Files
qwen3-8b-base-new-dpo-ultra…/margin_logs/margins.jsonl
ModelHub XC 66a4e63baa 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-0.3-20260430-192039
Source: Original Platform
2026-05-29 20:55:22 +08:00

478 lines
1.3 MiB

{"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-0.3-20260430-192039/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-0.3-20260430-192039/margin_logs/step_0000002.npy"}
{"epoch": 0.004188481675392671, "step": 3, "batch_size": 128, "mean": 0.01832614839076996, "std": 0.7961068153381348, "min": -2.465087890625, "p10": -0.9811218261718748, "median": 0.09198570251464844, "p90": 1.0257247924804687, "max": 2.0340728759765625, "pos_frac": 0.546875, "sample": [0.524932861328125, 0.011510848999023438, -0.1837158203125, -0.036182403564453125, 0.2940826416015625, -1.39227294921875, 1.01092529296875, 1.220855712890625, 0.0, 0.529876708984375, 1.30267333984375, -0.341064453125, -0.367462158203125, 1.58935546875, -0.38751220703125, 0.3181037902832031, -1.515472412109375, 0.780181884765625, 2.0340728759765625, -0.403778076171875, -0.51776123046875, 0.181732177734375, 1.18597412109375, 0.29137611389160156, 0.6016845703125, 0.24114990234375, 1.16424560546875, -1.40625, 0.062591552734375, -0.427764892578125, -0.00310516357421875, 0.08338546752929688, -0.22052001953125, 0.49853515625, -0.56280517578125, 0.18043136596679688, -0.79180908203125, -2.022705078125, -0.54766845703125, 1.338775634765625, -0.1134185791015625, 0.43536376953125, 0.492645263671875, 1.3878631591796875, 0.3704833984375, -2.465087890625, -0.0951385498046875, 1.49310302734375, -0.08245849609375, -0.10565185546875, -0.0323486328125, 0.22867965698242188, 0.480316162109375, 0.3202857971191406, -0.1647186279296875, -1.3834228515625, 0.44261932373046875, 0.286376953125, 0.3740348815917969, 0.16820144653320312, 0.0, 0.261749267578125, 0.618896484375, -0.916290283203125, 0.30229949951171875, -0.8498077392578125, -0.885833740234375, -1.86492919921875, 0.09748458862304688, 0.2501411437988281, -1.703125, 0.08648681640625, 1.641754150390625, -1.07122802734375, -0.0082855224609375, -0.5419921875, -0.322509765625, 0.914642333984375, -0.825927734375, -0.452362060546875, -0.0377655029296875, 1.0077972412109375, 0.9146995544433594, 0.20782089233398438, -0.12579345703125, 0.5657958984375, -1.656890869140625, 0.402679443359375, 0.12811279296875, 0.0706634521484375, 0.1820831298828125, 0.139923095703125, -0.64788818359375, -1.62554931640625, -0.07370376586914062, 0.4428558349609375, 0.12740325927734375, 0.7347030639648438, 0.4934501647949219, 0.18591880798339844, -0.2896232604980469, 0.2266845703125, 0.4495811462402344, 0.186859130859375, -0.075653076171875, 0.222015380859375, -0.6574249267578125, 0.51617431640625, -1.2311248779296875, -0.218292236328125, -0.9425048828125, -0.69793701171875, 1.0602569580078125, 0.157470703125, 0.853729248046875, 0.32324981689453125, 0.3885040283203125, -0.368560791015625, -0.751312255859375, -0.111053466796875, -0.0943603515625, -0.07050704956054688, 0.0230560302734375, -0.5987281799316406, -1.4273605346679688, 1.196380615234375, 0.43817901611328125, 1.314239501953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000003.npy"}
{"epoch": 0.0062827225130890054, "step": 4, "batch_size": 128, "mean": 0.14553971588611603, "std": 1.0315662622451782, "min": -3.81201171875, "p10": -0.9051239013671875, "median": 0.1633758544921875, "p90": 1.279827880859375, "max": 3.50323486328125, "pos_frac": 0.59375, "sample": [0.039398193359375, 0.3564453125, -0.06173896789550781, 0.8779296875, -0.38739013671875, 0.0054779052734375, 0.44548797607421875, 0.90631103515625, 0.0355224609375, 1.4578857421875, -0.768035888671875, 0.64202880859375, -0.4839019775390625, 0.39882659912109375, 0.5022087097167969, 1.197509765625, -0.2513427734375, -1.483642578125, 0.539703369140625, 0.556427001953125, -0.61041259765625, 2.5537109375, 0.4166107177734375, 0.950286865234375, -0.922149658203125, 0.9428253173828125, 3.50323486328125, 0.12942123413085938, -0.70428466796875, 0.20238494873046875, -1.213531494140625, -0.021955490112304688, 0.667449951171875, -0.77130126953125, 0.31787109375, -0.543853759765625, -0.2715911865234375, 0.5841064453125, 0.4074554443359375, -0.794708251953125, -0.06332588195800781, -0.139617919921875, -0.4978675842285156, 0.18438720703125, 1.636962890625, 0.36083984375, -1.2364349365234375, -0.67132568359375, 0.3572998046875, -0.407958984375, 1.604705810546875, 0.12248992919921875, 0.6966552734375, -1.513580322265625, 0.1396942138671875, -0.03452301025390625, -0.604461669921875, 1.28533935546875, 0.491455078125, 1.255889892578125, -0.83892822265625, -2.088714599609375, 0.143218994140625, -0.9458465576171875, -0.61749267578125, -0.701446533203125, -0.39776611328125, -0.3040618896484375, -0.3756866455078125, -1.3084716796875, 0.19573211669921875, 0.27508544921875, -0.81329345703125, -1.97601318359375, 1.8770599365234375, 0.726806640625, -3.81201171875, 0.008636474609375, -0.3629150390625, 0.5047073364257812, 0.516204833984375, -0.6651611328125, 0.03302001953125, -0.8978271484375, 0.9715118408203125, 1.839141845703125, 1.230682373046875, 1.258697509765625, 0.641357421875, 0.058349609375, -0.30648040771484375, -0.7152099609375, 0.18353271484375, 0.5328369140625, 3.032318115234375, -0.8855438232421875, -0.40985107421875, 0.102783203125, -1.2547683715820312, 0.5, 0.650604248046875, 0.9174346923828125, 0.4493675231933594, 1.134918212890625, -0.49505615234375, -0.0089111328125, -0.20189666748046875, 0.2224578857421875, -2.551513671875, 2.4564208984375, 0.37580108642578125, 0.629852294921875, -0.34902191162109375, -0.38055419921875, 0.7529449462890625, 1.402801513671875, -1.65838623046875, 0.850738525390625, 1.79547119140625, -0.6219978332519531, 0.5243682861328125, 1.779541015625, 0.412384033203125, 0.6911163330078125, 0.09576416015625, 1.2774658203125, 0.841064453125, 0.368408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000004.npy"}
{"epoch": 0.008376963350785341, "step": 5, "batch_size": 128, "mean": -0.0706949383020401, "std": 0.8736025094985962, "min": -2.37750244140625, "p10": -1.131884765625, "median": -0.02140045166015625, "p90": 0.8420852661132808, "max": 3.13677978515625, "pos_frac": 0.484375, "sample": [-0.8397979736328125, 0.94775390625, 0.14925384521484375, 1.15869140625, 0.4558868408203125, 0.7978057861328125, -1.345947265625, -0.8364105224609375, 0.6592864990234375, -0.957122802734375, 0.3447265625, -0.484954833984375, -0.52197265625, -1.201995849609375, -0.15874481201171875, -1.08294677734375, 0.5533370971679688, -0.42620849609375, -0.84063720703125, -1.89984130859375, 0.23248291015625, -0.08001708984375, 0.10900497436523438, -0.9393310546875, -0.562744140625, -0.590911865234375, 0.298553466796875, 0.6727752685546875, 0.5216751098632812, -0.06340408325195312, 0.0909423828125, -0.6227340698242188, 0.287567138671875, -0.1722564697265625, 0.5565032958984375, 0.3886871337890625, 0.50457763671875, 0.2039794921875, -1.31146240234375, 3.13677978515625, -1.101837158203125, 0.981689453125, 0.7449951171875, -0.3688201904296875, -0.25457763671875, -0.2839031219482422, -2.152191162109375, 1.8321990966796875, -0.4870185852050781, -0.27899169921875, -0.7794189453125, -0.1744384765625, -1.7828369140625, -0.03690528869628906, -0.09828948974609375, 0.945404052734375, 1.17071533203125, 0.140777587890625, -0.0449066162109375, -0.9674224853515625, 0.5711669921875, -0.932342529296875, 0.1449432373046875, -0.6170082092285156, 0.32476806640625, -0.107269287109375, -0.7974395751953125, 0.40020751953125, 0.389739990234375, -0.2164306640625, -0.9901123046875, 0.6169281005859375, 0.7737045288085938, 0.19327545166015625, 0.06053733825683594, -0.5353851318359375, -0.878997802734375, 0.55865478515625, -0.650787353515625, 1.3421173095703125, -1.69293212890625, 0.38482666015625, 0.3017578125, -1.814605712890625, 1.27294921875, -1.34393310546875, 0.26177978515625, -0.7581787109375, 1.7821044921875, -0.3070526123046875, -1.719390869140625, -0.6186943054199219, -1.729644775390625, 0.663330078125, 0.501953125, 0.095062255859375, -2.37750244140625, -0.105438232421875, 0.28754425048828125, 0.5134735107421875, -0.2442626953125, -0.8953857421875, 0.6403045654296875, -0.407318115234375, -0.2925567626953125, -2.176055908203125, -0.0482177734375, -0.001983642578125, -0.23984527587890625, -0.019317626953125, 0.517822265625, -0.54638671875, 0.556427001953125, 0.3475799560546875, 1.3243255615234375, 0.14888381958007812, 0.5818023681640625, -0.394317626953125, 0.7596435546875, 0.661834716796875, 1.30181884765625, 0.08411407470703125, 1.069091796875, -0.0234832763671875, -0.12847900390625, 0.4197845458984375, 0.1124420166015625, 0.460052490234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000005.npy"}
{"epoch": 0.010471204188481676, "step": 6, "batch_size": 128, "mean": -0.08236503601074219, "std": 0.8720070123672485, "min": -2.18719482421875, "p10": -1.1391540527343749, "median": -0.04315185546875, "p90": 1.0315353393554687, "max": 2.7886962890625, "pos_frac": 0.4609375, "sample": [-0.6715545654296875, -0.01788330078125, -0.1105499267578125, -0.29437255859375, 0.019287109375, 0.1816253662109375, 0.061279296875, -0.12037277221679688, 0.31430816650390625, 0.227783203125, -0.6344528198242188, 1.6425628662109375, 1.1640777587890625, 0.065093994140625, 1.051177978515625, 0.081634521484375, 1.352294921875, -0.16835975646972656, -0.3148651123046875, 0.862884521484375, 1.2294921875, 1.2322120666503906, -0.0097808837890625, 1.0071868896484375, -0.0325927734375, -1.926544189453125, -0.8642578125, -0.580963134765625, 0.4144287109375, 0.09852790832519531, -0.47893524169921875, -0.546905517578125, 0.803192138671875, -0.423370361328125, -0.42137908935546875, -1.538665771484375, -0.7944183349609375, -0.74334716796875, -0.8466949462890625, 1.605224609375, -0.7547454833984375, -1.10943603515625, -1.3526611328125, -0.631439208984375, 0.15895652770996094, -1.26092529296875, 0.24398040771484375, 0.93902587890625, 0.4265289306640625, -2.04388427734375, -0.14876556396484375, -0.32342529296875, 0.437469482421875, -0.5516357421875, -0.62030029296875, 0.182708740234375, 0.476715087890625, -0.6595001220703125, 0.758148193359375, 0.8397216796875, -0.07671737670898438, 1.052459716796875, 0.5865020751953125, 0.42522430419921875, -2.177276611328125, -0.2595977783203125, -0.178863525390625, 1.9437255859375, -2.18719482421875, 0.6762313842773438, 0.0985870361328125, 0.069244384765625, 1.249664306640625, -0.6092071533203125, 0.09157562255859375, -0.180755615234375, 0.25373077392578125, -1.8575515747070312, 0.0, -0.239532470703125, 0.6623001098632812, -0.323760986328125, -1.4062881469726562, -0.19540786743164062, -0.7731170654296875, 0.5247650146484375, 0.6165618896484375, -1.0784912109375, -0.3606147766113281, 0.2220611572265625, -0.84027099609375, -0.617401123046875, 1.0820770263671875, -1.8724365234375, 0.197784423828125, 0.223846435546875, -0.6092071533203125, 0.4148101806640625, -1.20849609375, -1.09710693359375, 0.3544921875, -1.345306396484375, -1.5577545166015625, -0.0537109375, -0.3219146728515625, 1.0231170654296875, 0.888580322265625, -0.71142578125, -0.67218017578125, 0.0709686279296875, -0.505126953125, -0.006885528564453125, 1.650238037109375, 0.08575439453125, 0.5205612182617188, 2.7886962890625, 0.90087890625, 0.7557373046875, -1.077392578125, 0.0132293701171875, 0.1343059539794922, -0.29547119140625, -0.352142333984375, -0.99951171875, -0.255401611328125, -0.6495361328125, -0.10426712036132812, 0.05634307861328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000006.npy"}
{"epoch": 0.012565445026178011, "step": 7, "batch_size": 128, "mean": -0.01760926842689514, "std": 0.9778647422790527, "min": -2.521514892578125, "p10": -1.1527557373046875, "median": -0.0391387939453125, "p90": 1.2637893676757812, "max": 2.94390869140625, "pos_frac": 0.4765625, "sample": [0.10516357421875, 2.407470703125, -1.8179931640625, -0.57818603515625, 1.272430419921875, -0.763214111328125, 0.8052825927734375, -0.639923095703125, 0.063751220703125, -0.7028961181640625, 0.262237548828125, 1.2600860595703125, -2.521514892578125, -0.6011734008789062, -1.1734619140625, -2.091064453125, 0.6705780029296875, 1.3582534790039062, 0.0, -0.035186767578125, -0.344757080078125, -0.300384521484375, -0.5658645629882812, -0.904144287109375, -1.234710693359375, -1.4742431640625, -0.061309814453125, -0.29047393798828125, 0.80987548828125, 0.3013153076171875, -0.30499267578125, 0.616119384765625, -0.41015625, 0.0706787109375, 1.5008544921875, 0.7596435546875, 0.2978973388671875, -0.0430908203125, -1.16888427734375, -0.4146270751953125, -0.67840576171875, -0.5442619323730469, -0.5784149169921875, 1.064453125, -0.825408935546875, 0.3489990234375, 1.208526611328125, 0.4909019470214844, 0.3590412139892578, 2.35626220703125, 1.369873046875, 0.39166259765625, 1.41259765625, -0.620269775390625, 0.43658447265625, 0.18088531494140625, 0.08355712890625, 0.26483154296875, -0.7800369262695312, 1.197113037109375, -0.2222747802734375, 2.1046142578125, 0.5535144805908203, -1.2601318359375, -0.8505859375, 1.055633544921875, -0.50286865234375, 0.2443408966064453, -0.545440673828125, -0.401153564453125, -0.5561370849609375, 2.94390869140625, 0.3054523468017578, -0.4264373779296875, 0.38800811767578125, -1.65399169921875, -0.24576187133789062, -2.095367431640625, -1.58587646484375, -1.881866455078125, -1.01934814453125, 0.18912124633789062, -0.35601806640625, -0.6205673217773438, 0.3406982421875, -0.4216156005859375, 0.48370361328125, 0.92401123046875, 0.563018798828125, 0.01854705810546875, -1.78118896484375, -0.3390960693359375, -0.31754302978515625, 0.034366607666015625, 0.3035736083984375, 0.1173553466796875, 0.208892822265625, -0.11430168151855469, 0.13054656982421875, -0.6001129150390625, -1.116363525390625, -0.178802490234375, -1.145843505859375, 2.524169921875, 0.640777587890625, -0.41375732421875, -0.7659912109375, 0.14678955078125, -0.002960205078125, 0.779449462890625, 0.5175323486328125, -0.33977508544921875, 0.26898193359375, 1.4598388671875, -0.4912109375, -1.097412109375, -0.215240478515625, 0.2488555908203125, 0.852569580078125, 2.52520751953125, -0.7639007568359375, -0.904541015625, 0.15386199951171875, -0.10896873474121094, -0.4220619201660156, -0.5596923828125, 0.507171630859375, 1.27783203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000007.npy"}
{"epoch": 0.014659685863874346, "step": 8, "batch_size": 128, "mean": -0.024483129382133484, "std": 1.0671980381011963, "min": -3.8980712890625, "p10": -1.43753662109375, "median": 0.038448333740234375, "p90": 1.362017822265625, "max": 4.04461669921875, "pos_frac": 0.53125, "sample": [-0.00482177734375, -0.28940582275390625, 0.4784698486328125, -0.764373779296875, -0.680419921875, -0.884765625, 0.1846923828125, 0.028917312622070312, 0.21782684326171875, -1.689453125, -0.17396163940429688, -1.44189453125, 0.470703125, 2.023162841796875, 0.1136474609375, -0.30492401123046875, -1.7570037841796875, 0.4501953125, 0.12678146362304688, 1.739227294921875, 0.65765380859375, 1.664398193359375, -1.307769775390625, -0.34332275390625, 0.3053436279296875, 0.5080413818359375, 1.0028076171875, 0.307403564453125, -1.4356689453125, -2.2381591796875, -0.954315185546875, 4.04461669921875, -0.354766845703125, -0.41632080078125, -0.0157470703125, -1.508514404296875, -0.006011962890625, -0.566162109375, -1.655548095703125, -2.0699462890625, -0.63555908203125, -2.2337646484375, 0.042694091796875, -0.6551361083984375, 0.79644775390625, -1.4843597412109375, 0.12140274047851562, 0.11559295654296875, -0.512542724609375, 1.11846923828125, -0.703399658203125, -1.1984100341796875, -0.422271728515625, 0.05278205871582031, 0.439910888671875, -1.633331298828125, 0.62982177734375, 0.543182373046875, -0.926300048828125, -1.300994873046875, 0.236968994140625, 1.3599853515625, -1.252197265625, 0.3123779296875, 0.341949462890625, 1.529541015625, -1.148773193359375, 0.18509674072265625, -3.8980712890625, 2.13751220703125, -1.024383544921875, 0.6575927734375, 0.114166259765625, -0.4223175048828125, 1.36676025390625, -0.288421630859375, 1.5262451171875, 0.6199188232421875, 0.07181167602539062, -1.7479095458984375, -0.41808319091796875, 1.60894775390625, -0.773406982421875, 0.6177902221679688, -0.5727081298828125, -0.04788017272949219, 0.9666900634765625, 0.0772705078125, 0.3359184265136719, 2.02587890625, 0.16143798828125, -0.524688720703125, 0.6632080078125, 0.03420257568359375, -0.3134307861328125, 0.4351959228515625, 1.097900390625, 0.63800048828125, -0.17073822021484375, 0.951019287109375, -0.3012237548828125, 0.10382080078125, 0.711761474609375, 0.2611083984375, -1.1550025939941406, 0.3427543640136719, -0.48974609375, -0.2498321533203125, 0.971435546875, -0.31549072265625, -0.27860260009765625, 0.738800048828125, 1.6197509765625, 1.1533203125, 0.97119140625, 0.829315185546875, -0.997528076171875, -0.776336669921875, 0.47235107421875, 0.0201568603515625, 0.1975555419921875, -1.7364959716796875, 1.4633026123046875, -0.88543701171875, 0.012577056884765625, -0.1329498291015625, 1.37200927734375, -0.14162826538085938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000008.npy"}
{"epoch": 0.016753926701570682, "step": 9, "batch_size": 128, "mean": 0.017218172550201416, "std": 0.9326652884483337, "min": -2.48931884765625, "p10": -1.06083984375, "median": -0.016033172607421875, "p90": 1.164772033691406, "max": 2.50469970703125, "pos_frac": 0.4765625, "sample": [-0.22533416748046875, -0.92138671875, -0.0656585693359375, 0.3764533996582031, -0.61187744140625, -0.6707763671875, 0.027099609375, -0.37918853759765625, 0.8023681640625, 0.089874267578125, -0.6894378662109375, -1.104827880859375, 0.761199951171875, -0.325408935546875, -0.950836181640625, 1.843505859375, 1.961151123046875, -0.83251953125, 1.1076202392578125, -0.5043258666992188, -0.03206634521484375, 1.132110595703125, -0.4488372802734375, 0.80908203125, -0.2378692626953125, -0.5244293212890625, -0.59503173828125, 0.282135009765625, 0.0624847412109375, -2.1162109375, -0.94036865234375, 0.021484375, -1.0530548095703125, -0.729278564453125, -1.138519287109375, -0.13901519775390625, 0.910308837890625, -0.3506317138671875, 0.531005859375, -0.5163116455078125, -0.496185302734375, -0.35907745361328125, 1.57147216796875, -0.3820343017578125, 0.503509521484375, -2.39398193359375, -1.06536865234375, -1.30645751953125, -1.2916259765625, -0.6797714233398438, -0.289154052734375, -1.05889892578125, -0.71502685546875, 0.6606292724609375, 2.400909423828125, 0.6192626953125, 0.17022705078125, 0.804656982421875, -0.91912841796875, 0.490081787109375, 0.1342926025390625, -0.7466354370117188, -1.1383056640625, -0.23682022094726562, -0.24181365966796875, 1.45648193359375, 0.6232070922851562, 1.78961181640625, -1.164886474609375, 1.2409820556640625, -0.71826171875, 0.14849281311035156, -2.48931884765625, 0.7961387634277344, 0.7373809814453125, -0.3125457763671875, -0.252166748046875, 0.9717254638671875, -0.3201637268066406, -0.483367919921875, 1.520477294921875, 0.43218994140625, -0.427398681640625, -0.0982666015625, 1.4231719970703125, -0.579345703125, 0.0, 0.0623626708984375, 0.78619384765625, -1.09375, 0.3170166015625, 0.45829010009765625, 0.8033905029296875, 0.36716461181640625, 0.8284454345703125, 0.3182373046875, -0.207489013671875, -1.552215576171875, -0.37821197509765625, 1.101593017578125, 0.15077590942382812, 0.0, 0.879150390625, 1.72265625, 0.865753173828125, 0.03914833068847656, 0.0, 0.26605224609375, 0.20965194702148438, 0.3343009948730469, 2.06689453125, -0.05728912353515625, 0.342132568359375, 0.448394775390625, -0.727142333984375, 0.260467529296875, -0.9447479248046875, -0.66162109375, -0.257537841796875, 0.11602783203125, -0.3388824462890625, 2.22412109375, 0.02503204345703125, -0.37335205078125, -1.8427200317382812, 0.615966796875, 2.50469970703125, -0.418609619140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000009.npy"}
{"epoch": 0.018848167539267015, "step": 10, "batch_size": 128, "mean": -0.0256330668926239, "std": 0.815056324005127, "min": -2.29925537109375, "p10": -1.1256378173828125, "median": 0.0090789794921875, "p90": 0.9759735107421874, "max": 2.1419677734375, "pos_frac": 0.5, "sample": [0.864501953125, -0.4927825927734375, 0.5458984375, -0.7081680297851562, 0.3497314453125, -0.4661865234375, -0.2989959716796875, 0.324310302734375, -0.359039306640625, -0.6201248168945312, 0.3349609375, 0.2426605224609375, -0.5581588745117188, 1.5885009765625, 0.4508056640625, 0.12837982177734375, 0.03272247314453125, -1.2257080078125, 0.02935791015625, 0.970123291015625, 0.6953125, 0.0287628173828125, 0.2906494140625, -0.58782958984375, 0.0801849365234375, 0.3462104797363281, 1.468017578125, -0.00775909423828125, -0.548126220703125, -0.1365966796875, -0.0597076416015625, 0.74169921875, -0.2681388854980469, 0.40594482421875, -0.6016197204589844, -0.12811279296875, 1.935302734375, -0.2031402587890625, 0.841827392578125, 0.5803985595703125, -1.13958740234375, -0.35010528564453125, -0.3387603759765625, -0.876678466796875, -1.3209228515625, -0.012298583984375, -0.7206993103027344, 0.42801666259765625, 0.242431640625, -0.2938385009765625, 0.2078380584716797, -0.47095489501953125, 0.92047119140625, -1.2161407470703125, -0.6980743408203125, -0.13958740234375, -0.6604766845703125, 1.138397216796875, 0.08138275146484375, 0.83697509765625, -1.1868438720703125, -0.630096435546875, -1.2460479736328125, 0.1416168212890625, -0.35306549072265625, -0.64801025390625, -0.4126739501953125, -0.57708740234375, 1.4874267578125, 0.018157958984375, 0.0626068115234375, -0.3055572509765625, 0.253997802734375, 0.5, -0.9148635864257812, 0.66912841796875, 0.10808563232421875, 0.9896240234375, -0.696075439453125, 1.3701171875, -0.6733016967773438, 2.1419677734375, 0.7362060546875, 1.971282958984375, 0.3904094696044922, 0.0223541259765625, 0.10223388671875, -0.16268157958984375, -1.1533203125, -1.67333984375, -0.1650848388671875, 0.05617713928222656, 0.09806060791015625, -0.30816650390625, -1.119659423828125, -0.7203216552734375, 0.18663787841796875, 0.3958740234375, 0.18270111083984375, -0.0574493408203125, 1.121063232421875, -0.4865875244140625, 0.083953857421875, 0.09637451171875, -1.67694091796875, -0.2976531982421875, -0.6641998291015625, -2.29925537109375, 0.7200088500976562, 0.417205810546875, -1.09503173828125, 0.07794380187988281, 0.87335205078125, 0.0, 0.34255218505859375, -1.21246337890625, -0.3318061828613281, -1.3179931640625, -0.37618255615234375, -0.83209228515625, 1.2730560302734375, 0.54998779296875, 1.59259033203125, 0.43956756591796875, 2.094451904296875, -0.986663818359375, -1.242645263671875, -0.648101806640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000010.npy"}
{"epoch": 0.020942408376963352, "step": 11, "batch_size": 128, "mean": 0.0076267048716545105, "std": 0.9092163443565369, "min": -2.878662109375, "p10": -1.1940521240234374, "median": 0.069091796875, "p90": 1.0300231933593746, "max": 2.668243408203125, "pos_frac": 0.5625, "sample": [-0.27288818359375, -1.88714599609375, -0.187225341796875, 1.2767333984375, 0.6073760986328125, 0.62493896484375, -0.499603271484375, -0.08099365234375, -0.8280029296875, 0.48895263671875, 0.23195648193359375, 0.3474884033203125, 0.976318359375, 1.281768798828125, 0.386505126953125, 0.1213226318359375, 0.50360107421875, -0.1808023452758789, 0.04559326171875, -0.38311767578125, -0.03728294372558594, 0.0782012939453125, 0.53717041015625, 1.3309326171875, -0.7260894775390625, -0.2166748046875, 0.683319091796875, 1.40142822265625, 1.3028335571289062, -0.6162643432617188, 0.11993408203125, 0.359954833984375, 0.3575592041015625, -1.1881103515625, -0.2057342529296875, 2.5740814208984375, -1.758880615234375, 0.3170051574707031, 1.78765869140625, 0.009128570556640625, 0.6156005859375, -1.33160400390625, -2.878662109375, 0.102294921875, -0.505126953125, -0.434661865234375, 0.01373291015625, 1.6077880859375, 2.2218017578125, -0.6322021484375, 0.5006866455078125, 0.28021240234375, 0.391021728515625, -1.35040283203125, -1.726898193359375, -2.724853515625, -0.3266754150390625, -0.01519775390625, 0.3763427734375, -0.301971435546875, 0.48052978515625, 0.812164306640625, -1.207916259765625, -0.933441162109375, 0.75567626953125, 0.14982986450195312, 0.0929718017578125, 0.0598297119140625, 0.29854583740234375, 0.118927001953125, 0.503143310546875, 0.1708526611328125, 0.391876220703125, -0.59796142578125, 0.20537567138671875, -0.1443634033203125, -0.27364349365234375, -0.5591926574707031, -0.048370361328125, 1.00433349609375, 0.5186309814453125, 1.271240234375, -0.33544158935546875, 0.21379852294921875, -1.4305877685546875, 0.32578277587890625, -0.5697174072265625, -0.100494384765625, -1.412200927734375, 0.69097900390625, 0.6611480712890625, -1.1060028076171875, -1.769744873046875, 0.24518203735351562, -0.81646728515625, -0.4005126953125, 0.4808349609375, -0.815216064453125, -0.3386802673339844, 0.0068798065185546875, -1.0129165649414062, -1.262786865234375, 1.4110107421875, -0.822174072265625, -0.04925537109375, 1.0899658203125, 0.3800048828125, 0.3324432373046875, 0.542572021484375, 0.040740966796875, 0.2575225830078125, -0.884033203125, -0.71844482421875, -0.7462234497070312, -0.821319580078125, -1.421661376953125, 0.0871124267578125, 0.631103515625, 0.255126953125, 0.9898834228515625, 0.906463623046875, 0.0599822998046875, 0.5740509033203125, 0.0406494140625, -0.303436279296875, -0.31946563720703125, -0.0916900634765625, 2.668243408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000011.npy"}
{"epoch": 0.023036649214659685, "step": 12, "batch_size": 128, "mean": -0.07332716882228851, "std": 0.8809601664543152, "min": -2.3472900390625, "p10": -1.169635009765625, "median": -0.06417465209960938, "p90": 1.0659210205078125, "max": 2.117431640625, "pos_frac": 0.46875, "sample": [-0.414306640625, 0.027984619140625, 0.3187255859375, 1.0479736328125, -0.021741867065429688, -1.1096267700195312, -1.0840911865234375, 0.109619140625, -1.1597900390625, -0.38994598388671875, -0.8470458984375, 0.153839111328125, -1.765106201171875, 0.3560943603515625, 0.663177490234375, 0.2767333984375, 0.4492950439453125, 0.1952972412109375, 0.18267822265625, 0.606903076171875, -1.871978759765625, 1.2513427734375, 2.117431640625, -0.1292572021484375, 0.221771240234375, -0.6113433837890625, -0.8159885406494141, -0.3720703125, 0.25384521484375, 0.2298126220703125, -2.3472900390625, 0.743988037109375, -0.499969482421875, -0.83099365234375, -0.637176513671875, 1.779510498046875, -0.11219024658203125, 0.6683807373046875, 1.207794189453125, -0.07930755615234375, -0.40093994140625, -1.888671875, 0.381378173828125, 1.643585205078125, 0.30345916748046875, -0.5249710083007812, -0.412384033203125, 0.4352874755859375, -0.109283447265625, 0.40911865234375, 1.2040252685546875, -0.2632293701171875, 0.8548583984375, -0.835205078125, -0.5449981689453125, -0.049041748046875, 1.06842041015625, 1.513153076171875, -1.211822509765625, -0.04742431640625, -0.81561279296875, -0.333740234375, 0.08331298828125, -0.699432373046875, 0.24551773071289062, 1.064849853515625, 0.36779022216796875, 1.52386474609375, -0.5474472045898438, 0.3030548095703125, -0.2388763427734375, -1.17822265625, 1.042572021484375, -0.1279296875, -0.31549835205078125, -1.433929443359375, 0.649078369140625, -0.6414031982421875, -0.7772674560546875, -1.59967041015625, 0.5744171142578125, -0.7030181884765625, -1.22308349609375, 0.9873580932617188, 0.36359596252441406, -0.726287841796875, 0.06491851806640625, 0.070953369140625, -1.335693359375, -0.7996826171875, 0.82940673828125, 0.38330078125, -0.680816650390625, -0.320037841796875, 0.136505126953125, 0.964599609375, -0.22723388671875, 0.12039947509765625, -1.032806396484375, 1.5191650390625, -0.197021484375, -0.2164154052734375, 0.80450439453125, 0.90496826171875, -0.8446044921875, 0.172576904296875, -0.181549072265625, 0.071380615234375, 1.6278839111328125, 0.61639404296875, -0.40960693359375, -1.23480224609375, 0.3309478759765625, -1.16595458984375, 0.8093414306640625, -0.014068603515625, 0.08712005615234375, -2.19488525390625, -1.032196044921875, -0.9058685302734375, 1.128997802734375, -0.382293701171875, -0.31805419921875, -1.04827880859375, 1.819366455078125, -1.0052604675292969, -1.2158966064453125, -0.217864990234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000012.npy"}
