324219, "max": 172.494384765625, "pos_frac": 0.7421875, "sample": [11.957019805908203, 67.99725341796875, -50.90013885498047, 19.29590606689453, 135.41848754882812, -19.156219482421875, 23.089874267578125, 9.060943603515625, 74.628173828125, 11.667961120605469, 26.029258728027344, 38.26483154296875, 71.80029296875, 16.378746032714844, 0.2977294921875, 33.27880859375, 130.13836669921875, 24.079238891601562, 1.018402099609375, 83.62179565429688, -5.5555267333984375, 21.303756713867188, 54.581939697265625, 28.387535095214844, 128.0029296875, 34.028472900390625, 30.694915771484375, 55.99058532714844, 2.544708251953125, 85.119140625, 20.179290771484375, -19.843063354492188, 133.01812744140625, -16.103240966796875, 2.840648651123047, 23.82232666015625, 114.25209045410156, -128.64285278320312, 41.65796661376953, -4.009233474731445, -34.248046875, 3.538299560546875, 22.544700622558594, -26.96014404296875, -6.206058502197266, 36.54498291015625, -0.6691722869873047, -109.53005981445312, 23.37274932861328, 14.125396728515625, 15.124374389648438, 130.5920867919922, 33.09748840332031, 22.02608871459961, 151.4212646484375, -47.88177490234375, 112.53662109375, 0.6672096252441406, -21.39727783203125, 125.06143188476562, 53.20732116699219, 93.91100311279297, -25.6513671875, -73.7225341796875, 16.491348266601562, 8.988616943359375, 172.494384765625, 110.99359130859375, 45.28905487060547, 78.79129028320312, 76.52787780761719, -51.57539367675781, 1.5513267517089844, 54.0186653137207, -0.11956787109375, -92.38188171386719, 0.0, 30.00701904296875, 123.1756591796875, 113.41323852539062, 99.66561889648438, 64.2371826171875, 76.17057800292969, 67.89419555664062, 8.35696792602539, 76.53688049316406, -24.92547607421875, -47.77435302734375, -8.485237121582031, 10.200309753417969, -57.44105529785156, 14.58477783203125, 115.212646484375, 95.09727478027344, 109.33561706542969, -83.31253051757812, 33.261993408203125, 8.669853210449219, 7.564369201660156, 99.42807006835938, 0.916473388671875, -48.427276611328125, 37.87559509277344, -16.5238037109375, 55.08293151855469, 32.253814697265625, 88.04788208007812, 7.5308685302734375, 10.068290710449219, -74.76799011230469, 17.433837890625, 21.716552734375, 15.989692687988281, -2.2171783447265625, 142.69699096679688, 82.35537719726562, 68.33160400390625, 55.47370910644531, 52.176300048828125, -66.41030883789062, -9.733451843261719, -3.8185348510742188, 27.812225341796875, 22.314117431640625, 83.03630065917969, -69.96710205078125, 2.3205337524414062, 18.3681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000248.npy"}
{"epoch": 0.5193717277486911, "step": 249, "batch_size": 128, "mean": 35.7943115234375, "std": 59.43977737426758, "min": -144.23471069335938, "p10": -28.20800018310547, "median": 25.317646026611328, "p90": 113.78224182128906, "max": 160.4222412109375, "pos_frac": 0.71875, "sample": [-10.630523681640625, 6.278415679931641, -40.34715270996094, -74.93719482421875, 21.467788696289062, 4.7980804443359375, -16.474990844726562, 11.869705200195312, -3.27728271484375, 24.365264892578125, -2.5280532836914062, -67.80150604248047, 92.54559326171875, 126.919677734375, -8.419830322265625, 68.98239135742188, 38.23297119140625, 114.22430419921875, 22.200225830078125, 114.07479858398438, 12.476242065429688, -125.988037109375, 105.6527328491211, 18.33489990234375, -27.787612915039062, 140.220947265625, 113.30471801757812, -14.910919189453125, 51.9803466796875, 75.31233215332031, 25.02098846435547, -8.597808837890625, 40.34783935546875, -14.160285949707031, 3.311798095703125, 28.801101684570312, 123.02041625976562, 54.021881103515625, 92.57070922851562, -84.2578125, 111.85671997070312, 101.18572998046875, -10.426651000976562, 106.25041198730469, 144.93521118164062, 30.007186889648438, -90.92683410644531, 92.968994140625, 59.293914794921875, 5.4108428955078125, -12.425537109375, 14.743339538574219, 156.9020233154297, 73.73650360107422, 113.6568603515625, -1.373992919921875, 15.296539306640625, 69.32576751708984, -29.18890380859375, 65.16903686523438, 37.22547912597656, -0.6124267578125, 1.2789306640625, 4.12286376953125, 96.8349609375, 9.54901123046875, 88.92404174804688, 3.005464553833008, 72.99134826660156, 89.79264831542969, -18.085784912109375, -57.52372741699219, 26.4326171875, 90.78106689453125, 26.01720428466797, 81.6954345703125, 160.4222412109375, 131.15338134765625, 105.7918472290039, 96.31217956542969, 141.04541015625, 76.72850799560547, 19.297462463378906, -37.2965087890625, 92.76614379882812, 131.5355224609375, -5.482322692871094, -21.432968139648438, -6.551425933837891, 87.65228271484375, 22.037948608398438, -44.17547607421875, 77.73883056640625, 66.93368530273438, 51.560394287109375, 89.46029663085938, 43.22442626953125, 25.614303588867188, 0.2051544189453125, -41.252410888671875, 114.6373291015625, 2.071929931640625, 9.865348815917969, 76.635986328125, 28.67810821533203, 9.7628173828125, -5.780769348144531, -144.23471069335938, -35.26715087890625, -11.768455505371094, -27.696136474609375, 9.815385818481445, 51.381744384765625, 2.0561676025390625, 17.76312255859375, -3.50921630859375, 77.5302734375, 39.1083984375, 94.00395202636719, 4.107635498046875, 100.97004699707031, 94.36770629882812, 63.14500045776367, -20.598358154296875, -3.212646484375, 32.434356689453125, 14.90434455871582, 128.1672821044922], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000249.npy"}
{"epoch": 0.5214659685863874, "step": 250, "batch_size": 128, "mean": 37.82052230834961, "std": 63.51063537597656, "min": -110.30291748046875, "p10": -30.37384033203125, "median": 34.527008056640625, "p90": 121.75363922119139, "max": 169.38804626464844, "pos_frac": 0.75, "sample": [86.22537231445312, 11.36688232421875, -57.31988525390625, 114.5560302734375, 88.53939819335938, 160.87481689453125, 146.4659423828125, 84.11039733886719, 39.80096435546875, -87.40487670898438, 87.46538543701172, 23.599924087524414, 104.595458984375, 30.866485595703125, 25.510345458984375, 128.4049072265625, 86.85226440429688, 18.702667236328125, 127.37127685546875, 98.2216796875, 119.99333190917969, 49.70867919921875, 22.315292358398438, 93.44467163085938, -30.142120361328125, 3.49261474609375, -110.30291748046875, 21.185028076171875, -6.55279541015625, 0.3077392578125, 128.07073974609375, -101.15692138671875, 48.464569091796875, 154.0721435546875, -10.322608947753906, 8.426513671875, 50.60260009765625, 1.7677021026611328, 126.94561767578125, -104.54312133789062, -2.4606266021728516, 50.148155212402344, 115.29116821289062, 29.62267303466797, 88.03359985351562, 74.84259033203125, -29.82427978515625, -80.06182861328125, -1.5839881896972656, 79.16514587402344, 87.21453857421875, 8.071319580078125, 61.107269287109375, -32.075439453125, 32.5762939453125, 14.56658935546875, -11.832778930664062, -29.72796630859375, 76.57584381103516, 5.2818603515625, 108.92953491210938, 126.20771789550781, 90.48019409179688, -1.0721569061279297, 50.97716522216797, 41.26513671875, 147.95159912109375, 95.34538269042969, -4.8541412353515625, 43.0670166015625, 81.82561492919922, 90.34822082519531, -30.914520263671875, 115.47601318359375, 36.47772216796875, -25.098297119140625, 30.693115234375, 107.0791015625, 125.86102294921875, 26.323671340942383, 20.146963119506836, -22.8233642578125, 38.7203369140625, -26.30157470703125, 108.16958618164062, 47.61459732055664, 1.1522979736328125, -103.80538940429688, 94.42074584960938, 133.39431762695312, -5.4586181640625, 169.38804626464844, -65.02304077148438, -25.10297393798828, -101.22579956054688, 2.049407958984375, 10.083564758300781, -80.12197875976562, -87.91473388671875, 95.49641418457031, 72.59306335449219, 87.5481948852539, 86.95635986328125, 78.12374877929688, 80.1892318725586, 58.86225128173828, -28.64947509765625, 38.6663818359375, 16.370586395263672, -0.9506893157958984, 6.14227294921875, 71.46572875976562, 144.3145294189453, -19.909622192382812, 39.77789306640625, 61.833396911621094, 97.89230346679688, 6.4813079833984375, 2.50421142578125, 3.1968231201171875, -24.650779724121094, 87.68891906738281, 43.904541015625, 3.001953125, 6.745025634765625, 22.713043212890625, 16.057144165039062, 3.4183120727539062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000250.npy"}
{"epoch": 0.5235602094240838, "step": 251, "batch_size": 128, "mean": 36.71636962890625, "std": 59.048851013183594, "min": -106.99430847167969, "p10": -26.937709045410156, "median": 23.369834899902344, "p90": 117.97225036621093, "max": 216.1912841796875, "pos_frac": 0.75, "sample": [80.095458984375, 43.174896240234375, 23.896957397460938, -14.366752624511719, -44.84490966796875, -42.290679931640625, -70.3084716796875, 50.319923400878906, 8.387771606445312, 19.8099365234375, 6.062397003173828, 64.16326904296875, 105.33309936523438, 82.26522827148438, 4.1777191162109375, -4.1151885986328125, 38.7935791015625, -0.80657958984375, 5.444944381713867, 139.08001708984375, -0.0742645263671875, 3.263427734375, -26.3212890625, 109.76788330078125, 87.76197814941406, 122.12548828125, -54.76959228515625, 105.04507446289062, 127.26629638671875, 22.571441650390625, 104.177734375, 5.045337677001953, -16.56671905517578, 40.591064453125, 27.456298828125, -66.73042297363281, 56.75048828125, -1.0935401916503906, 12.93115234375, -20.53186798095703, 82.17527770996094, 49.096099853515625, 102.15371704101562, 123.10577392578125, 121.79574584960938, 4.1219482421875, -2.639007568359375, -50.06968688964844, 3.386892318725586, 5.2228851318359375, -31.947425842285156, 16.23809814453125, 216.1912841796875, 28.30133056640625, 81.69573974609375, 92.36914825439453, 22.84271240234375, 108.07809448242188, 2.290752410888672, 66.71829986572266, 89.05278015136719, -106.99430847167969, 19.186458587646484, 76.31625366210938, 50.06158447265625, -33.65106964111328, 8.878097534179688, -4.713623046875, 25.07904052734375, -1.7475109100341797, 103.40380859375, 70.92587280273438, 4.2558746337890625, 4.097316741943359, 50.88409423828125, 8.965835571289062, 1.154022216796875, 29.8499755859375, -6.8542633056640625, 78.78582763671875, -0.39471435546875, 55.043548583984375, 2.3771772384643555, 3.8544769287109375, 79.67259216308594, 65.39790344238281, 118.41656494140625, -86.96327209472656, -2.45599365234375, 6.5551605224609375, 62.66594696044922, 141.1697998046875, 103.31234741210938, 17.899139404296875, 142.62277221679688, -6.937690734863281, 181.3809814453125, 117.78182983398438, 92.31475067138672, -6.666721343994141, 96.35031127929688, 32.1912841796875, 68.8360595703125, 43.51190185546875, 60.362884521484375, 32.28131103515625, 111.32994079589844, 3.65325927734375, 50.5430908203125, -87.153076171875, 1.496246337890625, 29.72412109375, 12.144699096679688, 3.1889114379882812, 4.968452453613281, -84.57386779785156, -27.7559814453125, -19.26416015625, 3.99151611328125, 142.99957275390625, 113.41470336914062, -26.587020874023438, 123.31349182128906, 129.73654174804688, -13.16583251953125, 4.336448669433594, 75.67101287841797, 86.1005859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000251.npy"}
{"epoch": 0.5256544502617801, "step": 252, "batch_size": 128, "mean": 38.10984802246094, "std": 67.49396514892578, "min": -140.8609619140625, "p10": -41.7647705078125, "median": 27.85845184326172, "p90": 126.28276824951172, "max": 201.428955078125, "pos_frac": 0.71875, "sample": [181.98538208007812, 129.50128173828125, 128.07876586914062, 58.8663330078125, -73.30473327636719, 40.13603973388672, -72.65303039550781, 12.535507202148438, 2.7763290405273438, -15.66070556640625, -4.139190673828125, -43.006134033203125, 14.611297607421875, 28.238616943359375, 201.428955078125, 38.00634765625, 119.44113159179688, 84.75846099853516, 108.93142700195312, 104.441650390625, 4.019830703735352, 80.30380249023438, -22.11419677734375, 126.88438415527344, -140.8609619140625, 5.565681457519531, 5.2884063720703125, 55.150474548339844, -33.25799560546875, 49.28790283203125, 0.9255828857421875, 118.9571533203125, 110.86260986328125, 98.34783935546875, 79.08084869384766, 16.179595947265625, -61.655792236328125, -9.614133834838867, 25.535980224609375, 192.2451171875, 72.70065307617188, 6.499061584472656, 80.01947021484375, 27.478286743164062, 24.8358154296875, 16.13140869140625, 26.075363159179688, -9.241409301757812, -0.348724365234375, 13.201370239257812, 106.06289672851562, 7.81756591796875, 126.02493286132812, 84.6077880859375, -20.9805908203125, -4.7857666015625, 90.21754455566406, -88.751708984375, 35.07666015625, 10.438583374023438, 84.26678466796875, -21.505027770996094, 35.436859130859375, 134.23846435546875, 16.783477783203125, 70.05213928222656, 115.64947509765625, 36.914306640625, 38.92340087890625, 118.9801025390625, -8.043609619140625, 99.3415298461914, 134.02935791015625, -7.302375793457031, -55.493560791015625, 109.4033203125, -82.78451538085938, 7.321678161621094, 73.6175537109375, 87.87956237792969, 81.13665771484375, 3.40313720703125, 37.3380126953125, 95.39739227294922, 2.0543975830078125, -3.0059280395507812, 0.0, 90.44812774658203, -99.0958251953125, 109.1583480834961, -0.031330108642578125, 99.1958236694336, -11.71734619140625, -22.0908203125, 35.0118408203125, 132.40072631835938, -2.1377410888671875, 169.17160034179688, 44.52996826171875, 112.3321533203125, -4.99468994140625, 64.74185180664062, 69.74008178710938, -1.0381927490234375, 127.80923461914062, 0.79168701171875, 95.74891662597656, -115.08633422851562, 19.123046875, 2.6250457763671875, -33.43890380859375, -41.232757568359375, 40.881103515625, -108.24383544921875, -98.57659912109375, 6.24005126953125, 92.95985412597656, 136.415771484375, -50.230133056640625, 102.80087280273438, 20.38037109375, 138.65570068359375, 42.1405029296875, -38.59442138671875, 15.752838134765625, 62.026153564453125, 17.607372283935547, 102.6627197265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000252.npy"}
{"epoch": 0.5277486910994764, "step": 253, "batch_size": 128, "mean": 31.62898063659668, "std": 61.256446838378906, "min": -132.21795654296875, "p10": -37.2059326171875, "median": 15.692764282226562, "p90": 119.05815124511719, "max": 150.48678588867188, "pos_frac": 0.75, "sample": [143.10903930664062, -7.88427734375, 10.720291137695312, 64.18708038330078, 18.56329345703125, 22.100357055664062, 2.2322616577148438, 73.54641723632812, 149.33612060546875, 26.275726318359375, 44.585205078125, 117.65521240234375, 5.302520751953125, -9.6331787109375, 61.096954345703125, -99.64935302734375, 72.27118682861328, -64.5414810180664, 102.3489990234375, -32.488616943359375, 3.22296142578125, -6.9216156005859375, 13.224905014038086, 8.770126342773438, 0.53436279296875, -32.140380859375, 31.679107666015625, 130.60464477539062, 96.63573455810547, 85.9364013671875, -3.2388763427734375, -12.931278228759766, -102.41964721679688, 27.40716552734375, 150.48678588867188, 8.11572265625, -16.20831298828125, 118.91519165039062, -132.21795654296875, -31.93243408203125, 119.3917236328125, 42.30450439453125, 122.94522094726562, 17.48748779296875, 17.245880126953125, -37.006256103515625, 104.02079772949219, 7.390380859375, -9.70556640625, 15.9427490234375, 118.4393310546875, 37.80023193359375, 115.49664306640625, 9.940277099609375, -50.23738098144531, 108.39361572265625, 9.79522705078125, 11.117927551269531, -37.671844482421875, 80.97047424316406, 48.30860900878906, -33.58447265625, 0.8050384521484375, 100.9681167602539, 21.1192626953125, 113.54234313964844, -41.528656005859375, -14.867790222167969, 100.8388671875, -29.2528076171875, 22.465179443359375, -12.259658813476562, -89.74964141845703, 5.92852783203125, -105.42327880859375, 49.297454833984375, -80.72836303710938, 4.109375, 0.1588592529296875, -30.16546630859375, 70.71162414550781, 1.2333297729492188, 52.2955322265625, 91.92630004882812, 35.75071716308594, -2.707733154296875, 27.499412536621094, 10.32867431640625, -46.88262939453125, 135.51919555664062, -50.500633239746094, 8.712150573730469, 76.81399536132812, 0.158111572265625, 6.73651123046875, 105.75645446777344, -42.59211730957031, 143.59127807617188, 8.025047302246094, 80.51664733886719, 88.67851257324219, 12.914604187011719, -35.848388671875, 1.719696044921875, 36.76551818847656, 119.47772216796875, 13.153427124023438, 126.50900268554688, 14.0947265625, -4.710750579833984, 26.61395263671875, 125.3665771484375, 92.56730651855469, 3.0922813415527344, 90.35214233398438, 12.548660278320312, 10.1075439453125, 15.442779541015625, 35.61444091796875, 145.24908447265625, 3.4988479614257812, 41.3299560546875, 131.24005126953125, 93.24014282226562, 10.103240966796875, 89.00785827636719, 65.5083999633789, 101.28121948242188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000253.npy"}
{"epoch": 0.5298429319371728, "step": 254, "batch_size": 128, "mean": 38.868927001953125, "std": 56.581905364990234, "min": -132.609130859375, "p10": -20.33158645629883, "median": 30.3883056640625, "p90": 113.27788696289062, "max": 152.6116180419922, "pos_frac": 0.75, "sample": [140.01821899414062, -0.168182373046875, -79.4365234375, 29.02198028564453, -11.554435729980469, -3.4868526458740234, 17.8846435546875, 18.0552978515625, -0.5022182464599609, 19.789932250976562, -3.2414627075195312, -13.163604736328125, -43.778594970703125, -4.188873291015625, 7.951698303222656, 123.527587890625, 3.84149169921875, 98.23898315429688, 49.41912841796875, -30.897308349609375, -37.638763427734375, 120.69281005859375, 21.7537841796875, 13.45037841796875, -3.757049560546875, 0.2315826416015625, 86.4005126953125, -10.469970703125, 94.53125, 67.5526123046875, 98.3692626953125, 131.59732055664062, -73.84246826171875, 77.11328125, 1.1319599151611328, 17.64642333984375, -20.538475036621094, 30.556476593017578, 5.740993499755859, 10.40008544921875, 66.13900756835938, 54.57221984863281, 83.25213623046875, 39.19050598144531, -48.60353088378906, 19.8193359375, 54.4285888671875, 47.901580810546875, 66.44677734375, 96.33697509765625, 53.90611267089844, 4.880496978759766, 35.39513397216797, 111.94903564453125, 79.8597412109375, 30.28436279296875, 92.4202880859375, -67.93649291992188, 31.333740234375, 78.22247314453125, 4.314567565917969, 16.896682739257812, 132.640869140625, 6.589271545410156, 41.8531494140625, 108.0191879272461, 40.41680908203125, 111.57183837890625, 114.87399291992188, 30.594436645507812, 9.571434020996094, 40.945709228515625, 127.0150146484375, 5.993125915527344, 45.015533447265625, -20.242919921875, -2.769775390625, 30.49224853515625, 1.7104759216308594, 2.3059158325195312, -1.991607666015625, -10.007530212402344, 128.0787353515625, 110.7159423828125, 70.5102767944336, 137.76568603515625, 74.16973876953125, 101.01649475097656, 17.208099365234375, 109.24981689453125, 21.75328826904297, 110.31852722167969, 111.87908935546875, 90.14990234375, 103.01067352294922, 24.5511474609375, 0.41497802734375, 109.04816436767578, -5.9652862548828125, -6.479644775390625, -15.21087646484375, 50.934417724609375, 50.323638916015625, -2.1961288452148438, 0.0, -28.678665161132812, 121.88717651367188, 85.83775329589844, 34.3590087890625, 22.247955322265625, 6.9649505615234375, 23.564048767089844, 34.284339904785156, 112.59384155273438, 79.34927368164062, -0.3567657470703125, -132.609130859375, -60.753936767578125, 138.28045654296875, 21.865859985351562, -65.223876953125, -68.430908203125, 4.605995178222656, 152.6116180419922, 143.4877166748047, 96.13995361328125, 107.4334716796875, 40.65611267089844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000254.npy"}
{"epoch": 0.5319371727748691, "step": 255, "batch_size": 128, "mean": 43.95281219482422, "std": 66.35061645507812, "min": -114.13975524902344, "p10": -22.387243270874023, "median": 32.69671630859375, "p90": 127.96644744873046, "max": 177.07553100585938, "pos_frac": 0.7734375, "sample": [17.008037567138672, -102.298828125, 136.88519287109375, 45.83555603027344, 43.92229461669922, -102.42889404296875, 2.249969482421875, 121.04794311523438, -11.7469482421875, 98.69087982177734, -8.04388427734375, 8.778915405273438, 9.681705474853516, 10.19061279296875, 146.10348510742188, 123.4420166015625, 6.449989318847656, 15.02020263671875, 139.26150512695312, 10.21405029296875, 109.93701171875, -22.367279052734375, -0.10601806640625, 99.29368591308594, 40.0323486328125, 125.67695617675781, 70.93297576904297, 104.33261108398438, -20.358673095703125, 27.507293701171875, 76.55929565429688, 10.711090087890625, 23.049835205078125, 15.36865234375, 51.95420837402344, 33.41493225097656, 118.97137451171875, 96.802734375, -75.71102905273438, -71.12889099121094, 177.07553100585938, 149.58111572265625, 128.0725860595703, 27.784988403320312, -81.5474624633789, -22.433826446533203, -112.25064086914062, 125.4393310546875, 90.5561294555664, 85.97294616699219, 116.53262329101562, 124.7022705078125, 88.94381713867188, 89.30523681640625, 144.21212768554688, 2.8337783813476562, 61.6763916015625, -85.6011962890625, -61.581298828125, 108.1695556640625, 22.843353271484375, 116.6337890625, 105.50997924804688, 98.94879150390625, 37.77020263671875, 65.70880126953125, 6.404449462890625, -96.43404388427734, 123.20539855957031, 17.615219116210938, 140.01708984375, 39.04638671875, 20.8558349609375, -3.2621917724609375, 2.850616455078125, 12.586003303527832, 3.516002655029297, 49.433380126953125, 10.792205810546875, 93.48694610595703, -5.82464599609375, 80.31967163085938, -114.13975524902344, 140.939208984375, 119.41995239257812, 92.88046264648438, -8.21051025390625, 45.98737335205078, 16.648239135742188, 16.20020294189453, 127.92095947265625, 55.6861572265625, 65.1114730834961, 123.15682220458984, -18.06640625, 146.97463989257812, 111.730712890625, 4.483734130859375, 3.8060302734375, -10.730064392089844, 1.4252080917358398, -37.514404296875, 71.69340515136719, 5.437255859375, 133.29623413085938, 77.41868591308594, 8.302871704101562, 102.44940185546875, -20.973838806152344, 60.4090576171875, 6.60479736328125, 10.171768188476562, -12.316665649414062, 115.56484985351562, 20.255218505859375, 31.978500366210938, -6.57110595703125, -12.16912841796875, 100.87077331542969, 63.5894775390625, 17.189102172851562, 106.89342498779297, -6.3367919921875, 136.56549072265625, -53.827880859375, 153.12173461914062, 25.605499267578125, -17.57616424560547], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000255.npy"}
{"epoch": 0.5340314136125655, "step": 256, "batch_size": 128, "mean": 43.77632141113281, "std": 69.45587158203125, "min": -129.81398010253906, "p10": -52.119688415527335, "median": 45.036787033081055, "p90": 126.90476074218749, "max": 161.60537719726562, "pos_frac": 0.7265625, "sample": [3.4335861206054688, 116.02536010742188, 123.99185180664062, -38.35791015625, 37.46949768066406, 82.31338500976562, 116.209716796875, 92.72296142578125, 113.14004516601562, -0.54730224609375, 98.44715881347656, -29.148269653320312, -31.1400146484375, -100.92333221435547, 42.42412185668945, 109.8121337890625, 135.51370239257812, -20.52685546875, 89.37956237792969, 104.22158813476562, 0.6589584350585938, 109.34217834472656, -20.87683868408203, 161.60537719726562, 17.837890625, 120.64198303222656, 106.49069213867188, -60.51971435546875, 137.68887329101562, 18.3896484375, 135.43170166015625, 42.59786605834961, 54.0582275390625, 41.044097900390625, 47.4757080078125, 12.921966552734375, -47.66461181640625, -120.03399658203125, 107.060302734375, 51.09593200683594, 138.40426635742188, 21.90863037109375, 65.45578002929688, 114.04605102539062, 54.64830017089844, 52.336334228515625, -67.99996185302734, -11.508308410644531, -42.69343566894531, 154.068603515625, 3.5117263793945312, -64.61691284179688, 133.41336059570312, 134.1309814453125, 4.7384185791015625, 2.692392349243164, -129.81398010253906, -35.20062255859375, 109.43811798095703, 72.53802490234375, 97.00933837890625, 17.82171630859375, 101.95761108398438, 137.03732299804688, -1.6760740280151367, 84.50450134277344, 58.22587585449219, -99.19103240966797, 97.66122436523438, -5.37420654296875, -33.540008544921875, -93.30938720703125, 119.44461059570312, 41.350067138671875, 41.81915283203125, 107.813720703125, -17.158477783203125, 116.74615478515625, 25.48797607421875, 135.31387329101562, 145.73828125, 86.45777130126953, 88.67961883544922, -2.769012451171875, 32.652008056640625, 61.29266357421875, -78.39756774902344, 8.892623901367188, 39.048858642578125, 125.0064697265625, -59.670654296875, 112.87156677246094, 99.24104309082031, 129.25616455078125, 15.697540283203125, 102.99109649658203, 49.935272216796875, -48.88356018066406, 18.433677673339844, 28.16741943359375, 38.759063720703125, 38.60920715332031, 126.25030517578125, 79.88067626953125, 77.96578979492188, 55.55755615234375, 14.530487060546875, 119.82614135742188, -62.039363861083984, -129.4972381591797, 5.715576171875, 124.54884338378906, 128.43182373046875, 14.973426818847656, -10.983787536621094, -0.709716796875, 125.34091186523438, -35.662353515625, -11.491252899169922, -62.932220458984375, 95.37423706054688, -5.53082275390625, -0.09534454345703125, 65.42731475830078, 20.15118408203125, 56.87031936645508, 69.45571899414062, 108.85276794433594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000256.npy"}
{"epoch": 0.5361256544502618, "step": 257, "batch_size": 128, "mean": 40.78329849243164, "std": 63.87449645996094, "min": -113.49085998535156, "p10": -25.95985641479492, "median": 27.6380615234375, "p90": 126.63329467773437, "max": 180.20147705078125, "pos_frac": 0.671875, "sample": [56.062835693359375, 27.704254150390625, 174.2938690185547, -19.23278045654297, -1.6803970336914062, 180.20147705078125, 21.969757080078125, 124.61447143554688, 100.63153076171875, 106.75072479248047, 65.95523071289062, 142.08868408203125, 106.26611328125, 70.64581298828125, -0.420013427734375, 28.46514129638672, 35.0196533203125, -27.341842651367188, -25.367576599121094, 126.82376098632812, 121.391845703125, -35.95214080810547, 77.75040435791016, 156.26431274414062, -18.1163330078125, -49.54351806640625, 25.04638671875, 23.535964965820312, -4.56298828125, -47.411476135253906, -22.426673889160156, -32.22662353515625, 112.6695556640625, 21.27606201171875, 84.96951293945312, 121.14486694335938, 176.93560791015625, 1.0965576171875, 132.3485107421875, 91.39839935302734, -21.56097412109375, 6.7362518310546875, -18.719932556152344, 105.82058715820312, 126.55166625976562, -33.024139404296875, -3.3498458862304688, 27.571868896484375, 14.132858276367188, -108.53089904785156, 86.40791320800781, 153.43649291992188, 63.04273223876953, 44.5518798828125, 108.26980590820312, 18.701690673828125, 78.9005126953125, -22.559402465820312, 69.32124328613281, 28.823211669921875, 28.725128173828125, 126.92041015625, 104.74362182617188, 14.885326385498047, -59.007598876953125, -19.833770751953125, 99.40286254882812, 3.2428455352783203, 47.47870635986328, 56.840728759765625, 0.21588134765625, 99.84857177734375, 73.8588638305664, 36.80511474609375, 21.089813232421875, 10.440841674804688, -113.49085998535156, -1.610595703125, 90.65171813964844, 121.974853515625, 1.0523834228515625, 2.725374221801758, 81.07345581054688, 30.13037109375, 18.615245819091797, -18.071380615234375, -1.69830322265625, 120.34201049804688, 58.31805419921875, -10.168609619140625, -17.58392333984375, 144.45806884765625, -30.661773681640625, 121.94354248046875, 62.39128494262695, 129.426025390625, 141.56320190429688, 66.86981201171875, -21.569000244140625, 8.317939758300781, 12.48895263671875, 13.676864624023438, 108.7685546875, 32.297454833984375, -5.8097076416015625, -12.3330078125, -5.99627685546875, -96.08810424804688, -15.1661376953125, 124.21871948242188, 25.575653076171875, 110.70703125, -18.81640625, 164.0715789794922, -36.82965087890625, -6.0897064208984375, 68.69989013671875, -22.339752197265625, 90.372314453125, 33.94720458984375, 42.45526123046875, -1.7126045227050781, -9.949302673339844, 5.76568603515625, -9.79836654663086, -12.655303955078125, -50.94471740722656, 107.52719116210938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000257.npy"}
{"epoch": 0.5382198952879581, "step": 258, "batch_size": 128, "mean": 45.445823669433594, "std": 58.25391387939453, "min": -138.42376708984375, "p10": -22.21260833740234, "median": 41.001747131347656, "p90": 116.82088317871093, "max": 152.26324462890625, "pos_frac": 0.7578125, "sample": [103.12110900878906, 21.77421760559082, 16.661148071289062, 4.708213806152344, 74.30805969238281, -25.859878540039062, 116.13665771484375, -25.96826171875, 47.42689514160156, 100.9178466796875, -44.345123291015625, 37.793792724609375, 59.1925048828125, 76.62686157226562, 112.14633178710938, 75.07414245605469, -23.480300903320312, 28.51739501953125, 113.878662109375, 90.10142517089844, 125.59536743164062, 71.149658203125, -37.512664794921875, 9.99188232421875, 148.81234741210938, -32.615203857421875, -138.42376708984375, 111.62744140625, 126.33392333984375, 39.64704895019531, 104.8192138671875, 91.06802368164062, -31.199462890625, 71.27227783203125, 78.36135864257812, 34.37188720703125, 3.598217010498047, -26.14215087890625, 59.18556213378906, 101.93048095703125, 115.27719116210938, 87.40756225585938, 6.4012451171875, 36.8861083984375, 12.350387573242188, 25.25775146484375, 23.400115966796875, 18.277587890625, 18.427886962890625, 15.33856201171875, 71.6954345703125, 6.328962326049805, 104.11083984375, 21.950775146484375, -122.22113037109375, 80.6808090209961, 46.61665344238281, -6.803619384765625, -104.07122802734375, 6.405120849609375, -16.866256713867188, 17.395614624023438, 33.07666015625, 35.860137939453125, 117.4168701171875, 3.324920654296875, 148.19229125976562, 115.8062744140625, -5.392669677734375, 98.65638732910156, 84.11552429199219, 64.16253662109375, -3.241546630859375, -6.0677337646484375, -0.9691400527954102, 45.32269287109375, 127.35675048828125, -57.33268737792969, 84.46016693115234, 29.64837646484375, -8.450458526611328, -21.6693115234375, 93.84185028076172, 102.01484680175781, -10.25567626953125, -0.9220504760742188, 95.23905944824219, 49.84039306640625, 116.56546020507812, -47.762115478515625, 46.81210708618164, 12.194892883300781, 36.81695556640625, 17.855426788330078, -5.650054931640625, 60.26043701171875, 96.8052978515625, 107.16664123535156, 10.96578598022461, -14.608245849609375, 4.2482452392578125, -3.4504547119140625, 56.83488464355469, -11.44482421875, 152.26324462890625, -2.2096023559570312, 107.117431640625, 130.0953826904297, 123.1865234375, 95.71507263183594, 42.3564453125, -16.27490234375, 147.9881591796875, 81.94064331054688, 137.75494384765625, 127.21954345703125, 76.37371063232422, 13.1390380859375, 72.47146606445312, 89.15101623535156, -12.9224853515625, 98.72834777832031, 82.40736389160156, 141.14620971679688, 48.74517822265625, 26.812408447265625, 24.181365966796875, -2.815216064453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000258.npy"}
{"epoch": 0.5403141361256545, "step": 259, "batch_size": 128, "mean": 41.11183166503906, "std": 62.3960075378418, "min": -116.46405029296875, "p10": -27.510217285156248, "median": 45.60906219482422, "p90": 120.6577392578125, "max": 159.2459716796875, "pos_frac": 0.734375, "sample": [95.82635498046875, -21.393539428710938, 123.01861572265625, 9.263837814331055, -28.769012451171875, 13.143203735351562, 15.7154541015625, 138.7379150390625, 112.4129638671875, -11.71832275390625, 58.61114501953125, 15.30859375, 45.13203430175781, 100.356201171875, 8.374755859375, 56.97505187988281, 99.92453002929688, -109.94206237792969, 79.70697021484375, 110.70944213867188, 63.16455078125, 146.6309814453125, -7.304656982421875, 30.995052337646484, 130.817626953125, 25.7296142578125, 65.3393325805664, 77.6753158569336, -87.33837890625, -9.10015869140625, 8.64788818359375, 9.50909423828125, -20.762420654296875, 90.76898193359375, 91.29043579101562, 42.882476806640625, 65.43505859375, 13.0712890625, 17.821044921875, 93.33749389648438, -37.360748291015625, 19.960479736328125, 128.57138061523438, 22.8044490814209, 1.9825439453125, -113.412109375, 83.3578109741211, 132.46792602539062, -2.1109619140625, 93.67193603515625, 88.73408508300781, 83.97947692871094, 118.59832763671875, 3.8299560546875, 15.221183776855469, 21.90874481201172, -14.297271728515625, 134.637451171875, 48.630584716796875, 13.123695373535156, 159.2459716796875, 70.73237609863281, 46.5528564453125, 2.4089393615722656, -7.1324005126953125, -90.02871704101562, 48.736419677734375, -36.0589599609375, -81.11531066894531, 93.33309936523438, 63.47703552246094, -10.276611328125, -5.0182037353515625, 107.67984008789062, 126.51640319824219, -18.306060791015625, 59.12969970703125, 0.0, -12.30767822265625, 113.60662841796875, 111.06352233886719, 110.19699096679688, -26.970733642578125, 17.28057861328125, 49.7440185546875, 62.65434265136719, 128.065185546875, -83.3178939819336, -0.5682449340820312, 50.524383544921875, -116.46405029296875, -19.1549072265625, 126.45817565917969, 133.406494140625, 25.6673583984375, 38.0736083984375, 79.8404769897461, -20.497772216796875, 103.8204116821289, -16.1319580078125, 72.94078063964844, 69.15071105957031, -24.571273803710938, 115.47576904296875, 16.04266357421875, 32.416259765625, 107.52352905273438, -94.77337646484375, 59.453643798828125, 119.64593505859375, 63.65576171875, 112.76057434082031, 11.100555419921875, -33.674072265625, 19.36622428894043, 46.086090087890625, 130.08847045898438, 16.522491455078125, 54.58145523071289, -22.24151611328125, 93.49838256835938, -47.99200439453125, -24.090866088867188, 112.28985595703125, 5.2054595947265625, 70.68673706054688, 99.43994140625, 92.55699157714844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000259.npy"}
{"epoch": 0.5424083769633508, "step": 260, "batch_size": 128, "mean": 16.67919158935547, "std": 62.652557373046875, "min": -145.8927001953125, "p10": -51.48412170410156, "median": 10.872360229492188, "p90": 105.11325531005858, "max": 148.81494140625, "pos_frac": 0.640625, "sample": [121.46487426757812, -15.724178314208984, 95.5520248413086, 13.3011474609375, -118.393798828125, 101.30461120605469, 39.523834228515625, 125.26953125, 20.258209228515625, 84.2230224609375, -18.665084838867188, 17.4559326171875, 0.0, -20.1187744140625, 21.55438232421875, 5.876262664794922, 3.36199951171875, 148.81494140625, 28.096405029296875, 39.98733139038086, 101.32670593261719, -145.8927001953125, -0.5028839111328125, 7.848876953125, 146.30807495117188, -50.587646484375, 17.80410385131836, 91.17098999023438, 29.44921875, 14.598030090332031, 30.863861083984375, -49.17950439453125, 16.125701904296875, 0.40224266052246094, 4.564298629760742, 142.9266357421875, -47.34600830078125, -112.42373657226562, -10.654800415039062, 13.947807312011719, 0.899688720703125, 17.3489990234375, 35.93682861328125, -69.419189453125, -29.749679565429688, 94.86961364746094, -81.2939224243164, -40.14358901977539, -50.39581298828125, 13.639568328857422, -1.10760498046875, -35.48992919921875, 116.84002685546875, 51.0205078125, 7.040874481201172, 12.139495849609375, 112.75897216796875, -119.90988159179688, 103.92169189453125, 86.42970275878906, -31.957366943359375, -3.210521697998047, 108.61610412597656, -10.260772705078125, 6.359767913818359, -90.4475326538086, 62.37873458862305, 107.89356994628906, -4.741455078125, 123.42239379882812, 9.841384887695312, 97.27413940429688, -43.483245849609375, 10.543670654296875, 14.513008117675781, -21.462051391601562, -17.94145965576172, 36.36863708496094, 13.56640625, 11.2010498046875, -99.35479736328125, -45.38954162597656, 139.61599731445312, 0.1418914794921875, 0.060394287109375, 30.9796142578125, 24.53363037109375, -18.814056396484375, -4.119010925292969, -101.76861572265625, -30.224212646484375, 2.981597900390625, -106.26771545410156, 130.70242309570312, 20.177093505859375, 8.581390380859375, -7.931186676025391, 57.14604187011719, -28.8568115234375, 112.53439331054688, 75.88873291015625, 67.22726440429688, 19.83197021484375, 53.90846252441406, 2.39398193359375, 6.34674072265625, -4.5948638916015625, 55.467002868652344, -56.679664611816406, 89.3165283203125, -5.880706787109375, 98.0323486328125, 97.64141845703125, 12.799896240234375, -50.761688232421875, 4.4512481689453125, -21.385528564453125, -112.51834106445312, -11.858596801757812, 34.82951354980469, 84.21710205078125, -53.1697998046875, 21.2403564453125, 11.243160247802734, 90.75440216064453, 0.5611495971679688, 62.16564178466797, -18.96258544921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000260.npy"}
{"epoch": 0.5445026178010471, "step": 261, "batch_size": 128, "mean": 35.786338806152344, "std": 66.4541244506836, "min": -118.34803771972656, "p10": -55.023414611816406, "median": 31.559307098388672, "p90": 119.84213104248046, "max": 186.53607177734375, "pos_frac": 0.7109375, "sample": [-12.570297241210938, -114.72271728515625, 43.02113342285156, -33.834197998046875, 93.3333740234375, 77.37103271484375, 95.55865478515625, -18.59527587890625, 20.020156860351562, 110.7591552734375, 125.02264404296875, 0.0, -16.67005729675293, -26.46246337890625, 6.310466766357422, 52.0411376953125, -17.248275756835938, 40.4864501953125, 119.66862487792969, 35.20428466796875, 71.49468994140625, -87.82655334472656, 109.13685607910156, -100.2711181640625, 80.18736267089844, -35.813323974609375, -15.148391723632812, 115.99313354492188, -69.40699768066406, -93.23865509033203, 87.95491027832031, 114.94241333007812, 120.24697875976562, 77.82246398925781, 126.70503234863281, 62.461822509765625, -14.387565612792969, 14.346492767333984, -47.7298583984375, 78.42765808105469, 14.348823547363281, 12.368621826171875, 126.99310302734375, -26.5443115234375, 1.4082412719726562, 19.518722534179688, -0.31871795654296875, 109.10395050048828, -85.33985900878906, 90.595458984375, 33.970733642578125, -23.64837646484375, 46.964996337890625, 137.2814483642578, 113.76321411132812, 49.056884765625, 70.55009460449219, -39.05488586425781, 108.40237426757812, 112.67176818847656, 114.46670532226562, 116.37541198730469, 73.38471984863281, 14.319122314453125, 87.23335266113281, -0.7124710083007812, 50.15526580810547, -73.26417541503906, 26.34551239013672, -1.4835739135742188, 105.3134765625, 186.53607177734375, 3.3690185546875, 146.07656860351562, 41.81178283691406, -6.98858642578125, 9.230316162109375, 2.5339717864990234, 8.873291015625, 11.937698364257812, 31.716583251953125, -6.0178985595703125, 111.57777404785156, -66.043701171875, 144.37060546875, 85.45318603515625, 38.50819396972656, 73.78692626953125, 17.366416931152344, -91.01327514648438, 15.794212341308594, 13.55194091796875, -92.10848999023438, 6.32843017578125, -56.6121826171875, 127.95806884765625, 106.87716674804688, 109.998046875, 111.23728942871094, 35.2882080078125, -54.34251403808594, 48.781776428222656, 0.9874763488769531, 101.10043334960938, 17.620590209960938, 50.114013671875, -63.290130615234375, 143.07333374023438, 123.80117797851562, -28.522605895996094, 123.73678588867188, -6.600532531738281, 31.40203094482422, -118.34803771972656, 41.820953369140625, 10.791763305664062, 11.28521728515625, 15.82989501953125, 84.99371337890625, 16.502288818359375, 82.74736022949219, 148.85467529296875, 11.857376098632812, -37.270751953125, -1.7907867431640625, 51.942718505859375, 4.4940338134765625, 88.86461639404297], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000261.npy"}
{"epoch": 0.5465968586387434, "step": 262, "batch_size": 128, "mean": 43.58715057373047, "std": 61.42725372314453, "min": -111.87860107421875, "p10": -26.811054992675775, "median": 31.24425506591797, "p90": 123.70712280273438, "max": 157.54742431640625, "pos_frac": 0.75, "sample": [102.1097412109375, 17.03948974609375, 108.722412109375, 29.78448486328125, 97.6171875, 3.6515426635742188, 147.0426025390625, 113.67044067382812, 24.799530029296875, 138.27572631835938, 156.88345336914062, -21.4251708984375, 74.19397735595703, 16.095138549804688, 12.55035400390625, 4.384366989135742, 97.00069427490234, 30.795074462890625, 5.637626647949219, 83.70133209228516, 123.1368408203125, -4.111724853515625, -66.30477142333984, 123.61135864257812, 46.7864990234375, 131.46923828125, 20.12115478515625, 27.503143310546875, 104.0018310546875, 143.090576171875, 28.443817138671875, -111.87860107421875, 29.820556640625, -17.036048889160156, -1.3913726806640625, -85.88923645019531, -45.482086181640625, 86.7554931640625, 1.91070556640625, 117.89932250976562, 31.693435668945312, -6.180278778076172, -7.946807861328125, -23.298736572265625, 34.91505432128906, 61.52368927001953, 71.64865112304688, -33.471282958984375, 54.909759521484375, -17.372955322265625, 75.75119018554688, 157.54742431640625, 100.02340698242188, 88.8323974609375, -36.85174560546875, -0.3729705810546875, 72.23104858398438, 44.248779296875, -25.37994384765625, 15.9158935546875, 31.853240966796875, -16.33050537109375, -6.9090423583984375, 72.79168701171875, 0.069976806640625, 12.47174072265625, 131.45211791992188, 12.163436889648438, 132.97872924804688, 46.313262939453125, -70.98489379882812, -1.4623565673828125, 126.8055419921875, 3.620391845703125, 70.87228393554688, -81.540283203125, -36.48899841308594, 113.02838134765625, 0.42669677734375, -31.633590698242188, -97.317138671875, -19.810258865356445, 0.0, 121.0248794555664, 147.00460815429688, 40.44822692871094, 15.184898376464844, 91.74930572509766, 23.46648406982422, 16.333541870117188, 104.74199676513672, 109.1923828125, 107.02604675292969, 19.1053466796875, 123.93057250976562, 78.88922119140625, 0.719573974609375, 74.31684875488281, 48.048370361328125, 113.27806091308594, 7.575969696044922, -8.24591064453125, 115.00115966796875, 114.1300048828125, -30.150314331054688, 12.582374572753906, 26.95220947265625, 28.334228515625, 126.18524169921875, 55.3150634765625, -2.668407440185547, 31.743240356445312, 142.97183227539062, 88.3660888671875, 109.69869995117188, 21.20648193359375, 123.20338439941406, 72.14022827148438, 115.81051635742188, 108.30952453613281, 17.459869384765625, -75.61577606201172, 49.33294677734375, 73.40008544921875, 8.789039611816406, 91.73927307128906, -1.508453369140625, -23.08673095703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000262.npy"}
{"epoch": 0.5486910994764398, "step": 263, "batch_size": 128, "mean": 36.55619812011719, "std": 65.1952896118164, "min": -142.76571655273438, "p10": -28.74175796508788, "median": 17.73635196685791, "p90": 129.35930786132812, "max": 189.30859375, "pos_frac": 0.65625, "sample": [132.83322143554688, 76.29376220703125, 128.72177124023438, 24.51312255859375, -5.332908630371094, 84.42900085449219, -24.3779296875, 6.071014404296875, 79.41545867919922, 57.26202392578125, -15.849235534667969, 83.82371520996094, 26.13671875, 23.45166015625, -43.808837890625, 4.539348602294922, -0.05860137939453125, -82.61923217773438, 51.8485107421875, -4.027587890625, 55.280487060546875, -43.885894775390625, -15.679794311523438, 83.70823669433594, -11.939849853515625, 123.09268188476562, 47.54252624511719, 77.38700103759766, -11.283935546875, -35.78912353515625, -25.721458435058594, 64.51449584960938, 133.80364990234375, 133.34552001953125, 142.3736572265625, -19.123153686523438, -10.844696044921875, -14.161651611328125, -54.5765380859375, 119.60548400878906, 7.56268310546875, -5.172637939453125, -8.921173095703125, 139.41073608398438, 54.642913818359375, 17.696725845336914, 96.35671997070312, -12.281234741210938, 117.09786987304688, 29.60637664794922, 92.703125, -19.50042724609375, -20.090179443359375, 89.57847595214844, 130.84689331054688, 82.55392456054688, -77.17816162109375, 5.739501953125, 107.9151611328125, 114.21830749511719, 0.0, 32.25970458984375, -142.76571655273438, 7.628925323486328, 70.84579467773438, 8.276519775390625, -0.973052978515625, 1.9790534973144531, 12.186477661132812, 32.28712463378906, 138.9183807373047, 3.9819717407226562, -61.21421813964844, -7.466848373413086, 12.425338745117188, 8.66058349609375, 87.62710571289062, 79.91796875, 70.414306640625, 2.95074462890625, -6.319610595703125, 7.695159912109375, -49.4052734375, 8.898727416992188, 15.585494995117188, 64.71466064453125, -90.84939575195312, 137.34088134765625, 123.4752197265625, 17.775978088378906, -23.43994140625, 138.84188842773438, 12.486083984375, -117.475830078125, -3.4670257568359375, -7.930999755859375, -87.22488403320312, 133.38739013671875, 19.317733764648438, -41.44244384765625, 96.25390625, 6.9651947021484375, 0.0, 8.3612060546875, 90.8309326171875, 71.57733154296875, 189.30859375, 31.255630493164062, 171.504638671875, -8.376655578613281, 106.375244140625, 3.1392898559570312, 125.66497802734375, 0.0, 117.7918701171875, 101.19393157958984, 21.393707275390625, 26.155006408691406, 106.8958511352539, -1.42156982421875, 118.13443756103516, 114.39506530761719, 65.03709411621094, -2.1114501953125, 168.2440185546875, -12.86984634399414, 89.60418701171875, -19.711517333984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000263.npy"}
{"epoch": 0.5507853403141362, "step": 264, "batch_size": 128, "mean": 45.739295959472656, "std": 60.31873321533203, "min": -119.25161743164062, "p10": -15.293597412109374, "median": 34.081979751586914, "p90": 131.32268981933595, "max": 188.14208984375, "pos_frac": 0.7421875, "sample": [-4.766990661621094, -12.685470581054688, 154.68862915039062, 64.74365234375, 137.35009765625, 106.5307388305664, 82.329833984375, 92.28536987304688, 6.703828811645508, 11.57452392578125, -5.185665130615234, 125.29360961914062, 41.6380615234375, 31.5574951171875, -16.24725341796875, 102.08988189697266, 83.47108459472656, 116.02565002441406, -17.56085968017578, 1.0582275390625, 113.61181640625, 116.41352844238281, -12.358245849609375, 70.5173110961914, -48.279884338378906, 162.043701171875, 118.84085083007812, 0.5954933166503906, 59.84442138671875, 65.453125, -28.012603759765625, 101.46815490722656, 124.73959350585938, 116.42146301269531, 144.51373291015625, 0.0, -8.688018798828125, 145.01031494140625, 65.76226806640625, 11.171180725097656, -19.999725341796875, 131.94998168945312, -17.896652221679688, 64.95843505859375, 78.42438507080078, 113.975830078125, -31.425201416015625, 150.31964111328125, -61.7135009765625, 35.3297119140625, 1.249359130859375, 15.299163818359375, 119.71282958984375, 59.85585021972656, 107.446044921875, -84.43768310546875, 120.36074829101562, 96.35027313232422, 38.10867691040039, 18.156715393066406, 52.077056884765625, -1.4584770202636719, 43.51362609863281, 40.20513916015625, 45.430328369140625, -23.090301513671875, -5.480224609375, 22.49468994140625, 17.877838134765625, 117.01815795898438, -1.026123046875, 10.336685180664062, -119.25161743164062, -13.908241271972656, 31.05184555053711, -2.9997634887695312, 32.83424758911133, 9.5340576171875, 131.19046020507812, 131.6312255859375, 87.07861328125, 1.669952392578125, -1.0404796600341797, 6.3249359130859375, -1.9436721801757812, 188.14208984375, -4.27020263671875, 0.0, 53.582275390625, 31.099109649658203, -43.949951171875, 40.38421630859375, 107.60823822021484, 130.69793701171875, 18.88836669921875, 100.859375, -3.9674720764160156, 39.609130859375, 89.85835266113281, 20.823989868164062, 131.64352416992188, 70.10501098632812, 1.1939697265625, -14.8848876953125, 11.84442138671875, 14.362701416015625, 30.7755126953125, 5.018524169921875, -14.57415771484375, 115.63794708251953, 1.3735809326171875, 48.38362121582031, 42.667510986328125, 170.31448364257812, 36.3267822265625, 142.74130249023438, -6.6123046875, 22.59862518310547, 138.88330078125, -7.239173889160156, -69.52324676513672, 11.378555297851562, 85.25630187988281, 8.528213500976562, 89.26071166992188, 1.9340534210205078, 37.50604248046875, 14.3060302734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000264.npy"}
{"epoch": 0.5528795811518324, "step": 265, "batch_size": 128, "mean": 42.75794982910156, "std": 65.38953399658203, "min": -112.38180541992188, "p10": -40.537384033203125, "median": 41.24702072143555, "p90": 131.96219177246093, "max": 174.814453125, "pos_frac": 0.703125, "sample": [-0.4078483581542969, 149.19354248046875, 117.3211669921875, -90.32366943359375, 43.436859130859375, 7.13421630859375, 111.3097152709961, 75.82537841796875, 83.29806518554688, -78.54640197753906, 52.8104248046875, 40.16192626953125, 90.02474975585938, -16.66552734375, 133.071044921875, 96.07951354980469, 37.523468017578125, 131.30059814453125, 100.64157104492188, 86.19550323486328, 174.814453125, 138.19473266601562, 146.70556640625, 16.010421752929688, 146.70538330078125, 97.00149536132812, 125.05032348632812, 136.20706176757812, 131.48696899414062, -6.6580047607421875, -50.471031188964844, 106.24343872070312, 148.4378662109375, 7.95269775390625, 81.367431640625, 131.13330078125, 75.71627807617188, 110.11495971679688, 44.175872802734375, 129.0648193359375, -17.227020263671875, 0.381500244140625, -54.383514404296875, 42.332115173339844, -8.466209411621094, -14.858078002929688, -17.107666015625, -39.0096435546875, -3.774444580078125, 95.32929992675781, 1.188812255859375, -16.566207885742188, -33.53276062011719, 2.5150146484375, -53.847267150878906, 146.585693359375, 65.372314453125, 5.114471435546875, 80.89817810058594, -112.38180541992188, 30.224842071533203, 104.69854736328125, 5.0396728515625, 96.29289245605469, 23.96270751953125, -41.01312255859375, 127.19877624511719, 23.54302978515625, 33.20855712890625, -23.58722686767578, -40.33349609375, 121.58660888671875, 134.4071502685547, 19.077651977539062, 0.0, -4.334951400756836, -32.963409423828125, -75.31494140625, 98.57791137695312, 115.61219787597656, 36.0458984375, 6.34271240234375, 86.63532257080078, 3.23968505859375, 110.1767349243164, 50.786529541015625, 0.0, -5.604682922363281, -10.919647216796875, 86.27606201171875, 68.80233764648438, -29.92291259765625, -0.094879150390625, 1.8066635131835938, 46.14051818847656, 5.070228576660156, 51.232666015625, 162.18630981445312, 14.731842041015625, -64.55398559570312, 18.36469268798828, 52.445281982421875, 70.01528930664062, 67.99126434326172, 50.34185791015625, 43.41741943359375, -101.48965454101562, 139.29641723632812, 109.4603271484375, -42.35662841796875, -5.300300598144531, 60.686920166015625, 7.74627685546875, -10.949386596679688, 65.48149108886719, -41.0516357421875, -0.0406341552734375, 136.35928344726562, 82.61589050292969, 88.64317321777344, 106.22097778320312, -12.086944580078125, 42.512245178222656, 28.96917724609375, 22.316247940063477, -81.89111328125, 23.853591918945312, 89.98345947265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000265.npy"}
{"epoch": 0.5549738219895288, "step": 266, "batch_size": 128, "mean": 44.83421325683594, "std": 61.50374984741211, "min": -140.57952880859375, "p10": -12.459148406982422, "median": 33.94297790527344, "p90": 120.20613403320313, "max": 175.34844970703125, "pos_frac": 0.828125, "sample": [42.96337890625, 97.8243637084961, 165.0311279296875, 21.101959228515625, -12.044891357421875, -12.775604248046875, 98.20072937011719, 3.86376953125, -18.25018310546875, 11.357666015625, 31.33349609375, 0.8917236328125, -3.5447540283203125, 53.290283203125, 80.52162170410156, -9.967926025390625, -10.124053955078125, 39.83650207519531, 36.22678756713867, -91.74177551269531, 115.54905700683594, 20.274246215820312, 3.5646514892578125, 31.501800537109375, 127.38449096679688, 87.01251220703125, -0.150665283203125, 49.39111328125, 135.50067138671875, 120.81607055664062, 1.9319953918457031, 154.09127807617188, -9.599334716796875, 46.33551025390625, 6.076793670654297, 106.45538330078125, -39.54351806640625, 9.41659927368164, 78.52461242675781, 92.55325317382812, 5.9732666015625, 25.3975830078125, 15.499252319335938, -25.81305694580078, 115.29743194580078, 102.12936401367188, 86.95924377441406, 117.4320297241211, 13.223186492919922, 28.647384643554688, 12.13677978515625, 100.51356506347656, -111.4136962890625, -106.62744140625, 0.1663036346435547, -39.87425231933594, 5.231706619262695, 40.943511962890625, 36.114105224609375, 77.57366943359375, 122.79299926757812, 89.58953857421875, 15.060638427734375, 17.9488525390625, 20.962926864624023, 85.973388671875, 94.55958557128906, 109.01094055175781, 13.8223876953125, 26.876693725585938, 32.1859130859375, 136.6373291015625, 66.194580078125, 46.91558837890625, 114.49454498291016, 160.16891479492188, 39.134033203125, -2.18341064453125, -75.97808837890625, -97.30307006835938, -140.57952880859375, 124.00517272949219, 13.583709716796875, 117.01869201660156, -13.36016845703125, 42.361785888671875, 41.143768310546875, 20.569091796875, 119.94473266601562, 32.658843994140625, -74.31625366210938, 5.05633544921875, 0.25469970703125, 31.187225341796875, 113.93316650390625, 61.70452880859375, 98.71363830566406, 92.15357971191406, 54.89276123046875, 5.8204345703125, 17.77880859375, 52.07177734375, 29.679779052734375, 143.98516845703125, 22.967981338500977, 23.434600830078125, -12.323524475097656, 0.7083358764648438, 38.459754943847656, -5.5872344970703125, 111.14251708984375, 134.45516967773438, 141.84812927246094, 7.549751281738281, 116.72254943847656, 9.603485107421875, 115.71249389648438, 83.45870971679688, 16.581642150878906, 35.22711181640625, 175.34844970703125, 15.750980377197266, 114.49546813964844, 8.669776916503906, 113.41152954101562, 99.92359924316406, 84.47994995117188, 119.02162170410156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000266.npy"}
{"epoch": 0.5570680628272251, "step": 267, "batch_size": 128, "mean": 37.98130416870117, "std": 68.42058563232422, "min": -139.72662353515625, "p10": -39.99794616699218, "median": 30.276992797851562, "p90": 122.42633209228515, "max": 206.2528076171875, "pos_frac": 0.7109375, "sample": [3.6999969482421875, 109.54312133789062, 100.57844543457031, 89.93893432617188, 159.43682861328125, 9.465301513671875, 16.310260772705078, 29.5194091796875, -30.5955810546875, -1.32781982421875, -0.20130538940429688, 27.43206787109375, -38.541259765625, 73.21484375, 18.4600772857666, -27.108154296875, 7.432899475097656, 141.37579345703125, 139.9988250732422, 34.7779541015625, -70.81227111816406, 151.90252685546875, 117.18342590332031, 7.759166717529297, 60.08205032348633, -21.3388671875, 53.41841125488281, 101.92137145996094, 88.4310302734375, -121.0006103515625, 102.86024475097656, 25.39958953857422, 28.313568115234375, -20.5518798828125, 13.986297607421875, 91.74748229980469, -17.245309829711914, 139.9696044921875, 122.023193359375, 80.80951690673828, -82.57049560546875, 142.83993530273438, 64.36421203613281, 94.20306396484375, 12.124000549316406, 177.67715454101562, 7.93463134765625, -24.27386474609375, 97.57000732421875, 126.06869506835938, -101.28123474121094, -90.62295532226562, 97.48477172851562, 104.44799041748047, 9.76519775390625, 170.5177001953125, 96.39472198486328, 93.24343872070312, 136.62789916992188, 32.58888244628906, 130.2325897216797, 70.59120178222656, -108.20831298828125, 12.1964111328125, 35.20649719238281, 2.9149627685546875, 74.42169189453125, -59.42837905883789, 54.710845947265625, 96.54461669921875, 11.25372314453125, 96.09332275390625, 67.018798828125, -139.72662353515625, 61.67320251464844, 79.65145874023438, 10.13623046875, 65.798583984375, 6.222984313964844, -0.30256080627441406, -43.66352081298828, 80.497802734375, -8.591957092285156, 40.96630859375, -10.160720825195312, 84.031494140625, -47.08807373046875, 52.81278991699219, 16.03167724609375, 87.96463012695312, -23.544565200805664, -26.942184448242188, 57.62690734863281, -6.6337890625, 2.5382232666015625, 104.1156997680664, 40.133514404296875, 77.80705261230469, 56.488525390625, -10.26513671875, 117.191162109375, 30.1883544921875, -26.21051025390625, 55.49169921875, -11.705528259277344, 25.897857666015625, 24.084869384765625, 115.44242858886719, -7.175422668457031, 206.2528076171875, -23.406845092773438, -43.396881103515625, 98.31763458251953, 30.365631103515625, 88.63479614257812, 1.352365493774414, 119.51028442382812, 106.44416809082031, -8.40692138671875, -24.795135498046875, 123.36698913574219, -136.0659637451172, -4.67828369140625, 0.732147216796875, 53.219566345214844, -83.72528076171875, -30.568984985351562, 10.75115966796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000267.npy"}
{"epoch": 0.5591623036649215, "step": 268, "batch_size": 128, "mean": 40.26398468017578, "std": 62.49504089355469, "min": -107.43635559082031, "p10": -22.23387145996093, "median": 27.052169799804688, "p90": 125.50874633789061, "max": 171.8702392578125, "pos_frac": 0.7734375, "sample": [30.217132568359375, 9.920234680175781, 87.8954849243164, 97.50027465820312, 117.86907958984375, 8.691192626953125, -2.20196533203125, -71.8183364868164, -13.096229553222656, -59.884368896484375, 2.9230499267578125, 159.37295532226562, 19.224365234375, 124.91205596923828, 1.1046142578125, 85.70663452148438, 171.8702392578125, -4.895181655883789, -0.12005615234375, 100.11431884765625, 110.34381866455078, 119.63490295410156, 53.20856475830078, -98.89964294433594, 55.639404296875, 31.510910034179688, -7.487478256225586, 49.06488037109375, 27.853607177734375, 38.880470275878906, 7.521123886108398, 92.97200012207031, 100.158447265625, 95.98812866210938, -12.739212036132812, 0.7602043151855469, -107.43635559082031, -67.82681274414062, -27.394363403320312, 170.147705078125, 108.14315795898438, 33.52782440185547, 26.52947998046875, 128.64019775390625, 2.7467041015625, 13.132369995117188, 69.25859069824219, -99.04122924804688, -44.597015380859375, -20.022232055664062, 30.274276733398438, 24.6082763671875, 52.880706787109375, 124.40786743164062, 5.882072448730469, 39.998291015625, 149.9845428466797, -3.9091033935546875, 68.39801025390625, 10.310821533203125, 161.96121215820312, 18.82464599609375, 117.07251739501953, -8.465911865234375, -50.648223876953125, 22.556106567382812, 127.58364868164062, 43.143341064453125, -0.225799560546875, -9.4708251953125, 98.1033935546875, 18.809417724609375, 109.60525512695312, 114.61976623535156, 17.976898193359375, -9.510631561279297, 45.654541015625, 137.72634887695312, 125.02899169921875, 60.79840087890625, 23.825347900390625, 71.80308532714844, 69.15032958984375, 1.8695793151855469, 72.45856475830078, 105.99653625488281, -70.49549865722656, 53.132904052734375, 45.43754577636719, 7.845947265625, 27.574859619140625, 126.628173828125, 150.84799194335938, 83.83792114257812, 0.3153076171875, 121.1697998046875, 17.159500122070312, 5.647857666015625, 39.8717041015625, 45.3475341796875, -9.438812255859375, 29.54608154296875, 52.5257568359375, 1.0506744384765625, 5.261510848999023, 63.136383056640625, 11.4979248046875, -39.008087158203125, 0.849945068359375, 44.089988708496094, 160.17984008789062, 12.394508361816406, 97.52230834960938, 120.50477600097656, 1.46484375, 142.45675659179688, -88.5760726928711, -7.2061920166015625, 168.47555541992188, 12.72711181640625, 0.3514842987060547, -38.016021728515625, 19.12006378173828, 19.772567749023438, 8.22674560546875, 12.455162048339844, -6.502105712890625, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000268.npy"}
{"epoch": 0.5612565445026177, "step": 269, "batch_size": 128, "mean": 38.816890716552734, "std": 68.00169372558594, "min": -177.93331909179688, "p10": -32.19518318176269, "median": 30.57683563232422, "p90": 123.9541015625, "max": 171.48809814453125, "pos_frac": 0.71875, "sample": [-102.34344482421875, 68.06058502197266, 107.38153076171875, -48.628173828125, 38.0404052734375, -1.978240966796875, 14.903938293457031, 115.77017211914062, 14.28302001953125, -43.521942138671875, 96.173095703125, -2.2490005493164062, -8.241943359375, 65.45071411132812, -30.4078369140625, 113.393798828125, 6.115570068359375, -2.37176513671875, 119.42682647705078, 89.08174896240234, 171.48809814453125, 153.3711395263672, 21.62762451171875, 137.68914794921875, 150.06686401367188, -22.751731872558594, 31.571014404296875, 116.46662902832031, 1.842529296875, 70.43405151367188, 3.6327552795410156, 132.17787170410156, 102.87579345703125, -86.73941040039062, 101.48526000976562, 0.0, 33.7310791015625, -54.74797058105469, 85.10429382324219, 101.07070922851562, -65.6143569946289, 17.920089721679688, 133.4632568359375, 93.38487243652344, 7.057586669921875, 87.1035385131836, 142.2609405517578, 0.0, 66.78219604492188, 42.98052978515625, 101.6579360961914, -17.667724609375, 123.61082458496094, -117.69879150390625, -73.46495056152344, 120.37391662597656, 98.3678207397461, -17.28998565673828, -107.0137939453125, -0.3542766571044922, 120.63400268554688, 31.097213745117188, 50.9346923828125, 7.915435791015625, 102.36475372314453, -24.25811767578125, 12.534332275390625, 82.05563354492188, 115.12812805175781, 12.12689208984375, 68.7138671875, 108.33261108398438, 0.0, 124.75508117675781, 14.794731140136719, 18.513153076171875, -128.12185668945312, 2.6767425537109375, 46.32916259765625, 5.924596786499023, -6.741485595703125, -26.58062744140625, 129.83111572265625, 20.79975128173828, 95.05258178710938, 30.05645751953125, 33.991241455078125, 54.302001953125, 89.53411865234375, 127.4102783203125, 44.02813720703125, 26.5552978515625, 10.3851318359375, -17.44091796875, 116.26510620117188, 25.07330322265625, -15.681594848632812, 107.83670043945312, 8.611351013183594, 93.45884704589844, 40.08894348144531, 45.13304138183594, 2.2973880767822266, -19.87676239013672, 4.9706268310546875, -108.21646118164062, 81.59540557861328, 130.613525390625, 8.284912109375, 80.70867156982422, -25.70220947265625, 10.112068176269531, 86.43063354492188, 1.71051025390625, 98.86109924316406, -36.365657806396484, -28.839065551757812, -5.828605651855469, 15.076461791992188, -8.729583740234375, 83.72006225585938, 129.50340270996094, 73.35687255859375, -177.93331909179688, -14.463531494140625, 16.656478881835938, 121.56040954589844, 150.04428100585938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000269.npy"}
{"epoch": 0.5633507853403141, "step": 270, "batch_size": 128, "mean": 46.27226638793945, "std": 72.71913146972656, "min": -172.24331665039062, "p10": -41.235260009765625, "median": 43.280616760253906, "p90": 132.48014221191406, "max": 206.52789306640625, "pos_frac": 0.734375, "sample": [-138.066162109375, 45.27903747558594, 104.61309814453125, 94.3980941772461, -41.65032958984375, 130.01983642578125, -12.113407135009766, 27.273345947265625, 77.4322509765625, -48.970123291015625, 122.13664245605469, 20.915313720703125, 21.5286865234375, 44.181427001953125, 21.565277099609375, 109.39962768554688, -12.872100830078125, 135.65968322753906, 88.93863677978516, -77.98078918457031, -41.946258544921875, 57.31220245361328, 7.199714660644531, 36.153167724609375, 24.405134201049805, -0.936767578125, 113.02178955078125, 109.7790756225586, -21.640869140625, 1.0251884460449219, 39.335968017578125, 117.10182189941406, 53.910125732421875, 3.331279754638672, 8.310653686523438, 9.929435729980469, -41.057373046875, 42.37980651855469, -144.73837280273438, 76.95468139648438, -55.61700439453125, 134.05120849609375, 26.4425048828125, 13.127532958984375, 120.02447509765625, 8.51104736328125, -5.1272735595703125, 131.93374633789062, 89.65724182128906, -104.1397705078125, -115.99005126953125, 122.54513549804688, 129.63043212890625, 129.15110778808594, 58.53296661376953, 54.11358642578125, 140.72341918945312, -35.313232421875, 122.2305908203125, 4.487030029296875, 30.386566162109375, -20.214996337890625, 11.26556396484375, 17.109447479248047, 126.14520263671875, 112.82878112792969, 139.61935424804688, 6.413543701171875, 123.7681884765625, 80.24200439453125, 91.83660888671875, -4.6153564453125, -129.93685913085938, 116.87000274658203, -51.276153564453125, 133.75506591796875, 64.6385498046875, 34.709346771240234, 101.74595642089844, -27.34783935546875, -12.698638916015625, 10.014205932617188, 80.49369812011719, 117.24560546875, 22.805957794189453, -6.7031402587890625, -0.289642333984375, -3.9139404296875, 0.33577728271484375, -3.7067337036132812, -36.576324462890625, -45.21232604980469, 102.67465209960938, 158.706298828125, -8.356231689453125, 51.84459686279297, 92.16436767578125, 133.79486083984375, 57.631103515625, -11.045654296875, 116.0927734375, 156.86090087890625, 25.242889404296875, 88.22650909423828, 85.65455627441406, -172.24331665039062, 38.199127197265625, -9.3616943359375, 142.21914672851562, 9.115753173828125, 71.91610717773438, 110.14910888671875, 65.60919189453125, 112.32492065429688, 126.49575805664062, 140.2442626953125, 139.19406127929688, 27.435256958007812, 10.89776611328125, 143.65036010742188, -18.47247314453125, 206.52789306640625, 119.72720336914062, 123.65663146972656, 95.89144134521484, 101.89599609375, 107.49676513671875, -29.4412841796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000270.npy"}
{"epoch": 0.5654450261780105, "step": 271, "batch_size": 128, "mean": 43.275421142578125, "std": 62.61290740966797, "min": -102.8133544921875, "p10": -16.557683563232413, "median": 30.026504516601562, "p90": 119.432666015625, "max": 206.30120849609375, "pos_frac": 0.78125, "sample": [102.6968002319336, -76.15821075439453, -8.027984619140625, 31.9268798828125, 15.082229614257812, -1.545440673828125, 12.998664855957031, -10.302295684814453, 33.19805908203125, -5.870338439941406, 19.393218994140625, 28.196945190429688, 119.01056671142578, 5.516204833984375, 14.496170043945312, 36.501129150390625, 85.0089111328125, 35.15382385253906, 30.976760864257812, 37.54963684082031, 20.735836029052734, 9.669105529785156, 117.318115234375, 104.38858032226562, 2.31463623046875, 4.541626930236816, -47.83544921875, 107.9417724609375, 25.918277740478516, -10.16693115234375, 105.66168212890625, 128.98236083984375, -3.446044921875, 42.6424560546875, 2.384490966796875, 25.22711181640625, 31.345947265625, 37.80082702636719, 90.57955932617188, -73.20733642578125, 112.00814819335938, 8.88897705078125, 1.546783447265625, 118.00796508789062, 103.3292236328125, 15.829315185546875, 46.066650390625, 0.299407958984375, 111.65293884277344, 9.71026611328125, -4.442108154296875, 75.49508666992188, 126.76637268066406, 60.328033447265625, 206.30120849609375, 150.8792724609375, 0.6268482208251953, 133.16357421875, 28.250701904296875, 107.6072998046875, 96.45458984375, 9.79998779296875, -5.294197082519531, 182.83282470703125, 103.91075134277344, -54.440826416015625, -21.776702880859375, 118.40359497070312, -7.901519775390625, 8.031402587890625, 117.96415710449219, 163.56927490234375, -31.9287109375, 91.00294494628906, -13.6734619140625, 20.948822021484375, -77.03089141845703, 29.643341064453125, 40.37628173828125, 92.42830657958984, -23.166259765625, -102.8133544921875, 133.6097412109375, -14.320960998535156, 158.50393676757812, 83.52458190917969, 22.429359436035156, -40.506683349609375, 39.980228424072266, 30.40966796875, 6.720211029052734, 133.76983642578125, 107.151611328125, -14.158615112304688, 5.74200439453125, 100.10221862792969, 8.963939666748047, 99.15323638916016, 113.61708068847656, 58.249244689941406, 18.421783447265625, 24.108016967773438, -4.94476318359375, 99.03936767578125, 149.53677368164062, -7.3884124755859375, -91.873779296875, 12.34027099609375, 17.150741577148438, 88.9197998046875, 37.190185546875, 103.47186279296875, 106.5738296508789, 9.698883056640625, 37.4498291015625, 15.5128173828125, 110.59487915039062, 119.49188232421875, 2.2309112548828125, -9.16259765625, 115.32881164550781, -88.98712158203125, 119.40728759765625, 144.66033935546875, 30.981639862060547, 0.88018798828125, 34.841163635253906, -67.4161376953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000271.npy"}
{"epoch": 0.5675392670157068, "step": 272, "batch_size": 128, "mean": 40.62211608886719, "std": 64.72366333007812, "min": -130.66751098632812, "p10": -25.155192565917968, "median": 28.587722778320312, "p90": 130.1058776855469, "max": 183.51556396484375, "pos_frac": 0.7578125, "sample": [85.08334350585938, 0.0, -7.7618408203125, 37.27470397949219, -88.82688903808594, 59.56976318359375, 157.35519409179688, 70.12066650390625, 108.35545349121094, 8.423309326171875, -24.866806030273438, 23.79610824584961, 49.737152099609375, 114.00508117675781, 68.13031005859375, 147.04864501953125, 144.369140625, 103.82992553710938, 46.86572265625, 18.903976440429688, 107.92666625976562, 9.485496520996094, -7.01116943359375, 7.111236572265625, 110.3736572265625, -24.284217834472656, 4.83381462097168, 159.3939208984375, 1.5304336547851562, 14.224365234375, 111.01043701171875, 52.74962615966797, 68.77951049804688, 3.2595863342285156, 130.0218505859375, 86.6845703125, 153.48251342773438, 130.30194091796875, 15.0902099609375, -2.2300186157226562, 82.61820220947266, -44.56268310546875, -76.13177490234375, 73.62730407714844, -9.122222900390625, 28.0167236328125, 108.98321533203125, -47.60858154296875, -1.7022171020507812, 1.7011260986328125, 135.43807983398438, 108.73713684082031, -14.920591354370117, 21.66693115234375, 115.17376708984375, 144.8094482421875, 35.35784912109375, 45.46575927734375, 96.69012451171875, 0.9537696838378906, 1.5819320678710938, 9.430831909179688, 9.476760864257812, 13.5965576171875, 91.8738784790039, 53.464881896972656, 183.51556396484375, 54.446136474609375, 133.462646484375, 8.77838134765625, 121.8607177734375, -44.578857421875, 58.60693359375, 9.6881103515625, -110.12908935546875, -49.133270263671875, -7.848114013671875, 47.038055419921875, 5.609527587890625, 114.53323364257812, 94.05361938476562, 50.4332275390625, -8.446441650390625, 153.5159149169922, 3.48529052734375, -130.66751098632812, 31.79799461364746, 13.72592544555664, 29.158721923828125, 4.757373809814453, -5.387733459472656, 85.04692077636719, 6.3837738037109375, -88.45852661132812, 76.63156127929688, 57.7252197265625, 52.21295166015625, 106.7682876586914, -22.27679443359375, 7.536548614501953, 0.0, -25.828094482421875, 16.60146713256836, -104.04696655273438, -19.786712646484375, 2.77459716796875, 22.031585693359375, -0.9876556396484375, -3.4985733032226562, 26.88067626953125, 99.90872192382812, 181.30438232421875, -57.05133056640625, 60.873756408691406, 119.93646240234375, -85.80352783203125, 12.830497741699219, 3.8115406036376953, 7.65492057800293, -4.395355224609375, 137.01034545898438, 91.91191864013672, 51.750144958496094, 40.80401611328125, 38.122528076171875, 129.10693359375, 106.00747680664062, 69.09724426269531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000272.npy"}
{"epoch": 0.5696335078534032, "step": 273, "batch_size": 128, "mean": 44.456390380859375, "std": 68.52577209472656, "min": -139.31442260742188, "p10": -35.740744018554686, "median": 35.01002502441406, "p90": 132.826513671875, "max": 196.7386474609375, "pos_frac": 0.6875, "sample": [17.351806640625, 155.15228271484375, 142.50860595703125, 9.46119499206543, 154.63702392578125, -8.521903991699219, 174.2379150390625, 21.33056640625, 61.79179382324219, 8.077255249023438, -24.83721923828125, 48.924705505371094, -2.812061309814453, 70.40959167480469, 196.7386474609375, -139.31442260742188, 68.29449462890625, 168.3514404296875, 12.74578857421875, 53.178985595703125, -73.18621826171875, 71.54841613769531, 119.05206298828125, 28.45895004272461, 134.9862060546875, 27.5211181640625, -108.04537963867188, 34.166168212890625, 108.48391723632812, -11.791610717773438, 15.545116424560547, -36.997894287109375, -22.06963348388672, 118.40475463867188, -21.0941162109375, -0.004547119140625, 10.014434814453125, 3.574432373046875, -7.050750732421875, 116.72494506835938, -7.220470428466797, -3.624908447265625, 1.0539932250976562, 46.94610595703125, 64.51673889160156, 25.055572509765625, 101.51790618896484, -0.5372543334960938, -68.3486557006836, 4.529815673828125, 74.39093017578125, -25.14208984375, 139.56826782226562, -18.19384765625, 84.73109436035156, -3.612213134765625, -0.5819091796875, 63.05291748046875, 25.339187622070312, 94.04459381103516, -48.84283447265625, 105.23812103271484, 8.01287841796875, 100.09742736816406, -23.53045654296875, 140.05279541015625, 96.46585083007812, -0.3732452392578125, 97.0496826171875, 2.517547607421875, 114.32229614257812, 16.362503051757812, -49.763824462890625, -5.150115966796875, 59.95611572265625, 131.7066650390625, 33.40678405761719, 124.09762573242188, 110.18679809570312, -73.30050659179688, 96.66698455810547, 110.82512664794922, 54.20161437988281, -35.20196533203125, 57.543701171875, 108.93990325927734, 29.304290771484375, 49.789772033691406, 134.58984375, 76.02023315429688, 68.98104858398438, -49.958065032958984, -3.16754150390625, 109.49423217773438, -135.20578002929688, 35.8538818359375, -93.368408203125, 116.36332702636719, 131.66146850585938, -0.7511749267578125, 145.6240234375, 93.6136474609375, 137.037841796875, 121.96952819824219, 126.4942626953125, -5.946877479553223, -9.762908935546875, 15.5223388671875, 117.7088623046875, 101.08918762207031, 17.828948974609375, 50.11444091796875, -3.3823089599609375, 42.082763671875, 24.201828002929688, 61.986541748046875, -55.395660400390625, 132.07080078125, 24.21002197265625, -6.842033386230469, -10.483585357666016, 126.27230834960938, 45.843414306640625, 111.66779327392578, 137.44052124023438, -42.772247314453125, 119.5986328125, -21.900894165039062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000273.npy"}
{"epoch": 0.5717277486910994, "step": 274, "batch_size": 128, "mean": 48.14777374267578, "std": 67.07520294189453, "min": -136.94454956054688, "p10": -17.829341125488277, "median": 39.58363342285156, "p90": 127.48556213378906, "max": 198.27001953125, "pos_frac": 0.796875, "sample": [113.90174865722656, 118.67523193359375, -44.080047607421875, 103.00152587890625, -0.9408340454101562, 68.74151611328125, 66.91546630859375, 28.893600463867188, 97.70723724365234, 10.671431541442871, 25.085418701171875, 106.46613311767578, 160.95358276367188, 91.3042984008789, 21.689109802246094, 89.87619018554688, 171.3912353515625, 143.76722717285156, -61.182098388671875, -10.138553619384766, 20.2930908203125, 147.93692016601562, 118.99948120117188, 40.747100830078125, -0.261474609375, -4.00469970703125, -108.56046295166016, -22.911895751953125, 12.482812881469727, 51.292022705078125, 16.953369140625, 39.484344482421875, 103.27296447753906, 64.72738647460938, 115.12272644042969, 39.68292236328125, 40.28697204589844, 108.15567016601562, 6.7021942138671875, 121.80743408203125, 141.92276000976562, -13.418725967407227, -10.346923828125, 7.132438659667969, -17.119308471679688, 119.93040466308594, 16.272232055664062, 19.193710327148438, 40.95794677734375, 151.375, -10.156242370605469, 7.40472412109375, 109.4393310546875, 64.40982055664062, -120.73446655273438, 27.310958862304688, -33.09809875488281, 29.73199462890625, 0.4646453857421875, 25.705230712890625, 19.661243438720703, 5.03056526184082, -7.940162658691406, 93.93846130371094, -72.92892456054688, -1.8399524688720703, 124.46144104003906, -0.3204307556152344, 61.73835754394531, 113.4403076171875, 11.08544921875, -136.94454956054688, 133.9261474609375, 45.42817687988281, 61.36810302734375, 50.8743896484375, 46.422332763671875, 38.68940734863281, 198.27001953125, 8.21929931640625, 178.80502319335938, 101.14372253417969, 22.27655029296875, 123.89126586914062, 0.1955413818359375, 59.64239501953125, 112.04714965820312, 126.66156005859375, 73.1944580078125, 126.34520721435547, 118.31006622314453, 27.954132080078125, -64.26528930664062, 80.83073425292969, 105.48611450195312, 106.0589828491211, 14.912322998046875, 130.59902954101562, 150.29544067382812, 1.0272369384765625, -87.04292297363281, 17.0679931640625, 5.17803955078125, -15.32147216796875, 20.989654541015625, 111.66258239746094, 11.037055969238281, 124.81340026855469, 35.022796630859375, 54.645751953125, 129.40823364257812, 96.27787017822266, -41.916259765625, 61.92033386230469, 10.613693237304688, -19.486083984375, 3.852783203125, 1.955657958984375, 19.395782470703125, 56.222686767578125, 12.688465118408203, -135.4012451171875, 117.69329833984375, -0.9261989593505859, 23.82632064819336, 102.23141479492188, 115.48663330078125, 171.74166870117188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000274.npy"}
{"epoch": 0.5738219895287958, "step": 275, "batch_size": 128, "mean": 55.019020080566406, "std": 65.01801300048828, "min": -113.09005737304688, "p10": -12.492962646484374, "median": 48.027137756347656, "p90": 135.5640869140625, "max": 169.27621459960938, "pos_frac": 0.7890625, "sample": [167.854736328125, -12.09600830078125, -2.3839778900146484, -98.5946044921875, 131.23187255859375, 89.28567504882812, 127.6240234375, 11.610595703125, 21.6563720703125, -91.71487426757812, 27.617660522460938, 148.20599365234375, -34.004547119140625, 138.1378173828125, 0.0, 9.6614990234375, 16.714263916015625, 93.2479248046875, 3.122629165649414, 5.523651123046875, 0.0, 0.0, 113.24044799804688, 113.22160339355469, 98.5699691772461, 3.5671310424804688, 141.980712890625, 101.33464050292969, 19.8289794921875, 62.16896057128906, 58.034820556640625, 6.361392974853516, 93.97313690185547, 126.69329833984375, 151.58090209960938, 146.72866821289062, 0.5922603607177734, -0.856414794921875, -113.09005737304688, 44.477317810058594, 119.55078125, 31.184661865234375, 61.214744567871094, 25.500091552734375, 15.24078369140625, -4.995903015136719, 2.6380157470703125, 124.359130859375, 134.4610595703125, 17.149871826171875, 133.78390502929688, -99.34994506835938, 2.512521743774414, 36.28031921386719, 91.89813232421875, 20.318328857421875, 93.34906005859375, 27.3349609375, 129.82284545898438, 0.0, 22.026214599609375, 68.37031555175781, 27.3707275390625, 131.3939971923828, -5.419189453125, 155.34744262695312, 156.51480102539062, 87.39566040039062, 50.47557067871094, 0.0, 10.31834602355957, 139.2080535888672, 65.04670715332031, 106.12701416015625, -9.585441589355469, 106.6171875, 131.84637451171875, 169.27621459960938, 47.67451477050781, 8.078666687011719, 110.02910614013672, 121.2476806640625, 81.66888427734375, 100.74298095703125, 84.46053314208984, 2.568340301513672, 23.09136962890625, 108.68400573730469, 66.08291625976562, 34.961212158203125, -18.352447509765625, 90.64445495605469, 9.43792724609375, 150.35595703125, -0.49273681640625, 124.09857177734375, -13.419189453125, 13.8468017578125, 164.55111694335938, 10.712661743164062, 60.73753356933594, 29.27630615234375, 128.94009399414062, -4.601036071777344, 128.17921447753906, 116.71012115478516, 13.321998596191406, -45.1143798828125, -0.9398193359375, -18.3326416015625, 130.83529663085938, -85.07913208007812, 108.107421875, 1.1951904296875, 48.3797607421875, 7.131473541259766, 18.907562255859375, 115.33880615234375, 163.81741333007812, 112.52730560302734, -30.2515869140625, 117.658447265625, 64.30253601074219, -18.229263305664062, 97.45022583007812, -15.21148681640625, 120.69668579101562, 60.315399169921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000275.npy"}
{"epoch": 0.5759162303664922, "step": 276, "batch_size": 128, "mean": 50.742774963378906, "std": 67.93109130859375, "min": -104.28842163085938, "p10": -29.347607421874983, "median": 34.338951110839844, "p90": 137.20826416015626, "max": 216.0758056640625, "pos_frac": 0.7578125, "sample": [216.0758056640625, 17.434165954589844, 108.95279693603516, -4.2284088134765625, 19.6207275390625, -0.740631103515625, 14.934326171875, 63.518798828125, 44.493408203125, 75.43587493896484, 97.48429870605469, 6.309295654296875, -2.524017333984375, 2.0179004669189453, -4.3616790771484375, 88.0949935913086, 47.87591552734375, 34.02789306640625, -40.7242431640625, 79.64714050292969, 140.38754272460938, 82.98027038574219, 8.159160614013672, 137.75872802734375, 81.99105834960938, -25.11578369140625, 1.563629150390625, -74.461669921875, -9.85601806640625, 119.01971435546875, 56.94921875, -3.0409393310546875, 69.57513427734375, 113.0984115600586, 77.99026489257812, 19.92681884765625, 33.4259033203125, -1.0057506561279297, 107.06706237792969, 52.08686828613281, 1.813079833984375, 22.239883422851562, 124.24368286132812, 36.697784423828125, 5.476619720458984, 155.01644897460938, 100.7845458984375, 134.66390991210938, 16.8406982421875, -7.82513427734375, 8.289115905761719, 122.15786743164062, 105.86209106445312, 31.7301025390625, 127.77861785888672, 10.9798583984375, 64.63121032714844, 23.0128173828125, 129.99099731445312, -39.22186279296875, 0.0, -81.84637451171875, 34.65000915527344, -91.83963012695312, 8.806343078613281, -40.982757568359375, 55.421112060546875, 164.57699584960938, -2.2416763305664062, -18.982208251953125, 65.06585693359375, -17.7030029296875, 100.99551391601562, 125.65838623046875, 109.13959503173828, 132.9759521484375, 127.09317016601562, -22.060882568359375, 35.48919677734375, 113.22813415527344, 130.6793670654297, 109.96443176269531, 108.98768615722656, 36.6490478515625, 12.266860961914062, 83.15548706054688, -63.82293701171875, 108.49327850341797, -3.2630233764648438, 32.64129638671875, -4.711395263671875, 186.04083251953125, 20.13897705078125, 18.765716552734375, 11.132461547851562, 10.103866577148438, -0.1569366455078125, 147.6383056640625, 13.731491088867188, 128.0303955078125, 130.31011962890625, 30.969863891601562, 139.40032958984375, 1.050048828125, -2.4648895263671875, 94.42056274414062, -39.604209899902344, -42.614986419677734, 28.002197265625, 143.0904541015625, 6.31781005859375, 116.10891723632812, 18.623390197753906, 159.9559326171875, 127.421875, -91.0726318359375, 158.4334716796875, 21.120941162109375, 107.97328186035156, -45.82464599609375, 152.87356567382812, 11.595306396484375, 113.76249694824219, -104.28842163085938, -75.6865234375, 136.97235107421875, 164.71807861328125, 122.59461975097656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000276.npy"}