{"epoch": 0.025130890052356022, "step": 13, "batch_size": 128, "mean": -0.056239962577819824, "std": 0.9144258499145508, "min": -2.997894287109375, "p10": -0.999749755859375, "median": -0.062023162841796875, "p90": 1.0119293212890625, "max": 2.59478759765625, "pos_frac": 0.4375, "sample": [-0.16951751708984375, 0.1502838134765625, -0.08837890625, 0.13385009765625, 0.31988525390625, -1.1141357421875, 1.6131591796875, -0.9874267578125, 0.2251129150390625, 0.32183074951171875, -0.060821533203125, -2.974212646484375, 0.27630615234375, -0.321624755859375, 0.11337089538574219, 0.5130863189697266, -0.48895263671875, 1.8165283203125, 0.8844146728515625, -0.22003173828125, -0.0637054443359375, -1.89776611328125, 1.4647216796875, -0.24579620361328125, 0.9835205078125, -0.26395225524902344, -0.010103225708007812, -0.867828369140625, -0.54376220703125, -0.82763671875, 0.452880859375, -0.34571075439453125, 1.62506103515625, -0.2572174072265625, -0.71038818359375, 0.08223724365234375, 0.18603134155273438, -0.8009033203125, 0.5535240173339844, -0.225860595703125, 0.4915142059326172, 0.231903076171875, 0.6451873779296875, -0.098541259765625, -0.149993896484375, 1.024383544921875, -1.02850341796875, -1.3409271240234375, 1.155059814453125, -2.53143310546875, -0.1403656005859375, -2.997894287109375, 0.2579841613769531, 0.285308837890625, 0.66229248046875, 0.0, -1.388916015625, -0.420562744140625, 0.32028961181640625, -0.0817413330078125, -0.4193572998046875, -0.725982666015625, -0.7129745483398438, -0.2109832763671875, 2.59478759765625, -1.877197265625, 0.480438232421875, -0.323028564453125, 0.19024658203125, -0.05132865905761719, 1.7415771484375, -0.09619140625, -0.05028533935546875, 1.454742431640625, -0.0960693359375, 0.878448486328125, 0.14447021484375, 1.41070556640625, 0.11579132080078125, -0.3021392822265625, 0.1440582275390625, 0.5432968139648438, 0.9280319213867188, -0.650299072265625, -0.2321014404296875, 1.006591796875, -0.4528465270996094, 0.058032989501953125, 1.330780029296875, 0.3300018310546875, 0.766754150390625, -0.00571441650390625, 0.77532958984375, -0.39881134033203125, 0.24358367919921875, -0.692596435546875, -0.06322479248046875, -0.76422119140625, -0.9310760498046875, -0.16546630859375, -2.842376708984375, 0.31406402587890625, -0.5404205322265625, -0.2935943603515625, 0.2791595458984375, -1.67291259765625, 1.887298583984375, 0.087890625, 0.4044036865234375, -0.074554443359375, -0.05264091491699219, 0.450836181640625, -0.4207305908203125, -0.2544994354248047, 0.34765625, 1.326812744140625, -0.546661376953125, -0.52850341796875, -0.8691558837890625, 0.29840087890625, -0.106689453125, -1.1573944091796875, -0.66778564453125, -0.44985198974609375, -0.024017333984375, 0.46620941162109375, -1.4161834716796875, -0.18436431884765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000013.npy"}
{"epoch": 0.027225130890052355, "step": 14, "batch_size": 128, "mean": 0.0517846941947937, "std": 0.8244470357894897, "min": -2.857391357421875, "p10": -0.6980644226074219, "median": -0.013751983642578125, "p90": 1.0444564819335935, "max": 2.42803955078125, "pos_frac": 0.4765625, "sample": [0.046600341796875, 0.7207260131835938, -0.6940841674804688, -0.4744873046875, -0.0207366943359375, -0.684844970703125, -0.1234283447265625, 1.92413330078125, 0.2174072265625, 0.9188995361328125, 0.35950469970703125, -0.29843902587890625, -1.736083984375, -0.210296630859375, 1.492218017578125, 0.41717529296875, 0.245819091796875, -0.185638427734375, -0.34407806396484375, 0.2529754638671875, -0.3622627258300781, -0.02695465087890625, 1.27569580078125, -0.588165283203125, 0.919403076171875, -0.50042724609375, -0.144989013671875, -0.37432098388671875, -1.782562255859375, -0.11541748046875, 0.19281005859375, -0.6060791015625, 0.09698486328125, -0.355926513671875, 0.07511138916015625, -1.14678955078125, 0.576263427734375, -0.5207500457763672, 1.01141357421875, -1.002288818359375, 0.11646652221679688, 0.472015380859375, -0.4761199951171875, -0.640625, 0.0, 0.020763397216796875, -0.35711669921875, 0.513275146484375, 0.01065826416015625, -0.7073516845703125, 0.6166229248046875, 1.6469573974609375, 0.1942310333251953, 0.88262939453125, 0.0457000732421875, -2.321136474609375, 0.2899017333984375, 1.111907958984375, -0.022186279296875, -0.036651611328125, -0.066162109375, -2.857391357421875, 1.35748291015625, 0.4342041015625, -0.0607147216796875, -0.565582275390625, 0.07794189453125, 1.0084228515625, 1.603729248046875, -0.25048828125, 0.256256103515625, -1.5064544677734375, -0.7998886108398438, 1.215576171875, 2.42803955078125, 0.9161453247070312, -0.4759368896484375, 2.21240234375, 0.79681396484375, -0.10999298095703125, -0.3104705810546875, -0.090576171875, -0.0012664794921875, -0.31798553466796875, 0.10614013671875, -0.3957405090332031, -0.41475868225097656, -0.5992660522460938, -0.3374481201171875, -0.22738075256347656, -0.50054931640625, -0.8507003784179688, 1.514801025390625, -0.2413482666015625, -0.8775634765625, 1.280059814453125, -0.36594390869140625, -0.4539031982421875, 0.028564453125, -0.247161865234375, 0.609039306640625, -0.4037628173828125, 0.2044525146484375, 0.7691650390625, 0.10888290405273438, 0.044830322265625, -0.2890472412109375, 0.4766082763671875, -0.0782623291015625, 0.7071380615234375, 0.308837890625, -0.1976318359375, -1.210968017578125, -0.6388015747070312, -0.00676727294921875, -1.88232421875, 0.21563720703125, 1.655914306640625, 0.32904052734375, -0.053630828857421875, 0.43212890625, 1.0155487060546875, -0.0419921875, 0.685089111328125, -0.09543609619140625, 0.72296142578125, 0.60296630859375, 0.52288818359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000014.npy"}
{"epoch": 0.02931937172774869, "step": 15, "batch_size": 128, "mean": -0.06755267083644867, "std": 0.9587298035621643, "min": -3.43017578125, "p10": -1.1253906249999999, "median": -0.043285369873046875, "p90": 1.0644752502441406, "max": 1.968414306640625, "pos_frac": 0.4921875, "sample": [0.865753173828125, -0.7525634765625, 1.10736083984375, 0.36102294921875, -1.1153564453125, 1.0790176391601562, -0.766998291015625, -0.16949462890625, 0.0551300048828125, 0.18991851806640625, -0.05438804626464844, -0.214111328125, -0.4063873291015625, -0.24335861206054688, 0.87274169921875, -2.3668212890625, 0.7855758666992188, 0.12207794189453125, 0.10990142822265625, -0.9066619873046875, 0.1649169921875, -0.6266555786132812, -0.809661865234375, -1.959442138671875, 0.0363616943359375, 0.4309844970703125, -0.1126708984375, -3.43017578125, -0.089508056640625, -0.20416259765625, 1.81060791015625, -0.171783447265625, -0.740478515625, -0.22605133056640625, 0.397308349609375, -0.20293426513671875, 1.5244140625, -1.0548095703125, 1.397369384765625, -0.3868560791015625, -0.19104766845703125, -2.211761474609375, 0.30712890625, -1.3316421508789062, 0.3170166015625, -0.86846923828125, 0.001522064208984375, -0.6595458984375, 0.1244354248046875, 0.029266357421875, -0.5380744934082031, -0.04460906982421875, 0.436676025390625, 0.049224853515625, 0.239990234375, -1.0059814453125, 0.3377685546875, -0.711578369140625, -0.0950469970703125, 0.51153564453125, 0.3943939208984375, 0.5809268951416016, 0.052001953125, -0.3644256591796875, -0.041961669921875, -0.1381378173828125, 0.2269744873046875, -0.613677978515625, 1.968414306640625, -1.769561767578125, 0.14725112915039062, -0.14132308959960938, 0.1600341796875, -2.54583740234375, -0.552154541015625, 0.8605499267578125, -0.21044921875, 1.1523895263671875, 0.5569915771484375, 0.8255767822265625, 1.799713134765625, -0.05667304992675781, -0.0760498046875, -0.7103271484375, -2.0782470703125, 0.29132080078125, -0.6015625, 1.753662109375, 0.1269702911376953, 0.6348876953125, -0.2633056640625, -0.4832916259765625, -0.1328277587890625, -1.5804901123046875, 0.916778564453125, 1.88671875, -0.1593494415283203, -2.0281982421875, 1.031494140625, 0.781982421875, -1.1488037109375, 1.658416748046875, 1.49127197265625, 1.3611526489257812, 0.1854248046875, 0.134735107421875, -2.83148193359375, -0.054187774658203125, -0.3342132568359375, -0.4702796936035156, 0.7351875305175781, 0.1136322021484375, -0.41495513916015625, 0.44075775146484375, 0.1016845703125, 0.32997894287109375, -1.88623046875, 0.39159393310546875, 0.32816314697265625, -0.14227294921875, 0.49249267578125, 0.14495849609375, -0.4171600341796875, -0.56085205078125, 0.7767333984375, 1.0582427978515625, -0.27472686767578125, -0.453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000015.npy"}
{"epoch": 0.031413612565445025, "step": 16, "batch_size": 128, "mean": 0.0075082927942276, "std": 0.8926595449447632, "min": -2.512847900390625, "p10": -1.1294403076171875, "median": -0.024200439453125, "p90": 1.0984191894531246, "max": 2.67529296875, "pos_frac": 0.4921875, "sample": [-0.1771392822265625, 0.27692413330078125, 0.41114044189453125, 0.454071044921875, 1.267547607421875, -0.1201934814453125, 0.431304931640625, 2.6004638671875, 2.67529296875, -1.48602294921875, 0.17696762084960938, 0.04431915283203125, 0.42195892333984375, -0.51513671875, 0.54345703125, -0.18865966796875, -0.32257080078125, 0.734954833984375, -1.76739501953125, 0.48486328125, 0.8189544677734375, 0.1584930419921875, 0.06640625, -1.089202880859375, 0.47906494140625, -0.309295654296875, 0.323760986328125, -0.091796875, -0.4280548095703125, -0.5142059326171875, 1.4932861328125, -0.04840087890625, -0.52032470703125, -1.421142578125, -0.069671630859375, -0.678680419921875, 1.75921630859375, -0.16414642333984375, -0.16097640991210938, -0.3877754211425781, 1.0174713134765625, -0.302947998046875, -0.1905670166015625, 1.211181640625, -0.5081939697265625, -0.18355751037597656, -1.134033203125, 0.1915130615234375, -0.833099365234375, -0.06675338745117188, 0.007755279541015625, 0.83209228515625, 0.29065704345703125, -2.41802978515625, -0.62054443359375, 0.7098388671875, 0.0158538818359375, -0.3885345458984375, 1.0430526733398438, 0.0709228515625, 0.54248046875, -0.056507110595703125, 0.6885223388671875, -0.7251129150390625, -0.290924072265625, 1.0604248046875, -0.34629058837890625, 0.131317138671875, -2.083984375, -1.675262451171875, 0.4574317932128906, 0.708282470703125, 0.7020187377929688, 1.026702880859375, 0.789825439453125, -1.349884033203125, -0.321380615234375, -0.05096435546875, -0.11730575561523438, -0.8228759765625, 0.01848602294921875, -0.6222763061523438, -1.4701995849609375, -0.611541748046875, -1.460174560546875, 1.564727783203125, 0.0, 0.43187713623046875, 0.44732666015625, -0.245147705078125, -1.132537841796875, -1.12811279296875, -0.72418212890625, -0.61309814453125, 1.31976318359375, -0.34368896484375, 0.16407394409179688, 1.2716827392578125, -0.4091949462890625, 0.6990776062011719, 1.00830078125, 1.300323486328125, 0.26483154296875, 0.070892333984375, -0.5353889465332031, -0.13564300537109375, 0.25787353515625, -0.09002685546875, -1.477508544921875, 1.18707275390625, 0.6346282958984375, -0.41316986083984375, -1.10302734375, -0.180755615234375, -1.029541015625, 0.174285888671875, 0.6954498291015625, 0.6605682373046875, -0.6207275390625, 0.263397216796875, 0.2639312744140625, -0.373779296875, 1.640838623046875, -2.512847900390625, 1.522552490234375, 0.35638427734375, 0.36627197265625, -0.563232421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000016.npy"}
{"epoch": 0.033507853403141365, "step": 17, "batch_size": 128, "mean": 0.0687805712223053, "std": 0.8452976942062378, "min": -2.40478515625, "p10": -0.8274032592773436, "median": 0.015352249145507812, "p90": 0.9908905029296873, "max": 3.10198974609375, "pos_frac": 0.5078125, "sample": [-0.00478363037109375, -0.6502532958984375, 3.10198974609375, -0.057464599609375, -0.4269561767578125, -0.2053375244140625, 1.090118408203125, -1.00201416015625, 0.377532958984375, 0.5679550170898438, 0.8388595581054688, -0.9790802001953125, -0.50433349609375, 0.12811279296875, 0.72503662109375, 0.5466995239257812, 0.6678466796875, 1.1739501953125, 0.84503173828125, 2.5211181640625, 0.6663818359375, 2.2060699462890625, -1.212677001953125, -0.5670166015625, 0.29352569580078125, 1.7274169921875, 0.30413818359375, 0.32611083984375, -1.6314697265625, -0.7880859375, 0.0959625244140625, -0.3408203125, -0.03520965576171875, 0.10821533203125, -0.5027008056640625, 0.1824951171875, 0.48150634765625, -2.40478515625, 0.12516021728515625, -0.24884796142578125, 0.83734130859375, -0.4470672607421875, -0.2676544189453125, -0.4046592712402344, -0.498809814453125, -0.08512115478515625, 0.2150421142578125, 0.0414276123046875, 1.4533843994140625, 0.6179046630859375, -0.54815673828125, 0.7918014526367188, -0.7889251708984375, 0.5670166015625, 0.06670379638671875, 1.28668212890625, 0.011844635009765625, 0.616119384765625, 0.07903289794921875, 1.088623046875, 0.9305496215820312, -0.32736968994140625, -0.46681976318359375, -0.9338150024414062, -0.663543701171875, -1.2078399658203125, -1.6097412109375, -0.19598388671875, -0.42624664306640625, 0.1815643310546875, -0.6375732421875, -0.7042236328125, -0.230194091796875, -1.4869232177734375, 0.01885986328125, -0.19873046875, 0.3449859619140625, 0.5607528686523438, 2.116180419921875, 0.07085800170898438, 0.968048095703125, -0.961456298828125, -0.6223297119140625, -0.655731201171875, -0.459564208984375, -0.132537841796875, 1.044189453125, 0.706573486328125, -0.6741943359375, 2.279876708984375, 0.25479888916015625, 0.7374267578125, 0.100616455078125, -0.7958221435546875, 0.669677734375, -0.450775146484375, -0.7545547485351562, -0.587646484375, 0.2991485595703125, -1.0316162109375, -0.6850090026855469, -0.564361572265625, -0.113677978515625, 1.29931640625, -0.44460296630859375, -0.186798095703125, 0.6602210998535156, -0.1605377197265625, 0.6248321533203125, -0.901092529296875, -0.2580680847167969, 0.19371795654296875, -0.683319091796875, -0.235626220703125, -0.94537353515625, -0.048004150390625, 0.2890777587890625, 0.91619873046875, -0.437957763671875, 0.40252685546875, -0.3318214416503906, 0.7021331787109375, 0.484130859375, 0.21469879150390625, 0.4478759765625, 0.5968780517578125, -0.0442047119140625, 0.7719573974609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000017.npy"}
{"epoch": 0.0356020942408377, "step": 18, "batch_size": 128, "mean": -0.06053653359413147, "std": 0.9244751334190369, "min": -2.75738525390625, "p10": -1.1078720092773438, "median": -0.042572021484375, "p90": 1.0479568481445312, "max": 3.060760498046875, "pos_frac": 0.484375, "sample": [-1.1094207763671875, -0.593658447265625, -0.3595848083496094, 1.0650787353515625, 0.3209800720214844, -0.61749267578125, -0.12503433227539062, -0.25653839111328125, 0.10538482666015625, 0.10464859008789062, 0.01190185546875, 1.18231201171875, -0.16475677490234375, 0.1507568359375, 1.8804473876953125, 0.2339019775390625, -0.59173583984375, 1.47906494140625, -0.003265380859375, 0.08327484130859375, 1.6071929931640625, 1.1349411010742188, -0.650543212890625, -0.3921356201171875, 0.678985595703125, -0.3817558288574219, -1.624664306640625, 0.8997650146484375, -0.3421783447265625, 0.3878669738769531, 0.404083251953125, -0.62225341796875, -0.1387786865234375, -0.532562255859375, -0.3936920166015625, -0.16275787353515625, 0.930908203125, 0.4611377716064453, -0.56658935546875, 0.81695556640625, 0.20774078369140625, 0.922576904296875, -2.3323974609375, 0.42083740234375, 0.46723175048828125, 0.082061767578125, -0.82720947265625, 0.073394775390625, 0.079864501953125, -0.197418212890625, 0.0769805908203125, 0.191986083984375, 0.7012481689453125, -0.02477264404296875, -2.41302490234375, -0.52166748046875, -0.5026702880859375, 1.0185546875, -0.664398193359375, 1.1395111083984375, 0.81292724609375, 1.08306884765625, -0.49066162109375, -2.17987060546875, 0.00567626953125, -0.30084228515625, -1.42791748046875, 3.060760498046875, -2.202972412109375, -1.67864990234375, 0.25346946716308594, 0.3286399841308594, -0.7794647216796875, -0.29351806640625, 0.2023162841796875, -0.06366348266601562, -2.75738525390625, -0.858123779296875, 0.34633636474609375, 0.1055450439453125, -0.74884033203125, -0.06037139892578125, -0.31512451171875, 1.8058624267578125, 0.195526123046875, 0.18650054931640625, -0.07472991943359375, -2.1702880859375, 0.44989013671875, -0.2928466796875, 0.75146484375, 0.0706024169921875, -0.1148223876953125, -0.21320343017578125, -0.64453125, 0.748779296875, 0.14267349243164062, -0.678375244140625, -0.636260986328125, 0.6269073486328125, -0.22121429443359375, 1.468994140625, -0.285125732421875, -0.46771240234375, 0.67901611328125, -0.76068115234375, -0.1590576171875, 0.67755126953125, -0.50628662109375, 0.6060333251953125, -1.090240478515625, -0.11080169677734375, 0.1439208984375, 1.550048828125, 1.2088470458984375, 1.028076171875, -0.73272705078125, 1.040618896484375, -0.30975341796875, 0.7481689453125, -1.39398193359375, 0.2469329833984375, -1.107208251953125, -1.531341552734375, -0.79888916015625, -0.619140625, 0.2340545654296875, -1.68988037109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000018.npy"}
{"epoch": 0.03769633507853403, "step": 19, "batch_size": 128, "mean": -0.21028956770896912, "std": 0.8335183262825012, "min": -2.551422119140625, "p10": -1.262001037597656, "median": -0.17927265167236328, "p90": 0.8420852661132812, "max": 2.35675048828125, "pos_frac": 0.40625, "sample": [-0.12717056274414062, 0.352691650390625, -0.6579437255859375, 1.7550048828125, -0.507080078125, -1.8154296875, -1.444488525390625, 0.035980224609375, -0.241455078125, -1.011962890625, 0.03021240234375, -0.24539566040039062, 0.648956298828125, -1.4000244140625, 0.6425971984863281, -0.6282005310058594, -1.1497650146484375, 0.537353515625, 0.9979934692382812, 0.20751953125, 0.40799713134765625, 0.18548583984375, -0.1937732696533203, 0.8001708984375, -0.575408935546875, 0.22772216796875, 0.05347442626953125, -1.1974716186523438, 0.428955078125, 0.5344009399414062, -1.06781005859375, 1.16900634765625, -2.1815185546875, -0.7637939453125, -0.925872802734375, -0.16477203369140625, -0.5052490234375, -0.046051025390625, -0.132598876953125, 0.10242080688476562, 0.14327239990234375, 0.084747314453125, -2.285614013671875, -0.9534759521484375, 1.6676025390625, 0.652587890625, -0.13463211059570312, -0.0782928466796875, -1.228057861328125, -0.440521240234375, 0.3155021667480469, -2.551422119140625, -0.9374237060546875, -0.3365325927734375, -0.44158935546875, 0.33258056640625, -1.42645263671875, -0.71533203125, 0.36985015869140625, -0.31613922119140625, -0.5692367553710938, -0.9399261474609375, -1.44610595703125, -0.16260910034179688, 0.783843994140625, -1.096435546875, -0.42169189453125, 0.8695526123046875, -0.6364593505859375, 0.8555755615234375, 0.021484375, -1.40057373046875, -1.450775146484375, -1.810760498046875, 0.34417724609375, -0.3064422607421875, 0.56036376953125, 0.19519805908203125, -0.9308319091796875, 0.5894775390625, -0.284332275390625, 0.101806640625, 0.10512161254882812, -0.15948486328125, 0.8363037109375, -1.0102081298828125, -0.19921302795410156, 1.10003662109375, -0.26818084716796875, -0.9427490234375, 0.322021484375, -1.0435791015625, 1.0428619384765625, 0.9045867919921875, -0.392425537109375, 1.28155517578125, 0.6103668212890625, -0.5161972045898438, 0.1829833984375, 0.3008842468261719, -0.5331649780273438, -0.5804290771484375, -0.9461517333984375, 1.163116455078125, -0.039794921875, -0.516998291015625, 0.891082763671875, -0.6898193359375, 0.1385498046875, -0.04803466796875, 0.18121337890625, -0.334320068359375, -0.286285400390625, -1.3412017822265625, -0.36083984375, -0.400177001953125, -0.7489013671875, -0.06976318359375, 2.35675048828125, 0.33165740966796875, -0.347381591796875, -0.054412841796875, 0.110107421875, -0.8368072509765625, 0.22611236572265625, -1.9452972412109375, -0.4453277587890625, -0.6658935546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000019.npy"}
{"epoch": 0.039790575916230364, "step": 20, "batch_size": 128, "mean": 0.005584493279457092, "std": 0.8136693239212036, "min": -2.2662353515625, "p10": -0.81669921875, "median": -0.0882720947265625, "p90": 1.0170806884765624, "max": 2.65484619140625, "pos_frac": 0.4375, "sample": [-0.2506561279296875, 1.1400146484375, 2.515380859375, 0.385498046875, -0.723876953125, -0.069000244140625, 0.32110595703125, -0.99639892578125, 0.248046875, -0.590362548828125, 0.096527099609375, -0.0875244140625, 0.4780426025390625, 1.166229248046875, 0.45330810546875, 1.2551422119140625, -0.0504150390625, -0.534332275390625, 0.653472900390625, 0.18472671508789062, -0.0932769775390625, -0.384307861328125, 1.77362060546875, 0.23668289184570312, 0.3429603576660156, 0.2623329162597656, 0.517547607421875, -0.1230621337890625, 0.15877532958984375, -1.3914794921875, -0.39483642578125, -0.16667938232421875, 0.23683929443359375, -0.118682861328125, -0.628814697265625, -1.7550048828125, -0.125518798828125, -0.17578125, -0.71697998046875, -1.6832122802734375, -1.630157470703125, -0.11407470703125, -0.37152099609375, 1.19476318359375, -1.122314453125, -0.30682373046875, 0.31984710693359375, -0.25284576416015625, -0.33856201171875, -0.6305084228515625, 0.6493988037109375, -0.19178009033203125, 0.0, -0.376617431640625, 1.569091796875, 0.41570281982421875, -0.3368797302246094, 2.65484619140625, 0.1663360595703125, -2.0711517333984375, 0.032135009765625, -0.360107421875, -0.410400390625, 2.143402099609375, 1.007476806640625, -0.8743896484375, -0.1055450439453125, 0.15308380126953125, -0.6991119384765625, -0.22552490234375, 0.21990966796875, 0.2722930908203125, -0.103515625, 0.0284576416015625, 0.007049560546875, -0.7802734375, -0.12945175170898438, -0.502227783203125, 0.8646392822265625, 0.9844512939453125, -0.089019775390625, -0.9546661376953125, 0.14766693115234375, 0.058216094970703125, -0.07940673828125, -0.3558349609375, -2.2662353515625, -0.177734375, -0.15189361572265625, -0.26654052734375, -0.27858734130859375, 1.42156982421875, 0.80810546875, -1.960662841796875, -0.2427215576171875, -0.04791259765625, 1.03948974609375, -0.7073326110839844, -0.20781707763671875, -0.06981468200683594, 1.71734619140625, 0.9046630859375, -0.005340576171875, -0.5450363159179688, 0.083251953125, -0.1863555908203125, -0.83984375, -0.83746337890625, -0.20830535888671875, -0.09417724609375, 0.2008056640625, -0.327850341796875, 0.49045562744140625, 0.650360107421875, -0.68023681640625, 0.21077728271484375, 0.4019775390625, -0.2552947998046875, -0.80780029296875, 0.26251792907714844, -0.617828369140625, 1.0009307861328125, 1.2010040283203125, -0.5390472412109375, 0.0020771026611328125, 0.2213134765625, 0.8140716552734375, 0.763824462890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000020.npy"}
{"epoch": 0.041884816753926704, "step": 21, "batch_size": 128, "mean": 0.02955658733844757, "std": 0.925585925579071, "min": -2.779510498046875, "p10": -1.1021636962890624, "median": 0.009002685546875, "p90": 1.17227783203125, "max": 1.98443603515625, "pos_frac": 0.5078125, "sample": [0.507904052734375, -0.332244873046875, -1.05218505859375, -0.2117767333984375, -2.103668212890625, -2.465362548828125, 0.28570556640625, 0.45013427734375, 0.4664764404296875, 0.05223846435546875, 0.4810943603515625, -1.0692138671875, 0.802642822265625, -2.03753662109375, -0.2652587890625, -0.3873748779296875, -0.7171630859375, 1.2660369873046875, 1.359466552734375, 0.06035614013671875, 1.492706298828125, -0.0033206939697265625, -1.797393798828125, 0.56622314453125, -0.3189849853515625, 0.50274658203125, 0.188079833984375, 1.521270751953125, -0.06317901611328125, -0.87890625, -0.06288909912109375, -0.627838134765625, -0.227264404296875, -0.5003662109375, 0.89501953125, 0.03298187255859375, 0.546234130859375, -0.263092041015625, -2.779510498046875, 0.893707275390625, -1.32794189453125, -0.444732666015625, -0.80841064453125, -0.3544578552246094, -0.1436920166015625, 0.72271728515625, -0.3463306427001953, 0.82135009765625, -0.29744720458984375, 0.96173095703125, -1.3170318603515625, 0.5823822021484375, 0.0, -0.37264251708984375, -0.12579345703125, 0.38663482666015625, 0.564208984375, -0.265350341796875, -0.8063182830810547, 0.5591049194335938, -0.34405517578125, 1.848388671875, 0.28646087646484375, 1.98443603515625, -0.0936279296875, 1.2763671875, 0.015625, -0.35791015625, 0.802215576171875, 0.299224853515625, 0.905181884765625, 1.0248870849609375, -0.291656494140625, -1.271331787109375, -0.06960296630859375, -0.9705810546875, -0.64093017578125, 1.749725341796875, -2.071014404296875, 1.31549072265625, -0.28795623779296875, -0.2644386291503906, 0.14599609375, 0.76251220703125, 0.42742919921875, 0.53369140625, 0.5904655456542969, 0.7230758666992188, 1.0651092529296875, -0.3520088195800781, -0.48602294921875, 0.5909347534179688, 1.690673828125, -0.04105949401855469, 0.524627685546875, 0.20006179809570312, -0.0005645751953125, -0.436676025390625, 1.7330474853515625, 0.6015472412109375, -0.85552978515625, -0.374114990234375, 1.44305419921875, 1.1693115234375, -2.51910400390625, -0.75341796875, -0.088714599609375, -1.5395660400390625, -0.3751678466796875, 1.1371917724609375, -0.327117919921875, 0.3849639892578125, 0.695526123046875, 1.17919921875, 0.9919204711914062, -1.179046630859375, 0.02559661865234375, 0.677978515625, -0.461639404296875, -0.57269287109375, -1.48345947265625, -0.443756103515625, 0.3965892791748047, 0.430206298828125, 0.00238037109375, 0.5183868408203125, 0.33000946044921875, 0.062042236328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000021.npy"}
{"epoch": 0.04397905759162304, "step": 22, "batch_size": 128, "mean": 0.1200462281703949, "std": 0.7564123272895813, "min": -1.76544189453125, "p10": -0.8100128173828125, "median": 0.06481170654296875, "p90": 1.184925842285156, "max": 1.959075927734375, "pos_frac": 0.5390625, "sample": [1.959075927734375, 0.4216117858886719, 0.26309967041015625, -0.395263671875, 1.29852294921875, -0.09822845458984375, 0.448944091796875, 0.19245147705078125, -0.468902587890625, 1.312713623046875, 1.04736328125, 0.22176361083984375, 0.03411865234375, -0.18990135192871094, -0.44329833984375, -0.98651123046875, 0.4627685546875, 0.475189208984375, -0.5493316650390625, -0.2656097412109375, 0.60125732421875, 0.8557586669921875, -0.303924560546875, 0.212005615234375, 0.72991943359375, -0.139678955078125, -0.262451171875, -0.645599365234375, 0.8115692138671875, 0.024005889892578125, -0.7833251953125, 0.3648834228515625, 0.0, 0.0794677734375, -1.65447998046875, -1.76544189453125, -0.09014892578125, 0.6083221435546875, 0.5069122314453125, -0.735107421875, 0.525177001953125, 0.6045074462890625, -1.0230255126953125, 1.9302978515625, 0.021026611328125, -0.7173309326171875, -0.839691162109375, -0.80364990234375, -0.53350830078125, -0.150726318359375, 0.13817977905273438, 0.05078125, -1.11395263671875, 1.0781707763671875, -0.2335662841796875, 0.6342239379882812, -0.22602462768554688, -0.888580322265625, 1.160980224609375, 0.99407958984375, -0.92578125, -1.448883056640625, 0.728759765625, -0.824859619140625, -0.01610565185546875, 0.13275909423828125, 1.4261474609375, 0.49981689453125, -0.3538055419921875, -0.7579345703125, -0.08273696899414062, -0.033050537109375, -0.0184326171875, 1.78515625, 0.13818359375, 0.586883544921875, -0.6868896484375, -0.11852264404296875, 1.46722412109375, 1.30865478515625, -0.5771484375, 0.221282958984375, 0.466094970703125, 0.39290618896484375, 0.0788421630859375, 0.19915771484375, 1.5804595947265625, -0.01007080078125, -0.68145751953125, -0.318878173828125, 1.2407989501953125, -0.6579208374023438, -0.5101127624511719, 0.6576461791992188, 0.391571044921875, 0.59368896484375, -0.76983642578125, -0.883697509765625, 1.5367431640625, -0.86920166015625, 0.215911865234375, -0.3030223846435547, 0.5174560546875, 0.983795166015625, -0.7469253540039062, 0.679412841796875, -0.2248992919921875, 1.3449249267578125, -0.1801300048828125, -0.27166748046875, 0.3686714172363281, -0.65606689453125, -0.13102340698242188, 0.0223846435546875, -0.3402862548828125, 0.82037353515625, 0.201385498046875, 1.014312744140625, 0.492218017578125, 1.249298095703125, -0.013885498046875, 1.038665771484375, -0.5887451171875, -0.9776153564453125, 0.09950637817382812, 0.965789794921875, 0.20662307739257812, 0.93011474609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000022.npy"}
{"epoch": 0.04607329842931937, "step": 23, "batch_size": 128, "mean": -0.019068002700805664, "std": 0.84830641746521, "min": -2.42779541015625, "p10": -1.1050704956054687, "median": 0.03585052490234375, "p90": 0.8102172851562498, "max": 2.96728515625, "pos_frac": 0.5234375, "sample": [-0.15013694763183594, -0.1764373779296875, -0.6282501220703125, 0.4617919921875, 0.311553955078125, 2.3349609375, -0.3175048828125, 0.4139251708984375, -0.8448333740234375, 0.28302001953125, 0.897857666015625, -0.565673828125, 0.12177276611328125, -1.556427001953125, -1.0233154296875, 0.13629913330078125, 2.17254638671875, 0.2800445556640625, -0.25968170166015625, -1.1474609375, 0.4042510986328125, -0.8155517578125, 0.433929443359375, -0.650787353515625, -0.6190643310546875, 0.0228729248046875, 0.5007781982421875, 2.04180908203125, 0.052978515625, 0.11107635498046875, 0.0, 0.45847320556640625, -1.3564376831054688, 1.0900955200195312, -1.13153076171875, 0.320648193359375, 0.048828125, -0.774139404296875, 0.7149658203125, -0.92596435546875, -1.121337890625, -2.42779541015625, 0.1651153564453125, -0.5213775634765625, 0.10101318359375, -0.4406585693359375, 0.0547637939453125, -0.49053955078125, 2.96728515625, -0.43138885498046875, -0.29163360595703125, -0.0003662109375, -0.9520416259765625, 0.64013671875, -0.29315185546875, 2.1724853515625, -0.6036148071289062, -0.34905242919921875, 0.0517120361328125, -0.4990997314453125, -1.170196533203125, 0.1658172607421875, 0.77203369140625, -0.26970672607421875, 0.1103515625, -0.40595245361328125, 0.16143798828125, 1.79901123046875, 0.6531829833984375, -1.089202880859375, 0.5665435791015625, -0.19048690795898438, 0.2742919921875, 0.0001220703125, 0.78070068359375, -0.1170654296875, 0.853240966796875, 0.2840862274169922, 0.697540283203125, 0.073272705078125, -0.3825531005859375, -0.19554710388183594, 0.9966049194335938, 0.483367919921875, -1.185943603515625, 0.613372802734375, -0.4901123046875, 0.6096343994140625, 0.60205078125, -0.71514892578125, 0.24445343017578125, -0.5099296569824219, -1.69476318359375, 0.11525344848632812, 0.43247222900390625, 0.553253173828125, 0.791778564453125, 0.881591796875, 0.11168670654296875, -0.1649169921875, 0.013702392578125, 0.714630126953125, 1.2378997802734375, 0.10356903076171875, -0.1455230712890625, -1.17132568359375, -1.09405517578125, -0.20587158203125, 0.3946533203125, 1.41461181640625, 0.16341400146484375, -0.5732421875, 0.4715156555175781, -0.6387176513671875, -0.37183570861816406, -0.34979248046875, 0.594818115234375, -0.0935516357421875, -0.220123291015625, -1.6258544921875, 0.4288330078125, -1.4240493774414062, -0.61181640625, 0.080047607421875, 0.4194793701171875, -1.5516815185546875, -1.0980987548828125, -0.7497024536132812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000023.npy"}
{"epoch": 0.048167539267015703, "step": 24, "batch_size": 128, "mean": 0.04736223816871643, "std": 0.988019585609436, "min": -1.947265625, "p10": -1.0855377197265625, "median": -0.0054779052734375, "p90": 1.2404922485351562, "max": 3.74755859375, "pos_frac": 0.484375, "sample": [0.6371002197265625, -0.779205322265625, -0.326934814453125, 1.2267913818359375, -1.0750732421875, 2.16693115234375, 0.317962646484375, 2.672943115234375, 0.4467124938964844, -0.883392333984375, 0.2946014404296875, 0.236297607421875, -0.44970703125, -0.5054893493652344, -0.09448623657226562, -0.53192138671875, -0.016387939453125, 0.44000244140625, 1.190765380859375, 0.30888938903808594, 2.7396240234375, 0.3555908203125, -0.765350341796875, 0.0, 0.24197006225585938, 0.034397125244140625, -1.947265625, -1.2814178466796875, 0.5209884643554688, 0.34185791015625, -1.18255615234375, 0.45075225830078125, -0.010955810546875, 2.4835205078125, -0.0758056640625, -0.3066864013671875, -0.50677490234375, -0.559173583984375, -0.698516845703125, -1.005035400390625, 0.1898651123046875, 0.5623397827148438, -0.13336563110351562, -0.53814697265625, 0.338897705078125, -0.038604736328125, 0.0, 0.838348388671875, -0.401397705078125, 0.04694366455078125, -0.2158203125, 1.054443359375, 0.108001708984375, 0.8799667358398438, -0.102294921875, 0.01192474365234375, -0.637847900390625, -0.18459320068359375, 1.313934326171875, 1.954345703125, -1.717681884765625, -0.24030303955078125, 1.2724609375, 0.28157806396484375, 0.42572021484375, 0.6947402954101562, 0.37371826171875, -0.6783294677734375, -1.109954833984375, -0.250213623046875, -1.070465087890625, -0.14044189453125, -1.073455810546875, -1.435028076171875, -1.7795562744140625, 0.2614288330078125, 0.280120849609375, -1.1772308349609375, 2.151702880859375, 0.07355499267578125, -0.7631072998046875, -0.722808837890625, -0.142578125, -1.45440673828125, -0.344696044921875, -1.482940673828125, -0.036376953125, -1.138397216796875, -0.593841552734375, 0.2290191650390625, 0.128631591796875, 0.730224609375, 0.67108154296875, -1.0355987548828125, 0.297119140625, -0.30098915100097656, -0.69677734375, 1.32684326171875, 0.7814178466796875, 1.51446533203125, 3.74755859375, 0.84881591796875, 0.601776123046875, 0.863677978515625, -1.5753021240234375, 1.61151123046875, -0.312713623046875, -0.671844482421875, 0.76702880859375, -0.64697265625, 0.697265625, -0.1167144775390625, 0.4141387939453125, 1.1505126953125, 0.1218719482421875, 0.201751708984375, -1.05120849609375, -0.401458740234375, -0.7783203125, -0.60308837890625, 2.37396240234375, 0.201446533203125, 0.13249969482421875, -0.02789306640625, -1.61566162109375, 0.4632415771484375, -1.01239013671875, -0.582275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000024.npy"}
{"epoch": 0.050261780104712044, "step": 25, "batch_size": 128, "mean": 0.06370164453983307, "std": 0.7688639760017395, "min": -2.289398193359375, "p10": -0.8320220947265625, "median": 0.049957275390625, "p90": 1.0004821777343749, "max": 2.2623443603515625, "pos_frac": 0.53125, "sample": [-0.3733062744140625, -0.80548095703125, 0.51275634765625, 1.3791656494140625, -0.209991455078125, 0.9976806640625, -0.674591064453125, 2.1280059814453125, 0.20550537109375, -0.98040771484375, 0.8116607666015625, -0.3393707275390625, -0.102142333984375, 0.590545654296875, 0.38329315185546875, 0.2728271484375, -0.100494384765625, 0.20407867431640625, 0.56781005859375, 0.0707855224609375, 0.0211334228515625, -0.03162193298339844, 0.239501953125, 0.1548004150390625, -0.80584716796875, 0.20672607421875, -0.1700439453125, -0.04358673095703125, 0.15901947021484375, -1.240142822265625, 0.07367515563964844, 0.2400665283203125, 0.0, -0.3423309326171875, 1.291961669921875, 0.09565925598144531, 0.4782257080078125, -0.2955322265625, -0.361846923828125, -0.97637939453125, 2.2623443603515625, 0.0, 0.62841796875, -0.1444091796875, -0.7394256591796875, 0.1425628662109375, 0.29479217529296875, 0.2503662109375, 0.917236328125, -0.835418701171875, -0.45453643798828125, 0.078857421875, 1.009552001953125, -0.4042015075683594, -0.32757568359375, -0.15625, 0.5680160522460938, -0.308990478515625, -0.6616363525390625, -0.6648406982421875, 0.8618240356445312, -0.703826904296875, 0.576812744140625, -0.248046875, -0.99468994140625, 0.31841278076171875, 1.709808349609375, -1.2120208740234375, -1.82232666015625, 0.3636322021484375, 1.112060546875, 0.395843505859375, 0.29439544677734375, 1.77532958984375, -1.3985595703125, -0.2659912109375, -0.24651336669921875, 0.4957847595214844, 1.6221466064453125, -0.553253173828125, -0.2664794921875, -0.0933990478515625, 0.7802581787109375, -0.77850341796875, 0.0520477294921875, -1.109649658203125, -0.4364299774169922, 0.08051490783691406, 0.6278228759765625, 0.2293548583984375, 1.00701904296875, -0.31500244140625, 0.5126495361328125, 1.78155517578125, 0.864837646484375, -0.908935546875, 0.0478668212890625, 0.4145050048828125, 0.72760009765625, 1.254150390625, -0.28701019287109375, 0.5400619506835938, -0.4473114013671875, -0.15753936767578125, -0.697021484375, 1.671142578125, -0.0482177734375, -0.6100616455078125, -1.21234130859375, -0.4794921875, -1.1067352294921875, 0.3919830322265625, 0.1478271484375, 0.865509033203125, -0.353240966796875, 0.6260833740234375, 0.632598876953125, -0.11631011962890625, 0.188812255859375, 0.03487396240234375, 0.3717536926269531, 0.1665630340576172, 0.045501708984375, -0.83056640625, -2.289398193359375, -0.3612518310546875, 0.7708740234375, -0.5425090789794922], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000025.npy"}
{"epoch": 0.05235602094240838, "step": 26, "batch_size": 128, "mean": -0.005920037627220154, "std": 0.7910383343696594, "min": -1.977325439453125, "p10": -0.9921295166015623, "median": -0.010845184326171875, "p90": 0.78472900390625, "max": 3.05535888671875, "pos_frac": 0.4765625, "sample": [-1.1827392578125, 0.040042877197265625, -0.005615234375, -0.536224365234375, 0.761871337890625, 0.42816925048828125, 0.2863311767578125, 0.31610107421875, 0.667388916015625, 0.37613487243652344, -0.20159530639648438, -0.628997802734375, 0.0, 0.5489959716796875, -0.619140625, -0.277496337890625, 2.0946044921875, -0.10601043701171875, 3.05535888671875, 0.509124755859375, -0.374786376953125, -0.152740478515625, -0.3082542419433594, 1.40692138671875, 0.16314697265625, 0.19132614135742188, 0.225067138671875, -0.553741455078125, 0.3781280517578125, 0.05780029296875, -0.7262802124023438, -1.49072265625, -0.34686279296875, -0.3587646484375, -1.802734375, 0.2520751953125, 1.2982025146484375, 0.21844482421875, -0.8461456298828125, 0.1206512451171875, 0.63671875, -0.2690601348876953, -0.04433441162109375, 0.78900146484375, 0.289947509765625, -0.41705322265625, -0.11212158203125, -1.160858154296875, -0.2611236572265625, -0.6207046508789062, -0.946929931640625, -1.30230712890625, -0.4108123779296875, 1.56072998046875, 0.801727294921875, -0.24725723266601562, -0.9443511962890625, -1.09759521484375, 0.14263916015625, 0.754150390625, 0.6809215545654297, -0.756988525390625, 0.5220947265625, 0.7819595336914062, -1.977325439453125, -0.48149871826171875, 0.714447021484375, -0.1422271728515625, 1.03741455078125, 0.1413421630859375, -1.2161483764648438, -0.30673789978027344, -0.3782196044921875, -0.355010986328125, 0.22086334228515625, -0.3175544738769531, 0.4902801513671875, 0.3512725830078125, -0.01763916015625, 0.3463897705078125, 0.0902099609375, -0.18829345703125, 0.06784820556640625, 2.525390625, 1.62286376953125, 0.42872047424316406, 0.34051513671875, -0.0938720703125, -0.9010810852050781, -0.3503570556640625, -0.6547698974609375, 0.6198043823242188, 0.16481781005859375, -1.936004638671875, 0.613800048828125, -0.1278228759765625, 0.7301177978515625, 0.13985252380371094, -0.13703346252441406, 0.062469482421875, -0.1351470947265625, -0.033443450927734375, 1.03875732421875, -0.898406982421875, 1.0144424438476562, -1.195709228515625, 0.1656036376953125, 0.0, -1.50054931640625, -0.01607513427734375, 0.432281494140625, 0.13335418701171875, 0.45880126953125, -1.10479736328125, -0.0648345947265625, -0.731536865234375, 0.27471923828125, 0.00640869140625, -0.653167724609375, 0.803741455078125, -0.197174072265625, -0.01727294921875, -0.15331268310546875, 0.78289794921875, -1.1099395751953125, 0.341949462890625, -0.513641357421875, -0.257965087890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000026.npy"}
{"epoch": 0.05445026178010471, "step": 27, "batch_size": 128, "mean": -0.03933677077293396, "std": 0.9862356185913086, "min": -3.346405029296875, "p10": -0.9931701660156249, "median": 0.025768280029296875, "p90": 0.8796356201171871, "max": 3.01385498046875, "pos_frac": 0.5078125, "sample": [1.157806396484375, -0.78082275390625, -0.3456840515136719, 0.08314132690429688, -0.3579864501953125, -1.3868408203125, -0.485687255859375, -0.5505828857421875, 1.109344482421875, 0.07186126708984375, 0.3101806640625, 0.684234619140625, 2.069732666015625, 0.02637481689453125, -0.19000244140625, 2.1380615234375, 0.7715225219726562, -0.48535919189453125, 0.0620269775390625, 0.72991943359375, 0.6205215454101562, 0.2883644104003906, 0.516998291015625, -0.152252197265625, 2.3358154296875, 2.89227294921875, 0.3663749694824219, 1.89605712890625, -0.015232086181640625, 0.38365936279296875, 0.3133544921875, 0.08087921142578125, 0.335174560546875, -0.198974609375, 0.3862762451171875, -1.01824951171875, -0.00189971923828125, 0.34393310546875, -0.6145782470703125, -0.1516265869140625, 3.01385498046875, 0.64801025390625, -2.374755859375, 0.04998016357421875, 0.7620849609375, 0.364349365234375, -0.4514732360839844, -0.982421875, 0.389739990234375, 0.09737396240234375, -2.393707275390625, 0.317901611328125, 0.09336090087890625, -0.31951904296875, 0.3414306640625, 0.780914306640625, -0.4127044677734375, -0.3611907958984375, 0.14196014404296875, 0.1568603515625, -0.75933837890625, -0.859466552734375, -0.4549407958984375, -0.143096923828125, 0.3309478759765625, -0.1539459228515625, -0.81219482421875, -0.5449104309082031, 2.17645263671875, -1.39605712890625, -1.475341796875, -3.346405029296875, -0.84429931640625, 0.22406005859375, 0.6136093139648438, 0.9914932250976562, -1.4608154296875, 0.760101318359375, -0.19036865234375, -0.01300048828125, -0.1101531982421875, 0.39593505859375, 0.390625, -0.54095458984375, -0.314361572265625, 0.035430908203125, 0.838134765625, -0.31542015075683594, -0.2910804748535156, 0.976470947265625, 0.244873046875, -0.5035324096679688, -3.3294677734375, 0.0286712646484375, -0.8415374755859375, -0.3548583984375, -1.9368896484375, -0.5154666900634766, 0.175018310546875, 1.588775634765625, 0.5826416015625, -0.03887176513671875, -0.201446533203125, -1.71405029296875, 0.0251617431640625, 0.824951171875, -1.554229736328125, -0.5882987976074219, -0.34558868408203125, -0.10133552551269531, -0.85638427734375, 0.09352874755859375, 0.046539306640625, 0.45245361328125, -0.47540283203125, 0.32928466796875, 0.38555145263671875, -0.5560150146484375, -0.2454833984375, -0.327850341796875, 0.097503662109375, 1.0249176025390625, -0.55364990234375, -1.798583984375, -0.505401611328125, 0.2906646728515625, -0.81695556640625, 0.1283893585205078], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000027.npy"}
{"epoch": 0.05654450261780105, "step": 28, "batch_size": 128, "mean": 0.1874668449163437, "std": 0.8731688261032104, "min": -3.4122314453125, "p10": -0.737689208984375, "median": 0.21732330322265625, "p90": 1.0083465576171875, "max": 3.15777587890625, "pos_frac": 0.625, "sample": [0.468902587890625, -0.54278564453125, 0.632110595703125, 0.848358154296875, 0.01480865478515625, -0.11825180053710938, 0.21929931640625, 0.5328254699707031, -0.03490447998046875, 0.8607177734375, -0.05430030822753906, -1.21319580078125, 0.752960205078125, 0.8024826049804688, 0.803497314453125, -0.642608642578125, -0.6100616455078125, -0.60955810546875, 2.1924896240234375, 0.79327392578125, 0.2153472900390625, -0.03759574890136719, 0.08907318115234375, -0.258697509765625, -0.0867462158203125, 0.100616455078125, -0.2958984375, -0.08330154418945312, 0.306793212890625, -2.6688079833984375, 0.999237060546875, 0.4262237548828125, -0.834320068359375, 1.156036376953125, 0.3181037902832031, -1.326904296875, 0.4328765869140625, 0.4306488037109375, 0.904052734375, 1.270111083984375, -0.6717147827148438, -1.57110595703125, 0.18255615234375, -0.0650634765625, 0.09649658203125, 0.94482421875, -0.010894775390625, 0.892608642578125, 1.64739990234375, -0.21739959716796875, -0.02496337890625, 0.27735137939453125, 0.125030517578125, -2.09088134765625, 0.100341796875, 0.5621795654296875, 0.732940673828125, 2.53369140625, 0.926513671875, 0.6102142333984375, -0.796905517578125, 0.3656463623046875, 0.6488494873046875, -0.4622993469238281, 3.15777587890625, -0.0406036376953125, -0.48798370361328125, 0.9399566650390625, 0.735107421875, 1.02960205078125, 0.150360107421875, 0.553955078125, 0.4353485107421875, -0.191619873046875, -0.71368408203125, 1.456268310546875, -3.4122314453125, -0.437774658203125, 0.249267578125, -0.56414794921875, 0.5160598754882812, -0.5217132568359375, 0.68560791015625, 0.97222900390625, 0.4423980712890625, 0.5555877685546875, 0.7030029296875, -0.254119873046875, 0.0035552978515625, 0.402740478515625, -0.23907470703125, 0.2492523193359375, 0.2864532470703125, 0.22222900390625, -0.1656513214111328, -0.21337890625, 0.637115478515625, 0.4736785888671875, 1.1124267578125, 0.950897216796875, -1.27459716796875, 0.3271942138671875, -0.6262054443359375, 0.8699951171875, -0.9672393798828125, -1.1094970703125, 0.6569061279296875, 0.20654296875, 1.446014404296875, 0.02496337890625, 0.051422119140625, 0.0896759033203125, 1.39947509765625, -0.2138843536376953, -0.228118896484375, 0.128021240234375, 0.3165283203125, 0.5867385864257812, 1.0601806640625, 0.9412841796875, -0.3897705078125, -0.793701171875, 0.802490234375, 0.188690185546875, 1.630859375, -0.5062961578369141, 0.0, -1.1891326904296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000028.npy"}
{"epoch": 0.05863874345549738, "step": 29, "batch_size": 128, "mean": 0.10563047230243683, "std": 0.9823217391967773, "min": -3.08062744140625, "p10": -1.059088134765625, "median": 0.1464080810546875, "p90": 1.155645751953125, "max": 4.9940185546875, "pos_frac": 0.546875, "sample": [-0.10164833068847656, -1.018402099609375, 1.517608642578125, 0.60015869140625, 0.418975830078125, -3.08062744140625, 0.644287109375, 4.9940185546875, -1.6192626953125, 1.41070556640625, -0.522003173828125, 1.558685302734375, 0.29034423828125, 0.244659423828125, 0.26674652099609375, 0.0, -0.70111083984375, 0.5235443115234375, 1.79833984375, 0.0197906494140625, -0.266387939453125, 0.18585205078125, -0.072601318359375, -2.23712158203125, -1.2296905517578125, 1.193115234375, 0.761016845703125, 0.1833343505859375, -0.404754638671875, 1.0538330078125, 0.32769775390625, -1.8714752197265625, 0.321746826171875, 0.902008056640625, -0.9121017456054688, 1.9210205078125, 0.72113037109375, -1.69500732421875, -0.63275146484375, -0.05693817138671875, 0.43353271484375, -0.09392166137695312, 0.290252685546875, 0.2114410400390625, -0.522247314453125, 0.145111083984375, 0.1983489990234375, -0.08697509765625, 0.18328857421875, 0.2612457275390625, 0.19976806640625, -0.3267822265625, -1.290130615234375, -0.2391204833984375, 0.793609619140625, 0.92657470703125, 0.99365234375, 0.5603179931640625, 1.286376953125, 0.31452178955078125, 0.67401123046875, -0.784332275390625, 2.71612548828125, 1.541046142578125, 0.43442535400390625, 0.6377639770507812, -1.4129638671875, 0.043731689453125, 0.0, -0.24127197265625, 0.226837158203125, -0.276824951171875, -0.561431884765625, -1.23590087890625, -0.35662841796875, -0.1016998291015625, -1.30340576171875, -0.29706382751464844, 0.147705078125, 1.132568359375, -0.479461669921875, -0.1683502197265625, -2.0648193359375, 0.01314544677734375, -0.0208282470703125, 0.0302886962890625, 0.7364349365234375, 1.691314697265625, -0.516876220703125, 1.13958740234375, -0.35088348388671875, 0.4140625, -0.02349853515625, -0.44512939453125, -0.164794921875, 0.2670745849609375, 0.0302734375, -0.0407257080078125, -0.20439910888671875, 0.5586395263671875, -1.258270263671875, -0.38653564453125, 0.564910888671875, -0.43221282958984375, 0.4951019287109375, 0.893768310546875, -0.4035186767578125, 0.36029052734375, -0.539306640625, -1.154022216796875, 0.862060546875, 0.36895751953125, -0.32281494140625, 0.55352783203125, 1.34930419921875, -0.90948486328125, 0.366241455078125, -0.7344818115234375, 0.9161376953125, -0.84149169921875, 0.2331981658935547, 0.311492919921875, -0.392303466796875, -0.5450897216796875, -0.05694580078125, 0.888031005859375, 1.501556396484375, 0.77325439453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000029.npy"}
{"epoch": 0.060732984293193716, "step": 30, "batch_size": 128, "mean": 0.119469553232193, "std": 0.9331782460212708, "min": -2.538116455078125, "p10": -1.0130767822265625, "median": 0.2371063232421875, "p90": 1.297637939453125, "max": 2.453704833984375, "pos_frac": 0.5859375, "sample": [1.294342041015625, 1.6430511474609375, -1.349700927734375, -1.35357666015625, -0.2820014953613281, 0.457611083984375, 0.8135528564453125, -0.4944000244140625, 0.21905517578125, 0.2552337646484375, 0.04399871826171875, -0.08990478515625, 1.5408935546875, 0.38507080078125, 0.554046630859375, -1.985198974609375, -2.538116455078125, 0.48602294921875, -0.307586669921875, -0.528594970703125, -2.149749755859375, -0.17950439453125, -1.03668212890625, 0.4924354553222656, -0.08538818359375, -1.466522216796875, 1.7449951171875, 0.3562641143798828, 0.31810760498046875, 0.9530029296875, 0.311767578125, -0.05200958251953125, -0.02215576171875, 1.8258056640625, -0.6686592102050781, 0.431640625, 0.519195556640625, 0.3133392333984375, 0.3317718505859375, -0.7469482421875, 0.20672607421875, 0.2494659423828125, 0.59661865234375, 1.63775634765625, 0.47210693359375, 0.342376708984375, -0.47974395751953125, -2.1892852783203125, 0.87261962890625, -1.8605575561523438, 0.39228057861328125, 1.1960906982421875, -0.37603759765625, 0.7260284423828125, 1.98345947265625, 0.17377471923828125, -1.002960205078125, 0.4237518310546875, -0.2503662109375, -0.7287139892578125, 1.69677734375, 0.98223876953125, 0.5242462158203125, -0.878448486328125, 0.04213714599609375, -0.6495285034179688, 0.292755126953125, -0.088623046875, 0.12081527709960938, 0.872314453125, 0.1406097412109375, -0.301727294921875, -0.781036376953125, 0.4492950439453125, 1.46783447265625, 1.58465576171875, -0.7476577758789062, -0.9497833251953125, 0.3263092041015625, -0.9194488525390625, 0.28995513916015625, 2.453704833984375, 1.305328369140625, 0.7600250244140625, 0.5137405395507812, -1.723480224609375, 1.58612060546875, -0.1157684326171875, 0.6210174560546875, -0.921295166015625, 0.0946502685546875, -0.1136016845703125, 0.8143310546875, -2.1854248046875, 0.848846435546875, -1.104736328125, 1.12957763671875, -0.09665298461914062, 0.33734130859375, 0.57598876953125, -0.59942626953125, 0.2247467041015625, 0.7647171020507812, 0.182861328125, 0.593414306640625, 0.49372100830078125, -0.7362060546875, -0.0516357421875, 0.444854736328125, 1.01531982421875, 1.95941162109375, -0.0226593017578125, 0.540740966796875, -0.55145263671875, -0.06112480163574219, -0.26416015625, 0.7437629699707031, 0.8017425537109375, -0.5343704223632812, -1.255279541015625, -0.354888916015625, 0.6126861572265625, 0.999237060546875, -0.418212890625, 0.03023529052734375, -0.17635345458984375, 0.6875457763671875, -0.37042236328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000030.npy"}
{"epoch": 0.06282722513089005, "step": 31, "batch_size": 128, "mean": 0.060995444655418396, "std": 0.8385193943977356, "min": -2.39691162109375, "p10": -0.9653533935546875, "median": 0.01201629638671875, "p90": 0.9943222045898437, "max": 2.505126953125, "pos_frac": 0.5078125, "sample": [-0.0666656494140625, 1.09771728515625, -0.5616455078125, -0.21044921875, -0.880096435546875, 0.8789138793945312, 0.2989654541015625, 0.5469970703125, 0.518646240234375, 0.671234130859375, 2.0283203125, -1.37762451171875, -2.39691162109375, -0.5584716796875, 0.192169189453125, -0.098724365234375, 0.7909183502197266, 0.3687744140625, 0.715118408203125, -0.2062225341796875, 0.24853515625, -0.70074462890625, 0.1663055419921875, 0.9885406494140625, 0.30530548095703125, 2.35797119140625, -1.3972015380859375, 1.44744873046875, -1.1240692138671875, 0.26378631591796875, 0.5445938110351562, 0.7661590576171875, -0.009845733642578125, -0.2664604187011719, 0.23260498046875, -0.847412109375, -0.9014739990234375, 0.0264129638671875, -0.2938232421875, 0.0, 0.117584228515625, -0.39467620849609375, 0.0, 0.37725830078125, 0.7384262084960938, -1.2470855712890625, -0.1202392578125, 0.5750885009765625, -0.44781494140625, -0.3027496337890625, 0.65838623046875, -1.364990234375, -0.064117431640625, 0.6879730224609375, -0.11791229248046875, -0.64093017578125, -1.872833251953125, -0.064422607421875, -0.963714599609375, -0.45511627197265625, 0.0, -0.20084381103515625, -1.7713623046875, -1.14398193359375, -0.395477294921875, -0.895111083984375, -1.1276702880859375, 0.750823974609375, -0.51629638671875, -0.24487686157226562, -0.3126373291015625, -0.96917724609375, 0.13257980346679688, 0.5983047485351562, -0.0810546875, 0.6161422729492188, -0.293121337890625, -0.034572601318359375, -0.136199951171875, 0.7522735595703125, 0.4562873840332031, -0.5359039306640625, 0.326751708984375, 0.2809429168701172, 1.2453384399414062, 0.0599517822265625, 2.505126953125, -2.07220458984375, -0.5066299438476562, 2.06304931640625, -0.520538330078125, 0.20979881286621094, 1.0078125, 0.079559326171875, 0.6487274169921875, -0.191009521484375, 0.854644775390625, -0.5945892333984375, -0.0489349365234375, 0.0, -0.764862060546875, -0.4058837890625, 0.697906494140625, 1.1492881774902344, -0.0246124267578125, 1.111114501953125, -1.234619140625, -0.19324111938476562, 0.756317138671875, 0.7437744140625, 1.198486328125, -0.06642913818359375, 0.5543365478515625, 0.817138671875, 0.5757293701171875, 0.01897430419921875, 0.17725753784179688, 1.573699951171875, -0.86590576171875, 0.52081298828125, 0.12371826171875, 1.8333053588867188, 0.3394737243652344, 0.313568115234375, 0.00505828857421875, 0.252349853515625, 0.0713958740234375, -0.12236785888671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000031.npy"}
{"epoch": 0.06492146596858639, "step": 32, "batch_size": 128, "mean": 0.09094801545143127, "std": 0.9035819172859192, "min": -2.63177490234375, "p10": -1.0369651794433594, "median": 0.058086395263671875, "p90": 1.2040252685546873, "max": 2.6119384765625, "pos_frac": 0.5390625, "sample": [-0.3760986328125, -0.4391498565673828, -1.749786376953125, -1.0224609375, -0.0791168212890625, -0.7714157104492188, -0.024944305419921875, 1.124237060546875, 0.5332794189453125, 0.777008056640625, -0.8927001953125, -0.5265655517578125, 1.621826171875, 0.0359039306640625, -0.229736328125, -0.6190757751464844, 0.09796142578125, -1.285919189453125, 0.8770751953125, 1.0645751953125, 0.3228302001953125, -1.45703125, 0.2410430908203125, 0.27239990234375, 0.49713134765625, -1.127197265625, 0.36754608154296875, -0.37681007385253906, -0.3485260009765625, 0.4600372314453125, 0.14068603515625, 0.22321319580078125, -1.0353012084960938, 0.91461181640625, 0.5640869140625, 0.789215087890625, 0.1107025146484375, -1.325225830078125, 0.22381591796875, -1.1483917236328125, 0.1520233154296875, -0.48162841796875, -2.63177490234375, 0.6251068115234375, 0.5628662109375, -0.05792236328125, 0.015960693359375, 1.2725143432617188, -0.0841064453125, 1.447723388671875, 1.12109375, -0.96435546875, -1.371978759765625, -0.22705078125, 1.357666015625, -0.060455322265625, 2.576568603515625, 0.6822509765625, 0.914093017578125, -1.32513427734375, -0.3268890380859375, -0.088043212890625, -0.3121185302734375, -0.9076995849609375, 0.947509765625, -0.7632904052734375, -0.9506072998046875, 0.207244873046875, 1.06396484375, 1.057159423828125, 2.39794921875, 0.35692596435546875, 0.27916717529296875, 0.018341064453125, -0.8699951171875, 0.93243408203125, 0.2918548583984375, 0.40667724609375, -0.38282012939453125, 1.231903076171875, 1.489715576171875, -0.83673095703125, -0.046886444091796875, -0.10920524597167969, 0.909027099609375, 0.5094757080078125, 1.19207763671875, 0.3866729736328125, -0.36578369140625, -0.71148681640625, -0.5624542236328125, -0.24929046630859375, -1.119232177734375, 0.10546875, -1.341339111328125, 0.078094482421875, -1.0408477783203125, 2.017791748046875, 0.20725250244140625, 0.9006500244140625, -0.069732666015625, -0.39727783203125, -0.44622802734375, -0.24201583862304688, -0.39105224609375, -0.2559967041015625, -0.37807273864746094, 0.17064666748046875, 0.3775482177734375, 0.07387542724609375, 1.793853759765625, -0.9675750732421875, 0.1295928955078125, 0.5383148193359375, 0.245697021484375, 0.5694732666015625, -0.481201171875, 1.147979736328125, 0.04229736328125, -1.5016326904296875, 0.1669921875, -0.017833709716796875, 1.538818359375, 1.74993896484375, -0.14508056640625, 0.008880615234375, -0.180633544921875, 2.6119384765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000032.npy"}
{"epoch": 0.06701570680628273, "step": 33, "batch_size": 128, "mean": 0.070505291223526, "std": 0.9937970042228699, "min": -3.083587646484375, "p10": -1.0046554565429686, "median": 0.08645439147949219, "p90": 1.2531372070312499, "max": 2.820220947265625, "pos_frac": 0.5625, "sample": [0.6953010559082031, 2.46002197265625, 1.1309967041015625, 1.3032379150390625, -0.36341094970703125, -0.3260955810546875, 0.6819381713867188, 0.520355224609375, 0.900665283203125, -0.519012451171875, 0.845458984375, -0.5562973022460938, 1.54998779296875, -0.272705078125, 0.0, -0.1781158447265625, 1.106292724609375, -0.347564697265625, -0.80621337890625, -1.30181884765625, 0.321014404296875, 0.5360031127929688, 0.14019393920898438, -1.062652587890625, 1.5755615234375, -0.0472869873046875, -0.05387115478515625, -0.0371856689453125, 2.1024169921875, 1.24896240234375, 0.58502197265625, -2.29541015625, 0.248992919921875, -3.083587646484375, 0.13686370849609375, 2.30859375, 0.1619873046875, 0.2571868896484375, -0.2375030517578125, -0.362701416015625, 0.26900482177734375, -1.3397293090820312, -0.5165176391601562, 1.26287841796875, -0.25885009765625, 0.7625885009765625, 2.35577392578125, -0.07027435302734375, 0.65008544921875, -0.9969329833984375, 0.1405181884765625, 0.039306640625, -1.162353515625, 0.0, -0.28076171875, 0.0531158447265625, 0.1356945037841797, 0.7542724609375, 0.7506103515625, 0.8353271484375, 0.9090576171875, -0.12335205078125, -0.7394943237304688, 0.0900115966796875, 0.31426239013671875, 0.67840576171875, -0.1855030059814453, 0.08289718627929688, -1.67431640625, 1.4246368408203125, -0.402099609375, 0.14525604248046875, -1.022674560546875, 2.820220947265625, -0.143218994140625, -0.458343505859375, 0.259063720703125, 0.019683837890625, -0.2264251708984375, -0.8076171875, -0.690643310546875, -0.2016754150390625, 1.7216796875, 0.2719573974609375, 0.38623046875, -0.885467529296875, -0.24810791015625, -1.864959716796875, 0.7313232421875, 0.5121307373046875, 0.062530517578125, 0.04473114013671875, 0.5332183837890625, -0.06671524047851562, 0.2095489501953125, -2.4615478515625, -0.9541778564453125, -0.552215576171875, -2.781280517578125, -0.0938720703125, 0.44284820556640625, 0.21478271484375, 1.1234588623046875, 1.67791748046875, 0.23955535888671875, -0.333648681640625, -0.7698974609375, 0.3797760009765625, -0.1536712646484375, 0.27484130859375, 0.33514404296875, 1.4921875, 0.14739990234375, 0.68365478515625, 0.384796142578125, -1.7008056640625, 0.3518218994140625, 0.045925140380859375, -0.173095703125, 0.28594970703125, 0.048187255859375, -0.0159759521484375, -3.03253173828125, 0.12146759033203125, -0.55206298828125, -0.4127235412597656, 0.330169677734375, 0.606689453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000033.npy"}