{"epoch": 0.5780104712041885, "step": 277, "batch_size": 128, "mean": 33.45337677001953, "std": 74.08748626708984, "min": -155.30255126953125, "p10": -64.53650665283203, "median": 29.082847595214844, "p90": 118.36764984130859, "max": 207.77264404296875, "pos_frac": 0.703125, "sample": [-1.292510986328125, 4.4551849365234375, -6.902761459350586, 115.19622802734375, 31.6368408203125, -22.49169921875, -114.28509521484375, -8.765289306640625, 107.79571533203125, 4.1756134033203125, 22.1341552734375, 84.93778991699219, 117.61430358886719, 97.25459289550781, -7.5118865966796875, 120.12545776367188, 93.99703979492188, -152.85858154296875, 79.7047119140625, 0.821502685546875, 30.330276489257812, 9.76535415649414, 92.07169342041016, 51.647735595703125, 92.20457458496094, 96.61129760742188, -0.7244882583618164, 1.250396728515625, 129.03347778320312, 0.42240142822265625, 83.70134735107422, -133.933837890625, -6.012481689453125, -39.810951232910156, -45.749916076660156, 12.538986206054688, 116.52198791503906, 6.359947204589844, 79.58289337158203, 64.70001220703125, -155.30255126953125, 3.083526611328125, -20.052696228027344, -26.460113525390625, 10.21136474609375, 85.36581420898438, 32.100555419921875, 105.98130798339844, 20.497879028320312, 46.758026123046875, 66.95489501953125, -10.495394706726074, 2.0645065307617188, 207.77264404296875, 49.7568359375, 0.0, 62.484466552734375, 89.91808319091797, 113.42449951171875, 27.835418701171875, 154.137451171875, 92.98574829101562, -111.0279541015625, -5.466180801391602, 0.93670654296875, 5.259796142578125, 91.82286071777344, -97.37210083007812, 177.41253662109375, 179.46685791015625, -65.87522888183594, 5.112152099609375, 155.42041015625, 54.716522216796875, 8.001800537109375, -152.6429443359375, 24.93695068359375, -72.865966796875, -78.6788330078125, 51.766292572021484, 19.23907470703125, 99.8246078491211, 54.194915771484375, -63.9627685546875, 75.21258544921875, 103.68568420410156, 103.42913818359375, 117.34848022460938, -91.73538208007812, 140.50286865234375, 2.769725799560547, 2.7779388427734375, -15.839569091796875, 68.76856231689453, 67.74139404296875, -24.517578125, 76.93838500976562, 101.74951171875, 38.54585266113281, 117.00448608398438, 7.5535888671875, 15.31939697265625, 110.30661010742188, 132.57748413085938, 33.81604766845703, 115.63096618652344, 48.032501220703125, 129.9212646484375, -52.27647399902344, 0.0, 132.98220825195312, 160.5496826171875, -23.166656494140625, 59.94232177734375, -39.0396728515625, -10.235107421875, -118.38931274414062, 125.33627319335938, 19.862869262695312, 49.77593231201172, 96.92582702636719, -108.90704345703125, -1.0495643615722656, -0.990631103515625, 8.454517364501953, 58.63050842285156, 32.416221618652344, -31.823740005493164], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000277.npy"}
{"epoch": 0.5801047120418849, "step": 278, "batch_size": 128, "mean": 31.6246337890625, "std": 74.94457244873047, "min": -173.7330322265625, "p10": -70.2983642578125, "median": 22.23367691040039, "p90": 126.67999877929687, "max": 205.45291137695312, "pos_frac": 0.7109375, "sample": [120.02410888671875, 97.06889343261719, -75.723876953125, 18.456954956054688, 13.090072631835938, 30.927200317382812, 66.08187866210938, 24.649429321289062, 13.235553741455078, 131.503173828125, 42.00494384765625, 118.33634185791016, 95.60882568359375, 31.191131591796875, 94.71873474121094, 108.38833618164062, -112.466796875, -17.412338256835938, 25.279891967773438, -3.708984375, 6.766899108886719, 75.27508544921875, 4.369964599609375, -25.368606567382812, 12.502471923828125, 173.7972412109375, 146.64138793945312, 37.54100799560547, 9.960693359375, -28.0899658203125, -167.75820922851562, 181.26226806640625, -95.98340606689453, 5.520172119140625, 73.97894287109375, 100.17880249023438, 81.50509643554688, -67.97314453125, 150.80426025390625, 20.90271759033203, 129.87567138671875, 148.43365478515625, 91.93844604492188, 12.74005126953125, -18.157630920410156, 56.614501953125, -1.044891357421875, 149.688232421875, 42.116859436035156, 53.45008087158203, 8.959318161010742, -1.58233642578125, 3.0865478515625, -87.90518188476562, -17.4588623046875, -173.7330322265625, 2.984699249267578, 155.07861328125, -79.84324645996094, -96.68096923828125, 53.58349609375, 79.8201904296875, 25.1275634765625, 8.496826171875, 10.31243896484375, 125.3104248046875, -59.238922119140625, 116.09767150878906, 115.89408111572266, 31.242538452148438, 157.07806396484375, 205.45291137695312, -51.66498565673828, -51.91288757324219, 75.44488525390625, -102.04808044433594, 101.07617950439453, 58.45808410644531, 74.82736206054688, 34.601409912109375, -130.79856872558594, 16.083953857421875, 3.253793716430664, 49.18701171875, 104.66250610351562, -12.555137634277344, 18.9136962890625, 53.25025177001953, 8.97698974609375, 105.11824035644531, 1.6942367553710938, -8.28436279296875, 115.21669006347656, 173.00262451171875, -22.703826904296875, 123.20379638671875, -42.05442810058594, 34.696380615234375, 90.79908752441406, -1.252960205078125, -9.902557373046875, 7.722320556640625, 23.56463623046875, -86.3387680053711, -14.9520263671875, 26.24622344970703, 131.89068603515625, -36.58240509033203, 25.699920654296875, 11.834178924560547, 111.68226623535156, 4.049015045166016, -0.71844482421875, 79.7188720703125, 4.930381774902344, -28.40160369873047, 31.228574752807617, -79.4815673828125, 93.44845581054688, -32.8070068359375, 5.362461090087891, 20.261459350585938, 10.9471435546875, 122.01023864746094, -121.12060546875, 73.02310180664062, -25.61695098876953, 116.23823547363281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000278.npy"}
{"epoch": 0.5821989528795811, "step": 279, "batch_size": 128, "mean": 52.2238883972168, "std": 67.58382415771484, "min": -120.3592758178711, "p10": -17.486212921142574, "median": 38.92644500732422, "p90": 141.04647521972657, "max": 214.60101318359375, "pos_frac": 0.796875, "sample": [40.87188720703125, 111.74295043945312, 39.32908630371094, -22.38323211669922, 7.472503662109375, -2.5172119140625, 67.91810607910156, 74.80120849609375, 5.544639587402344, 122.41732788085938, 80.00588989257812, 156.9869384765625, 38.5238037109375, -37.91375732421875, -2.998046875, 34.44635009765625, 136.52291870117188, 104.3839340209961, 36.88232421875, 102.33917236328125, 18.936965942382812, 40.574424743652344, -9.2581787109375, 35.49609375, 42.635398864746094, 26.908355712890625, 17.66790008544922, 5.741546630859375, -114.00465393066406, -6.927406311035156, 92.65390014648438, 162.45318603515625, 128.925048828125, 124.353759765625, 67.83041381835938, -35.277496337890625, 6.076454162597656, 104.38691711425781, -24.693771362304688, 12.15692138671875, 107.40714263916016, 120.23722839355469, 117.50730895996094, 146.08543395996094, 6.105224609375, 141.02951049804688, -16.085494995117188, 11.693756103515625, 0.1284027099609375, 98.79325866699219, 23.314727783203125, 30.06182861328125, 10.548446655273438, 39.37265396118164, 36.31571960449219, -7.6994476318359375, 214.60101318359375, -120.3592758178711, 109.1861572265625, 160.00363159179688, -13.364063262939453, -15.486328125, 18.927490234375, 23.3125, 3.01031494140625, -10.665939331054688, -21.939071655273438, 123.08937072753906, -2.7602157592773438, 181.3780517578125, -31.297210693359375, 185.94049072265625, 57.64494323730469, 87.85013580322266, 9.337600708007812, -115.24189758300781, 10.8720703125, 132.413818359375, 168.90115356445312, -102.85964965820312, 8.100799560546875, 24.189437866210938, 71.46163940429688, 90.18402099609375, 103.87680053710938, 85.93832397460938, 54.40106201171875, 8.340694427490234, 139.3859405517578, 21.270301818847656, 89.84698486328125, -22.182357788085938, 109.5567626953125, 46.727020263671875, 22.1105899810791, 17.28497314453125, 137.15719604492188, 2.319652557373047, 47.1002197265625, 130.0576629638672, 96.464111328125, 141.0860595703125, -10.915061950683594, 126.93386840820312, 154.14697265625, 137.47186279296875, 16.36663818359375, 10.406982421875, 18.26800537109375, 143.5951690673828, -2.216461181640625, 35.508087158203125, 43.572967529296875, 170.17724609375, 113.66700744628906, 54.5589599609375, 19.564453125, 43.201171875, -87.60597229003906, -20.754554748535156, 154.07562255859375, -7.0548095703125, 92.12347412109375, 3.087982177734375, 125.47000122070312, 64.552001953125, 37.1885986328125, 116.26608276367188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000279.npy"}
{"epoch": 0.5842931937172775, "step": 280, "batch_size": 128, "mean": 45.62397766113281, "std": 72.27284240722656, "min": -149.40576171875, "p10": -39.667322540283195, "median": 30.33648109436035, "p90": 143.03274230957032, "max": 177.91311645507812, "pos_frac": 0.7421875, "sample": [111.4737548828125, 115.73677062988281, 0.896240234375, 117.18112182617188, -3.9452152252197266, -149.40576171875, 133.87997436523438, 120.71209716796875, 155.254150390625, 23.942092895507812, -17.939285278320312, 31.604835510253906, 6.5518646240234375, 96.36224365234375, 17.53741455078125, -36.49327850341797, 144.3583984375, 91.4166259765625, 150.15744018554688, -6.102142333984375, 4.514801025390625, 143.7711181640625, -74.38662719726562, 145.11929321289062, -31.0250244140625, 108.3935546875, 26.76567268371582, 19.24041748046875, -9.084121704101562, 115.37063598632812, 0.0, 21.06256103515625, 31.2117919921875, -34.324981689453125, 46.601593017578125, 177.91311645507812, -5.546714782714844, 131.05198669433594, 11.287212371826172, -1.100738525390625, -7.395355224609375, 97.02296447753906, 8.866374969482422, 21.13958740234375, 101.45427703857422, 58.01005554199219, 109.85015869140625, -111.06925964355469, 142.79861450195312, 19.118927001953125, 9.362213134765625, 8.403984069824219, 6.85919189453125, 6.836175918579102, 105.4075927734375, 148.33367919921875, 62.21392822265625, -26.20172119140625, 3.4017486572265625, 118.30441284179688, 127.90374755859375, 33.91864013671875, 72.5511474609375, -77.79415130615234, -7.263153076171875, 47.69921875, -99.64178466796875, 137.12667846679688, 6.847137451171875, 90.76707458496094, 127.0477294921875, 14.067543029785156, 143.57904052734375, 146.16140747070312, 148.25689697265625, -23.73004150390625, 100.81684875488281, 86.03140258789062, 25.686279296875, 52.22314453125, 95.28399658203125, 138.578857421875, -47.550018310546875, 113.44892883300781, 132.28359985351562, -10.312355041503906, 16.0087890625, 46.351470947265625, -47.07342529296875, 0.409881591796875, 141.746337890625, 41.8775634765625, 93.77937316894531, -85.90974426269531, 66.22386169433594, 156.39141845703125, -8.899566650390625, 93.15443420410156, 137.61065673828125, 32.47015380859375, 131.0426483154297, 9.574165344238281, 120.40560913085938, -18.215133666992188, 17.930679321289062, 57.39459228515625, -0.5039749145507812, -5.1043701171875, 152.99957275390625, 6.656005859375, 20.802330017089844, 27.23638916015625, 6.902807235717773, 97.13880920410156, -66.94577026367188, 98.87983703613281, 118.53524780273438, 172.60760498046875, -74.21783447265625, -129.78550720214844, -19.3369140625, 10.710262298583984, -82.78239440917969, 29.461170196533203, 58.77970886230469, 7.123632431030273, 79.53570556640625, -57.8173828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000280.npy"}
{"epoch": 0.5863874345549738, "step": 281, "batch_size": 128, "mean": 46.97228240966797, "std": 66.25015258789062, "min": -128.32159423828125, "p10": -18.038628578186035, "median": 32.52156066894531, "p90": 138.20011291503906, "max": 244.490478515625, "pos_frac": 0.7421875, "sample": [107.5677490234375, 57.40593719482422, 0.686553955078125, 3.1811294555664062, 71.29605102539062, 10.910308837890625, 47.71636962890625, 76.56218719482422, 29.5657958984375, 143.63192749023438, 133.3375244140625, -102.15878295898438, -111.30313110351562, 149.15252685546875, 18.266368865966797, 28.514404296875, 110.65155792236328, 96.29608154296875, 4.3623504638671875, 13.343605041503906, 144.15257263183594, 11.426979064941406, -6.7805938720703125, -21.084381103515625, 121.29119873046875, -9.525833129882812, -46.29681396484375, 142.93182373046875, 19.585205078125, 27.420806884765625, 0.0, 48.2393798828125, -2.985015869140625, 45.58990478515625, 38.202205657958984, 124.42988586425781, 37.890045166015625, -8.82818603515625, 146.19908142089844, 97.97567749023438, -128.32159423828125, 38.903106689453125, 122.6063232421875, -18.489713668823242, 0.0, 29.39849853515625, 40.859375, 119.81576538085938, -22.7044677734375, 33.274383544921875, 90.935546875, 2.81475830078125, 76.71525573730469, 39.48945617675781, 1.2536773681640625, 166.04794311523438, -7.895416259765625, 38.728363037109375, -6.03839111328125, 63.87654113769531, 13.95892333984375, 40.850372314453125, -24.710205078125, -1.7288188934326172, 96.1346435546875, 68.17117309570312, 23.45526885986328, -10.973281860351562, -5.13018798828125, 5.861297607421875, 18.389816284179688, 137.04600524902344, 1.65289306640625, 138.731201171875, 125.48410034179688, 36.5264892578125, 135.42816162109375, 1.608642578125, 31.76873779296875, 92.26402282714844, 46.104774475097656, 91.02040100097656, 244.490478515625, 135.54092407226562, 129.4339599609375, -33.86424255371094, -4.49383544921875, 103.58692932128906, 38.55558776855469, 3.1090011596679688, 168.75811767578125, 31.031280517578125, 102.85935974121094, -3.6673011779785156, 19.130950927734375, -60.10382080078125, -8.240936279296875, 152.80511474609375, 119.90673828125, 16.17205810546875, -25.619293212890625, 48.742340087890625, 0.0, 149.53759765625, 7.675350189208984, -17.845306396484375, 100.56564331054688, -101.37725830078125, 158.04595947265625, 147.35049438476562, 117.49864959716797, 137.82260131835938, 128.15817260742188, -3.81011962890625, 137.97250366210938, 15.173370361328125, -14.967727661132812, 76.18325805664062, 11.807525634765625, 17.27996826171875, 22.62823486328125, -34.794036865234375, -6.8671112060546875, 132.512451171875, 16.029937744140625, 66.6412353515625, 33.65496826171875, -4.56201171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000281.npy"}
{"epoch": 0.5884816753926702, "step": 282, "batch_size": 128, "mean": 41.4055290222168, "std": 75.53102111816406, "min": -156.62344360351562, "p10": -43.86911315917968, "median": 28.287330627441406, "p90": 140.39600524902343, "max": 204.51904296875, "pos_frac": 0.734375, "sample": [135.0116424560547, -71.64926147460938, 41.653167724609375, 204.51904296875, 76.79714965820312, 10.320655822753906, -156.62344360351562, 140.38931274414062, 98.19564819335938, 4.3485565185546875, 10.964134216308594, 95.94143676757812, 22.647125244140625, -43.65142822265625, 10.2486572265625, 9.82647705078125, 158.64730834960938, -36.75946044921875, 124.757080078125, -7.101051330566406, 12.84857177734375, 41.659210205078125, -80.10659790039062, 39.902679443359375, 15.453826904296875, 195.95248413085938, 126.98896789550781, 41.37060546875, -2.1959228515625, 12.303375244140625, -41.61614990234375, 20.208755493164062, 112.282958984375, 55.73326110839844, -4.85662841796875, 17.681442260742188, 100.08682250976562, 32.53221130371094, 45.2235107421875, 45.07598876953125, 29.511886596679688, 95.10197448730469, 9.66131591796875, -69.40972900390625, 123.9576416015625, -32.03703308105469, -113.89633178710938, 38.95286560058594, 149.96035766601562, 12.707817077636719, 24.89239501953125, 9.422393798828125, -25.645111083984375, 11.516181945800781, -33.856964111328125, 173.04351806640625, 40.269866943359375, 1.2598419189453125, 17.320423126220703, -78.22792053222656, 6.887542724609375, -0.31097412109375, -26.63519287109375, 123.30490112304688, 134.49838256835938, 102.6763916015625, 99.59842681884766, -10.338584899902344, -10.145820617675781, 131.36651611328125, -102.97088623046875, -2.885589599609375, 127.44292449951172, 68.98265075683594, 18.932647705078125, 162.49310302734375, 120.312255859375, -73.39371490478516, 78.69122314453125, -27.83514404296875, -32.23860168457031, 10.464996337890625, -7.4508056640625, 162.3439178466797, 42.052642822265625, -71.39402770996094, 19.153396606445312, 15.175888061523438, 154.29388427734375, 138.18817138671875, 169.2073974609375, 27.062774658203125, 14.4884033203125, 75.06381225585938, -112.17926025390625, 52.052276611328125, 117.65238952636719, 140.41162109375, 8.330877304077148, 1.426361083984375, 126.94918823242188, 198.7237548828125, 113.77546691894531, 11.132671356201172, 112.91946411132812, -1.084381103515625, -139.51382446289062, 29.589263916015625, 35.27923583984375, 30.159534454345703, 128.3447265625, 30.406097412109375, 57.528045654296875, 103.16024017333984, 7.6207733154296875, 132.71514892578125, 147.95071411132812, 3.2411651611328125, 142.0277862548828, 59.012176513671875, 36.52537536621094, -97.71737670898438, -3.5124969482421875, -44.377044677734375, 120.9520263671875, -6.428802490234375, -0.69659423828125, 120.93017578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000282.npy"}
{"epoch": 0.5905759162303665, "step": 283, "batch_size": 128, "mean": 45.29396057128906, "std": 69.95589447021484, "min": -115.81240844726562, "p10": -52.32554473876953, "median": 32.85279083251953, "p90": 134.1828826904297, "max": 194.28399658203125, "pos_frac": 0.75, "sample": [-51.211517333984375, 25.008499145507812, 135.0124969482422, -25.24285888671875, 7.958051681518555, -52.893280029296875, 126.67918395996094, -64.71153259277344, 59.612060546875, 2.1696319580078125, -6.083099365234375, 129.44384765625, 68.27281951904297, 92.29932403564453, -1.8795337677001953, -48.58671569824219, -19.873779296875, 85.95092010498047, 3.305084228515625, -34.736968994140625, 188.4727783203125, 151.37985229492188, 32.229644775390625, -54.053955078125, 116.86007690429688, 117.33810424804688, 85.84488677978516, 13.102508544921875, 74.98503112792969, 14.715377807617188, 74.60997009277344, 9.097442626953125, 20.564193725585938, 115.89900207519531, 63.003150939941406, 81.41963195800781, 10.484420776367188, 0.7354907989501953, 86.6817626953125, 89.76785278320312, -25.047119140625, 114.67890930175781, -52.08222961425781, -84.23039245605469, 112.31938171386719, 107.2620849609375, 44.83038330078125, 19.16015625, -98.16677856445312, 175.16851806640625, 97.50106811523438, 131.92930603027344, 115.03822326660156, 103.6800308227539, 152.14585876464844, 2.54730224609375, 13.852828979492188, 31.233139038085938, 38.24957275390625, 90.19329833984375, 86.2039794921875, 68.6387939453125, 140.8270263671875, 136.46722412109375, 134.1199951171875, -6.3148193359375, 124.64036560058594, 0.51031494140625, 33.47593688964844, 28.864212036132812, -58.138145446777344, 101.28884887695312, -88.55921936035156, 145.39254760742188, -0.50665283203125, 81.94894409179688, 134.58824157714844, 109.26741027832031, 0.0, 2.155364990234375, -69.24445343017578, 23.6260986328125, 104.42703247070312, -109.13832092285156, 126.13909912109375, 100.11370849609375, 80.71089172363281, 26.612884521484375, -58.60125732421875, 39.10789489746094, 31.14820098876953, -7.5293731689453125, 10.152008056640625, -16.4847412109375, 50.853759765625, 118.75372314453125, 113.09732055664062, 70.66976928710938, -115.81240844726562, 112.12274169921875, 1.281015396118164, 29.36956787109375, 194.28399658203125, -3.6664352416992188, 30.7681884765625, 26.783554077148438, 116.1881103515625, -85.03643798828125, 19.45257568359375, 3.3170013427734375, 150.98745727539062, 20.97857666015625, -0.3684062957763672, 2.5743484497070312, 0.0, 93.78726196289062, 66.6961441040039, 193.76943969726562, 11.652381896972656, 5.362018585205078, 55.86036682128906, 111.86910247802734, 65.254150390625, 40.492156982421875, -8.611053466796875, 134.32962036132812, -70.24664306640625, -28.98975372314453], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000283.npy"}
{"epoch": 0.5926701570680628, "step": 284, "batch_size": 128, "mean": 45.50518035888672, "std": 81.31242370605469, "min": -144.66082763671875, "p10": -75.17524719238281, "median": 36.615135192871094, "p90": 146.97474060058593, "max": 252.93295288085938, "pos_frac": 0.7421875, "sample": [-101.53001403808594, 83.70968627929688, -83.8437271118164, -144.66082763671875, 113.39875793457031, 34.022674560546875, 144.71578979492188, 141.52850341796875, -60.17631530761719, 36.57548522949219, 102.81868743896484, 146.50177001953125, 37.67162322998047, 65.94532775878906, 26.8720703125, 101.32456970214844, 130.89352416992188, 63.72149658203125, 152.49932861328125, 7.320220947265625, 126.90289306640625, 211.32098388671875, -18.9720458984375, 136.04293823242188, -6.897375106811523, 15.123855590820312, 25.45880126953125, -74.7977294921875, 15.280731201171875, -42.311798095703125, -80.25338745117188, 39.610107421875, 97.89849853515625, -22.02240753173828, 5.1363525390625, 37.6414794921875, 90.88223266601562, 95.84530639648438, 77.37052917480469, 12.735687255859375, -118.6314926147461, 11.300079345703125, 42.77751159667969, 35.489349365234375, -7.873710632324219, -15.182769775390625, -23.4307861328125, 45.0849609375, 13.94012451171875, 188.41522216796875, 136.91317749023438, -101.96788024902344, 35.86187744140625, 19.333282470703125, 159.39263916015625, 94.04116821289062, 13.501708984375, -9.115852355957031, 148.07833862304688, 63.707977294921875, 111.83367919921875, 159.8898468017578, 49.16943359375, 2.5898818969726562, 137.1000213623047, 50.12530517578125, 168.7330322265625, 26.024169921875, 91.82298278808594, 11.076828002929688, 81.65193176269531, 32.67828369140625, -26.14775276184082, 51.137542724609375, 2.7061920166015625, 252.93295288085938, 9.55743408203125, 16.733627319335938, 8.865066528320312, -7.2466888427734375, 172.7975616455078, 131.8218994140625, 100.3488998413086, 150.06207275390625, -77.74957275390625, -134.1568603515625, 13.688837051391602, -19.75091552734375, 51.85450744628906, -16.92450714111328, 107.19561767578125, 19.31927490234375, 79.52276611328125, 120.16926574707031, 1.5000724792480469, -1.912109375, 5.99969482421875, -110.08302307128906, -3.480133056640625, 113.79293823242188, 143.32455444335938, 36.65478515625, 106.68400573730469, 13.0565185546875, 0.32694244384765625, 133.92578125, -22.810409545898438, 84.12930297851562, 24.0162353515625, 95.59736633300781, 132.9083251953125, 194.40359497070312, 104.16229248046875, -21.6077880859375, -134.191650390625, -104.605712890625, 104.54920196533203, -76.05612182617188, 1.6456432342529297, -41.75732421875, 172.85968017578125, 51.57240295410156, 108.57281494140625, 190.9793701171875, 113.38529968261719, -3.691211700439453, -89.61541748046875, 98.05557250976562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000284.npy"}
{"epoch": 0.5947643979057592, "step": 285, "batch_size": 128, "mean": 34.506629943847656, "std": 70.8833999633789, "min": -133.36563110351562, "p10": -43.95524597167969, "median": 26.9217529296875, "p90": 130.21058502197266, "max": 179.2769775390625, "pos_frac": 0.703125, "sample": [15.294769287109375, 87.21096801757812, 115.86590576171875, 31.91571044921875, 117.03909301757812, 131.41854858398438, 114.5670166015625, -120.49536895751953, 115.82369995117188, 29.782974243164062, 29.439605712890625, -15.133270263671875, 10.073738098144531, -90.91677856445312, -32.30999755859375, 35.45952606201172, 17.257965087890625, -43.87884521484375, -23.0037841796875, -15.044342041015625, -7.858367919921875, 114.19061279296875, -101.10968780517578, 28.861373901367188, 19.355003356933594, 26.218292236328125, 151.50396728515625, 60.38916015625, 101.27156066894531, 36.344329833984375, 179.2769775390625, -88.40850830078125, 9.234542846679688, 42.123443603515625, 131.64752197265625, 132.49610900878906, 120.6876220703125, -19.767791748046875, -28.711196899414062, 17.364593505859375, -44.133514404296875, -31.38226318359375, -18.902313232421875, -7.0501708984375, 12.204690933227539, 106.9654541015625, 56.06951904296875, 101.02810668945312, 8.950248718261719, 8.605514526367188, 129.69288635253906, 2.3858070373535156, 9.662765502929688, -16.718727111816406, 145.45584106445312, 1.5313377380371094, -6.77777099609375, -6.307146072387695, 118.11378479003906, 15.097007751464844, 46.498870849609375, 77.91571044921875, 126.96656799316406, 35.98637390136719, 48.98736572265625, 9.3883056640625, -106.58609008789062, 19.191326141357422, -104.79060363769531, 60.269775390625, -132.033935546875, 132.94207763671875, 29.95361328125, 1.1718063354492188, 98.62007141113281, -43.062103271484375, 68.91455078125, 31.76544952392578, 27.625213623046875, 15.73065185546875, 94.6753921508789, 115.16671752929688, 10.61236572265625, 1.7441253662109375, -14.349624633789062, 4.9368896484375, 32.128662109375, 127.87689208984375, 143.62322998046875, 110.85841369628906, 97.67514038085938, 118.7381591796875, -2.3507843017578125, -28.04345703125, -23.75872802734375, 4.0166778564453125, 37.0584716796875, 15.77783203125, -61.81919860839844, 19.45196533203125, 95.07295227050781, -50.25946044921875, 94.45536804199219, -35.52227783203125, -30.7330322265625, -26.575302124023438, 48.4541015625, -10.641963958740234, 138.53390502929688, 117.39913940429688, -133.36563110351562, 4.88671875, 30.859161376953125, -3.780670166015625, -6.889152526855469, -122.51870727539062, 124.26419067382812, 4.783042907714844, 123.01593017578125, 66.0594482421875, 63.049560546875, 41.64508056640625, 100.83441162109375, 134.34487915039062, -73.98541259765625, 152.609619140625, 156.77328491210938, 144.63784790039062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000285.npy"}
{"epoch": 0.5968586387434555, "step": 286, "batch_size": 128, "mean": 38.37523651123047, "std": 74.50286102294922, "min": -136.97711181640625, "p10": -57.74312973022461, "median": 28.905433654785156, "p90": 135.38223266601562, "max": 192.85784912109375, "pos_frac": 0.6953125, "sample": [-18.540740966796875, -24.62310791015625, 111.68313598632812, 13.997360229492188, 22.23462677001953, 26.9290771484375, 9.478485107421875, 8.523933410644531, 8.09619140625, 87.62654113769531, 32.74030303955078, 17.5223388671875, -14.723419189453125, 72.63623046875, -122.01852416992188, 153.23948669433594, 11.163070678710938, 150.2181396484375, 110.35255432128906, 134.9359130859375, -46.24267578125, 139.74569702148438, 14.8814697265625, 30.525619506835938, 17.808616638183594, 23.6669921875, -20.395904541015625, -6.3417205810546875, 105.3519287109375, -2.3627166748046875, -136.97711181640625, 1.7840728759765625, 127.50946807861328, 118.51905822753906, -57.74908447265625, 11.55327033996582, 37.59326171875, 93.20223999023438, 9.589469909667969, 71.56578063964844, -7.990203857421875, 130.59796142578125, 111.36871337890625, -94.23365020751953, 127.60198974609375, -48.523193359375, 37.6949462890625, 192.85784912109375, -20.016357421875, -7.10369873046875, 129.01467895507812, -132.68182373046875, 130.28790283203125, 97.64988708496094, 38.5001220703125, 147.17913818359375, 3.804931640625, 89.10577392578125, -99.67652130126953, 155.15817260742188, -10.374763488769531, 0.0, -5.4267578125, -98.8726806640625, 85.60247802734375, 51.74690246582031, -4.997318267822266, -38.082275390625, -3.5821075439453125, 55.40150451660156, -52.1868896484375, 114.67156982421875, 116.61134338378906, 47.1649169921875, -61.5360221862793, 53.297607421875, 125.78668212890625, 79.11102294921875, -84.72688293457031, 68.61798095703125, 80.35845947265625, 77.23529052734375, 18.993896484375, 189.8094482421875, 126.7457275390625, 36.56146240234375, 24.7689208984375, -31.923263549804688, 107.01707458496094, -106.68597412109375, 36.60741424560547, 156.24966430664062, 27.91619873046875, 50.42041015625, 136.42364501953125, 38.90234375, 73.49679565429688, 103.82948303222656, 100.93193817138672, 92.74974060058594, 159.7435302734375, 32.179351806640625, 141.39056396484375, 29.894668579101562, 64.13212585449219, -43.771728515625, 153.2235870361328, 27.259666442871094, -67.55941772460938, -17.36465072631836, -23.03558349609375, 114.5570068359375, 3.8632125854492188, -57.740577697753906, -99.48998260498047, 179.84674072265625, -95.78421020507812, -7.9271240234375, 20.89881134033203, 12.606758117675781, 22.286514282226562, 122.44485473632812, -3.1793289184570312, 16.890182495117188, 52.84930419921875, -24.57635498046875, 101.13041687011719, 13.331413269042969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000286.npy"}
{"epoch": 0.5989528795811518, "step": 287, "batch_size": 128, "mean": 42.005619049072266, "std": 70.39324188232422, "min": -165.11126708984375, "p10": -33.64141693115234, "median": 42.07514953613281, "p90": 126.6053009033203, "max": 167.93988037109375, "pos_frac": 0.7109375, "sample": [91.93040466308594, -25.892745971679688, -106.85919189453125, -10.71957015991211, -22.659835815429688, -20.145172119140625, 130.79226684570312, 118.20266723632812, 75.34638214111328, 148.37579345703125, 114.49954223632812, 17.093734741210938, 26.161579132080078, 124.09330749511719, 12.9217529296875, 15.543609619140625, -25.739120483398438, -73.42926025390625, -31.919418334960938, 93.45684051513672, -32.30381774902344, 167.93988037109375, 113.383056640625, -36.762481689453125, 38.771209716796875, -21.727874755859375, -13.043048858642578, 12.2054443359375, 118.2818603515625, 40.228363037109375, 12.009719848632812, 128.38168334960938, 106.3815689086914, 120.25762939453125, -5.502349853515625, 3.1489410400390625, 69.16282653808594, 103.40887451171875, -1.4428539276123047, 115.58389282226562, 95.96468353271484, 116.49041748046875, 19.031829833984375, 92.9713134765625, 133.9956512451172, 2.1348743438720703, -5.580780029296875, 125.843994140625, 116.2291259765625, 154.96688842773438, 108.57356262207031, 89.04052734375, 123.56118774414062, 102.48333740234375, 27.40777587890625, 43.87408447265625, 54.024986267089844, -69.54119873046875, 43.1982421875, -24.6324462890625, 114.95643615722656, 81.033935546875, 44.0780029296875, 35.1740837097168, 3.9110755920410156, 32.92791748046875, 60.938446044921875, 105.3538818359375, 55.52392578125, 75.25650024414062, -126.37493896484375, -10.856178283691406, 111.0582275390625, 133.848876953125, 48.75616455078125, 131.36599731445312, 119.5277099609375, 98.78530883789062, 15.736572265625, 98.34950256347656, 164.983642578125, -62.9217529296875, -105.03068542480469, 18.302154541015625, -49.313201904296875, 83.1882553100586, -29.914352416992188, 59.80078125, -28.0904541015625, 10.321365356445312, 74.59181213378906, 148.4807891845703, -165.11126708984375, -12.53790283203125, 84.07691192626953, -10.1290283203125, 69.48477172851562, 5.025718688964844, -117.22149658203125, 40.952056884765625, 3.1190185546875, 15.10076904296875, 10.458946228027344, 83.79763793945312, -28.601593017578125, -51.47700500488281, 111.89635467529297, -128.40591430664062, 150.3541259765625, -12.716651916503906, 92.46633911132812, -11.992965698242188, 118.24658203125, 43.47661590576172, 143.9767608642578, 8.361602783203125, 16.617034912109375, 83.03109741210938, 86.14883422851562, -10.72393798828125, 55.771728515625, 7.250919342041016, -24.5174560546875, -45.373748779296875, 16.47998046875, 131.3365478515625, -14.84423828125, 79.73828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000287.npy"}
{"epoch": 0.6010471204188481, "step": 288, "batch_size": 128, "mean": 40.931922912597656, "std": 74.0031509399414, "min": -131.84017944335938, "p10": -39.23995056152342, "median": 26.7718505859375, "p90": 134.86403198242186, "max": 204.9755859375, "pos_frac": 0.7109375, "sample": [76.634033203125, 3.8892364501953125, -24.463226318359375, -48.191162109375, 92.029541015625, -6.1578369140625, -21.77313995361328, -16.850677490234375, 45.721282958984375, 98.9105224609375, 19.05853271484375, 159.8170166015625, -0.115234375, 74.45791625976562, 62.233123779296875, -5.5664520263671875, -4.771385192871094, 47.60786437988281, 24.40460205078125, 0.0, 18.909576416015625, -11.3275146484375, 46.88201904296875, 8.87137222290039, 114.113037109375, -24.17584228515625, -30.118927001953125, 5.89556884765625, -34.8125, 18.150634765625, 160.14382934570312, 113.70458984375, 58.01702880859375, 88.46151733398438, 7.3171539306640625, 0.0, 2.158905029296875, 78.8191146850586, 10.21185302734375, 22.987152099609375, -6.1866912841796875, 79.84782409667969, -60.38848876953125, -90.79124450683594, 25.770999908447266, 115.23016357421875, 158.24533081054688, 150.47525024414062, -131.84017944335938, 2.4833831787109375, 125.87310791015625, -6.786048889160156, 138.15472412109375, 14.490684509277344, 6.4493408203125, -125.98397827148438, -14.91988754272461, 44.05706787109375, 121.49134826660156, -114.32723999023438, 200.31634521484375, -68.17689514160156, -20.193695068359375, 94.23529052734375, 173.3560791015625, 17.259185791015625, -17.350906372070312, -49.07427978515625, 125.09901428222656, 5.543586730957031, 124.3531494140625, 30.926544189453125, -1.9066009521484375, 128.27764892578125, 11.454193115234375, 20.238555908203125, 20.62939453125, 91.88035583496094, -35.403717041015625, 38.157676696777344, 96.6756591796875, 34.57225036621094, -20.094955444335938, -105.4600830078125, -24.645225524902344, -111.24893188476562, 12.879047393798828, 150.95870971679688, 133.31478881835938, 60.18963623046875, 38.744354248046875, 1.241668701171875, 93.00848388671875, -4.314760208129883, 143.29794311523438, 117.11241149902344, 98.9698715209961, -114.2275390625, 133.4537353515625, 25.82562255859375, 68.94232177734375, 82.10873413085938, -110.10528564453125, 172.22210693359375, 13.249580383300781, 47.191680908203125, 46.77587127685547, 40.96148681640625, 51.99908447265625, -15.2705078125, -67.39364624023438, 88.12724304199219, 29.242965698242188, 131.20123291015625, 101.1011962890625, 119.39715576171875, 27.71807861328125, 14.27960205078125, 11.87506103515625, 121.01515197753906, 75.70719909667969, 201.36105346679688, 40.792022705078125, 166.7957763671875, 131.5943603515625, 23.174468994140625, 204.9755859375, 107.94413757324219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000288.npy"}
{"epoch": 0.6031413612565445, "step": 289, "batch_size": 128, "mean": 28.795425415039062, "std": 72.87203216552734, "min": -145.87814331054688, "p10": -54.51197280883788, "median": 16.137451171875, "p90": 128.39861221313475, "max": 166.72930908203125, "pos_frac": 0.625, "sample": [-49.845458984375, -20.222015380859375, 15.686767578125, -78.1462173461914, -4.75640869140625, 2.431854248046875, 10.124114990234375, -19.63348388671875, 77.739501953125, -61.2745361328125, 123.1051025390625, -3.6209259033203125, 1.1401805877685547, 45.077880859375, -1.1071319580078125, 47.306793212890625, -29.0655517578125, 139.947509765625, 17.3409423828125, -50.381378173828125, 164.989501953125, 140.89405822753906, 146.99911499023438, 37.815147399902344, -6.2490234375, -20.928604125976562, -11.11932373046875, 4.2161865234375, 9.8720703125, 79.65022277832031, -3.2557525634765625, 122.80731201171875, -2.822174072265625, 18.65106201171875, 0.7235183715820312, 0.0, 7.513225555419922, 82.35577392578125, -28.455291748046875, 15.29547119140625, 18.1746826171875, -23.5931396484375, -105.68356323242188, -125.14312744140625, 76.078369140625, -30.2989501953125, 53.547607421875, 115.8292236328125, -6.662078857421875, -3.704559326171875, -7.9356231689453125, 122.29348754882812, -132.49923706054688, -133.10516357421875, 12.142822265625, 4.249053955078125, -19.841463088989258, -7.686286926269531, 97.1004867553711, 155.2989501953125, -78.4141845703125, 70.36651611328125, 79.13108825683594, 123.29296875, 36.75762939453125, 116.76227569580078, -51.613731384277344, -115.35617065429688, -83.61458587646484, -0.6270904541015625, 121.84405517578125, -39.09068298339844, 1.240509033203125, 105.98719787597656, 16.588134765625, 17.818771362304688, 63.448333740234375, 42.91400146484375, 88.27728271484375, 5.316009521484375, 68.06979370117188, -21.494842529296875, -145.87814331054688, 136.58770751953125, -18.393203735351562, 84.54570770263672, -10.690704345703125, 125.95567321777344, -40.73524475097656, 135.28390502929688, -6.78289794921875, 153.71942138671875, 119.58355712890625, 10.2178955078125, 34.82037353515625, 92.37066650390625, 107.3177490234375, 9.04449462890625, 92.86239624023438, 71.47525024414062, 132.43035888671875, 34.36681365966797, 149.66998291015625, 101.87599182128906, -21.392913818359375, 120.59483337402344, 28.9249267578125, 55.61610412597656, 59.0440673828125, 166.72930908203125, -102.772216796875, 99.2785873413086, 126.27516174316406, -33.115142822265625, 129.97618103027344, -62.26544189453125, -116.96853637695312, 30.120079040527344, -47.909019470214844, 127.7225112915039, 29.085662841796875, 29.829986572265625, 132.0474853515625, 5.0258636474609375, 17.425994873046875, -6.584796905517578, -22.8035888671875, 25.316650390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000289.npy"}
{"epoch": 0.6052356020942409, "step": 290, "batch_size": 128, "mean": 34.93117141723633, "std": 71.08599853515625, "min": -146.62374877929688, "p10": -48.57986450195312, "median": 20.308990478515625, "p90": 131.17540893554687, "max": 194.308349609375, "pos_frac": 0.6484375, "sample": [140.67515563964844, 20.54412841796875, -8.14508056640625, 93.86396789550781, 76.35482788085938, -146.62374877929688, 2.3995513916015625, 99.74053955078125, 0.511474609375, -22.1195068359375, 161.58624267578125, 93.32388305664062, 47.52630615234375, -12.279655456542969, 121.29026794433594, 92.98191833496094, 7.260841369628906, 35.5557861328125, 139.01123046875, 119.93807983398438, 119.96038818359375, -15.441596984863281, 0.8333740234375, -3.5195159912109375, 177.57968139648438, -3.4947509765625, -90.80509948730469, -97.64154052734375, -19.934783935546875, 1.13214111328125, 88.82080078125, 3.8584632873535156, -3.143280029296875, 64.974609375, 114.83013916015625, 153.21905517578125, 139.3192596435547, 109.62173461914062, 6.37030029296875, -120.17596435546875, 105.00914001464844, 69.58531188964844, -1.00811767578125, 112.32177734375, 76.03125, -0.325775146484375, 0.317840576171875, -66.1478271484375, 64.25141906738281, 122.09039306640625, -15.858062744140625, 14.637008666992188, 93.66326904296875, 18.456281661987305, 12.842140197753906, -27.825225830078125, 35.79212951660156, 105.21954345703125, -2.454803466796875, 20.724029541015625, 12.534027099609375, 14.109920501708984, -20.218109130859375, 104.011962890625, 18.45452880859375, 30.535980224609375, 27.53240966796875, 73.8740234375, 34.81866455078125, 93.7908935546875, -28.786285400390625, 49.31074523925781, -1.202972412109375, -2.4206771850585938, -18.500167846679688, 161.06878662109375, -16.68639373779297, -47.43719482421875, -0.5345458984375, 23.01007080078125, -2.494964599609375, -13.001327514648438, -70.12449645996094, 126.12420654296875, 120.49856567382812, 125.0279541015625, 133.7684326171875, -21.757171630859375, -137.88735961914062, 6.8254241943359375, 57.83213806152344, -29.799331665039062, 105.74917602539062, -26.102890014648438, -83.74191284179688, 6.1570587158203125, -69.0838623046875, -11.286514282226562, -70.19532775878906, 107.74627685546875, 10.026901245117188, -78.83199310302734, 43.5765380859375, 194.308349609375, 72.376953125, 41.7113037109375, 130.755859375, 135.4635009765625, 132.15435791015625, 38.26000213623047, -93.41902160644531, 20.0738525390625, 106.50053405761719, 44.83709716796875, 72.59124755859375, -51.24609375, -12.327568054199219, 6.705902099609375, -32.956695556640625, 104.12918090820312, -14.2315673828125, -5.1537628173828125, 22.7548828125, 97.34808349609375, -40.62571716308594, 50.2957763671875, 148.63156127929688, 138.87945556640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000290.npy"}
{"epoch": 0.6073298429319371, "step": 291, "batch_size": 128, "mean": 52.270042419433594, "std": 75.0965805053711, "min": -164.32791137695312, "p10": -50.061439514160156, "median": 42.45697784423828, "p90": 145.31977233886718, "max": 182.42987060546875, "pos_frac": 0.7265625, "sample": [150.5882568359375, -3.0434608459472656, 174.16598510742188, 16.381683349609375, 170.5462646484375, 80.95623779296875, 137.8028564453125, 123.54975891113281, 87.60540771484375, -74.66665649414062, -2.923614501953125, 104.73709106445312, 25.02245330810547, 2.020416259765625, 147.8668212890625, 12.9696044921875, 146.81430053710938, -15.78216552734375, 111.06240844726562, 27.53584861755371, 99.94488525390625, 5.88751220703125, 91.98764038085938, 15.40850830078125, -10.209197998046875, 131.70559692382812, 43.24671936035156, 7.5562286376953125, 106.98049926757812, -27.921234130859375, -132.2098846435547, -31.951496124267578, -56.45848083496094, 32.068607330322266, -164.32791137695312, 36.707275390625, 106.977783203125, 88.9686279296875, 55.21099853515625, 6.0299072265625, 144.67926025390625, -49.35948181152344, -89.21064758300781, 76.96009826660156, 140.5983123779297, 33.449310302734375, -5.711891174316406, 83.38993835449219, 114.80908203125, 52.96250915527344, 114.36724853515625, 83.57685089111328, 92.76082611083984, -16.96527099609375, 181.13934326171875, 107.99711608886719, 81.11932373046875, -35.74200439453125, -59.89860534667969, 114.69540405273438, 152.01751708984375, 11.31768798828125, -9.937835693359375, -4.76513671875, 137.41680908203125, -59.097076416015625, 22.876548767089844, 41.667236328125, -74.85621643066406, 142.52133178710938, -90.43336486816406, 14.988525390625, 28.792705535888672, 182.42987060546875, 102.49198913574219, 23.64813232421875, 104.40418243408203, 123.8466796875, 140.92144775390625, -4.983673095703125, 74.317138671875, 2.4258651733398438, -10.11346435546875, -23.6591796875, 32.30259704589844, 150.48809814453125, 91.89314270019531, -71.86581420898438, 0.0, -19.32269287109375, -51.6993408203125, 104.18829345703125, 28.914230346679688, -70.97286224365234, 181.32711791992188, 113.45487976074219, 140.628173828125, 89.81536865234375, 149.36849975585938, -26.65924072265625, 71.87759399414062, 122.41935729980469, 30.53472900390625, 113.26815795898438, 0.54156494140625, 64.0938720703125, 16.12969970703125, 34.232421875, 175.33184814453125, 27.4344482421875, -79.09707641601562, 108.5599365234375, -6.716377258300781, 76.88203430175781, 122.64057922363281, 13.89312744140625, 122.0111083984375, -0.990997314453125, 152.3619384765625, 135.06561279296875, 90.44784545898438, 21.630935668945312, -4.40496826171875, -19.261472702026367, 140.4260711669922, 144.35995483398438, 110.83863830566406, 41.52815246582031], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000291.npy"}
{"epoch": 0.6094240837696335, "step": 292, "batch_size": 128, "mean": 43.29698944091797, "std": 69.56321716308594, "min": -175.9437255859375, "p10": -36.92363891601562, "median": 36.65382385253906, "p90": 129.14954986572263, "max": 208.0966796875, "pos_frac": 0.734375, "sample": [25.4345703125, 3.818359375, 112.66936492919922, -63.36492919921875, 98.40276336669922, 113.24712371826172, 127.91700744628906, 187.04046630859375, 27.12158203125, 125.20254516601562, 191.49383544921875, 74.68759155273438, -28.89324951171875, -2.3245849609375, 54.488433837890625, 26.08837890625, -31.578954696655273, -8.447080612182617, 30.056610107421875, -4.8526611328125, 47.42918395996094, 39.12847137451172, 30.4786376953125, 66.21728515625, 43.90184783935547, 103.17709350585938, 104.26011657714844, 29.424102783203125, 20.057891845703125, -5.424930572509766, 119.72515869140625, 106.26126098632812, 1.3223915100097656, 39.37165832519531, 103.33355712890625, 28.154438018798828, 51.58502197265625, 6.50958251953125, -171.81170654296875, 22.689720153808594, 55.44111633300781, 52.530250549316406, 122.681396484375, 19.513328552246094, -4.6875, 7.759490966796875, 91.1058578491211, -92.7869873046875, -2.2684402465820312, 141.00897216796875, -51.50006103515625, 150.9176025390625, -57.90966796875, -16.434539794921875, -5.114288330078125, 85.88458251953125, -0.3133544921875, -53.9091796875, 75.9169921875, 13.33349609375, 110.55023193359375, 10.6029052734375, 124.32337951660156, -52.76481628417969, 0.046142578125, 13.066337585449219, 12.290618896484375, 21.508880615234375, 208.0966796875, 53.033203125, 4.98870849609375, 36.344879150390625, 91.0384521484375, 37.211212158203125, 113.06430053710938, 166.78863525390625, 33.23152542114258, 136.5096893310547, 27.898120880126953, -13.435430526733398, -53.926300048828125, -175.9437255859375, 165.97738647460938, 46.57234191894531, 132.02548217773438, 96.30325317382812, 58.35699462890625, -6.80657958984375, 7.515144348144531, 134.5389404296875, -35.343109130859375, -1.2988662719726562, -76.37690734863281, 15.884033203125, 60.035438537597656, -53.59632873535156, -8.042884826660156, -24.865982055664062, 20.91166877746582, 144.82049560546875, 93.06327819824219, 109.16954803466797, 20.153640747070312, 62.80934143066406, 26.601112365722656, 82.97674560546875, 36.9627685546875, -0.7574615478515625, -20.220809936523438, 38.130279541015625, 9.231414794921875, 45.422271728515625, 53.65185546875, -0.252349853515625, 127.79605102539062, 121.97027587890625, 179.97198486328125, -40.611541748046875, 83.28329467773438, 85.34829711914062, 144.47430419921875, -110.89959716796875, -5.784843444824219, 107.52651977539062, 41.93812561035156, 113.6861572265625, 110.09356689453125, 69.97874450683594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000292.npy"}
{"epoch": 0.6115183246073298, "step": 293, "batch_size": 128, "mean": 46.150718688964844, "std": 71.50701904296875, "min": -128.8282470703125, "p10": -31.53078460693359, "median": 35.225791931152344, "p90": 143.00631866455078, "max": 191.97265625, "pos_frac": 0.7421875, "sample": [140.0963592529297, -11.627288818359375, 128.21414184570312, 156.3038330078125, 37.67193603515625, -96.8695068359375, 16.611236572265625, 100.097412109375, 3.5768508911132812, 75.47086334228516, -5.7111968994140625, 72.58313751220703, 36.8671875, 29.169754028320312, 84.50776672363281, 13.814666748046875, -56.56537628173828, -1.61578369140625, -79.74606323242188, 16.970123291015625, 21.67150115966797, -0.380218505859375, 150.74224853515625, 117.0926513671875, 18.455230712890625, 0.0, 137.98760986328125, 37.628387451171875, 149.62127685546875, 28.517318725585938, 116.77667236328125, -29.65020751953125, -9.953689575195312, 2.98760986328125, 16.7864990234375, 11.493759155273438, 30.837646484375, 117.78883361816406, -0.742095947265625, 51.61865234375, 137.35728454589844, 0.6467151641845703, 152.08383178710938, 127.44107055664062, -26.9014892578125, -127.63534545898438, 138.40423583984375, 134.37088012695312, 176.1241455078125, 83.20765686035156, 67.19236755371094, 41.57061767578125, 112.702880859375, 125.53167724609375, 0.051055908203125, 112.60877990722656, 47.265655517578125, -14.94732666015625, 5.0758514404296875, 19.864990234375, -4.336662292480469, 142.70127868652344, 88.10269165039062, -74.53053283691406, -120.23868560791016, 59.20744323730469, 13.851299285888672, -33.837860107421875, 104.30215454101562, -11.150108337402344, -128.8282470703125, -35.306915283203125, 106.02131652832031, 191.97265625, -50.081451416015625, 11.1759033203125, -9.78875732421875, 113.38082885742188, -30.542037963867188, 38.679595947265625, 29.70623779296875, 43.156524658203125, 47.03643798828125, 16.607925415039062, 143.71807861328125, 33.31549072265625, 55.31353759765625, 125.68243408203125, 107.20645141601562, -24.840805053710938, -15.296875, 154.559814453125, 20.33495330810547, 127.48893737792969, 49.077850341796875, 29.229156494140625, 77.0086669921875, -109.78358459472656, 45.844940185546875, 40.62886047363281, 33.58439636230469, 173.71670532226562, 187.36129760742188, 165.6754913330078, 61.359619140625, 123.02650451660156, 20.358665466308594, 48.16015625, -45.20460510253906, 2.0994911193847656, -20.597503662109375, 20.150054931640625, -0.52728271484375, 75.64208984375, 187.08694458007812, 186.90179443359375, 26.81146240234375, 27.193313598632812, 116.20794677734375, -48.71022033691406, 82.78439331054688, -11.249305725097656, 8.654861450195312, 82.74862670898438, 41.725318908691406, -8.324859619140625, 39.85015869140625, 22.942657470703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000293.npy"}
{"epoch": 0.6136125654450262, "step": 294, "batch_size": 128, "mean": 39.347965240478516, "std": 73.17113494873047, "min": -129.07369995117188, "p10": -46.38988800048828, "median": 23.729618072509766, "p90": 136.25760498046876, "max": 200.98062133789062, "pos_frac": 0.7265625, "sample": [129.19375610351562, 109.12896728515625, 129.84793090820312, -2.3690185546875, -7.47857666015625, 9.291954040527344, 0.8203773498535156, 1.615325927734375, 15.041084289550781, 135.42498779296875, 45.50250244140625, 146.64572143554688, 106.37747192382812, -10.986618041992188, -3.841094970703125, 6.197687149047852, 118.84733581542969, 160.41651916503906, 131.88946533203125, 24.582015991210938, 118.8604736328125, 115.22244262695312, 105.57290649414062, -57.970237731933594, 30.754310607910156, 127.36572265625, 10.5277099609375, -1.82452392578125, 11.99859619140625, 20.370147705078125, 159.08404541015625, 27.868255615234375, 11.73394775390625, 93.35443115234375, -8.356651306152344, 19.4969482421875, -30.7314453125, -127.98724365234375, 42.240386962890625, 24.847686767578125, 76.2158203125, 157.8179931640625, -52.74359130859375, -98.1557846069336, 141.41363525390625, 55.96471405029297, 103.89867401123047, -4.279594421386719, -5.852714538574219, 38.13273620605469, 138.20037841796875, -8.412063598632812, 147.32571411132812, 132.872802734375, 71.80679321289062, 139.7510528564453, 140.4044189453125, -25.477203369140625, 17.595367431640625, 12.032081604003906, 100.90142822265625, 25.10357666015625, 31.240234375, 119.86744689941406, -119.05618286132812, 74.41417694091797, 104.60935974121094, 83.73114013671875, 12.767471313476562, 7.11846923828125, 14.870147705078125, 159.32838439941406, 146.200927734375, 0.0, 122.04114532470703, 200.98062133789062, -88.84373474121094, -7.1081085205078125, -13.020843505859375, 19.15045166015625, -6.735258102416992, 130.8489990234375, 6.660696029663086, -123.2886962890625, -17.215301513671875, 40.420928955078125, -37.78643798828125, 8.018363952636719, 88.48965454101562, -44.98872375488281, 26.8319091796875, 3.0632247924804688, 13.999427795410156, -128.3704833984375, 18.070281982421875, 98.90530395507812, 27.623077392578125, 2.2629776000976562, 5.235260009765625, -11.712081909179688, 22.877220153808594, 111.84697723388672, 13.849853515625, 43.97334289550781, 110.42545318603516, 94.42564392089844, -84.6739501953125, -115.81085205078125, 32.764404296875, -0.09369850158691406, -97.43187713623047, 43.51890563964844, 95.62454223632812, 11.261871337890625, 92.51962280273438, 80.8726806640625, 6.698272705078125, -27.146705627441406, 96.56721496582031, -9.659904479980469, 11.291748046875, -49.659271240234375, 80.26439666748047, 2.6599044799804688, 142.84432983398438, -129.07369995117188, 111.26703643798828, 132.820068359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000294.npy"}
{"epoch": 0.6157068062827226, "step": 295, "batch_size": 128, "mean": 43.528289794921875, "std": 75.8450927734375, "min": -129.27078247070312, "p10": -39.43000793457031, "median": 23.94578742980957, "p90": 139.6207733154297, "max": 220.5196533203125, "pos_frac": 0.7421875, "sample": [151.96514892578125, 18.756378173828125, -34.782501220703125, -113.16160583496094, -6.8507843017578125, 44.46095275878906, -44.73029327392578, 130.43597412109375, -104.24136352539062, -11.60601806640625, -15.81097412109375, 9.503669738769531, 136.1454315185547, 21.39990234375, -11.8082275390625, -21.399566650390625, -115.64085388183594, 102.13894653320312, -105.48077392578125, 40.610923767089844, 88.83576965332031, 175.00381469726562, -11.27117919921875, -129.27078247070312, -15.101470947265625, 107.94824981689453, 27.660110473632812, -39.20849609375, 96.73013305664062, -3.496917724609375, -8.70184326171875, 22.475204467773438, 35.63897705078125, 78.17886352539062, 106.03648376464844, 8.20843505859375, 58.22296142578125, 9.80010986328125, 67.0753173828125, 88.95187377929688, 178.3844757080078, -10.963340759277344, 120.4388427734375, 135.602783203125, 0.095184326171875, 39.02032470703125, 24.832618713378906, 164.6869659423828, 131.14462280273438, 115.67727661132812, -115.55682373046875, -74.224853515625, 137.2763214111328, 220.5196533203125, 133.0846710205078, 183.85565185546875, -25.445068359375, 99.48344421386719, 54.07318115234375, 23.058956146240234, -17.360992431640625, 11.663421630859375, 39.301971435546875, 2.0466079711914062, 116.38543701171875, 70.97466278076172, 18.11931610107422, 65.2647705078125, 120.87832641601562, 17.975509643554688, 6.1546783447265625, 9.682662963867188, 0.15267181396484375, 138.464599609375, -32.083831787109375, 93.79740905761719, 135.28036499023438, 120.82536315917969, 14.710739135742188, 95.5993423461914, 9.68752670288086, 14.818069458007812, 56.520973205566406, 67.96758270263672, 6.411712646484375, 19.050872802734375, 0.0, 62.4503288269043, 16.450721740722656, 16.072341918945312, 129.92044067382812, 165.8326416015625, 26.571502685546875, 45.30303955078125, 108.26141357421875, 0.13350296020507812, -121.404296875, 143.5018310546875, 78.53103637695312, 137.93968200683594, 110.20683288574219, -28.212547302246094, 12.591075897216797, 22.12255859375, 27.35638427734375, -97.12286376953125, 126.90991973876953, 14.859615325927734, -0.6879119873046875, 90.99900817871094, -72.62808227539062, 131.1562957763672, 188.1904296875, 144.36419677734375, 0.65179443359375, 163.66928100585938, 1.3922710418701172, 2.713714599609375, 22.291725158691406, -46.39593505859375, -39.946868896484375, 4.77716064453125, 142.31851196289062, -18.41473388671875, 120.07424926757812, 146.8582000732422, 51.860504150390625, -30.852783203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000295.npy"}
{"epoch": 0.6178010471204188, "step": 296, "batch_size": 128, "mean": 59.81519317626953, "std": 75.8868179321289, "min": -146.30560302734375, "p10": -21.30936126708984, "median": 56.9168701171875, "p90": 152.1791213989258, "max": 182.941650390625, "pos_frac": 0.796875, "sample": [115.19906616210938, 56.646331787109375, 24.7889404296875, 115.76588439941406, 116.37619018554688, -18.08538055419922, 0.9460906982421875, 0.5164394378662109, 60.94012451171875, 6.627071380615234, 109.775390625, -88.34500122070312, 44.67254638671875, 46.9141845703125, 9.43585205078125, 125.63644409179688, 133.70394897460938, 101.82464599609375, 119.01909637451172, -58.0606689453125, 128.62203979492188, 51.204620361328125, -12.7225341796875, 123.4583740234375, 59.314117431640625, 152.92861938476562, -9.498176574707031, 25.762065887451172, 140.93853759765625, -44.407806396484375, 13.05120849609375, 86.42558288574219, 165.4625244140625, 8.953655242919922, -121.57246398925781, 163.81863403320312, 49.381591796875, 127.8460693359375, 135.59226989746094, 38.121299743652344, -11.635055541992188, 22.231903076171875, 160.3497314453125, 127.13353729248047, 19.91522979736328, -29.50018310546875, 32.259002685546875, -24.065322875976562, 175.976806640625, 14.764144897460938, 158.52166748046875, 75.70222473144531, 5.293121337890625, 134.9603271484375, 152.86798095703125, 175.38211059570312, 11.784599304199219, 162.41024780273438, 57.187408447265625, 100.32839965820312, -3.11767578125, -16.18523406982422, 115.02578735351562, -81.22747802734375, 51.13604736328125, 88.11378479003906, 28.288177490234375, 143.9566650390625, -120.57206726074219, 138.1988525390625, 134.01126098632812, 7.348655700683594, -115.2945556640625, 114.47628021240234, 79.22802734375, 41.87994384765625, 164.5809326171875, 179.73223876953125, -3.1071434020996094, 113.651611328125, 47.99810791015625, 115.14691162109375, 62.474159240722656, -146.30560302734375, 148.58798217773438, 107.48713684082031, -1.774810791015625, -83.75552368164062, 151.273681640625, -6.6444091796875, 116.47271728515625, 5.662994384765625, 103.11984252929688, -42.549591064453125, 38.34197998046875, 0.0, 120.763671875, 122.90399169921875, 125.6463623046875, 112.0799560546875, 21.253021240234375, 138.15579223632812, 59.581329345703125, 124.93733215332031, 19.521713256835938, 40.346893310546875, 15.645751953125, 13.1708984375, 41.13262939453125, 123.41482543945312, 166.50357055664062, 86.57125854492188, 13.360055923461914, -17.271636962890625, 0.08351898193359375, 132.22515869140625, 124.23492431640625, 2.6445770263671875, -20.12823486328125, 151.88389587402344, 24.6396484375, 93.42562866210938, 20.72381591796875, 149.8447265625, 97.23037719726562, 182.941650390625, -128.333251953125, -7.2943267822265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000296.npy"}
{"epoch": 0.6198952879581152, "step": 297, "batch_size": 128, "mean": 47.71051788330078, "std": 72.83812713623047, "min": -142.51632690429688, "p10": -27.052622985839843, "median": 31.882732391357422, "p90": 144.96859130859374, "max": 177.15679931640625, "pos_frac": 0.6875, "sample": [170.4208984375, 14.428176879882812, 127.68661499023438, 145.3564453125, -34.572601318359375, 4.589572906494141, 129.17800903320312, 53.2894287109375, 144.41351318359375, -22.800323486328125, -3.117462158203125, -3.824462890625, -28.645492553710938, 5.9186859130859375, 153.89120483398438, -82.21804809570312, -9.563608169555664, 32.979278564453125, 118.36863708496094, 95.49132537841797, 11.159706115722656, 33.931610107421875, -2.4023971557617188, 46.8133544921875, -26.369964599609375, -44.754119873046875, -22.577117919921875, -142.51632690429688, 22.0045166015625, -15.507678985595703, -66.84517669677734, 99.92414855957031, 27.852066040039062, 83.68315887451172, 2.0488739013671875, 4.83575439453125, 139.46810913085938, 1.5789794921875, 177.15679931640625, 20.673004150390625, -12.309593200683594, 32.959373474121094, 74.28689575195312, -60.712860107421875, 3.83056640625, -17.734588623046875, -3.048553466796875, 30.80609130859375, -110.1968994140625, 134.86773681640625, 44.48193359375, -22.478668212890625, -0.815704345703125, -35.98260498046875, 0.54339599609375, 176.9830322265625, 131.237548828125, 107.58399963378906, -3.9648876190185547, -24.1160888671875, 103.64962768554688, 129.69863891601562, -119.3902587890625, 26.845703125, 160.69845581054688, 61.56500244140625, 16.41748046875, -24.049835205078125, 157.23489379882812, -24.13323974609375, 29.828399658203125, 105.88128662109375, 155.44418334960938, 147.9822540283203, 166.61163330078125, -58.972412109375, 43.28567886352539, 5.943122863769531, 104.10791015625, -6.8729400634765625, -80.5928955078125, 127.42948913574219, 76.63186645507812, -7.694007873535156, 1.7261486053466797, 29.257705688476562, 35.51341247558594, -8.486030578613281, 140.23899841308594, 7.019927978515625, 118.0703125, -16.27093505859375, 70.15435791015625, 118.17814636230469, 104.87506103515625, 96.69120788574219, 7.173797607421875, 139.95274353027344, -4.1507568359375, 160.7879638671875, 107.74380493164062, 123.39639282226562, 77.35568237304688, 134.53524780273438, -8.432075500488281, 101.62728881835938, 91.20388793945312, 12.41510009765625, 100.61466217041016, 124.92181396484375, 162.6295166015625, 122.880615234375, 9.21319580078125, 85.52816772460938, -2.45306396484375, -26.349716186523438, -17.6444091796875, 61.4268798828125, 121.59187316894531, 35.71307373046875, 20.017791748046875, 126.05621337890625, 160.12094116210938, -45.282012939453125, -4.728729248046875, 65.24993896484375, 116.86248016357422, 144.8023681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000297.npy"}
{"epoch": 0.6219895287958115, "step": 298, "batch_size": 128, "mean": 52.739112854003906, "std": 80.56024932861328, "min": -159.40118408203125, "p10": -38.99797973632812, "median": 36.231689453125, "p90": 153.0506561279297, "max": 289.2869873046875, "pos_frac": 0.734375, "sample": [196.47604370117188, 132.85858154296875, -20.223876953125, 29.364913940429688, -42.23223876953125, -42.9638671875, 5.85919189453125, -16.558135986328125, -23.08990478515625, 154.56399536132812, 88.66226196289062, -11.780624389648438, 8.0146484375, 17.650970458984375, 116.71377563476562, -91.28866577148438, 143.75881958007812, 88.75372314453125, 66.91060638427734, 103.32169342041016, 139.07577514648438, 131.36166381835938, 38.623382568359375, 194.41937255859375, 3.8773651123046875, -34.38262939453125, 126.6815185546875, -104.88397216796875, 127.95640563964844, 71.21701049804688, 138.95245361328125, -38.2999267578125, 122.90745544433594, 48.57472229003906, 32.84210205078125, -57.437774658203125, 8.010215759277344, 75.06906127929688, 135.06842041015625, -64.81465148925781, 119.15469360351562, -3.3856201171875, 122.72982788085938, -19.669036865234375, 42.55096435546875, 160.26077270507812, 39.38458251953125, 26.053451538085938, -122.31721496582031, -22.341659545898438, 32.49473571777344, 130.61724853515625, 19.64105224609375, 143.1153106689453, -74.40301513671875, 58.69508361816406, 124.29789733886719, 153.65036010742188, 12.4627685546875, 120.02102661132812, 180.91363525390625, -159.40118408203125, 128.24560546875, -14.146453857421875, -12.6678466796875, 187.36074829101562, 152.79364013671875, 44.2376708984375, -80.447265625, 0.0, 6.914482116699219, 8.967613220214844, 175.649169921875, 71.2938232421875, 25.038047790527344, -4.514598846435547, 119.025146484375, 40.02301025390625, 5.429191589355469, 128.85543823242188, -40.62677001953125, -24.250152587890625, 14.52273941040039, 71.5550537109375, 123.44107818603516, 6.244953155517578, 143.1864471435547, 133.92831420898438, 130.226318359375, 7.672882080078125, 201.06591796875, 37.6298828125, -29.92601776123047, 0.59222412109375, 14.598800659179688, 32.586456298828125, 69.91317749023438, 12.50787353515625, -5.730583190917969, -3.841552734375, -8.5655517578125, 124.99307250976562, 38.54090881347656, 135.744873046875, 192.20742797851562, 165.28890991210938, -20.61236572265625, 104.39799499511719, 289.2869873046875, 17.999130249023438, 138.46218872070312, 34.107177734375, 3.5663414001464844, 34.83349609375, 150.83526611328125, 18.90814208984375, -13.19873046875, 130.63427734375, 53.39495849609375, 167.562744140625, 4.748355865478516, 93.31768798828125, 17.035858154296875, 24.430328369140625, -70.06027221679688, -37.85577392578125, 76.05005645751953, -76.91706848144531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000298.npy"}
{"epoch": 0.6240837696335079, "step": 299, "batch_size": 128, "mean": 42.75865936279297, "std": 66.02165222167969, "min": -145.74545288085938, "p10": -30.631298828125, "median": 23.552597045898438, "p90": 133.02252807617188, "max": 175.530517578125, "pos_frac": 0.7109375, "sample": [102.60810852050781, 133.24520874023438, 18.080520629882812, 96.24093627929688, 20.759521484375, 2.304779052734375, 132.61959838867188, -98.12957763671875, 106.11151123046875, -145.74545288085938, 155.02810668945312, 43.973785400390625, 20.78301239013672, 106.81640625, -43.12506103515625, 128.38381958007812, -2.71234130859375, 31.91687774658203, 11.762298583984375, -122.03733825683594, 125.15845489501953, -1.845489501953125, 58.11163330078125, -74.79649353027344, 136.91033935546875, -5.2051544189453125, -52.269439697265625, -30.835052490234375, 115.62478637695312, 27.021076202392578, 92.87860107421875, -4.687774658203125, 138.20396423339844, -1.6952972412109375, -52.76179504394531, 78.0877685546875, 109.31085205078125, -11.86749267578125, -63.35419464111328, 107.39694213867188, 9.66741943359375, 161.30555725097656, 24.217910766601562, 11.479036331176758, -8.141807556152344, 1.2806396484375, 141.3753662109375, -4.469219207763672, -5.649322509765625, -68.87330627441406, -62.719970703125, 12.9671630859375, 8.921663284301758, 16.659912109375, 110.30975341796875, -36.270477294921875, 164.59617614746094, -6.53070068359375, 169.52862548828125, 76.49553680419922, 115.72215270996094, 18.58203125, 16.89031982421875, 106.06594848632812, -3.3474903106689453, 175.530517578125, -2.54144287109375, 10.164745330810547, 104.920166015625, 24.018035888671875, 108.89730834960938, 8.677648544311523, 58.7352294921875, 8.091339111328125, 114.82283020019531, 101.03163146972656, -2.1957473754882812, -17.447635650634766, -6.6153564453125, 125.478515625, -4.2024383544921875, 79.96646118164062, 53.19512939453125, 13.9801025390625, 46.00439453125, -6.6607513427734375, 133.319580078125, 44.486907958984375, 70.43795776367188, 94.64256286621094, 12.769744873046875, 93.77117919921875, 137.3969268798828, 11.553375244140625, 20.765213012695312, -9.41326904296875, 40.34695053100586, 3.9808578491210938, 117.69181060791016, -10.824928283691406, -30.543975830078125, 50.864715576171875, 36.080078125, -0.440826416015625, 123.39436340332031, 171.90176391601562, 14.48809814453125, 132.92709350585938, 27.07958984375, -1.18670654296875, 14.547271728515625, 2.277862548828125, 119.59497833251953, -33.04835510253906, 9.49737548828125, 11.1015625, -3.1185302734375, 33.74151611328125, 70.13185119628906, 38.571044921875, -13.107002258300781, 103.31546020507812, 142.64039611816406, 81.2587890625, 31.556884765625, 126.2823486328125, 23.087158203125, 77.10427856445312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000299.npy"}
{"epoch": 0.6261780104712041, "step": 300, "batch_size": 128, "mean": 43.75682830810547, "std": 76.83670806884766, "min": -122.28195190429688, "p10": -49.186711120605466, "median": 34.11672019958496, "p90": 141.10482788085938, "max": 236.2218017578125, "pos_frac": 0.6796875, "sample": [139.28536987304688, 100.21978759765625, 20.08755874633789, -93.92628479003906, 21.577606201171875, -27.272369384765625, 101.19805908203125, 93.25794982910156, 29.320547103881836, -47.66876220703125, 157.988037109375, 127.81320190429688, 143.35940551757812, 7.188770294189453, -3.9464950561523438, 135.82476806640625, -109.23031616210938, 28.764892578125, 79.56831359863281, -6.710479736328125, 149.24929809570312, 69.88800048828125, 43.4130859375, 86.64888000488281, 40.58551025390625, 6.762664794921875, 39.24810791015625, -14.65338134765625, 33.20330810546875, -28.099327087402344, 86.7586669921875, 82.96270751953125, 131.05459594726562, 117.74774169921875, 153.5474853515625, -46.544097900390625, -122.28195190429688, 16.804840087890625, 3.6144275665283203, 183.98159790039062, 58.019195556640625, -19.428955078125, -36.65411376953125, -2.7003173828125, 105.66818237304688, 128.15164184570312, -52.72859191894531, 134.31845092773438, 117.21226501464844, 168.77145385742188, 10.66537094116211, -111.22261047363281, -100.09425354003906, 31.977691650390625, 35.03013229370117, -4.990074157714844, 49.033905029296875, 68.44345092773438, 145.9178466796875, -8.836505889892578, 122.79991912841797, 125.76054382324219, -65.29595947265625, 154.64651489257812, 44.105926513671875, -0.323974609375, 1.94970703125, 102.07574462890625, 102.41635131835938, 127.04341125488281, 61.0460205078125, 100.38858032226562, -3.664548873901367, 11.5892333984375, -19.535263061523438, -98.62751770019531, 162.68153381347656, 52.441650390625, -32.75007629394531, 131.73516845703125, 140.13858032226562, 14.815677642822266, 45.326446533203125, -54.7091064453125, 110.5847396850586, 19.8153076171875, 0.89373779296875, 0.0, -13.290878295898438, 39.573760986328125, 169.9427490234375, -37.8956298828125, -120.15359497070312, 105.06207275390625, -1.7504425048828125, 135.79788208007812, 73.09892272949219, 5.89501953125, 112.9427490234375, 124.3685302734375, 12.922637939453125, 236.2218017578125, 7.48809814453125, 71.6107177734375, -15.007293701171875, 100.44558715820312, 122.52511596679688, -63.7611083984375, -34.92034912109375, -14.67742919921875, 84.40841674804688, -6.280096054077148, 20.871795654296875, -22.440460205078125, 92.10528564453125, 115.75988006591797, 47.685394287109375, 2.671051025390625, -95.38749694824219, -0.620452880859375, 174.82305908203125, 0.73785400390625, -85.75521850585938, -28.0006103515625, 130.02862548828125, 149.154541015625, -6.640678405761719, 6.819793701171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000300.npy"}
{"epoch": 0.6282722513089005, "step": 301, "batch_size": 128, "mean": 42.600975036621094, "std": 73.56432342529297, "min": -176.82672119140625, "p10": -33.75511016845703, "median": 27.058734893798828, "p90": 148.0282028198242, "max": 209.54608154296875, "pos_frac": 0.7265625, "sample": [14.279449462890625, 167.2244873046875, 129.84677124023438, 92.09674072265625, 19.43329620361328, 136.20394897460938, 36.226165771484375, -37.96531677246094, 110.31710815429688, 8.873224258422852, 3.8878116607666016, -1.58624267578125, 127.39935302734375, 13.657501220703125, 3.7822265625, 4.388526916503906, -27.501907348632812, -43.225006103515625, 118.21795654296875, 159.7623291015625, 14.295326232910156, 13.746414184570312, 88.2679443359375, 20.660736083984375, 35.98933410644531, 16.44659423828125, -0.39544677734375, 51.4986572265625, 11.5572509765625, 71.669189453125, 108.4910888671875, -10.646240234375, 0.0, 155.9037628173828, 122.55218505859375, 91.82259368896484, -15.499664306640625, 31.71697998046875, -32.109249114990234, 122.94865417480469, -18.934724807739258, 110.18798828125, 156.51902770996094, 23.18067169189453, -148.5697021484375, 5.014251708984375, -52.63433837890625, 111.81396484375, -30.195404052734375, 23.168670654296875, 60.58294677734375, -9.95281982421875, -33.54823303222656, 146.96595764160156, 172.32217407226562, 12.603546142578125, 38.91258239746094, 158.09646606445312, 1.585693359375, 209.54608154296875, 58.30768585205078, 115.97171020507812, 21.29498291015625, 34.27040100097656, -101.62176513671875, -75.93067169189453, 31.70489501953125, -176.82672119140625, -111.79251098632812, 95.71733093261719, 81.0968246459961, 56.05177307128906, -18.773590087890625, 15.929557800292969, 12.585433959960938, 29.5845947265625, 140.2520751953125, 34.93719482421875, 150.50677490234375, -72.60122680664062, 4.0223541259765625, -35.02642822265625, 171.2587127685547, 88.27261352539062, 150.51220703125, 28.1474609375, -5.385135650634766, -27.285446166992188, 114.77542114257812, -109.11444091796875, 133.3076171875, 37.48193359375, -19.276641845703125, 2.2588958740234375, 119.44189453125, -26.215530395507812, 79.40696716308594, 26.358734130859375, 70.06639862060547, 112.7464599609375, 119.42965698242188, 133.99224853515625, 151.78652954101562, -9.080596923828125, 27.75873565673828, 142.2288055419922, 33.72833251953125, -22.642654418945312, 158.21173095703125, -13.266197204589844, 19.15716552734375, 12.840621948242188, -12.600830078125, 18.568817138671875, 4.033164978027344, 105.6646499633789, -48.100189208984375, 30.754976272583008, 4.0433502197265625, 130.89195251464844, 166.22927856445312, -34.237823486328125, -22.094635009765625, 66.42510986328125, 23.858016967773438, 60.552215576171875, 106.74742126464844, -9.27325439453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000301.npy"}
{"epoch": 0.6303664921465969, "step": 302, "batch_size": 128, "mean": 51.147911071777344, "std": 76.79016876220703, "min": -147.13458251953125, "p10": -32.38484573364256, "median": 48.21120643615723, "p90": 143.9297882080078, "max": 179.5015869140625, "pos_frac": 0.7265625, "sample": [150.88307189941406, 52.149505615234375, 112.72048950195312, 145.36309814453125, 22.14997100830078, 110.65360260009766, 0.0, -8.768936157226562, -1.43682861328125, -61.06437683105469, -131.14553833007812, 73.51765441894531, 61.80292510986328, 142.44973754882812, -8.782867431640625, 13.561416625976562, 65.44390869140625, 47.086212158203125, -18.645523071289062, -72.4426040649414, 156.09323120117188, 132.749755859375, 164.50042724609375, -13.62054443359375, 113.66860961914062, 124.61654663085938, 179.5015869140625, 16.980072021484375, -104.3658447265625, 107.89073944091797, 29.538841247558594, 134.39797973632812, 122.16600036621094, 73.03793334960938, 48.22188186645508, 105.3468017578125, 170.53732299804688, 63.29637145996094, 60.239959716796875, 48.200531005859375, 112.27645874023438, -1.22509765625, -147.13458251953125, -5.675872802734375, 89.11275482177734, 82.07122802734375, 173.7598876953125, 41.89752197265625, 160.54254150390625, 140.5463409423828, -9.231689453125, 132.0022430419922, 12.505363464355469, 4.1287078857421875, -6.402473449707031, -94.83299255371094, 115.47613525390625, 17.011749267578125, 115.42022705078125, 38.83526611328125, 104.88726806640625, 96.87885284423828, 43.48748779296875, 3.9976730346679688, 40.69659423828125, 10.384185791015625, -6.071929931640625, 12.631935119628906, -5.479957580566406, -84.3802490234375, 111.0248794555664, 150.66659545898438, -26.781234741210938, -54.4874267578125, 135.4638671875, 114.4027099609375, -23.4285888671875, 130.84304809570312, 125.53547668457031, 46.69261932373047, 23.289459228515625, 56.5615234375, 131.80743408203125, -108.82879638671875, 143.94284057617188, -6.986907958984375, -143.08267211914062, 52.37224578857422, -19.390060424804688, -100.23018646240234, 11.243875503540039, -103.08453369140625, -3.85491943359375, 56.77032470703125, 5.56053352355957, 47.3262939453125, 105.44978332519531, 141.2032470703125, 140.69281005859375, -45.459938049316406, 2.046417236328125, 26.47137451171875, 53.210899353027344, -21.93597412109375, 121.75334930419922, 0.5310592651367188, 30.7269287109375, 177.34902954101562, 46.15739059448242, 108.05795288085938, -0.8784027099609375, 10.432762145996094, 48.5040283203125, -16.722137451171875, 130.61312866210938, 93.10487365722656, 178.57186889648438, 3.727764129638672, 123.28732299804688, 52.414398193359375, 24.217178344726562, 175.91259765625, 105.42507934570312, 136.04588317871094, 98.17778015136719, -8.673004150390625, 143.9241943359375, -7.3643798828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000302.npy"}
{"epoch": 0.6324607329842932, "step": 303, "batch_size": 128, "mean": 46.08173370361328, "std": 78.24722290039062, "min": -159.6595458984375, "p10": -39.53973541259766, "median": 41.779693603515625, "p90": 140.20360107421874, "max": 199.1978759765625, "pos_frac": 0.734375, "sample": [12.145965576171875, 17.81543731689453, -133.22543334960938, -106.77235412597656, 114.189208984375, -6.576507568359375, 114.15203857421875, 56.033416748046875, 127.78118896484375, 101.33889770507812, -46.5220947265625, 126.2644271850586, 143.94888305664062, 128.72830200195312, 41.247833251953125, -15.88189697265625, -9.55999755859375, 118.25872802734375, -8.176494598388672, 153.7154541015625, 99.72059631347656, 17.831298828125, 115.2097396850586, 38.04917907714844, 71.46728515625, -40.7156982421875, 91.11947631835938, -10.43084716796875, 158.2161865234375, 11.661598205566406, 138.7598876953125, 129.47457885742188, -39.03575134277344, 130.4451446533203, -10.731170654296875, 12.5032958984375, 6.2229156494140625, 46.6610107421875, 0.0, 7.9169769287109375, -105.84545135498047, 54.10527801513672, -13.431343078613281, 12.735702514648438, 10.188949584960938, 5.542510986328125, 129.4393768310547, 46.401641845703125, 90.11787414550781, 113.73123168945312, 93.03253173828125, -122.07560729980469, 118.001220703125, -159.6595458984375, 134.20880126953125, 134.96092224121094, 150.17742919921875, 86.81916809082031, 35.523040771484375, 120.96331787109375, 0.18241500854492188, -2.5466156005859375, 20.466461181640625, -11.114501953125, -64.36917114257812, 7.62310791015625, 94.85296630859375, 117.74517822265625, 134.38958740234375, 23.970809936523438, 159.00881958007812, -113.83828735351562, 19.728485107421875, -0.606475830078125, 128.82229614257812, 109.6881103515625, -1.216888427734375, 83.12985229492188, 133.102783203125, 187.48602294921875, 51.21006774902344, -155.973876953125, 111.087646484375, -123.15902709960938, 7.024068832397461, 64.263427734375, 112.5206298828125, 35.65711975097656, 77.31585693359375, 82.8697509765625, 11.0216064453125, -2.19287109375, 144.1956329345703, -2.476806640625, 146.88644409179688, 2.3497314453125, -5.003257751464844, 64.75320434570312, -30.27032470703125, 7.892791748046875, -28.23858642578125, -142.12399291992188, 53.57330322265625, -2.640869140625, 5.517036437988281, 68.08464813232422, 143.572265625, 31.046600341796875, 130.79908752441406, 106.01395416259766, 42.311553955078125, 54.481475830078125, 61.129249572753906, 117.5535888671875, -36.107940673828125, 199.1978759765625, 111.93612670898438, -103.94049072265625, 151.17657470703125, -1.0230865478515625, 150.6656494140625, 17.57818603515625, 176.302490234375, 24.096145629882812, 33.704681396484375, 9.01800537109375, 11.67010498046875, 48.44322204589844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000303.npy"}
{"epoch": 0.6345549738219896, "step": 304, "batch_size": 128, "mean": 40.73255157470703, "std": 73.80952453613281, "min": -142.81707763671875, "p10": -42.682830810546875, "median": 26.354629516601562, "p90": 138.6007537841797, "max": 176.68951416015625, "pos_frac": 0.71875, "sample": [0.0, 29.706268310546875, -11.850311279296875, 68.57208251953125, -38.805023193359375, 0.275238037109375, -138.1061553955078, 132.8240966796875, 124.589599609375, -23.74566650390625, 97.0601806640625, 124.4483642578125, 35.81681823730469, 94.76997375488281, -80.46139526367188, 40.7056884765625, 15.74163818359375, 144.65451049804688, 27.099578857421875, 11.069427490234375, 12.959548950195312, 12.351829528808594, -3.7961578369140625, -12.34716796875, 116.53425598144531, 129.87789916992188, 117.6619873046875, 23.787490844726562, 12.367218017578125, 1.8178253173828125, -16.023193359375, 7.815244674682617, 5.3730316162109375, -4.7012481689453125, 31.339614868164062, 22.512969970703125, 37.942359924316406, 42.20390319824219, 109.76446533203125, 106.99423217773438, -18.578079223632812, 98.76332092285156, 107.64527893066406, -44.265350341796875, 168.8031005859375, 21.9359130859375, -5.879905700683594, 100.87274169921875, -73.62014770507812, 40.9732666015625, 36.31208801269531, -53.39442443847656, 84.33998107910156, -142.81707763671875, 0.0, 140.48062133789062, -87.29354858398438, 94.77980041503906, 7.689727783203125, -8.1114501953125, -60.59674072265625, 22.802810668945312, -109.19467163085938, 100.7169189453125, 20.111328125, 64.01133728027344, -37.970123291015625, 46.0914306640625, 176.68951416015625, 122.48114013671875, -107.42886352539062, 20.613555908203125, 41.24620056152344, -93.94439697265625, -40.51385498046875, 158.66842651367188, 146.13369750976562, 2.1887855529785156, 149.20477294921875, -104.13079833984375, 130.02276611328125, 109.93356323242188, -26.405860900878906, -5.703495025634766, 138.51382446289062, 19.86480712890625, 140.87545776367188, -3.87213134765625, 132.22323608398438, -124.37286376953125, 10.212066650390625, 0.4626140594482422, 25.60968017578125, 141.9844970703125, 148.97303771972656, 9.916122436523438, -42.004608154296875, 120.89009094238281, 132.29920959472656, 138.8035888671875, 31.916839599609375, 33.28289794921875, 63.339202880859375, 10.865020751953125, 46.088035583496094, 133.08621215820312, 30.016006469726562, 158.4761962890625, 120.63859558105469, 133.13677978515625, 108.35842895507812, 49.287200927734375, -25.04095458984375, 3.78271484375, 84.01615905761719, 16.73883056640625, -22.179931640625, -6.51397705078125, 106.32093811035156, -7.3471527099609375, 25.352645874023438, 105.69104766845703, 2.833251953125, 127.35626220703125, -18.770553588867188, 130.8665771484375, 21.28033447265625, 158.04644775390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000304.npy"}
{"epoch": 0.6366492146596858, "step": 305, "batch_size": 128, "mean": 60.137386322021484, "std": 76.6485366821289, "min": -180.9935302734375, "p10": -13.812486267089838, "median": 51.470802307128906, "p90": 151.62968139648436, "max": 274.093017578125, "pos_frac": 0.796875, "sample": [7.3760986328125, 23.370586395263672, -11.952651977539062, 191.56411743164062, 37.450927734375, 3.8319091796875, 106.85423278808594, -54.759246826171875, 104.6259765625, 21.29936981201172, 92.2832260131836, 36.11402893066406, -10.62933349609375, 80.05014038085938, 19.402603149414062, -88.52197265625, -69.80191040039062, 25.521011352539062, 58.19944763183594, 2.888303756713867, 60.222259521484375, 140.24710083007812, 0.7978134155273438, 69.14543151855469, 5.1263580322265625, 17.833770751953125, -109.37640380859375, -2.9198150634765625, 32.37770080566406, 118.88130187988281, 22.899078369140625, 64.98150634765625, 123.02532958984375, 98.64120483398438, 148.43988037109375, 24.084915161132812, -0.1824188232421875, -4.0906829833984375, 130.32980346679688, -2.0128936767578125, -71.73567199707031, 50.76670837402344, -6.365116119384766, 110.9483871459961, 146.76300048828125, 135.1422882080078, 50.1317138671875, 156.98419189453125, 136.23040771484375, 52.174896240234375, -9.107088088989258, -109.88002014160156, 164.62933349609375, 152.96478271484375, 38.88307189941406, 62.3411865234375, 274.093017578125, 100.6827392578125, 175.31512451171875, 156.4891357421875, 19.405349731445312, -0.2617835998535156, -8.616455078125, 147.16220092773438, 36.967437744140625, 106.65416717529297, 150.77688598632812, 125.01158142089844, 86.23583984375, 123.33709716796875, -18.152099609375, 117.66197967529297, -68.76945495605469, 36.64497375488281, 121.93159484863281, -180.9935302734375, 24.918350219726562, 119.02540588378906, -2.035919189453125, 65.41213989257812, 59.269622802734375, 148.295166015625, 155.76844787597656, 9.612106323242188, 110.16720581054688, 108.67916870117188, -31.354583740234375, 194.31265258789062, 2.9372024536132812, 11.360488891601562, 154.591552734375, 109.38055419921875, 137.87646484375, 37.9583740234375, 60.56646728515625, 126.05209350585938, 52.2275390625, 146.44732666015625, 173.87860107421875, 52.19004821777344, 32.7415771484375, -4.22967529296875, 123.23171997070312, 127.08050537109375, 190.265380859375, 24.5826416015625, 32.4166259765625, 36.35064697265625, 47.57818603515625, 20.517822265625, 151.0574951171875, 1.1944656372070312, 28.999465942382812, 6.04034423828125, 18.799514770507812, -85.7364273071289, 150.60159301757812, 4.800148010253906, 12.545059204101562, -34.064697265625, 134.48507690429688, 53.3465576171875, 185.99017333984375, 131.93692016601562, 133.4664306640625, -71.90365600585938, -2.92822265625, 144.81715393066406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000305.npy"}
{"epoch": 0.6387434554973822, "step": 306, "batch_size": 128, "mean": 52.202308654785156, "std": 74.37667083740234, "min": -138.0650634765625, "p10": -21.311334228515623, "median": 46.70103454589844, "p90": 147.17316436767578, "max": 184.48748779296875, "pos_frac": 0.7421875, "sample": [37.50849914550781, -103.7686767578125, -103.2350082397461, 180.75149536132812, -135.26580810546875, 174.65911865234375, -2.401968002319336, 55.37129211425781, -13.264862060546875, 135.3865966796875, 25.915241241455078, 25.19024658203125, 139.59703063964844, 93.07002258300781, 25.32592010498047, 5.3026885986328125, -36.191650390625, 123.43832397460938, -3.2875823974609375, 21.9814453125, 42.51861572265625, 45.22728729248047, -38.191497802734375, 75.50286865234375, -138.0650634765625, 166.60369873046875, 168.881103515625, 123.4736328125, 114.46484375, -0.7625656127929688, 59.28887939453125, 81.27877807617188, 44.999755859375, 127.25848388671875, 147.31687927246094, 123.49016571044922, 169.75604248046875, 123.9713134765625, 69.756591796875, 99.17172241210938, 7.13262939453125, 110.69794464111328, -7.33233642578125, 10.378311157226562, 70.91976928710938, 91.94175720214844, 41.1402587890625, 69.66072082519531, 58.20623779296875, -78.36102294921875, -9.745574951171875, 146.2572479248047, -13.213096618652344, 177.83575439453125, 51.2230224609375, -136.46063232421875, 82.16918182373047, 77.95162963867188, -16.817779541015625, 175.5811767578125, -2.9463882446289062, 65.75543212890625, 102.68527221679688, 77.27690124511719, 120.98575592041016, 3.925729751586914, 1.7781524658203125, 2.9917774200439453, 105.423095703125, -20.818450927734375, 115.38729858398438, -112.85971069335938, 110.50828552246094, 40.49114990234375, 177.00762939453125, -1.1117725372314453, -43.716644287109375, 33.12168884277344, 18.463943481445312, 36.27471923828125, 13.802200317382812, 59.208892822265625, 118.38619995117188, 115.500732421875, 147.111572265625, -21.971298217773438, -21.10784912109375, 60.90972900390625, 84.89346313476562, 184.48748779296875, 68.00830078125, 46.233612060546875, 15.773529052734375, -21.7861328125, 47.16845703125, 115.8726806640625, 113.67251586914062, -11.160110473632812, 143.42315673828125, -16.6127872467041, 142.2293701171875, 66.75088500976562, 133.6918182373047, 97.74510192871094, 25.596328735351562, 17.547630310058594, 6.183403015136719, -3.9549102783203125, -3.106292724609375, -4.2312469482421875, -7.858020782470703, 3.58026123046875, 139.2018585205078, 102.60159301757812, 14.9437255859375, 172.36663818359375, 32.885765075683594, -3.0919342041015625, -19.0078125, 5.857940673828125, 108.19174194335938, 116.77879333496094, 0.44956207275390625, -108.41939544677734, 81.09307861328125, 23.464080810546875, 159.09115600585938, 147.68670654296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000306.npy"}