{"epoch": 0.06910994764397906, "step": 34, "batch_size": 128, "mean": 0.2425571084022522, "std": 1.0204863548278809, "min": -2.799407958984375, "p10": -0.8493148803710937, "median": 0.15316009521484375, "p90": 1.670782470703125, "max": 4.0716552734375, "pos_frac": 0.5625, "sample": [-0.31424713134765625, 0.5910797119140625, 0.78076171875, -0.5458450317382812, -0.0387420654296875, -1.5347747802734375, -0.6631927490234375, -0.20319366455078125, 0.3712921142578125, 4.0716552734375, -0.606475830078125, -0.1961669921875, 2.375030517578125, -0.47412109375, -0.2647247314453125, 2.37091064453125, 1.6702880859375, 0.458892822265625, 0.5585250854492188, 0.37290191650390625, 0.26827239990234375, -1.5155029296875, -0.1773681640625, 1.806640625, -0.55316162109375, 0.58441162109375, 3.2196044921875, 0.3793792724609375, 0.08550357818603516, -0.41998291015625, 1.03857421875, 1.3771820068359375, 1.317352294921875, -0.8666915893554688, 0.60601806640625, -0.8716888427734375, 1.713470458984375, -0.233856201171875, -2.799407958984375, -0.150909423828125, -0.0952911376953125, -0.9041748046875, 1.89202880859375, 2.03509521484375, 0.814544677734375, 0.36003875732421875, -0.5311279296875, 1.78753662109375, -0.08697509765625, -0.360107421875, -0.36397552490234375, -0.4949493408203125, -0.1894664764404297, 1.39410400390625, -2.05352783203125, 0.3672332763671875, -0.1431427001953125, 1.67657470703125, 0.8372802734375, -0.4144287109375, -0.416412353515625, -0.4857063293457031, 1.1372528076171875, 0.1409454345703125, -0.352508544921875, -0.68865966796875, -0.0478973388671875, -0.024785995483398438, 0.6256027221679688, 0.477325439453125, 0.121978759765625, 1.136260986328125, 0.39349365234375, -0.763153076171875, 0.66949462890625, 0.3765716552734375, 0.13461685180664062, -0.3230705261230469, 0.46782684326171875, -0.4897117614746094, 1.67193603515625, 0.785980224609375, 0.20819091796875, -1.3448333740234375, 0.4580841064453125, 0.42822265625, 1.101715087890625, 0.165374755859375, 0.24518585205078125, 0.102783203125, 0.06882476806640625, -1.792449951171875, 1.080322265625, -1.748199462890625, -0.0583038330078125, -0.66815185546875, -0.0490570068359375, -0.5582733154296875, 0.73431396484375, -0.15967178344726562, -0.40740966796875, 0.0804443359375, -0.036712646484375, -1.618438720703125, 1.509429931640625, -0.14177703857421875, -0.574005126953125, 0.1400146484375, -0.8464813232421875, 2.198455810546875, 0.928619384765625, -0.855926513671875, 0.476593017578125, -1.18719482421875, 0.4993896484375, 0.7083816528320312, -0.509613037109375, 0.7969970703125, 0.396331787109375, 1.006378173828125, 0.7775421142578125, 1.472991943359375, 0.49118804931640625, 0.546051025390625, 0.87030029296875, 0.2722930908203125, 1.8763427734375, 0.3007049560546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000034.npy"}
{"epoch": 0.0712041884816754, "step": 35, "batch_size": 128, "mean": 0.14058685302734375, "std": 0.8850899338722229, "min": -3.21527099609375, "p10": -0.9797348022460937, "median": 0.15148162841796875, "p90": 1.2468994140625, "max": 2.631072998046875, "pos_frac": 0.6171875, "sample": [-0.33039093017578125, 0.3543701171875, -0.0665283203125, 0.4617919921875, 0.3983154296875, 2.631072998046875, -0.48944091796875, -0.361328125, 0.860992431640625, -0.9622344970703125, 0.53082275390625, 0.3719024658203125, 1.083465576171875, 0.2893218994140625, 1.3165512084960938, 0.3486900329589844, -3.21527099609375, -0.8701171875, -0.6457366943359375, 0.12732887268066406, 0.63580322265625, 1.549896240234375, -1.7467041015625, 0.2559356689453125, 0.36248779296875, 0.7698211669921875, 1.2470703125, -0.5121917724609375, 0.47454833984375, 1.25152587890625, 0.6247100830078125, -0.4600372314453125, 0.8917236328125, 0.167755126953125, 2.21624755859375, 0.70831298828125, -0.019256591796875, -0.8311920166015625, 0.09297752380371094, 0.0, -0.612274169921875, 0.90045166015625, 0.6909866333007812, 0.99591064453125, 1.1202774047851562, 0.1705322265625, 1.701141357421875, -1.1632537841796875, 0.1352081298828125, -1.6025390625, -1.6139373779296875, -0.5308990478515625, 0.0823974609375, -0.3733673095703125, 0.9417724609375, -0.364776611328125, 0.4849853515625, -0.39886474609375, 0.9636001586914062, 0.07732582092285156, -0.73675537109375, 1.54913330078125, 0.88409423828125, 0.42346954345703125, -1.29559326171875, -0.025970458984375, 0.4850616455078125, 0.3167457580566406, 0.655029296875, 0.12385177612304688, -1.852874755859375, -0.00225830078125, -1.4827117919921875, 0.51641845703125, 0.029834747314453125, 0.654327392578125, -0.5064468383789062, 0.092010498046875, 1.313720703125, 0.4588623046875, 0.6846771240234375, 0.352630615234375, -1.597198486328125, 0.12027740478515625, 0.734832763671875, -0.3302459716796875, -0.1946258544921875, -1.1222152709960938, 0.693359375, 0.4805259704589844, -1.61474609375, 0.12250518798828125, 0.5235214233398438, 0.06864166259765625, 0.08263587951660156, -0.3336181640625, -0.392822265625, 1.685302734375, -0.220306396484375, 1.969635009765625, -0.636749267578125, 1.557098388671875, 0.0, 0.0546417236328125, 0.689300537109375, 0.209991455078125, -0.6246490478515625, 1.246826171875, 0.654541015625, -0.31087493896484375, -0.401214599609375, -1.02056884765625, -0.310638427734375, -0.06512451171875, 0.9565048217773438, -0.40972900390625, 0.11090087890625, 1.39984130859375, 0.0709991455078125, 0.20574188232421875, 1.05596923828125, 0.222991943359375, 0.1967315673828125, -0.05474853515625, 0.36310577392578125, -0.52606201171875, -1.2958526611328125, 0.225738525390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000035.npy"}
{"epoch": 0.07329842931937172, "step": 36, "batch_size": 128, "mean": 0.3770591616630554, "std": 1.0896759033203125, "min": -3.8856201171875, "p10": -0.72705078125, "median": 0.3636131286621094, "p90": 1.7369903564453124, "max": 3.273681640625, "pos_frac": 0.6640625, "sample": [3.273681640625, -0.675567626953125, 0.92340087890625, 0.5446624755859375, 1.195098876953125, 2.123992919921875, 1.3861083984375, -0.404388427734375, 0.165924072265625, -0.6360054016113281, -0.519744873046875, -1.5836944580078125, 0.55780029296875, -0.3570556640625, 1.0989227294921875, 0.6099090576171875, -0.5466690063476562, 1.405059814453125, 0.960601806640625, -0.34986114501953125, -0.03076934814453125, 0.68780517578125, 0.2973823547363281, 0.0, 0.22222137451171875, 0.9664306640625, 1.73223876953125, -2.2774658203125, -0.0176544189453125, -0.713897705078125, 2.40289306640625, 0.829803466796875, 0.056640625, 0.2021636962890625, -0.7811279296875, -0.1665496826171875, -0.1953887939453125, -0.1611175537109375, 0.1687030792236328, 0.062896728515625, 2.99920654296875, 0.71575927734375, 0.64556884765625, 1.0980377197265625, -0.5281219482421875, -0.717620849609375, 1.21856689453125, 0.5938262939453125, -1.8483810424804688, 0.994293212890625, 0.559326171875, 1.0279388427734375, 0.854949951171875, -0.7538986206054688, 0.5184326171875, 0.73309326171875, 0.3204803466796875, 0.7982025146484375, 2.172515869140625, 2.29498291015625, 0.0, -0.75885009765625, 0.674072265625, 0.5547103881835938, 1.1986083984375, 1.8372802734375, 0.07769012451171875, 0.8572540283203125, -0.3129768371582031, -0.0089263916015625, 2.07513427734375, 1.309539794921875, 1.9322509765625, 0.2866973876953125, 0.13543701171875, 0.0532073974609375, 0.118896484375, 1.39031982421875, -3.8856201171875, 0.537353515625, 0.891845703125, 0.8207855224609375, 1.2159576416015625, -0.15514373779296875, -3.0189208984375, -0.6597938537597656, 0.66607666015625, -0.249298095703125, 0.4449462890625, 0.4084625244140625, 0.85015869140625, 0.742401123046875, 0.6786651611328125, 0.6582489013671875, -2.1728515625, -0.0955810546875, -0.2919197082519531, 0.0854034423828125, -0.62835693359375, 0.40674591064453125, 1.1270751953125, 2.43023681640625, 1.19268798828125, 1.19091796875, -1.214263916015625, 0.1247100830078125, 1.486297607421875, -0.749053955078125, -0.11798095703125, -0.8038787841796875, -0.019998550415039062, 0.16291046142578125, -0.2806854248046875, 1.1043701171875, 0.5285415649414062, -0.3078765869140625, 1.748077392578125, 0.7308349609375, 2.3612060546875, 0.1229095458984375, 0.036956787109375, 2.97705078125, -0.8370819091796875, 0.3114452362060547, 0.49395751953125, 0.048259735107421875, 0.298614501953125, -0.707122802734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000036.npy"}
{"epoch": 0.07539267015706806, "step": 37, "batch_size": 128, "mean": 0.15439796447753906, "std": 1.1045141220092773, "min": -3.19122314453125, "p10": -1.0591064453125, "median": 0.13950729370117188, "p90": 1.3684204101562498, "max": 4.685302734375, "pos_frac": 0.5625, "sample": [-0.913238525390625, 0.4751434326171875, 0.5414581298828125, -0.1318359375, 1.552581787109375, -0.10166740417480469, 2.075439453125, 0.2752952575683594, 1.1276092529296875, 0.307220458984375, -2.21783447265625, -0.06979942321777344, -1.073974609375, -0.21097564697265625, -0.04449462890625, 3.598785400390625, 0.732330322265625, -0.076690673828125, 1.3912353515625, -0.81927490234375, 0.45375823974609375, 0.295501708984375, -0.016019821166992188, -1.68206787109375, -1.052734375, -0.0258941650390625, -0.11937713623046875, 0.855987548828125, 1.0503082275390625, -1.5368194580078125, 0.2988433837890625, -0.4251251220703125, -0.4157981872558594, -1.960357666015625, 1.143157958984375, 0.146392822265625, -3.19122314453125, 0.367889404296875, -0.30426025390625, 0.824615478515625, 0.2617950439453125, -0.18792724609375, -0.43474578857421875, 0.41680908203125, 1.65496826171875, 1.605010986328125, 0.0842599868774414, 0.35530853271484375, -0.06308746337890625, -0.9215087890625, 0.195892333984375, 0.413543701171875, -0.4012451171875, 3.22320556640625, -0.863800048828125, 0.115875244140625, 0.2661895751953125, 0.986297607421875, -1.009857177734375, 0.593292236328125, 1.2923583984375, 1.358642578125, 0.22425079345703125, -0.5238037109375, -0.61492919921875, 0.13768768310546875, 0.0859375, 0.45562744140625, 0.471588134765625, 2.62249755859375, 1.2394561767578125, 0.11376953125, -0.0107421875, 0.74591064453125, 0.633544921875, 0.900970458984375, 0.47802734375, 1.29901123046875, 0.397247314453125, -1.307464599609375, 0.1615142822265625, 0.76220703125, 0.9179458618164062, 0.0274658203125, -0.0319366455078125, -2.308258056640625, 0.12664794921875, -0.6368408203125, 2.198638916015625, 1.7274169921875, -0.536834716796875, 0.7371826171875, -1.215911865234375, -0.66522216796875, -0.29224395751953125, 0.3546867370605469, 0.908660888671875, -0.267730712890625, -0.9355850219726562, 0.50341796875, -1.4677734375, -0.5564613342285156, -0.88165283203125, -0.6042633056640625, 0.5740699768066406, -0.1529541015625, 4.685302734375, 1.13165283203125, 0.72381591796875, 0.5586814880371094, 0.141326904296875, -0.408905029296875, -0.0684967041015625, -1.137420654296875, 0.4412841796875, -0.85919189453125, -1.04571533203125, -0.7780914306640625, -0.266448974609375, -0.2303619384765625, 0.15924072265625, -1.2559051513671875, 0.27227783203125, 0.2402191162109375, 0.02496337890625, 1.8790130615234375, -1.7867431640625, 1.4822998046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000037.npy"}
{"epoch": 0.0774869109947644, "step": 38, "batch_size": 128, "mean": 0.024315327405929565, "std": 0.8978682160377502, "min": -3.151611328125, "p10": -1.02437744140625, "median": 0.07641220092773438, "p90": 1.006391906738281, "max": 2.853302001953125, "pos_frac": 0.5390625, "sample": [-0.470703125, 0.3779296875, 0.017541885375976562, 0.5651779174804688, 0.15364837646484375, 0.5768661499023438, -0.7731781005859375, 0.973876953125, 0.544525146484375, 1.0822601318359375, 1.5145263671875, -0.28514862060546875, 0.7235870361328125, -0.453216552734375, 0.330322265625, -0.1253662109375, -0.0095062255859375, 0.2076873779296875, 0.1301727294921875, -0.2617340087890625, -0.14453125, -0.527862548828125, 0.8426151275634766, 0.01961517333984375, -0.6448898315429688, -0.928558349609375, -1.327880859375, 0.57379150390625, 0.96435546875, -0.17396163940429688, -1.017059326171875, 1.090240478515625, -0.97174072265625, -0.3978271484375, 0.529327392578125, -0.7021636962890625, -0.172576904296875, -0.1251068115234375, 0.1464080810546875, -0.059234619140625, -1.0804290771484375, 0.09386444091796875, 1.75225830078125, -0.04349517822265625, -1.2774658203125, 0.4093780517578125, -1.01727294921875, 1.13800048828125, 0.823760986328125, 0.7545166015625, -0.904052734375, 0.03377532958984375, -2.247467041015625, -0.1758270263671875, 0.07826995849609375, -0.7221298217773438, -2.41064453125, 0.27447509765625, 0.074554443359375, -0.01031494140625, -0.1947174072265625, -0.8863525390625, 0.385955810546875, -1.0194244384765625, 0.297210693359375, -0.03529167175292969, 0.25508880615234375, -1.445404052734375, 1.685089111328125, 0.11767578125, 0.54931640625, -0.2076416015625, 0.86492919921875, 1.15753173828125, -0.121856689453125, -1.8818359375, 1.6884918212890625, -0.06027984619140625, -0.21250152587890625, -3.151611328125, 0.6093215942382812, -0.8069610595703125, -0.150299072265625, 0.4886474609375, 0.105438232421875, 0.8505859375, 0.234100341796875, -0.31341552734375, 0.585845947265625, -0.054492950439453125, 0.5905380249023438, 0.4894866943359375, 2.853302001953125, 0.3282623291015625, 0.36737060546875, -0.203216552734375, 0.33990478515625, 0.4345703125, 1.696197509765625, -1.4410400390625, 0.7368316650390625, 0.0, 0.44244384765625, -1.47393798828125, 0.3826904296875, -0.269683837890625, -0.496337890625, 0.6425857543945312, -0.443450927734375, 0.17095947265625, 1.822174072265625, -1.670654296875, -1.0359344482421875, -0.69140625, 1.65576171875, 0.1721649169921875, 0.66961669921875, 0.7056884765625, 0.33347320556640625, -0.61871337890625, -0.8763427734375, 0.2299652099609375, -1.5635986328125, 1.219268798828125, 0.01190185546875, 0.82598876953125, -0.3634357452392578, 0.46984100341796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000038.npy"}
{"epoch": 0.07958115183246073, "step": 39, "batch_size": 128, "mean": 0.4311588406562805, "std": 1.0081778764724731, "min": -1.703094482421875, "p10": -0.8006683349609375, "median": 0.37277984619140625, "p90": 1.7780548095703126, "max": 3.10333251953125, "pos_frac": 0.625, "sample": [0.5688629150390625, -1.235748291015625, -1.703094482421875, -0.1671600341796875, -0.2752685546875, 1.2798309326171875, -0.2069854736328125, 0.3763275146484375, -0.8111572265625, 0.93890380859375, 1.000244140625, 1.727935791015625, 0.86151123046875, 1.45428466796875, -0.504608154296875, 2.44110107421875, -0.5784912109375, 2.34161376953125, 0.8595123291015625, 0.388885498046875, -0.376495361328125, -0.078399658203125, 0.41524505615234375, -0.91656494140625, 2.99114990234375, 1.9188766479492188, 0.7384414672851562, 0.1593475341796875, -0.1178131103515625, 0.33270263671875, 2.4326171875, 0.443511962890625, 1.121429443359375, 0.25806427001953125, 2.4453125, -0.309600830078125, -0.30214691162109375, 1.0980148315429688, 1.342041015625, -1.1905059814453125, 0.36224365234375, 0.24559402465820312, 1.511871337890625, 0.807403564453125, 0.25399017333984375, 1.232940673828125, 2.22021484375, 0.00638580322265625, 0.4366912841796875, 2.66412353515625, -0.882843017578125, -0.6109466552734375, 0.4257965087890625, -1.0416259765625, 1.0537109375, -0.3935394287109375, -1.27880859375, 1.740875244140625, 1.00640869140625, -0.94775390625, -0.23933792114257812, -0.3471527099609375, -0.5463409423828125, -0.656341552734375, 0.5336990356445312, 0.21936798095703125, -0.05464935302734375, -0.21422767639160156, 0.8794708251953125, -0.1987457275390625, -0.20391845703125, 0.47235107421875, -0.2678966522216797, -0.7657470703125, 0.891815185546875, 2.072235107421875, 0.57666015625, -0.796173095703125, 1.1500320434570312, -1.12994384765625, 1.3961181640625, -0.13116073608398438, 0.53778076171875, 0.4760894775390625, 1.0968780517578125, 0.1050872802734375, 3.10333251953125, 1.73638916015625, 1.2433319091796875, -0.43344879150390625, 0.20354461669921875, 0.120697021484375, 0.3226280212402344, -0.531890869140625, 0.0, -0.192596435546875, 0.055927276611328125, -1.031707763671875, 1.7764892578125, 1.305572509765625, 1.781707763671875, -0.5383224487304688, 2.50543212890625, -0.1360626220703125, 0.995330810546875, 0.8754119873046875, -0.626800537109375, 0.4891357421875, 1.760833740234375, -1.383758544921875, 0.369232177734375, 0.42665863037109375, -0.299407958984375, 0.4201622009277344, 1.56365966796875, 0.015193939208984375, 0.904693603515625, 1.140960693359375, 0.605377197265625, -0.04317665100097656, -0.63568115234375, 0.5860137939453125, -1.28143310546875, 1.90716552734375, 0.34234619140625, 0.70831298828125, -0.4742088317871094, 0.7008743286132812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000039.npy"}
{"epoch": 0.08167539267015707, "step": 40, "batch_size": 128, "mean": 0.33518385887145996, "std": 1.1506041288375854, "min": -3.188751220703125, "p10": -1.0065414428710937, "median": 0.1920757293701172, "p90": 1.8826690673828124, "max": 4.490966796875, "pos_frac": 0.640625, "sample": [-0.24496078491210938, -1.77850341796875, -0.649261474609375, 1.3311767578125, -0.3164081573486328, 0.4826507568359375, -1.96783447265625, 1.097412109375, -0.1531524658203125, 3.1038818359375, 1.0635986328125, 1.0602264404296875, 0.707916259765625, -1.2357177734375, 0.7817840576171875, 0.3806266784667969, -0.23973846435546875, 1.896148681640625, -0.13019752502441406, 0.62451171875, -1.0807647705078125, 3.186370849609375, -0.2508678436279297, 0.26168060302734375, 0.6944808959960938, 2.218017578125, 0.0853271484375, 1.46685791015625, 0.0520782470703125, 0.657501220703125, -0.6734733581542969, 0.98577880859375, -1.1188735961914062, -0.10504150390625, 0.378662109375, -0.285308837890625, 0.2301788330078125, -0.5939788818359375, 2.799072265625, -0.8671875, 4.490966796875, 1.62091064453125, -2.48919677734375, -1.5482177734375, 0.19701766967773438, -0.5293998718261719, 0.045379638671875, 1.9454193115234375, 1.19110107421875, 1.9398956298828125, 0.25998687744140625, 0.409942626953125, 0.488128662109375, 0.171478271484375, 1.128448486328125, 0.96331787109375, 1.87689208984375, 0.5670013427734375, 1.43389892578125, -0.065185546875, -0.518768310546875, 0.027126312255859375, 0.1868133544921875, -3.188751220703125, -0.83404541015625, -0.91033935546875, 0.361907958984375, -0.5634002685546875, 0.595916748046875, 0.7489013671875, -0.256683349609375, -0.710723876953125, 0.6269302368164062, 1.813385009765625, 0.558135986328125, -1.1338119506835938, 1.03497314453125, 0.22581100463867188, 0.08349609375, -0.9747314453125, 0.0014801025390625, 1.740570068359375, 0.09674835205078125, -0.3627510070800781, -0.005157470703125, 0.5808601379394531, -0.6019515991210938, 2.3294677734375, -0.5328826904296875, 0.1871337890625, -1.15838623046875, -1.113983154296875, 0.813262939453125, 1.021026611328125, 0.344329833984375, 2.00616455078125, 1.49493408203125, -0.5067977905273438, 2.335205078125, 1.1023330688476562, 1.03662109375, 1.8726959228515625, -0.468505859375, 0.05818939208984375, -0.822723388671875, 1.9237060546875, 0.4298553466796875, 0.026845932006835938, 0.161285400390625, 0.171844482421875, -1.67999267578125, 1.05084228515625, -0.09149169921875, 0.13903045654296875, -0.394927978515625, -1.262847900390625, -0.5418701171875, 0.9431915283203125, 2.326873779296875, 1.256622314453125, 0.55230712890625, 0.09296989440917969, -0.42340087890625, 0.142791748046875, -0.0843353271484375, 1.0571746826171875, 0.138153076171875, 0.39642333984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000040.npy"}
{"epoch": 0.08376963350785341, "step": 41, "batch_size": 128, "mean": 0.22692617774009705, "std": 1.103472352027893, "min": -3.191436767578125, "p10": -0.9709808349609375, "median": 0.2468109130859375, "p90": 1.4909179687499998, "max": 2.9769287109375, "pos_frac": 0.609375, "sample": [2.280181884765625, 0.3263702392578125, -0.438201904296875, -0.7135467529296875, 1.462646484375, -1.288726806640625, -1.451263427734375, 0.5279998779296875, -0.4175262451171875, -0.3987884521484375, 0.8686981201171875, 1.819610595703125, 0.7241363525390625, -0.6235198974609375, -0.7689056396484375, 1.809967041015625, 0.529296875, 0.098663330078125, 0.24969482421875, 0.571868896484375, 0.27716827392578125, 1.324859619140625, -0.1673736572265625, 0.9986724853515625, -0.875457763671875, -0.3336334228515625, 0.67681884765625, 0.23101806640625, -0.7298126220703125, 0.5038585662841797, 0.7948760986328125, -3.1900634765625, -0.4842262268066406, 1.010009765625, 0.9251708984375, -0.4011058807373047, 2.08624267578125, 2.9769287109375, 0.991119384765625, 0.1397705078125, 2.7232666015625, 0.85821533203125, -0.2571563720703125, 1.048492431640625, 0.9429931640625, -0.650421142578125, 0.135528564453125, 0.1786956787109375, 1.03643798828125, 0.734893798828125, -0.49468994140625, -0.642425537109375, -0.586029052734375, 0.537384033203125, 0.7586669921875, -2.30645751953125, 0.40819549560546875, 0.62408447265625, 0.1185302734375, 0.7905044555664062, 0.37164306640625, -0.789031982421875, 1.04815673828125, -0.0623779296875, -1.014923095703125, 0.5913543701171875, -0.9289093017578125, 0.7258224487304688, 0.0, -0.448516845703125, -0.6020736694335938, 1.4593505859375, -0.4407196044921875, 0.5828857421875, -3.191436767578125, -2.5313720703125, 1.7501907348632812, 0.1950531005859375, 0.200164794921875, -1.289947509765625, 0.05584716796875, 0.1403045654296875, -0.5869216918945312, 1.0446701049804688, 0.365447998046875, 1.22052001953125, 0.7918930053710938, 1.3736572265625, 2.5355377197265625, 0.36395263671875, -1.40081787109375, 0.331298828125, 0.4101600646972656, -0.37884521484375, -0.991424560546875, 2.48480224609375, -1.5634765625, 0.053375244140625, 0.07240867614746094, 0.244232177734375, -0.12115478515625, 1.013763427734375, 0.959808349609375, -0.0614471435546875, -0.546173095703125, -0.02335357666015625, -0.1340484619140625, -0.6108551025390625, -0.64447021484375, -0.288818359375, 1.2477779388427734, 2.626373291015625, 0.39300537109375, 0.11785888671875, 1.556884765625, -0.075103759765625, 2.037109375, -0.96221923828125, 0.4012298583984375, -0.188629150390625, -2.51617431640625, 0.99285888671875, 1.66278076171875, 0.2493896484375, 1.3655242919921875, 1.128509521484375, 0.3956451416015625, -1.003662109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000041.npy"}
{"epoch": 0.08586387434554973, "step": 42, "batch_size": 128, "mean": 0.35740482807159424, "std": 1.1481380462646484, "min": -3.435760498046875, "p10": -0.7546035766601562, "median": 0.2852630615234375, "p90": 1.7707527160644532, "max": 3.683624267578125, "pos_frac": 0.6328125, "sample": [-0.7102203369140625, -0.4112548828125, 0.7682952880859375, 1.91534423828125, 0.093536376953125, 1.7294921875, -0.35400390625, 1.991424560546875, 1.631500244140625, -0.42955780029296875, 0.637176513671875, 2.93072509765625, 0.89691162109375, 0.100341796875, -1.4471435546875, -0.1523723602294922, 1.1839599609375, 0.43408203125, -0.10260772705078125, 3.683624267578125, -0.266082763671875, -0.3348846435546875, 0.1876373291015625, 0.289031982421875, 0.813507080078125, -0.3650665283203125, 0.682373046875, 0.10241508483886719, 0.76092529296875, -0.7191009521484375, 0.6237335205078125, 0.6810798645019531, 0.506317138671875, 1.216796875, 2.164306640625, 0.7854385375976562, 0.5125274658203125, 0.5042572021484375, -1.181243896484375, 0.117950439453125, -0.09476470947265625, -0.10108184814453125, -0.26385498046875, -0.2157440185546875, 1.12103271484375, 0.242645263671875, 1.2297439575195312, -0.665496826171875, 0.36395263671875, 0.24685287475585938, 1.03179931640625, 0.2435741424560547, 0.9921493530273438, 0.5647354125976562, -0.0742340087890625, -0.2901878356933594, 0.3523101806640625, -0.671417236328125, 0.6243743896484375, 0.04434967041015625, 0.56396484375, -1.228790283203125, -1.025665283203125, -1.344512939453125, -0.00469970703125, 1.578704833984375, -0.04981231689453125, 0.22107696533203125, -0.73553466796875, 0.3463935852050781, 0.851165771484375, 0.4670219421386719, 0.208343505859375, 2.32373046875, -0.46466064453125, 0.281494140625, -3.43084716796875, -0.791229248046875, 1.34747314453125, -0.5692558288574219, 0.7305679321289062, 2.66619873046875, 1.7130584716796875, -0.2998504638671875, 1.93170166015625, 0.520355224609375, 2.52484130859375, 0.687469482421875, 0.018810272216796875, 0.229095458984375, 1.5197601318359375, -2.0933837890625, 0.57989501953125, 0.5724525451660156, 1.487274169921875, 0.34002685546875, -1.9974365234375, 2.7135009765625, -0.11211395263671875, -0.2751617431640625, 1.0291900634765625, 0.4261627197265625, 0.0, 1.411651611328125, -0.1873779296875, -0.3005256652832031, 0.19207763671875, 0.1301422119140625, -0.1545257568359375, 1.0809326171875, 1.49456787109375, -1.883514404296875, 0.9946746826171875, 0.29808807373046875, 2.21826171875, 3.03594970703125, 1.770233154296875, -0.9070281982421875, 1.7719650268554688, -0.7389068603515625, -0.2018280029296875, 1.44549560546875, -3.435760498046875, 0.307098388671875, -2.084930419921875, 0.1939239501953125, -0.2906646728515625, -0.02086639404296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000042.npy"}
{"epoch": 0.08795811518324607, "step": 43, "batch_size": 128, "mean": 0.365061491727829, "std": 1.20891273021698, "min": -2.3779296875, "p10": -0.9992218017578124, "median": 0.2457122802734375, "p90": 1.8779296875, "max": 4.171142578125, "pos_frac": 0.625, "sample": [0.473236083984375, 1.8907470703125, 1.161041259765625, -0.16278076171875, 0.7872161865234375, 0.6944732666015625, 3.484893798828125, 0.371368408203125, -0.016384124755859375, -0.19281768798828125, 0.0538330078125, -0.6234283447265625, 0.862335205078125, 2.08099365234375, -1.0467529296875, -0.374847412109375, -1.0836563110351562, -0.6718292236328125, 0.7648162841796875, 0.243499755859375, 2.174560546875, 0.044403076171875, 1.57110595703125, 1.1485137939453125, -0.978851318359375, -1.4671783447265625, -0.849822998046875, -0.5498199462890625, -0.06962394714355469, 1.795989990234375, 2.61810302734375, 0.7896728515625, -1.64581298828125, -2.35394287109375, 3.25714111328125, 0.5901985168457031, -0.42876434326171875, 0.135467529296875, 0.5079154968261719, 0.5463638305664062, 0.6147613525390625, 0.548004150390625, -0.7149448394775391, -0.13714599609375, 0.161285400390625, -0.1783618927001953, -0.44140625, -0.5932846069335938, -2.3779296875, -0.746734619140625, -2.2183990478515625, 0.5981292724609375, 0.30740928649902344, 0.8979263305664062, 2.903076171875, 1.8724365234375, 1.0635986328125, 0.35369873046875, -0.7150726318359375, 1.3193359375, 0.6031036376953125, 0.0645294189453125, 1.117462158203125, 0.0457763671875, 1.4144287109375, -2.211517333984375, 1.60693359375, 0.2479248046875, 0.72357177734375, 1.721954345703125, 0.10760498046875, 0.871063232421875, 0.3780517578125, 1.3711700439453125, -0.8653564453125, -0.5970993041992188, -1.60498046875, 0.169158935546875, -0.280181884765625, 0.53619384765625, -0.0391693115234375, 0.4268798828125, 1.26171875, -0.66314697265625, 0.632781982421875, -0.03570556640625, 0.3884124755859375, 0.0081634521484375, 0.153717041015625, 0.193878173828125, 1.783172607421875, 1.40325927734375, -0.34234619140625, 0.2222747802734375, 1.9071044921875, -0.5283889770507812, -0.27043914794921875, 4.171142578125, -1.142181396484375, 0.8092117309570312, 0.027099609375, 1.4715957641601562, 3.405242919921875, 0.0536041259765625, -1.525299072265625, 1.12371826171875, 2.55889892578125, 0.31982421875, 2.1434783935546875, 2.0814208984375, -0.4433746337890625, -0.160675048828125, 1.1641616821289062, -2.243316650390625, -0.81268310546875, 1.6636810302734375, 1.1250152587890625, 0.6490325927734375, 0.874603271484375, 0.56744384765625, 0.0700531005859375, -1.70782470703125, -0.0063381195068359375, 1.28851318359375, -0.013946533203125, -0.45059967041015625, -0.2091217041015625, -0.09442710876464844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000043.npy"}
{"epoch": 0.09005235602094241, "step": 44, "batch_size": 128, "mean": 0.4912201762199402, "std": 1.1407415866851807, "min": -2.500396728515625, "p10": -0.937530517578125, "median": 0.4418067932128906, "p90": 2.0762207031249997, "max": 3.3604736328125, "pos_frac": 0.703125, "sample": [0.834136962890625, -0.3389129638671875, 2.23583984375, 0.5279159545898438, -0.8182525634765625, 2.25738525390625, 0.18080902099609375, 0.046764373779296875, 0.76226806640625, -0.06097602844238281, 0.9350738525390625, 1.207366943359375, 0.514862060546875, -0.40842437744140625, 1.493682861328125, 0.93603515625, -0.6173095703125, 0.22899436950683594, -1.68292236328125, 1.5722808837890625, 0.673004150390625, -0.580322265625, 0.790771484375, 0.43842315673828125, 1.0122299194335938, 0.9705123901367188, 1.751922607421875, 0.2897987365722656, -1.286773681640625, -0.22885513305664062, 1.3857269287109375, -0.968505859375, 0.7120361328125, -0.5735626220703125, 0.2655372619628906, 0.75482177734375, 0.858551025390625, -0.5163993835449219, 2.898284912109375, 1.4259185791015625, 2.29638671875, -0.35303306579589844, -0.12451171875, 2.806060791015625, 0.4451904296875, 1.82977294921875, 2.1444091796875, 2.4801025390625, 0.5653228759765625, 0.211273193359375, 0.1523876190185547, 0.13665771484375, -1.0302963256835938, 0.3431396484375, -0.1984710693359375, 1.3929290771484375, -0.974365234375, 2.1829833984375, 0.18534469604492188, -0.903045654296875, 0.2787628173828125, 2.28057861328125, 0.7854995727539062, 1.5653076171875, 0.68060302734375, 0.3285064697265625, 0.29541015625, -2.500396728515625, 2.0469970703125, 2.209259033203125, -1.29046630859375, 0.4869842529296875, 1.5938720703125, 1.3778076171875, 1.835357666015625, 1.072784423828125, 0.771240234375, -2.307891845703125, -0.400482177734375, -1.488311767578125, 2.174591064453125, 1.2313079833984375, 1.959503173828125, 0.21094322204589844, 0.23150634765625, 0.41794586181640625, 0.4337158203125, -0.92425537109375, 3.00811767578125, -0.4376373291015625, 0.456451416015625, 1.202880859375, -0.61419677734375, 1.35992431640625, 0.614715576171875, 1.650177001953125, 0.67388916015625, -1.22186279296875, -0.654022216796875, 0.285491943359375, 1.2618408203125, 0.11996650695800781, -1.2134246826171875, 0.519561767578125, -0.02129364013671875, 1.76251220703125, -0.8782958984375, 1.01153564453125, 0.149383544921875, 0.06378173828125, 3.3604736328125, 0.1571044921875, 0.49214935302734375, -0.379547119140625, 0.37609100341796875, 1.719451904296875, -0.06333160400390625, 0.6832122802734375, -2.11383056640625, -0.19779205322265625, -2.2115631103515625, 0.7050704956054688, 0.349395751953125, 0.974945068359375, -0.14495849609375, -0.6981887817382812, 0.2747650146484375, 1.6685791015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000044.npy"}
{"epoch": 0.09214659685863874, "step": 45, "batch_size": 128, "mean": 0.5517077445983887, "std": 1.4023394584655762, "min": -3.134033203125, "p10": -1.0079132080078124, "median": 0.3339691162109375, "p90": 2.319564819335937, "max": 6.4661865234375, "pos_frac": 0.671875, "sample": [-0.820648193359375, 0.6134796142578125, 1.20196533203125, -0.197418212890625, -0.2582130432128906, 1.0086669921875, 0.3536205291748047, 0.125579833984375, 1.3503150939941406, 0.07476806640625, 2.2750244140625, 0.83746337890625, 2.042724609375, 0.0389251708984375, 0.1456298828125, 0.173370361328125, -0.1401214599609375, -0.083251953125, 0.0, -3.134033203125, 0.829559326171875, 0.66278076171875, 0.16326904296875, 1.20086669921875, -1.184112548828125, 3.5167236328125, 0.0026702880859375, 1.073577880859375, 0.6136016845703125, -0.0462799072265625, -0.110931396484375, -0.3834075927734375, 1.845703125, 0.4410247802734375, 0.060626983642578125, 0.018218994140625, 0.341827392578125, 0.2237701416015625, 2.423492431640625, 2.66314697265625, 2.8883056640625, 0.69000244140625, 0.6948013305664062, 0.32611083984375, 3.55718994140625, 0.3574790954589844, 1.4539642333984375, -0.1277313232421875, -0.59716796875, 1.391815185546875, -0.000213623046875, 0.828704833984375, -1.567962646484375, 0.2474822998046875, 0.3433952331542969, 0.5461578369140625, -0.38488006591796875, 0.8246917724609375, -0.080780029296875, -2.290496826171875, -1.234619140625, 1.2789306640625, 0.08252716064453125, 0.151031494140625, -2.19329833984375, 1.241180419921875, 0.900390625, -0.16034698486328125, 0.842315673828125, 0.09326553344726562, 0.1346435546875, 2.643096923828125, 1.1300048828125, 0.7165946960449219, 3.433441162109375, 0.25390625, 1.2367401123046875, 0.070343017578125, 0.61614990234375, -0.217437744140625, 0.726409912109375, 3.6741943359375, 0.6348381042480469, 1.980804443359375, 0.19573974609375, -0.9119415283203125, -0.981903076171875, 0.1937103271484375, -0.5819931030273438, -1.975921630859375, -0.145599365234375, 3.7572021484375, -0.19696044921875, -2.269744873046875, 4.7030029296875, 0.20556640625, 1.162353515625, 0.405975341796875, 1.950927734375, 1.17242431640625, -0.07757568359375, -0.74151611328125, -0.14400482177734375, -1.4277496337890625, 1.95166015625, -1.068603515625, 0.727783203125, -1.922515869140625, 0.5680160522460938, 0.660614013671875, -0.5718231201171875, -1.1343841552734375, 1.18719482421875, 2.44097900390625, -0.05202484130859375, 1.6754150390625, 2.08502197265625, -0.21002197265625, 6.4661865234375, 2.213592529296875, 0.32550048828125, -0.092559814453125, 0.9287071228027344, 0.533782958984375, 2.9188232421875, -0.18521881103515625, -1.20233154296875, 0.9828567504882812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000045.npy"}
{"epoch": 0.09424083769633508, "step": 46, "batch_size": 128, "mean": 0.6709342002868652, "std": 1.3327085971832275, "min": -3.741119384765625, "p10": -0.8287475585937499, "median": 0.5205039978027344, "p90": 2.6155120849609372, "max": 4.52740478515625, "pos_frac": 0.6953125, "sample": [0.48358154296875, 4.52740478515625, 1.74609375, 0.36989593505859375, 0.205474853515625, 1.872100830078125, 0.923309326171875, 0.38427734375, -0.857757568359375, -0.9699478149414062, -2.2100067138671875, -0.220855712890625, 1.84808349609375, -0.6639404296875, 1.1338348388671875, -1.6414642333984375, -0.40594482421875, 0.1978931427001953, 0.777557373046875, -0.920745849609375, 0.010772705078125, 0.79168701171875, -0.59326171875, 1.3653717041015625, -0.6262359619140625, 0.09515380859375, -1.23699951171875, -0.0165557861328125, 2.8929443359375, -0.28925323486328125, -0.2874336242675781, 0.5198287963867188, 1.6372833251953125, 1.5864105224609375, 1.421112060546875, 3.72039794921875, 1.64727783203125, 0.18958282470703125, -0.639862060546875, 0.8302001953125, 1.261199951171875, 1.308990478515625, 1.052337646484375, 0.0373992919921875, 2.6826171875, -2.453857421875, 2.957366943359375, 0.3066368103027344, 0.69189453125, -0.31136322021484375, 2.2451171875, 2.3667449951171875, 0.7724990844726562, 0.54791259765625, 0.0188446044921875, 1.5150146484375, 0.3575439453125, 3.1004638671875, -0.6214675903320312, 3.163543701171875, 0.43953704833984375, 0.6872329711914062, 0.02303314208984375, -0.47100830078125, 2.66009521484375, -0.48423004150390625, 0.25679779052734375, -0.924591064453125, -1.147979736328125, 1.5809326171875, -0.816314697265625, 0.29872894287109375, 0.594757080078125, 0.227386474609375, -0.547515869140625, 1.7108154296875, -0.2810516357421875, -0.6249542236328125, -0.0075531005859375, 0.4901123046875, 0.93359375, 2.703125, 1.5379180908203125, 1.04278564453125, 1.71160888671875, 0.6931228637695312, -3.741119384765625, 0.017333984375, 1.0589752197265625, 1.795867919921875, -0.520263671875, 1.14727783203125, 2.040283203125, 0.6851882934570312, -0.18981552124023438, 3.796905517578125, 0.4850921630859375, -0.933807373046875, -1.12359619140625, 0.9074859619140625, -0.08239173889160156, 1.081298828125, 2.236053466796875, -0.4698333740234375, 3.6322021484375, 0.9336624145507812, 0.99493408203125, 0.2654876708984375, 1.5301513671875, 0.627777099609375, 2.596405029296875, -1.0186767578125, 0.4061279296875, 0.5600986480712891, -0.1823272705078125, 2.82550048828125, -0.194580078125, 1.402130126953125, 1.6536865234375, 3.250396728515625, 0.0, 1.570709228515625, 0.14656829833984375, -0.504180908203125, 0.11679840087890625, 1.77703857421875, 0.52117919921875, 1.89447021484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000046.npy"}
{"epoch": 0.09633507853403141, "step": 47, "batch_size": 128, "mean": 0.3118368089199066, "std": 1.44316828250885, "min": -3.73486328125, "p10": -1.3653411865234375, "median": 0.34197235107421875, "p90": 2.188339233398436, "max": 4.08868408203125, "pos_frac": 0.578125, "sample": [-1.3489990234375, -0.0414581298828125, -0.065185546875, -0.35755157470703125, 2.844024658203125, -2.338043212890625, -1.743133544921875, 0.1125335693359375, 1.732635498046875, -1.304840087890625, -0.454864501953125, 0.4481773376464844, 0.62530517578125, -0.4034576416015625, -1.82025146484375, -0.3021392822265625, 0.7630615234375, -0.2158203125, 1.275482177734375, -0.053009033203125, 1.267913818359375, -2.43017578125, 0.36220550537109375, -0.135528564453125, 0.548919677734375, 3.07049560546875, 0.33193206787109375, -0.5916748046875, -2.87701416015625, -1.9034271240234375, 1.6779632568359375, 0.0, 1.1349029541015625, -0.7457122802734375, -0.361907958984375, 1.25469970703125, 2.793487548828125, -0.41180419921875, 0.363433837890625, 0.5586795806884766, -1.403472900390625, 0.20340728759765625, 2.53814697265625, 2.8172607421875, 0.9440155029296875, 1.090972900390625, -0.1387939453125, 1.020263671875, 0.773345947265625, 0.5584964752197266, -0.1555938720703125, -0.16017532348632812, 3.36614990234375, -0.0125274658203125, -2.7298583984375, 2.562835693359375, -3.425811767578125, 1.8826904296875, 1.2643585205078125, 0.4822540283203125, -1.003173828125, 1.149566650390625, -0.3879852294921875, 1.09796142578125, 1.5306167602539062, -0.9597015380859375, 0.12188720703125, 0.7383880615234375, 0.5389404296875, 2.84320068359375, 0.783782958984375, 0.4785423278808594, -0.294921875, -0.37677001953125, 2.038421630859375, 1.2166061401367188, 0.0, 1.6881256103515625, 4.08868408203125, 1.481414794921875, 0.35201263427734375, -0.36721038818359375, 0.29811859130859375, -0.6629562377929688, 0.605987548828125, -0.228302001953125, 0.22212600708007812, 0.577117919921875, 0.17201995849609375, 1.2164306640625, 1.5052719116210938, -0.76531982421875, -0.038116455078125, 1.3699951171875, 1.2077560424804688, 0.3709716796875, 1.230072021484375, 0.5834197998046875, 3.16802978515625, 1.2120208740234375, -1.3280487060546875, 1.13665771484375, -0.2921600341796875, 0.6475830078125, -1.490142822265625, -1.98858642578125, 0.06796646118164062, -0.998291015625, -0.1527252197265625, -0.7193450927734375, -0.773101806640625, 0.6064910888671875, 3.10968017578125, 0.4511699676513672, 1.4245147705078125, 3.25982666015625, -1.1075210571289062, 0.6959514617919922, -0.5208587646484375, -1.05029296875, -0.353302001953125, 2.953948974609375, 0.8496780395507812, 0.188995361328125, 0.8970947265625, -3.73486328125, -3.5062255859375, 0.10016632080078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000047.npy"}
{"epoch": 0.09842931937172775, "step": 48, "batch_size": 128, "mean": 0.6784173250198364, "std": 1.4345860481262207, "min": -1.99151611328125, "p10": -0.7670806884765625, "median": 0.4258232116699219, "p90": 2.418202972412109, "max": 7.52197265625, "pos_frac": 0.6640625, "sample": [0.400634765625, 1.5050048828125, 1.3348541259765625, 1.1701278686523438, 0.683380126953125, 0.975067138671875, 0.8365478515625, 3.325592041015625, 1.600067138671875, 0.411224365234375, -0.5882644653320312, -0.7787857055664062, 0.163909912109375, -1.71697998046875, -0.4886360168457031, 4.64556884765625, -0.56695556640625, -1.7173004150390625, -1.566619873046875, 0.5988616943359375, 0.976348876953125, 0.5507278442382812, 0.5977783203125, -0.738922119140625, 0.67327880859375, -0.6514816284179688, -1.9678955078125, 1.05816650390625, -1.99151611328125, 2.1210479736328125, 0.5864944458007812, 0.15106201171875, 1.7547607421875, 0.074554443359375, 2.5565185546875, 1.2413063049316406, -0.0400848388671875, 3.14105224609375, 1.0122528076171875, 2.820709228515625, 0.409423828125, 3.271514892578125, 0.353271484375, -0.34142303466796875, 0.7949066162109375, -0.3482818603515625, -1.90386962890625, -0.042816162109375, -0.10211181640625, 0.46352386474609375, 5.748687744140625, 0.9974555969238281, 0.1030426025390625, -0.0605010986328125, -1.02325439453125, 0.777740478515625, 0.7846527099609375, -0.35108184814453125, 0.13504791259765625, -0.22607421875, 0.1781005859375, 0.24908447265625, 0.147552490234375, 0.44042205810546875, -0.527191162109375, 0.00982666015625, -0.777099609375, -0.533203125, 2.3781509399414062, 1.085845947265625, -1.349884033203125, 1.46807861328125, 0.0, 0.6865997314453125, 0.97393798828125, 0.07942771911621094, 2.51165771484375, 0.510894775390625, 0.15782928466796875, 1.860992431640625, 0.0962677001953125, 1.60577392578125, 0.988494873046875, 2.09246826171875, 0.62603759765625, 2.178863525390625, 1.970611572265625, 0.7007598876953125, -0.06975555419921875, -0.2545166015625, -0.9090118408203125, 1.7103424072265625, -0.5681915283203125, -0.33753204345703125, 2.56915283203125, 3.4365234375, -0.762786865234375, 1.1757659912109375, -0.7779083251953125, 1.6660537719726562, -0.5413894653320312, 2.2395172119140625, 0.0, 1.92669677734375, 1.62274169921875, -0.025146484375, 7.52197265625, -0.10085678100585938, -0.22857666015625, 1.4527206420898438, -0.1083831787109375, 0.07033920288085938, 0.256011962890625, 3.26861572265625, -0.09296417236328125, 0.068450927734375, 0.554290771484375, -0.6951770782470703, 0.31972503662109375, 0.8508090972900391, 1.3877105712890625, 1.338226318359375, -0.7781829833984375, 1.805633544921875, 0.2453765869140625, -0.042449951171875, 2.9080810546875, 1.331878662109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000048.npy"}
{"epoch": 0.10052356020942409, "step": 49, "batch_size": 128, "mean": 0.6090023517608643, "std": 1.6293094158172607, "min": -4.868408203125, "p10": -1.1821136474609375, "median": 0.4855060577392578, "p90": 2.8009185791015616, "max": 6.168731689453125, "pos_frac": 0.7109375, "sample": [0.8670730590820312, 1.2405242919921875, 0.4837989807128906, 1.1396636962890625, -0.003631591796875, 0.1605682373046875, -0.880523681640625, 4.644287109375, 3.8176422119140625, 0.8864173889160156, 0.8367156982421875, 1.5965576171875, -1.865997314453125, -0.119049072265625, 0.0760345458984375, 2.41888427734375, 1.5304412841796875, 1.4087295532226562, 0.34542083740234375, 2.507659912109375, 0.67742919921875, 0.10271835327148438, 3.62701416015625, 0.18267822265625, 1.65130615234375, -0.585906982421875, -1.21490478515625, 6.168731689453125, 0.5967864990234375, 2.257568359375, 0.2131805419921875, 2.296234130859375, 3.0879669189453125, -0.152313232421875, 1.792633056640625, 0.5959262847900391, 0.5381851196289062, 0.1917266845703125, 0.1252288818359375, -1.667083740234375, 0.3103370666503906, 1.941650390625, -2.1647796630859375, 0.5694961547851562, 1.3076248168945312, 0.681671142578125, 1.1132049560546875, 1.37628173828125, 2.021820068359375, -1.299285888671875, -0.457672119140625, -1.9506378173828125, 0.12839508056640625, 0.4387054443359375, 0.1567096710205078, 3.22418212890625, 0.5645065307617188, 0.22198486328125, -0.216278076171875, 1.248321533203125, 0.23621368408203125, 2.973419189453125, -1.168060302734375, 3.21435546875, 0.4180908203125, 0.59381103515625, 0.659423828125, -1.127960205078125, 2.6961669921875, 1.6832275390625, 1.2132568359375, 1.8863983154296875, -0.6313629150390625, -0.5030479431152344, -0.044345855712890625, 1.593536376953125, -3.4708251953125, 0.0965576171875, 2.6196136474609375, -2.28125, 0.46714210510253906, 3.31781005859375, 1.915130615234375, -0.10904693603515625, 0.34508514404296875, 0.487213134765625, -3.115966796875, 1.560089111328125, 4.0299072265625, 2.72698974609375, 0.652099609375, -0.9164581298828125, 0.33237457275390625, 0.16253662109375, 0.1593170166015625, -1.415069580078125, -0.9203643798828125, 0.578155517578125, 0.4035682678222656, -0.9432373046875, -1.50091552734375, -0.99603271484375, -1.86688232421875, 0.6501312255859375, 3.374908447265625, -0.2542724609375, 0.27996826171875, 0.79571533203125, 0.7588653564453125, 0.38707733154296875, 3.2293243408203125, -0.77685546875, 0.9841995239257812, -0.8267631530761719, 0.6008453369140625, 0.5331954956054688, 0.10705184936523438, 4.087310791015625, 1.0010910034179688, 1.0198516845703125, -4.868408203125, 0.10326385498046875, -0.887969970703125, 0.906036376953125, 1.460601806640625, -0.767791748046875, -0.2275390625, -0.5207595825195312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000049.npy"}
{"epoch": 0.10261780104712041, "step": 50, "batch_size": 128, "mean": 0.42275872826576233, "std": 1.6448469161987305, "min": -4.336090087890625, "p10": -1.2930999755859371, "median": 0.4724693298339844, "p90": 1.9042495727539062, "max": 5.809661865234375, "pos_frac": 0.65625, "sample": [0.659271240234375, -0.2008056640625, 1.44378662109375, 0.2205982208251953, -0.323211669921875, -0.2973785400390625, 1.4371337890625, -0.47765350341796875, -0.50494384765625, 5.809661865234375, -0.23974609375, -0.18646812438964844, 1.336181640625, 0.670318603515625, 4.50732421875, 1.3967666625976562, 1.4710693359375, -3.371429443359375, 1.418304443359375, -0.25325775146484375, -0.58746337890625, -3.98114013671875, -1.9989776611328125, 1.0632476806640625, -0.4879150390625, 1.626922607421875, 1.5399856567382812, 0.9342498779296875, 1.59564208984375, -1.535400390625, -1.64007568359375, 3.510498046875, 0.349609375, -0.575775146484375, 0.38262176513671875, -0.7239990234375, -0.34308624267578125, 0.81622314453125, 1.362335205078125, 0.2215576171875, -0.9270248413085938, -1.6619110107421875, 1.6927490234375, 1.31585693359375, -0.3960418701171875, -0.381317138671875, -0.27691650390625, 1.42059326171875, 0.25119972229003906, 0.7280197143554688, 1.02899169921875, 0.2401885986328125, -0.23581695556640625, 0.6219825744628906, 0.19092941284179688, 1.730499267578125, 2.369476318359375, -1.0760498046875, 2.07373046875, 2.42498779296875, 0.8095245361328125, 1.366180419921875, -4.336090087890625, 0.03636932373046875, -0.385589599609375, -0.6312255859375, -0.235931396484375, 0.698944091796875, 1.1273880004882812, -0.5382518768310547, 0.731475830078125, -0.3279571533203125, -0.3614521026611328, 0.207916259765625, 0.30803680419921875, -2.15447998046875, 1.856109619140625, 0.300750732421875, 2.870635986328125, 0.11111831665039062, -0.21291351318359375, 0.323089599609375, -3.96466064453125, 1.10174560546875, 0.5962467193603516, 1.8967742919921875, -3.844696044921875, -0.8944091796875, 1.468963623046875, -3.927947998046875, 1.1944580078125, -1.22222900390625, 2.007171630859375, 0.10321044921875, 0.4788665771484375, 1.239349365234375, 1.6211700439453125, 1.1677703857421875, 1.36151123046875, 0.63916015625, 1.20831298828125, 4.857696533203125, 0.0867462158203125, 1.5425872802734375, -0.247100830078125, -0.3726959228515625, 0.6072731018066406, 1.33502197265625, 0.7419891357421875, -1.458465576171875, 1.26971435546875, 2.5052566528320312, 4.2454833984375, 0.521942138671875, 0.1226806640625, -3.4844970703125, 3.59228515625, 1.92169189453125, 0.175537109375, 0.5664463043212891, 0.608642578125, 0.3434562683105469, 1.678955078125, 0.876739501953125, 0.282501220703125, -1.097412109375, 0.46607208251953125, 1.451446533203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000050.npy"}
{"epoch": 0.10471204188481675, "step": 51, "batch_size": 128, "mean": 0.6313219666481018, "std": 1.921681523323059, "min": -6.95806884765625, "p10": -1.7263702392578124, "median": 0.6102523803710938, "p90": 3.165399169921875, "max": 5.62158203125, "pos_frac": 0.6171875, "sample": [1.530731201171875, 0.8131179809570312, 2.1724853515625, 0.9743270874023438, 1.5373058319091797, -2.3306808471679688, 2.002532958984375, 4.3770751953125, -0.890350341796875, -0.0940704345703125, 1.759613037109375, 1.1240997314453125, 2.0111083984375, 3.75787353515625, -0.5363979339599609, -0.151397705078125, 0.673126220703125, -2.50396728515625, 0.960357666015625, 3.0767364501953125, -6.95806884765625, 1.021728515625, 1.275360107421875, 0.7876739501953125, 2.3079833984375, 5.6153564453125, -0.5246734619140625, -1.1656494140625, 3.7617034912109375, 0.18988037109375, -1.930389404296875, 1.8433837890625, 3.35430908203125, 1.3090057373046875, 1.0398025512695312, 3.1920166015625, 1.83843994140625, 1.6586837768554688, -0.5103759765625, 0.3536109924316406, 0.0044403076171875, 2.5123138427734375, -2.23162841796875, 2.4249267578125, -0.2641754150390625, 2.886444091796875, -0.3716278076171875, 0.6685791015625, -1.19580078125, -1.89825439453125, 3.518402099609375, 1.4182586669921875, -3.240020751953125, 0.764404296875, 1.0724334716796875, 1.65380859375, -0.22943878173828125, 0.8460540771484375, -1.28533935546875, 3.5336456298828125, -0.44384765625, 0.89453125, -1.0500030517578125, 1.4377288818359375, -3.125732421875, -1.762664794921875, 0.017351150512695312, -0.2744903564453125, -0.0651702880859375, 3.06585693359375, 0.13297653198242188, 1.095245361328125, -0.0985260009765625, 3.15399169921875, -2.14385986328125, 0.0, 3.735076904296875, -1.7108154296875, 1.2591400146484375, 2.247711181640625, 0.8043670654296875, -0.5525703430175781, -0.209228515625, 0.5519256591796875, -1.37115478515625, 0.01690673828125, -1.0641021728515625, -0.38922119140625, -2.4710693359375, 3.698516845703125, 0.3581733703613281, -0.260711669921875, 0.01617431640625, -0.14342498779296875, -0.33319091796875, 1.29644775390625, -0.2578887939453125, 2.19183349609375, 1.743743896484375, -0.9110260009765625, -1.316375732421875, -0.8192291259765625, 0.2816619873046875, 5.62158203125, 4.08544921875, 5.2733154296875, 1.8305282592773438, 2.08026123046875, 1.780731201171875, 0.466339111328125, 2.397674560546875, 0.05724906921386719, -0.3789215087890625, 1.9064254760742188, 1.422607421875, 1.62518310546875, -2.4558563232421875, -2.847930908203125, -0.6561279296875, 0.28350830078125, 1.57373046875, -0.809539794921875, 0.2003021240234375, -0.6140213012695312, -0.71600341796875, 0.9752044677734375, 0.043060302734375, 1.128570556640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000051.npy"}
{"epoch": 0.1068062827225131, "step": 52, "batch_size": 128, "mean": 0.9304392337799072, "std": 2.085083246231079, "min": -3.8623046875, "p10": -1.3544486999511718, "median": 0.6189498901367188, "p90": 3.8355072021484373, "max": 8.544464111328125, "pos_frac": 0.6796875, "sample": [0.6185150146484375, 1.78070068359375, -0.94244384765625, -0.3653564453125, 0.6913299560546875, -1.60052490234375, 3.39923095703125, 0.4800262451171875, 0.2063140869140625, 6.111297607421875, 0.155914306640625, 1.96844482421875, -0.7644195556640625, -3.725738525390625, 3.173248291015625, 5.29144287109375, 0.5556182861328125, 0.999267578125, -0.448394775390625, -1.4707717895507812, 0.249237060546875, 0.7096786499023438, 0.8748016357421875, 0.38747406005859375, 0.49098968505859375, -0.1863861083984375, -0.97503662109375, 4.171295166015625, 0.852142333984375, -3.0086517333984375, -2.7376708984375, -1.6152420043945312, -0.09435272216796875, 1.5399246215820312, -0.0516357421875, 0.7901077270507812, -1.0401077270507812, -2.135589599609375, 0.4176979064941406, -0.9068603515625, 4.3292236328125, 0.7462310791015625, -0.7351531982421875, 1.2113494873046875, 2.6598663330078125, 1.773223876953125, 0.1276092529296875, 3.80120849609375, -0.43829345703125, 4.35589599609375, 0.46697998046875, 2.588714599609375, 1.448883056640625, 0.6882171630859375, -0.1254558563232422, 1.889617919921875, -0.07667922973632812, -0.5714111328125, 0.38008880615234375, 0.46954345703125, 1.3510589599609375, -0.331939697265625, 2.436553955078125, 0.07244873046875, 3.83392333984375, 1.259033203125, -3.24285888671875, 0.11938858032226562, 1.3536758422851562, -0.17216205596923828, 0.041473388671875, 1.3818206787109375, 1.6331329345703125, 2.9807281494140625, -2.43353271484375, -2.586090087890625, 1.0097427368164062, 2.84796142578125, 2.2339935302734375, 1.92681884765625, 3.839202880859375, 6.3240966796875, -0.5646209716796875, 0.24338531494140625, 3.4481201171875, -3.8623046875, -0.3334693908691406, 4.009857177734375, 5.32403564453125, 2.0225830078125, 3.10760498046875, -0.09741973876953125, 0.0382232666015625, 4.51568603515625, -1.304595947265625, 0.39525794982910156, -0.5330276489257812, 4.8607177734375, 1.1149139404296875, -0.14823150634765625, 0.9189453125, 2.6568603515625, 1.240966796875, -0.79840087890625, 4.110870361328125, 2.7307167053222656, -1.780517578125, 0.6365814208984375, 8.544464111328125, 2.745513916015625, -0.750732421875, 2.46209716796875, 0.17156982421875, -0.5774154663085938, 2.56591796875, 0.6924610137939453, -0.033843994140625, 0.4059257507324219, 0.619384765625, 1.68212890625, 0.94659423828125, 1.6469879150390625, -2.4652099609375, -0.06758880615234375, 0.81005859375, 0.2000274658203125, 2.71038818359375, 0.12110137939453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000052.npy"}
{"epoch": 0.10890052356020942, "step": 53, "batch_size": 128, "mean": 0.7023434638977051, "std": 2.066988945007324, "min": -6.64056396484375, "p10": -1.526661682128906, "median": 0.4219779968261719, "p90": 3.074186706542968, "max": 8.76483154296875, "pos_frac": 0.6171875, "sample": [0.801727294921875, -0.528350830078125, 2.0410919189453125, -2.3017501831054688, 1.6687469482421875, -0.07366943359375, 3.196197509765625, -1.5008392333984375, -1.89617919921875, 0.12470245361328125, 1.4429931640625, -0.26055908203125, -0.2457427978515625, -4.19830322265625, -0.880035400390625, 2.111083984375, 1.71954345703125, -0.812896728515625, 1.846649169921875, 1.516845703125, -0.54949951171875, 0.5338134765625, -0.68426513671875, 2.2965087890625, 3.76434326171875, 5.797119140625, 2.8632583618164062, 4.402099609375, 1.90478515625, 5.351837158203125, -0.0350341796875, 3.87744140625, 0.010951995849609375, -0.1771087646484375, 0.9045009613037109, -0.15673828125, 2.669464111328125, -5.255828857421875, -0.40538787841796875, 1.341278076171875, 0.6732177734375, -0.34375762939453125, 0.41611480712890625, -1.79449462890625, -0.16135025024414062, -1.4744873046875, -0.70330810546875, -0.4951171875, 1.963134765625, -2.00592041015625, 0.283294677734375, 2.80450439453125, 1.1494064331054688, -3.14013671875, 0.48647308349609375, -1.0956573486328125, 0.26019287109375, 0.4148902893066406, -0.6395339965820312, 2.6983489990234375, 2.832183837890625, 0.20430374145507812, -0.21010589599609375, 0.898406982421875, 1.621307373046875, -0.43597412109375, 1.651123046875, -0.2589225769042969, 0.4556121826171875, 1.158477783203125, 0.9584388732910156, 2.94842529296875, 4.548370361328125, 1.8378753662109375, 3.971435546875, 2.439910888671875, 0.1495513916015625, 1.305145263671875, 1.83758544921875, 0.4261474609375, 3.642669677734375, 0.2965850830078125, 0.066497802734375, -0.151947021484375, -1.7411880493164062, 4.1107177734375, -3.499664306640625, 1.602142333984375, 0.3195152282714844, -1.253631591796875, 1.6843185424804688, -0.0030670166015625, 2.87640380859375, -0.49080657958984375, 3.2454833984375, 8.76483154296875, 0.2249755859375, -0.77886962890625, -1.4332656860351562, -0.476409912109375, -0.29877471923828125, 0.9994583129882812, -0.524139404296875, -1.5869140625, -0.3504180908203125, -1.672332763671875, -1.62347412109375, 2.865631103515625, 0.41780853271484375, -6.64056396484375, 0.1309967041015625, -0.11148834228515625, 2.5116119384765625, 3.0218963623046875, 0.48272705078125, -0.020751953125, 2.336669921875, 0.0, 1.585845947265625, 1.157318115234375, 1.670684814453125, 0.664794921875, 0.959564208984375, 0.289398193359375, 1.12646484375, 1.1798248291015625, 0.73077392578125, 3.732147216796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000053.npy"}