{"epoch": 0.6408376963350786, "step": 307, "batch_size": 128, "mean": 56.27093505859375, "std": 72.37808227539062, "min": -182.94259643554688, "p10": -18.97692108154297, "median": 64.11931228637695, "p90": 138.90304412841797, "max": 190.9200439453125, "pos_frac": 0.7734375, "sample": [142.84689331054688, 134.0459747314453, 12.531143188476562, 167.13592529296875, -119.3843994140625, -5.452180862426758, 86.28457641601562, 105.00033569335938, 87.85784912109375, 132.11444091796875, -159.88381958007812, -9.561590194702148, 87.29595947265625, 6.116302490234375, -5.161651611328125, 115.96006774902344, 134.32730102539062, 117.9393310546875, 12.068496704101562, 21.63262939453125, 59.1915283203125, -40.010650634765625, 73.163818359375, 8.917116165161133, 172.34295654296875, -28.979339599609375, 63.10040283203125, 13.05145263671875, 123.73654174804688, 159.69915771484375, 113.97244262695312, 25.05126953125, 29.00665283203125, 92.4519271850586, 109.18043518066406, -13.248992919921875, 142.36642456054688, 67.3720703125, 20.121734619140625, -6.448860168457031, 24.080718994140625, 95.8536376953125, 121.1801986694336, -13.2119140625, 128.03260803222656, 96.01779174804688, 137.3941192626953, -11.386245727539062, -10.237686157226562, 26.066497802734375, 83.70213317871094, 12.695690155029297, 138.8389434814453, 7.688899993896484, 75.43319702148438, 123.82496643066406, 111.5233154296875, 146.6992645263672, 16.5078125, 3.72747802734375, 16.406005859375, 69.65348815917969, -67.36373901367188, 110.09539794921875, -1.5789642333984375, -3.5905799865722656, 127.96047973632812, 75.15238189697266, -8.257804870605469, 5.361259460449219, -5.753566741943359, 133.1136474609375, -43.44256591796875, -38.32904052734375, 29.97943115234375, -18.479583740234375, 47.466705322265625, 130.75758361816406, 42.62884521484375, 24.446044921875, 138.7288818359375, -20.137374877929688, 65.13822174072266, 140.732421875, 163.68072509765625, 126.97085571289062, -26.656505584716797, 119.72808837890625, 6.53978157043457, 89.3646240234375, 16.42131805419922, -39.76017761230469, 82.85507202148438, 167.88018798828125, 101.88264465332031, 16.464447021484375, 132.74652099609375, -136.64004516601562, -15.68304443359375, 124.784912109375, 4.952301025390625, 175.454345703125, 31.820648193359375, 154.40078735351562, -182.94259643554688, 11.0545654296875, 2.27239990234375, 97.08651733398438, 33.8525390625, 94.0108871459961, -20.982330322265625, 22.2845458984375, 190.9200439453125, 120.95416259765625, 19.30072021484375, 101.74365234375, -2.0631103515625, -6.426910400390625, 80.90394592285156, 129.01409912109375, 99.01742553710938, 7.6591949462890625, 119.33021545410156, 137.23764038085938, 84.93724060058594, 3.1116676330566406, 139.0526123046875, 111.27268981933594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000307.npy"}
{"epoch": 0.6429319371727749, "step": 308, "batch_size": 128, "mean": 38.80859375, "std": 74.44212341308594, "min": -170.11886596679688, "p10": -42.27932434082031, "median": 17.398513793945312, "p90": 136.0209197998047, "max": 198.09182739257812, "pos_frac": 0.7421875, "sample": [145.01219177246094, 36.097496032714844, -42.943115234375, 145.05413818359375, 86.97369384765625, -96.53892517089844, -46.322418212890625, 52.662872314453125, 9.377288818359375, 86.49029541015625, 4.08709716796875, 150.32281494140625, 0.14414215087890625, -82.5496826171875, -5.331085205078125, 3.872894287109375, 115.77548217773438, -9.584671020507812, 121.7169189453125, 114.2740478515625, 126.4227066040039, 35.351318359375, 4.47265625, -86.17467498779297, 13.22979736328125, 115.87364196777344, -150.291259765625, -25.300331115722656, 14.479095458984375, 142.05642700195312, 116.68840026855469, 17.0711669921875, 34.639404296875, 104.50996398925781, -19.28887939453125, 1.7054443359375, 125.94198608398438, 7.994743347167969, 102.93494415283203, 10.518295288085938, -2.5988693237304688, 38.769020080566406, 183.178955078125, 0.18328857421875, -34.999542236328125, -6.38922119140625, 0.79156494140625, 137.35040283203125, -4.603975296020508, 118.56208801269531, 5.541557312011719, 24.234603881835938, 3.2458114624023438, 112.82381439208984, -7.577354431152344, 28.001922607421875, 124.69195556640625, 36.56884765625, -170.11886596679688, 135.45114135742188, 89.38517761230469, 10.654571533203125, -7.74755859375, -19.381988525390625, 198.09182739257812, 116.73908996582031, -2.854471206665039, -120.95828247070312, 13.487060546875, -2.9132003784179688, 96.23257446289062, -90.74547576904297, 144.06875610351562, 74.3266830444336, 10.3740234375, 17.725860595703125, 158.86074829101562, 184.41912841796875, 144.86294555664062, -1.2107810974121094, 79.91116333007812, 1.090576171875, 1.9303169250488281, 8.3065185546875, 71.201416015625, -123.92709350585938, 26.530654907226562, 6.785736083984375, -39.481842041015625, 108.62217712402344, 21.455657958984375, 13.974693298339844, 34.39239501953125, 10.118515014648438, 119.11198425292969, 109.8304672241211, 153.8323974609375, 130.58599853515625, 31.991897583007812, 122.334716796875, 27.167739868164062, 74.02734375, 176.9525146484375, -58.52375793457031, 36.371055603027344, 12.770454406738281, 11.36260986328125, -38.559627532958984, 118.90737915039062, 115.89459228515625, -74.32815551757812, 13.177001953125, -24.90447998046875, 2.8363418579101562, 2.6352310180664062, 99.1407470703125, -30.432266235351562, 13.137008666992188, -6.319549560546875, 6.509918212890625, -41.994842529296875, 98.26313781738281, 78.77627563476562, 123.64291381835938, -92.61723327636719, 48.7684326171875, 100.57101440429688, 67.71987915039062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000308.npy"}
{"epoch": 0.6450261780104712, "step": 309, "batch_size": 128, "mean": 57.46986389160156, "std": 81.36467742919922, "min": -123.54681396484375, "p10": -38.84646263122557, "median": 51.87038803100586, "p90": 159.88506469726562, "max": 285.1866149902344, "pos_frac": 0.734375, "sample": [61.93792724609375, 144.60858154296875, 134.4189453125, 113.70925903320312, -1.133575439453125, -64.60723876953125, 4.015960693359375, -102.15221405029297, 136.00234985351562, 44.578338623046875, 160.50811767578125, 13.332023620605469, -11.810150146484375, 7.6549530029296875, -90.18414306640625, 122.3488998413086, 12.73870849609375, 22.58684539794922, -9.887115478515625, 40.920684814453125, -9.259628295898438, 110.60848999023438, -10.579498291015625, -101.83302307128906, 13.01092529296875, 80.88348388671875, 2.2919769287109375, 285.1866149902344, 114.90020751953125, 53.195068359375, -7.9680023193359375, 168.302001953125, 108.40493774414062, 0.9897994995117188, -4.862274169921875, -11.456375122070312, 95.75896453857422, -48.37567138671875, 197.05703735351562, -90.25823974609375, 181.62786865234375, -66.71484375, -4.734260559082031, 58.181884765625, 122.29864501953125, -32.938560485839844, -1.06805419921875, -34.264190673828125, -71.70964813232422, 15.00091552734375, 131.1482696533203, -5.993091583251953, 14.817474365234375, -93.576416015625, 143.67807006835938, 122.88418579101562, 128.0736083984375, 13.69232177734375, 101.7298583984375, 14.42596435546875, 159.6180419921875, 4.702362060546875, 51.04576873779297, 221.2210693359375, -13.174690246582031, 168.2017822265625, 10.733993530273438, 118.32222747802734, 24.91132354736328, 43.605712890625, 42.76542663574219, 34.42749786376953, -0.8987503051757812, 177.1287841796875, 2.4136505126953125, 151.67529296875, 143.1405029296875, 104.12739562988281, 12.489456176757812, 162.52215576171875, 122.73751831054688, 131.6900634765625, 5.997184753417969, 87.76268768310547, 123.78153228759766, 145.57476806640625, -4.846748352050781, 93.17032623291016, 38.93560791015625, -34.762516021728516, 126.7032470703125, -123.54681396484375, 60.617462158203125, -17.237272262573242, 163.1650848388672, 122.6177978515625, 19.222213745117188, 132.89503479003906, 47.261260986328125, 9.353816986083984, -108.96075439453125, 59.05531311035156, 129.93057250976562, 158.43157958984375, 143.46701049804688, -2.98541259765625, 186.93988037109375, 33.40008544921875, 59.9332275390625, -98.96878051757812, 148.70877075195312, 55.73382568359375, 76.23028564453125, 143.78717041015625, 77.64614868164062, 131.22036743164062, 52.69500732421875, 165.83001708984375, -11.531707763671875, 144.53733825683594, 164.71441650390625, 65.33744049072266, 111.20573425292969, -70.93731689453125, 115.9957275390625, 39.424072265625, -2.377584457397461, 95.46450805664062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000309.npy"}
{"epoch": 0.6471204188481675, "step": 310, "batch_size": 128, "mean": 45.30121612548828, "std": 70.85225677490234, "min": -127.41961669921875, "p10": -31.953146743774415, "median": 23.5721435546875, "p90": 138.37528076171873, "max": 224.43792724609375, "pos_frac": 0.7421875, "sample": [224.43792724609375, 12.571624755859375, -0.3384284973144531, 182.46533203125, 38.81585693359375, 35.072723388671875, 25.91015625, 4.481021881103516, 122.50965881347656, 77.38693237304688, 112.81002044677734, 168.7476806640625, 66.53244018554688, 118.71380615234375, 23.50537109375, 37.17901611328125, 10.124588012695312, -5.157686233520508, 39.9102783203125, -56.22093200683594, -0.948577880859375, 4.18524169921875, 19.8544921875, 54.458648681640625, 75.41999053955078, -95.25172424316406, 126.08552551269531, -40.85498046875, 119.89247131347656, 0.685302734375, -4.537467956542969, -1.514312744140625, -23.03302001953125, 135.20285034179688, -0.352630615234375, -2.261383056640625, 63.969757080078125, 161.1915283203125, -53.576324462890625, -99.5416259765625, 114.97598266601562, -3.1548919677734375, 0.8304290771484375, 21.377288818359375, 103.60368347167969, 36.002105712890625, 83.40518188476562, 51.69220733642578, 22.469268798828125, 1.593780517578125, -10.70037841796875, 23.638916015625, 153.67343139648438, 111.534912109375, 96.1300048828125, 135.88681030273438, 173.9208984375, 19.050140380859375, -3.5381927490234375, -127.41961669921875, 1.6514244079589844, 160.09774780273438, 38.788841247558594, 22.029266357421875, -2.580718994140625, -112.3297348022461, 7.4204254150390625, 25.152252197265625, 113.3638916015625, 4.493631362915039, -16.910720825195312, 21.363601684570312, 7.6981048583984375, 32.24481201171875, 37.632598876953125, 112.1968994140625, 125.17276000976562, 162.47100830078125, 57.82318115234375, -7.7931060791015625, -32.19902420043945, 102.91255187988281, 111.86709594726562, -0.3890838623046875, -32.513519287109375, 109.2463150024414, 136.57174682617188, 5.6866912841796875, 114.22100067138672, -10.02215576171875, 13.554344177246094, -50.358154296875, 126.04806518554688, -85.278076171875, 44.45024490356445, 2.4319381713867188, 7.9587249755859375, 133.8812255859375, 45.448394775390625, 17.97552490234375, 49.35717010498047, 151.39108276367188, 4.721099853515625, -31.84777069091797, 135.40841674804688, -9.040496826171875, 0.6546478271484375, -37.373626708984375, -100.73674011230469, 101.03155517578125, -11.427734375, 145.46270751953125, 11.985107421875, 114.70883178710938, 7.028533935546875, -1.10296630859375, 18.56695556640625, 5.294635772705078, 8.589057922363281, 103.23194885253906, 42.759735107421875, 5.988044738769531, 133.1233367919922, 149.80148315429688, 197.62063598632812, 111.74845886230469, 142.58352661132812, 114.04460144042969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000310.npy"}
{"epoch": 0.6492146596858639, "step": 311, "batch_size": 128, "mean": 50.63996505737305, "std": 83.89373779296875, "min": -177.46844482421875, "p10": -52.225307083129834, "median": 35.329166412353516, "p90": 148.54703674316406, "max": 208.14215087890625, "pos_frac": 0.7265625, "sample": [6.209175109863281, -0.9268646240234375, 122.43344116210938, 118.24029541015625, 144.87283325195312, 27.94099998474121, 29.358108520507812, -1.0704803466796875, 143.02484130859375, 114.80224609375, 136.6785888671875, 6.742168426513672, -26.393695831298828, -26.496810913085938, 35.18687438964844, -2.2396240234375, 142.621337890625, 68.11338806152344, -81.94367218017578, 50.664825439453125, 174.7626953125, 137.26712036132812, 5.0242156982421875, 116.24246215820312, -1.392852783203125, -16.129119873046875, 156.93165588378906, 117.6529541015625, 5.900508880615234, 23.57012939453125, -97.41263580322266, -177.46844482421875, -176.54525756835938, 122.73960876464844, 29.166351318359375, 157.86050415039062, -121.55072021484375, -87.068603515625, 132.65753173828125, 50.231719970703125, -82.17540740966797, 104.53794860839844, 34.333953857421875, 2.31793212890625, -2.0196380615234375, 51.93817138671875, -14.307548522949219, 111.04641723632812, -97.68817138671875, 28.93885040283203, 110.38745880126953, 24.25225830078125, 2.45770263671875, 28.032806396484375, 121.6107177734375, -14.16455078125, 95.06971740722656, -17.72100830078125, -20.022216796875, 159.04376220703125, 157.90374755859375, 135.24642944335938, 140.23797607421875, 119.12845611572266, 124.43548583984375, -98.39068603515625, 36.0655517578125, 11.3536376953125, 18.680252075195312, 35.16968536376953, -97.83853149414062, -12.694080352783203, 170.5091552734375, 89.93478393554688, 20.4049072265625, 35.471458435058594, 156.99249267578125, 116.20297241210938, 24.051025390625, 144.06008911132812, 208.14215087890625, 4.2216796875, -102.5526123046875, 46.323516845703125, -0.21725845336914062, -28.212860107421875, 162.4538116455078, 152.64691162109375, 136.78262329101562, 24.57244873046875, 120.94187927246094, 98.73252868652344, 107.58158874511719, 59.97833251953125, 171.0426025390625, -125.55650329589844, 166.337158203125, -8.678985595703125, -121.25994873046875, 117.05258178710938, 7.9794464111328125, 146.12374877929688, -15.127105712890625, 84.26898193359375, 11.450592041015625, -4.9611053466796875, 146.78994750976562, 142.4305419921875, 9.285333633422852, 43.784759521484375, 127.91415405273438, 107.16244506835938, 123.26602172851562, 29.162689208984375, -36.22056579589844, 7.110481262207031, 112.21676635742188, 138.90939331054688, 139.7553253173828, 2.270030975341797, -39.48886489868164, -3.5347213745117188, 5.3771820068359375, 191.0389404296875, 105.47752380371094, 131.13851928710938, 80.65911865234375, -15.707351684570312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000311.npy"}
{"epoch": 0.6513089005235602, "step": 312, "batch_size": 128, "mean": 49.544517517089844, "std": 76.53087615966797, "min": -164.03924560546875, "p10": -40.37529754638672, "median": 38.257164001464844, "p90": 141.46051483154295, "max": 196.48471069335938, "pos_frac": 0.7421875, "sample": [35.187255859375, -4.028411865234375, -14.873443603515625, 126.70523071289062, 136.4546356201172, -164.03924560546875, 134.67892456054688, 119.81820678710938, -38.79863739013672, -39.81547546386719, -18.110107421875, 2.4310836791992188, 75.70661926269531, 110.7242431640625, 85.74960327148438, 156.66110229492188, 96.17562866210938, 130.71678161621094, -17.689346313476562, -108.075927734375, 7.59246826171875, 120.1689453125, 130.957763671875, -2.17840576171875, 8.055198669433594, 60.56806182861328, 57.57470703125, 196.48471069335938, -39.20751953125, -13.560768127441406, 151.35177612304688, -7.2665557861328125, 2.54705810546875, 156.071044921875, 73.00860595703125, 30.649322509765625, 54.8912353515625, 1.1979293823242188, 93.68522644042969, 109.76434326171875, 170.45571899414062, 148.2646026611328, 65.53327178955078, 125.660888671875, 110.96963500976562, 115.03459167480469, 59.289581298828125, 133.6292724609375, 126.17533874511719, 22.9501953125, 133.76712036132812, 135.13650512695312, 17.723861694335938, 3.8176651000976562, -43.527496337890625, 9.255889892578125, 37.98985290527344, 23.075180053710938, 0.1613922119140625, 189.8267822265625, 8.623023986816406, -27.610504150390625, 135.14999389648438, -110.79803466796875, 64.59435272216797, -33.7178955078125, 165.13504028320312, 114.61666870117188, -14.142791748046875, 117.26852416992188, 26.81463623046875, 127.203369140625, 24.442663192749023, -5.432216644287109, 98.09347534179688, 87.70611572265625, 2.0832901000976562, 144.3372802734375, -11.20294189453125, 111.39111328125, 140.2276153564453, 149.23660278320312, -42.256683349609375, 122.36599731445312, 22.380163192749023, 0.0, 59.4921875, -84.94442749023438, -113.53619384765625, -105.86259460449219, 19.4913330078125, 71.88201904296875, 6.026206970214844, 38.74623107910156, 89.0364990234375, 0.4532318115234375, -108.14002990722656, 17.411556243896484, 99.96820068359375, 1.3304443359375, 26.245742797851562, 5.275749206542969, 128.285888671875, 87.37002563476562, -5.806732177734375, 106.63638305664062, 126.04925537109375, -3.84808349609375, 172.83450317382812, 35.80780029296875, 135.43377685546875, 21.9423828125, 38.52447509765625, -4.768335342407227, -76.81903076171875, 70.19087219238281, 78.26806640625, 8.769500732421875, -41.681549072265625, 170.92373657226562, -88.90631103515625, 154.092529296875, -29.056915283203125, 133.21966552734375, 114.84827423095703, 23.8416748046875, 18.202442169189453, -55.16520690917969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000312.npy"}
{"epoch": 0.6534031413612565, "step": 313, "batch_size": 128, "mean": 46.392696380615234, "std": 82.20933532714844, "min": -149.59844970703125, "p10": -66.40892639160155, "median": 32.73625183105469, "p90": 151.55247955322264, "max": 222.920654296875, "pos_frac": 0.71875, "sample": [162.02947998046875, 137.65902709960938, -24.6239013671875, -25.187774658203125, -39.35809326171875, -1.9654388427734375, 57.10736083984375, 132.381103515625, 158.50830078125, -14.52789306640625, -8.854141235351562, 134.29269409179688, -88.53802490234375, 141.9130401611328, 96.48271942138672, 153.22718811035156, 140.07534790039062, 41.698944091796875, -8.132598876953125, -26.94970703125, 1.7262115478515625, 15.1402587890625, 4.100872039794922, 139.76840209960938, 118.27569580078125, -7.247406005859375, -2.6900634765625, 23.56048583984375, 110.64779663085938, 179.41436767578125, 166.66122436523438, 203.4036865234375, 49.77214050292969, 83.6715087890625, -42.05352783203125, 72.83905029296875, 69.11933898925781, 66.75234985351562, 173.87722778320312, 53.98674774169922, -64.4705810546875, 178.72515869140625, 149.45191955566406, 23.1583251953125, -78.89849853515625, 50.73973083496094, 24.361663818359375, -57.42034912109375, 116.14936828613281, 138.51666259765625, 31.111099243164062, 91.79672241210938, -136.74468994140625, 18.96254539489746, 63.0828857421875, 0.66552734375, 150.83474731445312, 59.593475341796875, 32.73724365234375, 2.680999755859375, 120.61901092529297, 3.8115081787109375, 122.84380340576172, 19.837379455566406, 9.732444763183594, 40.437408447265625, -70.93173217773438, 106.3362808227539, -103.27555847167969, -93.63711547851562, 161.66526794433594, -16.845123291015625, 23.167274475097656, 150.73220825195312, 18.99957275390625, -5.39019775390625, 31.190414428710938, 137.60691833496094, -106.8319091796875, 105.16026306152344, 11.289901733398438, -105.99594116210938, 103.405517578125, -43.61053466796875, 37.544647216796875, 128.2259063720703, 86.2943115234375, -8.354339599609375, 8.525491714477539, 4.800575256347656, 29.338623046875, 222.920654296875, 111.04794311523438, 55.71630859375, 130.69876098632812, 149.0302734375, 25.08924102783203, 162.24407958984375, -38.57904052734375, -145.78152465820312, 64.77886962890625, 47.53211212158203, 7.597021102905273, 159.27728271484375, 33.94342041015625, -81.80533599853516, -1.1790828704833984, -2.3093948364257812, 142.69424438476562, 119.38626098632812, 26.905029296875, 91.37400817871094, 32.735260009765625, 4.605926513671875, -2.2532405853271484, -8.819656372070312, 117.21980285644531, 30.955215454101562, -102.81613159179688, -103.88639831542969, -0.28348541259765625, -149.59844970703125, 104.02369689941406, 82.46019744873047, 19.0474853515625, 150.5112762451172, 23.697479248046875, 160.39442443847656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000313.npy"}
{"epoch": 0.6554973821989529, "step": 314, "batch_size": 128, "mean": 49.07756805419922, "std": 80.5748291015625, "min": -167.24441528320312, "p10": -49.73122177124022, "median": 32.27846908569336, "p90": 142.94266052246093, "max": 202.61813354492188, "pos_frac": 0.7109375, "sample": [32.02404022216797, 106.75143432617188, 117.90596008300781, 44.34552001953125, -10.6483154296875, 19.743927001953125, -76.457763671875, 106.82940673828125, -74.74585723876953, -123.71821594238281, -9.89483642578125, 68.48750305175781, 138.44622802734375, 156.2627716064453, 158.34490966796875, -64.159912109375, -61.2886962890625, 177.119140625, 130.5093994140625, 149.70193481445312, 107.64910888671875, -8.40460205078125, 28.40911293029785, 117.80036926269531, -2.5633544921875, 13.637748718261719, 12.615562438964844, 167.7655029296875, 6.223602294921875, 170.89862060546875, 7.499645233154297, -14.330947875976562, 8.47100830078125, -95.41354370117188, 135.6337127685547, 76.43438720703125, 20.067626953125, 141.62155151367188, 41.0179443359375, 142.961181640625, -38.99607849121094, 115.71661376953125, -72.52426147460938, 114.45404052734375, 5.272846221923828, 26.83814239501953, 180.34963989257812, 81.68728637695312, 9.760894775390625, 2.1939697265625, 134.40328979492188, 0.6986160278320312, 19.791213989257812, 100.1568603515625, -0.8558006286621094, 102.31008911132812, 32.53289794921875, 18.89202880859375, -16.905563354492188, -100.32023620605469, 132.44662475585938, 140.60000610351562, 113.91139221191406, 0.0, -2.3046875, -65.19032287597656, -160.020751953125, 124.48483276367188, 22.54364013671875, -18.549652099609375, -24.451568603515625, 19.300628662109375, 105.55300903320312, 136.12432861328125, 155.94789123535156, -46.426124572753906, -16.875953674316406, 160.37005615234375, -30.217742919921875, -167.24441528320312, 127.84517669677734, 55.186065673828125, 130.87008666992188, 84.43911743164062, 0.0, 138.31878662109375, -3.76458740234375, 95.99368286132812, 133.7894287109375, 24.503387451171875, 26.12140655517578, 123.0601806640625, 130.67308044433594, 33.399627685546875, 103.8643798828125, 142.93472290039062, 121.35419464111328, 25.018218994140625, 44.650390625, 7.121734619140625, 9.3726806640625, -26.61065673828125, -5.008026123046875, -36.664398193359375, 107.70840454101562, 5.797393798828125, 93.09806060791016, 119.67311096191406, 44.3193359375, 107.0326919555664, 165.15200805664062, 30.936172485351562, -45.77557373046875, -4.7785491943359375, 1.023406982421875, 202.61813354492188, 135.169189453125, -14.827079772949219, 17.420581817626953, 89.7923355102539, 131.12330627441406, 111.625244140625, -57.443115234375, 96.74151611328125, -159.2423095703125, 166.45651245117188, 101.45716857910156, -8.628410339355469], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000314.npy"}
{"epoch": 0.6575916230366492, "step": 315, "batch_size": 128, "mean": 55.17478942871094, "std": 78.46570587158203, "min": -161.2351531982422, "p10": -25.637631225585924, "median": 28.817989349365234, "p90": 158.60635681152343, "max": 248.20263671875, "pos_frac": 0.78125, "sample": [1.76470947265625, 58.232757568359375, 2.7980918884277344, 3.7248992919921875, 26.721633911132812, -47.1956787109375, -21.4892578125, 12.046707153320312, 163.78500366210938, -11.937652587890625, 30.986114501953125, -14.080078125, 122.0544662475586, 20.389190673828125, 121.53298950195312, -36.57244873046875, 47.03868103027344, 154.11639404296875, 54.210906982421875, -21.141815185546875, -77.87064361572266, -21.25360870361328, 2.870391845703125, -5.029762268066406, 175.5906982421875, 11.56640625, 115.29141235351562, 165.5228729248047, 195.09039306640625, 1.5384521484375, 24.131641387939453, 40.372344970703125, 151.0081787109375, 153.6781005859375, 29.398101806640625, 174.43301391601562, 154.22039794921875, 141.90103149414062, 103.63748168945312, -3.9345703125, 0.043487548828125, -35.317169189453125, 21.778076171875, 146.26239013671875, -40.66004943847656, 25.132980346679688, 7.81695556640625, 5.744903564453125, 13.57861328125, 8.629180908203125, 138.85906982421875, -4.608646392822266, 138.45382690429688, 158.36001586914062, 138.66549682617188, 141.36097717285156, 15.96124267578125, 8.276283264160156, 34.781585693359375, 76.02024841308594, 12.716506958007812, -0.479278564453125, -9.61166763305664, 91.22105407714844, 129.239013671875, 142.97708129882812, 106.30697631835938, 113.64339447021484, 74.77090454101562, 248.20263671875, -77.1636962890625, 32.20924377441406, 120.3773193359375, 22.085113525390625, 143.398193359375, -41.019775390625, -139.30279541015625, 166.86724853515625, 3.5227890014648438, 159.24346923828125, 147.012939453125, 7.5294036865234375, -1.9022979736328125, 135.88739013671875, 183.3431396484375, 10.681831359863281, 37.206939697265625, 110.70887756347656, 21.62457275390625, 174.7513427734375, 18.690322875976562, 45.76287841796875, -143.164794921875, -40.377105712890625, 146.68252563476562, 164.25213623046875, 80.918212890625, 41.214778900146484, -161.2351531982422, 146.7884521484375, 25.36407470703125, -6.521881103515625, -2.561279296875, 101.77059173583984, 9.977752685546875, -73.8656005859375, -49.736480712890625, 54.4852294921875, 25.507892608642578, 18.25982666015625, 150.83343505859375, 13.197952270507812, 124.56513977050781, 196.2389373779297, 14.821044921875, 107.21040344238281, 159.18115234375, 11.26788330078125, -6.393707275390625, 14.880767822265625, 90.81658172607422, -3.7790756225585938, 132.49087524414062, 28.237876892089844, 132.92520141601562, 88.41864013671875, 63.41192626953125, 17.500473022460938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000315.npy"}
{"epoch": 0.6596858638743456, "step": 316, "batch_size": 128, "mean": 55.4439582824707, "std": 80.75297546386719, "min": -144.56002807617188, "p10": -36.844712829589845, "median": 40.300933837890625, "p90": 163.4361145019531, "max": 249.99920654296875, "pos_frac": 0.734375, "sample": [137.7529296875, 74.71206665039062, 63.88775634765625, 37.424407958984375, -18.6591796875, -36.45329284667969, -11.18951416015625, 165.817626953125, 69.9586181640625, 167.65859985351562, 26.191436767578125, -37.758026123046875, 141.0171356201172, 34.58360290527344, -75.91810607910156, 24.534101486206055, 118.42539978027344, -58.28369903564453, 112.57431030273438, 10.84613037109375, 122.92822265625, 152.90927124023438, -24.29913330078125, 121.41561889648438, 123.50372314453125, 233.32687377929688, 41.64654541015625, 130.08755493164062, 41.8671875, 181.26124572753906, -9.968154907226562, 139.77505493164062, -24.305206298828125, 113.63143920898438, 166.63021850585938, 9.718658447265625, 86.70613861083984, 148.58905029296875, 157.50172424316406, 27.12762451171875, 4.488372802734375, 93.07540893554688, 6.07220458984375, 59.40924072265625, 11.450126647949219, 84.03050994873047, -14.795276641845703, 154.96673583984375, 27.052001953125, 1.015228271484375, 69.118408203125, 6.25823974609375, -144.56002807617188, -89.489990234375, -27.722381591796875, 193.90045166015625, 112.54248046875, 89.64826965332031, -1.5123291015625, -0.7696685791015625, -54.30851364135742, 53.27520751953125, 175.6612548828125, 139.21170043945312, 23.625579833984375, 23.93231201171875, -55.4403076171875, 249.99920654296875, 37.41123962402344, 106.49658203125, -6.5149993896484375, -1.1931610107421875, -32.99462890625, -59.5389404296875, 4.1073455810546875, 134.89389038085938, 75.88525390625, 145.18031311035156, 5.4742584228515625, 136.76510620117188, 124.18819427490234, 130.09925842285156, 162.41546630859375, -7.2029571533203125, -6.335027694702148, 51.29681396484375, 33.138763427734375, 182.36578369140625, -80.85982513427734, 145.829833984375, -96.9677734375, 70.2197265625, 0.0, -77.908203125, 17.15692138671875, 166.91796875, 166.44827270507812, 16.629669189453125, -1.1697235107421875, 179.43707275390625, 93.10842895507812, 16.615310668945312, -124.59257507324219, 13.06439208984375, -3.011627197265625, 56.140167236328125, 148.47918701171875, 0.08887481689453125, 98.41057586669922, 63.118988037109375, 176.00469970703125, 16.119140625, 153.05667114257812, -27.2486572265625, 22.273529052734375, 91.177978515625, -15.762020111083984, -125.38423156738281, 38.955322265625, 1.63153076171875, 92.76736450195312, 116.83430480957031, 113.9410400390625, 38.418548583984375, 98.65847778320312, 24.297805786132812, 131.43511962890625, -10.75469970703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000316.npy"}
{"epoch": 0.6617801047120419, "step": 317, "batch_size": 128, "mean": 50.974266052246094, "std": 76.3931655883789, "min": -145.70205688476562, "p10": -28.460504150390623, "median": 37.72870635986328, "p90": 149.9579360961914, "max": 238.36981201171875, "pos_frac": 0.734375, "sample": [129.7469940185547, -125.85407257080078, 0.0, 7.841148376464844, 191.39215087890625, -27.388999938964844, 159.1409149169922, 84.47212219238281, 10.247421264648438, 164.27679443359375, 13.3116455078125, 148.260986328125, 38.74340057373047, -12.127944946289062, 138.21493530273438, -105.4044189453125, -25.326278686523438, 139.29653930664062, 74.6527099609375, 12.362686157226562, 51.22528076171875, 10.3963623046875, 140.73117065429688, 114.53498077392578, 0.804473876953125, 135.12925720214844, 8.849395751953125, 1.1235218048095703, -7.973541259765625, 27.124237060546875, -3.2650108337402344, -58.190887451171875, 32.33122253417969, -6.225921630859375, -23.681915283203125, 103.45960998535156, 122.2207260131836, 165.70794677734375, 115.35714721679688, -40.94354248046875, -63.87303161621094, 32.67915344238281, 158.80050659179688, -132.45648193359375, -126.52810668945312, 147.42718505859375, 169.40087890625, 148.7906494140625, 17.647769927978516, 0.0, 77.50172424316406, 73.56375122070312, 36.99430847167969, 182.13192749023438, -56.26463317871094, -14.199142456054688, 17.81829833984375, 42.930633544921875, 2.2972869873046875, -3.278961181640625, 19.5572509765625, 153.32275390625, 137.80929565429688, 110.240234375, 0.0, 11.8634033203125, 19.96063232421875, -1.090576171875, 38.5671272277832, 105.39239501953125, 145.55291748046875, -6.393196105957031, -30.743408203125, -145.70205688476562, 164.54949951171875, 69.20252990722656, 151.88739013671875, -74.946044921875, 27.255006790161133, -9.924957275390625, 83.20698547363281, 119.860107421875, 110.39546203613281, -16.864410400390625, 46.19427490234375, 100.83270263671875, 47.40647888183594, -27.48211669921875, 109.33953857421875, 144.4736328125, -2.572052001953125, 8.820770263671875, 158.06442260742188, 11.661026000976562, 7.023712158203125, 0.8780612945556641, -1.7066497802734375, 127.8675537109375, 113.78103637695312, 238.36981201171875, 5.932258605957031, 112.629638671875, 87.25137329101562, 2.2496185302734375, -33.55140686035156, 40.20304870605469, 20.081268310546875, 38.463104248046875, 0.5180549621582031, 66.79641723632812, 108.37427520751953, 5.394172668457031, 132.59967041015625, 7.052223205566406, 96.92031860351562, 115.826904296875, 116.35016632080078, 67.60614013671875, 158.6465301513672, 15.370552062988281, -7.787776947021484, 70.850341796875, -34.144287109375, 149.1310272216797, 54.932342529296875, 121.51322174072266, -3.787364959716797, 127.41644287109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000317.npy"}
{"epoch": 0.6638743455497382, "step": 318, "batch_size": 128, "mean": 53.51047897338867, "std": 73.64446258544922, "min": -158.6318359375, "p10": -26.856939697265624, "median": 53.221527099609375, "p90": 145.30392761230468, "max": 186.9202880859375, "pos_frac": 0.7421875, "sample": [84.96086120605469, 21.9775390625, 137.709228515625, 131.6689453125, 18.472259521484375, -8.559906005859375, 33.8759765625, 21.186920166015625, 144.61557006835938, 44.46821594238281, 1.9327392578125, -22.945602416992188, 146.91009521484375, 29.433351516723633, 169.10601806640625, 14.17822265625, 129.10684204101562, -15.804336547851562, 103.00897979736328, 83.84024047851562, -23.6378173828125, -17.791351318359375, 129.97308349609375, 147.85072326660156, -16.898834228515625, 45.290008544921875, 121.68435668945312, 137.33868408203125, -115.23329162597656, 31.443267822265625, 1.680145263671875, -42.047119140625, 59.69061279296875, 72.00473022460938, 54.35398864746094, -24.847137451171875, 9.108956336975098, 59.05145263671875, 137.65567016601562, 173.93875122070312, 83.37066650390625, 120.29800415039062, 98.185791015625, 60.236724853515625, 34.74595642089844, 6.697906494140625, -135.68572998046875, 110.5995864868164, -2.4051513671875, 105.35159301757812, -149.97479248046875, 14.773815155029297, 89.9132080078125, 161.05645751953125, 119.47299194335938, 44.657928466796875, -21.153785705566406, 118.31159973144531, 140.67706298828125, -16.977294921875, 98.13520812988281, 114.64651489257812, 167.08572387695312, 49.19682312011719, -75.40768432617188, 53.42486572265625, -44.1142578125, 3.217498779296875, 153.23825073242188, -3.3873977661132812, 127.87521362304688, 20.90362548828125, 6.889862060546875, 89.61380004882812, 116.65289306640625, 130.74044799804688, 74.68231201171875, 127.71354675292969, 77.44293212890625, -7.726898193359375, 92.1253433227539, 73.03985595703125, 32.2384033203125, 133.55926513671875, 61.442596435546875, -36.7142333984375, 82.01725769042969, 161.501708984375, -42.37938690185547, -8.39599609375, -8.51629638671875, 52.19921875, 29.0924072265625, 108.324462890625, 53.0181884765625, 49.956817626953125, 95.2149658203125, -56.99214172363281, 119.013916015625, -18.619293212890625, 116.21936798095703, 186.9202880859375, 161.17852783203125, 44.651519775390625, -35.54803466796875, 126.342529296875, 62.8668212890625, 6.4538421630859375, -5.4878997802734375, -28.915924072265625, 152.28131103515625, 32.345367431640625, 57.7030029296875, 150.35723876953125, 37.06707000732422, 139.89932250976562, 56.22417068481445, 141.4361572265625, -25.974517822265625, 110.70693969726562, -128.9219970703125, 33.760955810546875, -1.0171432495117188, 28.850387573242188, 152.75205993652344, -8.03192138671875, -158.6318359375, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000318.npy"}
{"epoch": 0.6659685863874345, "step": 319, "batch_size": 128, "mean": 56.30867385864258, "std": 67.78380584716797, "min": -155.9391326904297, "p10": -21.462710571289062, "median": 42.550689697265625, "p90": 143.89747314453126, "max": 196.72674560546875, "pos_frac": 0.828125, "sample": [108.27507781982422, 3.1786956787109375, 25.16766357421875, 55.62879943847656, 30.2886962890625, 196.72674560546875, 105.1455078125, -43.139892578125, 58.60429382324219, 25.28369140625, -19.756103515625, 12.582719802856445, 106.44481658935547, 158.9373779296875, 106.16082763671875, 151.50860595703125, 7.2413330078125, 0.0, 41.324676513671875, 58.902130126953125, 10.215354919433594, 26.93401336669922, 93.54672241210938, 10.994125366210938, 5.1611328125, 126.00064086914062, 7.616241455078125, 57.3775634765625, -6.74700927734375, 105.28236389160156, -22.748870849609375, 10.02886962890625, 145.2066650390625, 5.135467529296875, 113.91680145263672, 48.46868896484375, 9.002822875976562, 80.4151611328125, 137.15283203125, 61.099090576171875, 4.100835800170898, 137.77975463867188, 42.76007080078125, -105.06961059570312, 1.6383590698242188, 101.1751708984375, 156.17877197265625, 8.59332275390625, -48.233489990234375, 144.32595825195312, 106.61178588867188, 136.04075622558594, 11.793426513671875, 21.449172973632812, 132.4925537109375, 26.39678955078125, 28.240753173828125, -68.39370727539062, 62.073326110839844, 132.2456817626953, -155.9391326904297, 50.69182586669922, 100.85594940185547, -76.43096923828125, 78.23609924316406, -20.9114990234375, 45.638763427734375, 21.974271774291992, 62.766021728515625, 120.4757080078125, 2.5884780883789062, 107.15412139892578, 101.38873291015625, 142.6332550048828, -2.6585235595703125, -13.860321044921875, 116.164306640625, 113.7706298828125, 143.71383666992188, -23.166580200195312, 16.24396514892578, 134.71270751953125, 127.46268463134766, 21.174346923828125, 36.06549072265625, 0.388519287109375, 159.9664306640625, 195.33810424804688, 95.15753173828125, 18.313934326171875, -13.392059326171875, 36.425750732421875, 96.28427124023438, -34.18011474609375, 128.9474334716797, 41.215850830078125, 28.852737426757812, 151.14523315429688, -42.45257568359375, 158.57177734375, 31.80755615234375, 11.70660400390625, 162.228515625, 42.34130859375, -56.600799560546875, 11.3370361328125, 119.41107177734375, 69.87799072265625, 139.0348358154297, -6.309837341308594, 38.76727294921875, 29.958419799804688, -16.943416595458984, 133.97073364257812, -25.784393310546875, 114.4295654296875, 117.45750427246094, 173.3720703125, -34.98113250732422, 31.30230712890625, 137.8092041015625, 12.26104736328125, 81.18946838378906, 0.21523666381835938, 141.29690551757812, 51.5213623046875, 32.53089141845703, 148.21163940429688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000319.npy"}
{"epoch": 0.6680628272251309, "step": 320, "batch_size": 128, "mean": 52.57035827636719, "std": 83.08792114257812, "min": -151.71575927734375, "p10": -46.646270751953125, "median": 53.5046501159668, "p90": 150.66273345947266, "max": 193.748046875, "pos_frac": 0.734375, "sample": [113.16983032226562, 74.87330627441406, -48.13885498046875, -140.05911254882812, 124.32247924804688, 148.35098266601562, -85.7664794921875, 8.142379760742188, 142.87387084960938, 148.41033935546875, 22.123016357421875, 115.25833129882812, 19.55084228515625, 9.4232177734375, 159.84725952148438, 94.76400756835938, -17.25732421875, 0.1759796142578125, 16.798095703125, 28.965057373046875, 16.485671997070312, 136.70875549316406, 38.44720458984375, 2.9275970458984375, 1.8821334838867188, -59.242279052734375, 166.69735717773438, 63.48027038574219, 61.47322082519531, 81.2206802368164, 102.72676086425781, 116.490966796875, 154.03219604492188, 18.460250854492188, 48.157928466796875, 137.05191040039062, -10.3052978515625, 150.81326293945312, 13.05963134765625, -0.5575790405273438, 5.722362518310547, 73.01475524902344, 113.26821899414062, -24.317108154296875, 160.68470764160156, 173.68185424804688, 112.64404296875, 26.3013916015625, 108.2222900390625, 93.40483093261719, -76.91868591308594, 13.92340087890625, -129.06375122070312, -27.039154052734375, 155.8153076171875, -19.097442626953125, 95.76788330078125, -15.910018920898438, 95.04847717285156, 149.93307495117188, 132.88128662109375, 147.60528564453125, 17.626983642578125, 140.22512817382812, 150.46041870117188, -29.9757080078125, 144.88302612304688, -17.48114013671875, 160.29248046875, 137.9935302734375, 23.777618408203125, -6.329498291015625, 26.039276123046875, -9.38372802734375, -151.71575927734375, 94.58995056152344, 10.90106201171875, 145.2608642578125, 9.9835205078125, 128.7287139892578, -127.76641845703125, 36.873016357421875, 7.885248184204102, -2.142181396484375, -17.411285400390625, 62.167762756347656, 95.73765563964844, 140.09649658203125, 89.02584838867188, -1.5077362060546875, -85.36070251464844, 49.303924560546875, 185.30157470703125, -7.0357666015625, 123.59230041503906, -26.980937957763672, 11.214035034179688, 99.785888671875, -26.012527465820312, 158.75143432617188, 84.91729736328125, -46.006591796875, 87.49017333984375, 150.5982208251953, 134.13812255859375, 193.748046875, 128.0665740966797, -19.291244506835938, 97.52334594726562, -118.29129028320312, 48.347442626953125, 65.53579711914062, -133.0587158203125, -4.819610595703125, 159.385009765625, 36.896636962890625, 143.00592041015625, 55.33305358886719, 127.31196594238281, 107.85484313964844, -28.005949020385742, 67.97329711914062, 24.62213134765625, -123.02320098876953, 177.21334838867188, 51.676246643066406, -143.8508758544922, 120.91026306152344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000320.npy"}
{"epoch": 0.6701570680628273, "step": 321, "batch_size": 128, "mean": 45.405372619628906, "std": 75.86480712890625, "min": -156.80581665039062, "p10": -34.710745239257804, "median": 30.993301391601562, "p90": 146.9872848510742, "max": 183.94287109375, "pos_frac": 0.7421875, "sample": [128.33474731445312, -43.80206298828125, 106.41317749023438, -143.79644775390625, 24.962890625, 164.48849487304688, 29.66400146484375, 6.888031005859375, -58.56507873535156, 169.49212646484375, 125.02308654785156, 18.705490112304688, -77.78223419189453, 62.54584503173828, 80.74212646484375, 3.16534423828125, 11.873908996582031, 138.84840393066406, 159.3734130859375, 4.187408447265625, 83.05425262451172, -70.38796997070312, 3.1715660095214844, -9.809814453125, 156.9847412109375, 108.69807434082031, -1.5495452880859375, 6.911895751953125, 103.69900512695312, 0.933868408203125, 122.67752075195312, 94.77952575683594, 102.22714233398438, 13.725448608398438, 38.78033447265625, -50.91107177734375, -29.51727294921875, 89.76687622070312, 5.5393829345703125, 131.35986328125, 37.44132614135742, 33.27813720703125, -6.9589080810546875, 63.719451904296875, 10.579071044921875, 160.74278259277344, -22.587753295898438, 144.2598114013672, 183.94287109375, 111.76571655273438, 36.165435791015625, 21.018508911132812, -19.909626007080078, 41.6644287109375, -1.4630508422851562, 8.890762329101562, 144.83392333984375, -9.040069580078125, 145.49392700195312, -22.25634765625, 57.489654541015625, 96.18270874023438, 34.30089569091797, 131.18307495117188, 67.78219604492188, 152.42007446289062, -32.740814208984375, 62.451812744140625, -63.72875213623047, -6.2041473388671875, -16.529403686523438, 59.05120849609375, -30.063705444335938, 164.71142578125, -150.64157104492188, 114.52850341796875, 162.8141326904297, 179.47409057617188, 32.322601318359375, 120.4639892578125, 19.059295654296875, -30.96337890625, 111.16680908203125, 91.78944396972656, 7.8451690673828125, 139.52993774414062, -20.664031982421875, 150.47178649902344, -90.57998657226562, 124.75558471679688, 16.348548889160156, 21.10882568359375, 107.52069091796875, 0.14862823486328125, 16.188426971435547, 173.87826538085938, 5.03448486328125, 76.39956665039062, 29.114410400390625, 88.38095092773438, 123.31256866455078, -4.378875732421875, 99.23873901367188, 22.971725463867188, -7.96624755859375, 116.5528564453125, 5.563243865966797, 4.491992950439453, 1.16680908203125, 1.4658660888671875, -39.3072509765625, -93.86953735351562, 9.600122451782227, -23.088851928710938, 40.44175720214844, 167.56298828125, -156.80581665039062, 142.78269958496094, 37.804901123046875, 72.45941925048828, 8.980712890625, 58.2642822265625, 5.8130950927734375, -1.9635086059570312, 139.2391357421875, 141.70172119140625, 0.0, -104.42550659179688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000321.npy"}
{"epoch": 0.6722513089005235, "step": 322, "batch_size": 128, "mean": 56.600040435791016, "std": 75.16079711914062, "min": -243.97259521484375, "p10": -30.02119140625, "median": 51.6467342376709, "p90": 152.82752685546873, "max": 191.9039306640625, "pos_frac": 0.75, "sample": [-54.882598876953125, -25.526824951171875, -53.0396728515625, 94.22152709960938, 30.32318115234375, 12.122917175292969, 50.96929931640625, -42.35260009765625, 136.42398071289062, 68.7808837890625, -13.994476318359375, 16.747299194335938, 138.88357543945312, 10.0867919921875, 144.9264373779297, 117.423828125, 35.5440673828125, -8.1805419921875, 112.92221069335938, 63.9185791015625, 52.09674072265625, -17.447235107421875, 163.26856994628906, 2.37652587890625, 0.0490875244140625, 40.837921142578125, -31.37030029296875, 94.87196350097656, 7.24456787109375, 122.24569702148438, 81.65545654296875, 131.5553436279297, 1.1134109497070312, -10.882720947265625, 126.98249816894531, 33.73638916015625, 32.07115936279297, 137.01885986328125, -28.521774291992188, 123.1463851928711, 137.54669189453125, 116.5048828125, -29.834945678710938, 16.14794921875, -5.463623046875, 19.7138671875, 112.98107147216797, -30.455764770507812, 97.98345947265625, -126.58113098144531, 124.83195495605469, 4.569305419921875, 139.5844268798828, 182.28622436523438, -8.94085693359375, 133.33648681640625, 118.22554016113281, -23.54437255859375, 104.12986755371094, 119.65560913085938, 158.54638671875, 58.08439636230469, 185.71136474609375, 191.9039306640625, -43.9132080078125, 123.42919158935547, 86.05082702636719, 35.3250732421875, 113.45533752441406, 13.465530395507812, 170.479248046875, 150.2720947265625, 112.19854736328125, 57.619232177734375, 10.058929443359375, 0.0, -59.4527587890625, 46.15155029296875, 62.89019775390625, -34.30366516113281, 154.0970458984375, 51.19672775268555, 134.1998291015625, 152.283447265625, -19.60028076171875, 54.539642333984375, 65.61923217773438, -10.235595703125, 3.9903564453125, 118.14949035644531, 164.88528442382812, -4.222259521484375, 29.254119873046875, -41.0050048828125, -12.4454345703125, 0.758209228515625, 93.71688842773438, 3.1020660400390625, -243.97259521484375, 74.63592529296875, 138.87486267089844, 64.89007568359375, 53.41064453125, 167.70257568359375, 127.81536865234375, -36.51653289794922, -2.0083770751953125, 188.2360076904297, 181.49081420898438, 6.4387664794921875, -12.716888427734375, 123.60823059082031, 23.928848266601562, 45.77497863769531, -4.752716064453125, 10.705860137939453, -35.20851135253906, 18.817230224609375, 20.31085205078125, 110.33833312988281, 88.58566284179688, 176.15731811523438, 94.7294921875, -5.76177978515625, 18.56756591796875, 144.31988525390625, 66.6181411743164, 163.4857940673828], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000322.npy"}
{"epoch": 0.6743455497382199, "step": 323, "batch_size": 128, "mean": 36.25709533691406, "std": 84.46714782714844, "min": -213.54241943359375, "p10": -67.84081268310545, "median": 22.316802978515625, "p90": 143.33808288574218, "max": 183.09933471679688, "pos_frac": 0.6953125, "sample": [14.268699645996094, 12.374420166015625, -129.7339324951172, -114.61224365234375, 127.38255310058594, 136.5218963623047, -63.90196228027344, 73.27449035644531, 16.957763671875, 180.8890838623047, -95.73298645019531, 131.44833374023438, 26.68212890625, -18.42083740234375, 16.157379150390625, 109.6396255493164, 150.0431671142578, -8.557449340820312, 70.70305633544922, -30.70843505859375, 179.05279541015625, 103.91427612304688, 22.297607421875, 122.7802734375, 23.17327880859375, 162.8499298095703, 14.77593994140625, 130.68014526367188, -88.114990234375, 142.40377807617188, -17.033782958984375, -180.6114501953125, -6.426811218261719, 9.82073974609375, 0.2233448028564453, 105.2911376953125, 20.9365234375, 24.30908203125, 80.0555419921875, 18.8625431060791, 26.370681762695312, 124.95343017578125, -28.90655517578125, 41.6937255859375, 132.04055786132812, -6.637176513671875, -19.258255004882812, -213.54241943359375, 148.13885498046875, 31.845962524414062, 7.094606399536133, 7.380481719970703, -64.55340576171875, 9.050605773925781, 52.55809020996094, 77.19467163085938, 49.17071533203125, 140.43505859375, 182.6610870361328, -25.955795288085938, 77.85284423828125, 83.30889892578125, 116.38910675048828, -21.71112060546875, 154.61489868164062, 22.33599853515625, 128.93106079101562, -16.527206420898438, -47.9342041015625, 86.78057861328125, 24.5806884765625, 2.815896987915039, -61.642852783203125, 19.58734130859375, 124.55158996582031, 67.84135437011719, 171.58969116210938, -75.51142883300781, 163.7713165283203, 2.937835693359375, 122.10272979736328, -128.91490173339844, 3.4270057678222656, 33.65838623046875, 20.657867431640625, 171.20355224609375, -48.5394287109375, 142.15838623046875, -134.92782592773438, 3.5589256286621094, 135.3658447265625, 57.963096618652344, -45.904571533203125, 110.52657318115234, -3.4300384521484375, -9.048412322998047, -94.03951263427734, -14.947463989257812, -30.504425048828125, -6.65203857421875, 20.52008056640625, -98.68344116210938, -11.082916259765625, 2.6182098388671875, 53.589874267578125, 49.034515380859375, 21.17580795288086, 2.2483367919921875, 131.1146240234375, 169.60833740234375, -2.7747764587402344, 64.68498229980469, 138.76927185058594, 134.0835418701172, -126.55152130126953, -123.56884765625, 75.50425720214844, 23.866546630859375, 108.87821960449219, 24.53516387939453, 21.55698013305664, -7.11968994140625, 106.60964965820312, 145.51812744140625, 112.82601928710938, -33.97991943359375, 0.9023780822753906, 183.09933471679688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000323.npy"}
{"epoch": 0.6764397905759162, "step": 324, "batch_size": 128, "mean": 60.54029083251953, "std": 73.45747375488281, "min": -109.32551574707031, "p10": -14.323461914062497, "median": 40.614370346069336, "p90": 154.81002044677734, "max": 199.703857421875, "pos_frac": 0.78125, "sample": [73.11500549316406, 156.92779541015625, 15.49261474609375, 127.38427734375, 24.42388916015625, 120.37646484375, 3.487823486328125, 13.566408157348633, 132.68634033203125, 33.48371887207031, 33.83349609375, 0.104461669921875, -47.8482666015625, 161.3011474609375, 6.5753173828125, 163.12356567382812, 124.88062286376953, 21.11162567138672, -13.44378662109375, 0.0, 2.95001220703125, 7.440956115722656, 128.43453979492188, -3.60711669921875, 4.087429046630859, 40.495697021484375, 134.85977172851562, -64.68356323242188, 128.36349487304688, -72.79244995117188, 6.936223983764648, 120.87139129638672, 145.4471893310547, 153.92877197265625, 128.6474609375, 125.36209869384766, -16.37603759765625, 52.594940185546875, 84.40496826171875, 0.0, -6.955596923828125, 131.3321533203125, 156.5653076171875, 112.34925842285156, 45.458465576171875, 129.92282104492188, 0.0, -0.72021484375, 110.1673583984375, -8.476226806640625, -76.46524047851562, 41.59465026855469, 47.36384582519531, 21.9766845703125, 107.65301513671875, 94.36483764648438, 9.40570068359375, 129.60760498046875, 159.21534729003906, 9.59344482421875, -90.96495819091797, 167.97323608398438, 10.214004516601562, 126.59130859375, 108.27088928222656, -4.914234161376953, 123.38285827636719, 130.02737426757812, 15.479156494140625, -99.8427734375, 154.05775451660156, 116.68070220947266, -109.32551574707031, 127.79771423339844, -2.120391845703125, -4.665802001953125, 31.113548278808594, 65.865478515625, 8.876022338867188, 199.703857421875, 182.07769775390625, 7.003826141357422, -23.7149658203125, 88.61746215820312, 178.34967041015625, 147.93536376953125, 19.816497802734375, 21.1392822265625, -33.926971435546875, 4.417755126953125, 135.2916259765625, 40.691741943359375, 12.041923522949219, -7.14422607421875, -0.721221923828125, 55.413330078125, 40.5369987487793, 157.6298828125, 132.16159057617188, 1.95928955078125, -40.77245330810547, -4.963754653930664, 7.6259765625, 34.369781494140625, -47.130859375, 188.6746826171875, 178.26580810546875, 156.74697875976562, 39.62849426269531, 5.558319091796875, 147.70098876953125, 109.33879089355469, -16.625465393066406, 124.75559997558594, 43.47760009765625, 129.4050750732422, 149.5872802734375, 129.19070434570312, -0.874725341796875, 147.1336669921875, 132.61859130859375, 1.5758705139160156, 132.84625244140625, 139.4562530517578, 139.35438537597656, 10.349655151367188, 3.06390380859375, 3.1211700439453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000324.npy"}
{"epoch": 0.6785340314136126, "step": 325, "batch_size": 128, "mean": 51.84834671020508, "std": 89.31425476074219, "min": -143.74063110351562, "p10": -46.160719299316376, "median": 30.55339813232422, "p90": 166.1050567626953, "max": 252.260986328125, "pos_frac": 0.65625, "sample": [74.41390991210938, 143.91122436523438, -115.0401611328125, 18.23779296875, 108.713623046875, 24.51641845703125, 57.74700927734375, 147.6397705078125, 34.20573425292969, -1.729095458984375, -0.320068359375, 10.04998779296875, 138.69906616210938, 141.84756469726562, -4.589508056640625, 4.991024017333984, 157.45620727539062, 62.378173828125, 88.54010009765625, 172.96176147460938, 139.19424438476562, 170.76185607910156, 92.62152099609375, -4.14442253112793, 132.48123168945312, 12.018394470214844, 12.451087951660156, 77.72103881835938, 126.93827819824219, 20.917755126953125, -104.62074279785156, 7.068046569824219, 86.84761047363281, 158.7709197998047, 164.00030517578125, 91.35624694824219, 21.548233032226562, 163.3155059814453, 29.443695068359375, -86.10789489746094, 122.46075439453125, -143.74063110351562, 153.63951110839844, -0.7108612060546875, -66.20606994628906, 0.275970458984375, 252.260986328125, -126.79524230957031, 182.42405700683594, -7.385406494140625, -1.1408843994140625, 99.594482421875, 160.8115234375, 10.92340087890625, 7.8406524658203125, -95.9357681274414, -8.44024658203125, -115.65111541748047, -7.945587158203125, -70.94052124023438, 64.57449340820312, 135.53619384765625, -5.053367614746094, -3.836029052734375, 94.38246154785156, 60.0235595703125, -15.021499633789062, -117.82611083984375, 191.74090576171875, 28.228424072265625, 114.11056518554688, 22.28997802734375, 239.3284912109375, -12.260467529296875, -9.1446533203125, -0.4007110595703125, -14.826141357421875, -16.08916473388672, 103.3692626953125, 147.44134521484375, -130.93869018554688, -11.982330322265625, -36.85260009765625, 171.0211181640625, -71.01106262207031, 98.05657958984375, 124.86468505859375, 40.10498046875, 45.86885070800781, -24.00128173828125, -15.084442138671875, 123.06546020507812, 169.58523559570312, -30.253646850585938, 2.4828548431396484, -125.14321899414062, -25.253746032714844, 164.9061279296875, 65.7828369140625, 134.767578125, 168.90255737304688, 33.38606262207031, 148.1147003173828, 198.25799560546875, 155.2193603515625, -37.569854736328125, 5.461395263671875, 21.010215759277344, 31.663101196289062, -0.53607177734375, -20.48455810546875, 148.50360107421875, 55.47563171386719, 135.02139282226562, 7.189872741699219, 124.6399917602539, 129.59396362304688, 185.81488037109375, 14.699432373046875, -0.8275146484375, -35.87983703613281, -31.013885498046875, 202.1884765625, -21.65869140625, -21.954544067382812, 202.05142211914062, 91.78915405273438, 148.4249267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000325.npy"}
{"epoch": 0.680628272251309, "step": 326, "batch_size": 128, "mean": 50.866722106933594, "std": 84.62596130371094, "min": -201.91961669921875, "p10": -68.15116729736327, "median": 57.9500732421875, "p90": 147.49191894531248, "max": 178.27198791503906, "pos_frac": 0.7109375, "sample": [141.08026123046875, 129.46084594726562, 125.74702453613281, 173.13768005371094, 119.45832824707031, 62.8646240234375, -42.69830322265625, -73.75112915039062, 132.27499389648438, -105.32115173339844, 145.09555053710938, -15.04498291015625, 136.17332458496094, 141.5375213623047, 22.237266540527344, 3.099925994873047, 5.313531875610352, 150.47357177734375, 178.27198791503906, 29.25189208984375, 57.514495849609375, 137.8963623046875, 5.5224761962890625, -32.69110107421875, 63.357574462890625, 103.45779418945312, 120.3719482421875, 94.7777099609375, 83.79985046386719, 16.62287139892578, -120.91741180419922, 149.60879516601562, 0.4232177734375, 60.67730712890625, -109.31289672851562, -106.29119873046875, 40.53656768798828, 92.4576187133789, -31.28009033203125, 139.6572265625, 89.0732421875, 110.67843627929688, -18.743194580078125, 134.85494995117188, 40.806427001953125, 98.98968505859375, -65.97352600097656, 135.44586181640625, -31.281494140625, 163.9154815673828, 94.899169921875, 109.14053344726562, 142.58724975585938, 17.42481231689453, 144.00286865234375, -123.50867462158203, -36.13800048828125, 24.4810791015625, 127.72479248046875, -10.498592376708984, 126.85917663574219, 113.8500747680664, 20.69390869140625, 43.29034423828125, 0.0, -201.91961669921875, 14.341983795166016, -0.89404296875, 131.51443481445312, -51.251800537109375, 14.978134155273438, 162.425537109375, 173.88601684570312, 83.63973999023438, 53.85919952392578, -52.6785888671875, 33.4468994140625, -73.23233032226562, 31.913101196289062, 124.03564453125, -2.952545166015625, 157.93475341796875, -5.392303466796875, 2.70703125, 11.69647216796875, 137.5714111328125, -32.99127197265625, -105.19058227539062, 109.61393737792969, 11.335609436035156, 135.2391357421875, 26.113922119140625, 164.00311279296875, -8.974029541015625, 84.35000610351562, -10.53729248046875, 85.60675811767578, 151.01824951171875, 24.2265625, 127.47329711914062, 155.37014770507812, 0.0, 133.44049072265625, 73.3831787109375, 59.41644287109375, 130.61962890625, -78.45875549316406, 144.16629028320312, 99.22946166992188, 0.0, 129.30111694335938, -142.88034057617188, -24.67401123046875, -93.54351043701172, 21.523452758789062, -112.60421752929688, 153.44210815429688, -8.11822509765625, 146.58468627929688, 42.85029220581055, 70.1445083618164, 26.473968505859375, -35.650390625, -41.99444580078125, 133.73106384277344, 151.79345703125, 58.385650634765625, 134.665283203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000326.npy"}
{"epoch": 0.6827225130890052, "step": 327, "batch_size": 128, "mean": 59.47547912597656, "std": 83.12676239013672, "min": -154.9361572265625, "p10": -25.550640106201172, "median": 41.74488067626953, "p90": 165.6436737060547, "max": 213.92877197265625, "pos_frac": 0.8046875, "sample": [34.67259979248047, 113.27339172363281, 121.99406433105469, 16.424461364746094, 194.41375732421875, -145.58993530273438, 129.1395263671875, 140.7344970703125, 13.160110473632812, 115.03939819335938, 4.960058212280273, 180.07943725585938, -6.842529296875, -136.8321533203125, 101.78031158447266, -19.51953125, -36.579315185546875, 11.32183837890625, 171.37362670898438, 166.822998046875, 119.00171661376953, 80.73772430419922, 19.84619140625, 48.10377502441406, 125.47976684570312, -118.99154663085938, 11.279769897460938, 109.75868225097656, 15.097137451171875, 127.74602508544922, 44.257537841796875, 176.61474609375, -12.539453506469727, 0.6152420043945312, 43.43064880371094, 113.89138793945312, 166.08499145507812, 8.814498901367188, 213.92877197265625, 108.76092529296875, -28.18304443359375, -10.868902206420898, 1.15008544921875, 114.23924255371094, 32.94452667236328, -13.779266357421875, 166.5078125, 37.372650146484375, 116.02435302734375, 11.618133544921875, 149.41415405273438, 6.452606201171875, -9.099761962890625, -154.9361572265625, 121.90375518798828, 17.27838134765625, 137.33197021484375, 121.60345458984375, 176.65985107421875, 124.7528076171875, 18.464706420898438, 137.71624755859375, 113.02568054199219, 22.281982421875, 164.75692749023438, 145.59375, 172.35821533203125, 73.91995239257812, 81.20654296875, -22.94720458984375, 39.139801025390625, 3.43487548828125, -11.89068603515625, -61.178138732910156, 103.3095703125, 19.95403289794922, 30.036102294921875, 61.9268798828125, 169.63394165039062, 8.41571044921875, -25.789337158203125, 3.739898681640625, 27.8209228515625, 118.3182373046875, 17.791229248046875, 165.57351684570312, 16.189163208007812, 0.0, 149.61248779296875, 14.765220642089844, 28.127166748046875, 157.093017578125, 186.52963256835938, 22.80035400390625, 135.16441345214844, 40.059112548828125, 140.5933837890625, 32.80186462402344, 139.0369415283203, 145.36569213867188, 71.37490844726562, 1.5364532470703125, -145.05419921875, -25.448341369628906, 94.66827392578125, 116.5032730102539, -100.98870849609375, 8.713897705078125, 85.9559326171875, 132.43580627441406, -127.04341888427734, 107.14407348632812, 165.807373046875, 8.440216064453125, -38.546112060546875, 133.95028686523438, -4.946533203125, 161.07769775390625, 1.3324871063232422, -6.0859527587890625, 152.48086547851562, 148.39315795898438, 136.83935546875, 6.2537384033203125, -111.55728149414062, 140.22332763671875, 23.459915161132812, 1.057281494140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000327.npy"}
{"epoch": 0.6848167539267016, "step": 328, "batch_size": 128, "mean": 54.003849029541016, "std": 76.77871704101562, "min": -143.1009521484375, "p10": -24.131689453125, "median": 36.007896423339844, "p90": 152.94262390136717, "max": 214.10040283203125, "pos_frac": 0.7578125, "sample": [30.63037109375, 121.4737548828125, 145.48655700683594, 36.69512939453125, 178.67364501953125, 34.5489501953125, -2.37689208984375, 126.66999816894531, 101.34823608398438, 117.60408020019531, 105.04298400878906, 89.492431640625, 5.122596740722656, 115.94970703125, 50.06438446044922, 0.31894683837890625, 11.659912109375, -9.433197021484375, 47.651123046875, 12.280776977539062, 87.38862609863281, 112.50340270996094, 90.77212524414062, -25.663978576660156, 180.79647827148438, 33.684783935546875, -3.07159423828125, 42.1298828125, 97.5875244140625, 119.63435363769531, -4.61419677734375, 121.45471954345703, 35.32066345214844, 157.96054077148438, 192.68292236328125, 100.46492767333984, 167.82371520996094, 33.24327850341797, 172.05709838867188, -0.066497802734375, 6.240966796875, 138.7404022216797, 86.12240600585938, -0.4830780029296875, 7.619415283203125, -7.290742874145508, -22.89013671875, 46.180908203125, -23.69586181640625, 139.4173583984375, 37.4901123046875, 9.7437744140625, -27.362945556640625, 165.08929443359375, 17.40838623046875, 33.60302734375, 10.17462158203125, 10.62969970703125, 58.12158966064453, -30.773292541503906, 29.84935760498047, 91.00482177734375, 156.40914916992188, 151.45697021484375, 127.23677062988281, 158.755126953125, 122.48310089111328, -72.179931640625, 135.93707275390625, -102.49978637695312, 14.837509155273438, 106.8970947265625, 122.63188934326172, 130.1482391357422, 28.809906005859375, 199.83050537109375, 177.143798828125, 21.1539306640625, 0.31622314453125, 123.14875793457031, -141.6796875, 8.887313842773438, 142.23971557617188, 40.158172607421875, -110.72146606445312, 7.506256103515625, 124.42178344726562, 125.96842193603516, 122.50833129882812, -21.0504150390625, -11.259246826171875, 169.1361083984375, 8.581523895263672, -1.9461822509765625, 5.37091064453125, -20.839828491210938, -5.5026092529296875, 21.575897216796875, -118.89630126953125, 80.0772476196289, 61.38038635253906, 35.287384033203125, 118.16822052001953, 1.751800537109375, -2.0175933837890625, 140.89907836914062, 143.9228515625, -143.1009521484375, -2.01080322265625, -25.14862060546875, 41.801483154296875, 20.970367431640625, 142.40924072265625, -30.732177734375, 27.852264404296875, 24.32220458984375, 0.3014984130859375, 108.60992431640625, -5.156944274902344, -136.4974822998047, -0.8443450927734375, 145.627197265625, 42.84260559082031, -56.750885009765625, 65.45263671875, 214.10040283203125, 123.70417785644531, 22.36407470703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000328.npy"}
{"epoch": 0.6869109947643979, "step": 329, "batch_size": 128, "mean": 45.39289855957031, "std": 88.98248291015625, "min": -172.125244140625, "p10": -49.86482543945312, "median": 25.57459259033203, "p90": 163.07624206542968, "max": 209.8957977294922, "pos_frac": 0.6796875, "sample": [-36.59515380859375, 7.4326324462890625, -4.8929595947265625, 41.97381591796875, 2.1099395751953125, -7.353084564208984, 7.2074737548828125, 27.045578002929688, 3.5901336669921875, 29.917083740234375, 142.33241271972656, -16.875144958496094, -157.4412841796875, 108.8804931640625, 86.54829406738281, 178.12091064453125, 8.82281494140625, -51.841552734375, 143.98719787597656, 109.21312713623047, 122.60446166992188, 19.3953857421875, 148.21267700195312, -15.54144287109375, 125.72596740722656, 153.6915740966797, -135.72122192382812, 120.75302124023438, 172.8124237060547, 89.32171630859375, 2.248199462890625, 22.05902099609375, -4.23046875, 156.09283447265625, -40.204132080078125, 41.707855224609375, 103.72030639648438, 113.9024658203125, 23.210433959960938, 98.25799560546875, -5.587646484375, 169.429931640625, 0.006378173828125, -51.24017333984375, 119.23492431640625, -49.275390625, -14.747756958007812, 152.66436767578125, 16.7825927734375, -172.125244140625, -4.922676086425781, 170.411865234375, -23.700714111328125, -20.5782470703125, 103.18377685546875, 107.87846374511719, 101.49710083007812, 29.598190307617188, 30.795822143554688, -140.01304626464844, 7.879642486572266, 149.34417724609375, -2.43463134765625, -99.26670837402344, -0.18891143798828125, -4.8458251953125, -0.5995674133300781, -16.88336181640625, 199.25286865234375, 15.501182556152344, 32.47715759277344, 0.51580810546875, 139.62286376953125, 208.606689453125, 16.705942153930664, 81.33587646484375, 63.485137939453125, -23.455596923828125, -110.2632827758789, 148.45159912109375, 135.43064880371094, -11.402496337890625, -166.41140747070312, 111.98747253417969, -10.9661865234375, 171.86509704589844, 26.464431762695312, 6.22412109375, 162.36477661132812, 2.187347412109375, 184.5235595703125, 1.0986251831054688, -34.28572082519531, 171.10427856445312, 130.06484985351562, -121.529052734375, 209.8957977294922, 164.736328125, -2.7355194091796875, 107.03961181640625, -5.1222381591796875, 13.794158935546875, 110.11007690429688, 48.934600830078125, 0.26665496826171875, 2.0387420654296875, 207.1768798828125, 99.210205078125, 140.8567352294922, 208.2225341796875, -16.063720703125, 104.94622802734375, 129.3260040283203, -6.799835205078125, 36.3343505859375, -128.58056640625, 24.68475341796875, -21.96453857421875, 67.73515319824219, -10.509979248046875, 75.70745086669922, 131.84719848632812, 1.3110065460205078, 124.98895263671875, 135.8101806640625, 46.261962890625, -61.15216064453125, -151.47207641601562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000329.npy"}
{"epoch": 0.6890052356020943, "step": 330, "batch_size": 128, "mean": 38.37641906738281, "std": 83.41674041748047, "min": -132.93170166015625, "p10": -67.40233306884765, "median": 24.459583282470703, "p90": 154.4720916748047, "max": 251.30242919921875, "pos_frac": 0.7109375, "sample": [-14.386810302734375, 137.4904327392578, 20.069183349609375, -106.82235717773438, 24.026145935058594, 101.56610107421875, 5.024383544921875, 174.62692260742188, 131.29893493652344, 66.64697265625, -56.52520751953125, 49.578277587890625, -10.330902099609375, -132.93170166015625, 102.03692626953125, -45.72380065917969, 155.06329345703125, 28.856109619140625, 17.22174072265625, 0.681854248046875, 147.78262329101562, -124.33473205566406, -32.56703186035156, 7.862060546875, 2.1676025390625, 114.25811767578125, 13.7432861328125, -64.6169204711914, 8.528671264648438, -4.6307373046875, 60.15287780761719, 22.490509033203125, 171.12289428710938, 109.69084167480469, 22.236244201660156, -6.650665283203125, -1.88824462890625, -69.51197814941406, -63.955963134765625, 26.41497802734375, 44.6640625, 168.26251220703125, 36.74552917480469, 41.232635498046875, 29.04351806640625, 112.18319702148438, 17.971485137939453, 187.85293579101562, 13.436576843261719, 92.75428771972656, 4.749485015869141, 154.21871948242188, 84.52749633789062, -122.51614379882812, 4.5831298828125, 24.893020629882812, -121.9983139038086, 35.732666015625, 18.73858642578125, 91.9359130859375, 13.395858764648438, 168.275634765625, 12.1998291015625, 90.59796142578125, -123.29158020019531, 2.380645751953125, 132.17247009277344, -72.6080322265625, 6.458641052246094, 13.166641235351562, 43.235504150390625, -122.55943298339844, 175.02777099609375, 105.98257446289062, -4.33819580078125, 139.34225463867188, -128.27528381347656, 174.60629272460938, -21.514678955078125, 3.333953857421875, 41.489990234375, 140.3231658935547, -8.037567138671875, -13.300689697265625, 26.65032958984375, 3.9863433837890625, 33.93670654296875, 151.74432373046875, 157.78326416015625, 109.58666229248047, 173.33885192871094, 11.8055419921875, 103.74280548095703, 42.15380859375, 12.821533203125, -34.08641052246094, 16.944671630859375, 27.880142211914062, 106.54645538330078, 135.64239501953125, 174.0618896484375, 34.34600830078125, -123.0362548828125, 93.44029998779297, -116.73513793945312, 146.9940948486328, -16.121337890625, 23.513015747070312, -42.86730194091797, 74.83938598632812, 142.72683715820312, -105.34648895263672, -6.901130676269531, -66.49819946289062, -2.26531982421875, 182.1131591796875, 37.86970520019531, 90.56230163574219, -1.99859619140625, 135.44261169433594, -21.75689697265625, 251.30242919921875, 46.73722839355469, -8.12799072265625, -25.8050537109375, 136.64990234375, 92.60636138916016, 33.12237548828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000330.npy"}
{"epoch": 0.6910994764397905, "step": 331, "batch_size": 128, "mean": 48.05012130737305, "std": 79.80277252197266, "min": -169.733642578125, "p10": -41.500363159179685, "median": 29.114044189453125, "p90": 162.01009826660155, "max": 191.266357421875, "pos_frac": 0.75, "sample": [4.241081237792969, 183.865234375, 164.62066650390625, -99.3896484375, 191.266357421875, 44.73858642578125, 33.732147216796875, -8.061126708984375, 44.08038330078125, 98.09866333007812, -4.582916259765625, -44.212738037109375, 68.68766784667969, -94.7125244140625, 181.95550537109375, -26.35162353515625, 6.040863037109375, 4.101169586181641, -9.894317626953125, -45.21148681640625, 13.009597778320312, 46.37188720703125, -27.2939453125, 37.66886901855469, 24.592647552490234, 154.03756713867188, 133.25726318359375, 69.55181884765625, 25.301727294921875, 71.85145568847656, 21.1961669921875, 43.1309814453125, -8.7508544921875, 67.399169921875, 59.070587158203125, 162.89883422851562, 3.8615875244140625, 98.24676513671875, 169.19677734375, 126.65331268310547, -55.214752197265625, 14.5611572265625, -2.7275161743164062, -158.8347625732422, 4.665166854858398, 9.011348724365234, 157.70106506347656, 105.987548828125, 86.53460693359375, 0.0, 55.496490478515625, -0.1141357421875, 126.19381713867188, -1.49481201171875, 8.83673095703125, 70.95243835449219, 37.92034912109375, 144.37171936035156, 38.962249755859375, 55.69915771484375, 24.06195068359375, 50.7088623046875, -41.04449462890625, -17.931930541992188, 178.8741455078125, 82.98926544189453, 161.62921142578125, 124.86422729492188, -38.28663635253906, 79.76499938964844, 6.007232666015625, 5.26654052734375, 150.29696655273438, -18.467620849609375, 41.2043342590332, 23.160701751708984, -47.39433288574219, 4.257080078125, 9.326713562011719, -87.10118865966797, 65.763916015625, 30.8201904296875, 120.87698364257812, 27.40789794921875, -169.733642578125, 186.928466796875, 20.948974609375, 153.81362915039062, -42.564056396484375, 26.158660888671875, 178.33514404296875, 98.23818969726562, 111.9758071899414, 179.83871459960938, 22.261428833007812, 164.76174926757812, 122.91331481933594, 9.84765625, -17.5653076171875, -6.716033935546875, -63.18414306640625, -16.25634765625, 150.45005798339844, 131.78204345703125, -139.72308349609375, 23.496109008789062, 188.37115478515625, 2.064817428588867, 133.72348022460938, -1.5699596405029297, 3.5122222900390625, 17.8714599609375, 140.2745361328125, 11.415130615234375, 118.04082489013672, 128.8643798828125, 14.58807373046875, 67.0419921875, 17.369503021240234, 129.61029052734375, -146.45706176757812, 23.096473693847656, 178.33975219726562, 137.6017608642578, 114.7633056640625, -31.641937255859375, 135.29873657226562, 22.401954650878906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000331.npy"}
{"epoch": 0.6931937172774869, "step": 332, "batch_size": 128, "mean": 59.89073181152344, "std": 86.75011444091797, "min": -178.87428283691406, "p10": -41.54210205078125, "median": 75.35160827636719, "p90": 159.04566802978513, "max": 231.75503540039062, "pos_frac": 0.7578125, "sample": [97.93269348144531, 9.85302734375, 17.85894012451172, 184.36090087890625, 125.07841491699219, 134.0445556640625, 87.92835998535156, 51.92780303955078, 190.346435546875, 132.276611328125, -2.3370819091796875, 8.926355361938477, 6.8096466064453125, -7.9076080322265625, 99.54133605957031, 106.68414306640625, -36.06097412109375, 62.44471740722656, 144.14453125, -32.00782775878906, 78.18637084960938, -5.374053955078125, 39.325927734375, 106.68772888183594, 1.5230178833007812, 29.481353759765625, -115.6845703125, 5.33221435546875, 13.761627197265625, 10.534072875976562, 135.45562744140625, 166.760498046875, 3.4375, 2.97186279296875, 154.3253936767578, 118.97750091552734, 137.595458984375, 169.3814697265625, 91.51846313476562, 105.881591796875, 139.44741821289062, 4.87078857421875, 121.27978515625, -0.7505569458007812, 138.15963745117188, 82.71844482421875, 170.21646118164062, 140.96617126464844, -3.6134185791015625, 24.94805145263672, 148.325439453125, 179.5484619140625, -108.0987548828125, -13.41217041015625, 175.09115600585938, 129.18045043945312, -134.31570434570312, 72.516845703125, 148.36468505859375, -178.77694702148438, 25.948814392089844, 157.30532836914062, 139.00274658203125, -1.481039047241211, 13.247795104980469, 8.382659912109375, 113.6043930053711, 20.663330078125, 13.650630950927734, -41.966827392578125, -6.640289306640625, 146.052978515625, 102.05271911621094, 20.217819213867188, -5.5999908447265625, 169.39816284179688, 111.1146240234375, 126.88946533203125, -41.360076904296875, 105.569580078125, -74.93504333496094, 121.53158569335938, -24.218551635742188, 3.26043701171875, 156.345947265625, -102.30197143554688, 95.48770141601562, 21.614166259765625, 6.8255615234375, 116.78353881835938, 152.65567016601562, 114.14736938476562, -97.03331756591797, 144.2841796875, 106.768798828125, 9.4281005859375, 125.71875, -35.584991455078125, -13.276870727539062, 92.872802734375, 162.90896606445312, 126.7960433959961, -119.70294189453125, -46.02880859375, 231.75503540039062, 187.14447021484375, 128.6372528076172, 18.268035888671875, 10.684158325195312, 38.336280822753906, 157.3899688720703, 126.307861328125, 110.9204330444336, 179.85079956054688, 53.95599365234375, 132.34539794921875, -60.22834777832031, 146.93038940429688, -11.340839385986328, 119.390869140625, 122.84927368164062, 173.89654541015625, -18.67150115966797, -108.39976501464844, 0.0, 11.057037353515625, -178.87428283691406, 4.8187255859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000332.npy"}
{"epoch": 0.6952879581151833, "step": 333, "batch_size": 128, "mean": 44.022254943847656, "std": 76.59392547607422, "min": -149.35794067382812, "p10": -45.083348083496084, "median": 37.46599197387695, "p90": 142.79199829101563, "max": 174.9945068359375, "pos_frac": 0.71875, "sample": [112.27336883544922, -149.35794067382812, -27.156200408935547, 158.63626098632812, 171.2469482421875, -42.80120849609375, 84.53775024414062, 99.027099609375, 106.09225463867188, 63.88104248046875, -50.40834045410156, 10.7769775390625, -16.841888427734375, 145.62860107421875, 139.8505859375, 91.83795166015625, 134.58462524414062, 46.248992919921875, 95.52117919921875, 116.65728759765625, -18.414413452148438, 46.42255401611328, 8.053131103515625, 145.147216796875, 136.05101013183594, 80.4112777709961, 55.03057861328125, 30.164794921875, 133.202880859375, 44.39422607421875, 9.582855224609375, 168.46409606933594, -13.0162353515625, 69.57745361328125, -140.90060424804688, 5.248847961425781, 52.73162841796875, -11.90777587890625, 136.35906982421875, 128.51754760742188, 65.81436157226562, 32.89994812011719, -9.704071044921875, 12.750701904296875, 105.10116577148438, 136.6629638671875, 4.133180618286133, 164.47518920898438, 23.7742919921875, -85.78657531738281, -61.087646484375, 76.83285522460938, 0.0, -5.980682373046875, 136.96926879882812, 7.33612060546875, 148.35687255859375, 161.11676025390625, 142.61956787109375, 51.09576416015625, -2.495025634765625, 94.80264282226562, -61.926055908203125, 135.13986206054688, 87.5889892578125, -35.32569885253906, 115.60537719726562, 174.9945068359375, 88.30177307128906, 32.1326904296875, -114.31715393066406, 78.00900268554688, 54.4893798828125, -66.00567626953125, 5.894340515136719, 122.03471374511719, -1.66729736328125, -2.80072021484375, 17.000747680664062, 123.85263061523438, 31.242149353027344, 47.84657287597656, 44.51246643066406, 158.0552978515625, -111.23165893554688, 58.3980712890625, 146.4215087890625, -9.44091796875, -40.77783203125, 146.41563415527344, 42.73019790649414, 114.73159790039062, -7.997489929199219, 107.73187255859375, -29.644989013671875, -17.975830078125, 25.87384033203125, 27.592529296875, -125.60130310058594, 113.57048034667969, -12.840911865234375, 122.12181854248047, -127.51669311523438, -16.883010864257812, 103.52435302734375, 131.6578826904297, 0.97235107421875, 43.692779541015625, 143.1943359375, 10.042510986328125, 122.53125762939453, 3.715452194213867, 124.68545532226562, 34.63531494140625, 40.188026428222656, -7.7119140625, -125.49922180175781, 15.848731994628906, 16.82965087890625, 26.438751220703125, -9.581817626953125, 18.73406982421875, -112.90495300292969, 14.053108215332031, -10.861587524414062, 34.74395751953125, 18.2974853515625, 2.244647979736328], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000333.npy"}
{"epoch": 0.6973821989528796, "step": 334, "batch_size": 128, "mean": 52.2901496887207, "std": 86.66947174072266, "min": -186.85093688964844, "p10": -35.377753448486324, "median": 40.97334289550781, "p90": 163.4613815307617, "max": 276.8004150390625, "pos_frac": 0.75, "sample": [124.37194061279297, 140.4451904296875, -29.037628173828125, -10.13397216796875, 0.0, -7.579559326171875, 27.022720336914062, -28.878860473632812, 35.9825439453125, 0.76171875, 36.624298095703125, 53.19566345214844, 18.0601806640625, 45.97557067871094, -35.05914306640625, 124.65592193603516, -3.089569091796875, 166.27325439453125, 50.07354736328125, -13.86846923828125, 119.16691589355469, 65.51766967773438, 201.81190490722656, 162.28372192382812, -70.486083984375, 35.879791259765625, 20.385726928710938, 37.808837890625, -2.5554580688476562, 14.580947875976562, 83.8837890625, -10.687751770019531, 0.2281017303466797, 162.5032958984375, 131.69479370117188, 59.912109375, -3.4480056762695312, 117.4017333984375, -167.60797119140625, 155.705078125, -9.915889739990234, 4.336395263671875, 22.27154541015625, 19.5797119140625, 8.794197082519531, 119.78207397460938, 12.89959716796875, -49.855316162109375, 132.26507568359375, 0.04412841796875, 45.837921142578125, -114.3350830078125, 155.5192413330078, 120.82672119140625, 190.39117431640625, 136.0059814453125, -4.746650695800781, 47.75373840332031, 9.140228271484375, 21.8839111328125, 13.177360534667969, 6.869956970214844, 192.3336181640625, 74.038330078125, -66.07414245605469, -127.26913452148438, 46.467010498046875, -159.88116455078125, 18.109161376953125, -20.58428955078125, 174.5277099609375, 44.6986083984375, -17.237884521484375, 182.38299560546875, 145.33880615234375, 129.92568969726562, 44.4864501953125, 16.436119079589844, 9.6124267578125, 116.72528076171875, 38.22021484375, 35.070491790771484, -5.05828857421875, 128.55194091796875, 106.29534912109375, 173.5725860595703, 78.09341430664062, 59.35382080078125, 71.094970703125, 63.80462646484375, 128.37493896484375, 134.84939575195312, 140.5762481689453, -16.590225219726562, 2.0676746368408203, 27.459129333496094, 0.23114013671875, -47.427978515625, 152.56460571289062, 174.47439575195312, 165.69691467285156, 18.81610107421875, 4.945930480957031, -125.226806640625, 147.0374755859375, -21.7750244140625, 113.74604034423828, 124.199951171875, 170.23678588867188, 142.2575225830078, -186.85093688964844, 194.66461181640625, 154.42120361328125, 0.3958873748779297, 128.4061279296875, -153.31503295898438, 12.39849853515625, 93.91864013671875, -36.121177673339844, 43.726470947265625, 79.68829345703125, 276.8004150390625, 145.9168701171875, 86.50057983398438, 68.55509948730469, -10.535690307617188, -88.415771484375, 195.13568115234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000334.npy"}
{"epoch": 0.6994764397905759, "step": 335, "batch_size": 128, "mean": 44.24415588378906, "std": 87.54065704345703, "min": -169.52288818359375, "p10": -79.31399154663086, "median": 27.861961364746094, "p90": 156.470556640625, "max": 208.48721313476562, "pos_frac": 0.703125, "sample": [-18.602935791015625, -145.82809448242188, 79.43291473388672, -41.50343322753906, 143.22207641601562, -83.03955078125, 178.5928955078125, -11.172271728515625, 50.60646057128906, 136.4762725830078, 72.72982788085938, -146.71905517578125, 43.13421630859375, -8.505189895629883, -3.0674095153808594, 22.9190673828125, 3.5051498413085938, -1.8720588684082031, -26.712005615234375, -77.22271728515625, -17.52008056640625, 174.26339721679688, -4.38323974609375, 5.594146728515625, 13.0447998046875, -0.06622314453125, 93.37918090820312, 2.7323226928710938, -34.101104736328125, 133.7515869140625, -169.52288818359375, 187.134765625, -21.738815307617188, 40.69525146484375, 49.43389892578125, 187.04928588867188, 7.0187835693359375, 128.59890747070312, 141.595703125, 62.08851623535156, 156.06777954101562, 143.03680419921875, 133.66009521484375, 122.04499816894531, 190.77102661132812, 157.41036987304688, 24.833541870117188, 127.3096694946289, 153.04782104492188, 160.79220581054688, -7.977561950683594, 61.85017395019531, 15.769502639770508, 119.9262466430664, 130.17041015625, 107.02742004394531, 191.42657470703125, 130.63165283203125, 4.6613311767578125, 39.9893798828125, 76.72988891601562, -134.29345703125, -88.50749206542969, -40.96661376953125, -119.25497436523438, 148.95513916015625, 93.00614929199219, 148.24032592773438, 75.37923431396484, 6.8381500244140625, 1.39349365234375, 0.0709228515625, 8.092269897460938, 58.950439453125, 193.11257934570312, 134.84539794921875, 26.281578063964844, -94.70014953613281, 10.502685546875, 24.388059616088867, 12.829483032226562, 131.81668090820312, -8.42449951171875, 101.38343811035156, 160.8837432861328, -14.543403625488281, -4.8270263671875, 1.6041717529296875, 132.3089599609375, 51.828338623046875, -50.28350830078125, 1.365936279296875, 123.55999755859375, 3.866209030151367, 13.621086120605469, -15.155548095703125, 10.7315673828125, 47.32331085205078, 80.05438232421875, -5.4805145263671875, 180.78469848632812, -131.55551147460938, 37.807464599609375, 29.442344665527344, 139.08868408203125, 21.507003784179688, 0.0, 153.77349853515625, 159.3087158203125, 130.75662231445312, -102.19793701171875, 100.38835144042969, 141.8600311279297, -121.15492248535156, -123.08468627929688, 208.48721313476562, 4.966705322265625, -3.8729019165039062, 8.67567253112793, -85.50460815429688, 11.372634887695312, 104.75406646728516, 57.2479248046875, 103.65666198730469, -77.71732330322266, -16.455902099609375, 78.36166381835938, 41.15739440917969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000335.npy"}