{"epoch": 0.11099476439790576, "step": 54, "batch_size": 128, "mean": 0.7727015018463135, "std": 2.168445110321045, "min": -5.765838623046875, "p10": -1.8357826232910155, "median": 0.5033683776855469, "p90": 3.7634246826171873, "max": 7.30413818359375, "pos_frac": 0.65625, "sample": [3.435791015625, 3.384063720703125, 2.244110107421875, 2.453369140625, 1.23974609375, 1.0129890441894531, -0.52264404296875, 3.78314208984375, 0.0420684814453125, -1.993560791015625, 0.21209716796875, -1.000335693359375, 0.10775375366210938, -2.076019287109375, 5.278564453125, -0.03354454040527344, 4.85797119140625, 1.0266952514648438, 0.251556396484375, 0.7564926147460938, 1.794830322265625, 4.7403564453125, 0.889373779296875, 1.6162109375, -1.3510284423828125, 0.5131149291992188, 2.97967529296875, -0.139007568359375, -0.12408447265625, -0.08041000366210938, 0.33916473388671875, 0.076812744140625, -2.662750244140625, 1.2290725708007812, 0.97650146484375, 3.176727294921875, 7.30413818359375, -4.66802978515625, 2.75732421875, -0.0476226806640625, 0.3336029052734375, 0.3229713439941406, 5.04925537109375, -0.156005859375, 4.8359375, 0.6367645263671875, -2.6944580078125, 0.0761871337890625, -1.2364959716796875, -1.81610107421875, 1.94049072265625, 5.929412841796875, 0.9022903442382812, 0.431427001953125, -0.10384941101074219, 0.02689361572265625, -0.96307373046875, 1.03973388671875, 1.7447662353515625, 0.9409942626953125, 3.0912303924560547, 1.55718994140625, -1.945281982421875, -1.296112060546875, 1.4355316162109375, 1.5995559692382812, -0.9644012451171875, 0.9006195068359375, 2.1242828369140625, 3.8743896484375, -0.1898651123046875, 0.5351943969726562, 1.9129638671875, 2.244903564453125, 0.541839599609375, 0.48964691162109375, 2.109100341796875, -0.3594970703125, 1.4521560668945312, -0.605438232421875, -0.5747909545898438, -1.8546829223632812, 0.493621826171875, 0.5168533325195312, 1.19146728515625, -0.631317138671875, 1.03045654296875, -0.9314727783203125, 2.8564529418945312, 0.6183013916015625, 0.213592529296875, 0.94775390625, -2.109954833984375, 2.483184814453125, 0.9691009521484375, 0.41597747802734375, -2.789886474609375, 5.592041015625, 5.1732177734375, 2.33319091796875, 2.468841552734375, 4.9532470703125, 3.3820953369140625, 0.943817138671875, 4.3734283447265625, -0.15183258056640625, -2.36279296875, 0.1714324951171875, -5.765838623046875, 1.782684326171875, -1.11712646484375, -2.9618682861328125, -0.20135498046875, -0.920166015625, 3.754974365234375, -0.1468658447265625, -1.8276824951171875, 0.3248291015625, 0.308197021484375, -0.6356353759765625, 0.066650390625, 0.31341552734375, 1.1289691925048828, -3.86541748046875, -0.787261962890625, -0.2102508544921875, 1.5499992370605469, -1.135223388671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000054.npy"}
{"epoch": 0.1130890052356021, "step": 55, "batch_size": 128, "mean": 0.6250678896903992, "std": 1.8671127557754517, "min": -6.28509521484375, "p10": -1.4466033935546876, "median": 0.6445255279541016, "p90": 3.012632751464843, "max": 7.405426025390625, "pos_frac": 0.625, "sample": [-1.44549560546875, 1.99920654296875, 0.0440673828125, 1.330291748046875, -0.8643646240234375, -0.946502685546875, 0.7505054473876953, 0.98956298828125, 1.89495849609375, -0.33135986328125, 1.80224609375, 2.3960952758789062, 3.150665283203125, 0.184051513671875, 0.992645263671875, 0.7862834930419922, 1.0430068969726562, -0.39312744140625, 0.7606277465820312, 1.2905960083007812, 0.2053070068359375, 0.807525634765625, 0.958892822265625, 1.9344482421875, 1.1051025390625, -0.43882179260253906, 0.8946990966796875, 1.9180908203125, 0.8749237060546875, -0.11126708984375, -1.071075439453125, -0.7374114990234375, -1.68365478515625, 2.450836181640625, 0.5955810546875, 1.5331344604492188, -0.3263397216796875, 0.93145751953125, -0.7898178100585938, 4.4466552734375, -2.6275634765625, 0.00618743896484375, -0.9019393920898438, 0.9896697998046875, -0.2301025390625, 0.84271240234375, 1.0024871826171875, -0.1081695556640625, 0.6737060546875, 4.3311767578125, 0.7447128295898438, -1.0159149169921875, -1.802001953125, 1.0960311889648438, 0.594757080078125, -1.5610275268554688, -2.128143310546875, 4.60418701171875, -1.5258636474609375, 1.804351806640625, -1.07257080078125, 0.804840087890625, 4.403167724609375, -0.6183204650878906, 1.5640869140625, -0.97119140625, -1.98046875, 3.514739990234375, 1.2932052612304688, -1.0903434753417969, 0.4825439453125, 0.139251708984375, -1.9211578369140625, -1.27252197265625, -0.8125267028808594, -0.2521533966064453, -0.096710205078125, 2.7405853271484375, -0.3995513916015625, 1.1370315551757812, -0.9755935668945312, 4.30010986328125, -1.65423583984375, 0.6614723205566406, -0.462554931640625, 0.18688201904296875, -0.9012508392333984, 2.9534759521484375, 3.715789794921875, 0.9550933837890625, -0.6185684204101562, 1.6629562377929688, 1.793426513671875, 0.316986083984375, 7.405426025390625, 0.3122272491455078, 2.048248291015625, -1.084869384765625, 0.6275787353515625, 3.5226287841796875, -0.522216796875, 1.6263961791992188, 0.8801612854003906, 1.73358154296875, 0.6680450439453125, -2.80474853515625, 5.90655517578125, 0.5683746337890625, 2.0100326538085938, -0.8746337890625, -6.28509521484375, -0.46196746826171875, 2.30389404296875, 0.253875732421875, -0.9132843017578125, -1.0622100830078125, 2.3584442138671875, -1.5552215576171875, 1.2801780700683594, -1.449188232421875, 1.2625579833984375, 0.1835174560546875, -0.6807861328125, 2.314056396484375, 0.208251953125, 4.506256103515625, 3.5276336669921875, 0.9475860595703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000055.npy"}
{"epoch": 0.11518324607329843, "step": 56, "batch_size": 128, "mean": 0.9080754518508911, "std": 2.301595449447632, "min": -5.58355712890625, "p10": -1.2497833251953123, "median": 0.6176986694335938, "p90": 4.101873779296874, "max": 7.396087646484375, "pos_frac": 0.671875, "sample": [0.02996826171875, -3.458709716796875, 2.828948974609375, 5.12310791015625, 0.4257354736328125, 1.960723876953125, 0.692779541015625, 1.1815185546875, 0.0248870849609375, 7.317596435546875, 0.71337890625, 2.4202880859375, -0.43231201171875, -0.5895233154296875, 1.09051513671875, 0.324615478515625, -0.17637252807617188, -0.5585250854492188, 2.81951904296875, -1.3179931640625, 0.05242919921875, -0.042999267578125, 2.2371826171875, -0.585693359375, -0.87359619140625, 5.2188720703125, 0.864227294921875, 4.4139404296875, 2.4026031494140625, 0.1609649658203125, 2.03411865234375, 3.012359619140625, 0.5682144165039062, -1.0138397216796875, 7.396087646484375, 1.9043426513671875, -3.7276763916015625, -0.08795928955078125, 0.726898193359375, -0.358489990234375, 5.4595184326171875, 1.209442138671875, 0.10054397583007812, -0.13047409057617188, 0.3965797424316406, 6.822021484375, 1.7254486083984375, -0.0201416015625, -1.220550537109375, -0.25662994384765625, -0.99041748046875, -1.0163192749023438, 1.0289459228515625, -4.6536865234375, -0.2206268310546875, -5.21343994140625, 0.17399024963378906, -0.5261173248291016, 1.713623046875, 0.17154693603515625, 1.8274002075195312, 4.3670654296875, 3.915771484375, 2.7250823974609375, -0.905181884765625, -1.51904296875, 1.904541015625, 2.1638946533203125, 5.124755859375, 0.9196395874023438, 0.273101806640625, -0.19962120056152344, -4.146240234375, 0.5078220367431641, 0.06668472290039062, -0.3166961669921875, -1.492340087890625, 4.61016845703125, 0.2037200927734375, 0.7283172607421875, -5.58355712890625, 1.120849609375, 2.707550048828125, -0.0399169921875, 0.64178466796875, 5.729736328125, 0.26861572265625, 3.0645751953125, -1.85626220703125, 1.93182373046875, -0.669097900390625, 5.8995361328125, 1.096893310546875, 1.5019760131835938, -2.115478515625, 3.98822021484375, 1.3553009033203125, -0.2962188720703125, 0.01287078857421875, 0.8281631469726562, 1.722869873046875, -0.9227752685546875, 1.0297088623046875, -2.3683624267578125, -1.7330322265625, 5.13916015625, -0.266693115234375, 1.099273681640625, 0.5936126708984375, 2.7631683349609375, 2.326690673828125, 0.75384521484375, 1.860443115234375, -1.135528564453125, 0.058734893798828125, 1.8141937255859375, -0.1582489013671875, 0.47747802734375, 1.32135009765625, 0.19403076171875, 1.735626220703125, -1.06890869140625, 2.037811279296875, 0.5225830078125, 1.45556640625, 3.403411865234375, 3.203033447265625, 0.7530136108398438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000056.npy"}
{"epoch": 0.11727748691099477, "step": 57, "batch_size": 128, "mean": 1.359413743019104, "std": 2.356470823287964, "min": -3.60491943359375, "p10": -1.2134201049804687, "median": 0.8377609252929688, "p90": 4.416717529296874, "max": 12.86895751953125, "pos_frac": 0.7265625, "sample": [-1.401763916015625, 2.9163818359375, -2.429443359375, 4.133819580078125, 0.17856597900390625, -1.36083984375, 4.976806640625, -0.04729461669921875, -2.21929931640625, -0.32562255859375, 0.4227294921875, -0.578643798828125, 3.6808013916015625, -0.10572052001953125, 3.6664657592773438, 0.5523433685302734, 2.9142303466796875, 1.646209716796875, 0.130859375, 2.015289306640625, 0.6278610229492188, 2.651458740234375, 0.26239013671875, -0.347564697265625, 0.09478759765625, 1.6627197265625, -3.60491943359375, 0.69989013671875, -0.15495681762695312, -1.2982254028320312, 2.079742431640625, 1.942108154296875, 0.886810302734375, 0.6332855224609375, 6.8841552734375, 12.86895751953125, 0.787841796875, 1.076080322265625, -1.4991188049316406, 1.46002197265625, 3.95562744140625, 0.7940330505371094, -1.14837646484375, 4.9267425537109375, 0.8453826904296875, 0.8907089233398438, 0.29482269287109375, -0.813262939453125, -2.022705078125, 2.286224365234375, 0.504364013671875, 0.9569320678710938, 0.19854736328125, 7.07952880859375, 0.42828369140625, 2.0433349609375, 1.3336410522460938, 1.1600799560546875, 2.620147705078125, -1.2537841796875, 0.863006591796875, -0.063629150390625, 0.5409870147705078, 2.66070556640625, 1.197906494140625, 4.56817626953125, 6.119140625, 3.584716796875, -1.1602935791015625, 1.365997314453125, 4.320831298828125, 0.041473388671875, -1.1961212158203125, -0.3392066955566406, 2.839324951171875, 2.13775634765625, 0.83013916015625, 3.562530517578125, 1.4257049560546875, 2.12353515625, 1.0233154296875, 0.70037841796875, 5.984161376953125, 4.71728515625, 0.45843505859375, -0.6536865234375, -1.5297012329101562, 3.18792724609375, -2.09906005859375, -0.917327880859375, 2.175525665283203, 2.92486572265625, 0.0598297119140625, 0.2895545959472656, -0.10235595703125, -0.5111427307128906, 2.575927734375, 0.16805267333984375, 0.35211181640625, 0.056915283203125, -3.49139404296875, -0.2920684814453125, 6.1033935546875, 3.7722015380859375, 0.09395217895507812, 5.61700439453125, 0.3621788024902344, 4.153167724609375, -0.2396087646484375, 0.6897659301757812, -0.161468505859375, 1.252227783203125, 2.79974365234375, -0.229522705078125, 1.9313812255859375, 1.86383056640625, 2.659454345703125, -0.022525787353515625, 0.9730281829833984, 0.13593482971191406, 3.87127685546875, -1.3342742919921875, 5.2923583984375, 1.48699951171875, 2.0690994262695312, -0.13360595703125, 4.58746337890625, 4.351806640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000057.npy"}
{"epoch": 0.1193717277486911, "step": 58, "batch_size": 128, "mean": 0.9785891175270081, "std": 2.129312515258789, "min": -11.2528076171875, "p10": -1.1729644775390624, "median": 0.9850883483886719, "p90": 3.348480224609375, "max": 6.371490478515625, "pos_frac": 0.75, "sample": [1.679473876953125, 3.81500244140625, 1.032440185546875, 2.509735107421875, 1.8511428833007812, -0.08104324340820312, 2.1046295166015625, -1.29364013671875, 2.36102294921875, 4.8531494140625, 2.128387451171875, 3.33026123046875, 1.2834701538085938, 5.77264404296875, 2.452789306640625, -0.7501678466796875, 3.6210250854492188, 2.58197021484375, 1.9735260009765625, -2.07098388671875, 6.371490478515625, 2.781524658203125, 1.997314453125, -1.4645233154296875, 0.5181884765625, 0.44134521484375, 1.464935302734375, 1.5775146484375, 1.1277923583984375, -1.02734375, 0.3333282470703125, 0.5968780517578125, 0.8179931640625, 0.12831497192382812, 3.3909912109375, -0.03826141357421875, 1.634002685546875, 1.5688705444335938, -0.38733482360839844, 0.1064453125, -3.8760986328125, 1.406524658203125, 0.87384033203125, 0.01007080078125, 3.58837890625, -1.1115798950195312, 0.3798103332519531, 5.6741943359375, 0.012722015380859375, -11.2528076171875, 0.735748291015625, 4.306671142578125, -1.378936767578125, -1.78509521484375, 0.1806640625, 0.4927520751953125, -0.7971420288085938, 1.2165641784667969, 0.661590576171875, 2.84228515625, 1.0012664794921875, 0.38663482666015625, 2.593017578125, 0.126800537109375, 3.198974609375, 1.15875244140625, 1.291717529296875, 1.4796142578125, -0.315948486328125, 0.1339111328125, -1.75445556640625, -1.9673233032226562, -1.121246337890625, 2.848907470703125, 1.2364501953125, -0.06744384765625, 6.1002197265625, 0.8161354064941406, -0.8941650390625, 0.1207733154296875, 2.0286865234375, 0.35980224609375, 4.823974609375, -2.745452880859375, 0.749176025390625, 0.9689102172851562, -0.06684112548828125, 0.6973190307617188, 0.12633514404296875, 1.99603271484375, -0.41161346435546875, 2.6526336669921875, 2.767578125, 1.2031097412109375, 1.51123046875, -1.038543701171875, 3.8533477783203125, -0.1959228515625, 0.482421875, 2.3211517333984375, 1.305389404296875, -0.1407794952392578, 0.5651931762695312, -1.751068115234375, 2.2539520263671875, 1.0737075805664062, 0.43628692626953125, 1.8648910522460938, 3.298004150390625, 2.09857177734375, -3.1346435546875, 2.578125, 0.702545166015625, 3.645843505859375, 1.0251693725585938, 0.7483749389648438, -0.0951690673828125, 0.38128662109375, 1.9107284545898438, 1.5285797119140625, 2.2512664794921875, 1.0065021514892578, 0.29376220703125, 1.840118408203125, -0.6478271484375, -2.660614013671875, 1.408355712890625, -0.257537841796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000058.npy"}
{"epoch": 0.12146596858638743, "step": 59, "batch_size": 128, "mean": 0.8722929954528809, "std": 2.547051191329956, "min": -8.394866943359375, "p10": -1.6437553405761718, "median": 0.5761680603027344, "p90": 3.9854797363281245, "max": 8.603546142578125, "pos_frac": 0.6484375, "sample": [-0.482147216796875, 2.2122039794921875, -0.77490234375, -8.394866943359375, 1.303619384765625, 2.448272705078125, 0.452911376953125, 2.240966796875, 1.0048828125, -0.427734375, 0.398529052734375, 0.3463287353515625, 2.2509613037109375, 1.9295196533203125, -3.4295806884765625, 0.44496917724609375, 0.49748992919921875, 0.19554901123046875, -0.872344970703125, 1.293365478515625, -0.44476318359375, 1.0021820068359375, 0.26915740966796875, -0.7266845703125, 1.2152862548828125, 1.2087554931640625, 0.956939697265625, 4.163726806640625, -3.335296630859375, 0.387451171875, -1.358551025390625, 1.2565727233886719, 0.7259521484375, 2.4819183349609375, 0.0, 5.624267578125, 0.015073776245117188, 2.2705841064453125, -0.28727149963378906, -1.968292236328125, 0.1048583984375, 3.55023193359375, 5.70819091796875, 6.7689208984375, 0.856842041015625, -3.883636474609375, 2.2700119018554688, 7.95928955078125, 3.5367431640625, 7.336669921875, 1.849334716796875, 0.955291748046875, 4.2171630859375, 4.73980712890625, 0.46920013427734375, 1.8301620483398438, -0.33467864990234375, 1.370513916015625, -4.188201904296875, -0.968994140625, 0.65484619140625, 2.030975341796875, -0.160430908203125, 0.04632568359375, 1.2118606567382812, 4.24456787109375, 5.1009521484375, -1.0731201171875, 3.05517578125, 1.8735275268554688, -0.5389556884765625, 1.884979248046875, -0.248016357421875, 7.99395751953125, -1.7127609252929688, -5.178466796875, 0.8218860626220703, -3.2684326171875, -0.166168212890625, 0.23168182373046875, 3.909088134765625, -0.486846923828125, 2.317779541015625, 1.816619873046875, 1.4252166748046875, -0.3000030517578125, 1.115814208984375, -1.6141815185546875, -1.7179412841796875, -2.28564453125, 1.89190673828125, 2.90032958984375, 0.0237579345703125, -0.17730140686035156, -0.4831695556640625, 3.7786102294921875, 1.31536865234375, -0.56707763671875, -1.0198822021484375, 0.0902099609375, -0.5884857177734375, 0.4254150390625, 0.06134796142578125, 2.0213165283203125, -0.4328575134277344, 1.894775390625, -0.569427490234375, -4.102630615234375, -0.9383163452148438, 0.7919921875, 2.601226806640625, 1.749420166015625, 5.07989501953125, 0.260711669921875, -0.82086181640625, -0.39478302001953125, 8.603546142578125, 1.8025131225585938, 2.077239990234375, -0.721405029296875, -3.601409912109375, 1.286376953125, 2.76043701171875, 0.72149658203125, -0.027044296264648438, 0.20642852783203125, 2.931243896484375, -0.4044189453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000059.npy"}
{"epoch": 0.12356020942408377, "step": 60, "batch_size": 128, "mean": 1.1233301162719727, "std": 2.387179374694824, "min": -6.98980712890625, "p10": -1.3310531616210937, "median": 1.177699089050293, "p90": 3.643627929687499, "max": 9.488494873046875, "pos_frac": 0.6875, "sample": [-4.951171875, 2.4412841796875, 4.736412048339844, -0.64263916015625, 2.207611083984375, 3.875152587890625, -0.3891010284423828, 3.168548583984375, 1.519622802734375, 2.35009765625, 0.773101806640625, -0.3344154357910156, -0.2851448059082031, 1.5624542236328125, 0.3046112060546875, 1.385986328125, 0.77581787109375, -1.1402587890625, 1.7717437744140625, 4.47314453125, 0.9049758911132812, -3.70953369140625, 2.655548095703125, -1.2799072265625, -1.4782905578613281, 0.34284210205078125, 1.322265625, 0.0, -0.44302940368652344, 3.9224853515625, -0.628997802734375, 2.92449951171875, 1.26861572265625, 2.5413665771484375, -0.3473663330078125, -0.5078125, 3.3739013671875, -3.488128662109375, 2.8755645751953125, -0.18238258361816406, -0.703857421875, 2.2228012084960938, 0.53021240234375, 3.544403076171875, 0.7689132690429688, -0.0977935791015625, -1.2583351135253906, -0.094207763671875, -1.135040283203125, 0.5925865173339844, 0.029415130615234375, 0.4896240234375, 1.550872802734375, 5.398895263671875, -0.031307220458984375, -1.73577880859375, 3.1353912353515625, -2.478759765625, 1.1579360961914062, 3.9068145751953125, 1.219757080078125, 1.99407958984375, 3.07891845703125, 0.6244735717773438, -6.98980712890625, 2.466766357421875, 1.54705810546875, 0.39191436767578125, -0.8032379150390625, -1.261566162109375, -1.4503936767578125, 2.803619384765625, 3.0699920654296875, 4.04144287109375, 2.213348388671875, 0.2701568603515625, 0.5051651000976562, 1.851837158203125, 1.209991455078125, 6.1954345703125, 3.3084259033203125, 1.2305374145507812, -1.1215629577636719, 0.7786712646484375, 6.553741455078125, 0.94482421875, 1.6266937255859375, -3.87353515625, 1.1974620819091797, 1.7331085205078125, 2.3998870849609375, 0.5289726257324219, -3.0927734375, -0.0049724578857421875, 2.7982177734375, 3.3866424560546875, -0.42324066162109375, 5.465087890625, -1.1720657348632812, -0.53411865234375, 0.2353515625, 2.89581298828125, 1.3247528076171875, -2.5904541015625, 4.86639404296875, 2.667388916015625, 0.469940185546875, 2.5239028930664062, 9.488494873046875, 0.0, 7.17523193359375, -2.38818359375, 1.7410888671875, 0.627777099609375, -0.17940902709960938, 0.7280731201171875, 2.500946044921875, 0.7809028625488281, 2.327056884765625, -3.7725830078125, 1.4970550537109375, 2.802703857421875, 2.3263511657714844, 2.82855224609375, 2.288360595703125, 0.0120391845703125, 3.4763641357421875, 2.96112060546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000060.npy"}
{"epoch": 0.1256544502617801, "step": 61, "batch_size": 128, "mean": 1.1290453672409058, "std": 2.4950859546661377, "min": -5.784942626953125, "p10": -1.3102073669433594, "median": 0.90203857421875, "p90": 4.489347839355468, "max": 8.4193115234375, "pos_frac": 0.703125, "sample": [-1.317230224609375, -0.862823486328125, 1.1340618133544922, 0.034149169921875, 6.3321533203125, 4.99761962890625, -0.136016845703125, 0.9477081298828125, -0.11115646362304688, 1.0387039184570312, 1.5556793212890625, -5.7352294921875, -1.18890380859375, 0.7718505859375, 1.024383544921875, 0.5849704742431641, -4.44854736328125, -0.8111114501953125, -2.0703125, -0.8177871704101562, 0.5391693115234375, 1.3356475830078125, 6.920562744140625, 4.145965576171875, 2.26397705078125, 0.205810546875, 3.08929443359375, -4.57763671875e-05, 3.71685791015625, 2.4369354248046875, 0.4347858428955078, 0.6681671142578125, 0.2701244354248047, -0.111541748046875, 3.346588134765625, 0.66119384765625, -0.11980438232421875, 4.4677276611328125, 1.632232666015625, 0.6870841979980469, -0.01373291015625, 1.47454833984375, 3.184326171875, -0.8893814086914062, 0.931121826171875, 0.3224029541015625, 4.731597900390625, 3.369598388671875, 1.828338623046875, 2.5099029541015625, 0.9212646484375, 2.39312744140625, 6.37408447265625, 4.80999755859375, 3.819244384765625, 2.85107421875, 1.7488555908203125, 1.1950340270996094, 5.9536895751953125, 0.7066802978515625, 0.6334991455078125, 2.4639930725097656, 1.13043212890625, -0.974395751953125, 0.2079620361328125, 4.539794921875, 2.552947998046875, 6.679290771484375, 2.64178466796875, 1.5774993896484375, 1.1052627563476562, 0.7066650390625, 1.800506591796875, -0.843170166015625, -0.7405624389648438, 0.140594482421875, 0.12430572509765625, 0.4215087890625, 1.38946533203125, -1.69342041015625, 6.267578125, 1.50238037109375, 2.82025146484375, 3.883026123046875, 0.5923309326171875, 2.6085968017578125, 0.15027618408203125, -1.3071975708007812, -0.6109809875488281, -0.196746826171875, -0.42296409606933594, -3.849700927734375, -1.71136474609375, 1.262054443359375, -0.199127197265625, 1.55499267578125, 0.9544010162353516, 5.012451171875, -1.4354248046875, 6.51190185546875, 0.8828125, 8.4193115234375, 2.02349853515625, 0.0, 0.3114013671875, 1.053466796875, 3.6057891845703125, 0.24517822265625, 1.2644996643066406, -5.784942626953125, 3.406158447265625, -2.468109130859375, -1.6566925048828125, -0.7814788818359375, -5.6737060546875, 1.78094482421875, -3.54901123046875, -0.455047607421875, 0.361236572265625, 0.15478515625, -0.38543701171875, 0.45117950439453125, -1.0516586303710938, 1.4768829345703125, 2.3207931518554688, 4.145965576171875, 2.6899566650390625, -1.2553329467773438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000061.npy"}
{"epoch": 0.12774869109947645, "step": 62, "batch_size": 128, "mean": 1.2035984992980957, "std": 2.5705418586730957, "min": -5.14129638671875, "p10": -1.7841064453125, "median": 0.8301277160644531, "p90": 4.406881713867187, "max": 9.761566162109375, "pos_frac": 0.6875, "sample": [4.157806396484375, 0.5892715454101562, 8.146240234375, 2.0819473266601562, 2.911163330078125, -0.061737060546875, 3.704742431640625, 1.0572509765625, -1.040771484375, 0.1634674072265625, -2.39361572265625, -1.1565170288085938, -0.60162353515625, 4.1917724609375, 2.2750244140625, 0.5084609985351562, 0.934600830078125, 2.5005569458007812, -0.351165771484375, -1.77099609375, 1.6520843505859375, 1.1477394104003906, 3.3107452392578125, 0.7612380981445312, -2.3636016845703125, 6.3167724609375, -5.14129638671875, 1.25921630859375, 0.5741310119628906, 0.70361328125, 0.70013427734375, -0.1055908203125, 1.90472412109375, 1.4936981201171875, 0.09205436706542969, 0.1761627197265625, 2.439697265625, -1.0681304931640625, 1.7147598266601562, -0.7142181396484375, 1.789520263671875, 2.78216552734375, 3.8311843872070312, 0.0844879150390625, -0.5132598876953125, 0.91748046875, 0.19839096069335938, 3.18572998046875, 3.4063720703125, 9.761566162109375, -4.6968994140625, 3.391021728515625, -1.012359619140625, 3.3836669921875, 3.35400390625, -3.3911285400390625, 5.4644775390625, 0.0, 4.157623291015625, -0.20114898681640625, -2.135284423828125, -1.6440696716308594, 5.148773193359375, -0.01697540283203125, 1.8619308471679688, -0.51397705078125, 0.43683624267578125, -1.3006591796875, 1.495941162109375, 0.96026611328125, 4.294036865234375, 0.899017333984375, 0.29936981201171875, -2.1269073486328125, 2.4462966918945312, 6.48309326171875, 0.9089508056640625, -1.814697265625, 0.32607269287109375, 1.7841854095458984, 0.03969573974609375, -0.1141204833984375, 1.870086669921875, 1.9150390625, -1.18670654296875, 3.111114501953125, 7.196258544921875, 2.25128173828125, 1.01568603515625, 4.50213623046875, 0.4843635559082031, -1.268341064453125, 7.0933837890625, 2.4463348388671875, 6.10107421875, 3.2028961181640625, -2.4163818359375, 5.590576171875, -2.2064208984375, -1.449493408203125, 4.53094482421875, -0.972442626953125, 0.9168014526367188, 0.351776123046875, 5.388275146484375, 2.4676971435546875, 2.272674560546875, 1.4906082153320312, 0.20496368408203125, 4.366058349609375, 3.084564208984375, 0.01532745361328125, 0.422760009765625, -3.51348876953125, 0.408538818359375, -0.7815322875976562, -2.433135986328125, -0.8108978271484375, 1.8712615966796875, 0.08320426940917969, 0.001800537109375, -0.444976806640625, 3.9189605712890625, 1.4613494873046875, -0.2890625, 0.759307861328125, -2.72698974609375, -0.5471038818359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000062.npy"}
{"epoch": 0.12984293193717278, "step": 63, "batch_size": 128, "mean": 1.310964822769165, "std": 3.013709306716919, "min": -6.24609375, "p10": -2.4903305053710936, "median": 1.1356830596923828, "p90": 4.673440551757812, "max": 12.55145263671875, "pos_frac": 0.671875, "sample": [2.563873291015625, 3.6121826171875, -0.3555431365966797, -6.24609375, -4.84228515625, -0.3690452575683594, 2.705963134765625, -0.1645660400390625, -0.2563629150390625, 2.617889404296875, -3.665374755859375, -2.61737060546875, -2.9342803955078125, -1.8164291381835938, 2.653900146484375, -0.6095809936523438, 3.197601318359375, 2.47222900390625, 7.594268798828125, 7.4296875, -3.429901123046875, -3.596832275390625, 12.55145263671875, -2.240142822265625, 2.21038818359375, 1.92755126953125, 2.08380126953125, -3.2089385986328125, 0.2407073974609375, 4.6982421875, 0.11142730712890625, 3.0487060546875, -0.8070220947265625, 1.0483360290527344, 1.2018470764160156, 8.80194091796875, -2.087860107421875, 0.8295745849609375, -1.5505218505859375, 4.60101318359375, -0.458160400390625, 0.84130859375, 4.7469482421875, -2.53594970703125, 0.35758209228515625, 3.29547119140625, 7.09600830078125, -6.181396484375, -3.294921875, 2.9178466796875, 0.03398704528808594, 4.5811767578125, 2.671051025390625, 0.0, -0.19283485412597656, 2.284027099609375, 4.113372802734375, -1.3915252685546875, -0.18333053588867188, 0.7416267395019531, 1.0138397216796875, 0.3289794921875, 1.06951904296875, 5.731201171875, -1.4970703125, 0.6584091186523438, 2.33209228515625, 3.541748046875, -2.3174514770507812, 4.745513916015625, 6.584716796875, 1.010009765625, 2.63214111328125, -1.6846923828125, 4.469482421875, -2.64459228515625, 3.3029098510742188, 1.3756446838378906, 1.523529052734375, -0.149566650390625, -1.543212890625, 8.35479736328125, 2.1996917724609375, -1.4493865966796875, 2.9978408813476562, 0.4251556396484375, 2.2855300903320312, 0.11428070068359375, 0.919342041015625, 2.5387420654296875, 2.566375732421875, 0.3404541015625, -2.4707794189453125, 4.662811279296875, 0.8590755462646484, 4.013031005859375, -0.0819091796875, 2.9168701171875, 3.0613250732421875, 3.64825439453125, 2.43670654296875, 1.2258453369140625, 3.985565185546875, -1.0789642333984375, 1.71380615234375, 0.540618896484375, 0.355316162109375, -0.18241119384765625, 4.0631103515625, 1.243194580078125, 0.685791015625, -0.1091461181640625, 2.783660888671875, 2.5018310546875, 1.8725814819335938, 6.3739013671875, 0.5180206298828125, 1.7732391357421875, 1.393798828125, 1.8350448608398438, -4.5972900390625, -0.0692291259765625, 2.64892578125, 2.435882568359375, -1.319488525390625, -0.5619087219238281, 4.578887939453125, 5.52484130859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000063.npy"}
{"epoch": 0.1319371727748691, "step": 64, "batch_size": 128, "mean": 1.5472369194030762, "std": 3.2232749462127686, "min": -7.6938934326171875, "p10": -2.0923858642578126, "median": 1.1270217895507812, "p90": 6.041735839843749, "max": 10.62713623046875, "pos_frac": 0.703125, "sample": [4.4097137451171875, -1.8875732421875, -4.242095947265625, 7.55255126953125, -2.143463134765625, 3.7132568359375, 1.9882965087890625, 1.8734130859375, 7.1209716796875, 0.61346435546875, 2.6147003173828125, 1.58441162109375, 7.25958251953125, 3.44927978515625, 0.9478912353515625, -3.3778533935546875, 8.9117431640625, 2.351837158203125, 0.9853668212890625, -4.77264404296875, 0.9887542724609375, 1.0880813598632812, -1.2661285400390625, -2.159637451171875, 0.931732177734375, -7.6938934326171875, 2.3211326599121094, 1.1216583251953125, 1.830810546875, 4.627899169921875, 0.986785888671875, 0.95147705078125, 3.93798828125, 0.4812507629394531, 1.1473388671875, -0.13480377197265625, 0.841094970703125, 2.4424972534179688, 1.7940902709960938, -0.045989990234375, 2.5263519287109375, -0.2919921875, -0.8712615966796875, 2.129058837890625, -0.0009765625, 10.62713623046875, 6.445068359375, -0.9396934509277344, 7.2451171875, 2.261138916015625, 3.07501220703125, -2.3401031494140625, 0.3269767761230469, 0.417694091796875, 0.7497882843017578, -0.7082595825195312, -2.8523101806640625, 2.759613037109375, 3.3295440673828125, 2.154022216796875, 2.7230224609375, 3.439228057861328, -1.531829833984375, 4.618621826171875, 1.415313720703125, -0.1096954345703125, 0.159393310546875, -4.935211181640625, 0.318084716796875, 2.577392578125, 1.03173828125, 2.3188934326171875, 9.04901123046875, 1.537872314453125, 0.21457672119140625, 2.0634765625, 2.4058990478515625, 0.60406494140625, -0.5708236694335938, -0.2458648681640625, -1.7796745300292969, 5.766937255859375, -0.2852210998535156, 0.12047576904296875, 4.30914306640625, 6.276519775390625, 0.603424072265625, 1.6975173950195312, 1.2793235778808594, -2.07049560546875, 3.542694091796875, 10.585601806640625, 5.94586181640625, 6.7813720703125, 1.0050201416015625, -0.831085205078125, -6.18804931640625, 0.9786224365234375, -1.872161865234375, 1.1752777099609375, 3.010711669921875, 5.345550537109375, 3.3739852905273438, 0.0, 0.20373916625976562, 0.26587486267089844, -0.77117919921875, 1.5377044677734375, 4.7303314208984375, -2.2489013671875, 5.17315673828125, 1.2138557434082031, -0.6412353515625, 6.6073150634765625, 1.13238525390625, -0.6584129333496094, -0.37207794189453125, 5.461456298828125, -2.8291015625, 1.1088333129882812, 0.0, 3.998443603515625, 6.26544189453125, 2.4935073852539062, 4.556068420410156, 3.542633056640625, -6.568115234375, -1.19683837890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000064.npy"}