{"epoch": 0.7015706806282722, "step": 336, "batch_size": 128, "mean": 43.28335952758789, "std": 77.40849304199219, "min": -147.67974853515625, "p10": -37.77248077392577, "median": 29.937782287597656, "p90": 152.1783218383789, "max": 246.4737548828125, "pos_frac": 0.6796875, "sample": [58.935089111328125, 6.609466552734375, 78.44534301757812, -17.474639892578125, 122.32612609863281, 60.775970458984375, -20.236175537109375, 125.55093383789062, 123.21630859375, 148.529052734375, 8.682151794433594, 40.03582763671875, 100.01915740966797, 71.4602279663086, 3.081645965576172, 11.93536376953125, -2.746337890625, 176.5284881591797, 8.479621887207031, 152.054931640625, 170.8925323486328, 40.97942352294922, 112.61876678466797, 134.16424560546875, -70.83453369140625, -16.11700439453125, 8.724365234375, 14.549659729003906, 16.495132446289062, 42.7183837890625, 123.20327758789062, 50.182533264160156, -3.4817466735839844, 138.76312255859375, 183.61273193359375, 155.1306915283203, 57.26948547363281, 204.89071655273438, 17.22119903564453, 8.18975830078125, -47.72023010253906, 132.971435546875, -43.922271728515625, 143.6048583984375, 46.2086181640625, -0.8880996704101562, -1.0655174255371094, 119.4635009765625, -35.13685607910156, 20.07437515258789, 31.55097198486328, 0.148651123046875, 17.330230712890625, 97.57125854492188, 37.12628173828125, -51.3909912109375, -7.3172607421875, -13.565765380859375, 101.74834442138672, -10.077484130859375, 92.41851806640625, 32.8687744140625, -76.9609375, 77.60009765625, -31.7366943359375, 100.15457153320312, 47.347808837890625, -26.65863037109375, 109.80934143066406, 136.27239990234375, 4.5406494140625, -0.11322784423828125, 144.4381103515625, 81.86294555664062, 5.810188293457031, -45.75299072265625, -7.004730224609375, 30.981658935546875, 36.524993896484375, -10.820114135742188, 3.636676788330078, 38.3880615234375, 29.122283935546875, -134.67343139648438, 127.77540588378906, 38.68794250488281, 116.92294311523438, 76.262451171875, 21.74382781982422, 145.4011993408203, -21.316131591796875, -121.43476104736328, 170.0037384033203, 157.67205810546875, 66.9403076171875, 59.44114685058594, 173.68438720703125, -5.210304260253906, -26.206329345703125, 114.50469970703125, -5.218994140625, 158.69773864746094, -17.13446044921875, -62.180206298828125, -131.08645629882812, 246.4737548828125, 15.689239501953125, 143.02430725097656, 192.16156005859375, -147.67974853515625, 7.037727355957031, -13.411506652832031, -1.036376953125, 152.4662322998047, 29.387405395507812, 9.619796752929688, 8.8262939453125, -44.85888671875, -111.5838623046875, 0.0, -20.589111328125, 30.4881591796875, 43.249664306640625, 158.2886962890625, -34.214263916015625, -24.157928466796875, -12.855026245117188, 55.84199523925781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000336.npy"}
{"epoch": 0.7036649214659686, "step": 337, "batch_size": 128, "mean": 52.15848159790039, "std": 85.8612060546875, "min": -132.9004364013672, "p10": -68.18069458007812, "median": 49.41328430175781, "p90": 162.74462890625, "max": 195.54241943359375, "pos_frac": 0.71875, "sample": [34.861114501953125, -31.8497314453125, 9.9844970703125, 27.82978057861328, 13.821647644042969, -2.891204833984375, -16.22785186767578, 177.37548828125, 149.00665283203125, 194.639404296875, 0.2530975341796875, 115.71200561523438, 21.970779418945312, 54.234375, 172.29376220703125, 161.73818969726562, -13.52276611328125, 47.790374755859375, -15.50634765625, 15.410568237304688, -98.46121215820312, -115.45829772949219, -71.23484802246094, -37.132720947265625, 26.95824432373047, 3.32183837890625, 53.98260498046875, 158.02427673339844, -2.0616111755371094, 146.96102905273438, -0.21417236328125, 170.66336059570312, 84.87069702148438, 106.91436767578125, 155.975341796875, 131.56460571289062, 45.75518798828125, 8.720146179199219, 101.526611328125, 108.18231201171875, -11.067787170410156, -38.085243225097656, -66.9774169921875, 143.356201171875, 60.831390380859375, 144.32958984375, -10.989273071289062, -120.98171997070312, 1.89404296875, 34.207763671875, 17.81158447265625, 132.82073974609375, -7.9420013427734375, 110.26910400390625, 65.20870971679688, -111.16375732421875, 97.13839721679688, 60.09629821777344, 35.68373107910156, 102.45989990234375, 1.0468597412109375, 25.940292358398438, 9.853622436523438, 170.16409301757812, 161.84072875976562, -28.494430541992188, 135.90554809570312, 134.57801818847656, -132.9004364013672, 89.47454833984375, 95.4222412109375, 171.46224975585938, 39.631317138671875, 24.8001708984375, 123.35993957519531, 5.710723876953125, -4.90155029296875, -25.976112365722656, 161.85995483398438, 134.3047637939453, 12.490758895874023, 68.88433837890625, 76.53962707519531, 155.08053588867188, 128.79624938964844, 195.54241943359375, -16.668371200561523, 101.66659545898438, -56.58137512207031, -98.27507019042969, 59.98402404785156, 149.3572235107422, 97.81721496582031, 185.15771484375, -111.46788024902344, 185.96258544921875, -54.375701904296875, -6.459930419921875, 115.65019989013672, 67.12692260742188, 79.6812973022461, -14.76348876953125, -91.8150634765625, 166.22613525390625, -113.61260986328125, -95.89652252197266, 174.78558349609375, 95.94422912597656, 9.626363754272461, 2.6484642028808594, -30.57525634765625, 94.505859375, 148.11209106445312, 175.37811279296875, -70.98834228515625, 164.80886840820312, 3.8043212890625, 4.941154479980469, -32.727783203125, 133.1365203857422, 135.63369750976562, 112.40000915527344, -114.5694580078125, 51.03619384765625, 13.76129150390625, 151.7674560546875, 149.7622833251953, 89.28173828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000337.npy"}
{"epoch": 0.7057591623036649, "step": 338, "batch_size": 128, "mean": 60.15818405151367, "std": 80.0336685180664, "min": -174.81983947753906, "p10": -27.993226623535158, "median": 44.38712501525879, "p90": 165.4822998046875, "max": 215.99560546875, "pos_frac": 0.7578125, "sample": [168.64242553710938, -18.68544578552246, 9.027679443359375, 133.25204467773438, 155.92544555664062, 172.98248291015625, 124.7035140991211, 134.67059326171875, -25.14812469482422, -36.809112548828125, 10.66781234741211, 141.44297790527344, 36.32159423828125, 17.300783157348633, 152.369140625, 12.385650634765625, 122.59805297851562, 101.5203857421875, -5.9194488525390625, -57.438690185546875, 162.1173095703125, 25.566162109375, -1.5058841705322266, 181.82586669921875, 5.336700439453125, 63.98876953125, 144.32568359375, 39.02751922607422, -9.358688354492188, 1.7643909454345703, 40.7330322265625, 10.022247314453125, -2.6964111328125, 24.546722412109375, 70.65653991699219, 122.2252197265625, 70.080078125, 157.77667236328125, 26.969940185546875, 36.70429992675781, 105.82907104492188, 32.053680419921875, 112.242919921875, 151.85369873046875, 170.13189697265625, 41.221435546875, -31.773712158203125, 146.13479614257812, 55.53173828125, 77.07199096679688, 33.65639877319336, -19.44866943359375, 164.12796020507812, -130.86300659179688, 53.194671630859375, -6.949371337890625, -11.850082397460938, 178.83615112304688, -105.47880554199219, 162.2049560546875, 132.7371063232422, 215.99560546875, 149.60984802246094, 45.293243408203125, 142.24859619140625, 59.168914794921875, -17.737152099609375, -12.601364135742188, -51.73663330078125, -3.5464248657226562, 7.165760040283203, 118.74545288085938, 181.18927001953125, 19.62469482421875, -27.988021850585938, 2.7467098236083984, 43.48100662231445, 154.67755126953125, 160.25808715820312, -1.0545616149902344, 73.1873779296875, 1.16510009765625, 132.41444396972656, -3.7537460327148438, 125.98698425292969, 24.09783935546875, -45.222625732421875, 10.526962280273438, 112.2398681640625, 138.37106323242188, 120.66140747070312, 178.5484619140625, 23.667755126953125, -4.00531005859375, -125.32713317871094, 155.51211547851562, 22.590423583984375, 188.7396240234375, -89.36251831054688, 78.79784393310547, 35.418853759765625, -28.00537109375, -1.380035400390625, 78.42343139648438, 0.7567138671875, 164.08831787109375, 109.24073791503906, 100.36848449707031, -36.892974853515625, 183.08233642578125, 48.09033203125, 62.167755126953125, 0.5713710784912109, 51.5528564453125, -4.842529296875, 175.95208740234375, 40.87890625, 53.47334289550781, 122.21124267578125, 82.7200927734375, -174.81983947753906, 3.59332275390625, 30.148193359375, 2.3502578735351562, 188.1968994140625, 183.0439910888672, -39.942962646484375, 135.0449981689453], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000338.npy"}
{"epoch": 0.7078534031413612, "step": 339, "batch_size": 128, "mean": 61.084228515625, "std": 85.2576675415039, "min": -181.04669189453125, "p10": -43.386681365966794, "median": 51.926490783691406, "p90": 159.89590454101562, "max": 220.68356323242188, "pos_frac": 0.7734375, "sample": [86.2127685546875, -45.6494140625, 65.9573745727539, 34.4241943359375, 129.21749877929688, -31.205337524414062, 31.478485107421875, -42.785552978515625, 104.80279541015625, -3.37890625, 182.28717041015625, 17.44854736328125, 4.18548583984375, 6.9694671630859375, 131.23419189453125, 2.790679931640625, 60.211212158203125, 67.50357055664062, 73.07012176513672, -60.10784912109375, 138.87985229492188, 13.49261474609375, 61.123199462890625, 103.83195495605469, 0.0, 216.10858154296875, -76.0052490234375, 138.93893432617188, 0.0, 16.270030975341797, 99.5744857788086, -44.78931427001953, 185.96734619140625, 137.90786743164062, 6.51727294921875, -149.58421325683594, 159.8436279296875, 163.71560668945312, 54.87544250488281, 32.83905029296875, 31.493011474609375, 143.10203552246094, -30.9676513671875, 139.852783203125, 180.69537353515625, 17.920623779296875, 138.94708251953125, -0.24736785888671875, 23.884246826171875, 88.91482543945312, 22.412445068359375, 6.69830322265625, 218.689208984375, 113.84962463378906, 125.0240478515625, 139.03411865234375, 25.576744079589844, 19.63074493408203, 143.13592529296875, 181.83572387695312, 159.09710693359375, -2.6025314331054688, -69.52847290039062, -81.72116088867188, -2.0374374389648438, 173.75686645507812, 5.302215576171875, 120.33200073242188, 123.2186279296875, 32.87982177734375, 33.706512451171875, 18.362823486328125, -130.39366149902344, 37.304718017578125, 137.59481811523438, 36.7554931640625, 137.58837890625, 150.49517822265625, -65.8673095703125, 126.31143188476562, 4.414192199707031, 129.92977905273438, 157.46632385253906, -8.48211669921875, 150.167236328125, 220.68356323242188, 113.42318725585938, 0.0, 82.9471435546875, 105.04623413085938, 122.55645751953125, 132.00726318359375, 146.05032348632812, 151.25692749023438, 160.10501098632812, -10.996002197265625, 55.230133056640625, -6.205169677734375, -31.9705810546875, 158.88540649414062, 122.75921630859375, -85.9300537109375, 29.30401611328125, 2.60516357421875, 12.7664794921875, 136.87249755859375, 140.28579711914062, 13.694580078125, -32.21238708496094, 3.8727340698242188, 171.700439453125, 136.73138427734375, -181.04669189453125, 150.21722412109375, 26.36114501953125, 11.658889770507812, 24.098236083984375, 63.135223388671875, 160.01788330078125, -71.81146240234375, -165.23878479003906, 186.50738525390625, 43.997222900390625, 48.9775390625, 150.017822265625, 25.331161499023438, -27.344078063964844, 144.72705078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000339.npy"}
{"epoch": 0.7099476439790576, "step": 340, "batch_size": 128, "mean": 49.71488571166992, "std": 82.43031311035156, "min": -142.33651733398438, "p10": -48.63021697998046, "median": 32.5490837097168, "p90": 155.0772689819336, "max": 221.1942138671875, "pos_frac": 0.734375, "sample": [198.1342315673828, 93.74767303466797, -21.339370727539062, 140.90817260742188, 1.3566741943359375, 150.03089904785156, -58.344482421875, -18.315780639648438, -53.8221435546875, 149.39183044433594, -68.39620971679688, 123.26649475097656, -74.10260009765625, 4.1993408203125, 0.0, 157.64840698242188, 55.85125732421875, -46.40510559082031, 116.92186737060547, 152.94906616210938, 144.82351684570312, 13.887374877929688, -31.680206298828125, 157.73556518554688, -95.80343627929688, 12.195556640625, -26.538482666015625, 137.81338500976562, 2.92431640625, 150.97618103027344, 183.4013671875, 3.6097545623779297, 112.61831665039062, 152.0721435546875, 31.49969482421875, -12.568466186523438, 0.5583782196044922, -116.742919921875, 29.386154174804688, 15.150009155273438, 146.01083374023438, 132.81869506835938, -110.66236877441406, 44.852630615234375, 80.08816528320312, 156.60377502441406, -27.740081787109375, 88.71315002441406, 14.155792236328125, -14.692337036132812, 84.33493041992188, 181.2969970703125, -14.646121978759766, -82.10198974609375, 8.458341598510742, -11.174774169921875, 135.87428283691406, 102.64962768554688, 31.983978271484375, 15.707107543945312, 5.3697509765625, -142.33651733398438, 159.85765075683594, 0.0, 191.82278442382812, 28.27655029296875, 68.32888793945312, 17.81938934326172, 148.83502197265625, 35.379150390625, 16.272064208984375, -5.608604431152344, 135.08644104003906, -105.05014038085938, 129.6124725341797, 41.3203125, 37.407562255859375, -104.13229370117188, 155.16360473632812, 10.87054443359375, 146.15960693359375, 213.6961669921875, -81.112060546875, 33.8807373046875, -15.057281494140625, 118.6676025390625, 70.28326416015625, 25.34082794189453, 141.776123046875, 22.486572265625, 39.413726806640625, -13.373270034790039, 78.54786682128906, -13.753036499023438, 137.69775390625, 102.08502197265625, -20.621810913085938, 4.31060791015625, 52.93621826171875, 41.017364501953125, 221.1942138671875, 9.451539993286133, 144.36331176757812, -40.24473571777344, 156.0467529296875, -33.967529296875, 20.205360412597656, -140.0645751953125, 137.21817016601562, 109.95736694335938, 133.626220703125, 33.11418914794922, 11.749927520751953, 107.73591613769531, 155.84442138671875, 155.04026794433594, 16.377044677734375, 82.03514862060547, 13.305046081542969, 50.545501708984375, 29.797210693359375, 61.940765380859375, -35.32177734375, 20.501178741455078, 62.57636260986328, 18.345550537109375, 132.06939697265625, -10.142288208007812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000340.npy"}
{"epoch": 0.7120418848167539, "step": 341, "batch_size": 128, "mean": 43.16305923461914, "std": 85.3323974609375, "min": -155.9368896484375, "p10": -63.266992187499994, "median": 26.22686004638672, "p90": 154.41173706054687, "max": 212.9647216796875, "pos_frac": 0.6640625, "sample": [6.2819366455078125, -32.140289306640625, 114.08566284179688, 145.06024169921875, 162.54425048828125, -42.013153076171875, 79.051513671875, 19.19000244140625, 31.437271118164062, 2.88909912109375, 135.140625, 38.21037292480469, 144.25865173339844, 10.94797134399414, 184.86163330078125, 174.36428833007812, -113.51284790039062, 4.0291748046875, -5.19256591796875, 57.9849853515625, 142.289794921875, 15.293365478515625, 129.4427490234375, -41.59562683105469, 180.68548583984375, 55.686309814453125, 10.846115112304688, 15.672134399414062, 9.149795532226562, -10.61761474609375, -1.721893310546875, 35.07159423828125, -15.599441528320312, -8.167083740234375, -94.58987426757812, 78.04904174804688, -61.81610107421875, -91.33767700195312, -0.49519920349121094, -5.7418060302734375, 4.39630126953125, 84.17262268066406, 169.55172729492188, 87.417724609375, 131.8994903564453, 5.04107666015625, 144.8365020751953, 23.822189331054688, -5.1917877197265625, 128.52015686035156, 101.55479431152344, 130.84896850585938, 184.52874755859375, -20.5172119140625, 148.69256591796875, 127.41128540039062, 54.169158935546875, 121.90554809570312, 12.41717529296875, -10.91986083984375, 105.888671875, -77.85040283203125, 212.9647216796875, -9.687942504882812, 115.48297119140625, -122.60671997070312, -84.13034057617188, 141.57066345214844, 163.11158752441406, 4.533905029296875, -3.3944091796875, 156.26123046875, 111.01458740234375, -21.541946411132812, 131.1085968017578, 142.75247192382812, 144.23348999023438, 33.35833740234375, 17.788803100585938, -125.02604675292969, 70.97958374023438, 4.350799560546875, -11.809066772460938, -13.839599609375, 197.43585205078125, 28.63153076171875, -21.25836181640625, 51.39495849609375, 47.387725830078125, 153.80166625976562, -93.17256164550781, 42.52325439453125, 183.86614990234375, 139.09225463867188, -124.74366760253906, -10.015472412109375, -13.362541198730469, -1.176565170288086, -10.329803466796875, 29.84075927734375, 35.822540283203125, 145.06195068359375, 22.2200927734375, -15.671630859375, -131.51480102539062, 37.86627197265625, 35.875, -34.632843017578125, -155.9368896484375, -28.98448944091797, 16.602935791015625, -66.65240478515625, -12.42730712890625, 155.83523559570312, -150.10072326660156, 67.35310363769531, 129.66299438476562, 153.02581787109375, 132.59825134277344, 51.41133117675781, 144.5972900390625, 7.25787353515625, 33.208412170410156, -19.2470703125, 188.54443359375, 13.308837890625, -10.512725830078125, 12.33096694946289], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000341.npy"}
{"epoch": 0.7141361256544503, "step": 342, "batch_size": 128, "mean": 52.6014404296875, "std": 77.93048095703125, "min": -171.50311279296875, "p10": -30.64202880859375, "median": 33.20100402832031, "p90": 155.7587905883789, "max": 213.590576171875, "pos_frac": 0.7421875, "sample": [-5.720977783203125, 23.880630493164062, 38.327880859375, 22.9271240234375, 135.556396484375, 205.54519653320312, 17.91677474975586, 101.16206359863281, 6.760368347167969, 39.552886962890625, -30.430267333984375, 91.9525146484375, -0.9448394775390625, 130.0742950439453, 108.40721130371094, 134.5941162109375, -7.722038269042969, 64.6910400390625, -118.6098403930664, 3.0659332275390625, 42.35015869140625, 111.02371215820312, 11.12554931640625, 22.786537170410156, -49.25988006591797, 12.294883728027344, 13.199363708496094, -42.151611328125, 62.633209228515625, 129.82968139648438, 13.243057250976562, 101.128662109375, 108.60722351074219, 63.68975830078125, -8.79327392578125, 14.737091064453125, 12.937057495117188, -11.970785140991211, -34.80781555175781, 31.71453857421875, 71.34974670410156, 175.35537719726562, 19.822021484375, -84.18783569335938, -17.154525756835938, -22.907623291015625, -29.865692138671875, 59.1278076171875, 53.367919921875, 3.513275146484375, 117.19491577148438, -4.327484130859375, -31.136138916015625, 157.99655151367188, 5.751312255859375, 185.49703979492188, 152.35565185546875, -3.3345947265625, 93.15740966796875, 123.70941162109375, 155.12864685058594, 158.65618896484375, 116.7869873046875, 127.9192886352539, 114.90318298339844, -124.58840942382812, -14.09759521484375, 0.0, -35.82550048828125, 48.02398681640625, 171.16241455078125, 5.4756622314453125, 176.8358154296875, 3.8741931915283203, 43.349143981933594, 133.7840576171875, 158.56674194335938, 36.15546417236328, 175.9431915283203, 4.039436340332031, -13.19927978515625, 143.57379150390625, 34.687469482421875, 113.78605651855469, -70.31845092773438, 7.4171295166015625, -171.50311279296875, -17.10650634765625, 108.44876098632812, 108.67801666259766, 150.10939025878906, 115.51824951171875, 26.987762451171875, 43.251129150390625, 26.13011932373047, 57.2877197265625, 126.72579956054688, 20.321151733398438, 5.6058502197265625, -112.15277099609375, 17.610382080078125, 133.87823486328125, -12.265975952148438, 213.590576171875, 19.141571044921875, 137.6750946044922, -19.460052490234375, 10.481475830078125, 87.27398681640625, -96.4693832397461, 157.2291259765625, 130.50192260742188, 29.4739990234375, 133.02435302734375, 135.84616088867188, 144.82635498046875, 20.04986572265625, 29.90228271484375, 129.34666442871094, 145.09893798828125, -13.119499206542969, -12.672592163085938, -35.5927734375, 189.95263671875, 123.11372375488281, -8.982492446899414, 181.294677734375, 11.3057861328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000342.npy"}
{"epoch": 0.7162303664921466, "step": 343, "batch_size": 128, "mean": 45.07551956176758, "std": 82.37698364257812, "min": -205.18939208984375, "p10": -41.44499206542969, "median": 23.06652069091797, "p90": 149.84053955078124, "max": 206.908203125, "pos_frac": 0.7109375, "sample": [-31.829818725585938, 86.25312805175781, 140.8697967529297, 6.906354904174805, 41.99652099609375, 74.79653930664062, 124.54547119140625, 21.37054443359375, -93.43759155273438, 26.171142578125, 148.1461944580078, -41.154998779296875, -8.11468505859375, -28.929779052734375, 104.623046875, 170.8837890625, 159.45132446289062, 43.96875, -10.640113830566406, -6.481271743774414, 205.6287841796875, 20.570343017578125, 154.5517578125, 148.24847412109375, 107.78231048583984, 20.004959106445312, 4.824287414550781, 145.029541015625, 20.54443359375, 131.18325805664062, 64.72503662109375, -115.29558563232422, 2.608123779296875, -10.08404541015625, -118.85009765625, 40.80065155029297, 17.507476806640625, 142.24862670898438, -57.693023681640625, 65.99627685546875, -42.53828430175781, 176.96554565429688, 40.150390625, 40.57818603515625, 0.0, -42.12164306640625, 17.59063720703125, 121.9365234375, 36.58594512939453, 201.85968017578125, -124.51473999023438, 23.697059631347656, 37.10645294189453, 12.859119415283203, 137.65509033203125, -33.665679931640625, 103.82945251464844, 67.55682373046875, -0.19886016845703125, 119.11457061767578, 128.7823028564453, 7.6497650146484375, -1.1273307800292969, 2.535369873046875, 130.4130401611328, 141.84527587890625, -8.06610107421875, 146.9522705078125, 19.1148681640625, 75.43292236328125, 91.55939483642578, -18.95599365234375, 53.861572265625, -22.097991943359375, -0.6497421264648438, 16.479759216308594, 11.115936279296875, 206.908203125, 124.60934448242188, 148.2150421142578, 33.01427459716797, 84.7373046875, -52.41290283203125, 13.8365478515625, 129.56973266601562, -21.96270751953125, 4.4246826171875, 34.6473388671875, 12.243194580078125, -27.1826171875, 4.771579742431641, 142.8262939453125, 141.04629516601562, 199.637451171875, 7.6429443359375, -11.774215698242188, 125.65934753417969, 95.72515869140625, 139.2864990234375, 22.43598175048828, 162.2366485595703, 133.31443786621094, -7.350006103515625, 53.25152587890625, -205.18939208984375, 153.55535888671875, -18.26434326171875, -5.678783416748047, -72.6767578125, -26.470794677734375, 8.821521759033203, 130.1312255859375, 154.61212158203125, 2.1668701171875, -132.50811767578125, 10.24749755859375, -7.1402130126953125, -106.905517578125, 57.6314697265625, 185.3341064453125, 92.87782287597656, 0.7777976989746094, 31.811614990234375, 7.556694030761719, 7.35986328125, 163.11257934570312, -14.19415283203125, -135.67678833007812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000343.npy"}
{"epoch": 0.7183246073298429, "step": 344, "batch_size": 128, "mean": 55.95680236816406, "std": 84.65853881835938, "min": -169.23663330078125, "p10": -39.505063629150385, "median": 39.64317321777344, "p90": 169.88348999023438, "max": 208.96807861328125, "pos_frac": 0.734375, "sample": [18.168212890625, -21.3243408203125, 125.10905456542969, 98.8294677734375, -20.04095458984375, 142.54922485351562, 143.1461639404297, 37.093536376953125, 126.71531677246094, -17.08868408203125, 123.22207641601562, 73.28038024902344, 127.63101196289062, -43.67426681518555, 113.02325439453125, 178.60858154296875, 15.4674072265625, 71.58474731445312, -156.82504272460938, 120.33572387695312, -51.959442138671875, 7.02508544921875, 0.35693359375, -33.022369384765625, 183.78009033203125, 105.08757019042969, -15.020425796508789, 169.2877197265625, 125.95028686523438, 10.8739013671875, 171.27362060546875, -38.426025390625, -6.86529541015625, 137.14129638671875, -169.23663330078125, 54.718353271484375, 143.41534423828125, 132.28585815429688, 12.360687255859375, -10.075103759765625, -147.5306396484375, 24.419677734375, 27.63568115234375, -0.8280715942382812, -5.6299896240234375, 19.763626098632812, -97.00726318359375, 118.20051574707031, 179.22671508789062, 114.54989624023438, -30.814453125, -113.76582336425781, -0.454620361328125, -47.839027404785156, 107.67138671875, 100.43746948242188, 174.27151489257812, -12.816741943359375, 186.677978515625, 167.20925903320312, 146.39393615722656, 144.5302276611328, 203.80307006835938, 187.02944946289062, 150.93467712402344, -4.912162780761719, 36.2236328125, 184.14254760742188, -13.876571655273438, 18.731201171875, -49.853363037109375, 14.285995483398438, 16.114288330078125, 197.54522705078125, 2.14654541015625, 42.19281005859375, 17.32916259765625, 67.96856689453125, 29.818603515625, 96.38871765136719, 32.087913513183594, 20.419464111328125, 127.1314697265625, 137.43112182617188, -6.535736083984375, 83.5062255859375, 2.3562164306640625, 146.38375854492188, 22.190277099609375, 54.68864440917969, 184.10113525390625, 114.69976043701172, 14.791976928710938, -86.5474853515625, -19.709068298339844, 160.52197265625, 66.23916625976562, -0.9199409484863281, 86.2984619140625, -30.877166748046875, -7.544242858886719, 33.625335693359375, 15.347625732421875, 208.96807861328125, -12.97686767578125, -82.1187744140625, 147.4649658203125, 36.38092041015625, 19.506309509277344, 3.917510986328125, 5.9556884765625, 119.02630615234375, 10.104507446289062, 163.28118896484375, 103.61451721191406, 45.944000244140625, 87.43667602539062, 82.862548828125, 58.41827392578125, 172.25473022460938, 105.6600341796875, 30.497398376464844, 63.608001708984375, 151.17227172851562, 78.04638671875, -42.02281951904297, 162.51266479492188, -141.80645751953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000344.npy"}
{"epoch": 0.7204188481675393, "step": 345, "batch_size": 128, "mean": 42.932220458984375, "std": 75.58804321289062, "min": -180.97637939453125, "p10": -26.044198608398435, "median": 20.825210571289062, "p90": 149.36648864746093, "max": 189.40341186523438, "pos_frac": 0.75, "sample": [54.761474609375, 109.20357513427734, 105.13983154296875, -130.02813720703125, 175.33380126953125, 99.03227233886719, 3.892333984375, -39.424285888671875, 13.981475830078125, 0.3303413391113281, 43.47767639160156, 11.760124206542969, -0.48879241943359375, -1.5221786499023438, 22.497299194335938, -6.01092529296875, 160.276123046875, 13.3580322265625, 13.18359375, -4.461395263671875, 11.760650634765625, -25.27789306640625, 127.96366119384766, 148.5430450439453, 3.1969566345214844, 37.3665771484375, 127.9827880859375, 28.429290771484375, -30.580596923828125, 0.6564369201660156, -10.081161499023438, 41.993438720703125, -21.436683654785156, 73.24017333984375, 0.36431884765625, 78.89718627929688, 57.8924560546875, 5.156463623046875, 17.519683837890625, 19.510589599609375, 47.679779052734375, -7.446531295776367, -33.99848556518555, 134.9267578125, 14.88916015625, 141.6223602294922, 12.506484985351562, 1.6930809020996094, 45.73970031738281, 187.15545654296875, 127.29889678955078, 129.04901123046875, 97.83853149414062, 161.03855895996094, 167.58909606933594, 11.958175659179688, 24.3719482421875, -27.832244873046875, 15.861740112304688, -6.971855163574219, 154.96615600585938, 100.04864501953125, 23.083404541015625, 23.4276123046875, 23.049972534179688, -66.66020202636719, -180.97637939453125, 15.161731719970703, -121.12286376953125, -22.518157958984375, 161.76162719726562, 168.16107177734375, 146.63758850097656, 22.632156372070312, 31.316970825195312, -22.451454162597656, 189.40341186523438, -92.83970642089844, 18.827239990234375, -3.2863082885742188, 150.57937622070312, 148.8466796875, 129.80291748046875, -140.58111572265625, 84.06488037109375, 25.76092529296875, -36.609588623046875, 181.9998779296875, 43.0216064453125, 116.95526885986328, 163.77891540527344, 139.19349670410156, 22.13983154296875, 2.32550048828125, 109.34677124023438, 14.413421630859375, 35.933326721191406, 141.54827880859375, 148.58761596679688, -24.985687255859375, 145.766357421875, -98.92132568359375, 127.43927764892578, -102.5025634765625, 16.9730224609375, 60.992706298828125, 2.484525680541992, 22.617828369140625, -10.588394165039062, -11.439697265625, -11.083419799804688, 120.13228607177734, 19.373016357421875, 15.619918823242188, 151.90008544921875, 131.08648681640625, 13.8858642578125, -1.0739173889160156, 0.0, 11.075912475585938, 2.0620651245117188, 18.791168212890625, 10.766227722167969, -0.4946708679199219, 140.65623474121094, 6.14276123046875, 83.47425842285156, 23.083740234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000345.npy"}
{"epoch": 0.7225130890052356, "step": 346, "batch_size": 128, "mean": 55.089622497558594, "std": 83.84135437011719, "min": -191.658935546875, "p10": -38.3590087890625, "median": 35.65809631347656, "p90": 166.04995422363282, "max": 187.450439453125, "pos_frac": 0.7109375, "sample": [-46.65638732910156, -37.82905578613281, 1.0270309448242188, -39.59556579589844, 113.82467651367188, 65.07647705078125, 168.72509765625, 168.07028198242188, 117.18907928466797, -7.231231689453125, 136.5637664794922, -21.137008666992188, -40.080169677734375, 22.385147094726562, 139.89535522460938, -135.64947509765625, -93.08505249023438, 3.867431640625, 21.3375244140625, 169.69400024414062, 23.024459838867188, -11.680526733398438, 23.2440185546875, -46.703155517578125, 44.36750411987305, 132.58489990234375, 8.598770141601562, 187.010986328125, -14.095703125, -2.907623291015625, 37.702545166015625, 124.05687713623047, -111.5026626586914, 35.701629638671875, -4.908447265625, 156.53826904296875, 140.57557678222656, 168.35531616210938, 16.074567794799805, 16.46234130859375, 77.37420654296875, 123.30142211914062, -131.17840576171875, 24.55303955078125, -7.4268646240234375, 27.148956298828125, -11.631317138671875, -4.4207611083984375, 21.328857421875, 45.37446975708008, 126.4437255859375, 20.1466064453125, 128.2970428466797, -29.467369079589844, 29.569610595703125, 90.20706176757812, 180.2880859375, 6.7026824951171875, -6.487434387207031, 140.15908813476562, 166.06689453125, 168.75372314453125, 28.08526611328125, 12.075416564941406, 23.91754150390625, 15.057472229003906, 168.31881713867188, -16.195762634277344, 159.17074584960938, 36.75079345703125, 145.038818359375, 117.11502075195312, 131.5290985107422, 184.94879150390625, 159.6368408203125, 139.0023193359375, -113.76655578613281, 166.04269409179688, 56.01789855957031, 30.70458984375, 120.442626953125, -122.96533203125, 156.24978637695312, 5.411407470703125, -6.03265380859375, -26.080337524414062, -21.527015686035156, 5.358634948730469, -25.2921142578125, 166.204345703125, 163.83819580078125, 35.61456298828125, -15.181037902832031, 152.26416015625, -1.8352279663085938, -14.615867614746094, 123.87037658691406, -191.658935546875, -0.06867599487304688, 156.92413330078125, 1.7243671417236328, 9.825454711914062, 125.64993286132812, 122.56393432617188, 176.7881317138672, 98.83467102050781, 127.56855773925781, 153.27154541015625, 7.2304534912109375, -1.834869384765625, 0.0, 94.38236999511719, -4.6334686279296875, 36.21470642089844, 131.65524291992188, 96.12130737304688, 72.62384033203125, 135.98956298828125, 44.35753631591797, -77.94523620605469, 129.03634643554688, 100.58615112304688, -40.70439147949219, 124.27859497070312, 150.55548095703125, 127.12080383300781, 4.394491195678711, 187.450439453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000346.npy"}
{"epoch": 0.724607329842932, "step": 347, "batch_size": 128, "mean": 46.66077423095703, "std": 80.87889099121094, "min": -203.03570556640625, "p10": -33.25379409790039, "median": 28.937698364257812, "p90": 156.55560302734375, "max": 191.43447875976562, "pos_frac": 0.7109375, "sample": [5.3008270263671875, 76.30169677734375, 140.93478393554688, 4.052337646484375, 66.30743408203125, 114.6455078125, 4.4630279541015625, 152.29949951171875, 16.92761993408203, 149.52735900878906, -203.03570556640625, 96.002685546875, 148.5822296142578, 27.253509521484375, 164.8592987060547, -106.67723083496094, -38.8590087890625, -1.5587921142578125, 132.42742919921875, 165.4757843017578, 129.36732482910156, -50.9036865234375, 39.584869384765625, 11.34527587890625, 19.948822021484375, 36.79827880859375, 145.49896240234375, 17.52813720703125, 10.740402221679688, -3.9266204833984375, 0.087493896484375, 115.614990234375, 176.5984649658203, 55.438232421875, 148.47021484375, 74.00112915039062, 157.22552490234375, -32.58179473876953, 51.155303955078125, 56.81322479248047, 9.669334411621094, 156.26849365234375, 176.48086547851562, 26.527328491210938, 37.80584716796875, 102.70025634765625, -14.864547729492188, 10.811767578125, 40.952728271484375, -109.23214721679688, -2.978912353515625, -8.053482055664062, 71.79124450683594, -15.571945190429688, -4.18182373046875, 90.48324584960938, -1.3279571533203125, 142.1671142578125, 80.61306762695312, -79.17985534667969, 108.91232299804688, 10.317596435546875, -1.6002731323242188, -16.72466278076172, -8.924179077148438, 105.53741455078125, 54.65162658691406, 46.25907897949219, 179.788818359375, 57.05891418457031, 13.60711669921875, -129.7857666015625, -9.44647216796875, 18.044570922851562, 191.43447875976562, 0.5733413696289062, -110.31812286376953, -15.813957214355469, -34.82179260253906, 3.989593505859375, 6.2319793701171875, 119.13751220703125, 161.60723876953125, -14.945068359375, 11.448169708251953, -148.74172973632812, 114.58016204833984, 181.7991943359375, 96.06378173828125, 90.31985473632812, -11.4564208984375, 104.19195556640625, -57.838836669921875, 143.99124145507812, 22.957443237304688, -10.239410400390625, 37.17271423339844, -2.0867347717285156, 162.73739624023438, 3.509063720703125, -172.9013671875, 3.10003662109375, 181.23394775390625, 146.2208251953125, 0.0, 18.9658203125, 18.936126708984375, 30.62188720703125, 124.61085510253906, 151.4185791015625, 31.0234375, 4.578765869140625, 146.44363403320312, 128.70469665527344, -10.8775634765625, -1.0401077270507812, 41.82159423828125, -4.593048095703125, 133.501953125, 40.0264892578125, -1.7320556640625, 163.66497802734375, 171.3033905029297, 58.08367919921875, 13.508384704589844, -38.176788330078125, 144.39523315429688, -8.358970642089844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000347.npy"}
{"epoch": 0.7267015706806282, "step": 348, "batch_size": 128, "mean": 42.65314483642578, "std": 78.45606994628906, "min": -194.35617065429688, "p10": -35.579786682128905, "median": 17.268707275390625, "p90": 149.77893829345703, "max": 194.40444946289062, "pos_frac": 0.671875, "sample": [72.83657836914062, 5.57855224609375, 3.7728424072265625, 4.07232666015625, -5.91595458984375, -13.946746826171875, 137.9159393310547, -8.113777160644531, 82.07510375976562, 112.29218292236328, 14.975982666015625, 145.22247314453125, 25.20025634765625, 130.75698852539062, 47.693267822265625, 11.43035888671875, 180.94967651367188, 120.701904296875, -58.303382873535156, 0.0, 15.32073974609375, 46.41441345214844, 114.54061126708984, -1.0285415649414062, -109.36941528320312, 176.75856018066406, -14.359390258789062, 1.8966827392578125, -9.51348876953125, 86.66021728515625, 127.52296447753906, 116.47344970703125, -9.599021911621094, -1.5693206787109375, 157.5047607421875, -31.87200927734375, 147.7740478515625, -4.326259613037109, 121.50825500488281, 158.72955322265625, 3.527006149291992, -15.85394287109375, -77.9271240234375, 27.46221923828125, -1.43017578125, 143.39524841308594, 36.118927001953125, 56.22441101074219, 23.544570922851562, 68.84508514404297, 79.72856903076172, 84.1075439453125, -82.89068603515625, 2.493480682373047, -11.898468017578125, 109.49111938476562, 160.31271362304688, 183.74636840820312, 157.88851928710938, -75.80230712890625, -8.859832763671875, -8.7049560546875, -35.4276123046875, 106.7657470703125, -0.049053192138671875, 19.2166748046875, 56.429046630859375, -2.544961929321289, 8.1259765625, 40.30987548828125, 139.23175048828125, -47.338836669921875, -139.34986877441406, -4.4685821533203125, -78.09259033203125, 40.433746337890625, -25.013717651367188, 5.4061737060546875, 99.30986022949219, 9.2625732421875, 186.96994018554688, 7.57525634765625, 144.51849365234375, 192.6692657470703, 1.7139892578125, 170.4195556640625, -35.93486022949219, -23.8704833984375, -30.0906982421875, 21.45684814453125, -10.548879623413086, 151.1717071533203, 135.55859375, -6.221839904785156, 194.40444946289062, -120.53460693359375, 6.992118835449219, 6.293178558349609, -57.72843933105469, 8.2415771484375, 149.18203735351562, 22.304122924804688, 129.0321502685547, -10.961122512817383, 101.09148406982422, 0.6998748779296875, 130.48825073242188, 120.95465087890625, -19.440338134765625, 138.02703857421875, 124.05158996582031, -16.7835693359375, 69.49444580078125, -62.47052001953125, 8.679519653320312, 45.458038330078125, 163.0171661376953, 41.11273193359375, 148.373779296875, 31.762771606445312, 128.38998413085938, 2.6703529357910156, -4.980628967285156, 12.933120727539062, 0.81549072265625, 33.06694030761719, -194.35617065429688, 79.54833984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000348.npy"}
{"epoch": 0.7287958115183246, "step": 349, "batch_size": 128, "mean": 56.07810974121094, "std": 82.9571533203125, "min": -193.72451782226562, "p10": -22.34122314453125, "median": 33.18568420410156, "p90": 161.66073455810547, "max": 231.1845703125, "pos_frac": 0.75, "sample": [132.28086853027344, 160.8781280517578, 58.881622314453125, -193.72451782226562, -13.804534912109375, -0.16904830932617188, 231.1845703125, 105.07470703125, -108.01847839355469, 55.745361328125, 43.68501281738281, -46.52043151855469, 135.71376037597656, 177.00152587890625, 123.09347534179688, 137.78729248046875, 188.3665771484375, 14.306777954101562, 117.69427490234375, 176.71328735351562, 74.12530517578125, 3.96392822265625, -22.4976806640625, -22.274169921875, -4.4782562255859375, 33.64398193359375, -6.34271240234375, -6.839658737182617, 176.17132568359375, 168.88121032714844, 62.73460388183594, 130.58734130859375, 100.2705078125, -45.64727783203125, 41.07763671875, 106.70281982421875, 186.7044219970703, -10.38519287109375, 18.0777587890625, 49.373077392578125, 7.131244659423828, 0.7478294372558594, 133.692626953125, -7.736053466796875, 8.719192504882812, 54.38714599609375, -10.415283203125, 204.0645751953125, 18.713958740234375, -12.699371337890625, -59.290771484375, -0.13375282287597656, -15.021148681640625, 163.48681640625, 157.06573486328125, 17.952102661132812, 6.1939544677734375, 24.463302612304688, -107.60186767578125, 79.49690246582031, 0.0, 201.10543823242188, 9.106689453125, 12.818115234375, 134.76119995117188, 26.444580078125, 146.9110107421875, 159.63873291015625, 153.18475341796875, -18.153196334838867, 158.71249389648438, -20.987548828125, -106.86669921875, 207.01470947265625, 4.2374267578125, 48.519012451171875, -7.425384521484375, 119.75851440429688, 4.028167724609375, 14.499008178710938, 32.727386474609375, 122.6833724975586, 136.4371337890625, 158.60980224609375, 129.08990478515625, 8.448944091796875, 66.38020324707031, 35.4946174621582, 148.72845458984375, 46.585296630859375, 89.80404663085938, -7.37432861328125, 147.41650390625, 9.53884506225586, -22.882293701171875, 6.0942230224609375, 145.4348907470703, 19.281448364257812, 28.545074462890625, -124.66433715820312, 34.9754638671875, 29.48381805419922, 10.722549438476562, -2.531646728515625, -107.942626953125, 24.752197265625, 8.758773803710938, 5.76893424987793, 17.823081970214844, 113.99676513671875, 3.98712158203125, 145.18572998046875, 29.635467529296875, 128.6055145263672, 143.99281311035156, 174.935546875, 148.72348022460938, -23.390625, 104.24514770507812, 189.87261962890625, 12.11474609375, 86.37301635742188, 139.718505859375, 112.20401000976562, -120.7286376953125, -1.3335723876953125, 16.364028930664062, 134.76351928710938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000349.npy"}
{"epoch": 0.7308900523560209, "step": 350, "batch_size": 128, "mean": 50.629417419433594, "std": 86.12879943847656, "min": -154.50970458984375, "p10": -62.26392517089842, "median": 42.33723449707031, "p90": 162.32890625, "max": 210.17044067382812, "pos_frac": 0.796875, "sample": [161.8968505859375, 75.46980285644531, 0.8988456726074219, -6.0269927978515625, 9.874034881591797, 1.6849822998046875, 49.42236328125, 180.72711181640625, 1.7440834045410156, 130.49026489257812, 31.1839599609375, 15.348739624023438, 159.87014770507812, -0.6331939697265625, 27.15625, 161.5938720703125, 148.2501983642578, 161.68692016601562, 22.657386779785156, -46.16446304321289, 149.01438903808594, 15.142715454101562, 3.374298095703125, 111.30250549316406, 104.73267364501953, 136.3097381591797, 43.47441101074219, 116.30972290039062, 193.4095458984375, 183.5391845703125, 44.328155517578125, 105.2957763671875, 15.974895477294922, 17.425704956054688, -46.61619567871094, 121.92466735839844, 14.891822814941406, 16.644973754882812, 55.6251220703125, 31.9283447265625, 26.317657470703125, 118.46763610839844, 114.35580444335938, -143.05191040039062, 156.9407958984375, -13.654296875, 6.372039794921875, 73.62841796875, -154.50970458984375, 5.6281585693359375, 41.77915954589844, 151.8367919921875, 204.0618896484375, 105.71003723144531, 28.4388427734375, 74.91266632080078, 178.45379638671875, 142.38827514648438, 60.732666015625, 189.80911254882812, 101.04678344726562, -78.43353271484375, 160.18690490722656, 9.03729248046875, -15.07830810546875, -56.09197998046875, 42.61256408691406, 81.15133666992188, -105.69134521484375, 9.02155876159668, 41.87528991699219, 129.4716796875, 2.8632354736328125, -112.50155639648438, -12.9892578125, 124.89736938476562, -8.030036926269531, 38.25975036621094, 14.732330322265625, 92.47430419921875, 179.0675048828125, 149.74090576171875, 109.13390350341797, 78.15542602539062, 32.216552734375, 144.22122192382812, 19.07110595703125, 168.990966796875, -20.657211303710938, 9.904632568359375, 180.10498046875, 56.81520080566406, 23.670318603515625, 45.54466247558594, 104.1063232421875, 42.06190490722656, 31.603302001953125, 139.1951904296875, 210.17044067382812, 66.11126708984375, 155.03854370117188, -141.66290283203125, 74.75445556640625, 5.007171630859375, -153.88531494140625, -99.35370635986328, -76.66513061523438, 3.2430038452148438, -147.06903076171875, 45.929595947265625, -145.6621551513672, -81.35496520996094, 11.312164306640625, 178.61605834960938, 78.17103576660156, 50.27484130859375, 10.947906494140625, 15.753829956054688, 165.173095703125, -12.791793823242188, -15.628753662109375, 50.8725471496582, 4.170684814453125, -7.000244140625, 163.3370361328125, -101.45307922363281, 47.48382568359375, 85.18241882324219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000350.npy"}
{"epoch": 0.7329842931937173, "step": 351, "batch_size": 128, "mean": 32.787132263183594, "std": 88.17457580566406, "min": -208.10818481445312, "p10": -62.77860412597655, "median": 13.790023803710938, "p90": 157.75110778808593, "max": 229.49365234375, "pos_frac": 0.6328125, "sample": [0.0, 139.5145263671875, 14.3026123046875, 194.60592651367188, -0.6534423828125, 179.07980346679688, 10.800949096679688, -46.64649200439453, 130.15850830078125, 2.14666748046875, 2.1403350830078125, -83.82170104980469, 15.8115234375, 86.83428955078125, 156.5762939453125, -129.1482391357422, 20.396564483642578, 109.53565216064453, -30.61986541748047, 19.20367431640625, 26.422683715820312, 136.23109436035156, -29.070343017578125, 201.3807373046875, 53.09968566894531, 156.01197814941406, 9.562915802001953, -15.603607177734375, -6.234893798828125, -0.83807373046875, 168.80490112304688, 176.79200744628906, 49.04930114746094, 14.558147430419922, -3.0674591064453125, -71.49871826171875, -3.6582584381103516, 136.786376953125, -97.44246673583984, -96.36335754394531, -136.28494262695312, 136.48785400390625, 13.458236694335938, 47.79949951171875, -96.09930419921875, 150.38612365722656, 8.413070678710938, 13.967193603515625, -23.4644775390625, 15.734405517578125, -35.20111083984375, -26.28570556640625, 17.510406494140625, -31.66583251953125, 19.18194580078125, -2.48272705078125, 152.13534545898438, -40.14678955078125, -11.845855712890625, 19.58868408203125, 170.57766723632812, 225.64035034179688, 38.50392150878906, 194.25091552734375, 153.88018798828125, -17.6326904296875, 25.69879150390625, 13.61285400390625, 125.35403442382812, -81.92356872558594, -4.4595794677734375, 120.07474517822266, -26.086456298828125, 7.389984130859375, 160.49234008789062, -157.31765747070312, 26.2269287109375, 14.558013916015625, 6.702972412109375, 18.177734375, 0.0, -1.7640380859375, 42.21568298339844, 83.334228515625, 140.0621795654297, -0.6711502075195312, 1.5850067138671875, 16.20763397216797, 123.3851318359375, 1.956085205078125, 32.881744384765625, 6.0924530029296875, 35.526611328125, -13.936714172363281, 0.0, 170.34185791015625, 169.01876831054688, -0.56689453125, 12.138866424560547, -14.569915771484375, 11.464950561523438, -208.10818481445312, 115.05995178222656, 156.34719848632812, 6.7254486083984375, 5.2453155517578125, 229.49365234375, 57.522491455078125, 153.79518127441406, -13.587814331054688, -0.3629264831542969, 3.3416900634765625, -13.417724609375, 47.87538146972656, -2.8774681091308594, 147.48049926757812, 19.349273681640625, -45.718994140625, -104.12667846679688, -151.15704345703125, 67.99523162841797, 107.11752319335938, -176.2923583984375, 34.911895751953125, -40.56304931640625, -59.041412353515625, 214.076416015625, 30.9195556640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000351.npy"}
{"epoch": 0.7350785340314137, "step": 352, "batch_size": 128, "mean": 58.7017822265625, "std": 81.8944320678711, "min": -176.67266845703125, "p10": -41.039965820312496, "median": 45.87675476074219, "p90": 170.01471099853515, "max": 205.19500732421875, "pos_frac": 0.75, "sample": [117.5723876953125, 22.457427978515625, 161.80712890625, -43.777587890625, 10.381744384765625, 189.2498779296875, 120.55374908447266, 110.12857055664062, 55.36976623535156, 2.1130828857421875, -59.9205322265625, 128.353515625, -129.49618530273438, 17.78790283203125, 142.8802947998047, 4.429264068603516, 33.1854248046875, 175.30859375, 141.54367065429688, 110.56253051757812, 194.55499267578125, -38.28369903564453, -39.86669921875, 175.3768310546875, 29.1044921875, -176.67266845703125, -15.734222412109375, 37.69868469238281, 19.3521728515625, 120.28548431396484, -1.3088512420654297, 63.72080993652344, 173.6629638671875, 16.39678955078125, 90.783935546875, -96.07563781738281, 18.214500427246094, 48.0030517578125, -15.003128051757812, 16.1663818359375, 109.99560546875, -44.6739501953125, -25.810150146484375, 127.3309326171875, -28.902679443359375, 22.964614868164062, 0.0, 130.21279907226562, 12.584716796875, 167.7240447998047, -1.8564453125, 181.09796142578125, 127.58544921875, 30.17950439453125, 144.793212890625, 146.22622680664062, 22.404556274414062, 0.0, 67.00662231445312, 83.0439453125, 109.42333984375, 1.9469223022460938, 17.762939453125, 87.033447265625, -57.235931396484375, -30.02703857421875, 26.75975799560547, 75.37445068359375, -79.51043701171875, 5.66204833984375, 110.01517486572266, 190.06729125976562, 69.150390625, 57.35026550292969, 7.982095718383789, -38.80621337890625, 173.94558715820312, -2.8263282775878906, -101.0174560546875, -88.30998229980469, 205.19500732421875, 170.114013671875, 169.97215270996094, 20.499794006347656, 33.226959228515625, 143.2621307373047, 2.30841064453125, -64.57012939453125, 0.286468505859375, 81.13531494140625, 92.15185546875, -8.473953247070312, 54.293701171875, 35.6234245300293, 34.90501403808594, 43.750457763671875, 129.59152221679688, 132.503662109375, -7.874114990234375, 178.0668487548828, 53.58311462402344, 117.12970733642578, 133.67491149902344, 178.32015991210938, 151.0324249267578, -28.50225830078125, 143.43328857421875, 154.26361083984375, 114.82121276855469, 4.321556091308594, 22.436676025390625, 133.800537109375, 164.4671630859375, -0.1438751220703125, -50.03662109375, -9.701187133789062, 29.167316436767578, 162.8760986328125, 130.7010498046875, 173.99659729003906, -24.057327270507812, -60.507293701171875, 132.2425537109375, 152.18325805664062, 67.82750701904297, 116.47347259521484, 20.93198013305664, 145.610107421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000352.npy"}
{"epoch": 0.7371727748691099, "step": 353, "batch_size": 128, "mean": 55.890960693359375, "std": 82.05420684814453, "min": -164.37258911132812, "p10": -18.769439697265625, "median": 40.77705383300781, "p90": 152.9528579711914, "max": 237.30020141601562, "pos_frac": 0.765625, "sample": [58.47587585449219, 133.8170166015625, 89.41854858398438, -67.533935546875, 87.01754760742188, 164.60137939453125, -118.50025177001953, 149.95361328125, 153.91122436523438, 13.15460205078125, 58.540771484375, 26.864334106445312, 7.62786865234375, 67.55038452148438, -7.334907531738281, 23.51312255859375, -15.903472900390625, -26.642974853515625, 53.142059326171875, -18.018524169921875, 112.24940490722656, 83.58538818359375, 0.7581100463867188, -164.37258911132812, 135.32723999023438, 33.41363525390625, 93.59791564941406, 99.707763671875, 223.522216796875, -0.2810707092285156, 166.291015625, 134.66357421875, 145.9229736328125, -18.300872802734375, 145.21395874023438, -143.10614013671875, 180.29440307617188, 91.5948486328125, 69.29461669921875, 237.30020141601562, 97.01075744628906, -97.34396362304688, 45.02439880371094, 154.24813842773438, 36.52970886230469, -3.905731201171875, 152.05258178710938, 226.22772216796875, 198.91458129882812, 128.4395294189453, 135.80604553222656, 86.07294464111328, -65.6183090209961, 113.34735107421875, 14.634902954101562, 208.81887817382812, 2.6166725158691406, -16.602027893066406, 126.3750991821289, 103.27351379394531, 180.2254638671875, 79.15194702148438, -29.13653564453125, -10.95538330078125, 0.7010955810546875, 91.83837890625, -125.65740966796875, -14.52252197265625, 16.17822265625, 26.221832275390625, 31.280502319335938, 20.666976928710938, 135.7064971923828, 146.7056884765625, 142.08079528808594, 127.26661682128906, 14.661933898925781, 137.97007751464844, 12.973892211914062, 2.8247909545898438, 138.5595703125, 34.52008056640625, 5.619384765625, 26.008819580078125, 22.610687255859375, 64.52752685546875, 32.28826904296875, 126.07879638671875, 120.79949951171875, 4.319446563720703, -9.817733764648438, -16.5272216796875, 132.3300018310547, 124.25115966796875, 24.743125915527344, -16.5380859375, -3.7821216583251953, 7.934173583984375, 10.933494567871094, -16.304931640625, 69.4803466796875, 10.37310791015625, 28.65057373046875, 3.7247314453125, 155.60121154785156, 53.065582275390625, 75.038330078125, 52.49200439453125, 145.81727600097656, -118.3873291015625, 27.160133361816406, -1.1261711120605469, 27.49533462524414, 61.00445556640625, 6.49193000793457, 0.0, 198.16830444335938, 115.32379150390625, -33.73260498046875, -19.862762451171875, 2.991790771484375, -146.73724365234375, -13.14312744140625, 19.867752075195312, 152.54212951660156, 105.9080810546875, 139.65802001953125, 131.18603515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000353.npy"}
{"epoch": 0.7392670157068063, "step": 354, "batch_size": 128, "mean": 35.758384704589844, "std": 90.45724487304688, "min": -168.9398193359375, "p10": -89.96158447265626, "median": 13.770380020141602, "p90": 148.22862854003907, "max": 224.40362548828125, "pos_frac": 0.625, "sample": [-9.674179077148438, -47.97447204589844, -116.77845764160156, 131.9322052001953, 45.55267333984375, 204.484130859375, 82.53402709960938, 9.215576171875, 1.1891822814941406, 4.925022125244141, 64.937744140625, 43.78631591796875, -89.74136352539062, 15.922012329101562, 147.7032470703125, -11.770484924316406, -14.577064514160156, -0.2449951171875, 9.82098388671875, -1.845306396484375, 113.79875183105469, 145.2522735595703, 87.71243286132812, -125.05667114257812, 25.381431579589844, 13.061603546142578, -10.161331176757812, 13.814895629882812, 140.02056884765625, 197.69564819335938, -94.20748901367188, -4.810546875, 0.5449752807617188, 56.190673828125, 176.35379028320312, 26.3824462890625, 13.72586441040039, -0.66192626953125, 145.9429931640625, 140.85302734375, 5.2811279296875, -60.1292724609375, 120.76145935058594, -15.447021484375, 143.1986083984375, 143.03167724609375, -32.3905029296875, 8.8065185546875, -157.3950958251953, 1.6484832763671875, -13.5814208984375, 195.84552001953125, 117.76632690429688, 172.42291259765625, 178.60415649414062, 177.09432983398438, 127.09894561767578, 132.267333984375, 33.0096435546875, 130.50393676757812, 95.29890441894531, -26.514602661132812, -15.059837341308594, -126.13174438476562, -10.4415283203125, 140.08924865722656, 9.937114715576172, 224.40362548828125, -1.132904052734375, -3.6086273193359375, 148.64111328125, -21.907379150390625, 126.74569702148438, 7.079469680786133, 193.632568359375, 30.403884887695312, 111.14543914794922, -121.91493225097656, -18.9677734375, 1.5284862518310547, -0.09771728515625, -141.17236328125, 158.47116088867188, 40.232635498046875, -21.185867309570312, 54.5704345703125, 29.960311889648438, 141.4942169189453, -40.8565673828125, -65.88870239257812, 18.248779296875, -0.5393753051757812, 148.05184936523438, 8.114822387695312, 89.77967834472656, 25.017044067382812, -9.09002685546875, -43.649810791015625, -90.47543334960938, -140.21343994140625, 118.77241516113281, -5.5521240234375, -0.8608932495117188, 128.02188110351562, 120.55789184570312, 75.69039916992188, 3.8643798828125, -4.8690185546875, 20.920257568359375, 182.75067138671875, 145.33914184570312, 84.16362762451172, -16.26093292236328, 156.02789306640625, -134.71961975097656, 76.0447998046875, -168.9398193359375, 126.460693359375, -129.0194091796875, -120.175537109375, 128.44325256347656, 24.749252319335938, -71.97348022460938, 12.703292846679688, -50.7996826171875, -40.05189514160156, 47.704071044921875, 24.456085205078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000354.npy"}
{"epoch": 0.7413612565445026, "step": 355, "batch_size": 128, "mean": 54.49286651611328, "std": 83.27344512939453, "min": -203.6497802734375, "p10": -33.404605102539065, "median": 39.286766052246094, "p90": 166.8793197631836, "max": 207.22998046875, "pos_frac": 0.7890625, "sample": [-2.0871047973632812, -203.6497802734375, 168.48995971679688, 46.547760009765625, 14.766571044921875, 11.822906494140625, 76.1072998046875, 1.4453125, 42.943878173828125, 61.3511962890625, 150.3543243408203, 11.31805419921875, -13.776870727539062, 129.97337341308594, 4.436977386474609, 145.42481994628906, 178.92684936523438, 148.6294708251953, -5.219032287597656, 2.10772705078125, -46.31028747558594, 18.80474853515625, 4.1567535400390625, 178.134765625, 13.2998046875, 166.39268493652344, -13.866756439208984, -5.912841796875, 38.812347412109375, 14.373138427734375, 26.29608154296875, 19.5018310546875, 40.6748046875, 180.92620849609375, 174.98077392578125, 49.29875183105469, 184.3197021484375, 39.76118469238281, 22.6019287109375, -33.384857177734375, 68.39264678955078, 149.69886779785156, 55.2911376953125, 25.651321411132812, -55.04121398925781, 104.2230224609375, -10.187957763671875, 78.61199951171875, 12.372840881347656, 0.28057861328125, 150.89657592773438, 15.131912231445312, 158.24615478515625, 13.094566345214844, 1.74169921875, 46.7623291015625, 4.325489044189453, 52.85495376586914, 47.54925537109375, -26.49795150756836, 168.42291259765625, 9.255373001098633, -33.45068359375, -13.492416381835938, 97.10441589355469, -95.32394409179688, 23.214141845703125, 207.22998046875, 168.01480102539062, 141.7494354248047, -0.18177413940429688, 174.15618896484375, -149.010498046875, 138.24676513671875, 0.2528114318847656, 66.97517395019531, 138.2798309326172, 80.43240356445312, 186.1588134765625, 67.83262634277344, 104.54623413085938, 148.49838256835938, 1.607940673828125, 1.26361083984375, 33.713287353515625, -54.0284423828125, 41.239532470703125, -4.744598388671875, 32.1553955078125, 33.596588134765625, 137.03793334960938, -100.19729614257812, -141.66799926757812, 2.345123291015625, -9.154632568359375, 41.293212890625, 19.449981689453125, 165.89752197265625, -37.088714599609375, 152.9554443359375, 38.508819580078125, -113.83406829833984, 198.03521728515625, 118.54257202148438, 174.96926879882812, 153.38104248046875, 155.33570861816406, 64.1937255859375, -81.99468994140625, -14.15130615234375, 11.985015869140625, 34.72029113769531, 15.815139770507812, 140.15798950195312, 123.05447387695312, 139.3017578125, 145.89431762695312, 125.6536865234375, 133.99417114257812, 120.95034790039062, 24.5902099609375, -4.93084716796875, 75.04421997070312, 164.21145629882812, 89.30413818359375, -128.47488403320312, 158.91136169433594, 31.160858154296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000355.npy"}
{"epoch": 0.743455497382199, "step": 356, "batch_size": 128, "mean": 45.9483757019043, "std": 80.01766204833984, "min": -170.8294677734375, "p10": -43.824514007568354, "median": 33.252349853515625, "p90": 148.5410354614258, "max": 180.11268615722656, "pos_frac": 0.6796875, "sample": [0.109130859375, 0.3927001953125, 144.42691040039062, -4.324432373046875, 75.52523803710938, -119.13330078125, 17.603973388671875, -46.661529541015625, 8.77653694152832, -2.364593505859375, -11.112144470214844, 25.076904296875, 26.388427734375, 107.59979248046875, -8.473274230957031, 3.0564708709716797, -11.54058837890625, -19.473876953125, 4.998291015625, 112.30825805664062, -55.748626708984375, 140.19271850585938, 36.328277587890625, 33.794189453125, -37.07780456542969, 172.631103515625, -113.41523742675781, 135.27374267578125, 3.529449462890625, 159.7606201171875, 35.972930908203125, 163.01953125, 7.313232421875, 106.77569580078125, 145.4972381591797, 153.45835876464844, 101.59658813476562, 47.379547119140625, 170.50755310058594, 113.24589538574219, -144.47373962402344, 143.96481323242188, 5.546638488769531, 8.308128356933594, 92.10940551757812, 30.92120361328125, -8.006301879882812, 174.63470458984375, 154.06210327148438, -6.798900604248047, 75.0255126953125, 59.396575927734375, 149.72459411621094, 171.605224609375, -0.017747879028320312, 38.34991455078125, -76.36611938476562, 136.87933349609375, 117.40412902832031, 58.806549072265625, 140.90634155273438, -131.1045379638672, 144.49302673339844, 83.11532592773438, 139.5825653076172, 139.1591796875, 135.56900024414062, -8.969032287597656, 1.8541030883789062, -10.950149536132812, -8.395156860351562, 23.173858642578125, 108.13546752929688, 24.370933532714844, -15.36407470703125, 28.57506561279297, 121.65080261230469, -0.5777587890625, 148.2871551513672, -85.17442321777344, 44.93158721923828, -35.003665924072266, 175.93777465820312, 32.71051025390625, -1.141448974609375, 149.1334228515625, 130.87149047851562, 103.70481872558594, 34.338558197021484, -0.08001708984375, -88.05294036865234, -4.201900482177734, -10.937896728515625, -170.8294677734375, 78.05220031738281, 15.994354248046875, 49.3929443359375, 0.0, 92.19271850585938, 50.985107421875, 118.58168029785156, 37.942222595214844, -7.789863586425781, -19.075347900390625, 0.5562610626220703, 18.03900146484375, 29.920040130615234, 127.84408569335938, -1.97027587890625, -83.79403686523438, 146.10794067382812, 180.11268615722656, 140.15667724609375, 119.31539916992188, -42.60865020751953, 154.20977783203125, 137.05020141601562, -46.9385986328125, 39.7218017578125, 136.55560302734375, -34.7327880859375, 34.93714904785156, -3.090606689453125, -127.20342254638672, 54.03082275390625, -11.907711029052734, 20.931442260742188, 133.864501953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000356.npy"}
{"epoch": 0.7455497382198953, "step": 357, "batch_size": 128, "mean": 51.19755554199219, "std": 85.56366729736328, "min": -174.14149475097656, "p10": -33.3188720703125, "median": 25.096245765686035, "p90": 169.22173767089842, "max": 242.17657470703125, "pos_frac": 0.7265625, "sample": [29.831085205078125, 147.26107788085938, -15.580289840698242, 1.74224853515625, 154.800537109375, 43.038055419921875, 0.0, -10.3509521484375, 105.78373718261719, 5.291717529296875, 149.6719207763672, -141.4463348388672, 4.651166915893555, 242.17657470703125, 200.08282470703125, 207.49609375, 9.546916961669922, -3.1539306640625, 26.452529907226562, 51.3602294921875, 15.7423095703125, 16.763427734375, 98.40022277832031, 149.913330078125, 155.16973876953125, -153.35922241210938, 219.39236450195312, 21.894615173339844, 27.279918670654297, 175.48895263671875, 22.96630859375, 13.12921142578125, -19.213760375976562, 7.849428176879883, 202.75527954101562, 61.286502838134766, -16.527557373046875, -18.92688751220703, 5.693450927734375, 118.24567413330078, 109.2608642578125, 197.473876953125, 61.4991455078125, 116.76724243164062, 132.77694702148438, -33.185516357421875, 6.8115234375, -135.47640991210938, 160.8419189453125, 8.678985595703125, -32.18157958984375, 34.301551818847656, 45.59527587890625, 3.66021728515625, 110.68894958496094, 128.3603973388672, 177.633544921875, 168.24264526367188, 35.98101806640625, 0.260589599609375, 142.93978881835938, 19.545257568359375, 29.51708984375, 23.080169677734375, 47.24017333984375, -7.6653594970703125, 143.9581756591797, 148.20346069335938, 77.636962890625, 33.659942626953125, -12.5084228515625, -72.54302978515625, -33.630035400390625, 4.5846099853515625, 2.5396881103515625, 146.0734405517578, 74.28996276855469, -59.618499755859375, 19.892105102539062, -35.3114013671875, -51.65069580078125, 4.9700469970703125, 9.465530395507812, 26.697998046875, 182.253662109375, 106.50906372070312, 156.0721435546875, -40.488189697265625, -47.84663391113281, -1.4695663452148438, -10.525611877441406, 14.567840576171875, 56.25569152832031, -24.66583251953125, 79.63992309570312, 140.61793518066406, -1.50054931640625, -108.37725830078125, 0.0, 171.50628662109375, -132.5988006591797, 0.0, 62.503814697265625, -3.96319580078125, 163.27923583984375, 22.491792678833008, 15.729896545410156, 95.04548645019531, 158.20358276367188, 140.31765747070312, 172.72415161132812, -174.14149475097656, 154.82858276367188, 57.648590087890625, 2.4511947631835938, 97.66378784179688, -4.9403076171875, 115.68995666503906, 23.739961624145508, 190.86065673828125, 184.65386962890625, 21.351383209228516, -4.821159362792969, -22.910446166992188, 144.33657836914062, -2.4927291870117188, 3.2750396728515625, 107.85516357421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000357.npy"}
{"epoch": 0.7476439790575916, "step": 358, "batch_size": 128, "mean": 56.77861404418945, "std": 87.36698150634766, "min": -175.01925659179688, "p10": -39.279184722900375, "median": 36.040496826171875, "p90": 167.66117248535156, "max": 201.8477783203125, "pos_frac": 0.7421875, "sample": [1.021728515625, 2.4810028076171875, 98.06622314453125, -24.792800903320312, 16.849853515625, 84.87654113769531, -35.398193359375, 115.18341064453125, 179.56964111328125, 201.8477783203125, 11.340564727783203, 6.252803802490234, 20.444198608398438, 153.2881317138672, 167.56292724609375, 153.09640502929688, 18.88885498046875, 180.71591186523438, 170.50062561035156, 35.83563232421875, -95.87367248535156, 139.48681640625, -62.84210205078125, -120.443359375, 52.35595703125, -16.896194458007812, 124.53392028808594, -3.3375244140625, 174.22943115234375, -11.50689697265625, 15.012100219726562, 114.74971008300781, 48.47351837158203, 123.77262878417969, 27.4490966796875, -6.32012939453125, 200.08511352539062, -19.126129150390625, 33.6129150390625, 122.25288391113281, -117.58610534667969, 159.87338256835938, 158.23028564453125, 170.33480834960938, 29.34754180908203, 155.38497924804688, 123.06175994873047, 150.35511779785156, 153.79177856445312, -0.04754638671875, -31.6131591796875, -13.222274780273438, -6.890289306640625, 25.549720764160156, 145.52488708496094, -11.016983032226562, 160.9615478515625, 162.54251098632812, -61.94291687011719, 198.880859375, -70.22793579101562, -175.01925659179688, 138.58349609375, -15.008949279785156, 6.2635650634765625, 169.3848876953125, 89.73724365234375, 29.73255157470703, 71.37347412109375, 10.8514404296875, 56.42930603027344, 12.1859130859375, 18.364166259765625, 30.836029052734375, 4.617218017578125, -116.67181396484375, 145.9617462158203, 88.3941650390625, -14.029060363769531, 180.15554809570312, 91.78701782226562, -115.2513427734375, 127.14259338378906, 48.532470703125, 167.89041137695312, 36.245361328125, 135.6827392578125, 96.86322021484375, 51.84654998779297, 166.49972534179688, 11.268089294433594, 186.09185791015625, 134.42459106445312, 104.73490905761719, 93.746826171875, 80.91860961914062, -29.605010986328125, 9.725860595703125, 8.81866455078125, 137.5626220703125, 141.0428466796875, 32.33099365234375, 125.36909484863281, 167.07977294921875, 173.69622802734375, 166.42559814453125, 156.15225219726562, -165.97247314453125, -15.908027648925781, -69.08248901367188, -121.29801940917969, 6.409965515136719, 131.6073760986328, 3.782562255859375, 27.15546417236328, -33.101097106933594, 7.801795959472656, 13.735939025878906, -0.8880996704101562, 32.68487548828125, 0.0, 113.68635559082031, 61.744415283203125, 43.9422607421875, 143.34652709960938, -48.33483123779297, 31.22998046875, -12.635482788085938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000358.npy"}
{"epoch": 0.749738219895288, "step": 359, "batch_size": 128, "mean": 53.05839157104492, "std": 82.35993957519531, "min": -199.35211181640625, "p10": -30.440655517578126, "median": 34.572265625, "p90": 161.98709106445312, "max": 218.2860107421875, "pos_frac": 0.75, "sample": [9.959587097167969, -55.9205322265625, 135.4571533203125, 176.20358276367188, 86.14596557617188, -88.54989624023438, 78.989501953125, 16.00365447998047, 33.925323486328125, 62.057037353515625, 36.04279708862305, 161.9384765625, 60.130733489990234, -1.6340713500976562, -13.08453369140625, -38.172760009765625, 112.5763168334961, 53.12040710449219, 22.468994140625, 18.6334228515625, 103.08313751220703, -11.709121704101562, 142.99020385742188, 168.0855712890625, -66.82144165039062, 130.21029663085938, 9.74639892578125, 35.12005615234375, 81.9146728515625, -180.168701171875, -1.5806884765625, 196.64804077148438, 6.65570068359375, 15.740974426269531, 2.3236541748046875, 87.72152709960938, -8.07058334350586, 139.6409912109375, 159.32278442382812, 165.197021484375, -26.891921997070312, -118.92160034179688, 136.7640380859375, 3.108123779296875, 120.4215087890625, 82.42630004882812, 7.081451416015625, 132.73043823242188, 132.24346923828125, 126.0760726928711, 147.45587158203125, 2.8203506469726562, 212.71627807617188, -53.07579040527344, 138.6507568359375, 0.35809326171875, 65.02398681640625, -173.64215087890625, 98.9018325805664, 41.8660888671875, 186.09417724609375, 105.99594116210938, -199.35211181640625, 70.76931762695312, 13.3818359375, 4.3398895263671875, -11.7325439453125, 52.577667236328125, 145.01669311523438, 89.322998046875, -3.127044677734375, 26.70361328125, -30.7281494140625, 34.02447509765625, 93.50720977783203, 143.1825408935547, -8.3321533203125, 131.33993530273438, 12.099594116210938, 141.94717407226562, -2.288177490234375, -22.441741943359375, 6.87957763671875, 64.8302001953125, 24.290817260742188, 18.9080810546875, 18.239593505859375, 118.934326171875, 62.07867431640625, 7.867156982421875, 165.2656707763672, 14.114700317382812, 165.8502197265625, 129.26263427734375, 22.00665283203125, -30.31744384765625, -111.15391540527344, 129.10159301757812, -16.82647132873535, -0.2999992370605469, 29.004058837890625, 181.54977416992188, 31.044158935546875, -75.34632873535156, 26.471939086914062, -22.85150146484375, 167.24496459960938, -15.71038818359375, 58.6522216796875, 139.74758911132812, 136.64456176757812, 113.51677703857422, 6.81036376953125, 112.166748046875, -9.076751708984375, 173.77529907226562, 161.02716064453125, 77.9095458984375, 86.33354187011719, -3.5308609008789062, -7.033378601074219, 218.2860107421875, 28.838836669921875, 31.86887550354004, 162.10052490234375, 8.869918823242188, 127.10537719726562, -33.733154296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000359.npy"}
{"epoch": 0.7518324607329843, "step": 360, "batch_size": 128, "mean": 47.775657653808594, "std": 74.1895523071289, "min": -202.56939697265625, "p10": -19.636337280273438, "median": 26.771984100341797, "p90": 154.0329116821289, "max": 217.27447509765625, "pos_frac": 0.734375, "sample": [67.0966796875, 120.79707336425781, -66.91403198242188, 158.69778442382812, -128.5227813720703, -9.079877853393555, -67.68486022949219, 15.76275634765625, 217.27447509765625, -46.96253967285156, 29.5926513671875, 105.02120971679688, -20.478256225585938, 138.61175537109375, 45.296142578125, 113.52664184570312, 15.474685668945312, 138.15631103515625, -19.404296875, 57.99858093261719, 170.01510620117188, 37.93653869628906, 14.54632568359375, 104.68008422851562, 123.69292449951172, 7.0782623291015625, 3.6149749755859375, 21.65655517578125, -1.5972900390625, 163.69406127929688, 35.096832275390625, 181.21205139160156, 96.29495239257812, 15.95062255859375, 53.07470703125, 6.20855712890625, -21.983245849609375, 133.17010498046875, -31.1820068359375, 108.93247985839844, 92.3955078125, 77.69876098632812, 144.3851318359375, 25.673248291015625, 99.47604370117188, 21.713245391845703, 17.784820556640625, 138.19200134277344, -8.21697998046875, 39.15087890625, -12.286758422851562, 138.30307006835938, 121.68905639648438, 57.4000244140625, 189.83126831054688, 0.220916748046875, 169.64744567871094, -38.63525390625, 146.78793334960938, 12.315437316894531, 77.80203247070312, 40.044921875, 154.141357421875, 116.85359954833984, 18.441009521484375, 1.8022689819335938, 168.75479125976562, 16.3912353515625, 50.74864196777344, -111.0675048828125, 0.3672943115234375, 74.99859619140625, 1.5508155822753906, 161.48777770996094, -18.0037841796875, 20.516597747802734, 60.42589569091797, 165.8853759765625, 142.12545776367188, 135.2767333984375, -9.879196166992188, 116.10113525390625, 6.322160720825195, -13.19279670715332, -202.56939697265625, 109.307373046875, 36.37225341796875, 8.360443115234375, -20.177764892578125, 8.898895263671875, 149.32199096679688, -18.047393798828125, 180.83993530273438, 53.88085174560547, -0.7434844970703125, -7.41064453125, 5.65020751953125, -124.14872741699219, 9.27972412109375, -1.0827751159667969, 50.16996765136719, 0.0, 23.373992919921875, 18.54754638671875, 153.98643493652344, -7.948734283447266, 63.2598876953125, 45.400634765625, -1.2629547119140625, 5.79460334777832, 55.445953369140625, -6.510009765625, -4.968879699707031, 119.8355712890625, 14.763153076171875, 115.27496337890625, -26.394004821777344, -18.81890869140625, 0.0, 10.824943542480469, 56.63599395751953, 27.87071990966797, 35.26106262207031, -13.177742004394531, -15.227825164794922, 17.925445556640625, 156.00518798828125, 153.712158203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000360.npy"}
{"epoch": 0.7539267015706806, "step": 361, "batch_size": 128, "mean": 64.15248107910156, "std": 89.71261596679688, "min": -172.5322265625, "p10": -37.51457366943359, "median": 71.3184585571289, "p90": 165.9190444946289, "max": 234.2999267578125, "pos_frac": 0.7265625, "sample": [134.03848266601562, -8.51422119140625, -104.49520874023438, 35.84368896484375, 32.89202880859375, -5.088409423828125, 1.5170021057128906, -25.7554931640625, 156.59954833984375, 49.96002197265625, 118.19159698486328, 15.953521728515625, 1.3336257934570312, 203.76580810546875, -1.3958187103271484, 155.92063903808594, -3.373777389526367, 14.779525756835938, -30.97845458984375, 1.1095352172851562, 156.078369140625, 185.31887817382812, -108.73677062988281, 151.44036865234375, -3.042083740234375, 183.76809692382812, 129.32542419433594, 134.4922332763672, 95.06087493896484, 188.2548828125, 10.402862548828125, -8.669235229492188, 113.28740692138672, 85.18156433105469, 15.047966003417969, 142.676513671875, -140.80426025390625, 92.51751708984375, 126.0613021850586, 100.05570983886719, 188.26605224609375, 196.64462280273438, 24.365036010742188, 87.66566467285156, 200.8594970703125, 24.376708984375, -110.68035888671875, 29.730087280273438, 86.60238647460938, -10.327617645263672, -1.1278438568115234, 32.89073181152344, 108.09161376953125, 0.0, 63.48750305175781, 14.7340087890625, 136.27296447753906, 96.18048095703125, 3.9111175537109375, -22.752655029296875, 34.99171447753906, 139.1689910888672, -20.484907150268555, 162.5299072265625, 108.01031494140625, -48.63336181640625, 54.4718017578125, 99.60840606689453, 24.625885009765625, -72.63247680664062, 80.41094970703125, -3.89276123046875, -124.6300048828125, -36.902923583984375, 116.5536880493164, 0.0, 59.657470703125, -38.94175720214844, 49.2774658203125, -10.233001708984375, 203.90341186523438, 152.88775634765625, 178.31155395507812, 79.1494140625, -171.23406982421875, 140.086669921875, -9.070404052734375, -52.09857177734375, 234.2999267578125, 133.29946899414062, -114.03416442871094, 92.29466247558594, -126.24462890625, 157.95489501953125, 174.46426391601562, 108.44712829589844, 113.72486877441406, 38.92730712890625, 151.43153381347656, 129.2685546875, 160.33941650390625, -7.027345657348633, 202.68959045410156, 149.25894165039062, 159.36456298828125, 150.80780029296875, 140.4493408203125, 45.11944580078125, 24.83148193359375, 148.42718505859375, 163.90106201171875, 57.157142639160156, -16.0567626953125, 106.93772888183594, 169.33920288085938, 134.65267944335938, -2.171356201171875, 154.57000732421875, 164.45326232910156, 107.45245361328125, 8.088455200195312, 132.63233947753906, -4.6459197998046875, 103.86865234375, 34.95508575439453, 18.08869171142578, 148.63058471679688, -172.5322265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000361.npy"}
{"epoch": 0.7560209424083769, "step": 362, "batch_size": 128, "mean": 54.72718811035156, "std": 83.46868896484375, "min": -175.34585571289062, "p10": -31.32783966064453, "median": 32.96535110473633, "p90": 154.66899108886716, "max": 247.35086059570312, "pos_frac": 0.75, "sample": [19.685325622558594, 91.74713134765625, -10.146076202392578, 62.023284912109375, 146.990478515625, 71.60145568847656, 136.14453125, -13.196807861328125, 85.17047119140625, 13.540130615234375, 117.60269165039062, -175.34585571289062, -140.91366577148438, -102.43602752685547, 205.96011352539062, 6.588312149047852, 151.1214141845703, -7.60638427734375, 41.72264099121094, 121.015625, 94.17959594726562, 143.3446044921875, 247.35086059570312, 49.2030029296875, 16.784423828125, 163.0686798095703, 126.2335205078125, -20.857177734375, -159.7138671875, 97.82610321044922, -24.789337158203125, -60.842369079589844, 23.541229248046875, -0.009521484375, -78.41991424560547, 132.1082763671875, 19.95270538330078, -12.992115020751953, 158.84799194335938, 14.119743347167969, 196.5457763671875, 101.10845947265625, 182.99969482421875, 146.9875946044922, 166.454345703125, 150.76174926757812, 131.1684112548828, -1.173736572265625, -15.9156494140625, 9.639045715332031, -37.79193115234375, 137.1788330078125, 141.1810302734375, -34.418914794921875, 2.0632247924804688, 4.19097900390625, -33.407470703125, 183.03515625, -4.68463134765625, 144.23260498046875, 148.62060546875, -20.651275634765625, 31.3841552734375, 126.655029296875, 110.96409606933594, 118.05084228515625, 21.752044677734375, 18.899276733398438, 115.13749694824219, 49.57452392578125, 169.59066772460938, 152.87799072265625, 139.80526733398438, 68.53671264648438, -3.472309112548828, 190.7684326171875, 116.02177429199219, 5.530937194824219, 119.63345336914062, 130.4058074951172, -38.80206298828125, 139.76773071289062, -12.022396087646484, 6.9220428466796875, -19.9737548828125, -97.97119140625, -30.436569213867188, 182.82456970214844, 8.766143798828125, 17.71826934814453, 26.10321044921875, 3.1813507080078125, -10.444290161132812, 13.532302856445312, 22.74029541015625, 12.828292846679688, 124.03788757324219, 41.30177307128906, 12.832138061523438, 1.96295166015625, -9.487930297851562, 144.27960205078125, 141.02548217773438, 21.68963623046875, 2.9237747192382812, 110.9674072265625, -142.86569213867188, 13.9659423828125, 25.82843017578125, 21.37541961669922, 175.07411193847656, 8.68647575378418, 121.81369018554688, 98.67222595214844, 32.09002685546875, -10.674674987792969, 32.053131103515625, 173.2841796875, 46.377838134765625, 33.840675354003906, -28.15447998046875, 83.48822784423828, 131.67047119140625, 151.20404052734375, 59.31329345703125, -89.586181640625, 76.46807861328125, 44.44270324707031], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000362.npy"}
{"epoch": 0.7581151832460733, "step": 363, "batch_size": 128, "mean": 61.934104919433594, "std": 84.77169036865234, "min": -192.89654541015625, "p10": -15.71838989257812, "median": 46.62335968017578, "p90": 169.36343841552736, "max": 273.35687255859375, "pos_frac": 0.78125, "sample": [147.9549560546875, -3.595977783203125, -12.65966796875, 168.16546630859375, 149.00177001953125, 67.9873046875, -1.0913219451904297, -123.29190063476562, 92.0384521484375, 121.48050689697266, -11.683807373046875, 7.869667053222656, 150.2139892578125, 93.45057678222656, 7.9600982666015625, 71.724609375, 67.88156127929688, 189.2365264892578, -6.64727783203125, -25.321382522583008, 23.124755859375, 134.58184814453125, 187.28805541992188, 65.43914794921875, 21.380264282226562, 176.921630859375, 187.74185180664062, 219.8983154296875, 12.745498657226562, 39.37409210205078, 149.37889099121094, 3.981353759765625, -13.15468978881836, -2.7021026611328125, 108.61949157714844, 1.6889667510986328, -2.3986358642578125, 63.7010498046875, 150.12074279785156, 1.95379638671875, 5.914741516113281, 21.936668395996094, 97.58868408203125, 124.6865005493164, 73.88697814941406, 81.3687744140625, -13.173995971679688, 107.924072265625, 169.3532257080078, 106.28160095214844, 107.3780517578125, 20.3554630279541, 273.35687255859375, 157.83062744140625, -128.99612426757812, 63.224609375, -2.421478271484375, 171.6693115234375, 34.57904052734375, 117.86029052734375, 26.10687255859375, 37.38197326660156, 167.6754913330078, 131.5249481201172, 182.5137939453125, 0.8570480346679688, 180.553466796875, 99.02764892578125, 39.1494140625, 6.338443756103516, 140.30630493164062, 4.46563720703125, 176.41180419921875, 224.38888549804688, 8.476150512695312, 157.53530883789062, -88.78921508789062, -11.139968872070312, 73.35197448730469, 57.073883056640625, 2.7682037353515625, 12.271453857421875, -7.135223388671875, 11.896484375, 49.97499084472656, -14.34405517578125, -3.6845703125, -5.59417724609375, -42.88116455078125, 11.922164916992188, 96.94488525390625, -126.6746826171875, 137.90768432617188, 9.799301147460938, 11.466201782226562, 169.38726806640625, 140.05519104003906, -18.9251708984375, 163.3359375, 95.57217407226562, 23.491241455078125, 124.05799865722656, -144.4739227294922, 165.36517333984375, 171.06072998046875, 131.25869750976562, 25.903526306152344, 51.93084716796875, 161.8047332763672, 60.49627685546875, -192.89654541015625, -36.98048400878906, -41.57075500488281, 0.7317581176757812, 20.3699951171875, 39.303619384765625, 8.09014892578125, 10.827651977539062, 131.0277862548828, 167.06698608398438, 164.34945678710938, 161.12515258789062, 31.65459442138672, 142.66925048828125, -67.37553405761719, 43.271728515625, 23.929931640625, -27.159027099609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000363.npy"}
{"epoch": 0.7602094240837697, "step": 364, "batch_size": 128, "mean": 45.611839294433594, "std": 85.04479217529297, "min": -204.018310546875, "p10": -29.920303344726555, "median": 31.42595672607422, "p90": 153.00510559082028, "max": 240.57763671875, "pos_frac": 0.671875, "sample": [65.69781494140625, -3.45782470703125, 97.27459716796875, 45.52214431762695, 112.35801696777344, 14.214309692382812, 128.249267578125, 118.876708984375, 121.06661987304688, 37.21910095214844, 47.43223571777344, 107.19415283203125, -8.98809814453125, -7.726287841796875, -15.197509765625, 80.34103393554688, 55.80084228515625, 31.953125, 4.556488037109375, 122.71524047851562, -148.9691619873047, -0.1638641357421875, 110.35798645019531, -156.61431884765625, 34.464111328125, -76.65261840820312, 183.60952758789062, 137.75582885742188, -9.263236999511719, 139.15286254882812, 5.2669677734375, 17.830368041992188, 57.665496826171875, 6.184276580810547, -23.53759765625, -4.81170654296875, -9.604095458984375, 136.7942352294922, -0.15191650390625, 140.796875, 112.73715209960938, 8.56842041015625, -180.22052001953125, 62.44874572753906, 132.80526733398438, 172.88473510742188, -57.71145248413086, 86.4105224609375, 17.4229736328125, -26.825180053710938, 150.81845092773438, 158.1072998046875, -9.973663330078125, 55.26910400390625, -11.613666534423828, -6.8186187744140625, 169.993408203125, -1.430429458618164, -204.018310546875, 216.3418731689453, 127.42529296875, -11.49078369140625, -8.6273193359375, -89.87460327148438, 34.24882507324219, 126.98341369628906, -26.1826171875, 172.66375732421875, -15.872528076171875, 7.772216796875, 36.0672607421875, 149.17320251464844, 80.28472900390625, -14.031982421875, 119.98245239257812, 9.31109619140625, 173.8330078125, 52.022239685058594, 16.094459533691406, 142.00006103515625, 117.72268676757812, -9.0042724609375, 27.616226196289062, -6.199310302734375, -5.5906829833984375, 85.85015869140625, 20.90300750732422, -22.73974609375, 46.361419677734375, 206.838623046875, 129.39874267578125, 12.989677429199219, -34.23773193359375, 30.898788452148438, 145.544921875, 127.7598876953125, 3.054950714111328, -0.7652320861816406, -28.069976806640625, 240.57763671875, 33.291534423828125, 161.95359802246094, -104.3043212890625, 204.20855712890625, 126.32205200195312, -51.89154052734375, 55.833099365234375, 26.081817626953125, 1.940826416015625, -17.571258544921875, 130.95846557617188, 1.60443115234375, -14.907630920410156, 182.7527618408203, 8.294075012207031, -3.85504150390625, 4.034698486328125, -112.9161376953125, 90.7169189453125, -99.98751831054688, 9.591625213623047, 58.03057861328125, 90.62005615234375, 11.97454833984375, -68.11380004882812, 105.0364990234375, 215.35203552246094, 82.16477966308594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000364.npy"}
{"epoch": 0.762303664921466, "step": 365, "batch_size": 128, "mean": 61.29475402832031, "std": 82.48385620117188, "min": -164.32034301757812, "p10": -25.117010498046874, "median": 48.44734001159668, "p90": 170.65591735839843, "max": 205.0859375, "pos_frac": 0.7578125, "sample": [0.0, 3.041278839111328, 106.81558227539062, 4.112762451171875, 166.01681518554688, 187.26104736328125, 13.548919677734375, 64.66799926757812, 0.0, 30.291259765625, 22.2138671875, 6.538909912109375, 12.520828247070312, 174.26861572265625, 14.702308654785156, 10.70513916015625, 19.9888916015625, 79.81633758544922, 14.067581176757812, 161.56333923339844, 43.79315185546875, 162.27694702148438, -7.432708740234375, 20.267669677734375, 63.22102355957031, 74.89190673828125, -22.790130615234375, 155.78793334960938, 136.96954345703125, 154.27194213867188, 154.64901733398438, -62.92725372314453, -26.74725341796875, 128.37106323242188, -124.817138671875, -75.66131591796875, 138.22341918945312, 48.30133819580078, 136.6435089111328, 81.80282592773438, -17.94108009338379, 141.90960693359375, 179.96063232421875, 136.135986328125, -29.383880615234375, 182.8092041015625, -0.86328125, 128.29090881347656, 48.75953674316406, 158.22683715820312, 138.24278259277344, -23.011093139648438, 82.177734375, -1.8135528564453125, 15.764801025390625, 8.08758544921875, 118.72941589355469, 163.87599182128906, 156.75299072265625, -28.429000854492188, 24.2825927734375, 145.500732421875, -36.526519775390625, 4.6336669921875, 5.340297698974609, 91.6112060546875, 127.55253601074219, 137.88693237304688, 81.67572021484375, 149.52645874023438, 187.1239013671875, -14.360519409179688, 180.27374267578125, 1.1952362060546875, 9.540283203125, 188.01211547851562, 140.2303466796875, -137.32119750976562, 171.6536865234375, 2.8006591796875, 179.19235229492188, 2.62213134765625, 197.08560180664062, 5.392669677734375, 31.821271896362305, -17.184768676757812, 147.97268676757812, 54.24772644042969, 3.781871795654297, -17.409019470214844, 66.09010314941406, 48.59334182739258, 81.23951721191406, 170.22830200195312, -6.928104400634766, 5.6316680908203125, 1.22039794921875, -17.991531372070312, 138.04714965820312, -7.90570068359375, 25.291976928710938, -111.43458557128906, -16.5386962890625, -4.405731201171875, 140.57656860351562, 0.0, 39.02735900878906, 11.139480590820312, 138.32423400878906, 182.434814453125, 71.46307373046875, 113.22586059570312, 5.739788055419922, -24.4183349609375, -50.378814697265625, -164.32034301757812, 138.71343994140625, 179.9686737060547, -37.137420654296875, 136.71823120117188, 34.93635559082031, -72.31561279296875, 205.0859375, 57.82958984375, 103.5050048828125, 136.9177703857422, 144.02859497070312, 105.85238647460938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000365.npy"}
{"epoch": 0.7643979057591623, "step": 366, "batch_size": 128, "mean": 57.970436096191406, "std": 83.00066375732422, "min": -196.42315673828125, "p10": -33.263424682617185, "median": 47.063026428222656, "p90": 164.3260711669922, "max": 216.2652587890625, "pos_frac": 0.734375, "sample": [34.70909118652344, 32.59541320800781, 216.2652587890625, 143.11122131347656, 124.54796600341797, 135.81027221679688, 164.66677856445312, 127.23098754882812, 107.4803237915039, 183.91848754882812, 167.06588745117188, 135.0627899169922, 116.91622161865234, -39.522064208984375, 31.217288970947266, 119.07716369628906, 113.00224304199219, 77.18894958496094, -139.38284301757812, 46.9290771484375, -196.42315673828125, 197.3388671875, -138.65264892578125, 17.333580017089844, 49.0655517578125, 124.70663452148438, 137.64695739746094, 76.42987060546875, 147.98080444335938, 91.16818237304688, -13.818023681640625, 139.73770141601562, 40.425201416015625, -33.90093994140625, 159.1677703857422, -176.72036743164062, -2.08477783203125, 116.301513671875, 16.329833984375, 24.89593505859375, 3.4207820892333984, -11.530975341796875, -6.7427978515625, -17.22380828857422, -8.355560302734375, 101.38014221191406, 145.43130493164062, -32.990203857421875, 47.19697570800781, -12.782958984375, 88.5174560546875, 33.812530517578125, 12.422454833984375, 69.58807373046875, 109.96537780761719, -16.7899169921875, 30.000396728515625, 18.830062866210938, 33.121307373046875, 17.912086486816406, -0.72576904296875, 143.0150604248047, 112.10414123535156, 3.159809112548828, 2.8325729370117188, 170.62225341796875, 73.75746154785156, 81.66461181640625, 96.53003692626953, -67.8006591796875, 138.71377563476562, 171.21356201171875, -58.922210693359375, 0.0, -25.961212158203125, 170.6235809326172, 58.516845703125, 119.44232177734375, 125.19677734375, -1.643951416015625, 6.887908935546875, 43.13429260253906, 181.21484375, 83.64178466796875, -5.581329345703125, 54.74365234375, 26.453338623046875, 191.5303955078125, -7.179779052734375, -8.633811950683594, -4.542236328125, -39.40777587890625, 107.50576782226562, 12.009536743164062, 70.80120849609375, 135.43521118164062, 103.521728515625, -5.44110107421875, 57.7479248046875, 158.03111267089844, 143.92857360839844, 16.753463745117188, 5.11785888671875, 191.44992065429688, 149.32894897460938, 29.784019470214844, 124.54765319824219, -21.506378173828125, 196.20590209960938, 0.9037799835205078, 3.856353759765625, 126.02474212646484, 168.98907470703125, 10.69769287109375, 14.165779113769531, -40.742401123046875, 154.2476043701172, -141.74510192871094, 127.56266784667969, 32.112060546875, 135.9755859375, 13.90899658203125, -0.11187744140625, 148.01309204101562, -42.569580078125, -10.62847900390625, 164.1800537109375, -34.44775390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000366.npy"}
{"epoch": 0.7664921465968586, "step": 367, "batch_size": 128, "mean": 59.99263000488281, "std": 79.73580932617188, "min": -130.8057861328125, "p10": -25.613371276855467, "median": 42.31800842285156, "p90": 169.03941192626954, "max": 197.62191772460938, "pos_frac": 0.765625, "sample": [132.7135009765625, 28.793060302734375, -17.75006103515625, 180.33523559570312, -1.909820556640625, 18.404815673828125, 46.067195892333984, -130.8057861328125, -8.157867431640625, 101.06536865234375, 146.1641845703125, 21.560407638549805, 5.874153137207031, -1.5004425048828125, 90.03977966308594, 187.31805419921875, 29.554641723632812, 168.80722045898438, -0.80560302734375, 9.20977783203125, 9.096832275390625, 194.52981567382812, 102.36758422851562, -41.75909423828125, 6.660896301269531, 183.33554077148438, -120.205078125, -12.84716796875, 58.79945373535156, -18.125747680664062, -26.114028930664062, 81.90821838378906, 3.306814193725586, -25.3988037109375, 144.36297607421875, -16.54498291015625, 87.34738159179688, 135.4883575439453, 128.1251220703125, -95.55499267578125, 157.5828857421875, 99.82859802246094, 9.839033126831055, 118.668701171875, 163.56854248046875, 2.2281265258789062, -75.39752197265625, 3.67523193359375, -8.993621826171875, -13.625473022460938, 48.354888916015625, 8.689956665039062, 99.23471069335938, 23.83416748046875, 143.05873107910156, 153.03421020507812, 139.39230346679688, 113.71318817138672, 82.08134460449219, -27.323822021484375, 12.9754638671875, -27.70391845703125, 189.86712646484375, 121.72463989257812, 15.360069274902344, 172.11953735351562, -26.934036254882812, 130.1964111328125, 197.62191772460938, 30.526844024658203, 69.35226440429688, 23.913360595703125, 39.288787841796875, 160.13925170898438, 170.775390625, 157.21359252929688, 8.72216796875, -24.45367431640625, 17.690338134765625, -44.946380615234375, 45.34722900390625, 169.58119201660156, 153.480712890625, 163.80169677734375, 18.135498046875, 25.184722900390625, 162.59620666503906, 22.033966064453125, -67.881591796875, 105.79580688476562, 154.419189453125, 122.85049438476562, 22.55059814453125, -12.889129638671875, 12.652002334594727, -5.725658416748047, -77.17986297607422, 171.08941650390625, 60.83011245727539, 167.88491821289062, -2.0548572540283203, 177.34478759765625, 154.09658813476562, 2.29608154296875, -10.636619567871094, 87.74162292480469, 82.82110595703125, 168.16937255859375, 6.2025146484375, 174.15167236328125, 81.72163391113281, 111.90054321289062, 28.067413330078125, -123.0196762084961, 62.50013732910156, 7.9351348876953125, 2.3799285888671875, 135.695068359375, 71.86117553710938, 148.81243896484375, 13.744405746459961, 66.32745361328125, 16.25457763671875, 5.4482421875, 193.7041015625, 114.54356384277344, -20.4671630859375, 88.3074951171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000367.npy"}
{"epoch": 0.768586387434555, "step": 368, "batch_size": 128, "mean": 67.98149108886719, "std": 82.43994903564453, "min": -153.535888671875, "p10": -20.571520233154295, "median": 58.242950439453125, "p90": 165.84650573730468, "max": 254.89761352539062, "pos_frac": 0.7890625, "sample": [9.604736328125, 179.6923828125, 121.83082580566406, 149.83251953125, 19.048736572265625, 158.365478515625, 49.02696990966797, -3.338714599609375, 5.739410400390625, -8.4593505859375, 118.44277954101562, -30.307937622070312, 93.48471069335938, 190.9368896484375, 117.72866821289062, 53.5191650390625, 208.74627685546875, 16.06951904296875, 68.99722290039062, 113.93521881103516, -21.272796630859375, -5.2777099609375, 46.798583984375, -20.270973205566406, 5.481689453125, 23.106414794921875, -5.954978942871094, 5.195812225341797, 48.90415954589844, 149.94952392578125, 1.9452438354492188, 47.63043212890625, -71.86811065673828, 90.30223083496094, 13.2987060546875, -153.535888671875, 20.826461791992188, 254.89761352539062, 137.55996704101562, 127.297607421875, 30.805023193359375, 164.98255920410156, 150.9119110107422, 133.2277374267578, 35.7218017578125, 20.796356201171875, 135.79574584960938, -7.9870758056640625, 24.650033950805664, -150.03048706054688, 6.2410736083984375, 144.54815673828125, 8.110832214355469, 124.00901794433594, -36.20233917236328, -8.921661376953125, 179.99252319335938, 136.07058715820312, -52.890869140625, 118.98727416992188, 141.3302001953125, 62.52972412109375, 131.63726806640625, 149.6012725830078, 169.64102172851562, 135.48876953125, -55.11236572265625, 82.02096557617188, 29.88812255859375, -7.8383941650390625, -3.720428466796875, 181.7408447265625, 176.20831298828125, 52.52754211425781, 28.519332885742188, 135.6738739013672, 147.5980224609375, 86.25662231445312, 160.47523498535156, 141.95596313476562, 2.872467041015625, 121.37060546875, 152.9653778076172, 53.9561767578125, -10.885307312011719, 144.66769409179688, 9.67193603515625, 65.73639678955078, -9.300643920898438, 120.86941528320312, -0.9177970886230469, 11.010887145996094, 164.35137939453125, 21.23627471923828, 115.1950454711914, 41.847557067871094, 102.58570098876953, 119.21978759765625, -1.1297073364257812, 124.88935089111328, 167.09579467773438, -90.76569366455078, 92.72016906738281, 179.133056640625, -77.197265625, 30.06451416015625, -8.8468017578125, 110.21865844726562, 203.92974853515625, 174.17193603515625, 27.76171875, 149.42825317382812, 26.626144409179688, 10.071769714355469, 18.019432067871094, 145.58555603027344, 165.31109619140625, 156.26251220703125, 139.8560791015625, 117.7569580078125, 38.23583984375, -143.2803955078125, -81.78717041015625, 5.91448974609375, 99.89335632324219, 167.4175567626953, -59.490478515625, 154.16201782226562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000368.npy"}
{"epoch": 0.7706806282722513, "step": 369, "batch_size": 128, "mean": 47.00896453857422, "std": 86.9924545288086, "min": -177.3115692138672, "p10": -44.083621978759766, "median": 24.035335540771484, "p90": 154.80687103271484, "max": 226.33462524414062, "pos_frac": 0.71875, "sample": [145.5439910888672, 28.575340270996094, 139.05210876464844, -66.91903686523438, 21.076080322265625, -7.979583740234375, 88.37101745605469, 4.1424560546875, 134.95883178710938, 0.59967041015625, 146.078125, 43.76104736328125, -39.373653411865234, 69.55398559570312, 181.79934692382812, 192.54147338867188, 119.00993347167969, 135.70863342285156, 21.11517333984375, 129.98538208007812, 185.23406982421875, 2.911701202392578, 104.3880615234375, 125.55316162109375, -36.95037841796875, -158.14422607421875, 141.22735595703125, -13.92962646484375, 16.65802001953125, 3.0846405029296875, 129.81004333496094, 27.19525146484375, 153.45559692382812, 92.1353759765625, -5.58197021484375, 137.54367065429688, -41.61053466796875, 82.8515625, 131.43875122070312, 33.43894958496094, -2.3929672241210938, 6.105712890625, 153.14141845703125, -36.063873291015625, 10.345947265625, -111.06585693359375, -4.3866729736328125, 129.34591674804688, -116.32687377929688, 179.21173095703125, -124.26593017578125, -131.503662109375, 85.73455810546875, -77.04833984375, 172.43649291992188, 121.3227767944336, -112.24937438964844, -6.788154602050781, 138.62411499023438, 5.447723388671875, -131.69857788085938, -1.02362060546875, -126.91545867919922, 156.9618682861328, 165.71392822265625, 59.26232147216797, 82.37168884277344, -1.8727264404296875, 161.539794921875, -41.70091247558594, 226.33462524414062, -14.322952270507812, 26.379249572753906, 20.0694580078125, 5.44207763671875, 199.94497680664062, 62.5703125, 1.61279296875, -46.230186462402344, 130.63546752929688, 1.4125289916992188, -177.3115692138672, 7.635154724121094, 54.002685546875, 17.09600830078125, 122.93924713134766, 158.78244018554688, 147.9065704345703, -42.505218505859375, 0.0, 21.691421508789062, 138.01507568359375, 56.572540283203125, 77.26033020019531, -8.2491455078125, 1.5154571533203125, 107.22830200195312, -21.507156372070312, 45.93060302734375, 151.5266876220703, 17.34051513671875, 145.06167602539062, -1.1873779296875, -43.163665771484375, 65.604248046875, -29.993955612182617, 142.71627807617188, 15.94970703125, 151.1552734375, -70.31002807617188, 37.68921661376953, 41.27326965332031, 15.0567626953125, 146.230224609375, 153.88330078125, 2.110595703125, 18.3555908203125, 170.82005310058594, -11.808074951171875, 111.593505859375, 1.8134403228759766, 174.05105590820312, 14.3173828125, 21.218109130859375, 13.949668884277344, -4.2924957275390625, 41.73388671875, 3.02862548828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000369.npy"}
{"epoch": 0.7727748691099476, "step": 370, "batch_size": 128, "mean": 60.398094177246094, "std": 88.6111068725586, "min": -193.98895263671875, "p10": -18.439773559570312, "median": 35.963523864746094, "p90": 165.253271484375, "max": 307.9776916503906, "pos_frac": 0.78125, "sample": [113.1590576171875, -153.98245239257812, 136.78659057617188, 34.95942687988281, 37.02569580078125, 48.79866027832031, -49.24286651611328, 5.546417236328125, 122.07017517089844, 14.181015014648438, 34.88254928588867, 96.78065490722656, -11.557273864746094, 21.4178466796875, 131.13946533203125, 10.2103271484375, -0.328643798828125, 226.71429443359375, 4.600013732910156, 26.01922607421875, 132.83453369140625, 29.727670669555664, 129.80364990234375, 141.6707000732422, -17.811798095703125, -2.5481643676757812, -12.13277816772461, 145.21499633789062, 6.803462982177734, 27.96734619140625, 27.935302734375, -19.186279296875, 139.01026916503906, 32.85895919799805, 18.788223266601562, 0.9793243408203125, 221.403076171875, 91.61666870117188, 4.505756378173828, 173.8114776611328, 125.23152923583984, -157.35960388183594, 164.56732177734375, 9.491561889648438, 125.94550323486328, 99.53147888183594, 114.37033081054688, 39.69001770019531, 38.449310302734375, 0.499481201171875, 118.75263977050781, 17.915054321289062, 9.131584167480469, 128.96617126464844, 105.45526123046875, -17.754989624023438, 140.88072204589844, -11.28546142578125, 158.94650268554688, -53.000144958496094, -21.50359344482422, 157.35818481445312, -11.6011962890625, -102.93136596679688, 8.527759552001953, 152.286376953125, 206.21441650390625, 65.7901611328125, 126.91595458984375, -59.70764923095703, 4.116180419921875, -8.655303955078125, 58.2210693359375, 158.82315063476562, 168.60244750976562, -18.119842529296875, 176.769775390625, 169.35107421875, 120.10556030273438, 155.3102264404297, 171.7769775390625, 146.4375457763672, 36.967620849609375, 208.15216064453125, 142.36300659179688, 149.2798309326172, -25.02684783935547, 8.997039794921875, -155.95538330078125, 147.989501953125, 133.97268676757812, 23.228208541870117, -104.2662353515625, 24.40985107421875, 307.9776916503906, -5.078235626220703, 20.8570556640625, 32.129547119140625, 152.22354125976562, 152.33230590820312, 158.6030731201172, 169.37945556640625, 13.049285888671875, 166.85382080078125, -10.106460571289062, 93.40997314453125, 80.21737670898438, 88.97055053710938, 38.695953369140625, 63.34942626953125, 185.76016235351562, -193.98895263671875, 8.786399841308594, -11.026458740234375, 1.7921981811523438, 144.4051971435547, 13.6817626953125, 7.7307891845703125, 159.3228759765625, 15.725799560546875, 12.947660446166992, -130.53924560546875, 31.518287658691406, 151.70553588867188, -9.776084899902344, 11.549362182617188, 124.33924102783203, -10.89886474609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000370.npy"}
{"epoch": 0.774869109947644, "step": 371, "batch_size": 128, "mean": 51.49903869628906, "std": 82.48046875, "min": -202.62359619140625, "p10": -33.894285583496085, "median": 33.989952087402344, "p90": 159.3288330078125, "max": 199.48004150390625, "pos_frac": 0.734375, "sample": [-0.6802291870117188, 107.88685607910156, 48.424560546875, 143.23641967773438, -11.686874389648438, 1.0311355590820312, 183.9775390625, 0.70648193359375, 60.03733825683594, 16.73681640625, 1.2498550415039062, -188.03836059570312, 10.4822998046875, 15.635711669921875, 28.081863403320312, -9.7264404296875, -61.682861328125, 94.97935485839844, 7.482666015625, -21.735015869140625, -24.63262939453125, 101.30291748046875, -9.186443328857422, 165.00570678710938, 65.39146423339844, 107.742431640625, -0.24781036376953125, 48.7225341796875, -21.787338256835938, 38.488922119140625, 131.61572265625, 163.86131286621094, -68.79011535644531, 17.000228881835938, -202.62359619140625, 76.94989013671875, -39.54119873046875, 1.7130584716796875, 123.3629150390625, 23.248687744140625, 161.67428588867188, 12.326004028320312, -98.57405090332031, 1.046142578125, 56.908172607421875, 3.638214111328125, 13.409423828125, 0.0, -32.324371337890625, 8.142486572265625, 173.5308837890625, 148.40945434570312, 107.564453125, -86.9393310546875, -139.01309204101562, -4.6835174560546875, 32.42106628417969, 174.87921142578125, 147.70510864257812, 11.254638671875, 16.276885986328125, 104.1962890625, 81.30799865722656, 190.4013671875, 36.72113037109375, 92.2364501953125, 115.54266357421875, 142.59771728515625, 163.29129028320312, -21.598922729492188, 27.977840423583984, 148.11383056640625, 125.40389251708984, -78.18927001953125, 122.04948425292969, 133.75827026367188, 124.31936645507812, 32.27232360839844, 160.12701416015625, 17.446487426757812, -41.31025695800781, -141.2745361328125, 18.04205322265625, 199.48004150390625, 115.3243408203125, -21.7918701171875, -108.44696807861328, 133.40017700195312, 138.4993896484375, 39.916664123535156, -0.0762176513671875, -18.41790771484375, 79.76705932617188, 111.14982604980469, 35.4327392578125, -1.6461410522460938, 140.18106079101562, 62.31640625, 22.83484649658203, 194.70004272460938, 27.84039878845215, -1.7643241882324219, 78.58395385742188, 100.04560852050781, 139.7233428955078, 139.3472900390625, -1.3411102294921875, 32.4071044921875, 31.765304565429688, 135.18252563476562, 137.31646728515625, -37.55741882324219, 64.64157104492188, 52.241371154785156, 12.64794921875, 145.96554565429688, 158.98675537109375, 28.788877487182617, 194.42596435546875, 145.58184814453125, -0.8516693115234375, -25.27056884765625, 32.54716491699219, -4.247039794921875, 72.92132568359375, 139.51312255859375, 113.39873504638672, 171.33311462402344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000371.npy"}
{"epoch": 0.7769633507853403, "step": 372, "batch_size": 128, "mean": 66.18814086914062, "std": 85.9235610961914, "min": -163.47372436523438, "p10": -17.62064056396484, "median": 56.75035095214844, "p90": 171.923486328125, "max": 246.87945556640625, "pos_frac": 0.765625, "sample": [171.7772216796875, -4.011810302734375, 134.9588623046875, 55.622074127197266, -1.7615947723388672, 135.86294555664062, 123.35659790039062, -6.539031982421875, 124.85645294189453, 166.77781677246094, 57.236846923828125, 66.97880554199219, 130.156494140625, 133.3656005859375, -23.162673950195312, 115.883056640625, 14.681732177734375, 131.09820556640625, -10.628007888793945, 17.75665283203125, 76.7440185546875, 58.73565673828125, 111.2628173828125, -0.597381591796875, -12.6087646484375, 135.96734619140625, 194.79046630859375, 93.39048767089844, 148.46749877929688, 130.04478454589844, -27.289352416992188, -19.633514404296875, 179.57058715820312, -9.86376953125, 88.603515625, 122.07862854003906, 187.4779052734375, -3.0012664794921875, 101.23677825927734, 56.07683563232422, -16.757980346679688, 177.48080444335938, 101.9238052368164, 129.53131103515625, 187.12399291992188, -162.60394287109375, 195.275146484375, 70.74154663085938, 11.442169189453125, 33.28167724609375, 18.587783813476562, 7.0907440185546875, 126.39283752441406, -3.870819091796875, 21.70367431640625, 29.8367919921875, 111.06622314453125, 155.67269897460938, 122.83531188964844, 35.213226318359375, 89.02706909179688, 30.428329467773438, 128.14031982421875, 18.699539184570312, 5.17236328125, 50.62284851074219, 150.31878662109375, 23.52935791015625, 166.10052490234375, -14.203380584716797, 3.640228271484375, 4.43988037109375, 13.926803588867188, 157.40298461914062, 173.02032470703125, 183.82861328125, -109.01553344726562, 180.226806640625, 22.73175048828125, -146.72927856445312, 150.76991271972656, 97.00025939941406, 40.66072082519531, 56.26385498046875, -137.48770141601562, 54.491600036621094, 53.2381591796875, 10.158828735351562, -5.876951217651367, 162.09300231933594, -11.684860229492188, -4.812347412109375, 151.96490478515625, 116.48297119140625, 44.8106689453125, -163.47372436523438, -98.67277526855469, 36.26727294921875, 125.67916870117188, -78.30718231201172, 157.93096923828125, -6.6899566650390625, -53.45477294921875, 6.887054443359375, 170.2556610107422, 139.16571044921875, -153.17620849609375, 172.2647705078125, 246.87945556640625, 184.7601318359375, 63.262603759765625, 204.31787109375, 163.3730926513672, 2.539764404296875, 52.921905517578125, 168.1851806640625, -1.6332168579101562, 134.019287109375, -28.3631591796875, 139.05911254882812, 110.851318359375, 139.24957275390625, 25.643096923828125, 0.97802734375, -3.6376304626464844, 117.89202880859375, 47.76214599609375, 42.276763916015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000372.npy"}
{"epoch": 0.7790575916230367, "step": 373, "batch_size": 128, "mean": 60.849998474121094, "std": 87.70860290527344, "min": -200.628173828125, "p10": -33.48042259216308, "median": 39.96833801269531, "p90": 179.84019775390624, "max": 200.75048828125, "pos_frac": 0.7890625, "sample": [12.2406005859375, 32.776344299316406, 160.33734130859375, 30.550582885742188, 0.6453590393066406, -21.538116455078125, 19.41619873046875, 187.6011962890625, 18.151611328125, 186.57278442382812, 147.21926879882812, -15.55987548828125, 8.297637939453125, 156.98060607910156, 114.60870361328125, 89.97575378417969, 1.6201858520507812, 50.556060791015625, -37.1181640625, 157.15216064453125, 63.536041259765625, 1.2671928405761719, 112.55998229980469, 12.68414306640625, -45.4461669921875, 61.278228759765625, -4.5386810302734375, -41.006622314453125, 37.96600341796875, 191.13037109375, 3.4169387817382812, -25.599746704101562, 23.288421630859375, 37.02960205078125, 49.6002197265625, 112.270263671875, 152.84963989257812, -17.76495361328125, 152.33197021484375, 78.380859375, 100.071044921875, 144.54937744140625, 169.93191528320312, 88.22147369384766, 20.7239990234375, 173.5289306640625, 23.395477294921875, -95.67498779296875, 45.4044189453125, 182.60772705078125, -31.921390533447266, 154.81268310546875, -115.39374542236328, -42.03125, 99.8365478515625, -87.87335205078125, 192.65386962890625, -7.551544189453125, 26.58837890625, 47.439697265625, 5.543272018432617, -200.628173828125, 194.2100830078125, 33.074951171875, 159.1796875, 15.818984985351562, -123.93960571289062, 166.56207275390625, 12.579238891601562, 128.8308563232422, 143.74038696289062, 160.12527465820312, 12.338447570800781, 86.95663452148438, 24.090621948242188, 56.48680114746094, 73.407470703125, 2.28759765625, 87.87002563476562, 7.747589111328125, -145.09564208984375, -15.25152587890625, 187.68789672851562, 1.3132171630859375, 41.970672607421875, 162.83670043945312, 156.62261962890625, 189.71026611328125, 197.24197387695312, 106.39486694335938, 4.41412353515625, 173.84934997558594, -2.33251953125, 20.850006103515625, 165.22801208496094, 162.99737548828125, 22.032440185546875, 189.5728759765625, -52.53750991821289, -138.76954650878906, 85.43187713623047, 155.265625, 200.75048828125, 155.1817626953125, 73.96499633789062, 15.099395751953125, 130.4699249267578, -13.407806396484375, 199.12088012695312, 16.781585693359375, 14.568653106689453, 15.151695251464844, 22.74407958984375, 103.70509338378906, -6.581211090087891, -7.010833740234375, 138.1448516845703, 6.7843475341796875, -83.54021453857422, 88.3560791015625, -3.485611915588379, 178.65411376953125, -15.253604888916016, 5.187164306640625, 166.28506469726562, 184.96917724609375, 127.83768463134766, 13.568084716796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000373.npy"}
{"epoch": 0.7811518324607329, "step": 374, "batch_size": 128, "mean": 41.45735168457031, "std": 87.52098083496094, "min": -180.69287109375, "p10": -67.98885955810546, "median": 22.538909912109375, "p90": 166.44430999755858, "max": 228.74993896484375, "pos_frac": 0.7109375, "sample": [218.016357421875, 0.0, 3.882080078125, 211.44635009765625, 5.42047119140625, -137.78823852539062, -8.790229797363281, 24.285614013671875, 173.23240661621094, 125.74079895019531, 14.86536979675293, 203.17596435546875, 18.251007080078125, -72.85874938964844, 73.45249938964844, 34.53289794921875, 65.1605224609375, 127.33599853515625, 19.47283935546875, 7.3270263671875, -128.52960205078125, -52.06298828125, 100.48409271240234, -121.99676513671875, 159.2683868408203, 27.421165466308594, 18.752777099609375, 11.93402099609375, 6.544771194458008, -21.22112274169922, 9.814453125, 8.725540161132812, -74.9481201171875, 62.72438049316406, -7.93475341796875, 14.994384765625, 25.29815673828125, -10.334716796875, 182.91146850585938, 139.7913818359375, 69.26837158203125, 10.426605224609375, 228.74993896484375, 1.0670242309570312, 62.820220947265625, 0.662506103515625, 2.5389328002929688, 95.37115478515625, -180.69287109375, -16.607643127441406, 32.869117736816406, -47.69781494140625, 165.68084716796875, 99.9613265991211, 23.622772216796875, -1.15478515625, -6.687744140625, 59.20281982421875, 90.34622192382812, -0.4038124084472656, 17.272598266601562, 180.51678466796875, 184.19622802734375, -39.706573486328125, -1.3331756591796875, 57.59581756591797, 48.7252197265625, 63.038330078125, -34.04620361328125, 72.87519836425781, 18.5396728515625, 57.5772705078125, 133.34463500976562, 119.59718322753906, -165.09417724609375, 118.56763458251953, -104.89471435546875, 10.5697021484375, 30.297821044921875, 145.05177307128906, 131.890625, 8.585784912109375, 61.753173828125, 148.99588012695312, 0.0, 147.60452270507812, 2.2624473571777344, -81.0316162109375, -40.31365966796875, -113.7857666015625, 6.925895690917969, 132.60238647460938, 152.88299560546875, 168.22572326660156, 22.28558349609375, -16.60369873046875, 198.27191162109375, 22.792236328125, 181.8306884765625, 57.24468994140625, 15.91143798828125, -42.89138412475586, -97.94670104980469, 29.66973876953125, 21.683622360229492, 4.126075744628906, 43.05877685546875, 187.01449584960938, -15.384590148925781, -117.50222778320312, -1.24969482421875, 100.57382202148438, 54.407501220703125, -3.5585479736328125, 160.72677612304688, -2.7471923828125, 145.4453887939453, 81.55459594726562, -65.90176391601562, 107.90814208984375, 210.61569213867188, 143.27566528320312, 147.8577117919922, 27.707839965820312, 31.541168212890625, -3.6706581115722656, 10.900283813476562, -118.837158203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000374.npy"}
{"epoch": 0.7832460732984293, "step": 375, "batch_size": 128, "mean": 51.72104263305664, "std": 86.84546661376953, "min": -180.92092895507812, "p10": -30.393327331542963, "median": 26.213375091552734, "p90": 164.18683166503905, "max": 250.65020751953125, "pos_frac": 0.7578125, "sample": [3.73065185546875, 66.35906982421875, 59.4903564453125, 11.391708374023438, 69.47705078125, 3.84259033203125, 21.573699951171875, -115.37669372558594, 6.841255187988281, 36.940582275390625, 180.3404541015625, 0.8005599975585938, 160.2578125, 162.91177368164062, 98.00393676757812, -8.94537353515625, 56.338409423828125, 114.80680847167969, 6.534914016723633, 8.695655822753906, 154.88665771484375, 165.68148803710938, 13.432296752929688, 58.621246337890625, 59.800384521484375, 156.41941833496094, -13.636146545410156, 8.407722473144531, -43.70068359375, 105.21009826660156, 12.701225280761719, 12.413951873779297, -4.4593048095703125, -116.57205200195312, 17.539939880371094, 161.414306640625, 113.135986328125, 163.5462646484375, 21.829566955566406, -25.084228515625, 30.443069458007812, 11.533958435058594, -120.81163024902344, 0.639373779296875, -180.92092895507812, 131.96446228027344, 151.82965087890625, 146.70211791992188, 15.734962463378906, -141.87594604492188, -133.4155731201172, -24.371673583984375, 79.81640625, -2.8729610443115234, 49.942962646484375, -4.39794921875, 136.34420776367188, -10.349136352539062, -33.480255126953125, -125.11874389648438, 177.94009399414062, 162.07904052734375, -14.21551513671875, 212.33941650390625, 46.83513259887695, 22.5555419921875, -49.145965576171875, 0.8178253173828125, 250.65020751953125, -7.290994644165039, 82.23262023925781, 12.991495132446289, 22.342941284179688, 90.13139343261719, 30.79388427734375, 181.3040771484375, 75.33441162109375, 5.62774658203125, 46.56488037109375, 10.472366333007812, 136.02435302734375, 98.18756103515625, -6.625213623046875, -29.070358276367188, -3.0818328857421875, 122.67880249023438, 2.094268798828125, 129.21035766601562, 118.0850830078125, 150.64541625976562, -142.70321655273438, 168.79083251953125, 201.0806884765625, -11.662429809570312, 205.8701171875, 218.44171142578125, -45.5303955078125, 163.48130798339844, 98.68077087402344, 167.16595458984375, -2.8350296020507812, -70.02375030517578, 162.79812622070312, -0.9585418701171875, 142.750244140625, 2.0416393280029297, 118.28945922851562, 167.4317626953125, 40.38372802734375, 5.74761962890625, -3.767974853515625, 175.07333374023438, 27.41180419921875, 40.7066650390625, 155.75006103515625, -28.871368408203125, 2.5786914825439453, 161.43280029296875, 2.0487213134765625, 17.16107177734375, 9.780670166015625, 25.01494598388672, 51.851654052734375, 109.72808837890625, 11.195281982421875, 138.8961944580078, 8.626922607421875, 104.98643493652344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000375.npy"}
{"epoch": 0.7853403141361257, "step": 376, "batch_size": 128, "mean": 51.55521011352539, "std": 89.88500213623047, "min": -161.96046447753906, "p10": -48.52994384765624, "median": 40.33270263671875, "p90": 166.92560424804688, "max": 252.197021484375, "pos_frac": 0.734375, "sample": [87.88362121582031, 50.869415283203125, 168.10447692871094, 252.197021484375, 122.91934204101562, 220.7252197265625, 111.22152709960938, 21.459209442138672, 52.91340637207031, -118.86485290527344, 162.84832763671875, -23.799240112304688, -139.75967407226562, 41.340240478515625, -21.067108154296875, 87.77558898925781, 3.7397079467773438, 43.618865966796875, 128.95477294921875, 168.56298828125, 108.31182861328125, -147.1336669921875, 1.5384597778320312, 11.882415771484375, 69.82159423828125, 26.314037322998047, -27.740921020507812, 179.719482421875, -2.6348419189453125, 147.95843505859375, 144.26040649414062, 59.79844665527344, 166.94235229492188, 39.325164794921875, -16.850906372070312, -20.143829345703125, 166.91842651367188, 83.1891860961914, 163.79144287109375, 11.6051025390625, 0.0, 18.531951904296875, 18.444931030273438, 118.79606628417969, -42.13960266113281, 188.71109008789062, -94.09634399414062, 140.60296630859375, 108.79066467285156, 131.46054077148438, -141.50904846191406, 102.27490234375, 44.375518798828125, 10.5115966796875, 148.84295654296875, 7.15118408203125, 25.888153076171875, 104.22676086425781, -43.47503662109375, -16.14752197265625, 122.44093322753906, -2.21368408203125, 210.06942749023438, 198.83203125, 25.407379150390625, 185.02935791015625, 158.39883422851562, 45.824684143066406, -45.541900634765625, -27.35894775390625, 79.85791015625, -55.502044677734375, 26.219329833984375, 2.8693923950195312, -5.778413772583008, 10.592620849609375, -113.36149597167969, 18.74315643310547, -75.800537109375, 170.96031188964844, 119.98965454101562, 57.523162841796875, 161.54519653320312, 123.6544189453125, 158.41458129882812, 129.23805236816406, 6.978424072265625, -161.96046447753906, 11.461029052734375, -109.31455993652344, 47.260772705078125, 2.062366485595703, 144.38470458984375, 3.5496902465820312, 7.364856719970703, 133.61697387695312, 80.04248046875, -12.997344970703125, -115.01473999023438, 159.48684692382812, 25.317596435546875, 149.782470703125, -96.42353820800781, 138.61337280273438, 17.474853515625, 21.34461212158203, 98.40605163574219, 35.79524230957031, 61.09039306640625, 140.08343505859375, -16.837799072265625, -39.195068359375, -39.995487213134766, 38.185943603515625, 95.10379028320312, -5.510265350341797, 33.922637939453125, -36.650917053222656, 53.4439697265625, -1.4244613647460938, 5.255191802978516, 66.15321350097656, 126.45382690429688, 33.97314453125, 175.26947021484375, -132.0759735107422, 190.643310546875, 164.135986328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000376.npy"}
{"epoch": 0.787434554973822, "step": 377, "batch_size": 128, "mean": 47.4456672668457, "std": 81.764892578125, "min": -155.246826171875, "p10": -33.22400817871093, "median": 23.35320281982422, "p90": 165.52721099853514, "max": 211.88421630859375, "pos_frac": 0.703125, "sample": [176.79766845703125, 58.41795349121094, 127.91566467285156, -13.198829650878906, 95.42730712890625, 9.597869873046875, 11.41424560546875, 149.03192138671875, 139.89930725097656, 14.815032958984375, 20.51219940185547, 52.03155517578125, 130.39181518554688, -55.564239501953125, 117.462646484375, 111.61695861816406, 177.1646728515625, -29.86822509765625, 1.0650482177734375, -29.237335205078125, 138.2093505859375, 2.0012054443359375, -1.134796142578125, 9.952892303466797, -13.026090621948242, -18.48914337158203, 67.20376586914062, -12.251129150390625, 163.1137237548828, 43.17893981933594, 109.92616271972656, 36.269775390625, 80.39860534667969, -106.198486328125, -8.005905151367188, 148.79983520507812, -4.300968170166016, -6.6376953125, -9.93048095703125, 41.466644287109375, 18.03338623046875, 96.41563415527344, 13.417594909667969, -65.5709228515625, 133.2139129638672, 10.192138671875, 188.557373046875, 62.89471435546875, 8.326614379882812, -108.83236694335938, 7.51983642578125, 197.86065673828125, -3.927490234375, 0.6179351806640625, 174.579345703125, 5.229026794433594, -78.44648742675781, -1.34600830078125, 138.78797912597656, 147.218017578125, 156.47885131835938, 28.8409423828125, 71.6053466796875, 91.06393432617188, -24.953277587890625, 144.90467834472656, 55.69026184082031, 0.0, -35.4100341796875, 40.9444580078125, 141.41390991210938, -2.67425537109375, 117.17167663574219, 141.5296630859375, -3.099334716796875, 62.8577880859375, 8.803092956542969, -32.287139892578125, -0.6755008697509766, 139.45941162109375, 18.38360595703125, 206.21697998046875, 9.1688232421875, -26.169815063476562, -15.996246337890625, -89.81987762451172, 16.895797729492188, 5.616020202636719, 163.12289428710938, 40.267333984375, 61.26055908203125, 29.16552734375, 8.584701538085938, 152.38345336914062, -54.501708984375, 174.99777221679688, 11.372085571289062, 23.5841064453125, 45.967742919921875, 132.34597778320312, 31.996063232421875, -122.44473266601562, -11.657081604003906, 85.1016845703125, 181.42694091796875, -69.13282775878906, 194.9622802734375, 34.107177734375, 72.97952270507812, 143.77297973632812, 19.21533966064453, 11.30728530883789, 25.907424926757812, 20.2215576171875, 180.50979614257812, 140.65673828125, 171.1372833251953, -131.03643798828125, -14.0296630859375, 211.88421630859375, 23.122299194335938, 18.009521484375, -98.26554870605469, 181.5140380859375, -11.277725219726562, -9.845733642578125, -155.246826171875, 62.65922546386719], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000377.npy"}
{"epoch": 0.7895287958115184, "step": 378, "batch_size": 128, "mean": 47.92448425292969, "std": 82.88127136230469, "min": -138.66326904296875, "p10": -40.16021118164062, "median": 27.337982177734375, "p90": 166.744612121582, "max": 206.54342651367188, "pos_frac": 0.7109375, "sample": [133.05902099609375, 3.66729736328125, 35.27642822265625, 125.89752197265625, -24.7100830078125, 3.3875999450683594, -28.36273193359375, 7.7867431640625, 177.96279907226562, 64.85455322265625, -8.714088439941406, 10.995651245117188, 27.700439453125, 130.68353271484375, 133.73831176757812, -51.12803649902344, -130.63375854492188, 174.39810180664062, 186.47308349609375, -25.576171875, 33.196258544921875, 38.93621826171875, 12.261688232421875, -42.4012451171875, 74.20367431640625, 181.86871337890625, 206.54342651367188, 143.8900146484375, 23.9306640625, 127.59054565429688, 3.9254608154296875, 137.64602661132812, 108.20164489746094, 22.51702880859375, 8.4219970703125, -138.66326904296875, 2.970987319946289, 2.1826725006103516, 63.0699462890625, 88.86700439453125, 0.0, 156.16317749023438, -7.4949951171875, 143.97433471679688, 183.88018798828125, -49.29193115234375, -5.284053802490234, 201.23468017578125, -14.335716247558594, 179.31488037109375, 110.093505859375, 113.693359375, -24.917861938476562, 4.36907958984375, 8.817535400390625, 0.9750595092773438, -28.89605712890625, 132.13534545898438, -18.93365478515625, 144.86972045898438, 62.78515625, -0.5311565399169922, 132.84365844726562, 144.89434814453125, 37.903656005859375, 120.38967895507812, -49.41044616699219, -55.973663330078125, 169.21807861328125, 105.16496276855469, 118.08523559570312, 23.327743530273438, 0.53692626953125, -26.67919921875, -119.853271484375, 165.00192260742188, -31.897445678710938, -90.28976440429688, 42.075286865234375, -2.72503662109375, 33.686737060546875, 6.8018035888671875, 82.804931640625, 70.8602294921875, 8.989990234375, 101.21920776367188, 179.75064086914062, 7.0464324951171875, 18.0189208984375, 68.88845825195312, 148.678955078125, 151.49600219726562, -24.465118408203125, 0.28594970703125, 56.0367431640625, -13.157135009765625, -70.920166015625, 7.919097900390625, 12.82696533203125, -20.980300903320312, -97.54232788085938, 175.88031005859375, -6.002655029296875, 54.02815246582031, -137.7311553955078, -3.755462646484375, 26.97552490234375, 124.43651580810547, -6.125701904296875, 33.21905517578125, 49.96392822265625, 53.06719970703125, 3.395233154296875, -14.489212036132812, 194.15872192382812, 159.51373291015625, 164.720703125, 165.68455505371094, 175.49505615234375, 18.569839477539062, -138.41302490234375, 107.98345184326172, 131.96148681640625, 46.060302734375, 29.835678100585938, -39.19976806640625, 16.485076904296875, -0.7784423828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000378.npy"}
{"epoch": 0.7916230366492146, "step": 379, "batch_size": 128, "mean": 51.26026153564453, "std": 88.91988372802734, "min": -175.3790283203125, "p10": -41.715652465820305, "median": 36.84034729003906, "p90": 167.3655258178711, "max": 203.48529052734375, "pos_frac": 0.7421875, "sample": [-22.655101776123047, 10.169868469238281, 1.8658447265625, 9.3974609375, -1.9348068237304688, 8.37823486328125, -0.09777069091796875, 145.75241088867188, 48.63951110839844, 1.9667739868164062, 100.21035766601562, 138.04049682617188, 41.718658447265625, 69.15087890625, 15.15179443359375, 8.880012512207031, 108.59053039550781, -118.47894287109375, 176.7978515625, 45.85820007324219, 118.84320068359375, 48.25714111328125, 154.11622619628906, -38.04913330078125, 10.920700073242188, -39.071380615234375, 152.7870330810547, 1.493499755859375, 91.05987548828125, -8.12563705444336, 30.970001220703125, 168.0061492919922, 170.58505249023438, 52.8331298828125, 48.47271728515625, 8.382965087890625, 73.69270324707031, 180.383544921875, 98.41787719726562, 11.942497253417969, -4.07525634765625, -112.22946166992188, 37.680084228515625, 15.509552001953125, 5.428497314453125, 195.27078247070312, 9.888805389404297, -62.418487548828125, -21.519866943359375, 7.68023681640625, 137.61260986328125, -4.000359535217285, 36.0006103515625, -153.50259399414062, -6.314727783203125, 11.344757080078125, 163.59259033203125, 148.31585693359375, 127.04200744628906, 127.04452514648438, 44.88896179199219, -50.579246520996094, 125.60342407226562, 80.2835693359375, 138.86410522460938, 203.48529052734375, 12.16534423828125, 169.52572631835938, 9.9835205078125, -32.36805725097656, -21.95147705078125, 0.0, -8.262832641601562, 115.63056945800781, 19.217742919921875, 174.06256103515625, 55.441497802734375, 152.75341796875, 171.9024658203125, 163.95152282714844, 8.66033935546875, 124.97480773925781, 6.4280548095703125, 1.96319580078125, 148.4052734375, 139.32110595703125, -21.349563598632812, 120.12236022949219, -23.539474487304688, 144.84689331054688, 112.33573913574219, -13.670265197753906, 179.2095947265625, 3.9750194549560547, 118.53079223632812, -11.83624267578125, 2.777273178100586, 131.27957153320312, -126.2635498046875, 201.85498046875, 196.05856323242188, -175.3790283203125, 17.6041259765625, 48.364532470703125, 160.9434814453125, 79.8795166015625, 62.357452392578125, -47.8856201171875, 152.07723999023438, 164.2862548828125, 177.73025512695312, 98.73358154296875, 161.867431640625, 79.07148742675781, -117.8033676147461, -137.65980529785156, 22.30394744873047, 4.029684066772461, 11.30865478515625, 167.09097290039062, -36.271881103515625, 119.9981460571289, -145.60073852539062, -81.13807678222656, -147.5843505859375, 18.95738410949707, -12.520660400390625, 166.19973754882812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000379.npy"}
{"epoch": 0.793717277486911, "step": 380, "batch_size": 128, "mean": 49.14793395996094, "std": 85.96485137939453, "min": -138.64149475097656, "p10": -46.574543762207014, "median": 30.11474609375, "p90": 168.66609191894528, "max": 251.01507568359375, "pos_frac": 0.71875, "sample": [163.03268432617188, 49.205169677734375, 108.06741333007812, 21.469131469726562, 132.1088409423828, 18.43194580078125, -134.5474395751953, 4.2593536376953125, 0.7626457214355469, -61.32275390625, 92.55624389648438, 65.04324340820312, 146.83740234375, -117.37842559814453, 30.45751953125, 15.802589416503906, 111.19207763671875, 43.74072265625, 140.97137451171875, -3.8019638061523438, 0.0, 46.8128662109375, 9.1451416015625, 25.376953125, 34.77630615234375, 149.129150390625, 7.045257568359375, 37.92230224609375, -0.9560394287109375, 9.977325439453125, 48.838134765625, 205.6151123046875, 12.225555419921875, 131.88381958007812, 1.131195068359375, -127.8699722290039, 139.376220703125, -4.198551177978516, 21.316986083984375, -20.1265869140625, 46.0419921875, 60.37428283691406, 164.92098999023438, -42.36773681640625, -18.70800018310547, 16.6297607421875, 16.593109130859375, 176.3348388671875, 175.77130126953125, -17.934967041015625, -67.17242431640625, 96.34678649902344, 66.5221939086914, -4.50067138671875, 22.156005859375, -83.2511215209961, -21.052783966064453, 13.38665771484375, 177.57839965820312, 142.52560424804688, -61.61981201171875, 157.4197540283203, 118.36093139648438, 31.362457275390625, 189.7735595703125, 163.5958251953125, -116.38677215576172, -5.517578125, 18.15447998046875, 22.840953826904297, 98.28250122070312, 183.0064697265625, 1.6978759765625, 115.55841827392578, 155.2166290283203, -5.071441650390625, 66.76608276367188, -40.2889404296875, 30.96514892578125, -138.64149475097656, 112.12002563476562, 251.01507568359375, -6.044670104980469, 136.2857208251953, 160.31915283203125, 24.591278076171875, 147.7703857421875, 164.7186279296875, 165.62100219726562, 53.16522216796875, 35.9293212890625, 215.10592651367188, -72.0020751953125, 56.28790283203125, -0.3007545471191406, 67.40293884277344, 0.73077392578125, -38.88922119140625, 179.43182373046875, 238.2822265625, 33.79914474487305, -0.616455078125, 50.48335266113281, 151.433837890625, 60.773529052734375, 16.96306610107422, 27.93865966796875, 45.72270202636719, -31.186691284179688, -32.34356689453125, -90.2825927734375, -100.65406799316406, 10.351394653320312, 10.532567977905273, 25.245956420898438, 116.09127807617188, -21.166839599609375, 46.91522216796875, 103.37078857421875, 224.007080078125, 29.77197265625, -16.1748046875, -34.149169921875, 179.4578857421875, 0.535552978515625, -56.39042663574219, 200.27114868164062, -11.284027099609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000380.npy"}
{"epoch": 0.7958115183246073, "step": 381, "batch_size": 128, "mean": 55.96220397949219, "std": 82.29882049560547, "min": -109.76397705078125, "p10": -34.747691345214825, "median": 41.674489974975586, "p90": 165.98414154052733, "max": 189.8955078125, "pos_frac": 0.703125, "sample": [133.923583984375, 184.44110107421875, 171.95294189453125, 44.396400451660156, 182.1665802001953, -47.16473388671875, 166.57106018066406, 178.59490966796875, 123.36671447753906, 5.395439147949219, 183.34666442871094, 22.289459228515625, 14.360549926757812, 144.09222412109375, 134.87368774414062, 153.37261962890625, 0.0, 11.8441162109375, 62.82325744628906, 129.191162109375, 136.75340270996094, 66.54388427734375, 4.0642547607421875, 189.8955078125, 46.569427490234375, 45.73108673095703, 135.36453247070312, -106.26652526855469, 0.0, -22.885986328125, 56.53204345703125, 129.69830322265625, -9.98797607421875, 145.4055633544922, -9.65951156616211, 32.998374938964844, 15.662139892578125, 64.78118896484375, 10.143280029296875, 115.69573974609375, 186.70892333984375, 85.43898010253906, -19.8232421875, 131.7860565185547, 131.31097412109375, 7.38482666015625, 165.73260498046875, -5.333915710449219, -90.64646911621094, 147.52308654785156, 149.08261108398438, 160.21157836914062, 43.58349609375, -0.1790008544921875, 178.141357421875, 163.732666015625, 11.09515380859375, 132.74615478515625, 167.00634765625, 67.56045532226562, -61.75798034667969, -13.2060546875, 90.43016052246094, 10.06591796875, -7.37445068359375, 46.63174819946289, 154.5245361328125, -3.5467529296875, -29.426101684570312, -87.82626342773438, 10.987251281738281, -17.492387771606445, -3.855438232421875, 150.76510620117188, -22.74542236328125, 24.821868896484375, -1.41119384765625, 103.82620239257812, 49.97442626953125, -91.106201171875, 161.99566650390625, -28.37366485595703, 145.67611694335938, 42.8316650390625, 177.80661010742188, -4.93780517578125, -25.73699951171875, -8.168586730957031, 157.03565979003906, 13.122833251953125, -6.811187744140625, -53.419281005859375, 138.00526428222656, 125.87876892089844, 105.4140625, -48.377044677734375, 2.534465789794922, 7.5963134765625, 31.354248046875, -106.29418182373047, 108.81498718261719, 8.898323059082031, -83.25897216796875, 9.78677749633789, -109.76397705078125, 126.70042419433594, -16.1826171875, 7.865243911743164, -15.179931640625, 141.48471069335938, -8.79736328125, 178.8424530029297, -72.23294067382812, 187.7716064453125, 11.696502685546875, 40.51731491088867, 51.57036590576172, 11.680961608886719, 131.3086700439453, 20.741119384765625, 151.74017333984375, 156.5985107421875, 34.88957977294922, -27.026611328125, -102.62460327148438, 129.82000732421875, 22.606475830078125, 75.54367065429688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000381.npy"}
{"epoch": 0.7979057591623037, "step": 382, "batch_size": 128, "mean": 69.19227600097656, "std": 86.93878936767578, "min": -133.4716796875, "p10": -33.46004180908203, "median": 70.82359313964844, "p90": 170.66821136474607, "max": 212.11224365234375, "pos_frac": 0.765625, "sample": [-106.88839721679688, 14.16253662109375, 62.371124267578125, 101.35614013671875, 104.51422119140625, 155.25912475585938, 156.80804443359375, -0.70123291015625, 21.898893356323242, 124.21177673339844, 131.43826293945312, -133.4716796875, 113.05596923828125, 14.312522888183594, 176.02685546875, 74.9969482421875, 46.737266540527344, 66.40037536621094, 141.300537109375, -13.437355041503906, 49.51141357421875, -127.15971374511719, 157.7438201904297, 5.347442626953125, 6.6910400390625, 182.42510986328125, 32.54425048828125, -35.44087219238281, 134.92660522460938, 109.26184844970703, -10.83013916015625, 131.27000427246094, 121.90701293945312, 34.927589416503906, 152.60577392578125, 105.8010025024414, -0.17874908447265625, 96.50446319580078, -6.356206893920898, 164.09530639648438, 27.330726623535156, 15.83123779296875, 166.00473022460938, 23.79791259765625, 160.56997680664062, 167.04290771484375, 173.11668395996094, 46.14642333984375, 39.043701171875, 9.619903564453125, 169.61886596679688, 135.91506958007812, 8.173999786376953, 70.65170288085938, -10.64837646484375, 116.07244873046875, 212.11224365234375, 187.69189453125, -38.82879638671875, -11.906890869140625, -32.35247802734375, -130.58895874023438, 210.870849609375, 37.676910400390625, 3.5060272216796875, 109.74716186523438, 192.10836791992188, 67.30233764648438, 184.52020263671875, 27.156585693359375, 190.14285278320312, 24.20880126953125, 37.17915344238281, -3.2938232421875, 165.288818359375, 106.53179931640625, 128.4869384765625, 20.742782592773438, 127.43409729003906, 163.48040771484375, 43.853515625, -32.611114501953125, 55.638214111328125, 139.14804077148438, -122.70448303222656, -3.361663818359375, -65.39761352539062, 162.3101806640625, -20.919418334960938, 162.7857666015625, 174.2135009765625, -20.798065185546875, 12.582565307617188, 178.1061553955078, 165.03167724609375, 55.2352294921875, 125.14899444580078, -117.15571594238281, 5.267425537109375, -12.853302001953125, 33.538665771484375, 70.9954833984375, 75.21714782714844, 158.42723083496094, 125.86422729492188, -67.60617065429688, 37.09220886230469, 146.91758728027344, -108.66629028320312, 138.74818420410156, 146.54119873046875, -9.861030578613281, 141.03106689453125, 48.58447265625, 82.47283935546875, 129.7035369873047, -78.51116943359375, 74.21829223632812, -14.639671325683594, 122.838623046875, 166.20639038085938, -19.958148956298828, 201.61260986328125, 142.2615509033203, 176.60922241210938, 127.76666259765625, 169.09388732910156, -96.8612060546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000382.npy"}
{"epoch": 0.8, "step": 383, "batch_size": 128, "mean": 52.38575744628906, "std": 83.70917510986328, "min": -150.0565948486328, "p10": -51.71762161254882, "median": 31.716331481933594, "p90": 166.31891479492188, "max": 210.356201171875, "pos_frac": 0.734375, "sample": [-7.094268798828125, 43.80767822265625, 141.9851837158203, 136.48770141601562, -28.321075439453125, 17.94677734375, -56.0068359375, 30.296051025390625, -1.20428466796875, 87.13969421386719, 135.4783935546875, -127.81867980957031, 137.18832397460938, 33.13661193847656, 0.0, -12.628639221191406, 18.371200561523438, -57.2987060546875, 50.5230712890625, -60.6434326171875, 19.44036865234375, 174.78073120117188, 148.51222229003906, 123.42947387695312, -113.59759521484375, 174.43028259277344, -69.20521545410156, -5.498615264892578, -66.84724426269531, 9.822402954101562, 80.54466247558594, 8.63916015625, -14.23233413696289, -39.25782012939453, 150.23873901367188, 1.8790283203125, 187.46658325195312, 172.948974609375, 34.21949768066406, 143.27377319335938, 94.43820190429688, 3.2904205322265625, 137.39682006835938, 209.172607421875, -150.0565948486328, -2.03143310546875, 45.63771057128906, 162.14898681640625, 91.00628662109375, 6.96136474609375, 165.785400390625, 2.8348388671875, -3.9098968505859375, 144.499755859375, 88.46139526367188, 20.99786376953125, 5.288055419921875, -9.513069152832031, 0.1401805877685547, -124.25619506835938, 132.4732208251953, 148.2294921875, 11.722908020019531, 167.56378173828125, -25.749267578125, 10.963531494140625, 161.982666015625, 4.9452056884765625, -2.43341064453125, 18.817901611328125, 111.25074005126953, 172.4132843017578, 132.231201171875, 133.052978515625, 56.965545654296875, 40.27176284790039, 0.443359375, -18.475677490234375, 25.12109375, 133.25804138183594, 3.3833770751953125, 116.86089324951172, 122.15025329589844, 52.085601806640625, 93.915283203125, 30.1339111328125, 29.7908935546875, 21.741092681884766, 64.48338317871094, -148.74163818359375, 168.24798583984375, 22.447141647338867, 57.92962646484375, 49.45850372314453, 33.63116455078125, 5.581422805786133, 0.7783966064453125, 185.0794677734375, 136.2235870361328, 130.18841552734375, 140.32431030273438, -65.1148681640625, 210.356201171875, -102.35111999511719, -85.925537109375, 23.826858520507812, 105.584716796875, 20.650970458984375, 0.0, 128.78228759765625, 150.82147216796875, -3.1758365631103516, 55.532958984375, 118.9117431640625, 93.80792236328125, -15.487274169921875, -38.857818603515625, 16.467365264892578, 173.60409545898438, 115.23037719726562, 8.565322875976562, 140.68423461914062, 192.35159301757812, -49.87938690185547, -20.7716064453125, -3.2874755859375, 168.11465454101562, 145.56942749023438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000383.npy"}
{"epoch": 0.8020942408376963, "step": 384, "batch_size": 128, "mean": 37.78759765625, "std": 88.837646484375, "min": -169.1575927734375, "p10": -72.08185806274415, "median": 21.186248779296875, "p90": 155.12958068847655, "max": 213.60537719726562, "pos_frac": 0.65625, "sample": [193.68682861328125, -169.1575927734375, 104.13232421875, -21.3482666015625, -23.8646240234375, 141.86422729492188, 43.56517028808594, 213.60537719726562, 29.58502197265625, 0.4944610595703125, 148.0780487060547, -18.597408294677734, -24.275039672851562, -71.45724487304688, 34.87481689453125, -116.802978515625, 150.47572326660156, -3.438922882080078, 178.06878662109375, -19.771392822265625, 5.715620040893555, 170.33871459960938, 132.8627166748047, 134.41864013671875, -23.927108764648438, 153.42315673828125, 23.373626708984375, 131.6002655029297, 18.67236328125, -72.7569580078125, 3.4622135162353516, -51.737060546875, -7.801605224609375, 54.029388427734375, -26.194732666015625, 128.8876190185547, 21.336700439453125, 31.92822265625, 145.6060028076172, -16.45000457763672, 20.700592041015625, -122.89572143554688, -1.0367050170898438, 15.489906311035156, 137.51046752929688, 148.5026397705078, 148.03567504882812, -158.86264038085938, 2.94952392578125, 21.035797119140625, 118.91315460205078, 183.39486694335938, -72.75306701660156, -1.09967041015625, -145.89419555664062, 38.954933166503906, 90.930908203125, 79.4775390625, -120.0982666015625, -20.717010498046875, 11.761734008789062, -122.47512817382812, 124.20285034179688, 135.70852661132812, 169.99615478515625, 47.691864013671875, 62.88653564453125, 56.19403076171875, 81.08685302734375, -5.011348724365234, -12.15114974975586, 109.02649688720703, 158.97201538085938, 11.654006958007812, 123.93344116210938, 155.49459838867188, -11.757568359375, 17.711883544921875, -17.812938690185547, 80.6220703125, 85.74516296386719, -114.26918029785156, 122.32545471191406, 80.75311279296875, 145.35693359375, 202.24853515625, 26.257568359375, 7.1185302734375, 78.68850708007812, 40.428466796875, 174.14044189453125, -11.008674621582031, 145.32687377929688, 64.47268676757812, -6.041015625, 3.45709228515625, -36.4278564453125, 170.5481719970703, 11.505096435546875, -10.139114379882812, -4.83355712890625, -48.52117919921875, 39.441162109375, -120.3612060546875, -146.796630859375, 26.093708038330078, 0.1596832275390625, 29.458845138549805, 11.226911544799805, 2.42462158203125, 184.17901611328125, -137.556640625, 53.329620361328125, -3.95904541015625, -3.9863719940185547, -11.847320556640625, 173.27899169921875, 154.97314453125, -71.79419708251953, 126.81124877929688, 136.85256958007812, -57.22113037109375, 1.8244247436523438, 15.642822265625, 25.850616455078125, 62.76495361328125, 15.34521484375, -63.233245849609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000384.npy"}
{"epoch": 0.8041884816753927, "step": 385, "batch_size": 128, "mean": 50.08042526245117, "std": 84.47056579589844, "min": -165.2415771484375, "p10": -41.36467895507812, "median": 28.80291748046875, "p90": 164.40193023681638, "max": 219.52288818359375, "pos_frac": 0.75, "sample": [-39.45758056640625, 98.6080322265625, 162.89500427246094, 108.2357177734375, 43.079925537109375, 28.8616943359375, 154.1473388671875, 28.187408447265625, 160.92388916015625, 140.2194061279297, 30.496002197265625, 0.64862060546875, -50.704429626464844, 20.774063110351562, 26.36865234375, 98.42204284667969, 56.55084228515625, 121.4206314086914, -76.9229507446289, -3.5287933349609375, 167.9180908203125, -38.492156982421875, 139.78787231445312, -87.80656433105469, 10.373916625976562, 158.2785186767578, 133.22549438476562, 11.187690734863281, 127.42333984375, 22.755218505859375, 5.7153778076171875, 197.87686157226562, 133.41067504882812, 12.650390625, 173.14666748046875, 219.52288818359375, 150.19296264648438, 28.744140625, 114.8216323852539, 217.75631713867188, 26.1932373046875, -32.57154846191406, 52.756072998046875, 111.62492370605469, 77.994873046875, -121.56706237792969, 135.3155059814453, 142.0094451904297, -132.32513427734375, 30.701019287109375, 211.61410522460938, 171.73193359375, 216.72216796875, -8.760269165039062, 125.12226867675781, 25.937164306640625, 62.745849609375, 0.9317836761474609, -160.2828369140625, 55.69507598876953, 16.596405029296875, 183.22296142578125, -12.6051025390625, 158.313232421875, 3.91455078125, 3.432830810546875, -29.31378173828125, 134.2125244140625, -2.516357421875, 64.71240234375, 99.3585205078125, 3.5832138061523438, -28.88507080078125, 10.1419677734375, 21.676284790039062, -24.696365356445312, 104.44523620605469, 139.00747680664062, 128.39105224609375, -95.91189575195312, 10.84234619140625, 102.89034271240234, 140.0924072265625, 101.66453552246094, 127.4395751953125, 11.390609741210938, -22.852859497070312, 28.2353515625, -165.2415771484375, 3.380779266357422, 54.817474365234375, -5.059803009033203, -2.9929656982421875, 25.782325744628906, 172.03662109375, 49.267669677734375, 0.0, 38.84388732910156, 69.19085693359375, 21.23809051513672, 2.0901870727539062, 11.784027099609375, 45.944793701171875, -35.399139404296875, 183.04644775390625, -11.024360656738281, -103.5262451171875, 113.2298583984375, -30.882537841796875, 186.90182495117188, 46.0205078125, 0.0, -72.3876953125, -94.43439483642578, 127.51428985595703, 3.185272216796875, 55.45513916015625, -45.8145751953125, -3.526458740234375, 0.749267578125, -75.74497985839844, 13.857734680175781, 87.37710571289062, 36.0152587890625, 126.70208740234375, 20.202880859375, 169.23410034179688, 16.37445068359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000385.npy"}
{"epoch": 0.806282722513089, "step": 386, "batch_size": 128, "mean": 61.615516662597656, "std": 89.6107177734375, "min": -232.64907836914062, "p10": -22.259399414062496, "median": 44.79817199707031, "p90": 175.75643005371091, "max": 236.07977294921875, "pos_frac": 0.7578125, "sample": [119.6903076171875, -3.7353668212890625, 5.5448760986328125, 126.3292465209961, 64.83447265625, -14.98681640625, 66.37652587890625, 169.3585205078125, 120.80181884765625, 137.3589630126953, 141.96148681640625, 236.07977294921875, -42.01141357421875, 34.31536865234375, 190.92221069335938, 39.86412048339844, 19.7027587890625, 31.08416748046875, 0.4221382141113281, 114.6268310546875, 111.76171875, 166.55267333984375, 25.4051513671875, 60.17108154296875, -12.40606689453125, -142.81983947753906, 40.550994873046875, -15.485870361328125, 138.36692810058594, 83.54302978515625, 32.924102783203125, -115.65936279296875, 188.37115478515625, 32.12713623046875, -18.596710205078125, 14.609130859375, 15.225982666015625, -31.954818725585938, 49.97003173828125, 155.13970947265625, 224.36944580078125, 169.01983642578125, 53.483062744140625, 36.90496063232422, 51.65876770019531, 131.36795043945312, 185.65530395507812, -73.45553588867188, 81.67184448242188, 202.57464599609375, 88.2308349609375, -23.65478515625, -89.4237060546875, -36.71759796142578, 194.34146118164062, 128.3973388671875, -9.831436157226562, 0.0, -80.68701171875, -20.549854278564453, -146.86767578125, -97.12139892578125, 125.24261474609375, 30.05999755859375, 1.7041969299316406, -7.946044921875, 51.644287109375, -1.6089096069335938, 156.17185974121094, -174.94200134277344, 181.6993408203125, 0.4332275390625, 137.29873657226562, 196.77224731445312, 195.95025634765625, 150.75851440429688, 6.174343109130859, -7.722259521484375, 157.12364196777344, 144.40785217285156, 141.5303192138672, 11.0885009765625, -232.64907836914062, 44.523193359375, -1.227264404296875, 166.0196533203125, 148.61636352539062, 13.46881103515625, 136.8333740234375, 1.819305419921875, 49.683990478515625, 18.59490203857422, -0.650390625, 0.0, 193.96661376953125, 143.73377990722656, -8.87957763671875, 7.739326477050781, 81.36273193359375, 137.16357421875, 159.13241577148438, 167.408203125, 37.325897216796875, 158.8518829345703, 12.634284973144531, 173.51715087890625, 30.9974365234375, -21.661376953125, -3.51751708984375, 151.42510986328125, 180.98141479492188, 45.073150634765625, 184.9829864501953, 89.41438293457031, 6.116058349609375, 26.217864990234375, 30.731246948242188, 5.36834716796875, 78.65643310546875, 142.86776733398438, 31.225086212158203, 173.42813110351562, 12.378421783447266, 10.842330932617188, 104.54193115234375, 0.0, 140.20794677734375, 55.975982666015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000386.npy"}
{"epoch": 0.8083769633507853, "step": 387, "batch_size": 128, "mean": 58.466941833496094, "std": 88.74607849121094, "min": -191.46038818359375, "p10": -39.0202880859375, "median": 40.4406852722168, "p90": 174.66542053222656, "max": 210.91140747070312, "pos_frac": 0.6953125, "sample": [55.749114990234375, 43.20867919921875, 103.79156494140625, 16.969135284423828, 6.7413330078125, 174.45489501953125, 191.4391326904297, -56.934356689453125, -143.58123779296875, -31.6976318359375, 0.0, 182.00912475585938, -11.739776611328125, 197.98110961914062, 131.99191284179688, 112.57666015625, 4.7897186279296875, -14.99847412109375, -28.089981079101562, 102.14443969726562, -45.99748229980469, 41.630821228027344, 158.20562744140625, -44.073455810546875, 52.53302764892578, 162.4799041748047, -99.13008117675781, -17.0635986328125, 161.99200439453125, 1.4543819427490234, 152.2904510498047, -63.674346923828125, 160.968017578125, 33.23783874511719, 149.0816650390625, 145.7061767578125, -14.0787353515625, 34.318634033203125, 69.29784393310547, -9.17156982421875, -38.558624267578125, 135.59152221679688, -9.446746826171875, -24.365875244140625, 179.50979614257812, 22.35846710205078, 193.4437255859375, 109.34130859375, -17.657751083374023, 0.291778564453125, -8.447044372558594, 180.15753173828125, -40.097503662109375, -12.949676513671875, 59.44769287109375, 24.42938232421875, 10.462066650390625, -125.81092834472656, 143.94955444335938, 157.78472900390625, 114.318603515625, 37.7188720703125, 63.245574951171875, 25.5587158203125, 137.566650390625, -122.13339233398438, 39.25054931640625, 34.978179931640625, 8.875030517578125, -12.30169677734375, -18.021377563476562, -12.811492919921875, 49.6859130859375, 184.32888793945312, 147.8077392578125, 128.53598022460938, 148.16929626464844, 38.660179138183594, -8.118698120117188, 112.24014282226562, 14.236175537109375, -6.701576232910156, 168.0194091796875, 19.989288330078125, 88.08847045898438, 173.85009765625, 126.84846496582031, 5.662200927734375, 61.66094970703125, 140.56588745117188, 0.9149627685546875, 164.97268676757812, 54.68992233276367, 23.176483154296875, 199.2457275390625, 82.65142822265625, 146.44796752929688, -54.508697509765625, 149.86212158203125, 160.8963165283203, 6.7313232421875, 68.68267822265625, -26.2327880859375, 144.01736450195312, -5.637210845947266, -61.893035888671875, -191.46038818359375, -15.31768798828125, 140.60183715820312, 210.91140747070312, 29.087066650390625, -4.298370361328125, 69.26445770263672, 110.76022338867188, 190.02349853515625, 140.51820373535156, 141.06114196777344, 13.29376220703125, 202.052978515625, 13.201118469238281, -23.5621337890625, -138.52096557617188, 175.15664672851562, 191.06988525390625, -0.7150650024414062, 153.9667205810547, -12.968589782714844, 139.60687255859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000387.npy"}
{"epoch": 0.8104712041884817, "step": 388, "batch_size": 128, "mean": 43.701263427734375, "std": 91.773193359375, "min": -175.33294677734375, "p10": -64.59671783447266, "median": 20.20804786682129, "p90": 162.58494873046874, "max": 259.8734130859375, "pos_frac": 0.671875, "sample": [97.1531982421875, -100.97891235351562, 142.89434814453125, 0.0, 259.8734130859375, 153.57398986816406, -4.637359619140625, 180.0159454345703, 42.21917724609375, 159.54901123046875, -149.03076171875, 143.6212158203125, -139.04425048828125, 4.324317932128906, 6.4962005615234375, 26.516571044921875, 61.59740447998047, 120.42422485351562, 198.75042724609375, -21.717529296875, 119.52081298828125, 200.474609375, -146.67941284179688, -26.930145263671875, -13.040679931640625, -40.10087585449219, 92.6920394897461, 123.85455322265625, 162.96011352539062, 171.45346069335938, 60.944305419921875, 17.13195037841797, 0.0, 12.673583984375, 20.371536254882812, -90.49575805664062, -18.24127197265625, 12.212860107421875, 113.08697509765625, 85.69921875, -64.36433410644531, 26.582374572753906, 119.78440856933594, 140.87355041503906, -28.211502075195312, -46.008697509765625, 166.13613891601562, -20.85767364501953, -6.386791229248047, 4.2937469482421875, 110.16998291015625, -19.648834228515625, 172.8035888671875, -43.587432861328125, 12.801651000976562, 130.26089477539062, -3.7359466552734375, 174.46292114257812, 87.777587890625, -29.464324951171875, -14.155288696289062, 62.792083740234375, 141.9774932861328, 6.068267822265625, 17.977697372436523, -54.95233154296875, 85.07467651367188, -170.40231323242188, 181.13038635253906, -107.40142822265625, 63.38427734375, 156.36776733398438, 129.37014770507812, 97.43697357177734, 27.312255859375, -0.75018310546875, -0.9580078125, 157.8916015625, 11.407707214355469, -135.39498901367188, -12.632797241210938, 180.87759399414062, 152.53770446777344, -17.778030395507812, 20.044559478759766, 9.05108642578125, 28.564163208007812, 135.68150329589844, 68.78947448730469, 18.321685791015625, -115.45474243164062, 23.92613983154297, -2.54376220703125, -175.33294677734375, -8.564373016357422, 180.16737365722656, 156.4842529296875, 142.46102905273438, 10.161741256713867, 154.2930450439453, 4.953559875488281, 119.78729248046875, 162.00042724609375, 191.84051513671875, -4.98248291015625, 10.034156799316406, 33.448028564453125, 27.87615966796875, -32.598114013671875, 1.770822525024414, 162.42416381835938, 10.93865966796875, -31.313705444335938, 9.723297119140625, 14.643600463867188, -65.13894653320312, 139.49911499023438, 13.922271728515625, 143.38186645507812, 71.742431640625, -106.7716293334961, -95.80018615722656, 32.909454345703125, 6.706695556640625, 31.9664306640625, 129.55027770996094, 122.96170806884766, -37.81555938720703], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000388.npy"}
{"epoch": 0.812565445026178, "step": 389, "batch_size": 128, "mean": 59.4467887878418, "std": 85.4936294555664, "min": -159.61572265625, "p10": -30.603165435791013, "median": 58.800079345703125, "p90": 161.9789840698242, "max": 194.13311767578125, "pos_frac": 0.7421875, "sample": [0.9051361083984375, -116.72303771972656, -19.00055694580078, 103.30545043945312, 145.34146118164062, 145.38165283203125, 26.68896484375, 25.96820068359375, -9.218505859375, 180.27099609375, 28.56439208984375, 164.0597686767578, 29.20819091796875, 185.1806640625, 21.28015899658203, 140.6405792236328, 15.371017456054688, 145.697509765625, 100.22540283203125, 187.67092895507812, -49.267822265625, 130.24368286132812, 167.0833740234375, 88.41537475585938, 4.667716979980469, 153.59219360351562, -156.57952880859375, 179.6114501953125, 35.65483093261719, 166.48049926757812, 33.93994140625, -159.61572265625, 129.1728515625, -29.601036071777344, -59.8939208984375, -1.87054443359375, -44.61699676513672, 132.900634765625, 179.38934326171875, 181.86447143554688, 141.50962829589844, 0.0, -109.29573059082031, -5.030693054199219, 144.464111328125, -92.026123046875, 161.9241943359375, 136.43743896484375, 147.77235412597656, 20.87359619140625, -10.055793762207031, 144.96707153320312, 41.3935546875, 12.478250503540039, 117.69865417480469, 95.33010864257812, 34.61424255371094, 49.434356689453125, -14.987579345703125, -3.7864913940429688, 38.75511169433594, -2.274169921875, 103.40385437011719, 149.60890197753906, -53.777099609375, 136.445556640625, 82.74606323242188, 139.9033660888672, 123.14605712890625, 114.35896301269531, 137.2557830810547, -32.94146728515625, 43.5419921875, 153.62208557128906, 2.92498779296875, 194.13311767578125, 94.85894775390625, 74.42669677734375, -29.014389038085938, -6.12823486328125, 57.717926025390625, 119.81600952148438, 31.51287078857422, 64.1217041015625, 17.0155029296875, 152.15679931640625, -153.65545654296875, 129.31182861328125, 120.55694580078125, 2.170318603515625, 136.365478515625, 24.77215576171875, 3.97119140625, 9.315494537353516, -4.945503234863281, 97.25006866455078, 9.139129638671875, -156.49749755859375, -26.18780517578125, 162.10682678222656, 3.97216796875, 155.2491455078125, -14.9029541015625, -23.667388916015625, 81.4302978515625, 174.21316528320312, 0.0, 64.49702453613281, 101.27571105957031, 187.1219940185547, 6.254495620727539, 59.882232666015625, -4.6131134033203125, 92.02337646484375, 87.79701232910156, 153.2149200439453, 125.29156494140625, 8.734893798828125, -23.837570190429688, 18.637001037597656, 102.63447570800781, 147.04757690429688, -138.0616455078125, 18.683029174804688, 139.90892028808594, -1.526641845703125, 84.24911499023438, 146.566162109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000389.npy"}
{"epoch": 0.8146596858638744, "step": 390, "batch_size": 128, "mean": 56.55153274536133, "std": 93.28559875488281, "min": -173.86553955078125, "p10": -59.46224670410156, "median": 52.014076232910156, "p90": 171.11105651855468, "max": 197.66748046875, "pos_frac": 0.7421875, "sample": [113.573974609375, 2.6117630004882812, 45.297943115234375, 19.069793701171875, 80.0958251953125, 137.77708435058594, 0.5295848846435547, 163.13113403320312, 126.2423095703125, 79.21035766601562, -35.48455810546875, 7.1274871826171875, 2.3871231079101562, -61.17279052734375, 33.828857421875, 120.12899780273438, 2.377704620361328, 175.06787109375, 154.79281616210938, 140.40304565429688, 83.96002197265625, -30.925262451171875, 158.78416442871094, 12.687973022460938, 60.98382568359375, 81.33599853515625, -123.85919189453125, 151.14669799804688, 150.59130859375, 112.5311279296875, 159.16107177734375, 129.92393493652344, -110.03492736816406, 42.925132751464844, 151.63485717773438, 35.3917236328125, 89.45967864990234, 20.054367065429688, -2.7362823486328125, 98.75895690917969, -172.0404052734375, 141.56399536132812, -61.57872009277344, 128.17288208007812, 49.349090576171875, -58.729156494140625, -20.380287170410156, -164.22854614257812, 140.18893432617188, -29.089508056640625, 54.67906188964844, -43.954681396484375, -135.70486450195312, 31.860870361328125, 191.9635009765625, 15.117464065551758, 22.910186767578125, 167.5153045654297, 132.4617156982422, -143.48623657226562, -108.39309692382812, 98.47079467773438, 155.54417419433594, 19.95690155029297, 186.75625610351562, -143.90878295898438, 195.58721923828125, 82.36514282226562, -15.336700439453125, 144.55667114257812, 58.4022216796875, 190.882568359375, 72.1258544921875, 0.0, -17.410888671875, -1.3375244140625, 171.05596923828125, -27.520065307617188, 139.47052001953125, 171.82211303710938, -19.02588653564453, 164.87484741210938, 185.3226318359375, -173.86553955078125, -16.99139404296875, 160.343017578125, 197.66748046875, 154.8679656982422, 3.4073486328125, 8.530672073364258, 9.940467834472656, 178.775146484375, 21.277496337890625, 63.273590087890625, 35.23289489746094, 123.8344955444336, -11.796791076660156, 167.09848022460938, 79.57635498046875, -54.38592529296875, 4.709394454956055, -88.69561767578125, 178.21200561523438, 154.46737670898438, -18.519248962402344, 39.472412109375, 187.501953125, 5.79608154296875, 1.4594154357910156, 6.6982421875, -37.2672119140625, 4.1959381103515625, -10.122932434082031, 127.320556640625, 92.3880386352539, 26.606689453125, 159.98553466796875, 152.89859008789062, 61.99424743652344, 166.9151611328125, 134.84814453125, 28.86962890625, 3.57647705078125, 172.38079833984375, 171.23959350585938, -81.84146118164062, 119.30009460449219, -2.2028961181640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000390.npy"}
{"epoch": 0.8167539267015707, "step": 391, "batch_size": 128, "mean": 54.168548583984375, "std": 79.53656768798828, "min": -148.11529541015625, "p10": -29.31005554199218, "median": 36.466888427734375, "p90": 157.11989135742186, "max": 314.52288818359375, "pos_frac": 0.7578125, "sample": [-6.503242492675781, -36.55986785888672, 4.54688835144043, 40.895355224609375, 36.40052795410156, 9.60821533203125, -0.968505859375, 5.481075286865234, 13.389545440673828, 109.42042541503906, 3.8195419311523438, 14.730377197265625, -14.703750610351562, 89.991943359375, 180.19482421875, 67.56594848632812, 50.80780029296875, 102.46843719482422, -25.866317749023438, -20.27698516845703, -16.753753662109375, 150.69427490234375, 73.087646484375, 62.876739501953125, -54.380828857421875, 45.10771179199219, 157.61447143554688, 22.95068359375, 1.8376502990722656, 34.484588623046875, 5.571868896484375, -62.8594970703125, 150.9765625, -8.971328735351562, 89.59759521484375, 91.056884765625, 126.2825927734375, 9.663864135742188, -13.255035400390625, 123.17901611328125, 9.491619110107422, 133.69325256347656, 141.8495635986328, 207.7890625, 151.8232421875, -1.454193115234375, 156.90792846679688, 12.267181396484375, -11.556716918945312, -2.313385009765625, 197.67156982421875, 63.59949493408203, -61.60957336425781, 60.942535400390625, 5.54132080078125, 109.41925048828125, 125.81228637695312, 88.28895568847656, 140.9742431640625, -34.57183837890625, 36.884735107421875, -119.52946472167969, 133.83370971679688, 24.80072784423828, 31.759521484375, 12.192138671875, 140.0316162109375, 8.253665924072266, 51.74481201171875, 5.5895843505859375, 0.616424560546875, 168.36328125, 20.932769775390625, 64.14783477783203, 128.3663787841797, -7.450469970703125, 123.84326171875, 70.25399780273438, -8.770736694335938, 151.06527709960938, -6.1463623046875, 84.36212158203125, 155.689453125, 314.52288818359375, 176.13543701171875, 7.628814697265625, -9.229568481445312, -34.6578369140625, 149.38894653320312, 11.334381103515625, 12.875, 1.0617008209228516, -147.09066772460938, 153.86859130859375, 44.120025634765625, 54.7025146484375, -33.93206787109375, 123.34547424316406, 175.65106201171875, 173.768798828125, -37.41754150390625, -46.32501220703125, 36.53324890136719, -105.37759399414062, 4.947967529296875, -2.413806915283203, 45.85045623779297, 129.93646240234375, 121.51220703125, 158.3641357421875, 13.291015625, 135.34378051757812, 141.6466064453125, 27.536148071289062, 85.43820190429688, 65.7296142578125, 174.47637939453125, 144.78335571289062, 205.0550537109375, 14.643798828125, -27.329193115234375, 9.730712890625, 31.97052001953125, 6.429267883300781, -6.942955017089844, 42.40386962890625, -148.11529541015625, 159.77511596679688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000391.npy"}
{"epoch": 0.818848167539267, "step": 392, "batch_size": 128, "mean": 69.38008880615234, "std": 81.57086181640625, "min": -150.14968872070312, "p10": -9.939663696289061, "median": 53.2689208984375, "p90": 184.09567565917968, "max": 221.16757202148438, "pos_frac": 0.765625, "sample": [157.12962341308594, 136.20999145507812, 60.82273864746094, 42.73468017578125, 46.76240539550781, 2.6127395629882812, 70.52487182617188, 85.78823852539062, 77.32659912109375, 106.2691650390625, 20.604995727539062, 87.62982177734375, 114.31745910644531, 32.65277099609375, -10.983489990234375, -2.144500732421875, -150.14968872070312, 217.29693603515625, 100.63471984863281, -101.28363037109375, 186.37518310546875, -6.961963653564453, 7.6764678955078125, 67.27450561523438, 177.73587036132812, 198.3367919921875, 0.0, -35.86590576171875, 150.78872680664062, 136.679443359375, 7.185005187988281, 31.2216796875, 187.03103637695312, 49.533416748046875, 48.739540100097656, 187.08441162109375, 183.56472778320312, 164.0196533203125, 139.1405029296875, 45.50877380371094, 143.3458251953125, 156.83895874023438, 14.350830078125, 27.028472900390625, -7.700408935546875, 20.99547576904297, -6.882316589355469, 5.617919921875, 112.73373413085938, 161.95823669433594, 49.659576416015625, 169.73587036132812, 193.38787841796875, 65.89497375488281, 183.85501098632812, 132.0284423828125, 8.992950439453125, 109.05084228515625, 181.504150390625, -5.756988525390625, -11.687049865722656, 108.0404281616211, -21.824737548828125, 183.7320556640625, 37.99396896362305, 55.55785369873047, -28.455734252929688, 131.97592163085938, 197.71224975585938, -1.0, 54.756744384765625, 177.08636474609375, 190.95489501953125, 135.03756713867188, 84.84967041015625, 187.16094970703125, 38.1636962890625, -128.09951782226562, -2.0289993286132812, -1.1565284729003906, 9.10223388671875, 155.698974609375, 6.6341552734375, 106.51837158203125, 17.785659790039062, -4.9996337890625, -9.4923095703125, 7.309478759765625, 22.645965576171875, -26.501174926757812, -0.59149169921875, 0.0, 136.28427124023438, 23.45489501953125, 151.06253051757812, 196.09149169921875, 54.53973388671875, 148.18960571289062, 184.6572265625, 2.021881103515625, 175.33175659179688, -0.582855224609375, 10.181610107421875, 1.5777530670166016, 55.1483154296875, 221.16757202148438, -1.9597396850585938, 192.230712890625, 22.001998901367188, 81.68624114990234, -116.32151794433594, 165.24603271484375, -9.456695556640625, 7.9679107666015625, 140.1512451171875, 26.5712890625, 60.50286865234375, -38.7779541015625, -8.62933349609375, 89.16514587402344, 51.99810791015625, 141.99331665039062, -30.54010009765625, 152.1612548828125, -26.31646728515625, 24.09278106689453, 156.50027465820312, 31.914337158203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000392.npy"}
{"epoch": 0.8209424083769633, "step": 393, "batch_size": 128, "mean": 51.15583419799805, "std": 87.60706329345703, "min": -188.21630859375, "p10": -36.35468444824218, "median": 33.54255676269531, "p90": 155.05066986083983, "max": 298.8204345703125, "pos_frac": 0.75, "sample": [14.550800323486328, 6.847137451171875, 4.1508941650390625, 101.88275146484375, 112.0399398803711, -64.82843017578125, 108.07515716552734, 102.81866455078125, 147.94534301757812, 2.68646240234375, 12.19183349609375, -2.98095703125, 82.41232299804688, -8.656944274902344, 141.97146606445312, 29.204193115234375, -126.99176025390625, 109.1004638671875, 70.67955780029297, 101.98779296875, 163.27374267578125, 51.51123046875, 12.1685791015625, 90.2227783203125, -40.7506103515625, 171.29501342773438, 157.7361297607422, 150.84866333007812, 32.536376953125, 28.43743896484375, 152.57696533203125, 1.123504638671875, 46.55364990234375, 144.842529296875, 27.242431640625, 66.59986877441406, -3.577667236328125, 153.44775390625, 24.713836669921875, 142.7625732421875, 12.2655029296875, -35.76654052734375, 49.9677734375, 156.99270629882812, -98.1456298828125, 161.251953125, -27.019866943359375, -26.776641845703125, -99.80351257324219, 79.21917724609375, 249.0992431640625, 60.495731353759766, 8.432586669921875, 20.015480041503906, 15.098426818847656, 60.43414306640625, 0.8056259155273438, -133.31883239746094, 145.12644958496094, 117.88536071777344, -174.85662841796875, 26.95330810546875, 14.426315307617188, -16.03082275390625, 20.276763916015625, 148.9873504638672, 107.85960388183594, 158.02755737304688, -166.8748779296875, -26.664276123046875, -37.727020263671875, 37.561134338378906, 29.454330444335938, 50.367645263671875, 15.858966827392578, 61.73541259765625, 193.93780517578125, -22.731277465820312, 46.677398681640625, -8.995540618896484, 9.115581512451172, -14.065086364746094, 142.28158569335938, -154.7701873779297, 17.234451293945312, 108.5041275024414, 147.3771514892578, 138.76736450195312, 140.11231994628906, 150.16073608398438, -25.885574340820312, 168.89291381835938, 147.34671020507812, 158.24111938476562, 159.5958709716797, 28.506195068359375, 34.548736572265625, -41.800872802734375, 154.21836853027344, -188.21630859375, 6.761955261230469, 35.843292236328125, 4.484619140625, -3.29180908203125, 298.8204345703125, 91.07147216796875, 38.07464599609375, 0.0, 8.356109619140625, 64.95782470703125, 206.099853515625, -41.8026123046875, 2.472076416015625, 9.684432983398438, -31.020843505859375, 122.29949951171875, 143.5833740234375, -9.563125610351562, -23.306289672851562, -4.2892608642578125, 0.1968212127685547, -26.381072998046875, 152.34954833984375, 153.99154663085938, 2.3472137451171875, 102.435302734375, 125.4095458984375, 145.04318237304688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000393.npy"}