{"epoch": 0.13403141361256546, "step": 65, "batch_size": 128, "mean": 1.0765533447265625, "std": 2.7456367015838623, "min": -7.891021728515625, "p10": -1.8444858551025387, "median": 0.8850250244140625, "p90": 4.630230712890625, "max": 7.96099853515625, "pos_frac": 0.7109375, "sample": [0.2887077331542969, -1.5406494140625, 0.46685791015625, 0.21247291564941406, 6.03570556640625, 1.1496505737304688, 1.9221878051757812, 3.8855133056640625, 2.7087249755859375, -5.049072265625, -2.826324462890625, -6.42327880859375, 1.12017822265625, 4.26885986328125, -1.1694259643554688, 3.25250244140625, 3.67840576171875, -2.3262939453125, 0.8107109069824219, 0.20745849609375, -3.697662353515625, 0.16718673706054688, 3.4594879150390625, 5.0179443359375, 2.2513256072998047, 1.4536285400390625, 4.53546142578125, 1.8105392456054688, -3.376373291015625, -0.0149078369140625, 1.118865966796875, -0.9743728637695312, -0.41104888916015625, 2.6936187744140625, 7.96099853515625, -0.90020751953125, 0.07580757141113281, 5.15948486328125, -0.16555023193359375, 0.159515380859375, 0.949005126953125, -3.017578125, 3.915618896484375, 1.6330490112304688, 6.26715087890625, -1.2804412841796875, 4.868408203125, 2.16253662109375, 5.14642333984375, 0.8526458740234375, 2.8058929443359375, 2.56842041015625, -0.52203369140625, 1.927764892578125, 2.5415878295898438, -0.487213134765625, 4.7357177734375, 1.2839889526367188, 0.0026092529296875, 2.9720458984375, -1.4429931640625, 0.32395172119140625, 1.19891357421875, -1.133880615234375, 0.98187255859375, 0.9099884033203125, 1.5478172302246094, -1.49371337890625, -2.54779052734375, -0.777130126953125, 2.47137451171875, 4.58502197265625, 1.180755615234375, 0.6286773681640625, 2.8085479736328125, 0.4862556457519531, -1.1322784423828125, -0.08181571960449219, 0.65447998046875, -1.0012588500976562, -5.57611083984375, -0.34661865234375, 0.0, 0.776458740234375, 3.5483245849609375, 1.0290145874023438, -1.0471038818359375, 0.8528900146484375, 4.2406768798828125, 2.925262451171875, 0.211578369140625, 3.277679443359375, 4.57208251953125, 2.4062347412109375, 5.6166229248046875, 0.3857421875, 3.0983428955078125, 2.3657989501953125, -0.5777912139892578, 7.157379150390625, 0.2105712890625, 3.6573333740234375, -7.891021728515625, 1.5039291381835938, 1.3664932250976562, 4.9224853515625, 0.118896484375, 5.85565185546875, 3.768035888671875, 6.183563232421875, 1.0176620483398438, 0.505645751953125, 1.019073486328125, 1.364044189453125, -6.32965087890625, -1.7402610778808594, 2.9336395263671875, -3.019989013671875, 0.0740966796875, 0.835174560546875, -2.087677001953125, -1.1020355224609375, 0.490020751953125, 0.39593505859375, 3.65106201171875, 0.0055065155029296875, 0.8600616455078125, -0.17491912841796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000065.npy"}
{"epoch": 0.13612565445026178, "step": 66, "batch_size": 128, "mean": 1.4377974271774292, "std": 3.779893398284912, "min": -10.190673828125, "p10": -2.5036346435546872, "median": 1.2566413879394531, "p90": 5.835099792480468, "max": 12.90771484375, "pos_frac": 0.7109375, "sample": [0.7438278198242188, 1.0652027130126953, 5.648895263671875, -1.78271484375, -1.493621826171875, 4.45196533203125, -1.127899169921875, 0.8646316528320312, 1.696563720703125, 1.394073486328125, -0.00330352783203125, -0.383026123046875, 2.2925071716308594, 1.2319717407226562, 1.7152862548828125, 0.128997802734375, -0.4823017120361328, 0.900604248046875, 9.714080810546875, 2.0351028442382812, 1.298583984375, -5.784698486328125, 5.234352111816406, 3.9893035888671875, 2.4185791015625, 2.77447509765625, 0.1477508544921875, 1.036712646484375, -0.6954612731933594, 6.4282989501953125, 3.109588623046875, 0.65740966796875, 11.7562255859375, -3.828857421875, 0.4792633056640625, -0.9271926879882812, 1.03900146484375, 10.226837158203125, 2.4265899658203125, -2.71356201171875, -4.58209228515625, 2.0895767211914062, 5.9572601318359375, 1.8544158935546875, 1.3466758728027344, 3.3834304809570312, -0.787841796875, 9.758636474609375, 8.7540283203125, -1.064239501953125, -7.97491455078125, 1.4764633178710938, 2.35064697265625, 12.90771484375, 9.45458984375, 0.4095001220703125, 0.5102386474609375, 3.0385284423828125, -10.190673828125, 1.10833740234375, -1.7047882080078125, 0.8478546142578125, 7.22283935546875, 1.3531646728515625, 4.61810302734375, -9.0379638671875, 0.7084503173828125, -1.21600341796875, 2.45074462890625, -0.2711944580078125, 0.985137939453125, 1.4617919921875, 6.345458984375, -3.23114013671875, 1.65557861328125, 1.28131103515625, 3.989715576171875, -1.2422332763671875, 4.216094970703125, -5.9626312255859375, 2.8684539794921875, 1.8609199523925781, -3.2713775634765625, 4.217620849609375, -0.7362899780273438, 1.5313873291015625, -1.1027069091796875, 0.66497802734375, 2.78021240234375, 1.2218017578125, 1.1415634155273438, 4.40594482421875, 3.63397216796875, -8.28973388671875, 0.014068603515625, -2.1334228515625, -3.8839263916015625, -2.413665771484375, 2.2700653076171875, 0.083770751953125, 7.40460205078125, 1.68206787109375, 3.431060791015625, 0.0, 3.4985809326171875, -1.3492813110351562, 0.9388275146484375, 7.0428314208984375, 1.4403152465820312, 4.47479248046875, 1.14825439453125, 2.43011474609375, -0.2827892303466797, -3.70159912109375, 0.277099609375, 3.085540771484375, -1.054595947265625, 3.50970458984375, 3.87103271484375, 5.038482666015625, -1.920501708984375, 0.931549072265625, 2.20404052734375, -1.7336196899414062, 3.083953857421875, 0.9241943359375, 5.782745361328125, 5.062408447265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000066.npy"}
{"epoch": 0.1382198952879581, "step": 67, "batch_size": 128, "mean": 1.3390545845031738, "std": 3.3055741786956787, "min": -7.22991943359375, "p10": -2.7951156616210935, "median": 1.2620229721069336, "p90": 5.1375579833984375, "max": 10.65850830078125, "pos_frac": 0.671875, "sample": [-0.7013473510742188, 4.143280029296875, -1.540802001953125, 8.61114501953125, 0.3137397766113281, -3.460205078125, 0.9428558349609375, 4.426483154296875, 4.740570068359375, -0.20383453369140625, 1.9107284545898438, 9.969207763671875, 1.3760147094726562, 2.36962890625, 3.64129638671875, 4.71875, 1.9262847900390625, 5.641937255859375, -0.02386474609375, -0.5462646484375, 1.3881072998046875, -1.4990005493164062, 4.37750244140625, -0.6555023193359375, 0.587432861328125, 5.907470703125, -3.90753173828125, -0.70379638671875, 1.80914306640625, -0.8450775146484375, -4.095855712890625, 0.1408100128173828, -0.7568359375, 1.8044281005859375, 1.0419921875, -4.464996337890625, 0.0, 1.36041259765625, 4.3785400390625, 1.987264633178711, -2.1057891845703125, 4.310211181640625, 3.477874755859375, 6.2399139404296875, 1.845794677734375, -1.712203025817871, 3.4565887451171875, -2.555908203125, -4.95428466796875, -0.1056060791015625, -3.50762939453125, 4.530303955078125, 0.0460662841796875, 7.924224853515625, -0.451934814453125, 3.096588134765625, 6.1337890625, -2.75982666015625, 1.501190185546875, 1.2642498016357422, 3.06146240234375, 2.80615234375, 3.70660400390625, -2.8774566650390625, 2.675567626953125, 0.4821033477783203, 0.3325843811035156, -5.51092529296875, 1.02630615234375, 0.339996337890625, 4.776824951171875, 2.5706787109375, 0.44325828552246094, 3.86737060546875, -7.1512451171875, 10.65850830078125, 8.93634033203125, 4.241058349609375, -0.047210693359375, 1.8554191589355469, 1.259796142578125, -1.7518596649169922, 2.8665771484375, 0.15570068359375, -4.040802001953125, -0.048564910888671875, 5.994697570800781, 1.168609619140625, 1.432586669921875, 4.12994384765625, 5.121124267578125, -7.15924072265625, 4.989715576171875, 3.1227569580078125, -5.097412109375, 4.2981719970703125, 0.540679931640625, 2.442840576171875, 2.5692138671875, 5.42724609375, 4.225372314453125, 1.3685226440429688, -0.6013717651367188, 1.5998382568359375, 0.8019561767578125, -0.122802734375, -1.2891464233398438, -7.22991943359375, -1.1551895141601562, 1.7997207641601562, 2.346466064453125, 3.495574951171875, -0.0867462158203125, -1.18157958984375, -1.1704368591308594, 5.1759033203125, 3.7078857421875, 0.8259544372558594, 5.509033203125, 1.167083740234375, 3.5069580078125, 1.0172195434570312, 2.60614013671875, 0.6936187744140625, 0.347381591796875, -1.0512924194335938, 1.090057373046875, -1.3961105346679688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000067.npy"}
{"epoch": 0.14031413612565444, "step": 68, "batch_size": 128, "mean": 1.7538150548934937, "std": 4.111147403717041, "min": -8.688995361328125, "p10": -3.1576004028320312, "median": 1.201319694519043, "p90": 6.420071411132812, "max": 16.822784423828125, "pos_frac": 0.6796875, "sample": [7.623138427734375, -0.06396102905273438, -1.0255126953125, -0.2318115234375, -1.1840476989746094, 0.05538177490234375, 0.052703857421875, 4.916351318359375, -0.5336036682128906, 1.5957870483398438, 0.989410400390625, -0.2814445495605469, 0.474456787109375, 6.541046142578125, 5.1488037109375, -3.894744873046875, 0.2052764892578125, 0.48572540283203125, 2.79290771484375, 2.1989173889160156, 7.292724609375, 0.4163818359375, -2.6379432678222656, 1.1719970703125, 16.822784423828125, 0.2709674835205078, 1.446359634399414, 4.93670654296875, 0.3393878936767578, -6.4886474609375, 5.84820556640625, 3.0789794921875, 11.777069091796875, 6.164794921875, 0.0986480712890625, 4.587432861328125, -0.0373382568359375, -0.4638214111328125, -4.36517333984375, -5.6888427734375, -0.808074951171875, -0.0743560791015625, 4.239749908447266, 2.60858154296875, 0.282318115234375, -4.71112060546875, 0.07633209228515625, 2.4466934204101562, 6.14276123046875, 6.13690185546875, 1.617584228515625, -3.217681884765625, 4.037353515625, 2.55169677734375, 11.061798095703125, -0.54290771484375, 4.644287109375, 10.74053955078125, -0.6017627716064453, 0.8716354370117188, 4.3280181884765625, 2.4777679443359375, -0.03974151611328125, 2.6523284912109375, -0.12623023986816406, -0.2747840881347656, -8.688995361328125, 3.9818115234375, 0.165374755859375, 10.542724609375, 0.46426963806152344, -3.986572265625, 1.812744140625, 1.2822341918945312, -5.74127197265625, 2.0697860717773438, -1.51153564453125, -2.7883987426757812, 11.8043212890625, -0.381439208984375, -3.1318511962890625, 1.7130126953125, 0.10158157348632812, 3.4757080078125, 2.11236572265625, 1.342864990234375, -1.047454833984375, 6.5693359375, -7.0426025390625, 9.648780822753906, 5.194122314453125, 1.160675048828125, 0.56573486328125, -2.32611083984375, 2.3502197265625, 6.163116455078125, -4.7729949951171875, -3.35540771484375, 5.3016357421875, 1.7108993530273438, 4.637908935546875, 0.9400882720947266, 2.6265640258789062, 6.36822509765625, 7.2041015625, 1.329254150390625, 5.3157958984375, 6.316558837890625, 2.3015289306640625, 4.52349853515625, 2.255523681640625, -3.244598388671875, 2.4438934326171875, 1.230642318725586, -0.49109649658203125, 1.2435646057128906, 4.944793701171875, 0.27984619140625, -1.020233154296875, 0.736907958984375, -0.8787994384765625, -2.05670166015625, 1.371978759765625, 11.533523559570312, 4.5863037109375, -2.370574951171875, -0.1510772705078125, 0.7970848083496094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000068.npy"}
{"epoch": 0.1424083769633508, "step": 69, "batch_size": 128, "mean": 2.137897491455078, "std": 3.882309913635254, "min": -9.811553955078125, "p10": -1.4301864624023437, "median": 1.8159637451171875, "p90": 7.0780090332031245, "max": 12.3739013671875, "pos_frac": 0.7265625, "sample": [4.8874969482421875, -0.484130859375, 2.5323638916015625, 4.9455718994140625, 1.58465576171875, -9.811553955078125, 4.3478546142578125, 1.9156417846679688, -1.14654541015625, -0.17637252807617188, 3.7835845947265625, 1.4504241943359375, 11.486785888671875, 0.0993194580078125, 8.0374755859375, -0.6294822692871094, 7.741943359375, 0.5664901733398438, 4.75506591796875, 0.97027587890625, -1.232330322265625, -0.000476837158203125, -3.2991371154785156, -5.96173095703125, 6.814697265625, 9.859283447265625, 5.02032470703125, 4.10906982421875, 0.75439453125, 0.791168212890625, -9.641571044921875, 4.80181884765625, -0.3924980163574219, 1.4480133056640625, 2.2109527587890625, -0.42905426025390625, 0.6349830627441406, 0.33164215087890625, 7.47857666015625, 1.092315673828125, 7.79437255859375, -5.990234375, 0.45273780822753906, 2.932342529296875, 1.80047607421875, 1.2685089111328125, -4.7037506103515625, 6.989715576171875, 2.6317901611328125, 0.5401840209960938, 6.3385772705078125, -0.165863037109375, 7.284027099609375, 5.376617431640625, 5.256378173828125, -5.35595703125, -3.94927978515625, 0.099456787109375, 3.462646484375, -0.28150367736816406, -0.17193603515625, 6.419708251953125, 0.16947555541992188, 5.186279296875, 0.3606719970703125, 2.9454345703125, 0.7367706298828125, 6.0291748046875, 4.2938232421875, 4.599853515625, 10.41326904296875, 3.6748504638671875, 2.2515945434570312, 0.26416015625, 3.09490966796875, -1.0742645263671875, 6.93182373046875, 1.2349319458007812, 3.0589752197265625, 2.1926422119140625, -1.5194072723388672, -0.427734375, 2.4781742095947266, 1.5523262023925781, -1.443206787109375, 1.905548095703125, 3.84844970703125, 0.8270263671875, 3.9417266845703125, 7.753082275390625, 6.79510498046875, 7.3853302001953125, 1.2097396850585938, 5.7106781005859375, -1.773040771484375, -1.327362060546875, 0.5846939086914062, 1.234405517578125, 9.366607666015625, 0.0271759033203125, 3.6014556884765625, 2.3370361328125, 12.3739013671875, 2.911041259765625, 4.98651123046875, 3.47259521484375, -2.32470703125, 2.75933837890625, -0.683197021484375, 1.94189453125, -0.70703125, -1.4246063232421875, -0.7645263671875, 6.4332275390625, 6.185882568359375, -7.3751220703125, -0.6779327392578125, -1.056427001953125, 2.6258163452148438, 8.852142333984375, 1.831451416015625, 3.45166015625, 0.77935791015625, -0.7097816467285156, -0.7244453430175781, 3.22381591796875, 6.980133056640625, 1.5813751220703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000069.npy"}
{"epoch": 0.14450261780104712, "step": 70, "batch_size": 128, "mean": 2.04455304145813, "std": 3.835651159286499, "min": -9.82305908203125, "p10": -1.557072067260742, "median": 1.6633148193359375, "p90": 7.202435302734375, "max": 13.921539306640625, "pos_frac": 0.6953125, "sample": [-0.37628173828125, 4.1178741455078125, 1.9184799194335938, -1.7282485961914062, -1.142913818359375, 0.8222503662109375, 4.006072998046875, 2.8544235229492188, -4.18463134765625, 9.026397705078125, -1.5670509338378906, -0.5611457824707031, -0.04815673828125, 3.2420654296875, 6.43402099609375, 4.28643798828125, 3.0280609130859375, 4.745513916015625, 0.5825347900390625, 1.8359375, 0.8396377563476562, 8.338104248046875, 0.49908447265625, 1.95867919921875, -5.17181396484375, 3.1895751953125, 0.6278228759765625, 2.7512054443359375, 2.07208251953125, -2.0630722045898438, 7.33441162109375, 0.881317138671875, -0.24979400634765625, 0.8062744140625, 1.7879638671875, 1.3861846923828125, 8.273178100585938, -4.42840576171875, -3.575103759765625, 0.223388671875, 1.7971725463867188, 0.30261993408203125, 10.3199462890625, 1.887420654296875, 1.849578857421875, -1.0441741943359375, -0.8879547119140625, -0.1264801025390625, -0.854339599609375, -1.3458251953125, 0.35333251953125, 1.0811386108398438, -0.86474609375, -0.13592529296875, -0.7662887573242188, 6.018768310546875, 8.4869384765625, 4.45770263671875, 7.080352783203125, 1.4496002197265625, 2.6811065673828125, 4.2506103515625, -0.7788314819335938, 0.6020050048828125, 2.611541748046875, 2.63189697265625, 0.028839111328125, 3.693756103515625, 1.4220314025878906, -0.721038818359375, -4.71942138671875, 0.95465087890625, -0.06853675842285156, 1.81597900390625, 0.7822799682617188, 7.4285888671875, -1.55279541015625, 1.637847900390625, 9.501953125, 1.784576416015625, 13.921539306640625, 4.718414306640625, 2.9512176513671875, 1.68878173828125, 5.6207733154296875, 2.670318603515625, -1.4753570556640625, 3.5126953125, -0.325408935546875, 11.757080078125, 4.738533020019531, 6.285614013671875, 0.0, 0.16925811767578125, 3.97113037109375, 8.1732177734375, 5.8520050048828125, 4.3211669921875, 0.64306640625, 12.1246337890625, 5.2073974609375, 5.41021728515625, 0.68341064453125, -2.5797462463378906, 2.86419677734375, -2.903717041015625, 7.1458740234375, 1.0269317626953125, 4.685272216796875, -2.2457427978515625, -0.8708267211914062, -1.2186813354492188, 0.955810546875, 2.7710113525390625, -7.458770751953125, -9.82305908203125, 2.693817138671875, -0.9792861938476562, 2.69000244140625, 6.07977294921875, 2.36151123046875, 2.5871429443359375, 1.840911865234375, -0.2878761291503906, 13.2017822265625, 0.0, 1.234527587890625, -0.5060195922851562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000070.npy"}
{"epoch": 0.14659685863874344, "step": 71, "batch_size": 128, "mean": 2.1108458042144775, "std": 4.566889762878418, "min": -13.1182861328125, "p10": -3.311775207519531, "median": 1.751068115234375, "p90": 8.103985595703124, "max": 13.4073486328125, "pos_frac": 0.7265625, "sample": [-0.523651123046875, 12.7354736328125, -1.0713996887207031, -5.8040771484375, -1.680572509765625, -5.497528076171875, -0.7804412841796875, 10.111083984375, -7.594573974609375, 4.244720458984375, 1.2726478576660156, 3.5638999938964844, -3.748382568359375, 4.1947021484375, 1.7166938781738281, 5.4716644287109375, 4.202667236328125, 0.848236083984375, 3.1614837646484375, 9.58087158203125, 2.3494415283203125, -0.2786865234375, 2.119781494140625, 13.4073486328125, 2.176666259765625, -1.70184326171875, 1.0672378540039062, 7.902984619140625, 3.1128463745117188, 4.43206787109375, -3.4738616943359375, 4.626655578613281, 2.099334716796875, 9.299072265625, -2.3951416015625, 4.4599609375, 8.074798583984375, 0.7279024124145508, 7.23309326171875, -4.35107421875, -0.590087890625, -1.3453216552734375, -2.111083984375, 2.9590682983398438, 2.25396728515625, -0.586273193359375, 1.90057373046875, -12.0380859375, -3.2423095703125, -0.2659149169921875, 1.5843353271484375, 5.693267822265625, -0.8789291381835938, 4.342563629150391, 1.3101158142089844, 5.6475372314453125, 10.724365234375, 1.521392822265625, -0.4906463623046875, -0.464935302734375, 4.5450286865234375, -1.533294677734375, 8.172088623046875, 7.3909912109375, 0.3656444549560547, 0.197265625, 0.06271171569824219, -1.273681640625, 2.70904541015625, 0.08957290649414062, 1.679168701171875, 13.1383056640625, 5.443267822265625, -4.9659423828125, -4.016284942626953, 4.5491943359375, 11.0338134765625, -6.752105712890625, -2.1179656982421875, 0.6269912719726562, 0.0322265625, -0.662841796875, 0.58062744140625, 2.602081298828125, 7.308746337890625, 6.629150390625, 1.642364501953125, 6.25006103515625, 6.239166259765625, 2.317108154296875, 1.765054702758789, 2.804290771484375, 0.3380126953125, 0.5191669464111328, 0.9616317749023438, 11.069381713867188, -0.17193603515625, 8.66656494140625, 3.330493927001953, 5.04193115234375, 6.391998291015625, -3.5531082153320312, 0.09880638122558594, 6.4041748046875, 9.325286865234375, 0.920013427734375, 1.050018310546875, 0.8298511505126953, -5.26171875, 1.749664306640625, 2.7054977416992188, 1.1046791076660156, 4.23297119140625, 1.1935348510742188, 1.752471923828125, 2.821187973022461, 2.5219955444335938, 4.8990478515625, 1.8178749084472656, 0.76727294921875, 2.4229278564453125, 1.7899093627929688, 4.287139892578125, 0.59295654296875, -0.48564910888671875, 11.84307861328125, -13.1182861328125, 3.2618560791015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000071.npy"}
{"epoch": 0.1486910994764398, "step": 72, "batch_size": 128, "mean": 2.030590772628784, "std": 4.200596809387207, "min": -11.745742797851562, "p10": -3.373735809326172, "median": 1.9655990600585938, "p90": 7.549382019042967, "max": 13.99420166015625, "pos_frac": 0.6796875, "sample": [3.9237060546875, 3.477081298828125, -2.2831573486328125, 1.73040771484375, 9.9122314453125, -2.562469482421875, 1.9186553955078125, 4.3251495361328125, 2.65814208984375, -0.7196006774902344, -0.3937530517578125, 2.27294921875, 1.7064971923828125, 2.44482421875, 2.7872314453125, 11.2607421875, -2.4327545166015625, 1.981842041015625, -2.963134765625, 3.132843017578125, 3.2225570678710938, 5.6240234375, 1.84454345703125, 10.377471923828125, 4.824432373046875, 0.8187732696533203, 0.39550018310546875, 3.242340087890625, -4.4373931884765625, 0.5389900207519531, 3.2205352783203125, -11.745742797851562, -1.733489990234375, -4.612937927246094, 1.8608551025390625, 2.644775390625, 1.182586669921875, -1.1398963928222656, 2.47271728515625, 0.33506011962890625, 4.8948974609375, -3.548828125, 5.484039306640625, 3.83551025390625, 7.9826202392578125, 2.0856704711914062, 2.9403839111328125, 0.05188751220703125, -4.852386474609375, -1.77032470703125, -3.3740615844726562, 5.9583740234375, -3.518707275390625, 0.0, 6.96124267578125, 7.36370849609375, 9.607192993164062, -0.29766845703125, 2.62896728515625, 10.19091796875, -0.08504486083984375, -3.895843505859375, 3.717273712158203, 5.735572814941406, 2.8879470825195312, 6.180908203125, 0.40146636962890625, 0.29906463623046875, -0.7447471618652344, 3.2281494140625, -3.8495330810546875, 1.9318313598632812, 1.2946853637695312, 3.4909515380859375, 8.06646728515625, 1.7859230041503906, 0.21988677978515625, 4.79547119140625, -0.567901611328125, 9.629714965820312, 1.704315185546875, 2.1438636779785156, -1.4381179809570312, -2.27081298828125, -1.5806427001953125, -3.37359619140625, 0.79486083984375, -0.0825042724609375, -1.8519477844238281, 8.577301025390625, -5.6058349609375, 6.97210693359375, -0.2834014892578125, 5.39208984375, 3.1262054443359375, 5.7250213623046875, 5.9579315185546875, 0.0160064697265625, 1.9493560791015625, 3.409332275390625, 2.98504638671875, 8.11810302734375, 0.0, 6.0292816162109375, -3.0100669860839844, 1.2753524780273438, 8.321884155273438, 13.99420166015625, 6.9294281005859375, 6.868537902832031, 3.9954681396484375, -1.240753173828125, -0.9660797119140625, -7.307403564453125, 11.1429443359375, -1.1026458740234375, -1.833984375, 0.0, 1.5225982666015625, 3.1500396728515625, 2.33990478515625, 2.4936447143554688, -5.570037841796875, 4.649688720703125, 2.300323486328125, -3.7687835693359375, 4.15008544921875, 6.9084930419921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000072.npy"}
{"epoch": 0.15078534031413612, "step": 73, "batch_size": 128, "mean": 1.9002209901809692, "std": 4.522438049316406, "min": -15.049530029296875, "p10": -3.2332481384277343, "median": 1.7521553039550781, "p90": 6.609985351562499, "max": 14.23797607421875, "pos_frac": 0.734375, "sample": [1.7122039794921875, 2.255706787109375, 6.2413330078125, 4.054473876953125, -0.716888427734375, 2.327301025390625, 1.335906982421875, -1.193389892578125, 1.3379974365234375, 6.235069274902344, 4.0101318359375, 2.512054443359375, 2.0096893310546875, 10.8101806640625, 0.3535175323486328, 0.1670074462890625, -7.48040771484375, -9.582366943359375, 0.966339111328125, 1.1305389404296875, 2.57904052734375, 0.31583404541015625, -0.036174774169921875, 2.501983642578125, 1.1775360107421875, 0.2794303894042969, 2.5869598388671875, 1.3135108947753906, 2.0313873291015625, 3.8678741455078125, -6.6871337890625, -3.6063690185546875, 4.822479248046875, 9.952423095703125, -1.760040283203125, 1.0715408325195312, -3.6719512939453125, 13.03411865234375, 5.462089538574219, 2.827362060546875, 3.9587554931640625, 1.29730224609375, 5.126922607421875, -2.42193603515625, 14.23797607421875, -1.69482421875, -2.240264892578125, -0.29473876953125, 2.5930099487304688, 12.32403564453125, -7.339141845703125, 1.78582763671875, 9.32623291015625, -15.049530029296875, 3.764678955078125, 1.6161651611328125, 3.587646484375, -4.67706298828125, 4.79119873046875, -2.2787628173828125, 3.407073974609375, 1.310302734375, 4.251777648925781, -0.97955322265625, -2.4933624267578125, 7.25762939453125, 4.896739959716797, -4.74420166015625, 8.018096923828125, 0.14283370971679688, -1.0650825500488281, -2.972412109375, 3.81024169921875, 4.3809814453125, 1.10302734375, 1.1419391632080078, -6.079132080078125, -6.98382568359375, -2.2947540283203125, 1.36468505859375, 5.349517822265625, 5.4375457763671875, 3.8878021240234375, 0.4513397216796875, -2.297332763671875, 3.1377182006835938, 0.3737144470214844, 2.839954376220703, -0.5311737060546875, 4.123443603515625, 0.9344291687011719, 1.47723388671875, 6.4884033203125, 3.2323989868164062, 5.2386474609375, 3.89068603515625, 2.462779998779297, 4.420989990234375, -7.071014404296875, 4.546142578125, 0.466522216796875, 5.77325439453125, 4.643272399902344, 3.86737060546875, 6.134552001953125, 6.539581298828125, 1.206634521484375, -0.38165283203125, 12.448516845703125, 7.526153564453125, 4.1324462890625, -1.2135009765625, 2.5131149291992188, -0.21436119079589844, 9.77252197265625, -3.307861328125, 0.21722984313964844, 6.028717041015625, -3.2012710571289062, 0.0689239501953125, 9.19720458984375, 1.7184829711914062, 6.774261474609375, 2.853729248046875, 1.278472900390625, 0.49399566650390625, 3.81854248046875, -2.75457763671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000073.npy"}
{"epoch": 0.15287958115183245, "step": 74, "batch_size": 128, "mean": 2.007657766342163, "std": 4.739091873168945, "min": -11.759796142578125, "p10": -3.6647094726562495, "median": 1.8002128601074219, "p90": 7.519618225097654, "max": 13.5355224609375, "pos_frac": 0.7109375, "sample": [3.6812744140625, 11.5870361328125, 3.3857421875, 0.49127197265625, 9.102142333984375, -1.6083984375, -4.308746337890625, 0.48626708984375, 6.4631500244140625, 6.44677734375, -0.672210693359375, 8.748672485351562, 0.0, 2.755645751953125, 1.1234130859375, 7.23834228515625, 5.07757568359375, 11.692779541015625, -2.4349822998046875, 1.198486328125, 1.43341064453125, 3.9891357421875, 1.86737060546875, 2.2493438720703125, 8.135955810546875, -5.87872314453125, -4.04547119140625, 0.30792236328125, -1.96417236328125, -3.1772918701171875, -1.8897705078125, -0.454345703125, -2.3555908203125, -0.34374237060546875, 5.51409912109375, 1.8088531494140625, 1.1655654907226562, 6.753879547119141, 0.2237415313720703, 7.3017730712890625, -0.7778091430664062, -1.03851318359375, -8.474395751953125, -8.666168212890625, 1.076080322265625, 4.483062744140625, 0.98529052734375, -2.48162841796875, -0.7196044921875, 4.693183898925781, -10.350006103515625, -8.92230224609375, -0.04266357421875, 2.923828125, 1.257415771484375, -0.7515869140625, 1.0887413024902344, 1.0549468994140625, -8.9449462890625, 6.115081787109375, 4.89263916015625, 4.1939849853515625, 2.455108642578125, 6.09552001953125, 6.29345703125, 1.18597412109375, 9.6385498046875, 5.70849609375, 1.8916244506835938, 13.086181640625, 1.16986083984375, 0.28643798828125, -1.2489852905273438, 5.413543701171875, -0.4738006591796875, 2.801971435546875, 2.317474365234375, 5.04058837890625, -0.1203460693359375, 5.4149627685546875, 3.2874526977539062, 0.5321884155273438, 1.8185272216796875, -11.759796142578125, 11.00738525390625, 1.0516510009765625, 0.732269287109375, 3.18084716796875, 1.5397720336914062, -0.7622604370117188, -0.5158538818359375, -3.50152587890625, 1.74615478515625, 6.101524353027344, 3.9569473266601562, 3.39508056640625, 5.348968505859375, -8.0130615234375, -5.865447998046875, 5.377685546875, 4.937408447265625, -5.083892822265625, 2.12353515625, 8.027923583984375, 2.29864501953125, 5.66961669921875, 2.92169189453125, -4.24432373046875, 6.2225341796875, -1.932861328125, 0.007228851318359375, 1.7915725708007812, 3.7463531494140625, 3.1412429809570312, 13.5355224609375, -0.5619277954101562, 0.9871444702148438, 4.929962158203125, 4.549896240234375, 0.447845458984375, 5.03369140625, 2.8648834228515625, 1.2071609497070312, 8.813728332519531, 12.64471435546875, 5.0689544677734375, 0.2707366943359375, 9.255279541015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000074.npy"}