{"epoch": 0.8230366492146597, "step": 394, "batch_size": 128, "mean": 50.8775749206543, "std": 80.98516082763672, "min": -138.6477508544922, "p10": -43.05140762329101, "median": 34.86332702636719, "p90": 166.52208709716797, "max": 209.73162841796875, "pos_frac": 0.6953125, "sample": [165.244384765625, 71.22787475585938, 169.02450561523438, 130.90350341796875, 52.61968994140625, 152.5877685546875, -12.15960693359375, -3.7770614624023438, 23.78466796875, 169.36587524414062, 88.19970703125, 5.1442718505859375, 34.15850830078125, 109.8349609375, -52.47540283203125, -2.5548248291015625, 160.33131408691406, -45.39788055419922, 156.69403076171875, -22.753555297851562, 9.249542236328125, 148.29901123046875, 203.3934326171875, -15.930084228515625, 0.9576950073242188, -70.35690307617188, 35.568145751953125, 137.0156707763672, 107.780517578125, 37.241302490234375, -34.19805908203125, 0.0, -28.96115493774414, -65.63429260253906, 164.487548828125, 12.918807983398438, 14.943084716796875, 113.27935028076172, 6.380584716796875, 48.394371032714844, 92.92495727539062, 89.36970520019531, 46.27642822265625, -138.6477508544922, 17.771682739257812, 105.82891845703125, 160.6392822265625, 60.74454116821289, 3.197296142578125, 18.341796875, -18.58636474609375, -42.0457763671875, 47.063690185546875, -14.7958984375, -20.678955078125, 175.5753173828125, 9.323806762695312, 68.20513916015625, 10.00799560546875, 56.45965576171875, 70.83116149902344, -53.470947265625, -0.1917724609375, -55.522247314453125, 161.3790283203125, 4.057073593139648, 44.283042907714844, 99.62088012695312, 20.89252471923828, 115.57992553710938, 143.24200439453125, 166.8201446533203, 175.40472412109375, 4.361761093139648, 200.35198974609375, -6.277099609375, -64.15289306640625, -33.17144775390625, -7.6354827880859375, 209.73162841796875, 18.068004608154297, -121.26531982421875, 70.1123046875, -21.401214599609375, -54.937744140625, 20.101882934570312, 40.33990478515625, 29.169265747070312, 8.264762878417969, 99.04891967773438, 0.0871429443359375, -12.066314697265625, 110.77984619140625, 13.987045288085938, -23.81573486328125, -28.334686279296875, 44.460968017578125, 51.24505615234375, 62.31758117675781, 60.533203125, -134.84620666503906, 147.64291381835938, 32.03118896484375, -30.012832641601562, 181.4775390625, 208.45248413085938, -28.316818237304688, 49.9478759765625, 89.46517944335938, 169.650634765625, 118.87850952148438, 133.1053466796875, 157.445556640625, 149.51364135742188, 167.44998168945312, -47.31744384765625, 185.72412109375, 17.79986572265625, -0.452545166015625, 166.39434814453125, 16.654815673828125, 135.22103881835938, -57.541900634765625, 133.0750732421875, -11.301406860351562, -14.599777221679688, -35.724639892578125, 147.87716674804688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000394.npy"}
{"epoch": 0.8251308900523561, "step": 395, "batch_size": 128, "mean": 53.3663330078125, "std": 81.7880859375, "min": -147.19381713867188, "p10": -25.495524597167968, "median": 28.301963806152344, "p90": 167.75290069580078, "max": 202.44580078125, "pos_frac": 0.7421875, "sample": [53.52886962890625, 7.30096435546875, -17.799057006835938, -6.970344543457031, -7.59564208984375, -16.829437255859375, 2.857086181640625, 200.9183349609375, 189.57305908203125, 26.943222045898438, 42.88776397705078, 49.841400146484375, 27.786346435546875, -1.260711669921875, 1.775064468383789, 37.92903137207031, 24.743080139160156, -81.36923217773438, 54.02113342285156, -21.408935546875, 42.54803466796875, 155.6292724609375, -19.3616943359375, -60.38880920410156, 177.69989013671875, 124.00595092773438, 20.596221923828125, 1.4894638061523438, 142.603759765625, 37.47407531738281, 131.30186462402344, 135.04537963867188, 107.9727783203125, 108.29023742675781, 46.70513916015625, 163.75149536132812, 100.09332275390625, 47.98419189453125, -0.01611328125, -2.9767990112304688, 101.7874755859375, 13.073856353759766, -13.189239501953125, 19.78839111328125, 163.70703125, 126.29269409179688, -13.406692504882812, -94.41082000732422, 131.93405151367188, 170.5384521484375, 187.626953125, -72.185791015625, 69.48428344726562, 149.202392578125, 163.48532104492188, 118.62283325195312, 42.773712158203125, 163.80059814453125, 15.8701171875, 113.84382629394531, -25.2978515625, -86.49496459960938, 21.606399536132812, -14.532501220703125, 154.3390655517578, -25.46893310546875, 17.85034942626953, 104.39311981201172, -13.319793701171875, 9.44171142578125, -14.813079833984375, 28.445068359375, -39.60406494140625, -85.41802978515625, 35.666412353515625, 82.39360046386719, 140.4140167236328, -2.2051162719726562, 16.570281982421875, 17.8515625, 23.935333251953125, 19.240882873535156, 202.44580078125, 167.17120361328125, 161.11627197265625, -21.09002685546875, 5.701751708984375, -55.6702880859375, 14.322357177734375, 56.31639099121094, 8.993743896484375, 94.02479553222656, -136.4552001953125, 98.50897979736328, -15.514404296875, 12.031730651855469, 179.07940673828125, 185.70042419433594, 169.2565460205078, 168.70370483398438, -7.260597229003906, 107.68362426757812, 26.59710693359375, -25.557571411132812, 126.85986328125, 91.53172302246094, 158.1798858642578, 10.54656982421875, 151.33642578125, -131.07936096191406, 0.94732666015625, 143.1578369140625, 178.60708618164062, -68.487548828125, 11.9171142578125, -147.19381713867188, 34.3310546875, 6.728414535522461, 177.36083984375, 175.12338256835938, 162.68023681640625, 167.3454132080078, 20.590972900390625, 7.72320556640625, 28.158859252929688, 113.03923034667969, 6.2445831298828125, 58.18304443359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000395.npy"}
{"epoch": 0.8272251308900523, "step": 396, "batch_size": 128, "mean": 52.85405731201172, "std": 87.9176025390625, "min": -200.9542236328125, "p10": -47.85791015625, "median": 45.70604705810547, "p90": 164.25335388183592, "max": 218.3746337890625, "pos_frac": 0.6953125, "sample": [129.6148681640625, 15.884063720703125, -10.051025390625, 167.05755615234375, 149.77542114257812, 16.198287963867188, -5.202232360839844, 17.124183654785156, -40.27001953125, 183.44393920898438, 103.02239990234375, 1.7071533203125, 96.86363220214844, 42.77691650390625, 1.3664398193359375, 49.058349609375, 167.76876831054688, -65.9599609375, 6.98919677734375, 137.45571899414062, -32.129302978515625, -9.730125427246094, -5.355316162109375, 218.3746337890625, -2.87884521484375, 34.778961181640625, 128.831787109375, 138.41485595703125, 57.41429138183594, 108.2179946899414, 75.6053466796875, -117.55397033691406, 70.60702514648438, 158.6112060546875, -124.10873413085938, 165.57852172851562, 16.512657165527344, -15.1202392578125, -5.144374847412109, 104.14845275878906, 41.67250061035156, 139.30804443359375, 74.9234619140625, 152.2802734375, 181.50384521484375, -8.1036376953125, 13.917028427124023, 13.85308837890625, -3.5400390625, 139.38681030273438, 190.915771484375, 182.34622192382812, 215.72171020507812, 5.411861419677734, 158.82113647460938, 82.5233154296875, 147.93853759765625, 0.5615234375, 13.80206298828125, -1.8741321563720703, 7.444122314453125, -25.677490234375, -74.03144073486328, -45.204010009765625, 96.4376220703125, -77.75199890136719, 50.802642822265625, -141.28317260742188, 13.746429443359375, -6.4532623291015625, 142.20462036132812, -6.5062255859375, -30.975875854492188, -2.4698104858398438, 143.12774658203125, 168.12930297851562, 65.73190307617188, 140.8021240234375, 20.64691162109375, -148.0240478515625, 71.57958984375, 114.3313217163086, 75.30553436279297, 136.17913818359375, 53.4644775390625, 163.6854248046875, 22.836170196533203, 186.5730743408203, -116.7251205444336, 152.9320068359375, -25.335205078125, 85.11053466796875, -49.066162109375, 107.60224914550781, 81.4742202758789, 72.37923431396484, 120.063720703125, -3.501953125, 17.25262451171875, 44.143798828125, 161.44589233398438, 65.07899475097656, -1.474517822265625, -200.9542236328125, -47.340087890625, 161.3651580810547, -11.419479370117188, 180.7054443359375, 158.9088134765625, 168.23069763183594, 0.0, 18.126953125, 156.43917846679688, 38.878082275390625, 36.99403381347656, 10.254356384277344, 0.0, 47.26829528808594, 93.60536193847656, -16.811691284179688, 49.699981689453125, -132.0245361328125, 82.97201538085938, 147.22781372070312, 147.36138916015625, -75.57415771484375, -96.70722961425781, 151.0143585205078], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000396.npy"}
{"epoch": 0.8293193717277487, "step": 397, "batch_size": 128, "mean": 50.84654998779297, "std": 87.17218780517578, "min": -173.7401123046875, "p10": -44.39212951660156, "median": 38.073760986328125, "p90": 164.79078979492186, "max": 201.4808349609375, "pos_frac": 0.703125, "sample": [-8.3031005859375, 7.776824951171875, -12.091117858886719, 19.68911361694336, 174.400146484375, 48.089752197265625, -109.15906524658203, 47.114013671875, -4.447669982910156, 18.02520751953125, 9.878280639648438, 14.93560791015625, 122.76661682128906, 20.18902587890625, 26.800186157226562, 166.80653381347656, 12.433792114257812, 110.06645965576172, 195.34188842773438, 136.01119995117188, -136.4497833251953, 25.691802978515625, -19.674163818359375, 87.70269775390625, 135.10726928710938, 148.909423828125, 0.0, -2.975830078125, 124.02279663085938, 41.897369384765625, 0.0, -36.788360595703125, 134.3988037109375, -14.321914672851562, -4.22393798828125, 2.320425033569336, 8.7244873046875, 151.8689422607422, -78.68142700195312, 140.35121154785156, 170.96658325195312, 33.35570526123047, 142.50225830078125, 165.96597290039062, -29.562835693359375, 0.42850494384765625, 138.84335327148438, 151.13577270507812, 201.4808349609375, -43.683685302734375, 197.18478393554688, -46.045166015625, 185.490478515625, -173.7401123046875, -109.15725708007812, -40.98150634765625, 44.306121826171875, -125.57754516601562, -0.116668701171875, -150.34564208984375, -36.30743408203125, 53.37614440917969, 164.01385498046875, 138.28262329101562, 88.99153137207031, 27.732128143310547, 133.036865234375, -10.17474365234375, 38.19561767578125, -1.3582839965820312, 59.36077880859375, 188.8082275390625, 74.19717407226562, 50.356903076171875, 146.77737426757812, 140.50961303710938, 143.55563354492188, -28.764495849609375, 97.90121459960938, 52.7559814453125, 150.24636840820312, 29.86578369140625, 172.79693603515625, 168.57754516601562, 37.951904296875, 0.8093032836914062, 144.1260528564453, -4.35614013671875, 11.300247192382812, 12.865447998046875, 132.14599609375, 17.378387451171875, 76.49272155761719, 9.722244262695312, 151.89434814453125, 143.5559539794922, 13.05440902709961, 16.6773681640625, 169.6171875, 62.76942443847656, 17.53759765625, -109.87908935546875, 40.01617431640625, -0.5024871826171875, 181.4965362548828, 164.28713989257812, 128.0610809326172, 4.787322998046875, -14.786720275878906, -19.545120239257812, -25.342132568359375, 163.171630859375, 131.2279052734375, -66.23565673828125, 43.225799560546875, 44.26898193359375, -126.59881591796875, -12.995010375976562, 140.66921997070312, 10.408935546875, -39.0238037109375, -47.81196594238281, 118.12844848632812, 126.45850372314453, 69.46923828125, -111.21044921875, 120.99285888671875, 122.69076538085938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000397.npy"}
{"epoch": 0.831413612565445, "step": 398, "batch_size": 128, "mean": 40.44446563720703, "std": 93.98536682128906, "min": -171.37347412109375, "p10": -79.88947906494138, "median": 22.035354614257812, "p90": 165.92162475585937, "max": 244.6943359375, "pos_frac": 0.65625, "sample": [13.58807373046875, 20.509857177734375, 147.88218688964844, -9.285736083984375, 244.6943359375, -96.09271240234375, 61.011322021484375, -3.982208251953125, 56.30351257324219, 166.21292114257812, -140.80703735351562, -59.423004150390625, -162.61276245117188, 13.252044677734375, 7.512672424316406, 22.75213623046875, 33.90362548828125, -5.871650695800781, 161.38442993164062, 171.34156799316406, 163.3011016845703, -135.10845947265625, -112.138671875, 134.25906372070312, 37.43096923828125, 119.03598022460938, 20.551788330078125, 140.48770141601562, 145.88681030273438, 114.03300476074219, -72.94523620605469, 9.083993911743164, 37.34429931640625, -171.37347412109375, -61.696136474609375, 138.94091796875, 25.99078369140625, 146.0438232421875, -0.9864501953125, -112.53247833251953, 1.9910297393798828, -98.8048095703125, 16.436767578125, 177.5277099609375, -0.7098150253295898, -54.260589599609375, 2.863800048828125, 135.43077087402344, 21.318572998046875, -10.507080078125, 32.596466064453125, -11.71051025390625, 106.24583435058594, -5.971954345703125, 137.265869140625, 54.88873291015625, 0.38406944274902344, -106.78988647460938, 23.602218627929688, -15.924755096435547, 49.418212890625, 41.4744873046875, 163.07546997070312, -29.580108642578125, 185.74609375, 7.366851806640625, 24.263633728027344, -147.0115966796875, 124.63711547851562, 171.34190368652344, 29.750930786132812, 197.49560546875, 15.159889221191406, 204.32827758789062, -11.712692260742188, -17.367645263671875, 9.94754409790039, 162.4027099609375, -30.92279052734375, -12.53759765625, 172.62863159179688, 176.62600708007812, 9.405235290527344, 35.37664794921875, 36.46387481689453, 6.22930908203125, -48.095680236816406, 140.6442108154297, 40.507171630859375, -50.305877685546875, 124.57974243164062, 119.22332763671875, -13.08294677734375, -148.893798828125, 73.6583251953125, -7.50482177734375, 145.23455810546875, -26.2091064453125, -13.52728271484375, 106.72245788574219, -17.27979278564453, -11.403717041015625, 142.9666748046875, 122.16923522949219, 13.868743896484375, 14.615631103515625, 60.70758056640625, 167.59130859375, 160.11288452148438, -33.09344482421875, -100.46375274658203, 182.9005126953125, 138.05902099609375, -50.003883361816406, 163.4949951171875, -4.002716064453125, -43.96165466308594, -167.88259887695312, 12.789703369140625, 165.79678344726562, 126.43280029296875, -6.87359619140625, 6.039604187011719, 214.02166748046875, 45.31634521484375, 29.014480590820312, 128.6407012939453, 86.60678100585938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000398.npy"}
{"epoch": 0.8335078534031414, "step": 399, "batch_size": 128, "mean": 47.99359130859375, "std": 81.43340301513672, "min": -168.9639892578125, "p10": -35.50116882324219, "median": 26.592424392700195, "p90": 163.28624877929687, "max": 208.14007568359375, "pos_frac": 0.703125, "sample": [-2.622833251953125, 138.72276306152344, 142.14300537109375, 147.67236328125, 11.1556396484375, -16.7706298828125, 0.140594482421875, 39.128326416015625, 36.94964599609375, 173.3492431640625, 47.73833465576172, -59.430023193359375, -36.176971435546875, 152.63902282714844, -2.7109603881835938, 56.19560241699219, 88.67454528808594, 5.2335205078125, 176.63307189941406, 29.264694213867188, -2.53546142578125, -77.72482299804688, -20.644500732421875, 161.93988037109375, -4.68182373046875, 0.75390625, 13.811172485351562, -132.33706665039062, -168.9639892578125, 130.09080505371094, -4.001556396484375, 162.63323974609375, 27.908588409423828, 173.61761474609375, 19.558509826660156, -8.992919921875, 145.37399291992188, -7.588676452636719, -0.4163055419921875, 182.82577514648438, 35.51251220703125, 159.5115203857422, 208.14007568359375, 150.41500854492188, -25.77117919921875, -11.248115539550781, 162.00897216796875, -53.141143798828125, 9.016855239868164, 145.84744262695312, -35.909637451171875, 41.61241912841797, 132.1986541748047, 102.91506958007812, 183.93414306640625, -7.96484375, 6.823635101318359, 8.5596923828125, -0.72076416015625, -9.137561798095703, 10.261566162109375, 165.04718017578125, -2.51788330078125, 25.276260375976562, 150.451171875, 11.592388153076172, 148.89039611816406, 20.31824493408203, 97.24855041503906, 43.52081298828125, 69.40440368652344, 112.10214233398438, -10.498123168945312, -128.78244018554688, -136.96627807617188, -131.51611328125, -12.897216796875, 33.86541748046875, 164.8099365234375, 171.24493408203125, 22.230926513671875, 145.6673583984375, 63.7745361328125, 147.41915893554688, 127.71322631835938, 62.48773193359375, 39.005859375, 35.22691345214844, 24.8748779296875, 143.4564208984375, 13.520740509033203, -41.803314208984375, 13.29238510131836, 39.96063232421875, 16.978759765625, 20.358779907226562, -34.263336181640625, 1.6524162292480469, 172.67214965820312, 7.6953887939453125, 77.59808349609375, 20.49005126953125, 61.989013671875, -48.471954345703125, 9.01165771484375, 167.1082763671875, -19.284454345703125, 197.41763305664062, 96.02505493164062, 61.4388427734375, 174.52220153808594, -107.90362548828125, -35.32611083984375, 37.79374694824219, 25.049602508544922, 23.491806030273438, 86.89320373535156, 111.91006469726562, -7.0020599365234375, 142.8846435546875, -32.7572021484375, 34.041748046875, 124.38951110839844, 84.36300659179688, -17.220245361328125, -2.253265380859375, 115.82371520996094, 17.221588134765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000399.npy"}
{"epoch": 0.8356020942408376, "step": 400, "batch_size": 128, "mean": 56.227325439453125, "std": 85.41888427734375, "min": -177.97555541992188, "p10": -38.95774230957031, "median": 35.08758544921875, "p90": 162.8911331176758, "max": 203.44583129882812, "pos_frac": 0.734375, "sample": [5.3216705322265625, -60.97071838378906, 11.800445556640625, 34.39990234375, 129.5244140625, 26.936622619628906, 116.54410552978516, 119.533935546875, 85.59812927246094, -100.12299346923828, 6.1981048583984375, 87.05914306640625, -177.97555541992188, 22.701698303222656, 169.82212829589844, 15.7967529296875, 43.837730407714844, 153.0606231689453, 124.50372314453125, 155.82833862304688, 193.51287841796875, -0.7705841064453125, 21.521286010742188, 124.43325805664062, 151.69369506835938, 114.32034301757812, -45.2578125, 136.50875854492188, -23.1851806640625, 136.18109130859375, 43.10050964355469, 8.384761810302734, 15.938911437988281, 127.62120056152344, 42.94303894042969, 164.60894775390625, 156.87847900390625, 107.286865234375, 23.370758056640625, -4.685302734375, -143.93646240234375, 141.40109252929688, 127.10261535644531, 161.67236328125, 24.45745849609375, 159.42575073242188, -1.0301513671875, 106.91748046875, 7.7882080078125, -15.736602783203125, 67.53582763671875, 136.96983337402344, 173.81292724609375, -117.09228515625, 0.47955322265625, -31.311660766601562, 9.067652702331543, 12.9473876953125, 128.12982177734375, -2.323680877685547, -1.333831787109375, 46.919036865234375, 170.40762329101562, 138.77288818359375, 31.342010498046875, -36.6680908203125, 19.608856201171875, 153.181640625, 74.10528564453125, 172.0990447998047, 8.4239501953125, 3.8884201049804688, -23.322433471679688, 27.568389892578125, -50.093536376953125, -83.22412109375, 163.78958129882812, 172.6386260986328, 3.315093994140625, -44.300262451171875, 92.98597717285156, 22.977752685546875, 135.21624755859375, 107.81173706054688, -3.5780372619628906, 162.77288818359375, 133.21096801757812, -24.680633544921875, 38.22306823730469, 47.49488830566406, 128.97698974609375, -29.890838623046875, -7.9940643310546875, -3.5604724884033203, 2.1209716796875, 152.693603515625, -24.388290405273438, 129.38951110839844, 2.762174606323242, -114.68310546875, 35.7752685546875, 172.48098754882812, 132.75994873046875, 25.333709716796875, -11.57342529296875, 22.081298828125, 131.41270446777344, -5.942192077636719, 9.695846557617188, 162.4952850341797, 145.01747131347656, 23.240585327148438, 155.55325317382812, -12.92236328125, 155.75088500976562, 166.2350311279297, 153.35592651367188, -21.078079223632812, 203.44583129882812, 181.37603759765625, 163.1670379638672, 158.90875244140625, 0.0, -72.9180908203125, -88.7137451171875, 3.125579833984375, -142.9169921875, 46.890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000400.npy"}
{"epoch": 0.837696335078534, "step": 401, "batch_size": 128, "mean": 65.35639190673828, "std": 83.63126373291016, "min": -220.37548828125, "p10": -22.843792724609372, "median": 54.174285888671875, "p90": 170.43948059082032, "max": 235.6143798828125, "pos_frac": 0.7890625, "sample": [1.75469970703125, 4.152191162109375, 25.261016845703125, -7.987823486328125, 18.7652587890625, 112.70948791503906, 16.4281005859375, 124.59160614013672, 92.35723876953125, -5.776580810546875, 149.9959716796875, 21.602935791015625, 42.26793670654297, 164.88137817382812, 86.916015625, 31.302200317382812, 77.9423828125, 144.33447265625, -9.47216796875, 130.49032592773438, -3.258270263671875, 34.092742919921875, -19.082015991210938, 75.06768798828125, -6.469696044921875, 162.09194946289062, 70.52593994140625, 1.5845603942871094, 149.1351318359375, 80.49264526367188, 176.34176635742188, 202.08709716796875, 140.80006408691406, 55.34991455078125, 15.58062744140625, 13.82952880859375, 142.41419982910156, 137.2388916015625, 8.714271545410156, 139.116455078125, 235.6143798828125, -18.800262451171875, 157.3070831298828, 174.33282470703125, 184.68106079101562, 148.617431640625, 33.422760009765625, 17.387054443359375, 148.87033081054688, 146.39505004882812, -96.3514404296875, 132.9009552001953, 27.05120849609375, 12.333229064941406, -220.37548828125, 136.70516967773438, 10.47540283203125, 136.75115966796875, 105.35183715820312, -66.35441589355469, -37.908660888671875, 136.79885864257812, 184.12322998046875, 141.6671142578125, 126.99197387695312, 97.89105224609375, -53.80377197265625, -89.6951675415039, 158.5375518798828, -19.387725830078125, -1.3013343811035156, 1.2140579223632812, 197.6143798828125, 72.8460693359375, 34.553253173828125, 3.610137939453125, 153.4803009033203, 91.3243408203125, 46.94683837890625, 79.2742919921875, 126.5078125, -25.85650634765625, 146.11392211914062, 170.730712890625, 162.76820373535156, 177.75128173828125, 45.32904052734375, -4.4672393798828125, 59.76824951171875, -24.337188720703125, 105.697021484375, 7.937896728515625, -9.931915283203125, -22.203765869140625, 68.97027587890625, -24.623199462890625, -24.75006103515625, 172.26077270507812, 201.0137176513672, 24.673095703125, 30.14825439453125, 0.364593505859375, 170.31466674804688, 154.19448852539062, 173.37026977539062, 17.323516845703125, 5.88330078125, 12.464141845703125, -22.071510314941406, 72.06944274902344, 31.586029052734375, 0.53863525390625, 3.22283935546875, 12.981670379638672, -7.961647033691406, 2.1648025512695312, 52.9986572265625, 142.35618591308594, 149.8453369140625, -26.95661163330078, 111.24110412597656, 150.4495391845703, 157.44189453125, 0.10462188720703125, 196.2541961669922, -173.1920166015625, -69.16896057128906, 155.03662109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000401.npy"}
{"epoch": 0.8397905759162304, "step": 402, "batch_size": 128, "mean": 55.54277801513672, "std": 86.05521392822266, "min": -157.66180419921875, "p10": -43.00240325927732, "median": 39.04113006591797, "p90": 161.22025604248046, "max": 235.96127319335938, "pos_frac": 0.7421875, "sample": [-6.0155029296875, -3.465057373046875, -129.0476837158203, 45.945098876953125, 152.41212463378906, -20.271453857421875, -69.71853637695312, -28.0775146484375, 118.76889038085938, -157.66180419921875, -131.0048828125, 134.23367309570312, 21.083740234375, -6.521366119384766, 5.0479736328125, 208.62600708007812, 142.93971252441406, 130.419921875, 13.055068969726562, -28.3699951171875, 142.1310272216797, 2.77386474609375, 13.926429748535156, 181.076171875, -131.7713165283203, 4.548233985900879, 2.38677978515625, 145.91448974609375, -153.64468383789062, 35.09832763671875, 127.1937255859375, -1.40802001953125, 96.78396606445312, 156.82952880859375, 9.70819091796875, 0.0, 209.9066162109375, 154.939208984375, -18.382171630859375, 145.57992553710938, 47.5626220703125, 30.961944580078125, 144.3462677001953, 13.567916870117188, 114.7488784790039, 146.09762573242188, -67.18521118164062, -8.806732177734375, 173.71612548828125, 151.2305145263672, 25.38690185546875, -56.85142517089844, -37.067108154296875, 129.8668212890625, 126.16177368164062, 48.560882568359375, 70.79203796386719, 69.23283386230469, 2.781829833984375, 122.43550872802734, 12.580974578857422, 192.849853515625, 166.22705078125, 11.57769775390625, 123.0625991821289, -22.211166381835938, -82.82194519042969, 89.01959228515625, 42.30409240722656, -143.26785278320312, 154.7763671875, 148.63388061523438, 106.44635009765625, -2.1707763671875, 137.3832244873047, -6.79449462890625, 235.96127319335938, 4.703929901123047, 74.16580200195312, 169.94338989257812, 160.86874389648438, 162.25457763671875, 30.69476318359375, 114.92520904541016, 14.1134033203125, 38.761932373046875, 170.3526611328125, 116.4722900390625, -68.03921508789062, 136.27996826171875, 122.39395904541016, 77.80230712890625, 12.041145324707031, -0.7183761596679688, 36.38446044921875, 90.47906494140625, 5.9811859130859375, 80.61651611328125, 17.840576171875, 159.142333984375, 158.64450073242188, 162.0404510498047, 141.65896606445312, -26.023391723632812, 195.27337646484375, -5.3549957275390625, 39.32032775878906, 9.62274169921875, 27.83544921875, 166.0419921875, 13.236427307128906, 134.08950805664062, 87.92462158203125, 95.24526977539062, -64.02447509765625, 6.83087158203125, -0.3318634033203125, 50.985107421875, 61.31181335449219, 3.915802001953125, 123.23944091796875, 1.006814956665039, 6.018909454345703, -58.508575439453125, -15.2249755859375, 25.00201416015625, 117.05738830566406, -1.8833980560302734], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000402.npy"}
{"epoch": 0.8418848167539267, "step": 403, "batch_size": 128, "mean": 74.53414916992188, "std": 76.04407501220703, "min": -140.15432739257812, "p10": -13.524829101562498, "median": 59.3546142578125, "p90": 169.29732971191405, "max": 222.8450927734375, "pos_frac": 0.8203125, "sample": [-20.623641967773438, 136.83619689941406, 6.565460205078125, 159.91184997558594, 126.8472900390625, 53.60760498046875, 159.52597045898438, 106.15624237060547, 22.915679931640625, 26.2144775390625, 163.59027099609375, 167.4619140625, 44.91778564453125, 53.217529296875, 10.231338500976562, 93.02853393554688, 18.614501953125, 82.114501953125, 34.7088623046875, 9.688552856445312, 165.1747589111328, -16.98220443725586, -34.689849853515625, 33.326019287109375, 115.73786926269531, 140.97039794921875, 170.2808074951172, 47.1068115234375, 33.98748779296875, 187.1090087890625, 198.15423583984375, 4.222991943359375, -0.24967193603515625, 122.70704650878906, -1.279083251953125, 109.23038482666016, 16.0830078125, 28.8056640625, 0.0, -15.192840576171875, -0.12941741943359375, 155.490234375, 194.10134887695312, 93.30047607421875, 161.32220458984375, 158.59100341796875, 150.24172973632812, 20.900482177734375, 140.67916870117188, 199.67926025390625, 14.2308349609375, 42.08085632324219, 13.207305908203125, 1.2176170349121094, 172.91574096679688, 36.882568359375, 30.853729248046875, -52.100494384765625, 120.80450439453125, -11.463027954101562, -19.69281005859375, 74.3343505859375, 2.31463623046875, 174.07177734375, 144.53028869628906, 19.42230987548828, 192.54867553710938, 1.1699695587158203, 47.227378845214844, 25.4686279296875, 26.62725830078125, 143.064697265625, 20.897476196289062, 82.826416015625, -12.809967041015625, 94.84770202636719, 115.3623046875, 163.49893188476562, 4.342132568359375, 89.17172241210938, -1.6440296173095703, 56.661376953125, 178.43817138671875, 14.434173583984375, 163.61705017089844, -18.604095458984375, 11.789901733398438, 143.77853393554688, 130.96295166015625, 76.32746887207031, 214.93719482421875, 93.38128662109375, 222.8450927734375, 188.2080078125, 52.91859436035156, 174.75048828125, -34.114410400390625, 131.5521240234375, 150.411865234375, 133.1883544921875, 126.71815490722656, 90.83924102783203, 147.90390014648438, 33.007598876953125, -140.15432739257812, 168.87583923339844, -23.33843994140625, 153.29519653320312, -1.3125686645507812, 143.91644287109375, -52.56842041015625, -2.0227279663085938, 23.78619384765625, 61.1273193359375, -25.64697265625, 43.75660705566406, 48.18089294433594, 2.2313079833984375, 146.70941162109375, -4.703521728515625, -120.63677978515625, 148.52249145507812, 151.51788330078125, 97.54794311523438, 142.48025512695312, 114.05327606201172, 132.7989501953125, 57.5819091796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000403.npy"}
{"epoch": 0.8439790575916231, "step": 404, "batch_size": 128, "mean": 51.22079849243164, "std": 88.89654541015625, "min": -201.86456298828125, "p10": -39.125186157226565, "median": 31.53076171875, "p90": 168.1268768310547, "max": 228.68209838867188, "pos_frac": 0.71875, "sample": [170.85110473632812, 165.09298706054688, 35.115936279296875, 179.6184844970703, -136.28897094726562, 158.34890747070312, 182.85049438476562, 2.2189788818359375, -80.27909088134766, 0.0, -149.22329711914062, 79.65106964111328, 27.927406311035156, -23.0262451171875, -82.27523803710938, 149.15623474121094, 12.90277099609375, 174.3924560546875, 1.8453407287597656, -51.8199462890625, 163.31051635742188, 63.161041259765625, 19.81341552734375, -10.87261962890625, -19.659133911132812, 113.107177734375, -201.86456298828125, 153.80520629882812, -38.081298828125, 10.142509460449219, 9.228302001953125, -39.063232421875, 33.871063232421875, 11.277618408203125, 94.55940246582031, -9.970466613769531, 3.8235530853271484, -6.13604736328125, 228.68209838867188, 122.68759155273438, 156.36032104492188, -4.1413726806640625, -10.123626708984375, -53.21990966796875, 143.84115600585938, 60.66844940185547, 69.72589111328125, 29.190460205078125, 5.781848907470703, 20.00811767578125, 97.22184753417969, 8.015899658203125, 160.73220825195312, -24.4140625, 46.843902587890625, 191.5565185546875, -130.45965576171875, 16.94952392578125, -14.660491943359375, 20.64398193359375, 197.09298706054688, 179.27920532226562, -134.60476684570312, 145.830810546875, 103.66409301757812, 145.19290161132812, 148.3404541015625, 0.9775390625, -5.0533599853515625, 155.72532653808594, 136.68829345703125, 2.571126937866211, 12.73651123046875, -9.071136474609375, 28.1473388671875, 120.5029296875, 11.76171875, -16.8448486328125, 185.43487548828125, 41.11296081542969, 86.00172424316406, 20.378921508789062, 17.41107177734375, 63.80804443359375, 6.030242919921875, -32.49342346191406, -17.895248413085938, -15.384674072265625, 180.17764282226562, -81.0201416015625, -5.733089447021484, 120.27267456054688, 161.22073364257812, 40.900390625, -137.0163116455078, -39.269744873046875, 43.07292175292969, 11.604705810546875, 166.0970458984375, 70.97979736328125, -38.42302703857422, 8.55523681640625, 134.58615112304688, 55.687255859375, 166.9593505859375, 98.2574691772461, 34.04474639892578, 66.46690368652344, -115.5955581665039, 176.48928833007812, 55.15093994140625, 198.24407958984375, 162.4460906982422, 54.961883544921875, 28.09783935546875, 1.6062164306640625, 57.49687194824219, -11.049346923828125, 85.50262451171875, 144.33226013183594, 26.187286376953125, -3.6809234619140625, -2.161285400390625, 134.96075439453125, 143.83840942382812, 118.31459045410156, 165.5712890625, 191.38427734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000404.npy"}
{"epoch": 0.8460732984293193, "step": 405, "batch_size": 128, "mean": 58.28684997558594, "std": 82.87339782714844, "min": -145.5242919921875, "p10": -29.730575561523434, "median": 37.87266540527344, "p90": 173.62618255615234, "max": 214.692138671875, "pos_frac": 0.765625, "sample": [14.89532470703125, 5.7745361328125, 11.248687744140625, 0.0, -43.593963623046875, -3.516326904296875, 8.77716064453125, 143.16595458984375, 98.13531494140625, 155.46255493164062, 3.3225555419921875, 19.052490234375, 67.47882080078125, 47.20118713378906, 141.52122497558594, -62.772918701171875, -145.5242919921875, 22.663986206054688, 41.79225158691406, 190.01275634765625, 68.55796813964844, 157.12939453125, -1.1534996032714844, -15.627960205078125, 138.33920288085938, 11.102325439453125, -29.008956909179688, 56.5234375, 173.01390075683594, 32.075904846191406, 128.09136962890625, 142.73611450195312, 7.749542236328125, -9.71051025390625, 55.14533233642578, 141.10484313964844, 37.405029296875, -92.283447265625, -60.981163024902344, 131.5135498046875, -2.7252235412597656, 186.31553649902344, 143.18154907226562, 1.0888175964355469, -12.508544921875, 214.692138671875, 146.82809448242188, 93.03618621826172, 121.18994140625, 65.1849365234375, 8.60662841796875, 75.45870971679688, 175.05484008789062, 104.75981140136719, 7.740936279296875, 2.50469970703125, 134.70034790039062, 155.51504516601562, 133.88034057617188, -26.6429443359375, -107.54671478271484, 102.49943542480469, 101.08624267578125, -12.052536010742188, 11.999664306640625, -5.6911773681640625, 62.17010498046875, 7.993682861328125, 19.55462646484375, 208.1013946533203, 63.593292236328125, 54.87347412109375, 157.2034912109375, 96.697265625, 191.14801025390625, 199.71240234375, -31.414352416992188, 121.9794692993164, 12.136558532714844, 37.0888671875, 198.07550048828125, 11.24020767211914, -32.72479248046875, -85.82089233398438, 152.44033813476562, 25.393489837646484, 0.39037322998046875, 194.056396484375, 9.333274841308594, 115.20883178710938, 135.52230834960938, 23.17675018310547, -91.79605102539062, 7.0256195068359375, 164.23919677734375, 185.031005859375, 11.13101577758789, 7.120574951171875, 154.40414428710938, 126.10492706298828, -90.88186645507812, 175.80758666992188, 1.75067138671875, 167.693603515625, 22.044464111328125, 50.382598876953125, 12.798107147216797, 155.40545654296875, 63.465423583984375, 38.340301513671875, 42.85076904296875, 1.0961151123046875, 172.187255859375, -2.874156951904297, -6.0861663818359375, -13.649993896484375, -22.713241577148438, 4.602510452270508, -28.475692749023438, 3.6465015411376953, -65.54583740234375, 182.27020263671875, 211.264892578125, -9.20965576171875, -73.64513397216797, 169.83245849609375, 97.05828857421875, 85.93405151367188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000405.npy"}
{"epoch": 0.8481675392670157, "step": 406, "batch_size": 128, "mean": 47.19438171386719, "std": 90.7179946899414, "min": -186.20376586914062, "p10": -49.30468902587891, "median": 25.77743148803711, "p90": 158.63853759765624, "max": 208.48388671875, "pos_frac": 0.6953125, "sample": [141.92103576660156, -28.736412048339844, -35.4326171875, 19.33843994140625, -146.483642578125, 195.11199951171875, 58.46662902832031, 149.42361450195312, 112.34620666503906, 159.73316955566406, -18.06566619873047, 0.5791492462158203, -22.188232421875, 173.99053955078125, 59.792816162109375, -26.70721435546875, 136.36087036132812, -4.082469940185547, 99.74513244628906, 24.243057250976562, -125.92674255371094, -48.14457702636719, 15.366851806640625, 119.49862670898438, 147.36767578125, 1.639251708984375, 107.05073547363281, 192.45654296875, 27.51165771484375, 101.90818786621094, 58.47528076171875, 158.50770568847656, -151.64266967773438, -1.55438232421875, 23.529388427734375, -129.42933654785156, -35.28271484375, -5.852043151855469, -13.971794128417969, 150.73995971679688, -44.73486328125, -49.82493591308594, -125.967529296875, -3.84576416015625, 32.06067657470703, 164.50381469726562, 19.086837768554688, 141.65155029296875, 134.43423461914062, 158.8712158203125, -1.4025650024414062, 21.400115966796875, 154.34442138671875, 106.89981842041016, -186.20376586914062, -6.135772705078125, 137.5159912109375, 147.37960815429688, 108.30803680419922, 18.40496826171875, 2.50360107421875, 152.99835205078125, -119.6077880859375, 172.89691162109375, 22.224639892578125, 28.0079345703125, 7.791717529296875, 11.671539306640625, -82.10015869140625, 149.27630615234375, 208.48388671875, 151.05352783203125, 11.092453002929688, 35.92730712890625, 172.49942016601562, 39.614471435546875, 4.2595062255859375, 145.50796508789062, 15.048717498779297, 185.44805908203125, 152.69778442382812, -17.520408630371094, -145.5126953125, 11.74782943725586, 157.47523498535156, 133.87567138671875, 113.78048706054688, -9.541549682617188, 25.6749267578125, 109.88729858398438, 60.07673645019531, -15.355911254882812, 117.79719543457031, 133.8840789794922, -90.0484848022461, 207.126220703125, 122.60218811035156, 2.5488662719726562, 127.60015869140625, 40.892578125, 45.38737487792969, 142.82965087890625, 36.8680419921875, 150.40158081054688, -25.22711181640625, 50.88621520996094, -117.49160766601562, 183.00576782226562, 3.0311279296875, 35.480072021484375, -30.773590087890625, 86.94488525390625, -45.761199951171875, 122.17376708984375, 201.90603637695312, -1.1403884887695312, 2.4238739013671875, 25.87993621826172, -4.97930908203125, -17.680034637451172, -74.67913818359375, 20.922119140625, 6.452484130859375, -8.857879638671875, 158.538818359375, -49.08172607421875, 3.1351547241210938, 17.649005889892578], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000406.npy"}
{"epoch": 0.8502617801047121, "step": 407, "batch_size": 128, "mean": 56.314788818359375, "std": 86.86022186279297, "min": -140.29202270507812, "p10": -36.683718872070315, "median": 33.17655944824219, "p90": 167.56112518310547, "max": 210.75869750976562, "pos_frac": 0.703125, "sample": [-10.676719665527344, 27.38226318359375, 100.15447998046875, -14.191619873046875, 151.66424560546875, 14.757984161376953, 162.8245849609375, 106.50637817382812, -36.14093017578125, 21.73918342590332, -1.6864700317382812, -57.076812744140625, -131.10891723632812, -3.399749755859375, -115.29759216308594, 133.85324096679688, -0.5469818115234375, -0.3198394775390625, 14.50439453125, -27.1162109375, 137.62310791015625, -2.64385986328125, -9.627197265625, 126.90435028076172, 167.04342651367188, -73.28828430175781, 44.805335998535156, 97.58436584472656, -10.680305480957031, -13.16754150390625, -140.29202270507812, 104.11460876464844, -37.88751220703125, 35.223419189453125, 106.11372375488281, 172.087890625, 159.32583618164062, 173.17623901367188, -29.471267700195312, -2.7659149169921875, 64.79171752929688, 25.12677001953125, 31.12969970703125, 3.4058761596679688, -36.91864013671875, 191.20733642578125, 71.66666412353516, 119.22865295410156, 186.0767822265625, 9.3707275390625, 113.36677551269531, 144.645751953125, 7.202850341796875, 188.06240844726562, 10.01763916015625, 148.79855346679688, -68.61019134521484, -0.8999557495117188, 26.947113037109375, 104.28860473632812, 136.3005828857422, 25.415557861328125, 10.338897705078125, 21.730552673339844, 160.90704345703125, 73.4847412109375, -100.4205322265625, -19.021316528320312, 129.03317260742188, 168.7690887451172, 18.750762939453125, -26.866867065429688, 139.75421142578125, 162.22518920898438, 188.1317138671875, 15.707645416259766, -2.267974853515625, 40.161949157714844, 0.0, 198.07266235351562, 159.52032470703125, 11.029167175292969, 131.61094665527344, -2.86456298828125, 164.13613891601562, -97.8023681640625, 185.32611083984375, 105.83145141601562, 16.12896728515625, 134.173828125, 161.87542724609375, 2.393463134765625, 17.65997314453125, 125.15554809570312, 108.33969116210938, 12.708961486816406, -36.583038330078125, 35.673179626464844, 38.151123046875, 126.14224243164062, 163.2919921875, -20.307723999023438, 196.41848754882812, -126.45782470703125, 78.43618774414062, -118.57177734375, 103.90866088867188, 77.72061157226562, 165.760009765625, 7.74951171875, 149.05624389648438, 169.87725830078125, 49.9908447265625, -128.42758178710938, 178.4793701171875, 7.33154296875, 37.48741149902344, -30.71439552307129, 10.00506591796875, 136.56585693359375, 145.9532470703125, 145.99774169921875, 21.27459716796875, 7.612970352172852, -5.8858184814453125, 210.75869750976562, 157.26016235351562, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000407.npy"}
{"epoch": 0.8523560209424084, "step": 408, "batch_size": 128, "mean": 57.503326416015625, "std": 84.24321746826172, "min": -191.06216430664062, "p10": -30.195040893554683, "median": 42.085205078125, "p90": 165.19392089843748, "max": 194.36245727539062, "pos_frac": 0.7265625, "sample": [-3.4085311889648438, 2.597991943359375, -41.07746124267578, 63.695343017578125, 13.442733764648438, 157.21780395507812, 46.72962951660156, 185.28619384765625, 168.58856201171875, 147.83489990234375, -28.229644775390625, 17.79290771484375, 86.25215148925781, 118.34342956542969, -41.841064453125, 153.52291870117188, 20.64215087890625, -11.251953125, -4.9887542724609375, 100.10379791259766, 0.934234619140625, 20.070281982421875, 8.139602661132812, -11.32586669921875, 179.7576904296875, 147.25418090820312, 182.9752197265625, 170.3896484375, -13.390106201171875, -1.73834228515625, -22.965194702148438, 151.60830688476562, 136.03927612304688, 133.45098876953125, -34.8170166015625, 114.6656494140625, 191.89425659179688, 125.19023132324219, 155.43252563476562, 42.44989013671875, -32.496551513671875, -183.95294189453125, -191.06216430664062, 12.7706298828125, 177.47625732421875, 143.17955017089844, 5.7609100341796875, 19.123756408691406, -27.377197265625, 156.74391174316406, 28.438873291015625, 118.144775390625, -8.785720825195312, -7.4261322021484375, 87.15937805175781, 41.01806640625, 137.0216064453125, 143.55886840820312, -29.20867919921875, 73.34664916992188, 159.09620666503906, 146.35072326660156, 193.52685546875, 3.5219078063964844, -16.9139404296875, 163.078857421875, -40.754486083984375, -17.339797973632812, 134.98455810546875, -114.22396850585938, 172.0103759765625, -11.199462890625, 133.91119384765625, 13.87255859375, 76.91557312011719, 160.59420776367188, -18.90594482421875, 85.73185729980469, -1.021820068359375, 26.038925170898438, 159.6064453125, 78.75025939941406, 17.565567016601562, 40.40228271484375, -24.69677734375, 3.76934814453125, 142.26742553710938, 147.70278930664062, 120.884033203125, -57.99546813964844, 29.368988037109375, 12.345268249511719, 51.282623291015625, 155.76724243164062, 159.48187255859375, -1.7911224365234375, -151.41497802734375, 64.76754760742188, 167.72808837890625, 41.72052001953125, -40.89830017089844, 158.05458068847656, 15.554595947265625, 58.400535583496094, 54.54002380371094, -5.8175048828125, 16.61402130126953, 9.422882080078125, 6.844879150390625, 48.943206787109375, 57.83655548095703, 13.592086791992188, 23.729904174804688, 0.0, 11.44866943359375, 194.36245727539062, 65.53550720214844, -2.118406295776367, 118.52362060546875, -40.474090576171875, -113.17243194580078, 27.60125732421875, 164.10784912109375, 46.407379150390625, 185.19992065429688, 85.38507080078125, 124.14262390136719, 179.1707763671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000408.npy"}
{"epoch": 0.8544502617801047, "step": 409, "batch_size": 128, "mean": 59.342437744140625, "std": 92.57434844970703, "min": -160.5845947265625, "p10": -43.98504905700683, "median": 37.91862487792969, "p90": 174.4914123535156, "max": 232.7021484375, "pos_frac": 0.7421875, "sample": [154.68157958984375, 134.42333984375, 149.39434814453125, 58.124267578125, 184.514404296875, -41.758544921875, 123.54339599609375, 180.89971923828125, 3.55120849609375, 41.158782958984375, 25.443016052246094, 8.848236083984375, 22.336959838867188, 3.43646240234375, 28.87762451171875, 197.69833374023438, 25.282936096191406, 22.3365478515625, 178.0145263671875, -144.40194702148438, -11.088424682617188, -59.10016632080078, -1.451979637145996, 167.98219299316406, 145.88735961914062, -103.2626953125, 14.867218017578125, -110.49996948242188, 165.5050506591797, -157.44891357421875, 170.0352783203125, 147.75735473632812, -87.48697662353516, 44.24957275390625, -6.64703369140625, 189.96514892578125, 22.94757080078125, 40.32513427734375, 189.3043212890625, 172.98150634765625, 129.8721923828125, 165.44229125976562, -112.63665771484375, 29.003158569335938, 2.355926513671875, 32.531982421875, 83.20858764648438, 155.88241577148438, 15.97784423828125, 194.96563720703125, 156.07757568359375, 140.79702758789062, 122.7554931640625, 145.56246948242188, 232.7021484375, -42.499237060546875, -93.68924713134766, 72.27511596679688, 21.663818359375, -7.723175048828125, 210.53167724609375, 122.38064575195312, 19.404403686523438, 86.13619995117188, 152.36517333984375, 52.972991943359375, 22.828521728515625, -38.022003173828125, -4.167518615722656, -36.20606994628906, 154.2091064453125, -28.030715942382812, -124.58340454101562, 172.10865783691406, 1.1574935913085938, 122.56716918945312, 183.92398071289062, 215.29937744140625, 56.759864807128906, 26.30377197265625, 6.642658233642578, -43.249698638916016, 96.49105834960938, 0.0, 110.06765747070312, 221.6754150390625, 35.512115478515625, 172.07037353515625, 152.22317504882812, -137.81951904296875, -18.358627319335938, 155.59255981445312, 27.64691162109375, -10.961814880371094, 23.94140625, -1.267934799194336, 129.79583740234375, 157.55044555664062, 143.10838317871094, 144.30091857910156, 31.748130798339844, -0.05794525146484375, 110.814697265625, 81.613525390625, 43.541839599609375, -10.560264587402344, 21.42840576171875, -45.70086669921875, -4.0052337646484375, -11.218833923339844, 44.39080810546875, 3.992940902709961, -160.5845947265625, 32.42491149902344, 144.81895446777344, 93.45175170898438, 5.998378753662109, 136.611572265625, 30.805084228515625, -41.787689208984375, 151.474609375, 11.597930908203125, 7.890357971191406, 128.9393310546875, -69.02587890625, 131.76992797851562, 193.55484008789062, 63.252899169921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000409.npy"}
{"epoch": 0.856544502617801, "step": 410, "batch_size": 128, "mean": 51.925167083740234, "std": 89.16024017333984, "min": -151.64950561523438, "p10": -41.53123931884765, "median": 42.831443786621094, "p90": 172.23135681152343, "max": 233.7294921875, "pos_frac": 0.6953125, "sample": [72.3250732421875, -42.94767761230469, 14.851966857910156, -38.4949951171875, -18.05632781982422, 150.87759399414062, 4.877599716186523, 29.120208740234375, -32.185577392578125, 130.7237091064453, 213.70498657226562, -23.647674560546875, 39.56562042236328, -77.15948486328125, 45.13581085205078, -149.938232421875, -16.97723388671875, 41.515655517578125, 162.0611572265625, 113.701904296875, 172.19769287109375, -116.95028686523438, 65.08129119873047, 12.813270568847656, 155.35128784179688, -7.2044525146484375, 122.27276611328125, 114.51348876953125, 108.89338684082031, 139.67233276367188, 19.198486328125, 102.96267700195312, 63.1572265625, 149.21798706054688, 0.07025718688964844, 124.54544067382812, 17.338119506835938, 63.828765869140625, 34.093902587890625, 168.5986328125, 162.72711181640625, 46.68865966796875, -8.413070678710938, 158.86724853515625, -140.21615600585938, 133.57608032226562, -147.83517456054688, -9.9473876953125, -19.744094848632812, -22.9732666015625, 93.74755859375, 133.70132446289062, 61.21147155761719, -51.744293212890625, -25.07628631591797, 45.337646484375, -5.804939270019531, 152.19223022460938, 87.34915161132812, 46.921051025390625, -47.15924072265625, 189.232177734375, 91.0838623046875, 182.42462158203125, -2.9251842498779297, 10.270217895507812, -28.817718505859375, -2.0540771484375, 129.80340576171875, 44.14723205566406, 130.0640869140625, 145.5029296875, 34.94847106933594, -20.8609619140625, -68.22146606445312, 0.0, 37.168251037597656, 26.082061767578125, 195.63818359375, 170.5738525390625, 197.27490234375, 7.705474853515625, 4.4951171875, -50.89447021484375, -35.013816833496094, 137.269287109375, 20.618370056152344, 20.1522216796875, -28.565658569335938, 28.71080780029297, -138.85716247558594, 69.30596923828125, -13.850799560546875, 90.53654479980469, -40.9241943359375, 198.45315551757812, -151.64950561523438, 172.30990600585938, 6.539344787597656, 40.67478942871094, 56.67938995361328, -22.531410217285156, 135.93780517578125, -150.33062744140625, 55.369476318359375, 75.58515930175781, 102.02424621582031, 141.85800170898438, 216.9228515625, 117.8089599609375, 174.01321411132812, -21.89599609375, 138.55862426757812, 96.74491882324219, 1.635589599609375, 57.20001220703125, 194.8818359375, -32.852935791015625, 44.21870422363281, 4.7085723876953125, 4.253131866455078, 178.984619140625, 203.77130126953125, 21.4490966796875, 85.56785583496094, -35.132415771484375, 233.7294921875, -3.2000999450683594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000410.npy"}
{"epoch": 0.8586387434554974, "step": 411, "batch_size": 128, "mean": 62.6776237487793, "std": 85.62287139892578, "min": -200.288818359375, "p10": -20.912091064453122, "median": 40.73374557495117, "p90": 166.09908752441405, "max": 292.1177978515625, "pos_frac": 0.765625, "sample": [124.53529357910156, 165.06646728515625, 34.27978515625, 143.7149658203125, 24.04779052734375, -2.7888336181640625, -1.2754859924316406, 49.50401306152344, 40.32366180419922, 136.35858154296875, 152.237060546875, 157.41348266601562, -2.9257736206054688, -18.634979248046875, -125.3363037109375, -133.73919677734375, 200.16754150390625, -19.40234375, 131.29833984375, 121.37876892089844, 41.143829345703125, 59.501617431640625, 144.8212127685547, 46.26806640625, 132.496337890625, 32.594970703125, 28.117340087890625, 292.1177978515625, 133.71652221679688, -17.8052978515625, 173.855224609375, 41.81129455566406, 38.336097717285156, 21.344100952148438, 132.03106689453125, 31.705047607421875, 174.54974365234375, 150.286865234375, 66.86441040039062, 26.315021514892578, 6.839447021484375, 30.481658935546875, -4.61004638671875, 71.27253723144531, 5.2285003662109375, -85.36363983154297, 1.12713623046875, 0.7003746032714844, -19.889007568359375, 101.03741455078125, 141.3275146484375, 27.673065185546875, -11.582733154296875, 3.9227523803710938, 15.296375274658203, 164.86700439453125, -30.764137268066406, 99.39077758789062, 38.68621826171875, 95.8976058959961, -0.3295764923095703, 34.735992431640625, -10.309417724609375, -81.32537841796875, 139.43441772460938, 139.99716186523438, 118.32754516601562, 165.30996704101562, 24.403762817382812, 187.852783203125, 5.203554153442383, 138.4871368408203, 30.579833984375, 167.94036865234375, 20.08197784423828, 186.19500732421875, 22.845550537109375, 3.466461181640625, 154.88519287109375, 139.86172485351562, 33.185508728027344, -200.288818359375, 172.76406860351562, 14.498653411865234, 9.906814575195312, -132.70169067382812, 43.54344177246094, -2.572071075439453, 149.1894989013672, 197.58969116210938, 139.04721069335938, -92.75161743164062, 31.382232666015625, 32.863433837890625, 15.509780883789062, 91.6004638671875, 112.79109191894531, 203.24334716796875, 131.86953735351562, -48.49116516113281, 155.78848266601562, -23.299285888671875, -74.15927124023438, -13.23638916015625, 9.504648208618164, 36.44525146484375, 132.79693603515625, 45.10345458984375, 34.37370300292969, 127.9397964477539, -31.08517837524414, 104.80535888671875, 154.21568298339844, -7.865287780761719, 65.73014831542969, -7.5705413818359375, 89.83467864990234, 208.0501708984375, 143.1748046875, 160.12168884277344, 146.72686767578125, 193.6753387451172, 126.45797729492188, 182.66378784179688, 81.5496597290039, -6.846038818359375, -72.20075988769531, -7.61627197265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000411.npy"}
{"epoch": 0.8607329842931937, "step": 412, "batch_size": 128, "mean": 50.285247802734375, "std": 91.56178283691406, "min": -185.19976806640625, "p10": -48.801788330078125, "median": 35.79646301269531, "p90": 165.916943359375, "max": 241.27224731445312, "pos_frac": 0.6640625, "sample": [-129.1075439453125, 162.8245849609375, 5.757423400878906, 176.571044921875, 14.208984375, 97.6456298828125, 37.1265869140625, 112.5784912109375, -33.404449462890625, -18.735321044921875, 34.669189453125, 32.5302734375, -3.4457168579101562, -15.024576187133789, 21.802642822265625, -185.19976806640625, 0.215240478515625, -26.359771728515625, -111.65376281738281, 28.522705078125, -2.2530059814453125, 2.0587081909179688, 19.18499755859375, 83.48760986328125, 162.31663513183594, 157.12469482421875, 172.43600463867188, 32.31293487548828, -18.383285522460938, 34.29138946533203, 120.58346557617188, 124.69244384765625, -20.699737548828125, 162.69869995117188, 183.93661499023438, 155.69338989257812, 164.51693725585938, 17.318771362304688, 149.032958984375, 70.9154052734375, 131.43263244628906, -25.630126953125, 168.00494384765625, -128.96450805664062, 84.63972473144531, -19.80322265625, -3.3685760498046875, -26.543670654296875, -148.64300537109375, 49.264923095703125, 124.212646484375, 176.16954040527344, 196.76412963867188, -20.6278076171875, 169.3126220703125, 41.27520751953125, -10.690017700195312, 85.73158264160156, -85.14336395263672, -34.188873291015625, -79.93710327148438, -7.795560836791992, 204.36257934570312, 68.25634765625, -2.0825653076171875, 165.83932495117188, 129.67672729492188, 241.27224731445312, 46.3221435546875, 62.6263427734375, -103.11822509765625, 61.232391357421875, -48.81866455078125, -20.20123291015625, 44.771453857421875, 34.7830810546875, 188.75001525878906, 163.6575927734375, 40.3760986328125, -163.43609619140625, 36.809844970703125, 96.24571228027344, 156.83103942871094, -1.674468994140625, 166.09805297851562, 3.213775634765625, 158.62081909179688, 154.4305419921875, 179.9132080078125, -90.32499694824219, 11.205429077148438, 165.42129516601562, 147.3910675048828, -111.07730102539062, -6.350616455078125, 46.59858703613281, -14.556167602539062, 126.66680908203125, -11.82373046875, 128.69337463378906, -13.412139892578125, -0.14287757873535156, 198.76504516601562, 43.04400634765625, 146.14068603515625, 6.79023551940918, -48.7945556640625, 105.2204818725586, 15.593719482421875, 115.54322814941406, 155.282470703125, 26.146011352539062, 0.0, 52.14253234863281, -12.99755859375, 150.70965576171875, 3.1364593505859375, 103.24044036865234, -129.2607421875, 5.258953094482422, 25.659515380859375, 138.82192993164062, -41.76702880859375, 54.35931396484375, -10.742935180664062, -11.453826904296875, 133.11544799804688, 161.25161743164062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000412.npy"}
{"epoch": 0.86282722513089, "step": 413, "batch_size": 128, "mean": 47.20831298828125, "std": 81.0724105834961, "min": -167.78411865234375, "p10": -44.418617248535156, "median": 23.29983139038086, "p90": 161.6335464477539, "max": 235.73263549804688, "pos_frac": 0.7265625, "sample": [21.389020919799805, -19.698501586914062, -1.0945281982421875, 1.9265213012695312, 7.625885009765625, 146.16796875, 53.32977294921875, 20.3016357421875, -167.78411865234375, -12.34014892578125, 174.67266845703125, 39.84320068359375, 177.26788330078125, -48.08953857421875, 31.962066650390625, 15.637771606445312, 10.54791259765625, 174.1390380859375, 60.768218994140625, -79.25129699707031, 49.83367919921875, 14.676403045654297, 93.10224914550781, 15.599365234375, 28.406478881835938, -7.044425964355469, 8.910713195800781, 187.477294921875, -0.15686988830566406, 105.89978790283203, 152.76834106445312, -57.658203125, -37.988197326660156, 153.44151306152344, -39.085205078125, 151.2225799560547, 161.3467254638672, 143.7894744873047, 148.65447998046875, 10.206230163574219, 114.32673645019531, -3.9746475219726562, -94.57818603515625, 3.3978271484375, 61.68853759765625, -31.66766357421875, -96.7061996459961, 54.28462219238281, 175.62966918945312, 84.5601806640625, -43.93927001953125, 3.158660888671875, -18.391998291015625, 93.09283447265625, 23.45941162109375, 124.5763168334961, 162.30279541015625, -24.629730224609375, 41.22868347167969, 89.18584442138672, 167.92572021484375, 25.37261962890625, 14.3294677734375, 15.319236755371094, 34.4571533203125, 200.60882568359375, -101.30296325683594, 55.95562744140625, 153.74395751953125, -52.56951904296875, 146.5010986328125, 42.7342529296875, -21.04498291015625, 235.73263549804688, 13.307182312011719, -5.1688232421875, -3.051483154296875, 23.834518432617188, -18.887115478515625, 103.84300231933594, 7.1236724853515625, 146.7926025390625, -48.4349365234375, 137.12332153320312, 139.125244140625, -22.569900512695312, 12.5074462890625, 77.9615478515625, 4.658702850341797, -49.34423828125, 156.37942504882812, 38.95429992675781, -45.53709411621094, 129.14422607421875, 18.4962158203125, -19.409496307373047, 28.580581665039062, 15.471450805664062, 149.76358032226562, 11.725616455078125, 9.761432647705078, -3.8264389038085938, 5.8200531005859375, 124.34396362304688, 110.24505615234375, 72.1094970703125, 3.59637451171875, 15.794952392578125, 111.125732421875, 83.27983093261719, 23.14025115966797, 28.196624755859375, 119.62194061279297, 154.38540649414062, -10.602088928222656, -111.47982788085938, 47.59722900390625, 163.90853881835938, -137.38436889648438, 204.2186279296875, 16.3121337890625, 183.10037231445312, 4.8997344970703125, -0.20841217041015625, 176.76263427734375, 133.13140869140625, -32.98316955566406, 19.91534423828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000413.npy"}
{"epoch": 0.8649214659685864, "step": 414, "batch_size": 128, "mean": 55.56306457519531, "std": 94.12638854980469, "min": -201.99542236328125, "p10": -41.7703369140625, "median": 35.453834533691406, "p90": 176.7974884033203, "max": 276.96636962890625, "pos_frac": 0.703125, "sample": [120.04287719726562, -134.5511932373047, 98.86128234863281, 150.29855346679688, 1.3277816772460938, 141.94192504882812, 67.707275390625, 60.1019287109375, 11.420822143554688, -114.0560302734375, -7.6136322021484375, 7.1411895751953125, 176.32803344726562, 23.28289794921875, 6.1950225830078125, -2.1427974700927734, -25.604949951171875, -113.04351806640625, 152.214599609375, -8.49167251586914, -201.99542236328125, -79.24462890625, 207.5831298828125, 8.1805419921875, 130.48614501953125, -141.8672637939453, -40.273193359375, -2.5369720458984375, 140.54824829101562, 141.41256713867188, -17.02566146850586, -25.951171875, 137.98208618164062, 153.87806701660156, -55.952484130859375, 172.0145721435547, 143.884033203125, -5.6145782470703125, 196.22122192382812, -25.783050537109375, 25.83154296875, 160.294189453125, 103.05035400390625, -10.015228271484375, 126.01832580566406, 69.39120483398438, 6.6903076171875, 8.078506469726562, 39.1651611328125, 198.38519287109375, -5.9112396240234375, -14.125053405761719, -3.2678985595703125, 202.14657592773438, 144.17710876464844, -7.967803955078125, 149.46243286132812, 174.7128448486328, 175.69296264648438, -93.38638305664062, 126.37579345703125, 2.8500499725341797, -0.1277008056640625, 150.15957641601562, 9.78839111328125, 43.32379150390625, 135.70034790039062, 113.93360900878906, 56.966796875, -0.64959716796875, 148.02679443359375, -8.1748046875, 21.56866455078125, 191.1893310546875, 143.22802734375, 163.35638427734375, 149.68130493164062, 1.2701416015625, 37.646209716796875, 124.87938690185547, 6.2684783935546875, -40.4251708984375, -129.71630859375, 198.65493774414062, 276.96636962890625, 161.51461791992188, 177.89288330078125, 102.29102325439453, 5.670936584472656, 138.01535034179688, -187.73944091796875, 10.804229736328125, 33.26145935058594, 148.13064575195312, 12.227977752685547, 180.42816162109375, -30.94415283203125, -2.2958984375, -7.5243072509765625, 11.642105102539062, -116.47061920166016, 77.3778076171875, 222.90921020507812, 38.39329528808594, 9.074653625488281, 12.830558776855469, 150.48538208007812, 110.11967468261719, 96.56440734863281, 16.663543701171875, -44.9090576171875, 75.50534057617188, 196.45703125, 58.955841064453125, -4.015237808227539, 11.590423583984375, 119.68130493164062, 183.01962280273438, 24.006851196289062, 38.03279113769531, 11.705703735351562, 0.094512939453125, -54.7525634765625, -8.84222412109375, 78.660888671875, 62.880523681640625, 180.66610717773438, -6.4595489501953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000414.npy"}
{"epoch": 0.8670157068062827, "step": 415, "batch_size": 128, "mean": 65.45309448242188, "std": 88.37609100341797, "min": -172.2530517578125, "p10": -27.726220703124998, "median": 55.82218933105469, "p90": 174.4617492675781, "max": 242.1009521484375, "pos_frac": 0.734375, "sample": [18.146223068237305, 132.68125915527344, 16.921531677246094, 34.828582763671875, -37.86836242675781, -5.7359771728515625, 146.04251098632812, -1.058074951171875, 139.21307373046875, 10.022735595703125, 58.69062042236328, -1.7974433898925781, 173.09185791015625, 103.40934753417969, -26.471343994140625, -8.615104675292969, 171.02447509765625, 4.700927734375, 143.63262939453125, 42.92735290527344, 148.73931884765625, 20.1197509765625, 183.732666015625, 3.3109130859375, 145.0452423095703, 37.197113037109375, 84.01728820800781, 161.02947998046875, 161.18997192382812, 116.86880493164062, -132.4167938232422, 193.2086181640625, 180.40951538085938, 25.1611328125, -137.54376220703125, 22.4532470703125, 56.99981689453125, 174.0836181640625, -8.856491088867188, 98.46934509277344, 116.18778228759766, 18.509246826171875, -0.42449951171875, 17.831451416015625, 61.60163879394531, -23.6383056640625, 158.184814453125, 43.338623046875, 175.34405517578125, 138.1912078857422, -11.81414794921875, 29.244720458984375, -101.7227783203125, 64.451904296875, 149.6626739501953, 167.38238525390625, -52.800750732421875, 167.48040771484375, 35.441925048828125, -28.013885498046875, 54.644561767578125, 153.04859924316406, -111.30902099609375, 127.95132446289062, 242.1009521484375, -21.70660400390625, 24.39215087890625, 58.24005889892578, -33.845703125, -0.6835613250732422, -26.861679077148438, 186.9791259765625, 140.0426025390625, 30.077682495117188, -0.1563262939453125, -7.58660888671875, 156.581298828125, 195.00009155273438, 118.75872802734375, -42.29779052734375, 80.12875366210938, 8.23907470703125, 116.82659912109375, -60.28662109375, 172.4271240234375, -23.970916748046875, -119.3988265991211, 72.23450469970703, 104.78076171875, -23.988006591796875, 2.5438232421875, 0.7589111328125, 42.765350341796875, 153.951416015625, 104.4217529296875, -27.602935791015625, 58.082275390625, 114.44583129882812, 156.77130126953125, -172.2530517578125, 196.57550048828125, 37.722442626953125, 158.94036865234375, 18.074356079101562, 26.989410400390625, 180.71133422851562, 86.21432495117188, -21.093978881835938, 162.38031005859375, 136.55104064941406, 148.49481201171875, 52.78217315673828, 148.23721313476562, 125.58627319335938, -110.53495788574219, 211.3936767578125, 22.940750122070312, 188.48057556152344, 186.8955078125, -6.121105194091797, 4.46875, 21.269195556640625, 137.383544921875, -4.1319732666015625, 145.42999267578125, 137.3638153076172, -1.9643325805664062, 211.26641845703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000415.npy"}
{"epoch": 0.8691099476439791, "step": 416, "batch_size": 128, "mean": 62.33583068847656, "std": 88.72042846679688, "min": -188.14898681640625, "p10": -28.31707305908203, "median": 43.10325622558594, "p90": 171.54027252197264, "max": 224.602783203125, "pos_frac": 0.7578125, "sample": [163.48873901367188, 202.13369750976562, 175.15188598632812, 27.491455078125, 3.5174407958984375, 2.8660125732421875, -29.09326171875, 14.7899169921875, -21.299026489257812, 123.75975036621094, -37.43829345703125, 224.602783203125, 46.35453796386719, 30.9217529296875, -1.60687255859375, 17.269264221191406, 12.501068115234375, 184.2470703125, 207.326171875, 98.54994201660156, 130.01895141601562, 147.9359130859375, 42.48851013183594, 170.19505310058594, -15.365081787109375, 38.00019836425781, -29.370620727539062, 137.1766357421875, -88.76535034179688, 23.245132446289062, 125.04730224609375, 156.30484008789062, 145.4656982421875, -6.65777587890625, 9.71707534790039, 172.35922241210938, -138.2758026123047, 159.62680053710938, 25.91961669921875, 17.8406982421875, 29.228271484375, 35.876380920410156, -19.797714233398438, -188.14898681640625, -23.169063568115234, 172.26553344726562, 27.994552612304688, 125.48487854003906, 121.93719482421875, 172.73016357421875, 38.413330078125, 171.2294464111328, 12.50897216796875, 19.21483612060547, -65.8316650390625, 153.00570678710938, 167.19351196289062, 136.52723693847656, -5.3272857666015625, -12.281402587890625, 127.311767578125, 73.73516845703125, -135.9912109375, -19.793365478515625, 196.75936889648438, 142.39529418945312, 0.17035675048828125, 133.377685546875, -41.733154296875, 138.14280700683594, 1.3028564453125, -27.984420776367188, 43.71800231933594, 151.57977294921875, 202.60342407226562, 138.42225646972656, 90.60546875, 152.0185089111328, -14.66259765625, 1.1307754516601562, 89.9827880859375, 27.34967041015625, 160.412353515625, -27.114051818847656, -95.42608642578125, 110.28654479980469, 52.452117919921875, 87.09365844726562, -2.740325927734375, 23.513084411621094, 38.130897521972656, 132.83094787597656, 50.81334686279297, 109.88186645507812, -158.79736328125, 132.436767578125, 60.969696044921875, 0.627105712890625, 27.58929443359375, 163.3480682373047, -127.44869995117188, 138.95480346679688, 44.893798828125, 16.667938232421875, 56.6400146484375, -10.34564208984375, 186.99334716796875, 129.27642822265625, 205.23257446289062, -16.165481567382812, 3.8928375244140625, 199.3974609375, 46.631591796875, 21.07232666015625, -13.3052978515625, 155.49855041503906, 149.4639892578125, 114.9007568359375, 38.59497833251953, -6.601947784423828, 28.43170166015625, -85.09503173828125, 165.81204223632812, -13.457984924316406, 157.12594604492188, 17.235260009765625, 141.32354736328125, 159.12269592285156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000416.npy"}
{"epoch": 0.8712041884816754, "step": 417, "batch_size": 128, "mean": 54.21757125854492, "std": 90.99718475341797, "min": -166.98886108398438, "p10": -60.956655883789026, "median": 42.480464935302734, "p90": 169.63470458984375, "max": 220.3701171875, "pos_frac": 0.7109375, "sample": [-2.4135208129882812, 41.55088806152344, 32.40553283691406, 97.83638000488281, -29.747406005859375, 83.51815795898438, 36.51286315917969, -45.660491943359375, 180.82601928710938, 182.15420532226562, 6.9923095703125, 39.20654296875, 4.466436386108398, 136.08724975585938, -151.0333251953125, 7.210868835449219, 9.151016235351562, -7.5203704833984375, -11.138916015625, 151.53396606445312, 164.18560791015625, -28.20257568359375, 181.09097290039062, -83.4854736328125, -101.87589263916016, -144.54837036132812, 106.50128173828125, 4.782310485839844, -37.33445739746094, 194.62161254882812, 67.53729248046875, -8.845611572265625, 4.4654541015625, 101.14192962646484, 112.79133605957031, 169.85089111328125, 22.7716064453125, -114.4605712890625, 138.59129333496094, 134.90609741210938, 157.65188598632812, 26.561767578125, 110.88618469238281, -3.54803466796875, 135.23486328125, 132.28707885742188, 16.335159301757812, 72.9052734375, 123.56529235839844, 82.21037292480469, 26.291048049926758, -0.6401939392089844, -15.790603637695312, -101.14434814453125, -51.46575927734375, 97.7342529296875, -4.327880859375, 154.44802856445312, 77.79336547851562, -14.0162353515625, 35.80694580078125, 14.005218505859375, 3.1680736541748047, 14.0284423828125, 168.88900756835938, 1.5520782470703125, 160.35250854492188, -27.4300537109375, -166.98886108398438, 3.0562057495117188, -25.957244873046875, 176.533203125, 137.54150390625, 66.06210327148438, 51.09881591796875, -84.93292236328125, 111.44340515136719, 160.753662109375, 0.0, -2.087860107421875, 118.78141784667969, 0.0, 44.54254150390625, 144.48306274414062, 113.97975158691406, 132.5550994873047, 61.28778076171875, 4.6167449951171875, 130.04153442382812, -19.9722900390625, -87.54229736328125, 169.53607177734375, 152.6168212890625, 153.0679931640625, 169.54205322265625, 46.947021484375, -152.94435119628906, 102.57713317871094, 213.9638671875, 18.176483154296875, 165.94415283203125, -90.02159881591797, -83.10208129882812, 72.74234008789062, 155.87362670898438, 173.1789093017578, 185.42428588867188, -43.617950439453125, -121.44509887695312, 18.24127197265625, 13.9395751953125, -11.947341918945312, 8.974185943603516, 160.40695190429688, 182.05392456054688, 162.13128662109375, -14.570404052734375, 48.71051025390625, 172.77780151367188, 6.964241027832031, 51.00872802734375, 27.716445922851562, 43.41004180908203, -4.1734771728515625, 138.50912475585938, 118.96408081054688, 220.3701171875, 198.8399658203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000417.npy"}
{"epoch": 0.8732984293193717, "step": 418, "batch_size": 128, "mean": 46.780662536621094, "std": 84.49758911132812, "min": -181.43997192382812, "p10": -49.07896690368652, "median": 32.7696533203125, "p90": 158.26808013916016, "max": 231.0614013671875, "pos_frac": 0.734375, "sample": [-36.017974853515625, 148.59768676757812, 32.507659912109375, 49.8751220703125, -86.77685546875, -138.19480895996094, 0.75531005859375, 180.87939453125, -46.751739501953125, -181.43997192382812, 37.960052490234375, 80.32476043701172, 27.760772705078125, -139.60992431640625, 55.431854248046875, 90.23538208007812, 127.5160903930664, -107.20684814453125, 77.81387329101562, 16.923675537109375, 163.37191772460938, 121.73202514648438, -15.642822265625, -0.3048076629638672, -0.096923828125, 9.045074462890625, -16.193344116210938, 133.31637573242188, 44.98779296875, 184.77435302734375, 142.42041015625, 38.5137939453125, 157.4115447998047, 127.80044555664062, 17.900115966796875, 79.693115234375, -24.497283935546875, 138.95648193359375, -48.588897705078125, 2.1368408203125, -0.97027587890625, 173.14031982421875, 1.306732177734375, 139.015380859375, 4.6194000244140625, 19.008575439453125, 1.6290645599365234, 36.809112548828125, 28.054595947265625, -17.860092163085938, 1.731170654296875, 161.76751708984375, 119.43595886230469, 28.967681884765625, 21.491790771484375, -8.844146728515625, 7.66192626953125, 25.40252685546875, 65.83695983886719, 109.89804077148438, -75.2657470703125, 133.579833984375, 160.26666259765625, 38.94874572753906, 33.031646728515625, 9.774192810058594, 132.91012573242188, -91.66442108154297, -64.04171752929688, 7.053596496582031, -53.13739013671875, 9.012733459472656, 136.13084411621094, 82.58773803710938, 86.37484741210938, 177.2877960205078, 115.7264404296875, -10.524368286132812, 83.35842895507812, -37.755035400390625, 39.27423095703125, 146.25125122070312, -9.775482177734375, 66.05657958984375, 110.16207885742188, 153.53411865234375, 147.1179656982422, -29.489974975585938, 0.45526123046875, -71.10870361328125, 14.07568359375, -27.041305541992188, -20.276824951171875, 130.75955200195312, 141.6233673095703, 15.454078674316406, 168.24417114257812, 115.90177917480469, 76.10411834716797, 152.96176147460938, -27.6434326171875, 0.932891845703125, 8.763921737670898, 29.262741088867188, -111.84710693359375, 39.260894775390625, 49.2469482421875, 141.54660034179688, -170.50677490234375, 164.40423583984375, 36.82219314575195, 16.12628173828125, 48.9683837890625, -3.468179702758789, 154.30682373046875, 52.54248809814453, 231.0614013671875, 163.30526733398438, 19.476730346679688, 217.75485229492188, 42.022491455078125, -21.174850463867188, -9.4248046875, 21.7755126953125, -50.22246170043945, 156.8729705810547, 214.21273803710938, 14.185928344726562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000418.npy"}
{"epoch": 0.875392670157068, "step": 419, "batch_size": 128, "mean": 63.7318229675293, "std": 97.34812927246094, "min": -222.5828399658203, "p10": -55.98374328613279, "median": 51.755828857421875, "p90": 174.7431610107422, "max": 245.2799072265625, "pos_frac": 0.7578125, "sample": [49.66864013671875, 16.270553588867188, 3.475830078125, 169.8056640625, 28.299461364746094, 24.946258544921875, 6.774990081787109, 19.317672729492188, 56.58447265625, -101.57275390625, -123.32247924804688, 80.20355224609375, -130.23272705078125, 161.88316345214844, 2.5349502563476562, 26.52581787109375, 204.7623291015625, 22.193450927734375, 138.5954132080078, 169.23739624023438, 171.17002868652344, 153.60382080078125, 175.141845703125, 3.179931640625, 180.478759765625, -1.499114990234375, 35.08784484863281, 8.231964111328125, -26.969635009765625, -156.6815185546875, 161.98837280273438, 151.63369750976562, 136.86207580566406, -3.5794448852539062, 52.07098388671875, -25.29693603515625, 86.02627563476562, -0.37227630615234375, 151.9295654296875, -149.86500549316406, 169.93673706054688, 3.7760162353515625, 163.53421020507812, 137.13314819335938, 245.2799072265625, 10.308143615722656, 208.09718322753906, -69.897705078125, 52.51873779296875, 188.44662475585938, 83.365478515625, 132.25970458984375, -19.62993621826172, 49.765037536621094, 40.7437744140625, 163.4257049560547, 202.56085205078125, 156.88372802734375, 7.01129150390625, -11.840118408203125, -105.62039184570312, 54.7220458984375, -5.663135528564453, 3.471813201904297, 223.484130859375, -134.28321838378906, 154.61038208007812, 176.8202362060547, 10.283584594726562, 39.858551025390625, -39.68773651123047, 146.21731567382812, 141.25863647460938, 21.820419311523438, 120.04486083984375, 0.0, -97.28439331054688, -51.378173828125, 14.815933227539062, 0.0, 23.472824096679688, 136.61875915527344, 142.2378387451172, 173.1225128173828, -10.202926635742188, 152.17791748046875, 18.371063232421875, 171.9462890625, 110.52859497070312, 19.931228637695312, 170.8441619873047, 51.440673828125, 175.26922607421875, 89.70281982421875, -1.14532470703125, 12.74212646484375, 130.67454528808594, 174.57229614257812, 149.86508178710938, 171.27685546875, 192.71194458007812, 13.5479736328125, 217.90875244140625, 17.48675537109375, 163.6924285888672, -40.12042236328125, 147.9771728515625, 150.2017364501953, 57.81658935546875, 141.62890625, -77.10986328125, -66.73007202148438, 176.80892944335938, 148.6033477783203, -222.5828399658203, 92.12753295898438, -147.1370849609375, -14.92901611328125, 147.31912231445312, -21.815521240234375, 104.76483154296875, 154.45950317382812, -12.303009033203125, 20.952789306640625, 8.549270629882812, 58.880760192871094, 138.45037841796875, 26.803680419921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000419.npy"}
{"epoch": 0.8774869109947644, "step": 420, "batch_size": 128, "mean": 60.017486572265625, "std": 88.06338500976562, "min": -170.03903198242188, "p10": -39.237255477905265, "median": 35.422760009765625, "p90": 168.3437042236328, "max": 222.42617797851562, "pos_frac": 0.765625, "sample": [-89.1287841796875, 18.65325927734375, 28.642776489257812, -169.3427734375, -24.657852172851562, -27.380226135253906, 27.636444091796875, 160.41720581054688, -43.830810546875, 100.2186279296875, 163.90234375, 4.7926025390625, -86.76986694335938, 139.43038940429688, 128.59901428222656, 10.939498901367188, 147.25994873046875, -115.05333709716797, 139.36892700195312, 115.13660430908203, 33.10162353515625, 57.262901306152344, -5.050300598144531, 36.7640380859375, 6.1294708251953125, 169.19552612304688, 11.160675048828125, 19.059669494628906, 127.96755981445312, 30.642791748046875, 4.4808197021484375, 204.4892578125, 41.86650085449219, 149.45855712890625, 14.831329345703125, -4.5093994140625, 132.36819458007812, 118.762939453125, 187.66751098632812, 11.490997314453125, 194.01177978515625, 58.2276611328125, 22.408912658691406, 149.07571411132812, 20.62396240234375, 33.5181884765625, 117.7100830078125, -4.423614501953125, -3.294708251953125, 101.09608459472656, -9.990966796875, 118.2779541015625, 143.7703094482422, 161.692138671875, 132.1142120361328, 135.46844482421875, 0.0, 31.034561157226562, -37.76165771484375, -136.37905883789062, 134.88250732421875, 132.83673095703125, 2.45849609375, 4.776519775390625, 14.702117919921875, 34.08148193359375, -3.926055908203125, 186.38388061523438, 206.1827392578125, 138.952880859375, 144.65975952148438, -94.89222717285156, 30.283935546875, 93.08738708496094, -37.258819580078125, 177.69125366210938, 122.63247680664062, 21.3148193359375, 165.09774780273438, 32.9747314453125, -24.877227783203125, 10.850669860839844, 222.42617797851562, -47.184661865234375, 150.1178741455078, 60.43424987792969, -8.58746337890625, -42.68031692504883, 164.1409912109375, 178.8718719482422, 57.02899169921875, 24.512054443359375, 170.48095703125, 167.9786376953125, 11.434783935546875, 13.892242431640625, 105.37574768066406, 57.48333740234375, 198.72726440429688, -8.703369140625, 187.09982299804688, 120.82598876953125, -155.3121795654297, -1.84075927734375, 29.537811279296875, 160.23037719726562, 96.17608642578125, 10.13909912109375, -68.22384643554688, -84.39501953125, 89.63031005859375, 101.87548828125, 120.34510803222656, 159.9884033203125, 156.14797973632812, 20.07177734375, 134.48919677734375, -170.03903198242188, 20.46044921875, 180.900634765625, 163.0860595703125, 117.21240234375, -35.454345703125, -8.409967422485352, 4.251617431640625, 103.43759155273438, 0.7851028442382812, 57.422607421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000420.npy"}
{"epoch": 0.8795811518324608, "step": 421, "batch_size": 128, "mean": 66.1441421508789, "std": 83.57243347167969, "min": -146.30496215820312, "p10": -25.814960479736328, "median": 65.47415161132812, "p90": 168.1654846191406, "max": 203.36920166015625, "pos_frac": 0.75, "sample": [202.61688232421875, 94.915771484375, 121.90389251708984, 43.82159423828125, 133.15008544921875, 105.396484375, 131.7952880859375, 11.86322021484375, -146.30496215820312, -10.711463928222656, 147.511474609375, -30.51483154296875, 10.207977294921875, 119.38009643554688, 30.012401580810547, 95.22272491455078, 38.46160888671875, -138.98837280273438, 29.89556884765625, 120.68287658691406, -44.740257263183594, -46.66624450683594, 67.44187927246094, 13.637763977050781, 141.42807006835938, 199.42120361328125, 159.69833374023438, 42.9080810546875, 10.306533813476562, 96.5500717163086, 0.0, -92.583740234375, 68.01382446289062, 11.221359252929688, 10.308425903320312, 81.77207946777344, 167.96588134765625, 166.51327514648438, 27.409469604492188, 156.51446533203125, 180.9678955078125, 152.58673095703125, 44.39501953125, 36.45582962036133, -5.974918365478516, 20.725791931152344, 144.67715454101562, -23.816253662109375, 194.10751342773438, 132.5997314453125, 133.8851776123047, -15.244743347167969, 128.50772094726562, 103.27789306640625, 42.893157958984375, 175.92059326171875, -115.9275131225586, 158.0794219970703, 9.92669677734375, 63.50642395019531, 37.42828369140625, -24.439109802246094, 199.79714965820312, 184.048095703125, 159.4053497314453, 81.66409301757812, 152.77491760253906, 160.56768798828125, 157.85411071777344, 148.73785400390625, 155.3102569580078, 179.2061767578125, 103.51638793945312, 158.7947998046875, 102.82620239257812, 78.56341552734375, -19.001434326171875, 39.8101806640625, 164.68399047851562, 12.034339904785156, 111.07034301757812, 5.5619964599609375, 41.710601806640625, 118.8201904296875, -4.3773193359375, 109.1800537109375, 35.32835388183594, -22.904312133789062, 147.12582397460938, -12.250755310058594, 129.92123413085938, 127.89509582519531, 195.30130004882812, 105.306884765625, -26.396770477294922, 55.598175048828125, 113.20098876953125, -117.28756713867188, -141.11163330078125, 98.334716796875, -26.22845458984375, -23.590469360351562, 35.5257568359375, -10.00238037109375, 20.772186279296875, 168.6312255859375, 7.03033447265625, -8.5184326171875, 120.9867935180664, -63.121673583984375, -28.287261962890625, 159.43310546875, 35.14556884765625, 135.87417602539062, -22.096923828125, 170.35284423828125, 108.46263122558594, 0.0, -22.9434814453125, -25.63774871826172, 203.36920166015625, 82.95135498046875, 27.15380859375, 0.0, 17.623382568359375, -4.27301025390625, 189.16436767578125, 6.07464599609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000421.npy"}
{"epoch": 0.881675392670157, "step": 422, "batch_size": 128, "mean": 45.837371826171875, "std": 91.79230499267578, "min": -184.61126708984375, "p10": -60.87030944824218, "median": 38.81547546386719, "p90": 168.65575408935547, "max": 208.145751953125, "pos_frac": 0.703125, "sample": [14.97955322265625, -18.142478942871094, 157.81312561035156, -41.93009948730469, 126.13766479492188, -16.845008850097656, 157.10687255859375, -148.25350952148438, 28.187164306640625, 1.0234375, 126.67056274414062, 5.03240966796875, 60.4588623046875, 156.80441284179688, 130.75396728515625, -184.61126708984375, 9.637451171875, 12.6219482421875, -21.131378173828125, 122.83991241455078, 117.3414535522461, 39.881988525390625, 106.88650512695312, 29.560379028320312, -46.509857177734375, 151.97467041015625, -140.44866943359375, 128.09910583496094, 131.1702117919922, 179.94903564453125, 40.953369140625, 4.988037109375, -17.249282836914062, 79.82254028320312, 8.1851806640625, 132.97622680664062, 6.400613784790039, -115.10137939453125, 181.59901428222656, 10.682891845703125, 8.9613037109375, 163.25741577148438, -59.6026611328125, 46.509429931640625, 59.0137939453125, -163.37225341796875, -13.0843505859375, 208.145751953125, 52.4224853515625, 155.5452117919922, 42.1583251953125, 179.58615112304688, 129.73684692382812, 16.564590454101562, 108.46554565429688, -2.547760009765625, 37.27906799316406, 14.825128555297852, 20.741943359375, 74.00079345703125, 137.50558471679688, -41.19829177856445, -31.47669219970703, -145.00140380859375, 185.45294189453125, 4.7308349609375, 118.08991241455078, 11.6239013671875, 2.3265380859375, 184.5010986328125, 88.33963012695312, -174.36541748046875, 162.0086212158203, -25.79168701171875, 63.0030517578125, -30.473072052001953, 144.51284790039062, -63.828155517578125, 173.86123657226562, 6.976890563964844, 181.79541015625, -2.8227920532226562, 134.39736938476562, 43.45001983642578, 180.98272705078125, 42.48931884765625, -100.86178588867188, 10.9072265625, 168.65042114257812, -0.3152637481689453, -14.656753540039062, -18.17333984375, 43.21026611328125, -23.871826171875, 144.26341247558594, 124.44789123535156, -41.04034423828125, -1.9587860107421875, 50.599639892578125, 54.36402893066406, 66.26438903808594, 14.922805786132812, 192.0582275390625, 134.63970947265625, -107.57620239257812, 142.18133544921875, 52.169403076171875, -130.41650390625, -29.983551025390625, 102.05975341796875, 0.0, -120.15493774414062, 168.66819763183594, 142.3699951171875, -110.52134704589844, 33.68585205078125, 84.35687255859375, 134.18527221679688, 37.74896240234375, -1.7006607055664062, -28.211532592773438, 111.76140594482422, 5.004493713378906, 169.5750274658203, 63.07618713378906, 17.629348754882812, -0.8473167419433594, 188.66085815429688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000422.npy"}
{"epoch": 0.8837696335078534, "step": 423, "batch_size": 128, "mean": 64.61135864257812, "std": 91.01239776611328, "min": -159.0968017578125, "p10": -36.88289184570312, "median": 53.6314697265625, "p90": 178.8287094116211, "max": 206.43814086914062, "pos_frac": 0.75, "sample": [0.895172119140625, 179.1288604736328, -15.907333374023438, 13.31201171875, 160.00155639648438, 145.84112548828125, 17.373210906982422, -13.923583984375, 2.4362831115722656, 146.23690795898438, 98.528076171875, 140.07586669921875, 29.780487060546875, 133.88723754882812, -121.578369140625, -11.688385009765625, 158.79931640625, -94.893798828125, 175.89202880859375, 14.151626586914062, 9.64950942993164, -32.00927734375, 23.5396728515625, 141.3419189453125, -140.846923828125, 120.68026733398438, 143.82354736328125, 14.650283813476562, 155.9329833984375, 182.18658447265625, 6.549774169921875, 22.9608154296875, 69.9249267578125, 55.10870361328125, 113.41497802734375, 134.94921875, 168.9168701171875, 2.749969482421875, -20.473358154296875, 70.41476440429688, 61.5499267578125, 171.2921142578125, -10.0919189453125, 155.78231811523438, -0.0125579833984375, 25.159507751464844, 9.31878662109375, 195.520263671875, -39.82818603515625, -36.46319580078125, 183.03921508789062, -38.87693786621094, -57.629638671875, 56.925262451171875, 172.3294677734375, -34.734619140625, -111.0869140625, -109.03458404541016, 178.7000732421875, 167.81146240234375, 31.509689331054688, -0.6341400146484375, 146.30596923828125, 12.6834716796875, -13.056442260742188, 121.84135437011719, 9.881187438964844, 165.04934692382812, 13.436561584472656, 34.543853759765625, 127.78252410888672, 155.78524780273438, 126.75076293945312, 137.67581176757812, 27.784027099609375, 62.524017333984375, -37.8621826171875, 165.36068725585938, 200.21490478515625, 52.15423583984375, 17.7322998046875, 175.39932250976562, 16.073287963867188, 174.20596313476562, 188.55438232421875, 57.71977233886719, 102.08026123046875, 23.571121215820312, -3.6069488525390625, 185.75143432617188, 141.50991821289062, 149.79209899902344, 181.80914306640625, -6.696258544921875, 199.00765991210938, 125.2567138671875, 3.092428207397461, -10.182708740234375, 180.59890747070312, -6.17291259765625, 3.8225555419921875, 135.92202758789062, 36.14752197265625, 184.39523315429688, 206.43814086914062, -36.1551513671875, -4.8961334228515625, 131.24705505371094, 180.0423126220703, -17.695831298828125, -127.35072326660156, 45.101776123046875, -93.52377319335938, 33.7677001953125, 17.852161407470703, 168.33261108398438, 57.87646484375, 144.48150634765625, 170.38787841796875, 55.4422607421875, 130.55328369140625, 142.05059814453125, 159.31857299804688, -145.23385620117188, 30.453948974609375, 46.64533996582031, -2.7834911346435547, -159.0968017578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000423.npy"}
{"epoch": 0.8858638743455497, "step": 424, "batch_size": 128, "mean": 62.7999382019043, "std": 80.6575698852539, "min": -161.88540649414062, "p10": -30.059655761718744, "median": 56.81661605834961, "p90": 165.02873229980466, "max": 215.90362548828125, "pos_frac": 0.7578125, "sample": [63.30027770996094, 51.445281982421875, 0.66461181640625, -4.97430419921875, 47.303443908691406, 12.73297119140625, -14.749931335449219, -4.45135498046875, 32.940696716308594, 7.17919921875, 3.94989013671875, -121.1611328125, 215.90362548828125, 51.596099853515625, -6.315032958984375, 103.32730102539062, 10.3671875, 188.78274536132812, 163.50039672851562, 144.10028076171875, -5.867153167724609, 7.7919921875, 191.63134765625, 146.05465698242188, 99.4661865234375, 104.1156234741211, -0.19038009643554688, -14.143218994140625, 111.72857666015625, 123.5132827758789, 116.9526138305664, 190.17184448242188, 151.67813110351562, -51.69405746459961, 122.51937866210938, 139.4269256591797, 34.4110107421875, 141.67636108398438, 168.5948486328125, -81.72509765625, 19.197265625, 156.42648315429688, 136.26605224609375, -161.88540649414062, -0.08612060546875, 187.22747802734375, 174.00955200195312, 70.59677124023438, -150.5156707763672, 59.219482421875, -24.664058685302734, -13.88873291015625, -5.01007080078125, 161.12452697753906, 90.5184326171875, 138.89581298828125, 56.208587646484375, 153.56588745117188, 123.20651245117188, 55.332000732421875, 19.55743408203125, -7.559051513671875, 159.79953002929688, 145.49319458007812, 186.637451171875, 133.77670288085938, 23.164520263671875, 36.4351806640625, -63.0947265625, -48.51397705078125, -37.77850341796875, 71.17050170898438, 136.8714599609375, 188.57275390625, 196.55120849609375, 157.37171936035156, 7.399452209472656, 57.424644470214844, 132.88352966308594, 96.681884765625, 63.922576904296875, 110.53787231445312, -1.9522628784179688, 2.9221572875976562, 170.49986267089844, 74.79476165771484, -28.745452880859375, 0.0, -37.60273742675781, 31.433929443359375, 12.943634033203125, 90.48421478271484, 148.106689453125, -8.372161865234375, 87.908935546875, 151.19215393066406, 78.99075317382812, 142.535888671875, 151.20933532714844, 129.0113525390625, 16.925189971923828, -56.153076171875, 126.2933349609375, 139.17703247070312, 117.60470581054688, 1.2271709442138672, -8.856670379638672, 7.7698974609375, -26.14886474609375, 6.141483306884766, 126.44973754882812, -75.35413360595703, 170.72494506835938, 148.7503662109375, -37.42889404296875, 7.903913497924805, 15.914602279663086, 41.8788948059082, 27.8182373046875, 67.37081909179688, 25.715728759765625, 97.9035415649414, -33.126129150390625, 19.053146362304688, 83.3670654296875, 14.251998901367188, 2.234905242919922, 181.01634216308594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000424.npy"}
{"epoch": 0.8879581151832461, "step": 425, "batch_size": 128, "mean": 52.62913131713867, "std": 88.51565551757812, "min": -203.65121459960938, "p10": -34.69642333984375, "median": 24.081283569335938, "p90": 172.68167419433593, "max": 220.58853149414062, "pos_frac": 0.71875, "sample": [143.19482421875, 150.57614135742188, 52.99334716796875, 59.500694274902344, -16.64410400390625, 15.065338134765625, 43.62664794921875, 127.56880187988281, 16.304840087890625, -14.169174194335938, 157.41748046875, 23.3349609375, 170.40097045898438, 6.937744140625, 132.2901153564453, -1.0330657958984375, 121.76429748535156, 143.52694702148438, 1.5274581909179688, 22.4525146484375, 77.20269775390625, 140.4568328857422, 24.3309326171875, -3.5543899536132812, 153.84707641601562, 101.11090850830078, 4.149452209472656, 19.61651611328125, -105.22418212890625, -3.32000732421875, 45.66448974609375, 8.0360107421875, -19.427642822265625, 14.7947998046875, 181.68673706054688, -10.278892517089844, -10.845001220703125, 187.21746826171875, 4.4794921875, 18.270790100097656, -19.50836181640625, 188.04135131835938, 138.9805908203125, 31.729583740234375, 5.2691650390625, 4.9854736328125, -36.123748779296875, -1.92327880859375, -139.27685546875, 210.60760498046875, 77.7818603515625, 143.02780151367188, -101.190673828125, 49.6893310546875, 169.73406982421875, 5.9153289794921875, 145.16310119628906, 220.58853149414062, 37.29962158203125, 23.831634521484375, -35.88700866699219, 23.475326538085938, 8.000526428222656, 40.33876037597656, 179.35748291015625, -43.16154479980469, 92.98666381835938, -3.9918651580810547, 162.6326904296875, 143.08035278320312, 140.36843872070312, 8.68353271484375, 192.2279052734375, 7.038053512573242, -18.398345947265625, -4.634552001953125, -10.94253921508789, 28.730377197265625, -19.048324584960938, 116.17938995361328, -0.31035614013671875, 163.9542236328125, 28.843658447265625, 90.03224182128906, 174.38450622558594, -111.5115737915039, 14.531929016113281, -34.18617248535156, 28.422378540039062, -22.24139404296875, 17.6273193359375, -25.000112533569336, 28.361618041992188, -141.21168518066406, -58.87481689453125, -80.17678833007812, 113.3321533203125, 0.8798828125, 139.32098388671875, -125.59332275390625, 45.933258056640625, 215.55792236328125, 17.874298095703125, 156.4185333251953, 31.34355926513672, 104.63636779785156, 164.44332885742188, 3.448638916015625, 5.955020904541016, 160.0531005859375, 0.0, -11.284942626953125, 155.02508544921875, 179.26904296875, 212.36264038085938, -4.215459823608398, 18.621028900146484, 158.9897003173828, -203.65121459960938, 40.59588623046875, 189.09710693359375, 171.95188903808594, 124.47053527832031, 157.20643615722656, 176.84774780273438, -76.5102310180664, -7.303558349609375, 2.2997360229492188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000425.npy"}
{"epoch": 0.8900523560209425, "step": 426, "batch_size": 128, "mean": 57.54353713989258, "std": 79.62841796875, "min": -145.26931762695312, "p10": -18.151017761230467, "median": 32.59047508239746, "p90": 167.7069900512695, "max": 201.13772583007812, "pos_frac": 0.765625, "sample": [142.87718200683594, 2.4219970703125, 1.7080392837524414, -35.809112548828125, 182.28463745117188, 23.597610473632812, 121.19673156738281, 2.8105506896972656, -119.22317504882812, 16.281707763671875, -105.00662231445312, -46.38238525390625, -48.85279846191406, 134.62261962890625, 150.8121795654297, 159.89356994628906, 158.62896728515625, 25.73699951171875, 138.53392028808594, 15.032470703125, -44.287689208984375, 52.4798583984375, 182.36251831054688, 19.133697509765625, 4.2683258056640625, 23.765838623046875, 168.9540252685547, 177.59066772460938, 143.37759399414062, 176.9569091796875, 116.30010986328125, 4.035209655761719, 163.626220703125, 112.10494232177734, 152.08901977539062, 3.5917835235595703, 1.4355010986328125, 23.474594116210938, 167.17254638671875, 178.7451934814453, 39.64208984375, 177.35365295410156, 8.416183471679688, -5.181549072265625, 159.82489013671875, 160.56561279296875, -2.1475830078125, 121.37458801269531, 14.008941650390625, 14.4161376953125, -99.90129089355469, 21.735992431640625, 66.44010162353516, 3.191986083984375, 32.3441276550293, 159.33615112304688, -61.50518798828125, 41.449562072753906, 159.31631469726562, 127.75833129882812, 120.353515625, 139.40194702148438, -30.6552734375, -3.425914764404297, -1.8160247802734375, 87.78936767578125, 60.41203308105469, 133.70916748046875, 1.7877349853515625, -9.38531494140625, 66.966796875, 23.039764404296875, -14.545433044433594, 12.857696533203125, 0.11355972290039062, 196.10040283203125, -24.42156982421875, -0.217681884765625, 69.85243225097656, 135.9508056640625, -15.321640014648438, 150.0986328125, 67.64837646484375, 72.44125366210938, 62.53369140625, 32.836822509765625, 35.26153564453125, 16.19330596923828, 9.304656982421875, 140.08758544921875, 131.6544189453125, 18.421478271484375, 0.3378429412841797, -19.311386108398438, 179.26779174804688, 39.36473083496094, -5.12158203125, 174.99008178710938, -5.801300048828125, -8.359724044799805, 97.45488739013672, 11.44891357421875, 0.0, 10.446174621582031, -12.370574951171875, 190.57998657226562, 72.1983642578125, 3.4152755737304688, 5.78216552734375, 50.53776550292969, 113.2935791015625, 26.5887451171875, -115.4764404296875, -2.1617431640625, -8.210296630859375, 137.50680541992188, -17.653717041015625, 161.56451416015625, 150.365966796875, 116.7601318359375, 65.47209167480469, 4.291742324829102, -145.26931762695312, 109.23333740234375, 201.13772583007812, 49.4312744140625, -16.19146728515625, 178.22108459472656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000426.npy"}
{"epoch": 0.8921465968586387, "step": 427, "batch_size": 128, "mean": 77.34085083007812, "std": 84.48245239257812, "min": -160.3155517578125, "p10": -22.737698364257813, "median": 80.4819564819336, "p90": 178.92105712890626, "max": 288.11834716796875, "pos_frac": 0.7890625, "sample": [17.78295135498047, 178.99505615234375, 13.098556518554688, -28.300537109375, -86.27435302734375, 15.48583984375, 137.1448211669922, 89.671142578125, 139.81338500976562, 35.18328857421875, 118.61474609375, 54.03385925292969, 183.29891967773438, 38.6617431640625, 154.216064453125, 168.45794677734375, 153.02806091308594, 7.4778900146484375, 144.44332885742188, 190.05535888671875, 188.75033569335938, 50.17054748535156, -21.919021606445312, 118.30694580078125, 152.63458251953125, 140.3885498046875, 132.42880249023438, 114.78717041015625, 52.91947555541992, 52.45250701904297, 206.12442016601562, 170.50637817382812, 107.5250244140625, 37.4471435546875, 0.53759765625, -3.6353836059570312, 124.46351623535156, 8.162742614746094, -1.454437255859375, -33.267913818359375, 172.70570373535156, 23.14984130859375, -84.82026672363281, 65.38980102539062, -119.6868896484375, 138.5120086669922, -0.7362079620361328, 177.49143981933594, -23.192901611328125, 146.9801025390625, 178.88934326171875, 31.07562255859375, 148.0543212890625, 78.4615478515625, 195.91061401367188, 174.737060546875, 206.14108276367188, -8.986541748046875, 13.519638061523438, -10.796798706054688, 136.7197265625, 105.27670288085938, 192.56814575195312, 18.29559326171875, 97.09771728515625, 56.661041259765625, -18.633941650390625, 149.97531127929688, 147.81126403808594, -46.507720947265625, -160.3155517578125, 171.93185424804688, 108.49624633789062, -22.598602294921875, 125.99462890625, 20.59340476989746, -23.062255859375, 149.5381317138672, 118.35232543945312, 15.104454040527344, 199.64517211914062, -110.70001220703125, 147.5704803466797, -37.84765625, 24.721389770507812, 107.76608276367188, 140.91287231445312, 128.77011108398438, 82.50236511230469, 120.7088623046875, 139.24301147460938, 173.92514038085938, -29.6131591796875, 48.9813232421875, 161.6793212890625, 138.02859497070312, 180.69375610351562, 54.5467529296875, 4.42523193359375, 125.52042388916016, -14.472274780273438, 14.118606567382812, 41.8897705078125, -27.742050170898438, 7.6884002685546875, 288.11834716796875, -7.372184753417969, 120.76945495605469, 2.7664031982421875, 165.15283203125, 179.23028564453125, 44.10060119628906, 137.26014709472656, -3.82269287109375, 46.45013427734375, -4.3566436767578125, 138.329833984375, 67.02633666992188, 0.8682861328125, 5.579345703125, 167.57757568359375, 27.87847900390625, -0.7808685302734375, -18.051551818847656, 29.27349853515625, 159.81793212890625, 196.1521453857422, 166.3851776123047], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000427.npy"}
{"epoch": 0.8942408376963351, "step": 428, "batch_size": 128, "mean": 73.73387908935547, "std": 82.36717987060547, "min": -143.21957397460938, "p10": -14.019719696044918, "median": 61.49098205566406, "p90": 168.39948730468748, "max": 261.3689880371094, "pos_frac": 0.8125, "sample": [31.052780151367188, -2.364044189453125, -58.40342712402344, 164.30950927734375, 184.42385864257812, -19.477325439453125, 56.353424072265625, 56.74151611328125, -94.95034790039062, 11.190109252929688, 261.3689880371094, 23.043670654296875, 62.75531005859375, 110.6561279296875, -7.10882568359375, 165.30517578125, 25.6458740234375, -7.9677734375, 15.803773880004883, -5.446868896484375, 112.61947631835938, -3.319671630859375, -143.21957397460938, 194.9454345703125, 159.70193481445312, 185.154296875, 148.79908752441406, 144.03823852539062, 122.25346374511719, 229.5904541015625, 47.34507751464844, 174.65216064453125, 2.2784385681152344, 86.10296630859375, -17.34429931640625, 13.128158569335938, -2.651123046875, 135.97119140625, 34.981353759765625, 134.64144897460938, 60.4185791015625, -5.866630554199219, -142.57992553710938, 134.4879150390625, 157.60604858398438, 199.48080444335938, -37.535736083984375, 35.284202575683594, 2.8174591064453125, 12.607681274414062, 147.56988525390625, 121.48114013671875, 123.90315246582031, 106.59078979492188, 25.424835205078125, 133.94619750976562, 152.3774871826172, 106.2850341796875, -43.0372314453125, 215.717529296875, -121.85444641113281, 196.16259765625, 147.82920837402344, 54.61212158203125, 151.89739990234375, 5.94757080078125, 135.43768310546875, 131.0280303955078, 134.22564697265625, 50.71038818359375, 18.018165588378906, 60.96826171875, 155.41555786132812, 152.1793212890625, 226.28253173828125, 15.94757080078125, 48.74072265625, 1.31817626953125, 79.28517150878906, 22.992446899414062, 52.321922302246094, 20.73773193359375, 150.79718017578125, 62.013702392578125, -16.3631591796875, 6.544403076171875, -17.408935546875, 22.71490478515625, 102.09011840820312, 136.8539581298828, 242.36248779296875, 167.79315185546875, 169.81427001953125, 3.60888671875, 90.5888671875, -90.00994873046875, -16.9190673828125, 173.34716796875, 130.09295654296875, 0.0, 24.420440673828125, 161.38282775878906, 154.10079956054688, 14.553298950195312, 148.22171020507812, 10.124061584472656, -13.015388488769531, 115.38424682617188, 120.3232421875, 70.67230224609375, 48.667877197265625, 140.98056030273438, 20.421409606933594, 47.929893493652344, 161.43612670898438, 75.85763549804688, 20.790191650390625, 85.33615112304688, 126.12467956542969, -1.61968994140625, 33.8505859375, 28.0723876953125, 155.2869873046875, 0.9809780120849609, -12.137420654296875, 148.690185546875, 133.94708251953125, 119.44879150390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000428.npy"}
{"epoch": 0.8963350785340314, "step": 429, "batch_size": 128, "mean": 56.45586013793945, "std": 100.05728912353516, "min": -191.11973571777344, "p10": -61.845851135253895, "median": 33.26865768432617, "p90": 183.36752014160155, "max": 213.5445556640625, "pos_frac": 0.6796875, "sample": [155.1338653564453, -184.3035888671875, -6.0472412109375, 127.00897216796875, 163.8657684326172, 159.3376922607422, 27.22589111328125, 27.296829223632812, -77.40333557128906, 123.6864013671875, 5.978691101074219, -148.4072723388672, 6.546875, 153.61856079101562, 188.96221923828125, 155.51405334472656, 1.7281169891357422, 154.0430908203125, -37.814178466796875, 22.54681396484375, 114.7275619506836, 135.27011108398438, 8.591144561767578, 161.72531127929688, -88.00155639648438, 175.20361328125, 117.2762451171875, 191.37413024902344, 24.15802001953125, 83.30126953125, -127.12826538085938, 143.4471893310547, 138.4990234375, 159.06658935546875, 30.801864624023438, 33.8975830078125, -124.06060791015625, -11.05712890625, 164.23309326171875, 213.5445556640625, 164.13360595703125, 150.13832092285156, -76.39021301269531, -7.561103820800781, 87.43817138671875, -6.24920654296875, 201.2847900390625, 104.280029296875, 8.35626220703125, -39.220703125, 9.258371353149414, 186.97158813476562, 193.50820922851562, 166.50445556640625, 11.457351684570312, 63.342742919921875, 202.794189453125, -103.86892700195312, -123.35986328125, 152.20755004882812, -3.3443603515625, 47.341217041015625, 161.68649291992188, 180.74728393554688, -33.0601806640625, 18.28693389892578, 51.6746826171875, 48.729339599609375, 32.639732360839844, 107.37483215332031, -22.652923583984375, -57.152496337890625, -38.9151611328125, -57.4569091796875, 147.74783325195312, -191.11973571777344, 175.41796875, -13.735939025878906, 16.34759521484375, 167.68580627441406, -3.2550315856933594, 0.0, 163.9291534423828, 199.75619506835938, 167.72723388671875, 63.478912353515625, 165.91734313964844, 0.49150848388671875, -1.6598663330078125, 5.236324310302734, 196.19851684570312, -30.84942626953125, 116.87925720214844, -15.688430786132812, 12.935226440429688, 51.952972412109375, 20.49996566772461, 4.455085754394531, -1.9207763671875, 120.74972534179688, -58.20484924316406, -18.244140625, -70.34152221679688, 16.234481811523438, -37.63592529296875, 107.5760498046875, 127.0289306640625, 186.50018310546875, -154.6334686279297, 123.95521545410156, 186.47561645507812, 152.21633911132812, -50.55523681640625, 179.1372833251953, 184.412841796875, -17.640838623046875, 14.879369735717773, 4.2636871337890625, 192.61517333984375, 182.91952514648438, -23.12286376953125, 138.84210205078125, -18.398452758789062, -29.764358520507812, 64.32424926757812, -7.884574890136719, 112.50238037109375, -144.59628295898438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000429.npy"}
{"epoch": 0.8984293193717278, "step": 430, "batch_size": 128, "mean": 51.42303466796875, "std": 91.53382873535156, "min": -161.12298583984375, "p10": -51.421191406249974, "median": 33.04975128173828, "p90": 173.782373046875, "max": 219.30715942382812, "pos_frac": 0.7109375, "sample": [-68.57623291015625, -18.720504760742188, -135.55809020996094, 5.498138427734375, 124.47967529296875, 126.27377319335938, -134.9462127685547, 123.88043212890625, -10.027984619140625, 21.203285217285156, 139.01751708984375, 40.730342864990234, 34.42578125, -13.846267700195312, 67.65925598144531, -161.12298583984375, 179.3292236328125, 20.8497314453125, 1.4983081817626953, 152.36801147460938, 77.97946166992188, -99.72024536132812, -16.927536010742188, 131.23760986328125, 14.232177734375, 33.53765869140625, -44.06903076171875, 0.0, 175.08502197265625, 92.9739990234375, 94.80211639404297, -22.57208251953125, 146.48138427734375, 46.90850830078125, 18.509368896484375, 0.0, 72.80142211914062, -1.8651809692382812, 3.74298095703125, 130.37646484375, -97.55377197265625, 190.35952758789062, -34.52789306640625, 150.15328979492188, 186.62973022460938, 110.56149291992188, 41.50396728515625, 20.352848052978516, -23.8671875, 219.30715942382812, 123.44758605957031, -13.5775146484375, 136.47425842285156, 15.7413330078125, 162.63092041015625, 137.8302001953125, 32.56184387207031, -76.10244750976562, 161.84918212890625, -150.8759765625, 150.5287322998047, 163.62379455566406, 175.03860473632812, 161.32003784179688, 6.8385467529296875, 63.02874755859375, 159.19656372070312, 141.3852996826172, 9.925054550170898, 183.26885986328125, 25.22998046875, 31.279251098632812, 200.96551513671875, 144.2001953125, -24.073944091796875, -23.88048553466797, -3.35791015625, 11.353851318359375, 1.0294303894042969, 108.0545654296875, 150.53265380859375, 37.076873779296875, -43.98822021484375, 11.73141860961914, -23.93663787841797, 113.67404174804688, 82.80036926269531, 204.60577392578125, 0.0, 5.36468505859375, 40.350563049316406, -26.00396728515625, 105.93704223632812, 13.270523071289062, 193.6788330078125, 117.73866271972656, 29.810302734375, 173.24398803710938, 7.057392120361328, -117.89381408691406, 193.93995666503906, 182.59262084960938, 97.91387176513672, -31.50567626953125, -117.01510620117188, 94.93084716796875, 111.6695556640625, 4.825660705566406, 178.6549072265625, 136.17996215820312, -101.44635009765625, 6.7749786376953125, 61.597198486328125, -12.9031982421875, 30.925567626953125, 165.90692138671875, 9.405914306640625, -1.443023681640625, 1.6899871826171875, -139.58567810058594, -9.819061279296875, -151.1160888671875, 66.793212890625, 131.8805694580078, 123.44922637939453, 27.9345703125, -34.68731689453125, 153.77200317382812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000430.npy"}
{"epoch": 0.900523560209424, "step": 431, "batch_size": 128, "mean": 47.484619140625, "std": 86.35225677490234, "min": -153.85464477539062, "p10": -42.94634094238281, "median": 29.330291748046875, "p90": 173.65429077148437, "max": 231.4261474609375, "pos_frac": 0.6875, "sample": [-9.671234130859375, -153.85464477539062, 55.537994384765625, 147.12413024902344, 136.419921875, 80.50259399414062, 23.971847534179688, 76.93716430664062, 11.04052734375, -99.04232788085938, -3.13653564453125, 27.887359619140625, 6.63751220703125, 39.2220458984375, -37.17047119140625, 58.538330078125, -36.558868408203125, 125.73480224609375, 113.33302307128906, 13.096735000610352, -2.75067138671875, 127.70077514648438, 51.58098602294922, -124.90604400634766, 106.94271850585938, 145.30050659179688, 89.78086853027344, -53.44683837890625, 152.7372283935547, 198.68075561523438, 31.015731811523438, 50.86631774902344, 124.4462890625, -121.9344253540039, -54.06123352050781, 145.53982543945312, 98.01290893554688, -21.602890014648438, -5.562828063964844, 109.87290954589844, -37.22014236450195, -145.46051025390625, 6.641462326049805, 16.22796630859375, 134.47711181640625, 48.31414794921875, 178.5634765625, -11.921676635742188, -36.572166442871094, -142.42294311523438, 4.4324188232421875, 33.85662841796875, 1.8183174133300781, 146.01113891601562, -0.077972412109375, 150.90679931640625, 44.19390869140625, 188.0606689453125, 159.6099853515625, -20.886764526367188, 0.1837158203125, 12.326915740966797, 165.48338317871094, 27.17596435546875, 19.5706787109375, 108.354736328125, 63.35626220703125, 175.23651123046875, -22.77240753173828, -22.454803466796875, 172.9761962890625, -20.499786376953125, -3.97119140625, 183.739013671875, -2.25994873046875, 220.17138671875, 4.67547607421875, 206.0982666015625, 44.942047119140625, 38.445587158203125, 186.56857299804688, 231.4261474609375, 93.71841430664062, -51.505340576171875, -0.9555034637451172, -14.4886474609375, 140.53140258789062, 32.29945373535156, -42.296234130859375, 18.975387573242188, 204.81488037109375, 140.97996520996094, 171.1580352783203, -30.9503173828125, 4.355804443359375, 215.89068603515625, -41.644989013671875, 187.563232421875, -3.881307601928711, 11.52459716796875, 35.366668701171875, 51.56303405761719, 9.99444580078125, 5.0186767578125, 30.773223876953125, 4.226165771484375, 145.10110473632812, 58.71916198730469, 123.11470794677734, -13.885177612304688, 35.482818603515625, -14.108154296875, 9.524459838867188, 142.066650390625, 24.50689697265625, 101.08982849121094, -12.324203491210938, -55.89056396484375, -54.80221939086914, 32.55375671386719, -10.725639343261719, -44.4632568359375, 16.93183135986328, 5.299468994140625, -88.93780517578125, 201.06492614746094, 138.44943237304688, 34.14421081542969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000431.npy"}
{"epoch": 0.9026178010471204, "step": 432, "batch_size": 128, "mean": 62.42559051513672, "std": 87.76649475097656, "min": -178.02029418945312, "p10": -30.102261352539063, "median": 40.05729675292969, "p90": 181.60302429199217, "max": 223.750732421875, "pos_frac": 0.6875, "sample": [-11.573760986328125, 161.83181762695312, 4.684173583984375, 161.09210205078125, 204.83612060546875, 192.2630615234375, 153.74697875976562, 63.082244873046875, -47.284912109375, -152.232177734375, 187.6671142578125, -3.359344482421875, 163.28106689453125, 39.7607421875, -9.070907592773438, 15.602149963378906, 35.27325439453125, 179.4910125732422, -40.28181457519531, 6.634941101074219, 128.2126922607422, -178.02029418945312, 60.94677734375, -0.3895435333251953, -16.496490478515625, 6.931333541870117, -24.6605224609375, 159.7451629638672, 62.25363540649414, -9.993255615234375, 26.30279541015625, 167.5222930908203, 193.32763671875, 169.19444274902344, -30.67633056640625, -6.44244384765625, 145.76968383789062, -36.41168212890625, 180.58602905273438, 21.01739501953125, 127.59140014648438, 138.9742431640625, 9.59429931640625, 2.8038291931152344, -2.1500015258789062, -0.070770263671875, 59.614288330078125, 12.183876037597656, -107.43710327148438, -3.000762939453125, 127.63436126708984, 102.93186950683594, 14.136207580566406, 184.44815063476562, -21.00408172607422, 10.759521484375, -79.16155242919922, 151.97607421875, -126.18701171875, 0.23616790771484375, -7.075408935546875, 24.160537719726562, 151.15065002441406, -11.666778564453125, 149.9692840576172, 5.753089904785156, 184.94168090820312, 137.44540405273438, 137.81979370117188, 132.6856231689453, -61.23091125488281, 108.64799499511719, 6.702056884765625, -15.72488021850586, 19.987815856933594, 191.0284423828125, 136.94940185546875, 197.528564453125, -21.559890747070312, 154.1171875, 44.37103271484375, 2.5800704956054688, 223.750732421875, -14.675750732421875, 148.1497802734375, 110.0225830078125, 168.0508575439453, 125.84060668945312, -29.856231689453125, 129.88381958007812, 11.641830444335938, -13.245342254638672, 0.0, 183.97601318359375, 166.19781494140625, -19.403594970703125, 11.9053955078125, -38.02838134765625, -9.4312744140625, 63.0926513671875, 59.754058837890625, -4.401325225830078, -22.459697723388672, 148.17376708984375, 111.7259292602539, 120.74945068359375, 195.62881469726562, 148.26174926757812, 196.3704833984375, 159.75247192382812, -0.05767822265625, 134.82342529296875, 40.353851318359375, 196.40643310546875, -4.3375091552734375, 47.39836120605469, 71.12127685546875, 118.34452819824219, 4.12982177734375, 139.96286010742188, 76.775390625, 146.1162109375, -34.84088134765625, 4.298961639404297, 38.09294128417969, 124.57072448730469, -4.265727996826172, -36.461517333984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000432.npy"}
{"epoch": 0.9047120418848168, "step": 433, "batch_size": 128, "mean": 50.44794464111328, "std": 86.96658325195312, "min": -189.319091796875, "p10": -32.14140930175781, "median": 46.49896812438965, "p90": 146.3062316894531, "max": 291.43646240234375, "pos_frac": 0.75, "sample": [121.4268798828125, -148.87298583984375, -177.91006469726562, 142.53237915039062, 96.69161987304688, 39.348487854003906, 120.98562622070312, -9.89068603515625, -18.614721298217773, 37.7318115234375, 35.060516357421875, 21.736328125, 141.25155639648438, 83.41815185546875, 69.683349609375, 24.36517333984375, -11.9765625, 56.14789581298828, 60.63507080078125, 17.453575134277344, 40.01678466796875, 14.301513671875, 132.21722412109375, -16.545852661132812, -0.13054466247558594, 46.535518646240234, 179.62075805664062, 5.205841064453125, 187.17950439453125, 151.17730712890625, 140.17483520507812, -22.084915161132812, 137.26551818847656, 50.61529541015625, 89.62026977539062, 186.17977905273438, 140.33090209960938, 124.81565856933594, -129.96270751953125, 76.1248779296875, -9.660606384277344, 26.21112060546875, -0.243011474609375, 43.13032531738281, 55.28782653808594, 83.79818725585938, 107.48091125488281, -1.1888847351074219, 24.0059814453125, 136.11776733398438, 144.2186279296875, 41.04880142211914, 112.37461853027344, 8.723114013671875, 48.7001953125, -97.85546875, 44.90966796875, 135.18467712402344, 13.617042541503906, 9.972579956054688, -20.23272705078125, 58.851287841796875, -118.46817779541016, 124.20915222167969, 166.80014038085938, 70.08113098144531, 133.2952117919922, -189.319091796875, 50.30580139160156, 138.50521850585938, -32.711761474609375, 18.024627685546875, 155.13641357421875, -7.166145324707031, 6.21697998046875, 31.509536743164062, 116.94686126708984, 131.78518676757812, 140.3181610107422, -173.50567626953125, -111.41006469726562, 159.23052978515625, 93.70341491699219, -28.766845703125, 27.7596435546875, -3.782958984375, 153.3223876953125, -6.343357086181641, 71.64044189453125, 40.663177490234375, 136.38894653320312, -2.1015472412109375, -177.829833984375, 14.895177841186523, 20.267608642578125, 121.66778564453125, 102.14910888671875, 166.17015075683594, 120.86343383789062, -113.40625, 54.01617431640625, -14.027755737304688, 2.8404998779296875, 291.43646240234375, 38.4173583984375, 116.50834655761719, 154.524658203125, 144.1964874267578, 153.5301055908203, 1.6457977294921875, 59.885833740234375, 18.4361572265625, 87.20516967773438, -31.89697265625, 31.660736083984375, 100.61297607421875, 46.46241760253906, -5.485553741455078, 136.9766082763672, 192.90704345703125, 33.6787109375, 121.49603271484375, -153.1338348388672, 69.62977600097656, -22.11870574951172, -37.23956298828125, 115.3682861328125, 4.446250915527344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000433.npy"}
{"epoch": 0.9068062827225131, "step": 434, "batch_size": 128, "mean": 59.03070831298828, "std": 89.20281982421875, "min": -174.70468139648438, "p10": -28.076291656494135, "median": 31.471858978271484, "p90": 168.174609375, "max": 236.5435791015625, "pos_frac": 0.6875, "sample": [85.60575866699219, -13.533828735351562, -11.857757568359375, 4.88665771484375, 165.20753479003906, 20.525894165039062, 133.06549072265625, 127.54609680175781, -22.28314208984375, 230.43170166015625, -9.520437240600586, 152.9109344482422, -76.3408203125, 2.4350738525390625, -9.449443817138672, 7.345703125, -20.233978271484375, 128.6246337890625, 115.51007843017578, -1.06024169921875, 16.80255126953125, 218.28897094726562, -5.97955322265625, 236.5435791015625, 164.77151489257812, -80.19512939453125, 0.0, -16.17589569091797, -7.563301086425781, 137.69993591308594, 191.4747314453125, 152.50167846679688, -20.299667358398438, 172.96920776367188, 137.16421508789062, 144.5692901611328, -9.58941650390625, 42.350738525390625, 25.39227294921875, 192.3798828125, 53.388946533203125, 115.11160278320312, 176.33712768554688, -46.605224609375, 33.26008605957031, -139.44235229492188, 108.67526245117188, 124.9249267578125, 31.0880126953125, 106.46768188476562, 167.98040771484375, 166.5340118408203, 114.956787109375, 226.81329345703125, 27.572113037109375, 125.45597839355469, 148.25396728515625, 11.3966064453125, -31.929779052734375, 168.62774658203125, 21.502159118652344, 53.842041015625, 201.4849853515625, -18.757789611816406, 4.566802978515625, -2.69158935546875, 79.95291137695312, 161.16445922851562, 21.8685302734375, 31.85570526123047, 191.13092041015625, 5.2161407470703125, -16.44580078125, 173.8751220703125, -61.338134765625, 2.8425025939941406, -33.051910400390625, 187.99655151367188, 5.79766845703125, 132.89047241210938, 161.82757568359375, -34.468902587890625, 79.44949340820312, 10.4300537109375, 41.164520263671875, 153.125, 158.09512329101562, -173.2921142578125, 136.4677734375, -20.149017333984375, 59.51438903808594, -53.250640869140625, -118.6551513671875, 0.0, 144.75119018554688, 28.26921844482422, 20.751861572265625, 165.56689453125, 166.254638671875, 12.166305541992188, -18.003753662109375, -5.709014892578125, -23.0290470123291, 135.47222900390625, -10.011566162109375, 9.470165252685547, -5.625322341918945, -26.42479705810547, 139.39981079101562, 98.29139709472656, 43.524505615234375, 2.443767547607422, 0.0, 112.2818603515625, 118.55311584472656, -6.065765380859375, 22.221939086914062, 150.19720458984375, 92.8092041015625, 83.14825439453125, -174.70468139648438, 17.37811279296875, 25.126220703125, -6.673828125, -116.84103393554688, 165.74169921875, 113.29728698730469, 146.1541748046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000434.npy"}
{"epoch": 0.9089005235602095, "step": 435, "batch_size": 128, "mean": 53.88923645019531, "std": 83.74420166015625, "min": -172.7618408203125, "p10": -26.2843162536621, "median": 39.2042236328125, "p90": 163.99048919677733, "max": 223.18353271484375, "pos_frac": 0.765625, "sample": [134.30677795410156, 10.662139892578125, 6.297389984130859, 69.81165313720703, 33.40745544433594, 177.12425231933594, 153.31394958496094, -113.74983215332031, 128.48638916015625, -7.043304443359375, -10.93411636352539, -19.682220458984375, -172.7618408203125, 45.36920928955078, 4.915609359741211, -18.9283447265625, 72.24346923828125, 69.98394775390625, 160.07968139648438, 112.92691040039062, 12.36541748046875, 26.93907356262207, 165.4764404296875, 0.0, 3.080310821533203, -9.7532958984375, 60.2490234375, 23.25550079345703, 131.9505615234375, 15.786849975585938, 136.91604614257812, 151.3516845703125, 132.50592041015625, -24.218605041503906, -8.71209716796875, 200.21502685546875, 72.08425903320312, 161.277099609375, 41.954345703125, 8.306610107421875, 169.01654052734375, 168.3169403076172, 167.3831787109375, 24.503524780273438, 149.07046508789062, 19.918853759765625, 144.247802734375, -31.10430908203125, 163.35365295410156, 35.1024169921875, 143.37432861328125, 91.07763671875, 196.186279296875, 143.22991943359375, 29.936662673950195, 4.493904113769531, 52.422210693359375, 20.486404418945312, 70.35914611816406, -76.31251525878906, 11.111831665039062, 194.67828369140625, 8.63541030883789, -132.21693420410156, 17.842918395996094, 14.05987548828125, 40.581390380859375, 168.68527221679688, -111.36338806152344, 11.2716064453125, 27.648773193359375, -2.9239654541015625, 9.98468017578125, 179.48141479492188, 180.27645874023438, -9.341766357421875, 141.6116485595703, 158.3197021484375, -80.51522827148438, 102.15116882324219, 4.1219482421875, 144.58111572265625, 60.93988037109375, 139.3513641357422, -131.04791259765625, 9.0302734375, 78.47552490234375, 4.409616470336914, -0.23273849487304688, -10.20550537109375, -7.581920623779297, 139.27609252929688, 115.44900512695312, 223.18353271484375, 107.48286437988281, 158.44337463378906, 59.39112854003906, 122.30352783203125, -23.711570739746094, 59.5472412109375, -6.6206817626953125, 41.57740783691406, 103.75648498535156, 3.5105361938476562, 20.409759521484375, 16.115966796875, 37.827056884765625, 7.8351898193359375, 132.0955810546875, 5.604621887207031, -9.61541748046875, -63.58888244628906, -136.83169555664062, -49.283958435058594, -1.58612060546875, 206.72824096679688, 10.649658203125, 49.706878662109375, 103.28916931152344, -135.85360717773438, 47.888031005859375, 128.4549560546875, 89.25341796875, 0.0143585205078125, -31.382125854492188, 151.34725952148438, 49.170166015625, 122.24098205566406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000435.npy"}
{"epoch": 0.9109947643979057, "step": 436, "batch_size": 128, "mean": 37.96335220336914, "std": 83.52153778076172, "min": -171.21209716796875, "p10": -48.976739501953126, "median": 20.363834381103516, "p90": 159.8475112915039, "max": 212.32208251953125, "pos_frac": 0.671875, "sample": [107.00039672851562, -24.2626953125, 196.93450927734375, 133.77304077148438, 3.213838577270508, 29.88416290283203, 156.9557342529297, 30.062332153320312, -36.25128173828125, 95.6881103515625, -104.19973754882812, 56.31988525390625, -64.11837768554688, 34.413818359375, 2.2993621826171875, 185.90924072265625, 28.662353515625, -18.96619415283203, -49.9984130859375, -80.2608642578125, 13.088546752929688, 120.96403503417969, 135.62686157226562, 121.73443603515625, 173.33547973632812, -26.98772430419922, 153.47711181640625, -0.556243896484375, 139.56494140625, 23.02734375, 37.20098114013672, 8.1090087890625, -4.359195709228516, 22.03814697265625, 161.8014678955078, 44.11405944824219, -11.259979248046875, 46.455543518066406, -0.170440673828125, -30.155868530273438, 105.50469970703125, 10.9722900390625, 17.25659942626953, 105.65412902832031, 167.29678344726562, 30.8382568359375, 6.2299957275390625, 4.9031982421875, -16.72021484375, 159.01010131835938, 32.052581787109375, -22.65423583984375, 31.520751953125, -13.22601318359375, 130.57965087890625, -31.64551544189453, 182.87136840820312, -171.21209716796875, 5.416595458984375, 163.25225830078125, 145.3472900390625, 39.619415283203125, 12.009552001953125, 61.696895599365234, 52.460662841796875, 137.62359619140625, -4.0289306640625, -135.93966674804688, 94.918212890625, 134.32081604003906, 2.013427734375, 212.32208251953125, 62.339111328125, -117.05873107910156, 22.454879760742188, 61.06414794921875, 12.40554428100586, 125.00774383544922, 10.2891845703125, -2.038848876953125, 74.04513549804688, -24.025344848632812, 15.219390869140625, -6.5780029296875, 60.857421875, 67.44375610351562, -4.1331329345703125, -75.32875061035156, 41.387481689453125, -67.22134399414062, 170.61978149414062, -15.055191040039062, -4.294414520263672, -47.19482421875, 14.527191162109375, 22.052978515625, 211.422607421875, -0.2787017822265625, 3.9908065795898438, -30.676010131835938, -15.395660400390625, -163.956787109375, 18.68952178955078, 132.21011352539062, 104.80780029296875, 11.64080810546875, -136.28939819335938, 139.8448944091797, 210.549560546875, -12.9793701171875, 9.502449035644531, 27.908157348632812, -26.18670654296875, 123.04135131835938, 18.624679565429688, 114.78550720214844, 1.8853759765625, 199.98382568359375, -68.1192398071289, 10.558013916015625, -48.53887939453125, 47.186431884765625, -152.39715576171875, 99.77627563476562, -27.4879150390625, 173.82147216796875, 32.27680969238281, -8.04913330078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000436.npy"}
{"epoch": 0.9130890052356021, "step": 437, "batch_size": 128, "mean": 64.31895446777344, "std": 90.3431396484375, "min": -156.16964721679688, "p10": -20.54325256347656, "median": 56.489646911621094, "p90": 175.98243408203123, "max": 230.43408203125, "pos_frac": 0.7734375, "sample": [25.4783935546875, 159.076416015625, 65.753173828125, 141.010498046875, -155.52919006347656, 181.9456787109375, 5.661651611328125, 7.8634185791015625, 75.69427490234375, 13.13226318359375, 168.6180419921875, 21.8876953125, 130.28367614746094, 69.03884887695312, 10.987075805664062, 148.24789428710938, 0.2769775390625, 4.6997833251953125, 46.0054931640625, 147.1226806640625, 23.047409057617188, 181.75778198242188, 230.43408203125, 162.7007598876953, 16.865966796875, 29.73992919921875, -2.6332244873046875, 144.42666625976562, 144.6614532470703, 98.57029724121094, -16.340576171875, 2.63275146484375, 8.198455810546875, 70.09034729003906, -4.2176513671875, 48.504730224609375, -138.7830810546875, -24.164688110351562, 9.050125122070312, 30.781494140625, 145.66038513183594, 57.370452880859375, 168.56024169921875, 156.12460327148438, -43.135650634765625, 4.790435791015625, 30.12567138671875, 16.6749267578125, 181.1300048828125, 137.6815185546875, 133.87066650390625, 195.8028106689453, 123.36212921142578, 161.46670532226562, 8.836868286132812, 39.739044189453125, -22.79998779296875, -14.601696014404297, 185.02032470703125, 177.10540771484375, 144.03948974609375, -2.63629150390625, 173.15896606445312, 185.78323364257812, 209.74383544921875, 99.84544372558594, 1.8310737609863281, 87.76348876953125, -19.055618286132812, -19.576080322265625, 65.54670715332031, 7.73370361328125, -34.289825439453125, 172.2952880859375, -152.41683959960938, 153.58456420898438, 0.9488668441772461, 55.60884094238281, -156.16964721679688, 0.0, 153.0301513671875, -121.45065307617188, 6.40814208984375, -8.969318389892578, 175.50115966796875, 134.42465209960938, -59.91339111328125, 112.35345458984375, -136.48159790039062, 136.89700317382812, 184.59783935546875, 142.6199951171875, -2.7936553955078125, 128.95108032226562, 81.68301391601562, -2.1866455078125, 146.8820037841797, 23.96747589111328, 83.94354248046875, 41.203887939453125, 180.06785583496094, 183.96295166015625, 114.85708618164062, 170.64199829101562, 28.696029663085938, -19.154502868652344, 72.3173828125, 114.9971694946289, 160.87738037109375, -128.4564971923828, 4.6296539306640625, 0.50323486328125, 142.25753784179688, -154.91387939453125, -7.98577880859375, 148.78408813476562, 52.301513671875, 148.666259765625, 131.50985717773438, -1.3384246826171875, 107.48312377929688, -13.1773681640625, 26.725311279296875, 73.11395263671875, 0.0, 16.824066162109375, 134.3164520263672, 224.550048828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000437.npy"}
{"epoch": 0.9151832460732985, "step": 438, "batch_size": 128, "mean": 31.133052825927734, "std": 87.43158721923828, "min": -163.3929443359375, "p10": -75.43216781616206, "median": 12.539566040039062, "p90": 153.6873062133789, "max": 222.7574462890625, "pos_frac": 0.609375, "sample": [25.219566345214844, -162.99105834960938, 117.12966918945312, -14.327133178710938, -36.85057067871094, 144.29486083984375, 53.437896728515625, 34.57844543457031, 0.0, 12.189422607421875, 173.59982299804688, -45.61285400390625, 16.409767150878906, 153.1529541015625, 41.0682373046875, 9.909271240234375, 14.89923095703125, 154.9341278076172, 135.08462524414062, -18.818870544433594, -39.821258544921875, -0.52606201171875, 42.437744140625, -0.22946929931640625, -101.281494140625, -64.86756134033203, -21.59228515625, 121.88102722167969, 9.3800048828125, 49.40606689453125, 10.832307815551758, 184.37582397460938, 130.96295166015625, -11.009994506835938, 30.8048095703125, 176.82818603515625, -138.656005859375, -9.893905639648438, 63.91009521484375, -163.12640380859375, -163.3929443359375, -107.63163757324219, 72.900390625, 129.00177001953125, -0.8060760498046875, -11.272796630859375, 2.553680419921875, 26.424224853515625, -145.03421020507812, -2.544597625732422, -2.42913818359375, -136.395751953125, -1.9299468994140625, 15.385955810546875, 7.58062744140625, 25.636268615722656, 113.826904296875, -38.294189453125, 47.481903076171875, 112.74079132080078, 11.481338500976562, -20.192888259887695, -0.9679632186889648, -143.65667724609375, 117.7685546875, -6.578113555908203, 24.466278076171875, 33.72489929199219, 168.55233764648438, -119.63179016113281, 64.5242919921875, -38.822349548339844, 163.10394287109375, -17.386375427246094, 25.7725830078125, 66.85745239257812, 120.19020080566406, -100.08291625976562, 152.05401611328125, 147.81100463867188, -29.958953857421875, 146.45538330078125, -41.55962371826172, -3.056121826171875, 192.17233276367188, -10.885848999023438, 135.4990692138672, 72.49795532226562, -108.8140869140625, 12.143310546875, 70.18476104736328, 14.555381774902344, 1.5622310638427734, -9.08526611328125, 170.9493408203125, 138.66156005859375, 10.422531127929688, 162.54124450683594, 216.3824462890625, -23.298370361328125, -16.3426513671875, -40.671844482421875, 33.06634521484375, 113.47021484375, 124.9863052368164, -28.967041015625, 14.974639892578125, 20.96997833251953, -29.914520263671875, 142.3911590576172, 171.42230224609375, 11.080317497253418, -50.7344970703125, 53.692081451416016, 0.0, 5.3278961181640625, 120.35440063476562, 10.093681335449219, -26.56523895263672, 86.47718811035156, 165.66827392578125, -100.57803344726562, 7.744817733764648, 151.3665771484375, 222.7574462890625, -5.417228698730469, 12.88970947265625, 28.208465576171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000438.npy"}
{"epoch": 0.9172774869109948, "step": 439, "batch_size": 128, "mean": 52.737518310546875, "std": 88.25176239013672, "min": -190.4892578125, "p10": -64.5521629333496, "median": 32.864036560058594, "p90": 165.76710968017576, "max": 212.88885498046875, "pos_frac": 0.71875, "sample": [-95.67669677734375, -103.63937377929688, -142.86111450195312, 192.14413452148438, -61.08428955078125, 163.461669921875, 14.186676025390625, 18.47661590576172, 64.30342102050781, 154.8469696044922, -3.8493709564208984, 19.688629150390625, 10.123945236206055, 4.451324462890625, 126.58570861816406, -102.41412353515625, 9.66412353515625, 114.8740234375, -2.4297027587890625, 32.364898681640625, 184.569580078125, 151.34335327148438, 7.679046630859375, 35.367462158203125, 2.04473876953125, 139.88427734375, 55.28240203857422, 144.43023681640625, 168.32376098632812, 18.82072639465332, -65.5821304321289, 92.3963623046875, 14.034420013427734, 194.38836669921875, -20.444122314453125, 116.89627075195312, 194.92071533203125, 61.622772216796875, 212.88885498046875, 25.943939208984375, -33.7744140625, 17.20296859741211, 146.51031494140625, 132.04759216308594, 3.5631103515625, 142.84060668945312, 141.05673217773438, 13.4727783203125, 4.0051422119140625, 2.18756103515625, 0.94482421875, 16.630050659179688, -83.6357421875, 15.744476318359375, 42.52513122558594, 159.90203857421875, 134.85426330566406, -24.28083038330078, 72.31465911865234, 0.0, -67.47122192382812, 0.0, -30.8228759765625, 14.905220031738281, -72.439453125, 162.07080078125, 23.40667724609375, -64.11074829101562, 10.016656875610352, -35.79296875, -1.67987060546875, 150.4085693359375, 15.97222900390625, -23.930068969726562, -7.5369110107421875, 81.45430755615234, -6.464099884033203, 108.81147003173828, -42.370269775390625, 125.85910034179688, -34.043121337890625, 104.66728973388672, 151.21588134765625, 25.09979248046875, 73.78359985351562, 91.15711975097656, -87.07682800292969, -3.74932861328125, -90.45496368408203, -6.387451171875, 164.67140197753906, -136.5579833984375, 0.0, 171.131103515625, 189.1168212890625, 100.52042388916016, 187.02395629882812, 148.64944458007812, 40.249114990234375, -38.646209716796875, 44.9981689453125, -66.16548156738281, 136.3772735595703, 33.36317443847656, 54.08924865722656, 17.892822265625, 120.06796264648438, 122.73384094238281, 132.03564453125, 37.546630859375, 158.77618408203125, 73.1195068359375, 50.42155838012695, 163.97003173828125, 146.98886108398438, 0.21460533142089844, -190.4892578125, 185.2416534423828, 111.67776489257812, 179.40771484375, 158.4263916015625, 131.83865356445312, 18.832916259765625, -18.644500732421875, 179.74510192871094, 0.0, 172.1484832763672, 152.99099731445312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000439.npy"}
{"epoch": 0.9193717277486911, "step": 440, "batch_size": 128, "mean": 42.44549560546875, "std": 90.38330078125, "min": -178.97467041015625, "p10": -83.26473922729492, "median": 25.234268188476562, "p90": 160.14338836669918, "max": 232.92987060546875, "pos_frac": 0.6796875, "sample": [-37.932342529296875, 0.0, 33.38720703125, -140.20245361328125, -15.728168487548828, 80.54647827148438, -141.01608276367188, 114.35791015625, -9.621528625488281, 108.41421508789062, -21.027435302734375, 28.4515380859375, -97.16912841796875, 1.663299560546875, 74.607666015625, -123.81135559082031, 103.58366394042969, 153.25149536132812, 127.134765625, 139.74560546875, 45.340576171875, 178.41156005859375, -2.1188278198242188, 23.639892578125, -63.882720947265625, -5.824432373046875, -178.97467041015625, 116.74760437011719, 78.6612548828125, 75.9110107421875, 179.97607421875, 202.46957397460938, 10.794742584228516, -14.200439453125, -19.24224853515625, 53.7421875, 133.09841918945312, -151.23483276367188, 0.0, 51.7420654296875, -160.684326171875, -4.263397216796875, 9.7303466796875, 113.83490753173828, 168.1952362060547, -8.891716003417969, -100.48367309570312, 0.491485595703125, 16.33202362060547, 138.8738250732422, 193.74551391601562, -14.098548889160156, 132.20819091796875, 0.44222259521484375, 39.173248291015625, 178.84664916992188, -2.745746612548828, -124.65754699707031, -45.143707275390625, -36.13128662109375, -45.778717041015625, 0.2630615234375, 6.277923583984375, -39.690460205078125, 232.92987060546875, 139.91793823242188, 194.45291137695312, 132.3246307373047, 150.41058349609375, 19.462005615234375, -0.0067272186279296875, 26.76873779296875, 13.762313842773438, 141.03640747070312, 153.31649780273438, 6.7878570556640625, -2.514068603515625, -91.28782653808594, 17.416961669921875, 17.305084228515625, 43.45611572265625, 125.53083801269531, 147.4842529296875, 73.57492065429688, 180.22898864746094, -97.06338500976562, 29.92291259765625, 180.2464599609375, 40.036376953125, -100.19572448730469, 185.24859619140625, -36.35328674316406, 147.3797607421875, -14.949005126953125, 139.2587890625, 34.53633117675781, 31.426284790039062, 29.526485443115234, 137.6123809814453, 41.32708740234375, 154.75238037109375, -39.921226501464844, 20.3428955078125, 19.776351928710938, 47.809295654296875, 68.740966796875, 6.3449554443359375, 176.99566650390625, 127.41191864013672, 70.32546997070312, 12.165863037109375, 54.89015197753906, 23.699798583984375, 154.7929229736328, -39.567657470703125, 156.69259643554688, 143.1455078125, 21.239532470703125, 16.419586181640625, 105.89776611328125, -26.91033935546875, 175.2286376953125, 17.623580932617188, -82.24398803710938, -5.576019287109375, 16.565017700195312, -85.64649200439453, 142.1720428466797], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000440.npy"}
{"epoch": 0.9214659685863874, "step": 441, "batch_size": 128, "mean": 61.35826873779297, "std": 81.83232116699219, "min": -165.3822479248047, "p10": -18.965844726562498, "median": 45.62001037597656, "p90": 172.67332916259764, "max": 211.16650390625, "pos_frac": 0.78125, "sample": [-11.55523681640625, 29.697998046875, -130.23696899414062, 47.7320556640625, 0.11338233947753906, -11.775482177734375, 180.49545288085938, -10.64752197265625, 3.14263916015625, -5.902099609375, -46.76287841796875, 211.16650390625, 27.607337951660156, 124.95733642578125, 44.02008056640625, 139.48208618164062, 155.48309326171875, 150.4234619140625, 81.09060668945312, 75.19309997558594, 63.8934326171875, 128.68125915527344, 12.232891082763672, 18.36273193359375, 165.48089599609375, 77.78350830078125, 148.8629150390625, 161.2284393310547, 43.14691162109375, 5.248626708984375, 13.340484619140625, 127.096923828125, 94.86482238769531, -13.045196533203125, 141.0005645751953, 112.7745361328125, 16.2335205078125, 161.85946655273438, 131.6289825439453, -45.397377014160156, 158.5006103515625, 142.2352752685547, -73.41350555419922, -5.01251220703125, 173.8779296875, -1.7401752471923828, 44.043548583984375, 36.504119873046875, 142.028564453125, 175.66192626953125, 163.8682403564453, 8.44580078125, 49.372100830078125, 18.03905487060547, 17.693572998046875, 13.72283935546875, -3.83514404296875, 148.0858154296875, 19.20806884765625, -20.592071533203125, 1.9940185546875, 31.9168701171875, 19.621414184570312, 142.9610595703125, 0.248199462890625, 0.0, 3.0125503540039062, 90.78416442871094, 50.81532287597656, 193.81674194335938, -66.8115234375, -67.22773742675781, -10.852327346801758, 10.456039428710938, 0.0, 138.9404754638672, 55.27386474609375, 51.316680908203125, 141.794189453125, 47.19647216796875, 13.430099487304688, 149.37771606445312, 24.181060791015625, -76.39395141601562, 47.557098388671875, 65.40460205078125, 142.39625549316406, 96.90547180175781, 0.00164794921875, 30.2554931640625, 3.81756591796875, 5.7071533203125, 39.152069091796875, 127.43789672851562, 141.1243896484375, 8.52423095703125, 168.8263397216797, -110.20122528076172, 172.324462890625, 51.78143310546875, 171.37521362304688, -165.3822479248047, 61.476715087890625, 3.476715087890625, -67.78292846679688, -0.2897491455078125, 193.99911499023438, -109.42465209960938, 157.5994415283203, -18.268890380859375, 30.399559020996094, 176.5443115234375, 174.65200805664062, 173.4873504638672, 188.5584716796875, 105.53933715820312, 62.132102966308594, -4.4232177734375, -30.94818115234375, 9.03369140625, 202.35736083984375, -6.7213592529296875, 139.44696044921875, 186.6768798828125, 174.77651977539062, 113.47428894042969, 132.89553833007812, 36.63446044921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000441.npy"}
{"epoch": 0.9235602094240838, "step": 442, "batch_size": 128, "mean": 47.50719451904297, "std": 95.62236785888672, "min": -201.4508056640625, "p10": -77.98026733398437, "median": 29.37994384765625, "p90": 170.2274627685547, "max": 245.800537109375, "pos_frac": 0.71875, "sample": [138.13235473632812, 13.390350341796875, 21.262557983398438, -27.409748077392578, 205.7708740234375, 162.96649169921875, 0.08175277709960938, -9.287307739257812, 170.18411254882812, 3.2301177978515625, 85.18502807617188, 22.679733276367188, -93.24381256103516, -8.740768432617188, 148.95169067382812, -145.2242431640625, 28.010101318359375, 22.00531005859375, 12.043121337890625, -134.23394775390625, 13.19610595703125, 39.33620834350586, -4.363319396972656, -5.2627716064453125, 123.31301879882812, -5.9205780029296875, 128.47119140625, -77.3399658203125, 188.47544860839844, 173.17996215820312, -143.10733032226562, 5.602630615234375, -53.36883544921875, 148.07550048828125, 120.07868957519531, 145.31143188476562, -20.163803100585938, 4.3034515380859375, -130.6827392578125, 8.452896118164062, 20.522735595703125, -28.782676696777344, 124.11727905273438, 18.711952209472656, -25.300537109375, -160.98977661132812, 14.608749389648438, 36.5211181640625, 10.619926452636719, -22.01483154296875, 15.477996826171875, 5.442596435546875, 105.61460876464844, 203.99273681640625, 7.011545181274414, 101.0853271484375, 135.0973358154297, 0.0, 81.98818969726562, -120.5367431640625, -59.834869384765625, 11.692314147949219, -145.8994903564453, 150.4245147705078, -123.27191162109375, 179.1704864501953, 76.75799560546875, -4.7423858642578125, 71.76263427734375, 131.22879028320312, 0.22562599182128906, -201.4508056640625, 178.43734741210938, -9.754608154296875, 90.75430297851562, 66.88388061523438, 201.76422119140625, -22.77374267578125, 125.35440063476562, 46.81495666503906, 30.749786376953125, 61.893829345703125, -3.9682769775390625, 201.46441650390625, 145.2366943359375, 170.61463928222656, -79.47430419921875, -140.09527587890625, -49.0054931640625, -35.96714782714844, 17.306915283203125, 101.00982666015625, 14.00201416015625, 115.41400146484375, 16.351226806640625, 139.67681884765625, 105.28675842285156, 140.09666442871094, 103.63735961914062, 155.14263916015625, 182.30673217773438, 169.093017578125, 154.55703735351562, 170.32861328125, 49.95037841796875, 245.800537109375, 8.7261962890625, -55.70482635498047, -38.36286926269531, 202.57388305664062, 123.62832641601562, 142.32814025878906, 59.431549072265625, 133.7948760986328, 10.576065063476562, 50.86665344238281, 11.377845764160156, 66.864501953125, 144.35653686523438, -52.70562744140625, 75.75709533691406, 140.822509765625, 156.1728515625, 26.47711181640625, -134.9076385498047, 63.7576904296875, 138.5106658935547, 65.09805297851562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000442.npy"}
{"epoch": 0.9256544502617801, "step": 443, "batch_size": 128, "mean": 49.42631530761719, "std": 87.37177276611328, "min": -174.15423583984375, "p10": -43.144260406494126, "median": 27.277313232421875, "p90": 167.92184906005858, "max": 225.21240234375, "pos_frac": 0.7109375, "sample": [103.97076416015625, 9.428985595703125, 129.22903442382812, 20.262237548828125, -19.1976318359375, 54.1768798828125, 181.6952667236328, 10.281005859375, 178.897705078125, -12.849479675292969, 13.723907470703125, 58.33612060546875, -7.790000915527344, 96.9022216796875, -12.030548095703125, -120.12384033203125, -15.440521240234375, 104.84510803222656, 163.86346435546875, -7.280162811279297, 41.22666931152344, 82.40167236328125, 71.53868103027344, 42.07965087890625, -18.855712890625, 30.471282958984375, 24.48846435546875, 152.32730102539062, 30.415145874023438, -53.779136657714844, -26.07275390625, 47.159393310546875, -144.6671142578125, 5.911205291748047, -8.177032470703125, 22.276611328125, 163.652099609375, 23.44573974609375, 156.90103149414062, 11.325489044189453, 170.335693359375, -17.049346923828125, -9.829254150390625, 1.716400146484375, 22.880538940429688, 50.304466247558594, 225.21240234375, 5.8155059814453125, 71.3670654296875, -29.377960205078125, 67.0098876953125, -166.34146118164062, -61.219451904296875, -70.46044921875, 57.7169189453125, 142.87794494628906, 110.05384826660156, 214.90933227539062, 72.87773132324219, 184.25735473632812, 119.4066162109375, 30.066162109375, -7.595008850097656, 139.67947387695312, 164.80999755859375, 4.725273132324219, -2.2161865234375, 40.05425262451172, 146.5503692626953, 144.401123046875, -7.8182220458984375, -5.1163330078125, 184.4295654296875, 15.029960632324219, 167.1874542236328, 0.385589599609375, -4.210121154785156, 146.71205139160156, -54.028106689453125, 10.646560668945312, -12.48809814453125, 169.63543701171875, 163.41387939453125, 117.09249877929688, 85.16116333007812, -59.169471740722656, 14.360572814941406, -98.57998657226562, 133.6665496826172, 197.91070556640625, 196.15838623046875, 143.66473388671875, 41.463714599609375, 22.702957153320312, -98.48472595214844, 127.45947265625, 11.107734680175781, 191.07061767578125, 125.15138244628906, 145.06024169921875, 66.22900390625, -18.082275390625, -154.42959594726562, 159.25645446777344, -5.187255859375, 2.093505859375, 47.081390380859375, 128.907470703125, 55.871002197265625, 7.9695587158203125, 143.7017059326172, -7.323699951171875, -38.586456298828125, 8.510303497314453, 3.3793182373046875, -11.313987731933594, 77.79483032226562, 170.11643981933594, 8.664596557617188, -174.15423583984375, 160.36572265625, 143.08201599121094, 173.90016174316406, 4.480064392089844, -2.91455078125, -120.41622924804688, 10.33197021484375, 9.756500244140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000443.npy"}
{"epoch": 0.9277486910994764, "step": 444, "batch_size": 128, "mean": 53.940185546875, "std": 90.65724182128906, "min": -192.8154296875, "p10": -41.04326019287109, "median": 32.13762855529785, "p90": 172.6047119140625, "max": 281.0177001953125, "pos_frac": 0.7109375, "sample": [3.688976287841797, -3.7058563232421875, 172.406005859375, 22.25628662109375, 75.94070434570312, 56.36541748046875, -24.38946533203125, 4.43511962890625, 19.417457580566406, -0.860687255859375, -8.668746948242188, 32.728851318359375, -38.322174072265625, 39.843170166015625, 144.27589416503906, 27.2049560546875, 132.9621124267578, 58.440887451171875, -24.70562744140625, 33.25818634033203, 115.16450500488281, 29.1043701171875, 62.16986083984375, -40.41796875, 18.676239013671875, -32.91998291015625, 151.00616455078125, 16.806442260742188, 6.8186187744140625, 58.4422607421875, 15.575424194335938, 149.70657348632812, 136.06686401367188, -17.081497192382812, 168.76710510253906, -27.993438720703125, -20.442977905273438, 39.679656982421875, 160.6331787109375, 143.997314453125, 162.77166748046875, 203.05145263671875, -16.54815673828125, -4.365427017211914, 43.0654296875, 150.78286743164062, 19.98675537109375, -135.5823974609375, -123.72317504882812, -40.040313720703125, 216.9146728515625, 124.33071899414062, -73.68209838867188, 200.31829833984375, -5.117835998535156, 72.1744384765625, 224.0003662109375, 171.9215087890625, -4.0638427734375, -11.682037353515625, 1.8995113372802734, 281.0177001953125, 136.98133850097656, -30.608871459960938, 68.956298828125, -8.354461669921875, 182.27737426757812, 131.67030334472656, -80.88374328613281, 205.74612426757812, 140.62217712402344, 91.49887084960938, -58.15338134765625, 177.98590087890625, -15.184600830078125, 55.41778564453125, 185.28497314453125, 158.5029296875, 156.64434814453125, 71.65116882324219, 166.52139282226562, 1.161722183227539, 1.0112152099609375, 191.89389038085938, 4.725482940673828, 155.67852783203125, -8.593864440917969, 143.95486450195312, 138.83929443359375, -109.96526336669922, 3.273468017578125, 186.2960205078125, 8.762626647949219, 133.8996124267578, 40.583580017089844, 27.926239013671875, 45.426025390625, -12.144622802734375, 9.8719482421875, -62.38134765625, -19.222900390625, 8.078521728515625, 175.58950805664062, 4.8999481201171875, -83.73469543457031, 37.01800537109375, -56.410194396972656, 150.0845947265625, -42.50227355957031, 132.25613403320312, 5.884485244750977, 63.294647216796875, 9.859771728515625, 118.95578002929688, 95.12200927734375, 31.546405792236328, 173.068359375, 135.26060485839844, 143.9595947265625, 130.96649169921875, -3.4776535034179688, 12.352169036865234, -59.96185302734375, 1.1378154754638672, -170.77627563476562, 24.53045654296875, 128.8245391845703, -192.8154296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000444.npy"}
{"epoch": 0.9298429319371728, "step": 445, "batch_size": 128, "mean": 43.262535095214844, "std": 89.1049575805664, "min": -159.2586212158203, "p10": -57.87196960449218, "median": 18.01885223388672, "p90": 170.0952194213867, "max": 215.68130493164062, "pos_frac": 0.671875, "sample": [8.876861572265625, -158.89222717285156, 183.42129516601562, 79.80368041992188, 43.048095703125, 183.5557861328125, -5.121952056884766, 46.33772277832031, 167.2791748046875, -116.70730590820312, 9.417388916015625, 76.93739318847656, 149.4348907470703, -65.70059204101562, -3.8187255859375, 3.412200927734375, 52.350494384765625, 12.742385864257812, 199.9697265625, 18.74920654296875, 171.90182495117188, -6.272674560546875, 6.539299011230469, -107.6894302368164, 28.344146728515625, 129.77529907226562, 48.523681640625, -2.472766876220703, 157.15093994140625, 24.002464294433594, 192.41448974609375, 79.65017700195312, 34.5565185546875, -0.40196990966796875, -24.62579345703125, -12.029800415039062, 9.210147857666016, 44.2647705078125, 147.66744995117188, 94.44686889648438, 7.82049560546875, 169.76992797851562, 188.6600341796875, -1.54449462890625, 88.62913513183594, 4.08152961730957, 0.5637588500976562, 28.1702880859375, 11.096672058105469, -126.75323486328125, -37.566200256347656, 32.86395263671875, 172.803955078125, -41.135650634765625, 137.75790405273438, 0.13962936401367188, 48.08184814453125, 172.12322998046875, 14.99468994140625, 170.85423278808594, 125.77958679199219, 67.68511962890625, 215.68130493164062, -1.9346923828125, 155.7670440673828, 129.38563537597656, 116.48104858398438, -159.2586212158203, 50.701568603515625, 189.57650756835938, 62.533233642578125, 96.4144287109375, -24.380294799804688, 131.46176147460938, 133.99832153320312, 5.7728118896484375, -18.011138916015625, -133.77923583984375, 158.91802978515625, 130.62110900878906, -0.7655029296875, -71.0162353515625, -44.516082763671875, -125.5865478515625, 3.9666748046875, 33.39592361450195, 16.485824584960938, -44.9112548828125, 115.55805206298828, 0.6078109741210938, 49.085540771484375, 137.85275268554688, 186.69223022460938, 145.76336669921875, 43.129638671875, 151.33099365234375, -3.025514602661133, -137.74667358398438, 168.16690063476562, -4.45855712890625, 10.113151550292969, 74.19216918945312, 166.11898803710938, 15.373222351074219, -6.2822265625, 17.306869506835938, 8.66558837890625, 18.7308349609375, 145.697998046875, -2.500946044921875, -42.90582275390625, -12.200912475585938, 0.0, -14.857860565185547, 12.105552673339844, -67.34268188476562, 2.8555755615234375, -7.03936767578125, -0.4484100341796875, 161.62960815429688, -54.516845703125, -51.28857421875, -144.74520874023438, -2.1001434326171875, 53.8790283203125, 161.87686157226562, -79.49472045898438, 179.89735412597656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000445.npy"}
{"epoch": 0.9319371727748691, "step": 446, "batch_size": 128, "mean": 61.36278533935547, "std": 89.7020263671875, "min": -200.15628051757812, "p10": -30.38847961425781, "median": 43.31239700317383, "p90": 174.72107543945313, "max": 232.12435913085938, "pos_frac": 0.734375, "sample": [101.28842163085938, 214.852783203125, 4.719779968261719, 4.27935791015625, -0.6246795654296875, 36.412261962890625, 138.9294891357422, 53.751686096191406, 148.13832092285156, -15.957015991210938, 57.57086181640625, -79.35661315917969, 118.09646606445312, 40.47692108154297, 8.178497314453125, 42.90330505371094, 204.12960815429688, 18.660110473632812, -15.662490844726562, 16.80474853515625, 12.963577270507812, 126.25857543945312, 175.45452880859375, 152.640380859375, -6.997533798217773, -48.94905090332031, 217.75665283203125, -1.6279449462890625, 9.716835021972656, -81.11129760742188, 12.710845947265625, 12.296409606933594, -30.139739990234375, 38.83545684814453, -54.955841064453125, 20.927352905273438, 161.25946044921875, 3.2508926391601562, 17.1246337890625, 43.72148895263672, 173.796630859375, 3.9448318481445312, 93.93065643310547, 19.768905639648438, 45.087493896484375, 144.06436157226562, 174.40673828125, 115.8116455078125, 163.33551025390625, -200.15628051757812, 84.25030517578125, 120.85684204101562, 197.30987548828125, 3.6170654296875, 133.3717803955078, 152.7814483642578, 36.35821533203125, 48.481201171875, -187.98095703125, 108.4310302734375, 172.47506713867188, -0.6412467956542969, -36.342681884765625, 0.0, 39.296966552734375, 157.51919555664062, 41.120628356933594, 106.86815643310547, 146.25830078125, 143.27955627441406, -11.70574951171875, 152.88433837890625, 123.02919006347656, 154.41046142578125, 148.686767578125, 11.747039794921875, 160.31192016601562, 132.64744567871094, -48.664947509765625, 0.0, 81.06576538085938, -22.715133666992188, 194.28805541992188, -8.2769775390625, 28.720504760742188, -82.27987670898438, 97.3814697265625, 207.56182861328125, 148.72705078125, 84.07501220703125, -30.9688720703125, 232.12435913085938, 195.21136474609375, 161.3372802734375, 140.49392700195312, 88.62399291992188, 102.1505126953125, 137.835693359375, -11.490795135498047, 96.54461669921875, 19.068622589111328, -12.587432861328125, 189.7884521484375, 190.60171508789062, -184.7088623046875, -137.16162109375, -11.589218139648438, 22.820068359375, -10.678192138671875, 2.13677978515625, 45.01795959472656, 0.0, -27.226211547851562, -12.62249755859375, 135.4713134765625, -43.3485107421875, 187.27700805664062, 41.423583984375, 205.03631591796875, 13.979934692382812, 6.223388671875, 55.60398864746094, -15.5286865234375, -10.647552490234375, 158.37353515625, 64.57831573486328, 140.95704650878906, 98.39187622070312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000446.npy"}
{"epoch": 0.9340314136125655, "step": 447, "batch_size": 128, "mean": 64.2353515625, "std": 81.35118865966797, "min": -183.69403076171875, "p10": -21.494393920898432, "median": 54.224552154541016, "p90": 165.4056884765625, "max": 213.75970458984375, "pos_frac": 0.765625, "sample": [-50.60191345214844, 10.741130828857422, 139.93560791015625, 138.66505432128906, 30.506088256835938, 8.495540618896484, 185.00112915039062, 11.562255859375, -33.21484375, 153.26718139648438, 74.3375244140625, 135.31394958496094, 44.8023681640625, 168.32611083984375, 157.64149475097656, 52.70362091064453, 0.0, -0.031646728515625, 179.08233642578125, 10.5885009765625, -13.423080444335938, 36.391448974609375, 146.31329345703125, 23.13134765625, 130.844482421875, -11.2047119140625, 1.555267333984375, 128.9946746826172, -133.25146484375, 134.1114501953125, -118.85174560546875, 25.19677734375, 126.76316833496094, 116.92947387695312, -16.213603973388672, 23.116668701171875, 157.4975128173828, 164.42156982421875, 23.435272216796875, 134.2815704345703, 3.5254364013671875, 150.52243041992188, 174.5466766357422, 169.44967651367188, 35.177154541015625, -2.146270751953125, 146.40847778320312, 135.85699462890625, 162.7727508544922, 0.0, 163.364013671875, 34.79493713378906, 64.426513671875, 130.8224334716797, -34.20555114746094, -37.93553161621094, -5.9649658203125, 4.56005859375, -109.45941925048828, 158.82472229003906, 12.364486694335938, 70.09003448486328, 0.602752685546875, 65.60362243652344, -20.281997680664062, 194.4166259765625, 139.295166015625, -2.22662353515625, -49.06407165527344, 15.891937255859375, 66.75234985351562, -30.118362426757812, -24.323318481445312, 9.1722412109375, 97.23831176757812, 0.8399887084960938, 75.80120849609375, -34.359954833984375, 63.4500732421875, 168.86276245117188, 212.00830078125, -3.27691650390625, -3.6046295166015625, 64.9422607421875, -183.69403076171875, 0.0, 6.933204650878906, 137.82269287109375, 11.847015380859375, 168.03997802734375, 50.51824951171875, 35.493194580078125, 31.028564453125, 129.22195434570312, 203.3958740234375, 99.758544921875, 153.13241577148438, 213.75970458984375, 0.0, 162.0762176513672, 110.89987182617188, 133.6171875, 27.79559326171875, 25.200225830078125, 35.512786865234375, 94.27020263671875, 30.6317138671875, -115.281982421875, 160.7013702392578, 78.09808349609375, 7.627777099609375, 134.02462768554688, 55.7454833984375, 165.77294921875, 123.3939208984375, 116.16856384277344, 13.677459716796875, 133.0575714111328, -8.69097900390625, -14.28438663482666, 58.940704345703125, 165.248291015625, 143.1204833984375, 102.93081665039062, 9.00775146484375, 178.25350952148438, -5.8492431640625, 110.62354278564453], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000447.npy"}
{"epoch": 0.9361256544502617, "step": 448, "batch_size": 128, "mean": 55.57023620605469, "std": 87.77481079101562, "min": -179.46368408203125, "p10": -27.642460632324216, "median": 35.72099304199219, "p90": 165.86783599853516, "max": 223.1385955810547, "pos_frac": 0.7421875, "sample": [16.607192993164062, 9.279953002929688, -127.87686157226562, 12.90570068359375, 76.81781005859375, 173.9146728515625, 6.13104248046875, -179.46368408203125, 110.23831176757812, 145.78579711914062, -14.67095947265625, -134.54791259765625, 52.123443603515625, 36.412445068359375, 3.9437255859375, 106.67201232910156, 139.4053955078125, 163.246337890625, 62.33343505859375, -7.884878158569336, 104.60022735595703, 173.18365478515625, 223.1385955810547, 5.598152160644531, 89.44696044921875, 26.689407348632812, 48.56793212890625, -22.771198272705078, 14.926116943359375, 9.42376708984375, -10.63079833984375, 34.85955810546875, 173.121826171875, -96.79783630371094, 140.058349609375, -42.93182373046875, 153.8218994140625, 166.71319580078125, 65.89266967773438, 116.08380126953125, 76.61947631835938, 24.22315216064453, 179.23486328125, 0.3846282958984375, 76.27900695800781, 5.30702018737793, 84.01438903808594, 109.60574340820312, -3.2742156982421875, 201.90234375, -12.1025390625, -6.24005126953125, -10.2998046875, 106.82687377929688, 3.7674636840820312, -13.405975341796875, 35.029541015625, 118.1944580078125, -72.33740234375, 175.67117309570312, 134.55784606933594, 67.4327392578125, 13.420135498046875, -0.5150527954101562, 163.82046508789062, 200.84420776367188, 30.372501373291016, 12.479103088378906, 156.96981811523438, 192.3104248046875, 156.04104614257812, 102.19236755371094, 117.25572204589844, 41.642608642578125, 10.647140502929688, 122.53814697265625, 171.86753845214844, 152.47105407714844, 145.4510040283203, 105.46926879882812, -18.726303100585938, 172.34625244140625, -2.399578094482422, 154.43089294433594, -5.8074798583984375, -141.84153747558594, -135.90969848632812, -148.212890625, 152.52078247070312, 15.809234619140625, -13.33209228515625, -13.624114990234375, 158.2563934326172, 157.0810089111328, 19.51959228515625, -131.388671875, 62.6204833984375, -1.7647838592529297, 16.74267578125, 149.55148315429688, 146.70303344726562, 14.568479537963867, 165.5055389404297, 8.105335235595703, 117.04766845703125, -27.130401611328125, -77.42131042480469, 95.16017150878906, 28.180267333984375, -28.837265014648438, 159.55198669433594, -25.22479248046875, -15.222991943359375, 10.82281494140625, -45.695335388183594, 22.44476318359375, 7.1805419921875, 17.90069580078125, 117.42904663085938, 0.4407997131347656, 125.21654510498047, -15.5667724609375, 7.834079742431641, 158.36148071289062, 220.1163330078125, 137.8375701904297, 65.9266357421875, 74.84797668457031], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000448.npy"}
{"epoch": 0.9382198952879581, "step": 449, "batch_size": 128, "mean": 63.368194580078125, "std": 86.59453582763672, "min": -178.8145751953125, "p10": -25.624473190307615, "median": 58.46906280517578, "p90": 174.28140258789062, "max": 208.6260986328125, "pos_frac": 0.7421875, "sample": [149.82485961914062, 102.11532592773438, -35.50550842285156, 119.6441650390625, 57.91229248046875, -13.42474365234375, 8.765228271484375, 186.35061645507812, -4.5988616943359375, 5.289821624755859, 16.73333740234375, 38.667205810546875, 137.57406616210938, 83.6756820678711, 5.8961944580078125, 138.89407348632812, 193.71133422851562, 169.64657592773438, 16.1845703125, -2.413545608520508, -106.45877075195312, 129.8190155029297, 186.88232421875, 122.2975845336914, 54.87664794921875, 100.96460723876953, -44.65118408203125, 44.9881591796875, 7.029065132141113, 173.62689208984375, 194.88446044921875, 167.75531005859375, 156.35693359375, 181.17340087890625, 25.492523193359375, 159.60458374023438, 19.59686279296875, 105.44438171386719, 124.10113525390625, 125.57843017578125, 169.46524047851562, 28.94415283203125, 129.5263671875, 14.646797180175781, 208.6260986328125, 198.67294311523438, 1.5440673828125, -4.580949783325195, 185.16253662109375, 153.6337127685547, -51.559059143066406, 63.404048919677734, 12.85931396484375, 23.936660766601562, 4.022672653198242, 142.21788024902344, 19.909530639648438, 128.46417236328125, 5.27337646484375, 92.79881286621094, -18.156402587890625, 43.728363037109375, 157.17694091796875, 18.396697998046875, 110.32410430908203, -4.373252868652344, 62.359283447265625, 60.199462890625, -8.59649658203125, 153.16574096679688, -58.52642822265625, 72.80482482910156, 158.21768188476562, 142.1856689453125, -178.8145751953125, -1.3805084228515625, 2.0748443603515625, 144.85269165039062, 59.02583312988281, 156.3822021484375, 143.14202880859375, 96.92301940917969, 206.04586791992188, 30.19378662109375, 120.73560333251953, 110.72822570800781, 157.18817138671875, 18.23028564453125, 125.06967163085938, -27.374027252197266, -5.180564880371094, 32.408233642578125, -2.860870361328125, 183.3585205078125, 166.56884765625, 175.80859375, -122.66387939453125, -10.790399551391602, 115.38372802734375, 63.34283447265625, -37.340087890625, 137.5103759765625, -23.79863739013672, -24.371063232421875, 89.83067321777344, -23.600738525390625, -56.113677978515625, -9.957168579101562, 146.73529052734375, 19.5931396484375, 6.078620910644531, 136.10409545898438, -24.874664306640625, -3.2294273376464844, 37.204864501953125, -8.278411865234375, 142.33929443359375, 103.0751724243164, -1.71783447265625, 179.27249145507812, 34.164581298828125, -158.22914123535156, 180.30300903320312, 21.042213439941406, 60.484161376953125, -157.78741455078125, -124.86001586914062, -7.02911376953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000449.npy"}
{"epoch": 0.9403141361256544, "step": 450, "batch_size": 128, "mean": 48.42830276489258, "std": 85.13880920410156, "min": -166.33737182617188, "p10": -29.225253295898437, "median": 26.19338607788086, "p90": 170.75942993164062, "max": 205.60940551757812, "pos_frac": 0.6953125, "sample": [0.0, -116.4962158203125, 176.56752014160156, -11.801490783691406, 65.49160766601562, -4.137916564941406, 147.91920471191406, -13.12689208984375, 171.11767578125, 50.9664306640625, 118.24298095703125, 181.16009521484375, -1.3303680419921875, 93.05621337890625, 150.61264038085938, 10.890811920166016, -18.452316284179688, 122.14093017578125, 14.794570922851562, 16.72035789489746, -8.871658325195312, 75.05421447753906, 182.89413452148438, 168.47828674316406, 122.97008514404297, 28.23773193359375, -14.821525573730469, 17.043914794921875, 91.34217834472656, 21.982025146484375, 17.124340057373047, -5.9494476318359375, -8.5146484375, 1.79583740234375, 56.27649688720703, 13.84259033203125, 200.2664794921875, -139.1448974609375, 85.54501342773438, -68.05026245117188, -2.265043258666992, 173.58123779296875, 24.123088836669922, 70.03948974609375, 16.152175903320312, 50.60778045654297, -25.41071891784668, 99.12686157226562, 182.88546752929688, 98.32968139648438, 1.2158203125, 160.5565185546875, 16.62075424194336, 10.452224731445312, 99.75848388671875, -6.7787628173828125, 35.750885009765625, 183.62034606933594, 10.553253173828125, -140.16229248046875, 46.289642333984375, -166.33737182617188, 146.97982788085938, 46.4853515625, -28.81622314453125, 178.7613525390625, 147.73675537109375, 3.787832260131836, 117.13720703125, 113.74652099609375, -0.6869049072265625, 121.22451782226562, -0.8992919921875, 165.77947998046875, 3.782318115234375, 24.991073608398438, 178.13934326171875, 134.97528076171875, -111.06242370605469, 17.107803344726562, 108.27484130859375, 44.5169677734375, -18.91485595703125, 2.429351806640625, 178.1190185546875, 33.46466064453125, -1.5410690307617188, 9.399749755859375, 94.34146118164062, -15.431289672851562, 19.3968505859375, 93.44184875488281, 170.60589599609375, 205.60940551757812, 25.831512451171875, 156.43075561523438, -35.973785400390625, -123.16094970703125, -30.179656982421875, -116.1500244140625, 15.10977554321289, 73.37057495117188, 116.79766845703125, 128.0747833251953, 188.5986328125, 41.078887939453125, -119.59324645996094, 103.03970336914062, 10.31097412109375, 157.72171020507812, 160.18955993652344, 6.5087738037109375, 26.555259704589844, -17.80377197265625, 162.02041625976562, -115.48394775390625, -3.52197265625, -13.83929443359375, 40.87677001953125, 97.69552612304688, 0.0, -12.728446960449219, 105.57305908203125, 41.181739807128906, -12.36224365234375, 151.4371337890625, -22.722579956054688, -99.48947143554688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000450.npy"}
{"epoch": 0.9424083769633508, "step": 451, "batch_size": 128, "mean": 53.56208038330078, "std": 86.14354705810547, "min": -167.6767578125, "p10": -33.7187286376953, "median": 29.932331085205078, "p90": 163.02423706054688, "max": 227.46914672851562, "pos_frac": 0.75, "sample": [62.415740966796875, 23.109825134277344, -23.8497314453125, 7.376625061035156, 192.8547821044922, 149.8489990234375, 58.45997619628906, 2.4134140014648438, -152.15353393554688, -27.8221435546875, 113.22322082519531, 7.435914993286133, 184.53466796875, 20.845428466796875, 25.44842529296875, 111.88421630859375, 118.6048583984375, 9.998626708984375, 175.93374633789062, -85.7603759765625, 155.51904296875, -5.762451171875, 10.85821533203125, 68.46755981445312, 200.98568725585938, -0.120697021484375, 227.46914672851562, 13.932830810546875, -15.134872436523438, 31.944793701171875, -65.34712219238281, -4.477073669433594, -167.46348571777344, 174.77899169921875, 61.99639892578125, 39.89225769042969, 22.663047790527344, 106.90304565429688, 38.29595947265625, 135.65066528320312, 17.492794036865234, 149.32199096679688, 133.98733520507812, 142.54296875, 126.7802505493164, 162.52015686035156, 14.404403686523438, -1.920928955078125, 188.00192260742188, -0.01318359375, 6.0218963623046875, 189.48736572265625, 22.342147827148438, 201.70635986328125, 142.0413818359375, 150.18896484375, -29.126564025878906, 63.780029296875, -167.6767578125, 130.91050720214844, 136.76846313476562, 89.22796630859375, 24.8524169921875, 161.0589599609375, 2.9873428344726562, 55.27391052246094, -82.44731903076172, 1.1133499145507812, 82.14067077636719, 111.20082092285156, 2.5252227783203125, 8.079345703125, -30.7135009765625, 154.60137939453125, 10.197906494140625, -48.2415771484375, 78.24685668945312, 9.7576904296875, 148.50137329101562, 44.1417236328125, 0.105499267578125, -2.7450180053710938, -30.8160400390625, 64.53009796142578, 163.28643798828125, -27.942550659179688, 15.674148559570312, 99.22555541992188, 20.004776000976562, 15.268821716308594, 124.46514892578125, 123.66039276123047, -22.7689208984375, 160.75782775878906, 20.579574584960938, -150.85382080078125, 27.91986846923828, -40.491668701171875, -15.161224365234375, 164.18789672851562, 46.13726806640625, 156.83419799804688, 63.918426513671875, 162.911865234375, -15.170623779296875, -96.56489562988281, -95.866943359375, -76.34500122070312, 97.39852905273438, 25.73388671875, 149.5458984375, 21.445777893066406, 171.55276489257812, 154.43045043945312, -5.119148254394531, 15.734962463378906, 89.88656616210938, 3.917327880859375, 68.71321105957031, 223.911376953125, 44.028018951416016, -2.0700531005859375, -22.690200805664062, 6.352821350097656, 132.3751220703125, 126.42633056640625, -42.854644775390625, 134.5651092529297], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000451.npy"}
{"epoch": 0.9445026178010472, "step": 452, "batch_size": 128, "mean": 52.03689956665039, "std": 83.453369140625, "min": -155.71038818359375, "p10": -29.168626403808588, "median": 33.171966552734375, "p90": 170.0419952392578, "max": 222.1640625, "pos_frac": 0.7578125, "sample": [-64.8765869140625, 142.98574829101562, 39.263671875, 205.5987548828125, -58.79301452636719, 7.747779846191406, 10.881843566894531, 176.73300170898438, -11.040390014648438, -25.234092712402344, 171.80252075195312, 0.627685546875, 5.68853759765625, 56.94916534423828, -1.929473876953125, 7.65716552734375, 65.62979125976562, -4.8442535400390625, -6.3780517578125, 16.8465576171875, -0.12091445922851562, -136.7188720703125, 103.61743927001953, 151.01016235351562, 135.78201293945312, 56.81719970703125, 54.081085205078125, 23.472320556640625, 0.47565460205078125, -102.93435668945312, -101.85781860351562, 187.95620727539062, -16.06375503540039, 194.88827514648438, 92.808349609375, -17.28631591796875, 175.08755493164062, 168.13375854492188, -15.241294860839844, 137.28219604492188, 93.5204086303711, 12.4263916015625, 6.53143310546875, 73.75677490234375, 137.77627563476562, 12.861227035522461, 38.77100372314453, 140.6016387939453, 100.41921997070312, 19.14653778076172, 150.058837890625, 35.6309814453125, 120.99264526367188, -10.237930297851562, 10.977684020996094, -52.13482666015625, 13.872024536132812, 18.156402587890625, 22.174530029296875, 22.226165771484375, 11.252616882324219, 41.931854248046875, -93.04165649414062, 93.28964233398438, -58.423675537109375, 144.05043029785156, 141.8638916015625, 166.25473022460938, 15.66070556640625, 14.5460205078125, 3.591400146484375, 169.8135986328125, 27.03199005126953, 15.44183349609375, 152.66441345214844, 222.1640625, 150.4416046142578, -16.84820556640625, 50.089813232421875, 51.921844482421875, -140.86044311523438, 105.66925048828125, 175.70384216308594, -4.576446533203125, 170.57492065429688, 149.46302795410156, -5.4111785888671875, 142.3766632080078, 183.00918579101562, 30.71295166015625, 40.36637878417969, 47.2889404296875, 18.846920013427734, -1.0885696411132812, 6.012367248535156, -36.58653259277344, 148.02955627441406, 46.14048767089844, 58.164215087890625, -17.6058349609375, 207.73226928710938, 51.0245361328125, 21.922637939453125, 133.71078491210938, 42.377113342285156, 2.446748733520508, -27.77191162109375, 18.705299377441406, 168.34146118164062, 164.4245147705078, 210.70635986328125, 171.8060302734375, 13.359512329101562, 37.43914794921875, 161.001220703125, 0.0, 76.7515869140625, -1.6809425354003906, -131.64529418945312, -32.42762756347656, 39.45635986328125, 51.226470947265625, -155.71038818359375, 48.3712158203125, 4.783966064453125, 164.07754516601562, 6.3218994140625, 0.012914657592773438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000452.npy"}
{"epoch": 0.9465968586387434, "step": 453, "batch_size": 128, "mean": 60.24262619018555, "std": 82.1954574584961, "min": -170.15814208984375, "p10": -26.75479431152343, "median": 50.432228088378906, "p90": 165.0129180908203, "max": 232.05572509765625, "pos_frac": 0.78125, "sample": [27.087806701660156, 141.62002563476562, 7.504730224609375, 112.82696533203125, 156.1962890625, 5.9704132080078125, 19.182464599609375, -3.2535552978515625, 146.98635864257812, 97.14846801757812, 70.9632339477539, 81.57142639160156, 81.32652282714844, 70.0836181640625, -18.93695068359375, -48.7630615234375, 79.08370971679688, -170.15814208984375, 24.384613037109375, 164.35354614257812, -24.45501708984375, 158.11410522460938, 162.51593017578125, 139.8048553466797, 67.23931884765625, -3.5223388671875, -48.531768798828125, 158.064697265625, 49.115325927734375, 156.65403747558594, 21.432174682617188, 133.3251190185547, 39.24517822265625, 134.2176513671875, 34.8780517578125, 31.253692626953125, 34.8753662109375, -2.790618896484375, 177.33753967285156, 178.37603759765625, 6.840667724609375, 166.55145263671875, 31.47503662109375, 96.60244750976562, 135.66534423828125, -110.59530639648438, 13.372589111328125, 20.21807861328125, 12.65020751953125, -68.2435302734375, 7.5496826171875, 146.40640258789062, 6.1375579833984375, 19.606002807617188, -155.589111328125, 80.05282592773438, -5.22418212890625, -50.223419189453125, 50.76448059082031, 12.41168212890625, 126.40968322753906, -7.226776123046875, 0.0, 126.96043395996094, 99.36836242675781, 22.824859619140625, 118.40172576904297, 10.918365478515625, 148.08639526367188, 25.34466552734375, 94.46438598632812, 70.69195556640625, 51.27032470703125, -150.4156494140625, 21.251129150390625, 168.04345703125, 188.016357421875, 159.81887817382812, -2.846872329711914, 145.23236083984375, -15.196853637695312, 183.35989379882812, 84.21914672851562, -8.87957763671875, 10.666900634765625, 22.79010009765625, 181.56430053710938, -19.946533203125, -32.120941162109375, 180.62255859375, 42.8096923828125, 88.15640258789062, -48.5462646484375, -35.145172119140625, 160.50711059570312, 20.912139892578125, -146.1204833984375, 154.02403259277344, -8.849044799804688, 232.05572509765625, 59.00526428222656, 46.297515869140625, 26.42707061767578, 76.62757110595703, 62.585357666015625, 62.47039794921875, 1.6678314208984375, 3.2195892333984375, 176.43063354492188, 61.35517883300781, 156.7859649658203, 126.44601440429688, 25.65252685546875, 150.02041625976562, 50.0999755859375, 124.39579772949219, 175.38894653320312, 178.39892578125, -9.778030395507812, 1.7850341796875, 145.8233642578125, 6.287986755371094, 124.44947052001953, -18.92755126953125, 151.8443145751953, -46.09063720703125, 64.45893859863281, 205.6748046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000453.npy"}
{"epoch": 0.9486910994764398, "step": 454, "batch_size": 128, "mean": 55.40950012207031, "std": 84.36637115478516, "min": -207.1859130859375, "p10": -21.824474334716793, "median": 24.696109771728516, "p90": 173.85077667236328, "max": 225.37881469726562, "pos_frac": 0.78125, "sample": [46.1578369140625, 162.2484893798828, -144.39395141601562, -20.87890625, -6.152984619140625, 10.07904052734375, 175.58392333984375, 143.12466430664062, 9.689697265625, 38.43656921386719, 21.018447875976562, 20.835283279418945, -10.82463264465332, 123.24250030517578, 23.294723510742188, 110.8195571899414, -136.0162353515625, 6.311103820800781, 26.5458984375, -2.8588428497314453, 118.09072875976562, 30.67369842529297, 21.079498291015625, 3.471221923828125, 206.08416748046875, 225.37881469726562, -1.37677001953125, 183.255859375, -207.1859130859375, 7.361480712890625, 19.203392028808594, 148.28089904785156, 71.9992446899414, 127.61677551269531, 160.1689910888672, 20.306480407714844, 178.00338745117188, -27.213272094726562, 17.02227783203125, 23.718032836914062, 159.556884765625, 16.237625122070312, 177.78155517578125, 135.19180297851562, 25.67418670654297, 144.29598999023438, 112.27487182617188, 1.9146728515625, 176.64138793945312, -12.8336181640625, 17.375167846679688, 58.7432861328125, 33.983917236328125, 0.760467529296875, 142.01248168945312, -16.27166748046875, 120.70559692382812, 4.3230743408203125, 160.38104248046875, 176.50704956054688, 123.49822998046875, 145.6753692626953, 182.6177978515625, 109.57392883300781, 187.01995849609375, 166.249755859375, 155.05308532714844, 22.2384033203125, 15.892868041992188, -0.511962890625, 178.11578369140625, 153.44924926757812, -1.808502197265625, -60.14802169799805, 11.12451171875, 32.82206726074219, -24.030799865722656, 68.11776733398438, 16.426223754882812, -159.5029296875, -20.469192504882812, 1.6492691040039062, 174.15908813476562, -40.5234375, -0.798797607421875, 58.00811767578125, 145.14657592773438, 104.027099609375, -24.57940673828125, 155.9537353515625, 78.3056640625, 13.649147033691406, 19.11126708984375, 173.71864318847656, 201.82083129882812, 2.92657470703125, 21.09931182861328, 8.378860473632812, 169.1110076904297, 2.950450897216797, -33.261566162109375, 12.31170654296875, 165.30984497070312, 39.24609375, 28.89837646484375, 26.41497802734375, 165.25033569335938, 38.59242248535156, -61.577178955078125, 144.01568603515625, 2.5321044921875, 11.965362548828125, 71.10009765625, 2.2976837158203125, 23.27044677734375, 133.64784240722656, 128.08819580078125, -3.311279296875, -60.26690673828125, 32.687477111816406, 15.785873413085938, -10.85382080078125, 30.372894287109375, 165.2047119140625, -14.820037841796875, -1.77294921875, 11.111038208007812, -92.79814147949219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000454.npy"}
{"epoch": 0.9507853403141361, "step": 455, "batch_size": 128, "mean": 46.65873718261719, "std": 83.4156494140625, "min": -174.9473876953125, "p10": -36.60500869750976, "median": 31.044464111328125, "p90": 157.12589569091796, "max": 221.64794921875, "pos_frac": 0.703125, "sample": [140.79241943359375, 78.80965423583984, 24.47314453125, -10.13690185546875, 126.38031005859375, 62.29345703125, 85.19215393066406, 161.93922424316406, 14.415847778320312, -21.12010955810547, -4.9615325927734375, 89.33428955078125, -4.995880126953125, 112.21668243408203, 54.49769973754883, 156.3839569091797, 166.5214385986328, 136.27084350585938, 210.70858764648438, -47.02557373046875, 8.7449951171875, 38.287109375, 10.404281616210938, 150.56878662109375, 221.64794921875, 158.56263732910156, 154.16390991210938, 15.95208740234375, 36.819400787353516, 5.83648681640625, -18.727264404296875, -166.54885864257812, -0.8428630828857422, 25.874755859375, 24.341064453125, 6.757415771484375, -10.48455810546875, -23.99334716796875, -46.32823944091797, 19.30279541015625, 31.626358032226562, -8.75677490234375, 129.10897827148438, 204.45437622070312, 15.473876953125, 39.969482421875, 143.06170654296875, -3.9697265625, 157.203857421875, 25.70196533203125, 12.143447875976562, 10.408538818359375, 31.647567749023438, -7.341697692871094, -5.2449188232421875, -0.66876220703125, 12.659515380859375, -147.39578247070312, -38.91779327392578, 0.5476627349853516, 21.447471618652344, -29.528701782226562, 6.53038215637207, -35.24943542480469, -28.971649169921875, -49.05028533935547, 92.48123168945312, 52.46134948730469, 139.14187622070312, 107.40216064453125, -144.54254150390625, 123.33487701416016, 162.03115844726562, 154.25466918945312, 42.5770263671875, -15.033782958984375, -15.9644775390625, 163.02386474609375, 157.0924835205078, 81.40013122558594, 57.097381591796875, 2.3920822143554688, 8.352752685546875, -98.59667205810547, 152.18612670898438, -1.1624221801757812, -157.6514434814453, 77.5246353149414, -5.171142578125, 25.838390350341797, -6.039306640625, 45.02630615234375, -0.17138671875, 34.45735168457031, 30.462570190429688, 124.36013793945312, 202.20220947265625, 201.09698486328125, 109.33914184570312, -35.61381530761719, -0.85736083984375, 6.257404327392578, 128.611572265625, 101.1300048828125, 5.063873291015625, -3.3440475463867188, 14.41436767578125, 147.99078369140625, 26.8741455078125, 63.64171600341797, -100.08549499511719, 143.33917236328125, 80.331787109375, 70.53060913085938, 151.33187866210938, -93.84375, -174.9473876953125, 47.0633544921875, -58.74700927734375, 150.8000946044922, 139.70339965820312, 63.676361083984375, 173.08151245117188, 51.54283142089844, 94.83404541015625, 51.27178955078125, 165.4870147705078, 32.3582763671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000455.npy"}
{"epoch": 0.9528795811518325, "step": 456, "batch_size": 128, "mean": 67.49485778808594, "std": 94.82898712158203, "min": -176.32989501953125, "p10": -44.522900390625, "median": 39.68766784667969, "p90": 178.52467803955076, "max": 312.10498046875, "pos_frac": 0.7265625, "sample": [39.279052734375, -8.178218841552734, -49.698089599609375, 153.71978759765625, 33.55760955810547, -7.433723449707031, 150.77378845214844, -56.885772705078125, 177.3643798828125, 109.408203125, 168.11859130859375, -176.32989501953125, 19.607498168945312, -0.5574951171875, -28.07763671875, 190.78573608398438, 129.095703125, -46.273712158203125, 7.57586669921875, 19.495086669921875, 170.24237060546875, 96.29525756835938, 4.5570068359375, 159.67376708984375, 1.125472068786621, -18.808032989501953, 100.86279296875, 65.292236328125, 153.8319854736328, -67.87017822265625, 140.31106567382812, -126.79736328125, 138.9358673095703, -43.772552490234375, 88.509033203125, 171.96392822265625, 104.05316162109375, 143.66995239257812, 108.52323913574219, 127.21017456054688, 211.71405029296875, -19.414306640625, 178.16619873046875, 22.69336700439453, 181.38446044921875, 151.89605712890625, 193.12799072265625, -3.825775146484375, 11.239990234375, 154.005859375, 38.5364990234375, 77.13725280761719, -7.47589111328125, 37.724021911621094, -51.5875244140625, 15.974365234375, 3.219970703125, 18.364398956298828, 26.79425048828125, 200.022216796875, 12.16046142578125, -6.492219924926758, 16.39775848388672, -112.90491485595703, -18.380767822265625, 173.468994140625, 143.26976013183594, -135.6247100830078, 30.01007080078125, 21.2823486328125, 13.482269287109375, 129.34828186035156, -5.116558074951172, 131.62356567382812, 40.096282958984375, 127.70266723632812, 204.710205078125, 130.24053955078125, -46.86761474609375, 63.20831298828125, 141.45213317871094, 154.9014892578125, -137.23635864257812, 178.01565551757812, 8.82421875, -7.117034912109375, 20.444534301757812, 195.3463134765625, 177.29977416992188, 186.8297576904297, 128.34066772460938, 193.61148071289062, -3.1799697875976562, 174.6653289794922, 0.0, 3.718149185180664, 120.00550842285156, 170.3856964111328, 131.5994873046875, 120.53962707519531, 15.06048583984375, 10.882080078125, 125.5987548828125, -1.1678314208984375, 98.23318481445312, -124.39044189453125, 151.11416625976562, 3.677642822265625, 10.94036865234375, -15.52655029296875, -1.357513427734375, 118.39222717285156, 112.37698364257812, 124.62542724609375, 179.3611297607422, -0.4203681945800781, 258.7738037109375, 36.566619873046875, 312.10498046875, 212.09957885742188, 167.42071533203125, 30.979034423828125, -11.95318603515625, -3.583892822265625, 163.8519744873047, -36.02398681640625, -122.36474609375, 171.1566162109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000456.npy"}
{"epoch": 0.9549738219895288, "step": 457, "batch_size": 128, "mean": 59.99661636352539, "std": 89.77043914794922, "min": -191.87173461914062, "p10": -37.72562255859374, "median": 39.866180419921875, "p90": 174.33487243652343, "max": 210.6656951904297, "pos_frac": 0.71875, "sample": [177.54978942871094, 32.723663330078125, -63.079978942871094, -6.0249481201171875, 142.70343017578125, 40.0694580078125, 157.0591583251953, 169.01104736328125, 187.13882446289062, 15.320632934570312, 156.74310302734375, -0.7333297729492188, 188.61984252929688, 114.182373046875, 6.83587646484375, 193.56027221679688, -10.855560302734375, 31.990386962890625, -125.39229583740234, 171.99017333984375, 146.1558837890625, 149.8203582763672, 110.09355163574219, 207.9906768798828, 134.02230834960938, 48.493438720703125, 167.95675659179688, 13.019638061523438, 168.59783935546875, 18.551971435546875, 80.88116455078125, -8.253692626953125, -2.741485595703125, -35.79510498046875, 144.68875122070312, 166.03640747070312, -27.969749450683594, 41.2498779296875, -97.56802368164062, 112.12506103515625, -21.3502197265625, 90.11602020263672, -28.791275024414062, 0.67791748046875, 5.959632873535156, -191.87173461914062, 185.67620849609375, 26.678436279296875, 178.29690551757812, -9.531829833984375, -44.0185546875, 70.24554443359375, 114.32263946533203, -52.202392578125, 135.20103454589844, -4.647680282592773, -150.18035888671875, 146.1923828125, 43.00408935546875, 25.12054443359375, -19.3170166015625, -19.919822692871094, 20.426605224609375, -25.83307647705078, 29.144500732421875, 64.92086791992188, 128.949951171875, 23.87676239013672, 210.6656951904297, -11.030487060546875, 23.522140502929688, 160.30230712890625, 140.56973266601562, 39.66290283203125, 21.110137939453125, -144.35560607910156, -45.572601318359375, 35.057708740234375, 23.205902099609375, 96.34449768066406, 22.96142578125, 163.57012939453125, 120.21867370605469, 111.8328857421875, 108.95233154296875, -23.19390869140625, 121.8138427734375, -2.11932373046875, 153.11181640625, 195.49526977539062, 7.550142288208008, -16.38214111328125, 163.435791015625, -127.498779296875, -46.34686279296875, 162.88433837890625, -11.81072998046875, 119.24285888671875, -18.879730224609375, 171.44415283203125, 142.36178588867188, 27.15216064453125, 8.671722412109375, 47.527008056640625, -42.23016357421875, 189.92474365234375, 31.875, 108.5214614868164, 12.966995239257812, 116.70494079589844, 177.19500732421875, 83.17316436767578, 3.72589111328125, 177.2036590576172, 163.69003295898438, 136.59963989257812, 4.481239318847656, 21.186914443969727, -137.93850708007812, 96.68470764160156, 184.23617553710938, 1.160776138305664, 0.0, 173.10910034179688, -1.8337020874023438, 146.53866577148438, 51.25714111328125, -8.056365966796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000457.npy"}
{"epoch": 0.9570680628272251, "step": 458, "batch_size": 128, "mean": 53.458011627197266, "std": 88.09800720214844, "min": -173.86154174804688, "p10": -30.730648803710938, "median": 23.533554077148438, "p90": 175.8804168701172, "max": 225.79896545410156, "pos_frac": 0.7578125, "sample": [13.654098510742188, -47.5653076171875, 11.492889404296875, -152.61566162109375, 191.94677734375, 15.738273620605469, -42.617279052734375, 171.15191650390625, -9.71160888671875, -42.905181884765625, 19.04815673828125, -85.25074768066406, -44.92303466796875, 0.940460205078125, 2.207341194152832, 112.73463439941406, 30.433990478515625, 65.41818237304688, -21.0574951171875, 55.106109619140625, 179.50009155273438, -1.5218582153320312, 91.85198974609375, -3.61395263671875, 23.55194091796875, 4.488176345825195, 159.57843017578125, 177.21951293945312, -10.15576171875, 142.13400268554688, 21.67413330078125, -13.50726318359375, 133.7583770751953, 204.87515258789062, 39.006256103515625, 18.837188720703125, 15.790191650390625, 45.786590576171875, 170.76248168945312, -29.722976684570312, 175.3065185546875, -3.3040599822998047, -138.22808837890625, 23.515167236328125, 8.3798828125, 5.485595703125, 4.42041015625, 12.014419555664062, 7.195369720458984, 43.38829040527344, 4.755592346191406, -38.28663635253906, -22.618759155273438, 163.4049072265625, 142.9569091796875, -30.752288818359375, 1.96807861328125, 1.5709457397460938, 126.86527252197266, 196.7867431640625, 35.80438232421875, 2.575127601623535, 217.5130615234375, 59.60040283203125, 121.16464233398438, 4.380157470703125, 19.97259521484375, -116.61495208740234, -26.578125, 193.259521484375, 157.61495971679688, 52.239990234375, -30.72137451171875, 29.06976318359375, -73.75775146484375, 151.53038024902344, 109.47856140136719, -19.12567138671875, 9.342262268066406, 136.021484375, 154.91439819335938, -26.26641845703125, -9.746383666992188, 28.054443359375, 195.74935913085938, 27.047012329101562, 225.79896545410156, -1.5806884765625, 37.56951904296875, 148.25830078125, 32.518310546875, 191.67715454101562, 141.5752716064453, 131.3513946533203, 205.44937133789062, -3.8558273315429688, 135.87664794921875, -5.6087646484375, 5.695568084716797, 128.4368896484375, 43.197113037109375, 13.277587890625, 7.91259765625, 162.23031616210938, 122.17105102539062, 75.01641082763672, 8.559989929199219, 6.4803924560546875, 154.5393524169922, 171.1913604736328, 40.715728759765625, 191.4231414794922, 164.92662048339844, 125.87261962890625, 159.09564208984375, 17.396484375, 158.4859619140625, 2.1281280517578125, -169.00457763671875, -173.86154174804688, 134.4430694580078, 178.77301025390625, 17.409561157226562, 5.2210693359375, 132.3043670654297, 32.746612548828125, 5.523582458496094, -19.577491760253906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000458.npy"}
{"epoch": 0.9591623036649215, "step": 459, "batch_size": 128, "mean": 41.245330810546875, "std": 91.57390594482422, "min": -182.79913330078125, "p10": -59.814389038085935, "median": 18.68212890625, "p90": 170.75897827148438, "max": 221.1358642578125, "pos_frac": 0.6796875, "sample": [4.4969024658203125, 3.1561126708984375, 14.544790267944336, 149.9592742919922, 0.50225830078125, 148.06417846679688, -140.06576538085938, 42.413818359375, 20.62701416015625, 29.53948974609375, 57.740264892578125, 111.00041198730469, 58.40055465698242, -15.729949951171875, 14.169792175292969, 6.197399139404297, 174.83523559570312, 97.53054809570312, -17.713590621948242, 40.72857666015625, 72.4169921875, 19.1455078125, -58.316741943359375, 30.092437744140625, 28.070465087890625, -11.539932250976562, 142.8574981689453, 1.7658004760742188, 150.77630615234375, -22.46912384033203, -17.834482192993164, 0.0, 3.9673004150390625, -37.55523681640625, 159.07855224609375, -119.46377563476562, 77.8591079711914, -18.817462921142578, -15.17578125, 61.87786865234375, 164.57843017578125, -14.031280517578125, 18.1865234375, 64.2310791015625, -146.57806396484375, 79.31218719482422, -22.129638671875, 25.552490234375, -15.27542495727539, 180.2735595703125, 4.323951721191406, 14.064443588256836, -3.2283554077148438, 209.11264038085938, 17.497467041015625, -1.0132064819335938, 45.04582214355469, -147.41049194335938, 165.19374084472656, 221.1358642578125, 128.2342529296875, -126.01943969726562, 172.0108642578125, 136.44717407226562, 150.165283203125, 189.21963500976562, 83.6136474609375, 139.10406494140625, -148.85086059570312, -151.91864013671875, 12.8712158203125, 42.9794921875, 152.13182067871094, 154.86679077148438, -16.112442016601562, -168.17938232421875, 188.693115234375, 111.97649383544922, 8.8875732421875, -63.30889892578125, -182.79913330078125, -4.3929443359375, -5.19305419921875, -110.279052734375, 170.32452392578125, 82.0343017578125, -18.1768798828125, 171.772705078125, 0.0, 11.154441833496094, 141.9103546142578, 0.17730712890625, 23.89471435546875, 205.47100830078125, -103.75830078125, 186.82220458984375, -5.64410400390625, 131.1865234375, 15.304397583007812, 33.90418243408203, -63.868408203125, 134.82437133789062, 17.07684326171875, -14.749008178710938, -11.416671752929688, 69.80379486083984, 165.57669067382812, 14.3814697265625, 5.0638427734375, -6.3250732421875, -43.234832763671875, 20.262847900390625, 19.6273193359375, -41.435791015625, 156.18634033203125, 184.28961181640625, -0.307861328125, 16.156753540039062, 167.137939453125, 29.561126708984375, -5.068046569824219, 180.73464965820312, 196.61196899414062, 18.21875, 37.17071533203125, 6.25732421875, 53.316253662109375, 127.04995727539062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000459.npy"}
{"epoch": 0.9612565445026178, "step": 460, "batch_size": 128, "mean": 41.2523193359375, "std": 93.97315979003906, "min": -188.951416015625, "p10": -86.8520751953125, "median": 26.0264892578125, "p90": 159.2156234741211, "max": 226.2147216796875, "pos_frac": 0.703125, "sample": [12.320419311523438, 82.48068237304688, -120.89242553710938, 148.8907470703125, 56.708465576171875, 155.94955444335938, -156.96270751953125, 125.68875885009766, -44.9072265625, 180.62298583984375, -122.47344970703125, 16.193328857421875, 14.67144775390625, 13.385498046875, 29.983074188232422, -8.7762451171875, -91.62149047851562, 12.738555908203125, 159.4025115966797, 182.12588500976562, -22.1195068359375, 29.4224853515625, 56.7701416015625, 118.26345825195312, 41.8492431640625, 9.478240966796875, 119.03659057617188, -7.341094970703125, 195.62030029296875, -28.328857421875, -14.46533203125, 124.61677551269531, 54.9271240234375, 226.2147216796875, 107.52206420898438, -59.94194030761719, 123.00845336914062, -146.97283935546875, 217.97775268554688, -19.9073486328125, -169.7383575439453, 159.13552856445312, -92.81683349609375, 1.8632640838623047, 89.50656127929688, 142.89349365234375, -36.1253662109375, 137.09886169433594, 7.169036865234375, -81.93450927734375, 91.81784057617188, -0.9672584533691406, 0.7801685333251953, -0.15242767333984375, -35.10900115966797, 20.101974487304688, -7.6056060791015625, -2.55853271484375, -2.7257823944091797, 27.835052490234375, 128.72396850585938, 73.37457275390625, 142.68215942382812, 10.330284118652344, 19.941970825195312, -89.03970336914062, 0.27422332763671875, 51.95880126953125, 154.71591186523438, -0.09420013427734375, 8.4365234375, 70.03717041015625, 24.5072021484375, 19.99517822265625, 101.43829345703125, 120.0927734375, 153.81573486328125, 146.58465576171875, 27.43096923828125, 30.903377532958984, 186.46640014648438, 28.225021362304688, -10.97021484375, 213.5772705078125, 50.4075927734375, 219.28811645507812, 174.38397216796875, 160.14691162109375, -134.60227966308594, 172.88250732421875, 123.21514892578125, 127.34872436523438, -128.0740966796875, 7.7650604248046875, -43.3079833984375, 9.36865234375, 150.38514709472656, -188.951416015625, 119.68917846679688, -52.178741455078125, 61.830963134765625, -139.0594482421875, 145.43072509765625, 60.472015380859375, 15.991973876953125, -85.91452026367188, -9.833564758300781, -68.14842224121094, 158.47369384765625, 37.70025634765625, 24.62200927734375, 140.20065307617188, 33.708709716796875, 155.4205322265625, -113.755126953125, 3.513956069946289, 5.456291198730469, 139.8254852294922, 10.110748291015625, 155.25216674804688, 163.99842834472656, 10.723426818847656, 41.9075927734375, -84.8973388671875, -49.3565673828125, 5.106819152832031, 22.742935180664062, 37.96995544433594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000460.npy"}
{"epoch": 0.9633507853403142, "step": 461, "batch_size": 128, "mean": 54.18479919433594, "std": 92.51085662841797, "min": -180.932861328125, "p10": -43.67797470092773, "median": 37.59489440917969, "p90": 177.0810806274414, "max": 217.890380859375, "pos_frac": 0.703125, "sample": [17.480690002441406, -147.946044921875, 178.77297973632812, 20.61431121826172, 50.145111083984375, 145.45089721679688, 150.86578369140625, 23.6871337890625, -59.99369430541992, 179.87088012695312, 149.55145263671875, 0.0, 14.79254150390625, 217.890380859375, 128.75450134277344, 136.01431274414062, 137.13909912109375, 53.610748291015625, -3.882232666015625, 10.05426025390625, 35.83262634277344, 176.86520385742188, 103.12584686279297, -86.60433959960938, 69.84799194335938, -10.24444580078125, 13.76611328125, 189.39077758789062, -37.91902160644531, 147.9947509765625, -0.6432342529296875, 192.88845825195312, 55.26373291015625, 145.39942932128906, -37.57505798339844, 150.02584838867188, 188.60348510742188, 17.63134765625, -157.134033203125, -40.240753173828125, 178.21746826171875, -150.5667724609375, 2.091796875, 16.877685546875, 121.0460205078125, 3.1673202514648438, 120.64008331298828, 133.8612060546875, 177.5847930908203, -5.702667236328125, 136.0752716064453, -42.50373077392578, 23.697662353515625, -2.8897552490234375, 10.662445068359375, -13.082595825195312, 138.7413330078125, -177.1796875, -46.417877197265625, 105.41213989257812, 128.4656982421875, 1.947509765625, -13.514846801757812, 136.40310668945312, 183.32589721679688, 23.231719970703125, -180.932861328125, 2.8335113525390625, 170.2771453857422, 8.317596435546875, 5.141578674316406, 158.31028747558594, -0.48151397705078125, 166.92913818359375, -13.466339111328125, 191.98150634765625, 81.55596923828125, 148.33740234375, -58.441253662109375, -34.781524658203125, 148.2061767578125, 5.70037841796875, 133.45387268066406, 85.37948608398438, 153.30984497070312, 55.378578186035156, -21.4940185546875, -95.49737548828125, -7.907686233520508, 112.98583984375, 87.50094604492188, 120.66371154785156, 12.46661376953125, 137.16879272460938, 84.99005126953125, 186.5352325439453, -3.706390380859375, 38.302520751953125, 150.852294921875, -138.40240478515625, 98.5472183227539, 0.9878158569335938, 180.69544982910156, 20.06207275390625, -20.47625732421875, -15.7493896484375, 29.731185913085938, 148.703125, 152.34890747070312, -10.393348693847656, -37.58882141113281, 108.29255676269531, -2.164600372314453, 133.2239990234375, 4.1280670166015625, 2.0362319946289062, 70.60127258300781, 89.01167297363281, -145.48757934570312, 42.09064483642578, 36.88726806640625, -5.9380645751953125, -76.8240966796875, 183.9783935546875, -18.42098045349121, 64.883056640625, 133.73159790039062, 168.55047607421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000461.npy"}
{"epoch": 0.9654450261780104, "step": 462, "batch_size": 128, "mean": 68.39488220214844, "std": 95.40206909179688, "min": -175.595947265625, "p10": -32.15923233032226, "median": 68.57703399658203, "p90": 181.86306457519532, "max": 230.221923828125, "pos_frac": 0.7578125, "sample": [140.29757690429688, 57.4560546875, 5.9003753662109375, 67.4281005859375, 171.88442993164062, 40.107574462890625, 151.07505798339844, 107.39263916015625, 93.41949462890625, 185.55770874023438, -142.38369750976562, 138.0728759765625, 186.85174560546875, 20.2991943359375, 182.17437744140625, 8.475860595703125, 107.98799133300781, 139.244384765625, 95.03347778320312, 13.3143310546875, 55.17205810546875, 139.66018676757812, 134.03411865234375, 46.64501953125, 56.1696662902832, -4.16009521484375, 169.66453552246094, -1.2741050720214844, 151.0517578125, -113.88514709472656, -140.03713989257812, 75.10601806640625, -5.20269775390625, 10.914825439453125, -4.340362548828125, 151.71234130859375, 183.0277862548828, -109.28756713867188, -9.873046875, 13.08941650390625, 142.546142578125, 153.34243774414062, 163.36822509765625, 7.2396697998046875, 221.2900390625, 110.21831512451172, 138.0882110595703, 172.77890014648438, 120.68841552734375, 3.513153076171875, 181.72964477539062, 7.8929901123046875, 143.05606079101562, -135.91445922851562, 13.246253967285156, -14.479690551757812, 178.27267456054688, 40.80450439453125, 9.074165344238281, 16.7193603515625, 148.57901000976562, 203.3341064453125, 35.50825119018555, 160.0594024658203, -18.37652587890625, 98.17748260498047, -130.62078857421875, 122.7933349609375, 230.221923828125, -23.780349731445312, 160.12890625, 166.21804809570312, 155.4475555419922, 133.65234375, 137.24679565429688, 90.2314453125, -24.52740478515625, 25.243011474609375, 9.525114059448242, -28.445159912109375, 13.622814178466797, -27.4537353515625, -1.038482666015625, 2.41131591796875, -154.72174072265625, 69.72596740722656, 145.2147674560547, 220.49801635742188, -97.65742492675781, 118.72273254394531, -66.78256225585938, 193.38302612304688, 62.700286865234375, 44.473114013671875, 117.95274353027344, 228.68695068359375, -42.12708282470703, 109.5870361328125, 130.38076782226562, 158.80667114257812, 116.66107940673828, -5.804718017578125, 225.69851684570312, 33.442626953125, -17.488433837890625, 18.167327880859375, 141.18878173828125, -14.826812744140625, 35.24462890625, 12.87762451171875, 151.4832763671875, 6.0790557861328125, 62.49151611328125, 175.82098388671875, -142.26101684570312, 204.7244873046875, -30.644851684570312, 121.30615234375, 137.43373107910156, -19.326217651367188, 186.08261108398438, 30.71014404296875, -1.66424560546875, -35.692787170410156, 124.0462646484375, 137.8140869140625, 158.32237243652344, -175.595947265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000462.npy"}
{"epoch": 0.9675392670157068, "step": 463, "batch_size": 128, "mean": 48.06439971923828, "std": 83.73226165771484, "min": -164.88775634765625, "p10": -31.182516479492186, "median": 26.243423461914062, "p90": 166.8438507080078, "max": 210.89453125, "pos_frac": 0.71875, "sample": [113.63327026367188, 51.79966735839844, 44.331764221191406, 2.5304946899414062, 14.75091552734375, -2.5606746673583984, 187.1422119140625, 10.24560546875, -161.9346923828125, 80.38693237304688, 157.12310791015625, 0.234710693359375, -149.0861053466797, -40.59459686279297, 4.2945556640625, 175.78607177734375, 39.347747802734375, 121.20040893554688, 15.28961181640625, -1.7350273132324219, 18.15087890625, -164.88775634765625, -20.47540283203125, 5.360908508300781, 66.0508804321289, 9.36102294921875, -88.71844482421875, 37.24946594238281, 171.60751342773438, 173.4332275390625, -150.10948181152344, 110.12691497802734, 9.087844848632812, -30.883148193359375, 7.61822509765625, 182.83319091796875, 168.24169921875, 154.22874450683594, 210.89453125, -49.213134765625, 60.14532470703125, -20.432281494140625, 103.40560913085938, 198.727783203125, 77.05546569824219, 6.92474365234375, 187.304931640625, -1.5977325439453125, 27.8460693359375, 4.038810729980469, 15.515022277832031, 30.91436004638672, 167.71865844726562, 59.23486328125, 8.105300903320312, 20.022430419921875, 178.52041625976562, -23.602325439453125, 157.64324951171875, 132.97335815429688, 85.997802734375, 158.9610595703125, 6.275787353515625, -10.260574340820312, -25.9005126953125, 19.899185180664062, 166.46893310546875, -7.286376953125, 64.52774047851562, 53.4447021484375, 155.902099609375, -31.88104248046875, -6.37310791015625, 64.7197265625, 50.341583251953125, -9.1007080078125, 121.06637573242188, -26.935806274414062, -23.447044372558594, 46.721466064453125, 150.3266143798828, 13.04107666015625, 99.04083251953125, 123.89718627929688, 75.36294555664062, 6.462974548339844, 22.283588409423828, 140.5589599609375, 73.74139404296875, -18.786043167114258, -15.717315673828125, 113.98974609375, -35.80842590332031, 0.0, -8.4755859375, 3.560274124145508, 6.812068939208984, 183.514404296875, 29.003936767578125, -13.642204284667969, 89.40693664550781, 124.35214233398438, -14.914199829101562, 2.6073951721191406, 16.220054626464844, -104.2293701171875, 24.640777587890625, 11.543701171875, 115.81915283203125, -0.9932708740234375, 9.498077392578125, -65.4163818359375, -26.95458984375, 106.32059478759766, -1.55242919921875, 152.18386840820312, 156.85324096679688, 50.952877044677734, 158.04861450195312, 120.94091796875, 199.2459716796875, 39.216705322265625, 79.709716796875, 142.64974975585938, 125.2635498046875, 157.45587158203125, -157.99639892578125, -33.54541015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000463.npy"}
{"epoch": 0.9696335078534032, "step": 464, "batch_size": 128, "mean": 52.34285354614258, "std": 84.72042083740234, "min": -198.1204833984375, "p10": -33.3259033203125, "median": 31.13860511779785, "p90": 162.9767807006836, "max": 201.0535888671875, "pos_frac": 0.7421875, "sample": [0.0, 95.93058776855469, 31.477619171142578, 147.03506469726562, -10.953353881835938, 129.88116455078125, 141.222900390625, -31.964752197265625, 166.00958251953125, -0.051025390625, 57.083099365234375, 71.99957275390625, 141.65565490722656, 22.785125732421875, 130.83995056152344, 26.478750228881836, 157.79043579101562, -1.145263671875, 149.65826416015625, 116.50288391113281, 46.1883544921875, 153.29550170898438, 9.602807998657227, 17.843048095703125, 55.199249267578125, 154.34805297851562, 44.290184020996094, -148.16226196289062, 201.0535888671875, 16.755294799804688, -0.808197021484375, -15.735260009765625, 24.349319458007812, 187.04258728027344, 173.04913330078125, 168.03616333007812, 100.4642333984375, 5.609458923339844, -11.1407470703125, 113.91477966308594, 6.444091796875, 25.6455078125, 168.123291015625, 93.99551391601562, 121.00732421875, 0.51806640625, -11.86590576171875, 17.604736328125, 1.773712158203125, 87.51588439941406, 14.409500122070312, 22.24847412109375, -0.2047576904296875, 136.96701049804688, 16.89242172241211, 100.15716552734375, 142.61480712890625, 65.22119140625, -153.8201904296875, 116.25802612304688, -2.714996337890625, 145.35476684570312, -7.458457946777344, 152.50244140625, 180.1663818359375, -8.2669677734375, 110.09249114990234, 26.666000366210938, 31.81219482421875, 105.11907958984375, -101.91033935546875, -198.1204833984375, -113.71253967285156, 3.1342849731445312, 28.314910888671875, 16.25164794921875, -16.76251220703125, -49.701904296875, 117.52252197265625, 33.229042053222656, 162.8374786376953, 189.41583251953125, 154.6431884765625, 183.289794921875, -39.11419677734375, 163.30181884765625, 1.5188941955566406, -36.501922607421875, 9.5899658203125, 3.439040184020996, 41.841278076171875, -102.68429565429688, 159.39724731445312, -6.43951416015625, -63.135528564453125, -1.3409957885742188, -1.5720596313476562, 138.2301025390625, 0.08843994140625, 10.24223518371582, 131.54833984375, 171.5597686767578, 26.98790740966797, 167.39181518554688, 166.51275634765625, -36.645660400390625, 109.10977172851562, 0.685394287109375, 139.642333984375, -116.27426147460938, 14.731414794921875, 9.031959533691406, -15.512371063232422, 72.89920043945312, 154.0992431640625, 47.78769302368164, 126.86192321777344, 145.92086791992188, 136.75331115722656, -175.806640625, 18.628936767578125, 31.8931884765625, -23.293792724609375, 97.24934387207031, 16.135379791259766, -24.370834350585938, 48.0572509765625, 30.799591064453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000464.npy"}
{"epoch": 0.9717277486910995, "step": 465, "batch_size": 128, "mean": 67.17460632324219, "std": 90.44564056396484, "min": -177.541259765625, "p10": -27.609762573242183, "median": 49.013389587402344, "p90": 178.75726470947265, "max": 214.56051635742188, "pos_frac": 0.7265625, "sample": [-2.7696075439453125, -8.550048828125, 10.406829833984375, 59.99348449707031, 163.718505859375, -2.5570316314697266, -26.4273681640625, 183.57943725585938, 10.299072265625, 27.621246337890625, 3.7273406982421875, 10.58612060546875, -5.90435791015625, 8.972503662109375, -123.947998046875, -37.107666015625, 178.30029296875, 157.01644897460938, 194.26080322265625, 173.9261474609375, 154.73797607421875, -37.40350341796875, 41.35618591308594, 160.6048583984375, 33.46783447265625, -0.7282524108886719, 3.5714111328125, 164.52261352539062, 177.36758422851562, 140.71304321289062, -2.133331298828125, 143.8998260498047, 34.10760498046875, 125.72512817382812, -4.3506011962890625, -10.1466064453125, -36.26446533203125, 180.23330688476562, 138.30043029785156, 212.77786254882812, -1.622955322265625, 156.8812255859375, -1.43780517578125, -30.368682861328125, -44.593475341796875, 34.98382568359375, -6.080657958984375, 51.00291442871094, -24.98443603515625, 133.32627868652344, 133.127197265625, 87.57229614257812, 105.80044555664062, 135.34706115722656, 104.88911437988281, -100.23656463623047, 200.62567138671875, 38.909088134765625, -9.397613525390625, 14.097930908203125, 22.42242431640625, 56.54505920410156, 44.058685302734375, 114.772216796875, 169.48687744140625, 22.704910278320312, 186.220947265625, 206.83123779296875, -11.92742919921875, 22.278921127319336, 89.47763061523438, 156.32891845703125, 149.52915954589844, 135.45933532714844, 117.64519500732422, -81.03857421875, -6.25164794921875, 46.962066650390625, 27.490585327148438, 101.15625762939453, 61.80059814453125, 9.298980712890625, 4.343017578125, 12.95367431640625, 5.720741271972656, 137.05340576171875, 187.54678344726562, -92.3292236328125, 55.47637176513672, -15.194854736328125, 176.92572021484375, 199.92431640625, 214.56051635742188, 91.74494171142578, 133.99969482421875, 91.66900634765625, 188.673095703125, 169.4296875, 167.75250244140625, 4.252449035644531, 167.8123779296875, 164.79977416992188, 188.97158813476562, 170.45941162109375, 10.212020874023438, 144.34536743164062, -17.843658447265625, 166.97418212890625, -148.15216064453125, 47.02386474609375, -89.13278198242188, 1.416229248046875, 170.03713989257812, 164.661865234375, 156.17063903808594, 179.8235321044922, -137.571533203125, 147.502197265625, 40.281341552734375, 10.474065780639648, -21.036941528320312, 112.714111328125, -15.294921875, -17.256179809570312, 160.5013427734375, -177.541259765625, -10.5718994140625, 79.47161865234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000465.npy"}
{"epoch": 0.9738219895287958, "step": 466, "batch_size": 128, "mean": 69.29045867919922, "std": 83.43473815917969, "min": -159.51522827148438, "p10": -14.19885330200195, "median": 68.55582427978516, "p90": 166.66830291748047, "max": 219.563720703125, "pos_frac": 0.8125, "sample": [7.212215423583984, 119.25468444824219, 140.0783233642578, 155.73330688476562, 122.3453369140625, 169.42257690429688, 60.367034912109375, 0.0, 99.429443359375, 71.70964050292969, 158.04530334472656, -25.751018524169922, 126.94229125976562, 110.00439453125, 7.47998046875, 210.96563720703125, 87.6572036743164, 0.03469085693359375, 139.42120361328125, 79.63519287109375, 124.71014404296875, 25.192073822021484, 88.23002624511719, -129.78262329101562, 32.61480712890625, 47.58843994140625, 132.35501098632812, 26.83453369140625, 134.60366821289062, 117.78309631347656, -2.0704345703125, 156.36380004882812, 4.596199035644531, 131.4666748046875, 144.78729248046875, 1.9958000183105469, 160.911376953125, -7.6789093017578125, 154.34674072265625, 85.82136535644531, 215.2163848876953, 9.86328125, 210.80020141601562, 52.291595458984375, 137.94342041015625, 179.142822265625, 23.204307556152344, -63.01153564453125, -122.15798950195312, 0.774627685546875, 8.129638671875, 143.66421508789062, 6.821014404296875, 146.89974975585938, 99.04110717773438, 161.51425170898438, -12.598907470703125, -157.56329345703125, 39.73602294921875, 161.01675415039062, 126.95033264160156, 146.05032348632812, 115.73161315917969, 84.94561767578125, 17.879623413085938, 131.95785522460938, 163.739990234375, 132.16455078125, 187.67251586914062, 141.54180908203125, 0.0, 150.32937622070312, 210.9327392578125, -29.799896240234375, -13.176239013671875, 22.32781219482422, 135.4144744873047, -28.592666625976562, 87.61575317382812, 219.563720703125, 6.994895935058594, 10.7586669921875, 201.634033203125, 87.94345092773438, 30.265762329101562, -159.51522827148438, 20.54498291015625, -8.81292724609375, 27.207763671875, 124.8453598022461, 152.86135864257812, 18.318099975585938, 19.981101989746094, -19.954635620117188, 202.7393798828125, 3.8329601287841797, 47.788787841796875, 119.24853515625, -69.57742309570312, -136.79013061523438, 65.40200805664062, 26.80908203125, -2.3776092529296875, 168.4998321533203, 22.75385284423828, 0.0, 165.88336181640625, 20.377288818359375, -0.22383689880371094, 94.94747924804688, 180.05755615234375, 111.90960693359375, 0.4978790283203125, 133.83139038085938, 95.56978607177734, -5.6695556640625, 25.3973388671875, 31.20013427734375, 21.687889099121094, 2.728759765625, 25.853240966796875, 10.718734741210938, 193.48171997070312, 145.29959106445312, -22.398468017578125, -16.58495330810547, 147.04641723632812, 31.532440185546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000466.npy"}
{"epoch": 0.9759162303664921, "step": 467, "batch_size": 128, "mean": 52.23879623413086, "std": 91.57190704345703, "min": -150.31246948242188, "p10": -54.90548706054686, "median": 24.528358459472656, "p90": 172.1668487548828, "max": 209.38482666015625, "pos_frac": 0.671875, "sample": [163.69033813476562, 143.28028869628906, 108.29067993164062, 100.74375915527344, 42.619300842285156, -12.979888916015625, 79.11175537109375, 35.991920471191406, 99.1842041015625, -1.3704833984375, -150.31246948242188, -141.59646606445312, -7.92718505859375, -13.548843383789062, 20.796661376953125, 148.82461547851562, -98.089111328125, 192.51052856445312, -7.2303466796875, 110.71155548095703, -20.64208984375, 132.89920043945312, -17.842910766601562, 119.59791564941406, 160.62875366210938, 80.34527587890625, 23.73211669921875, 134.99520874023438, 0.15302276611328125, 12.359619140625, 22.881378173828125, 149.42276000976562, 71.746337890625, 163.26422119140625, 178.49386596679688, 88.76641845703125, 25.324600219726562, 180.35916137695312, 76.81697082519531, -66.58056640625, 182.72195434570312, -0.994049072265625, 0.1526947021484375, 177.27169799804688, 20.67498779296875, 152.09027099609375, -16.429466247558594, -38.543060302734375, 173.47018432617188, -125.1038589477539, 12.266265869140625, -128.18386840820312, -38.039146423339844, 154.0630645751953, -50.50921630859375, 140.9893798828125, -38.19411849975586, -43.17950439453125, 150.13018798828125, 196.31686401367188, 18.40362548828125, 2.7545166015625, -0.8924694061279297, 10.41986083984375, 120.83755493164062, 16.75897216796875, 0.6789779663085938, 101.84166717529297, 21.45465087890625, -37.617740631103516, 7.102935791015625, -142.24610900878906, 161.56613159179688, 0.0, -1.8134002685546875, -4.6526031494140625, 81.62734985351562, 187.0851287841797, 4.576263427734375, -2.4884719848632812, 0.0, 43.9163818359375, 147.56869506835938, -9.581878662109375, 114.75701904296875, 98.06474304199219, 139.1680908203125, -39.35791015625, 201.87078857421875, -107.91304016113281, -107.95855712890625, 189.6004638671875, 163.4146728515625, 136.53363037109375, 171.6082763671875, 0.4011821746826172, -23.917068481445312, 22.485626220703125, -9.020660400390625, 111.29115295410156, -35.143035888671875, 161.85610961914062, 159.67935180664062, 39.15380859375, -17.654281616210938, 16.741195678710938, -89.21881103515625, -65.1634521484375, 13.573494911193848, 94.73321533203125, 170.80364990234375, 140.0537872314453, -114.69847106933594, 126.27218627929688, 9.236061096191406, 6.28369140625, -10.440277099609375, 105.79238891601562, 209.38482666015625, 42.98122787475586, 160.53225708007812, 100.029541015625, 50.96466064453125, 163.5226593017578, 181.22000122070312, 182.83642578125, -91.75167846679688, -23.736236572265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000467.npy"}
{"epoch": 0.9780104712041885, "step": 468, "batch_size": 128, "mean": 45.19153594970703, "std": 80.70736694335938, "min": -132.331298828125, "p10": -44.88482818603516, "median": 22.493072509765625, "p90": 160.40887451171875, "max": 235.9881591796875, "pos_frac": 0.6953125, "sample": [34.118377685546875, 190.016357421875, 142.9199981689453, -39.393653869628906, 30.126632690429688, 18.7847900390625, -25.919448852539062, 158.57058715820312, 44.94609832763672, 32.661529541015625, 31.320068359375, -42.71124267578125, 21.81500244140625, 182.99795532226562, -74.42831420898438, -13.413284301757812, -3.7159423828125, 167.32174682617188, -91.96937561035156, -22.8887939453125, 97.93894958496094, 37.716636657714844, -107.4854965209961, 40.65380859375, -11.730621337890625, 208.18988037109375, 48.64227294921875, 113.717041015625, -0.423095703125, -13.599414825439453, 2.39117431640625, 16.941726684570312, 181.23251342773438, -24.94304847717285, 31.867431640625, -8.510833740234375, -18.617095947265625, 159.77212524414062, -110.4853515625, 23.171142578125, 7.966209411621094, 25.27947998046875, -0.473663330078125, 146.61795043945312, -5.0135498046875, 3.6654815673828125, 70.52008056640625, -92.49684143066406, 38.44206237792969, 4.828025817871094, -5.5976104736328125, 3.803569793701172, 44.513946533203125, -2.682861328125, 13.399932861328125, 17.7684326171875, 10.66876220703125, 154.2541046142578, 128.46420288085938, 153.62110900878906, 163.544677734375, -24.65496826171875, -132.331298828125, 10.096282958984375, 235.9881591796875, 7.158660888671875, 15.287849426269531, 175.23208618164062, 2.59661865234375, 77.71337890625, 54.1363525390625, 103.42532348632812, -26.2244873046875, 76.8858642578125, 145.15447998046875, 154.62677001953125, -7.62103271484375, 17.62860107421875, 7.664520263671875, 120.4647216796875, -53.04068374633789, 191.79351806640625, 30.722064971923828, 168.2700958251953, 34.61529541015625, 126.19208526611328, 137.864013671875, -19.35452651977539, 137.02679443359375, 111.86615753173828, -10.828811645507812, 5.576198577880859, 2.170562744140625, 190.1700439453125, 0.005615234375, 35.82281494140625, 122.14471435546875, 3.7770004272460938, 62.543426513671875, 0.5364990234375, 150.37417602539062, 155.57308959960938, 50.07422637939453, 5.05816650390625, -44.334197998046875, -29.388671875, -46.960662841796875, 65.49578857421875, 51.756103515625, 96.5628662109375, -26.026779174804688, -5.38214111328125, 126.3291015625, -56.06672668457031, -54.698760986328125, 15.5374755859375, 147.3634490966797, 7.7456512451171875, -4.8465118408203125, -109.07833099365234, 144.69053649902344, 153.9927978515625, 28.306915283203125, 169.68487548828125, -46.16963195800781, 161.89462280273438, -51.741241455078125, 148.9771728515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000468.npy"}
{"epoch": 0.9801047120418848, "step": 469, "batch_size": 128, "mean": 59.457427978515625, "std": 92.26290130615234, "min": -207.69906616210938, "p10": -28.198161315917968, "median": 41.65576171875, "p90": 171.72006225585938, "max": 216.89678955078125, "pos_frac": 0.765625, "sample": [19.172882080078125, 167.71578979492188, 0.0, 13.924686431884766, 31.51153564453125, 137.95045471191406, -31.765823364257812, 29.63177490234375, 8.90838623046875, 110.89279174804688, 122.77651977539062, 34.7269287109375, 51.86871337890625, -150.48983764648438, 26.445556640625, -9.609466552734375, 190.47067260742188, 154.52235412597656, 22.307861328125, 133.14581298828125, -10.021469116210938, 52.60943603515625, 5.3440399169921875, -1.1275901794433594, -155.81607055664062, 165.4541473388672, 19.8206787109375, 163.1295166015625, 171.86553955078125, 49.183868408203125, 137.80010986328125, 54.162841796875, 212.19256591796875, 161.27517700195312, 1.8699302673339844, 105.42393493652344, 49.84405517578125, 83.27435302734375, 164.73573303222656, 7.88360595703125, -2.7826385498046875, -27.87548828125, 31.083282470703125, 6.5740966796875, 139.53036499023438, -3.3162841796875, 23.351181030273438, 23.07537841796875, 50.774017333984375, 37.7291259765625, 133.419677734375, 173.1590576171875, 161.4828338623047, 177.57806396484375, -144.03955078125, 127.84526824951172, 151.33514404296875, 195.73019409179688, 158.10028076171875, -107.651611328125, 5.338909149169922, -9.1009521484375, 186.1861572265625, -153.2550048828125, -105.39533996582031, 65.8182373046875, 23.858535766601562, -3.1798095703125, -29.581787109375, 171.2607879638672, 7.2760162353515625, 135.5014190673828, 79.14312744140625, 158.7515411376953, -207.69906616210938, 12.387619018554688, 45.5823974609375, 148.45960998535156, -7.2019500732421875, 216.89678955078125, 162.40232849121094, 135.64328002929688, 34.452606201171875, 0.542694091796875, 165.10986328125, -14.839080810546875, -195.75384521484375, 5.277280807495117, 60.796600341796875, 22.03736114501953, 58.04640197753906, 183.37396240234375, -5.3157501220703125, 163.79710388183594, -35.72784423828125, -27.503082275390625, 0.47907257080078125, 1.5887908935546875, -14.5260009765625, 167.47999572753906, -28.951065063476562, 0.9177322387695312, -27.01324462890625, 203.24862670898438, 5.341207504272461, 1.05828857421875, 171.65771484375, 101.76042175292969, 149.06655883789062, 184.06585693359375, 151.8802490234375, 7.04296875, 166.48583984375, 47.2919921875, -1.417755126953125, 74.91595458984375, 24.551130294799805, 153.427734375, 130.27072143554688, 34.91857147216797, 179.989990234375, 156.682373046875, 109.03614807128906, -23.654510498046875, 65.72146606445312, 36.0367431640625, 194.87237548828125, -75.1766357421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000469.npy"}
{"epoch": 0.9821989528795811, "step": 470, "batch_size": 128, "mean": 66.28998565673828, "std": 90.79705810546875, "min": -170.69195556640625, "p10": -38.2270439147949, "median": 48.94532775878906, "p90": 185.0393493652344, "max": 222.97303771972656, "pos_frac": 0.78125, "sample": [-16.1136474609375, 18.41778564453125, 170.72787475585938, 39.972686767578125, -1.9581756591796875, 140.73397827148438, 5.39385986328125, 150.99053955078125, 162.223876953125, 25.08514404296875, -30.7772216796875, 4.3896942138671875, -1.9984550476074219, 159.45068359375, 113.34088134765625, 144.51904296875, 170.40985107421875, 2.292003631591797, -13.480560302734375, 50.5386962890625, 122.57215881347656, -64.52804565429688, -65.58737182617188, -129.92031860351562, 205.97607421875, 142.51422119140625, 58.90946960449219, 21.2110595703125, 160.40223693847656, 156.13140869140625, 215.1802978515625, 131.40774536132812, -55.609962463378906, -1.61981201171875, 102.50032043457031, -93.58221435546875, 192.97421264648438, 101.92843627929688, 26.005767822265625, -98.35118103027344, 11.58868408203125, 201.99008178710938, 129.095947265625, 19.21014404296875, 9.956329345703125, 152.04293823242188, 188.1663818359375, 47.351959228515625, 46.54124069213867, 117.02261352539062, 10.749626159667969, 118.9114990234375, 24.8475341796875, 36.99853515625, -17.71673583984375, 146.172119140625, 129.91860961914062, -1.9871826171875, 146.26246643066406, 17.03973388671875, 66.22452545166016, 4.106884002685547, 157.1476287841797, 34.60784912109375, 21.361709594726562, 22.55145263671875, 6.920722961425781, -170.69195556640625, 17.818023681640625, 156.56793212890625, 9.97369384765625, 137.21096801757812, 97.20327758789062, -11.620803833007812, -120.107666015625, -61.50818634033203, -74.30657958984375, 131.14093017578125, 172.70997619628906, 148.05657958984375, 35.23248291015625, 45.14179992675781, 95.91326904296875, 201.03042602539062, 83.00495910644531, -20.515541076660156, 6.22930908203125, 186.37710571289062, 184.94973754882812, 185.24844360351562, -24.962478637695312, 163.4849853515625, 130.1499481201172, 146.18234252929688, -2.3056640625, 62.459259033203125, 21.245803833007812, 131.6348419189453, 125.73736572265625, 3.11492919921875, 168.23477172851562, 175.85952758789062, 120.30931091308594, 222.97303771972656, 18.653778076171875, 171.9163818359375, 1.0141372680664062, 5.432769775390625, -12.745697021484375, -148.85086059570312, 9.433868408203125, -66.99143981933594, 193.66278076171875, -27.64324951171875, 125.30042266845703, 203.1903076171875, 3.046875, 46.25140380859375, 73.0631103515625, 66.6153564453125, 191.90423583984375, -124.5335693359375, 66.14041137695312, 31.208343505859375, 195.38589477539062, 164.73226928710938, 169.50491333007812, -19.506744384765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000470.npy"}
{"epoch": 0.9842931937172775, "step": 471, "batch_size": 128, "mean": 49.59154510498047, "std": 86.67208099365234, "min": -157.8643035888672, "p10": -32.90140647888184, "median": 27.62055206298828, "p90": 168.21133270263672, "max": 231.537841796875, "pos_frac": 0.6953125, "sample": [-97.50297546386719, 154.66578674316406, 28.87103271484375, 89.30960083007812, 146.83657836914062, 202.95404052734375, 150.2255401611328, -114.38150024414062, -12.89760971069336, 151.4904022216797, -49.86250305175781, -21.758056640625, 28.807281494140625, 121.75202178955078, -2.678680419921875, 103.210693359375, 18.573394775390625, 26.177902221679688, -24.808021545410156, -20.80621337890625, 27.2972412109375, 27.656585693359375, 19.770263671875, 24.15618896484375, -36.11268615722656, -9.377098083496094, 188.4248046875, 0.863372802734375, -26.88446044921875, 35.22602844238281, 191.11270141601562, 50.811737060546875, 63.22308349609375, 10.49005126953125, -4.0279693603515625, 89.46379089355469, 136.5976104736328, 117.39012145996094, 25.65557098388672, 15.770416259765625, 27.584518432617188, -4.79290771484375, 158.92254638671875, 169.8243408203125, -28.886505126953125, 101.1981201171875, 28.260848999023438, 195.14744567871094, 142.47634887695312, -4.669166564941406, 4.9544219970703125, 12.07000732421875, 163.92718505859375, 95.593505859375, -121.68276977539062, 148.01953125, 120.16233825683594, 2.951719284057617, -5.20452880859375, 29.866249084472656, 209.3316650390625, 231.537841796875, 13.740606307983398, -1.777191162109375, -32.97526931762695, -4.248449325561523, -5.1851806640625, 6.3728790283203125, -157.8643035888672, 22.739349365234375, 98.22518920898438, 29.181564331054688, 146.9186248779297, 25.639556884765625, -106.91001892089844, 33.27728271484375, 33.710205078125, 95.72711944580078, 47.99535369873047, -32.8697509765625, 92.62130737304688, -18.09820556640625, -5.76422119140625, -14.658103942871094, 51.3582763671875, 133.62030029296875, 10.826461791992188, 2.3226699829101562, 145.282958984375, -5.549163818359375, 141.12576293945312, 153.76287841796875, -8.527244567871094, 55.22007751464844, 7.541748046875, 168.6137237548828, 168.03887939453125, -70.34011840820312, -9.227890014648438, -27.995407104492188, 32.33208084106445, 17.195159912109375, 10.59649658203125, 83.75909423828125, 139.9349365234375, 171.71665954589844, 45.119964599609375, 224.65805053710938, 94.83067321777344, 8.763214111328125, 139.91921997070312, 126.468505859375, 180.93295288085938, -132.279541015625, 167.26531982421875, 27.313201904296875, 149.56918334960938, 0.0, 44.934112548828125, -132.12892150878906, 180.66314697265625, 158.74038696289062, 181.5048828125, -17.3179931640625, 24.39410400390625, -17.100341796875, -112.08207702636719, -136.13519287109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000471.npy"}
{"epoch": 0.9863874345549738, "step": 472, "batch_size": 128, "mean": 62.66654586791992, "std": 91.16357421875, "min": -156.60653686523438, "p10": -33.362562561035155, "median": 39.34715270996094, "p90": 171.84937133789063, "max": 322.153564453125, "pos_frac": 0.75, "sample": [322.153564453125, -8.601608276367188, 2.441152572631836, -97.37672424316406, 4.9492034912109375, 2.62969970703125, -11.569499969482422, 142.10194396972656, 4.152557373046875, 17.204254150390625, 108.87738037109375, 43.756988525390625, 171.20535278320312, 30.327667236328125, -66.65331268310547, 147.33428955078125, -18.78595733642578, 10.150825500488281, -37.4427490234375, 85.55357360839844, 15.648651123046875, 31.258636474609375, 126.48541259765625, 43.449737548828125, -14.334907531738281, 175.11575317382812, 130.77999877929688, -122.2119140625, 3.9567527770996094, 147.53713989257812, 40.394439697265625, -82.93782043457031, 19.827362060546875, 202.3546142578125, 38.259765625, -27.175750732421875, 145.86143493652344, 162.66647338867188, -33.99342346191406, 5.5466156005859375, 23.49713134765625, 128.70216369628906, 115.69674682617188, 127.03140258789062, -27.939254760742188, 20.9700927734375, 15.010955810546875, -0.927337646484375, 14.21575927734375, 16.0306396484375, 138.5631103515625, 55.260467529296875, 170.64895629882812, -105.7283935546875, 140.2071533203125, 141.88385009765625, 162.05419921875, 145.5357666015625, 125.69874572753906, 32.46861267089844, -12.188201904296875, 78.06170654296875, 4.057460784912109, -1.7314281463623047, 142.619873046875, 167.2513427734375, 135.25997924804688, 65.79096984863281, 120.07614135742188, -2.88702392578125, 37.86375045776367, 34.240989685058594, 28.239990234375, 161.28904724121094, 142.48681640625, 205.41494750976562, 173.35208129882812, -8.901058197021484, 165.4862518310547, 22.636764526367188, -18.36749267578125, 230.55801391601562, 159.5040740966797, 202.34930419921875, 38.29986572265625, 204.62957763671875, 196.87103271484375, 167.5596923828125, 143.67483520507812, -156.60653686523438, -72.66796875, -106.52703857421875, 94.44059753417969, 70.296142578125, 2.938720703125, 130.96356201171875, 141.31597900390625, 9.203994750976562, -33.092193603515625, 106.61630249023438, 163.18310546875, -13.486503601074219, 169.26162719726562, 129.96487426757812, 19.13854217529297, 114.06564331054688, -6.56512451171875, 20.8072509765625, -2.8361873626708984, 1.15789794921875, 185.17166137695312, 175.51673889160156, 4.777460098266602, 16.88720703125, 45.667938232421875, -136.85426330566406, -91.78004455566406, 232.8172607421875, -13.01556396484375, 49.98944091796875, 179.47811889648438, 100.63482666015625, -107.27105712890625, 102.3497314453125, 129.3660125732422, 127.48529052734375, -30.980567932128906, -13.772407531738281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000472.npy"}
{"epoch": 0.9884816753926702, "step": 473, "batch_size": 128, "mean": 63.91862487792969, "std": 88.84867858886719, "min": -175.65724182128906, "p10": -28.586326599121083, "median": 46.696075439453125, "p90": 177.835400390625, "max": 213.70172119140625, "pos_frac": 0.765625, "sample": [10.86553955078125, 178.44439697265625, 55.84757995605469, -3.5069217681884766, 133.78553771972656, 7.083648681640625, 26.780029296875, 16.47894287109375, 43.06744384765625, -15.2462158203125, 79.44223022460938, 25.084701538085938, -65.64276123046875, 19.857141494750977, -112.95630645751953, 54.27608108520508, 45.39849853515625, 8.384765625, 77.82852172851562, 146.9498291015625, 118.39698791503906, 140.73800659179688, 189.3740234375, 213.70172119140625, 78.50927734375, -119.27532958984375, -15.468948364257812, 124.22886657714844, 109.11373901367188, 26.613800048828125, 179.01080322265625, 172.90667724609375, 158.24063110351562, 37.61285400390625, -14.382820129394531, -53.991455078125, 26.658172607421875, -119.21752166748047, 203.56515502929688, -6.440574645996094, 132.45068359375, 139.94622802734375, 37.629425048828125, -15.98337173461914, -13.789169311523438, -25.734848022460938, -43.42144775390625, 184.2193603515625, 61.6895751953125, 26.23114013671875, 19.452850341796875, -17.078903198242188, 0.04080390930175781, 185.61221313476562, -20.08270263671875, 177.57440185546875, -21.247314453125, 140.1330108642578, 193.8231201171875, -164.00283813476562, 71.70388793945312, 179.32687377929688, 182.50827026367188, 3.787872314453125, 29.597991943359375, -41.354095458984375, 209.54525756835938, 168.6395263671875, 79.77102661132812, 190.19525146484375, 127.03125, 137.43026733398438, 118.77902221679688, 19.243179321289062, 4.228515625, 144.63592529296875, 144.50393676757812, 16.70909881591797, 13.713775634765625, 150.10577392578125, -3.5942230224609375, -35.239776611328125, 167.3436279296875, 134.2689971923828, 134.32501220703125, 172.49246215820312, -175.65724182128906, -114.13013458251953, 152.99981689453125, -16.509765625, 30.25640869140625, 147.42279052734375, 2.9272079467773438, 45.322967529296875, -0.88800048828125, -52.555999755859375, 20.237232208251953, 30.582916259765625, 66.001220703125, 1.6975555419921875, 177.490478515625, 138.7395477294922, 136.3655242919922, 0.0, 147.17291259765625, 98.28506469726562, -137.79934692382812, 140.13865661621094, 27.715744018554688, 134.89892578125, 145.37457275390625, 15.878036499023438, 3.234577178955078, 170.50772094726562, 27.049285888671875, -4.669769287109375, 4.280670166015625, -14.301742553710938, 28.28594970703125, 49.1466064453125, 153.7185516357422, 146.6861572265625, 157.93972778320312, 58.697425842285156, 205.57275390625, 137.66102600097656, 168.53599548339844, 47.99365234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000473.npy"}
{"epoch": 0.9905759162303664, "step": 474, "batch_size": 128, "mean": 47.041358947753906, "std": 84.84364318847656, "min": -180.2906494140625, "p10": -46.73231887817382, "median": 32.82466125488281, "p90": 158.04876098632812, "max": 249.53497314453125, "pos_frac": 0.7421875, "sample": [-30.187637329101562, 27.89105224609375, 249.53497314453125, -3.8736953735351562, -9.47488021850586, -30.48876953125, 131.9577178955078, 187.31454467773438, 69.3724365234375, 150.7232666015625, 7.57965087890625, 32.572509765625, -7.74853515625, -126.3499755859375, 7.741146087646484, -136.86077880859375, 4.3366241455078125, 1.392242431640625, 106.42993927001953, 51.40502166748047, 80.36409759521484, 171.04188537597656, 3.6992645263671875, -69.01583862304688, 152.17245483398438, -29.171722412109375, -91.503662109375, 45.732421875, -18.250892639160156, 72.15838623046875, 140.6026611328125, 49.5865478515625, 0.3194122314453125, 136.41900634765625, 37.759552001953125, 133.167236328125, 220.49993896484375, 21.664588928222656, 0.5887527465820312, 179.54229736328125, 132.02618408203125, 157.0709228515625, 179.49569702148438, 170.1886749267578, 66.98138427734375, -3.457378387451172, 0.0, 25.28082275390625, 23.919677734375, 9.994575500488281, -2.4907379150390625, -8.201416015625, -50.36970520019531, 20.038864135742188, 71.418701171875, 115.53839874267578, 21.393043518066406, -45.173439025878906, 160.33038330078125, 147.9370880126953, 7.7780914306640625, -8.6651611328125, -5.8769378662109375, 150.71871948242188, 129.0493927001953, 25.387100219726562, 150.733154296875, 79.54006958007812, -36.4307861328125, 11.211013793945312, 92.55752563476562, 30.939945220947266, -133.13246154785156, 12.055500030517578, 41.733978271484375, -25.603271484375, 178.34765625, 125.20219421386719, 99.08518981933594, -5.804351806640625, -58.07305908203125, 19.833648681640625, -0.6895942687988281, 4.80595588684082, 0.2203216552734375, -180.2906494140625, -52.75005340576172, 139.55259704589844, -160.4772186279297, 144.4469757080078, 100.21897888183594, -174.95132446289062, -12.76580810546875, 2.14178466796875, 131.52938842773438, 8.930145263671875, 55.795166015625, -12.56365966796875, 33.076812744140625, 116.458984375, 77.13748168945312, 145.0771484375, 165.196533203125, 76.89239501953125, 203.25784301757812, 95.30593872070312, 124.74313354492188, 17.797515869140625, 48.137725830078125, 23.41326904296875, 34.342437744140625, 38.55323791503906, 38.76983642578125, 87.92794799804688, 35.64385986328125, 47.507049560546875, 142.01519775390625, 178.46115112304688, -57.81184387207031, 38.437103271484375, 31.144058227539062, 164.1047821044922, 33.9608154296875, -124.69219970703125, 26.71038818359375, 23.903900146484375, 28.7705078125, 140.74533081054688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000474.npy"}
{"epoch": 0.9926701570680628, "step": 475, "batch_size": 128, "mean": 49.42664337158203, "std": 87.19001770019531, "min": -187.32089233398438, "p10": -51.33792991638183, "median": 32.974647521972656, "p90": 157.3899383544922, "max": 239.41799926757812, "pos_frac": 0.6953125, "sample": [185.0662841796875, 142.76193237304688, -77.58753204345703, 142.5583038330078, 152.01922607421875, 30.70806884765625, -187.32089233398438, 204.81924438476562, -94.47490692138672, -48.89013671875, -0.4953460693359375, -130.764892578125, -49.62518310546875, -46.02715301513672, 10.202728271484375, 131.4536895751953, 122.98469543457031, 148.9073944091797, 157.697021484375, -10.22823715209961, 184.41189575195312, -15.979934692382812, -27.269363403320312, 49.245018005371094, 169.21791076660156, 152.5423126220703, 139.30084228515625, -9.549041748046875, 219.66070556640625, -1.693399429321289, 41.752593994140625, 92.886474609375, -15.523773193359375, 153.11891174316406, 43.79109191894531, 20.159568786621094, -121.0146255493164, 28.306472778320312, 1.1669292449951172, 157.25833129882812, 134.20162963867188, 39.71648406982422, -129.4477081298828, 61.62571716308594, 84.42265319824219, -15.525665283203125, 129.63278198242188, -11.70904541015625, 148.373046875, -28.340362548828125, 14.18222427368164, -83.2622299194336, 149.3514404296875, 142.66879272460938, -100.204833984375, 152.31341552734375, 179.09445190429688, -2.3970470428466797, 21.525659561157227, 104.95477294921875, 148.6619873046875, 21.6390380859375, 6.8333282470703125, -20.874305725097656, 25.6868896484375, 102.07713317871094, 21.302871704101562, 44.822998046875, 8.267822265625, 31.2596435546875, 15.6512451171875, 160.8008575439453, 72.66075134277344, -15.53778076171875, 131.00454711914062, -37.94561767578125, 0.9093723297119141, 239.41799926757812, 39.591552734375, -1.8522796630859375, 28.099349975585938, 190.66366577148438, 153.89300537109375, -33.542388916015625, 146.70159912109375, -55.3343391418457, 54.562042236328125, 5.0496826171875, 53.600341796875, 48.047569274902344, -0.5366744995117188, 35.40434265136719, 127.49813842773438, 130.65367126464844, -2.25408935546875, 150.92153930664062, 140.02325439453125, 160.45211791992188, 32.87825012207031, 88.71624755859375, 12.180778503417969, -106.07220458984375, 36.174957275390625, 0.1857147216796875, -7.026275634765625, -32.767852783203125, -100.04364013671875, 69.91143798828125, 124.42703247070312, 107.75303649902344, 86.29814147949219, 92.6650390625, -2.1321640014648438, 12.424346923828125, 148.7011260986328, -33.3160400390625, 59.96577453613281, -74.99063110351562, -91.22589111328125, 25.746734619140625, 182.61404418945312, 33.071044921875, 19.116092681884766, 190.98117065429688, 73.56022644042969, 10.8316650390625, -8.346546173095703, 9.31829833984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000475.npy"}
{"epoch": 0.9947643979057592, "step": 476, "batch_size": 128, "mean": 58.3709716796875, "std": 88.33688354492188, "min": -174.85910034179688, "p10": -32.21426086425781, "median": 32.15055847167969, "p90": 171.28668823242185, "max": 232.03359985351562, "pos_frac": 0.7109375, "sample": [-27.382080078125, -114.90200805664062, -4.204376220703125, -35.725341796875, 17.646636962890625, 142.92112731933594, -2.1993942260742188, -8.167953491210938, 17.29693603515625, 95.78047180175781, 129.9754638671875, 92.8924560546875, 164.5984344482422, 191.87625122070312, 122.7068099975586, 201.538330078125, -31.06451416015625, 179.76832580566406, 141.21810913085938, -6.276554107666016, -8.37103271484375, 35.3193359375, 16.776123046875, 23.695953369140625, 129.83316040039062, -16.601783752441406, 159.84426879882812, -2.989023208618164, -174.85910034179688, 182.7468719482422, 17.2921142578125, 31.665252685546875, -37.75639343261719, 181.54238891601562, 158.73724365234375, 14.374359130859375, 169.85101318359375, -62.401573181152344, 28.34967041015625, -14.933624267578125, -0.233154296875, -169.8988037109375, 127.57647705078125, 21.443439483642578, 137.2499542236328, 32.6358642578125, 134.88253784179688, 184.32366943359375, 16.961135864257812, -70.682861328125, 110.30465698242188, 166.526611328125, 11.32415771484375, 11.6759033203125, -1.6638603210449219, 193.29708862304688, 9.733978271484375, -17.599609375, 167.0236358642578, 29.912750244140625, 148.29119873046875, 118.78043365478516, -39.527862548828125, 166.67776489257812, 11.83636474609375, 183.12100219726562, 175.83334350585938, 22.4595947265625, 133.41830444335938, 60.744781494140625, 28.6400146484375, -17.511642456054688, 208.6401824951172, 37.259918212890625, 41.11626434326172, -1.3199100494384766, 138.87875366210938, -129.73648071289062, 153.22695922851562, 43.5640869140625, 145.00416564941406, 136.50564575195312, 136.98928833007812, 1.884033203125, 9.045623779296875, 3.2924346923828125, -3.3400497436523438, 132.73385620117188, -160.31040954589844, 117.80103302001953, -12.021194458007812, 88.1412353515625, 174.6365966796875, 3.1044769287109375, 123.86589050292969, -17.774581909179688, -13.47265625, 85.25789642333984, 3.73968505859375, 139.13534545898438, -34.897003173828125, 35.662750244140625, -74.08642578125, 232.03359985351562, 17.8260498046875, -24.713027954101562, -8.836883544921875, 11.469459533691406, 100.3521728515625, 102.2991943359375, 48.51611328125, 149.6640625, 156.2008056640625, 31.046279907226562, 6.061126708984375, -17.8477783203125, 131.86358642578125, -20.043746948242188, 1.8516693115234375, 112.12542724609375, 161.30047607421875, 132.51187133789062, -89.69732666015625, 168.97618103027344, -24.098541259765625, 175.14498901367188, 124.24295043945312, 92.73919677734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000476.npy"}
{"epoch": 0.9968586387434555, "step": 477, "batch_size": 128, "mean": 63.42995071411133, "std": 93.24760437011719, "min": -188.31533813476562, "p10": -44.5405029296875, "median": 55.84518051147461, "p90": 174.91624450683594, "max": 226.05364990234375, "pos_frac": 0.7109375, "sample": [-116.44338989257812, 69.21685791015625, 147.0586395263672, -108.49462890625, 108.22479248046875, 36.72796630859375, -63.711181640625, 121.73358154296875, -27.969192504882812, 88.53758239746094, -8.520280838012695, 173.59820556640625, 10.625747680664062, 141.90097045898438, 20.6475830078125, -10.09515380859375, 200.22695922851562, 92.70925903320312, -188.31533813476562, -7.77252197265625, 1.9114990234375, -32.72430419921875, 85.21835327148438, 171.50897216796875, 197.9575958251953, 177.99166870117188, 133.58004760742188, -8.790487289428711, 17.722442626953125, 27.4395751953125, 145.00454711914062, 169.39459228515625, 38.78570556640625, 210.42446899414062, 133.56414794921875, 47.022369384765625, 205.58099365234375, 56.80889129638672, -11.828804016113281, 0.0, 16.4476318359375, 157.02073669433594, -2.787200927734375, -29.2203369140625, 128.65740966796875, -16.817779541015625, 12.2017822265625, 143.26373291015625, 153.80938720703125, -104.8447265625, -4.621772766113281, 12.54937744140625, -26.252761840820312, 164.34527587890625, 28.346771240234375, 14.910614013671875, 145.14523315429688, 27.726806640625, 150.78466796875, 65.29951477050781, 139.97552490234375, 163.29153442382812, 226.05364990234375, -8.4879150390625, 124.0789794921875, 121.37362670898438, 184.4959716796875, -109.94384765625, 66.11963653564453, -63.00323486328125, 118.00814056396484, 163.2216796875, 118.22221374511719, 143.69613647460938, 118.54119873046875, 184.18014526367188, -44.414764404296875, -5.86260986328125, 1.4916534423828125, 151.58111572265625, -24.38250732421875, 161.05593872070312, 118.62384033203125, 0.08089447021484375, -29.034317016601562, -154.10430908203125, 157.20559692382812, 104.1229248046875, 78.61103820800781, 129.56365966796875, -3.696627616882324, 15.67041015625, -13.216236114501953, 194.45428466796875, -5.95867919921875, 149.72142028808594, -114.89176940917969, 132.77023315429688, 39.969390869140625, 142.49005126953125, -93.83499145507812, 212.12237548828125, -27.45147705078125, -145.64736938476562, 17.587890625, -19.909820556640625, -44.833892822265625, -4.86578369140625, 190.57464599609375, 97.15017700195312, 41.7969970703125, 217.48223876953125, 164.03036499023438, 178.39920043945312, 0.96533203125, 158.271240234375, 158.18850708007812, -55.062591552734375, 164.86631774902344, 13.1302490234375, 54.8814697265625, 135.35787963867188, 142.3528594970703, 50.63268280029297, 3.6116409301757812, 95.98150634765625, 52.185455322265625, 6.9731903076171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.4-20260429-230725/margin_logs/step_0000477.npy"}

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:2b4e5bed25bd4783993e7d8bed8c0dcb604850687b6133068c9ebbcfe9593fce
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:923d8d34ebce2d992a81fde80eff4a6f422a21f64bc55e9cf132991c17436c41
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:365eaa665a0c7698658a73751f5c9745cb25e589c8d6d59ae4d66c263fe118a2
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f9d650311b777002ed24a99c6d73576d35b9f45ba5eb7d1ed3e9522683b55860
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cdda08f20f65b69506f6dd23591cc359bf424bf70b1417fa6606b2bc72464d89
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a5d35c35bac92fb3124f3dd1b4345bdfb3c081fb671eca6fa6c3d0710b9284ad
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:dd45e93aeb656ec00bb096434251e13ea893db59340aae1d7eb57a266ec89812
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:1aff882b4bf90005134838a08afb6ea4ab780bf4c28741accbed2153b7c9ccfa
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:383152ce46636f4d929d25bf3607953690b4e7afa12f3fc70f27c90a8591e056
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:299cb19be087ae28ff6f5ea0c13d506eb095564220c668f570923a303d9882a8
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d1e94ffbf8bc0ceb49b9d3df6d2f0d9c8e88b18f6dba32e062139f03bd0e8816
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:6084efa51250d20d4a8f4682694462cad75c7833c94422723378fa1008ee0683
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:6079b5e7dd07868b06b6436b6ce379bb0340496a99fcedf451c3d0f76aa26e11
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:0b4718beabb3853df88b8dab12b2be5e55bc2a66de298fc0c337239ed967d009
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:900c819d6dc36e4e28049573c7c86f6a0622ec62802158774c93f6a7f9a1b5f0
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:58e1556aec5f8ceea9f7bcb64b1836a2df6b88c084eb45d0b9cbd8b706a72076
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:4fc96f49feb6399bfec97bc05fd65236c4180faff7f84c3ca1eea5561c3c7d6a
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b5fc63392dbd09efc69cc10324de46018d10bfc3750131aa7d956c9cd5ce2dbf
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d4f93bb718943ae9a0c9392b5bddfb2d13d4bd434db4d58a203cc5fe74a39efe
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:1f153a2be6ed30b2b11841ecfbe5d2012deea83de98a7ead07d029310b3b444f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:77eb38ba70ed91b415db5b526a4d14083d69cea1838d352ba9684a8fb1c2f828
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ee008349c50fbaa6c8851fb0efbe87582164ffadc48c1082f58d6b7d9517c558
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:bbf2b5c186b744f35191e20fef8e13ac68487dc48a59afc455f1373888b81249
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:11491aecb6c58ea82f1d3d0b7c17c3d7cc624a898ea19388c2e8c36f3ae0aacf
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f3c3bd3d43ea5ab17ae3561f13d1158b596483c61d64684cde6b4310f1191cd0
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7bf28729f46ee57e3c826cd266a57bd75a7734fbac6444e6f9a1630da63181c8
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:224e72e5446c26ce2215b522b7bb0009dc0f1ff90bbadeaaf656bc0a53baab7b
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:166dc0d6f7c9874182b18bc33895be0755ecfba34337c770df618ca5a4a4d5ab
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c8be2f7160f796682a15fe9ef8d004bac4a9096eda1c82d51dc3d187831b2ab0
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f21246e36c09a40b4da8b57f3b9c421ceb32c0e6b177df6972d1c0cff79f935e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:9e85074517d7227a9a2315eb195d284456b282caeec54a4f89e55f0d7038610c
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:8a73f0d26e0519ce7d26d26235585ec3b8d92c5673fca828c07ae0217d996066
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:140ce4590da6a857d55cbe6363d727b00f814a9484147496ad22f2749c0e20d7
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c07a7f561875c60a9e5731bbb0a0cac39e1ed63c4af06ba871c6f945862f99e1
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:9ca497fe2c1b885dc431d1a1811cd8073da8801867407bcddf63f51f2139f6ba
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:042794f917e1a522690f5fddb197c5e5d02313933003d35c7691796cfa199934
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:5dc62a7db07391dfa7460dd3346d1ad3240d24599c862eec2026b080085d28c4
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:15714a61f2b0c59be0572239816066f97bdac7d10f95b7068e43bf197e8bfc69
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b0d33f7a5597668371a67d41bbd4b9d07630b8cc9b6ac76c99b6b0a9e357da1e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d8742b9f82aaf4a57c91ae864ccb458c1436f7833dd80ce21c65152a807cb694
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e53f7932cf744e0ac8bad1b28b5222b65452f9bd9ebdd100392eb082b49a24d5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:013f01bbbd02400c064b35bc43c0ad9ec6ae9601ae12d6971f7d0a5625c60306
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e967af86ba32f054b4b4be97f3c49cfb9e74fc30b5aa13b2f8719e1c86be176e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:df9568b6e5697a612fc50791930a43feb1eb64a80177a6ca5d98b6ef6c01ea09
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:502578a73d9d34a37cbd939adba69c1caa2af6562bd5c9b6631edb8b57af11a1
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:4061786c3eca245cbf8cf9edc0bf1d97e849755b0e48d43b5f027bee61a6b5d5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:55b6c86c1d9afa7ecff08ce6a79ef659624615f012d8bc93f49e05453873d5da
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:51554462e97a7369d1a1b5973483e4bf73f50cbc5ca3e3f4af4b37f154c8a174
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d09c73e3db3a5479d61d765bd1f6956d77763be041e81beafa1256732f0c2e3f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:5d606ccd07007fe9eb53dec68f22071f025d3577a18ff071b7ce383f26ab7448
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:6cbc213e2742ce5ff192e2e4d953ed2aab18dc1c90365e355878e84b802c324d
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:8c6a7fc0a2b163d4f6109eab1ba45f4f556261f8b1d09ba9a61c80dab8ef9c7d
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cb542ae41f02e9003ce67b2a770d99bea6e500ba77e99c67490e1a6fcb439b81
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d49ae27c145f247458072a585c7784755da3ee2553c5cf0c4db953d2f35e560f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:647100369c5e8bd0891c0a920feb258685e007a279177cc027862d93bbbbde5d
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e6ab759b67317dc4c9d5a02ab6266d88dfe1dec531350f201297104966c3f0b6
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:1223b066c817855adc2e7adc7d39549e096de4b6e3e63cf2be8430082d885925
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:53a2dea5bba6cbabc4c86b7b9ebb305e19c21d45da813000ff6430c44c04e329
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:069580d05064dcb532f2920e057fbb1b66fa3019ecb02b42f454f6a16867ddc4
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:61544ff659bee313b2dffc98091d4283891df66f0b0967b77e9c3a4fda24e30c
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:413f583c51f184fc0bd8c2fb400f6746228bb4caa195798e689dc79b796a9662
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:eba65a650be6453fad0d21b896d99f9394b3cdbe065440cef6cd9211577b05c3
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b41ea46226e49ed19c8db7fbdafa27df49231ba0f303a14ae5fd04fbe8cf970f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:47e5575cc0927bea17c288e799e88aa006c51599ab3d1e96e1089d5d233f62b5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:40c27f638f637101f69cd71f6f8ba1d17521ab1480c1bb7609346859ac3841b2
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:3a7875641a98e416a5fc91ad2232ca1e94bba7c62bac76f536f70d450262298c
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cc3d37d575962e5c5b0a10e4d1ec915b6aa3207875e755d6f3c980a1be216eb4
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:03f7fc9901ec9ef842178ce6a6d2b7dd57dba7cc27e59f46554e3a84db479802
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:df39f3f2cabd8fb9c2e7ecbd5ab174f7236195b6cb5fdf6e8ad5608a8d82774e
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:22d95f32f6d1ea88daed7d8db8116b2ed56e410347d0224e77219e44fbdbbd8d
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:57478ee9fc5ae58cf17e4047243f736463ab1f3cc2c939df065c820ca4f416f7
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c5dde88ec56efab5e7ad61a54c1805c00cb46bb8f52794551d2db94a65f70fac
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:310e9fa098df59c8915744accb11163752b3a859b4b4ee2c72fe1115c1ea6886
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cf76937c11c2e897548e7b6c2ee11ff193003b7126fc279a686c446d9833a929
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7d611cc21b4f36af9c60e3c2e31faf5b09fb84e4ce3603bb4397d524cf65f164
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:44ffd355cafab3646abe703d65bdf5898bd83c8a0968619267c15113f9d2698b
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d78995fbffe29b2edc0bbf0933051fa925ca8297b6e2e8c2ac7b2fb267fff490
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:9bc987de8cb74c26aafe6a7d69abd7106c4bf2b77f86248e9c0d89103b834a42
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:df785c2b634de216d017c2de174707e5709b46b7462ae249aba26b95d5fb3b0f
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d278f5b616ca2918a9b86d753f58e82eb3a173738e7bec673225b01781ca17e4
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:595605453bf9d6672a79923dc81c2204596dcf3d1aa0f2b8869757b853c4cefc
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a17266942023411ad9b30291958a05a9ba29cfd577dd830f45d8c851e89f6659
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:92120d7e43b4745465467cf902ff2c19fdb0dea8d94a22dc5037b8e823882c42
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:69787e889bb195d0923e5341a5bf7a2f9518db80c16c7802aa6084ba728e36a1
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:25944ae53492c6113a4becc7d9680a0d2bbda808076243ec154b7272ab07306a
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a65101dd80e4856db7a2b2fca2c0c0c440ed3a73ad22103ac76622e647775aba
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:6eea58c4abed0e66c5d181210a98051449dbd4818b5faa3a4b3bca4a31f80041
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:7228a70c583573c168fe24216cb58ad3adc8af507f069638d000e8c8aab1f791
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:4b72f64f90765f643db18c1c154491917d6436e81ec55d8e9b8d4efa17473dcb
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d7266c62291f5425572ca7e30534a074eab9901f88678a5597d8a5a60a88f3de
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f6af19e5f2b96b69ebb1c23d0018518e5599d25f4870d04a87c416c9d65739d3
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d0a3675c828cbd0773e9aac1af357b58d97bd652168ef0872ca53d36f226cd95
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a1db3ea3df4633f33beb500e753cb80dc0b969567fcf9d9a8c150469de6afac4
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:71b9a9d67c7ad76772fc9bc48c6f33ba2839e8f3880163daa585141276c22fda
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:4437984407a488c74a40fea838b425229ac4710ddea5dd2afef55b09f72425c5
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:273ab5f84a050684c45ac5a4728960f6e792224f5e79f166088be526f7077046
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:8594768d962905d859269641996aca28bbae4bc1dfb1e134bb1451c17d5f44f3
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:ccc82b2362890660a22f7c35d32f5d39ff5bccb14f70563fad7449af4911a17d
size 640

View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:a00e2ca51d5ae7a02483b7435d563a29c4cae67cd3cf4f6e421cca1a6380bda6
size 640

Some files were not shown because too many files have changed in this diff Show More