{"epoch": 0.1549738219895288, "step": 75, "batch_size": 128, "mean": 2.655945062637329, "std": 4.472104549407959, "min": -10.3468017578125, "p10": -2.671162414550781, "median": 2.2647953033447266, "p90": 7.795520019531247, "max": 14.27874755859375, "pos_frac": 0.75, "sample": [2.144195556640625, 2.8544692993164062, 1.3049163818359375, 1.6302337646484375, -1.845916748046875, 7.08935546875, 6.5463714599609375, -2.61151123046875, 4.264739990234375, 2.9205551147460938, 0.8677902221679688, -4.7119140625, 2.3850479125976562, -0.4845314025878906, 6.864776611328125, 0.19780921936035156, 14.27874755859375, 2.01666259765625, 6.2896728515625, 4.3093109130859375, 7.45245361328125, 6.627216339111328, 6.429443359375, -3.42236328125, 0.741790771484375, 12.88177490234375, -4.8441162109375, 6.10260009765625, 0.571868896484375, 1.1769561767578125, 3.6124267578125, 7.0840301513671875, 5.39874267578125, 2.8721923828125, 0.1930084228515625, -1.5547504425048828, 3.6299896240234375, 2.9947509765625, 4.04022216796875, 1.29156494140625, 8.809722900390625, 5.833751678466797, 0.978851318359375, -9.9442138671875, 2.104248046875, -1.02264404296875, 0.7310161590576172, 0.0322113037109375, 9.44183349609375, 0.8108367919921875, 4.806396484375, 4.1947021484375, 2.1900177001953125, 0.7653045654296875, 5.485595703125, 4.330879211425781, -4.934478759765625, 11.0728759765625, -0.5120697021484375, 7.19097900390625, 0.9803466796875, 2.358030319213867, 1.379241943359375, 7.149383544921875, 1.9485969543457031, 0.9868011474609375, 7.1602783203125, -1.8982009887695312, 2.3395729064941406, 1.073333740234375, 2.4433441162109375, -2.164520263671875, 1.2740554809570312, 10.4661865234375, 10.8201904296875, 0.37613868713378906, -10.3468017578125, -6.728179931640625, -1.741790771484375, 3.434589385986328, 11.249603271484375, 5.8904266357421875, 6.28875732421875, -4.055213928222656, -0.08966445922851562, 9.028457641601562, -0.04033660888671875, -0.38922119140625, 2.5911712646484375, 7.1983489990234375, 4.37738037109375, -0.214691162109375, 4.9765625, 1.58056640625, 5.92266845703125, 12.072296142578125, 0.081817626953125, 4.95654296875, -0.517181396484375, 6.7156219482421875, -6.5073089599609375, -0.02508544921875, 4.933563232421875, 9.71575927734375, 6.69683837890625, 5.6953125, -0.4922218322753906, 4.5637969970703125, 1.672698974609375, -3.603302001953125, 6.54345703125, -0.18901634216308594, 1.289520263671875, 5.819499969482422, 4.447139739990234, 8.59600830078125, -0.036407470703125, 1.5721588134765625, 4.494781494140625, -1.304443359375, 2.8355865478515625, 0.917572021484375, 4.925441741943359, 9.29656982421875, -2.8110198974609375, -2.8103485107421875, -7.352867126464844, 0.1863555908203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000075.npy"}
{"epoch": 0.15706806282722513, "step": 76, "batch_size": 128, "mean": 2.9792046546936035, "std": 5.129618167877197, "min": -9.860443115234375, "p10": -1.9141906738281245, "median": 1.9240169525146484, "p90": 9.78920440673828, "max": 22.171646118164062, "pos_frac": 0.7421875, "sample": [6.477294921875, -3.765716552734375, 1.2905349731445312, 3.1735992431640625, 2.5305023193359375, 5.7060546875, 1.135406494140625, 1.2889537811279297, 2.5273399353027344, 4.792938232421875, 9.7216796875, 0.2101459503173828, 1.763153076171875, -0.925323486328125, 15.1783447265625, 5.72509765625, -0.6491546630859375, -1.7928466796875, 1.6752777099609375, -1.5035171508789062, 13.192413330078125, 0.5717239379882812, 4.157073974609375, 2.3287506103515625, 2.5281829833984375, 11.980377197265625, 2.041168212890625, 2.9904937744140625, 1.0874519348144531, 0.899566650390625, 1.87261962890625, 0.09122848510742188, 5.8012847900390625, 8.8330078125, -9.688232421875, 15.6146240234375, 4.023284912109375, 5.8502349853515625, 2.68939208984375, 3.32220458984375, 9.56292724609375, 5.20550537109375, 6.639373779296875, 6.675689697265625, -6.05224609375, 1.1221637725830078, 3.8694229125976562, 4.1509552001953125, 0.1195526123046875, 9.946762084960938, 0.0, 4.735809326171875, 0.30670166015625, -0.18951416015625, 7.621391296386719, -3.11053466796875, 11.180450439453125, -7.123077392578125, -2.543548583984375, -4.216094970703125, 6.1402130126953125, 11.380523681640625, -2.19732666015625, 1.89801025390625, 3.3811187744140625, 1.1562061309814453, 3.951995849609375, 3.900543212890625, 5.304050445556641, -1.685791015625, -1.672210693359375, 10.859161376953125, -2.6278076171875, 0.8724288940429688, -4.3095245361328125, 0.0, 9.519149780273438, 4.417060852050781, 5.4986419677734375, 12.746734619140625, 2.1490249633789062, -0.7101325988769531, 1.559417724609375, 1.3330535888671875, -0.010986328125, 1.7409820556640625, 1.9500236511230469, 7.4063720703125, 1.8871002197265625, 3.0976104736328125, 3.9459991455078125, 1.2451705932617188, -0.5620956420898438, 1.7759552001953125, -0.16915512084960938, -1.5553207397460938, -4.8898162841796875, 0.450164794921875, -0.010833740234375, 0.0422515869140625, 7.430419921875, 1.1988143920898438, 7.41796875, 12.673187255859375, 8.900390625, 17.465606689453125, 12.266815185546875, 0.595977783203125, 2.698638916015625, -3.685791015625, 0.1159515380859375, -1.2559051513671875, 22.171646118164062, -0.56317138671875, -9.860443115234375, -0.1141204833984375, 4.1334228515625, 3.8591156005859375, 8.7969970703125, 1.0419540405273438, 2.2700042724609375, 8.0560302734375, 2.212188720703125, 1.2632904052734375, -1.2109489440917969, 2.2723312377929688, 0.0243682861328125, -0.6927947998046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000076.npy"}
{"epoch": 0.15916230366492146, "step": 77, "batch_size": 128, "mean": 2.309375047683716, "std": 5.445695400238037, "min": -15.97662353515625, "p10": -3.5792945861816405, "median": 1.5818519592285156, "p90": 9.406378173828125, "max": 19.66790771484375, "pos_frac": 0.640625, "sample": [-0.737152099609375, 1.386322021484375, 2.072540283203125, 3.9360809326171875, 1.111572265625, 0.6122283935546875, 8.92486572265625, 3.82086181640625, -0.25969505310058594, 3.7177810668945312, -3.37890625, -0.4008216857910156, 9.759140014648438, 5.3194580078125, 11.496337890625, 1.4555816650390625, 0.3452606201171875, 3.1932220458984375, -15.97662353515625, 5.9736480712890625, -1.5047683715820312, -0.812957763671875, -1.859283447265625, -1.683837890625, 7.08636474609375, 9.29083251953125, 5.898193359375, 2.535633087158203, 4.4541473388671875, 4.470977783203125, -4.793243408203125, -0.406646728515625, -0.099853515625, 7.806182861328125, -1.49493408203125, 1.1016254425048828, 1.0643463134765625, 4.28387451171875, 9.83441162109375, -1.2078475952148438, -0.1749420166015625, 0.8851547241210938, 6.921356201171875, 1.351806640625, 4.147117614746094, 3.1748046875, 0.40012359619140625, 19.66790771484375, -1.661712646484375, 1.258514404296875, 5.45782470703125, -2.64404296875, 4.095733642578125, 9.3941650390625, 6.7343292236328125, 6.122123718261719, 3.194183349609375, -2.02374267578125, 1.822601318359375, 0.08778190612792969, -1.2969818115234375, -3.5547561645507812, 7.62640380859375, 18.23614501953125, -1.72711181640625, 6.154510498046875, 4.205902099609375, -1.5771484375, 11.27044677734375, 2.5669097900390625, 9.43487548828125, 2.54278564453125, -8.838897705078125, -3.708709716796875, -3.6365509033203125, -1.9290313720703125, 4.436614990234375, 1.5743484497070312, 0.7206878662109375, 0.2971649169921875, -2.6953125, 11.285720825195312, -2.7079696655273438, -3.2855224609375, 1.58935546875, 3.91619873046875, 9.243499755859375, -8.5712890625, -3.8045654296875, 3.9805221557617188, 2.1672286987304688, 2.1681365966796875, -2.004791259765625, 3.3757781982421875, 4.8914031982421875, -4.68768310546875, -1.2298431396484375, -6.70379638671875, 9.718994140625, 4.1219482421875, -4.884521484375, 5.0873260498046875, -1.2588939666748047, 1.183685302734375, 7.136100769042969, 1.23614501953125, -0.2783203125, 12.02301025390625, 13.320068359375, -0.1897735595703125, 2.1736717224121094, -6.14947509765625, -5.230743408203125, 17.85284423828125, -0.4184913635253906, 4.688812255859375, -4.7762908935546875, 2.84033203125, -2.252065658569336, -0.88360595703125, 5.8079833984375, 2.4644317626953125, 4.647491455078125, 6.893310546875, 5.30035400390625, 14.028076171875, -2.9438323974609375, 0.0707550048828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000077.npy"}
{"epoch": 0.1612565445026178, "step": 78, "batch_size": 128, "mean": 3.0111918449401855, "std": 5.300633907318115, "min": -12.927078247070312, "p10": -2.992021179199219, "median": 2.5556259155273438, "p90": 9.910678100585937, "max": 20.9434814453125, "pos_frac": 0.7265625, "sample": [-0.60467529296875, 6.01190185546875, -5.90155029296875, 5.383392333984375, 9.51861572265625, 2.0887985229492188, 2.3054122924804688, 3.960052490234375, -3.0245132446289062, 5.4840545654296875, -0.35610198974609375, 2.08966064453125, 4.593257904052734, 3.839599609375, -0.9791450500488281, 3.91180419921875, 3.7555694580078125, 6.5892181396484375, 10.783706665039062, 0.055389404296875, 0.0, 9.975128173828125, -2.5141448974609375, 2.418792724609375, 1.1286087036132812, 1.6073074340820312, 5.67620849609375, 2.8272151947021484, -3.8111038208007812, 4.666778564453125, -2.9780960083007812, 4.95654296875, 10.204605102539062, 0.4701995849609375, 8.172698974609375, -1.68939208984375, 1.784515380859375, 16.281341552734375, 0.705902099609375, -4.6980743408203125, 4.3559722900390625, 4.049652099609375, -0.6959991455078125, 20.9434814453125, 11.29425048828125, -3.188323974609375, 3.7896575927734375, 3.065155029296875, -0.41497802734375, 3.531524658203125, 3.9285888671875, -12.927078247070312, 2.0717544555664062, 2.7209014892578125, 0.1673736572265625, -0.6817283630371094, 6.7141876220703125, 4.8339385986328125, 3.2881126403808594, -1.3148956298828125, 10.837890625, 0.56732177734375, 4.274635314941406, 5.8822479248046875, 2.570667266845703, -3.8983230590820312, 1.1097259521484375, 3.49139404296875, 3.7152099609375, -9.626617431640625, 12.280670166015625, 2.200164794921875, 2.92987060546875, 2.0694046020507812, 9.53802490234375, 6.620147705078125, -0.3410663604736328, 2.9908599853515625, -1.2259674072265625, 1.8233795166015625, -4.814239501953125, -1.1647167205810547, 9.670806884765625, 15.70697021484375, 1.4734954833984375, 1.511077880859375, -0.6873626708984375, 3.00421142578125, -4.899078369140625, 4.883758544921875, 3.0336837768554688, -2.5303497314453125, 9.481048583984375, -0.8458251953125, -7.2547607421875, 2.0084686279296875, 2.595458984375, -3.8742828369140625, 11.7073974609375, 0.6027679443359375, 0.4690513610839844, 1.5783462524414062, -1.9522171020507812, 9.883056640625, 6.954010009765625, 3.6212310791015625, 6.5211181640625, -0.0112457275390625, 2.0020370483398438, 2.1495208740234375, 14.335205078125, 0.403778076171875, 2.69134521484375, 8.207763671875, 3.51348876953125, 2.5405845642089844, 6.4202880859375, 13.920867919921875, 9.4658203125, -3.816650390625, -1.863250732421875, -0.15900421142578125, 18.421539306640625, 1.3533363342285156, 1.727630615234375, 2.8873748779296875, -1.5236968994140625, 8.052032470703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000078.npy"}
{"epoch": 0.16335078534031414, "step": 79, "batch_size": 128, "mean": 3.0458428859710693, "std": 7.065212249755859, "min": -42.799072265625, "p10": -3.2361297607421875, "median": 2.156036376953125, "p90": 10.63734436035156, "max": 23.9051513671875, "pos_frac": 0.6953125, "sample": [-0.18297576904296875, 5.5961456298828125, 9.888320922851562, -2.444305419921875, 5.04620361328125, -2.046478271484375, 1.5575408935546875, -0.9913711547851562, 3.0673980712890625, -2.5529937744140625, 12.450469970703125, -1.199188232421875, 0.0, 6.14727783203125, -0.36041259765625, -0.93328857421875, -0.65753173828125, -2.7748565673828125, -5.828453063964844, 0.21551513671875, 2.2104339599609375, -3.358795166015625, -3.245208740234375, -1.7914505004882812, -3.23223876953125, 4.0323486328125, 2.3578643798828125, 0.0, 9.04150390625, 0.4297618865966797, -6.7969970703125, 11.1309814453125, 6.070884704589844, 5.4748077392578125, 3.3423614501953125, 1.3346328735351562, 2.39373779296875, 2.1016387939453125, 4.029045104980469, 0.1048583984375, 3.7349853515625, -8.242515563964844, 10.268798828125, 0.44909095764160156, -0.460479736328125, 6.1206817626953125, 7.13334846496582, 16.089508056640625, 2.2886276245117188, 3.5282745361328125, 0.2515869140625, -3.619354248046875, -42.799072265625, -3.746246337890625, 0.9499969482421875, 8.141265869140625, 0.46820068359375, 15.54302978515625, 9.99700927734375, 10.06182861328125, -0.9031219482421875, 5.248809814453125, 3.1366424560546875, 5.5359344482421875, 11.4912109375, 2.44329833984375, 3.44647216796875, 2.736347198486328, -2.2264251708984375, 1.8460693359375, -1.71826171875, 4.374137878417969, 0.6022720336914062, 2.064422607421875, -1.1861572265625, -4.12530517578125, 8.872283935546875, -0.72076416015625, 10.51617431640625, 1.872528076171875, -3.2940597534179688, 18.47930908203125, 8.36212158203125, 7.046356201171875, 3.697906494140625, 6.28448486328125, 4.017189025878906, 0.5855178833007812, 0.21280670166015625, 0.9931221008300781, 0.26030921936035156, 5.344169616699219, 0.7824935913085938, 0.781585693359375, 2.0345840454101562, 23.9051513671875, 0.05096435546875, 5.10797119140625, -4.7310638427734375, 9.0443115234375, 14.931976318359375, 10.920074462890625, 4.308197021484375, 5.0801544189453125, -5.016448974609375, 8.761951446533203, 7.97503662109375, -0.40799713134765625, 8.269622802734375, -0.8938140869140625, -4.01861572265625, 18.897796630859375, 0.186614990234375, -0.9904632568359375, 4.801822662353516, 6.510475158691406, 6.547119140625, 4.97735595703125, 0.8726577758789062, 5.6495361328125, -0.4354248046875, 10.298248291015625, 11.7235107421875, 21.387451171875, -0.540985107421875, -2.768035888671875, 0.988037109375, 13.79449462890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000079.npy"}
{"epoch": 0.16544502617801046, "step": 80, "batch_size": 128, "mean": 2.5625574588775635, "std": 6.140864849090576, "min": -22.22509765625, "p10": -2.9001068115234374, "median": 1.8758087158203125, "p90": 11.040957641601562, "max": 22.84234619140625, "pos_frac": 0.6640625, "sample": [2.3118972778320312, 4.19610595703125, -1.5398101806640625, 1.0835342407226562, 0.611572265625, -1.89410400390625, 1.609588623046875, -5.74920654296875, 7.97674560546875, -3.4477691650390625, 1.012420654296875, -2.18341064453125, 5.97265625, 5.366943359375, 4.21331787109375, -2.6895523071289062, -2.517181396484375, 10.9376220703125, 3.9035797119140625, 4.271759033203125, 16.375946044921875, 0.43428802490234375, 2.47320556640625, 5.93023681640625, -0.283843994140625, 11.755889892578125, 6.9482421875, -3.089141845703125, 3.8093528747558594, 10.408843994140625, 2.7752227783203125, 0.7187385559082031, 3.849365234375, -1.6299591064453125, -11.80511474609375, 1.658660888671875, -22.22509765625, 6.35009765625, 0.0, 12.72943115234375, -2.644683837890625, 9.18927001953125, -1.385986328125, -2.23358154296875, 3.041229248046875, -7.3187255859375, -0.26483154296875, 4.91094970703125, -1.4111328125, -2.1451950073242188, 1.2681751251220703, 1.27642822265625, 2.3127975463867188, 4.990325927734375, 8.6890869140625, -0.04967308044433594, 0.7391319274902344, -0.15203094482421875, -1.4714813232421875, 12.039154052734375, 3.11993408203125, 1.3648147583007812, 5.985240936279297, 5.253265380859375, -4.367523193359375, 1.53839111328125, 3.3741455078125, 0.41188812255859375, 6.4601287841796875, 2.7869091033935547, -2.54425048828125, 4.1654510498046875, 3.1681432723999023, -4.8239898681640625, 18.14300537109375, -0.87347412109375, -1.7041091918945312, -0.032501220703125, 4.779693603515625, -8.6158447265625, 12.282623291015625, 4.074737548828125, -1.033294677734375, 10.9962158203125, 2.60565185546875, 2.9014930725097656, -5.5196990966796875, 1.18939208984375, -1.2199935913085938, 17.72332763671875, 6.667808532714844, 2.09295654296875, 11.44281005859375, 3.9168701171875, 3.9543418884277344, 12.661102294921875, 7.03021240234375, 1.2799453735351562, -10.860916137695312, 0.33341217041015625, -6.88104248046875, 11.145355224609375, -2.819091796875, 5.92333984375, 4.07537841796875, 9.20355224609375, 0.0687103271484375, -1.0576324462890625, 4.630645751953125, 8.2730712890625, -1.962066650390625, 0.26184844970703125, 0.333221435546875, -0.28711700439453125, 0.096099853515625, -5.499687194824219, 3.6860084533691406, -1.3576202392578125, 0.9714012145996094, -2.3568572998046875, 5.1134490966796875, 12.327239990234375, 9.23223876953125, 3.3921966552734375, 22.84234619140625, -0.063812255859375, 3.816741943359375, 14.78082275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000080.npy"}
{"epoch": 0.16753926701570682, "step": 81, "batch_size": 128, "mean": 2.996581554412842, "std": 6.531432151794434, "min": -17.71063232421875, "p10": -3.041569137573242, "median": 2.534198760986328, "p90": 11.687435913085936, "max": 16.459259033203125, "pos_frac": 0.6875, "sample": [0.017564773559570312, -0.12551307678222656, 10.094917297363281, -10.851486206054688, 4.9907379150390625, -0.352386474609375, -7.589141845703125, 3.3591079711914062, 16.459259033203125, 7.812255859375, 15.940536499023438, 13.14215087890625, -1.8946685791015625, -2.831573486328125, 7.546142578125, 0.6642990112304688, 4.484527587890625, 8.32623291015625, 5.238800048828125, 4.9127044677734375, 0.7505035400390625, 11.58184814453125, -0.21344757080078125, -2.6928176879882812, 4.681724548339844, 3.9504241943359375, 2.2739410400390625, -0.109405517578125, 9.187103271484375, 7.2333984375, 12.25634765625, 2.9933624267578125, 7.946601867675781, 3.93310546875, -2.51373291015625, 3.3003692626953125, -4.424163818359375, 15.420166015625, 12.896697998046875, 3.116180419921875, -3.6660919189453125, 0.1180419921875, 10.489776611328125, -0.06198883056640625, -0.649383544921875, 2.039337158203125, -15.656982421875, -2.753387451171875, 7.508544921875, 10.3785400390625, 0.5095977783203125, 3.017791748046875, -6.142631530761719, 9.16192626953125, -4.7273712158203125, -0.3654022216796875, 6.73577880859375, 4.2860565185546875, 1.8194427490234375, -15.083389282226562, -3.325103759765625, 3.0417633056640625, 4.913482666015625, 1.49591064453125, 0.31192588806152344, 4.050506591796875, 0.35858154296875, 0.8877716064453125, 2.898406982421875, 4.00457763671875, -0.052276611328125, -1.5572433471679688, -0.4911346435546875, 11.933807373046875, 14.915283203125, 9.408966064453125, -2.0037078857421875, 5.2913970947265625, 2.028745651245117, 7.112945556640625, -12.27313232421875, 5.974617004394531, 10.7117919921875, -3.0031890869140625, 2.07635498046875, 1.6376113891601562, -0.8890457153320312, -2.31658935546875, 0.5167694091796875, 2.9203109741210938, -17.71063232421875, -1.664794921875, -15.339019775390625, 1.23236083984375, 2.6988983154296875, 1.7151756286621094, 0.0400390625, -0.16573715209960938, 13.2982177734375, 13.100784301757812, 10.564605712890625, 14.4171142578125, 4.3972625732421875, -2.099578857421875, 9.726409912109375, 2.5777359008789062, -0.7177734375, 1.4634246826171875, -0.016681671142578125, 6.810455322265625, -3.131122589111328, 5.429878234863281, 0.46976280212402344, -0.35369873046875, 2.49066162109375, 10.235008239746094, 7.5612030029296875, -1.9907341003417969, 10.738525390625, 6.2918701171875, 7.243499755859375, 13.36932373046875, 14.54339599609375, 10.092742919921875, 1.68438720703125, 0.9944686889648438, 6.2779541015625, -1.1639556884765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000081.npy"}
{"epoch": 0.16963350785340314, "step": 82, "batch_size": 128, "mean": 3.0780720710754395, "std": 5.86331844329834, "min": -15.13677978515625, "p10": -3.675746154785156, "median": 3.310028076171875, "p90": 10.701971435546875, "max": 21.21453857421875, "pos_frac": 0.734375, "sample": [2.3040771484375, 4.41522216796875, 0.1625518798828125, -5.9002838134765625, -0.649688720703125, 6.3970947265625, 1.671142578125, 11.26318359375, 7.5015411376953125, 0.4636688232421875, 0.48419189453125, 5.74310302734375, -4.6346282958984375, -2.8191070556640625, -2.4084014892578125, 7.4491424560546875, 4.891937255859375, 3.4969863891601562, 12.256851196289062, 5.971427917480469, 1.207275390625, -3.336212158203125, 7.353416442871094, -1.8432693481445312, 0.6674880981445312, 9.943756103515625, -9.510833740234375, 4.625152587890625, 5.267333984375, 3.669921875, 17.5262451171875, 2.832763671875, 11.720794677734375, 5.2064361572265625, 10.71551513671875, 12.93817138671875, -0.8534049987792969, 4.789466857910156, 4.1373748779296875, -15.13677978515625, -3.6324920654296875, -0.0555267333984375, 4.1220703125, 3.7599334716796875, -2.7904815673828125, 10.951019287109375, -1.703277587890625, 8.28875732421875, 19.21551513671875, -1.07598876953125, -1.659088134765625, 17.604019165039062, 5.256114959716797, 4.6457672119140625, 3.483428955078125, 0.6898860931396484, 10.6961669921875, -3.77667236328125, -1.2100830078125, 3.3184356689453125, 3.3413162231445312, -5.722259521484375, -13.145660400390625, 3.522584915161133, -4.5595703125, 1.144073486328125, -4.0733642578125, 1.4058303833007812, 10.51312255859375, 3.7626495361328125, -0.6458854675292969, 6.2994232177734375, 1.785797119140625, 3.7418975830078125, 0.1757659912109375, 3.1652679443359375, 1.0582275390625, 1.4356575012207031, 4.697704315185547, 1.0985946655273438, 1.0040245056152344, 21.21453857421875, -3.121368408203125, 8.15478515625, -0.240966796875, 2.8936691284179688, 7.431732177734375, 9.573772430419922, 1.9174652099609375, 2.857757568359375, 3.448211669921875, -2.1318511962890625, 2.999378204345703, 4.940673828125, 7.200736999511719, 6.920562744140625, -6.20391845703125, 2.7736968994140625, 5.228790283203125, 11.61053466796875, 6.613983154296875, -0.1257648468017578, 0.9451980590820312, 4.6607208251953125, 6.93157958984375, 3.3844127655029297, 0.0061931610107421875, -0.57061767578125, -4.47607421875, 1.6415252685546875, -3.0389022827148438, 3.07879638671875, 6.689208984375, 7.664146423339844, -3.51885986328125, 11.548248291015625, 12.8477783203125, 3.3016204833984375, 4.388580322265625, 5.274444580078125, 4.020751953125, 5.62835693359375, 10.021209716796875, 6.169677734375, 1.1101551055908203, 0.7440547943115234, -10.5758056640625, -3.952911376953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000082.npy"}
{"epoch": 0.17172774869109947, "step": 83, "batch_size": 128, "mean": 3.5550053119659424, "std": 6.968375205993652, "min": -14.417083740234375, "p10": -3.4512481689453125, "median": 2.7752227783203125, "p90": 11.539912033081055, "max": 29.136474609375, "pos_frac": 0.6875, "sample": [7.164764404296875, 0.0, 10.02618408203125, 3.41180419921875, 1.785919189453125, -0.1491851806640625, -1.0555267333984375, 9.72662353515625, 2.205078125, 1.0543365478515625, -2.2546539306640625, -14.417083740234375, -0.37268829345703125, -11.557037353515625, 0.90582275390625, 0.72857666015625, 4.5614776611328125, 9.98309326171875, 8.087265014648438, 2.4969139099121094, 9.887420654296875, 6.786529541015625, -1.410186767578125, 5.81744384765625, 10.976806640625, -4.420654296875, -1.655029296875, -0.7666549682617188, -1.585235595703125, 11.561859130859375, 5.973869323730469, -0.589141845703125, 3.104644775390625, 1.6633453369140625, -6.834320068359375, 13.98797607421875, 2.98583984375, 2.832061767578125, 19.663162231445312, 1.2987213134765625, 4.454380035400391, -1.725372314453125, 2.8822708129882812, 4.5589752197265625, -2.378082275390625, 10.6729736328125, 5.164794921875, 4.533966064453125, -1.09173583984375, -2.825592041015625, 6.3537750244140625, 5.679412841796875, 18.9351806640625, 10.477783203125, -11.4749755859375, 1.2942657470703125, 9.711669921875, 3.98724365234375, 29.136474609375, 0.72174072265625, -3.479339599609375, 11.530506134033203, 4.41156005859375, 6.3213348388671875, 2.248626708984375, -1.75079345703125, 0.1173095703125, 0.47747039794921875, -3.2843780517578125, 8.102462768554688, -7.8053131103515625, 6.0735931396484375, -0.29308319091796875, -2.370758056640625, -3.439208984375, 3.403961181640625, 15.91790771484375, -3.571868896484375, 4.22296142578125, 9.467445373535156, -3.655487060546875, 16.17315673828125, 5.31842041015625, 3.249786376953125, -13.9007568359375, -6.434539794921875, 6.517738342285156, 14.368255615234375, -8.337310791015625, 10.530914306640625, 0.17843055725097656, -1.4244003295898438, 0.2440948486328125, -2.024749755859375, 5.868114471435547, 17.783050537109375, 16.0103759765625, 1.389617919921875, 2.5201873779296875, 1.305511474609375, 16.4637451171875, -0.438262939453125, 6.7421875, 2.7183837890625, -1.141510009765625, 0.37674713134765625, 7.5491790771484375, 6.22552490234375, 2.350677490234375, -1.84478759765625, 8.4478759765625, -3.7183990478515625, 9.563674926757812, 9.60552978515625, 3.02105712890625, 3.7255706787109375, 2.6612396240234375, 17.9384765625, 20.0068359375, -2.73760986328125, 1.846832275390625, -0.7872467041015625, 8.968154907226562, -0.8458099365234375, 4.428863525390625, 6.403923034667969, 2.8564453125, 1.9933013916015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000083.npy"}
{"epoch": 0.17382198952879582, "step": 84, "batch_size": 128, "mean": 2.79119610786438, "std": 6.929223537445068, "min": -21.245849609375, "p10": -4.181997680664063, "median": 2.9218368530273438, "p90": 11.892144775390625, "max": 18.792755126953125, "pos_frac": 0.6953125, "sample": [-1.3683452606201172, 7.923614501953125, 16.415191650390625, 0.03850555419921875, -2.723236083984375, -4.848358154296875, 5.9256591796875, 4.3928985595703125, 5.932273864746094, -4.1673583984375, 0.8254852294921875, 8.8551025390625, 1.2394943237304688, -2.4499588012695312, 3.4061279296875, 4.128639221191406, 5.8917999267578125, 0.44654083251953125, -0.19688796997070312, 9.68902587890625, -3.2022171020507812, 8.39459228515625, 0.985321044921875, 5.40087890625, 8.61175537109375, 2.17755126953125, -8.493408203125, 6.647216796875, 6.7938232421875, 3.829925537109375, -3.318359375, -18.09539794921875, -21.245849609375, 12.3468017578125, 3.0095672607421875, -1.8109130859375, 6.5985107421875, 2.7645034790039062, 0.3212890625, -0.8779373168945312, 6.034759521484375, 2.8341064453125, 12.12127685546875, -4.7056427001953125, 11.7939453125, -2.78399658203125, 1.7266082763671875, -3.2380447387695312, 12.49066162109375, 4.171295166015625, 18.580322265625, -16.66839599609375, 9.5450439453125, 1.6011962890625, 3.7481155395507812, -3.3973388671875, 3.3747634887695312, -4.607269287109375, 6.589599609375, 7.9776153564453125, 5.690277099609375, 3.701904296875, -0.208831787109375, 5.1485595703125, 12.971038818359375, 1.01025390625, 3.815032958984375, 0.9016571044921875, -0.8812255859375, 2.5899505615234375, -2.58160400390625, 4.463165283203125, 2.016326904296875, -0.056671142578125, 2.5467071533203125, 12.48846435546875, 7.780250549316406, 8.920616149902344, -11.7628173828125, 4.797515869140625, 3.0628662109375, 2.17926025390625, -2.904052734375, 2.1339340209960938, -4.216156005859375, -0.3345050811767578, -0.65771484375, 8.573715209960938, 16.288177490234375, 0.7935791015625, 10.960113525390625, 4.0067596435546875, -5.303985595703125, 6.06524658203125, -0.29389190673828125, -0.1012115478515625, 4.0014801025390625, 2.6669559478759766, 3.93096923828125, 3.1496734619140625, 5.2561492919921875, -0.9090995788574219, 2.138397216796875, 2.2639083862304688, 8.707923889160156, 5.874088287353516, -3.242767333984375, 18.792755126953125, -12.637451171875, 10.371349334716797, 17.4327392578125, 2.0793609619140625, 6.363128662109375, 4.274169921875, 16.4810791015625, 7.75457763671875, -7.97882080078125, -15.117881774902344, 5.09906005859375, -2.737396240234375, 0.4141044616699219, 12.702987670898438, 13.9844970703125, -2.90618896484375, 4.430633544921875, 5.51043701171875, 1.1872711181640625, -3.0501861572265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000084.npy"}
{"epoch": 0.17591623036649215, "step": 85, "batch_size": 128, "mean": 2.0377633571624756, "std": 5.745347023010254, "min": -15.7637939453125, "p10": -5.457533264160155, "median": 2.454191207885742, "p90": 9.429136657714844, "max": 16.3424072265625, "pos_frac": 0.6640625, "sample": [4.1799774169921875, -3.102783203125, -1.9162712097167969, 16.3424072265625, 6.428188323974609, 5.0584716796875, -0.97308349609375, -0.1383056640625, 6.4932861328125, 3.5478515625, 2.4921226501464844, -2.8061981201171875, 2.1138572692871094, 8.400192260742188, 6.10565185546875, 3.566375732421875, 6.06475830078125, 5.636741638183594, 5.398468017578125, 2.7579879760742188, 1.51055908203125, 0.8714599609375, 9.407516479492188, 14.486175537109375, 1.4995155334472656, -0.93377685546875, -2.027191162109375, -5.1663360595703125, -7.487579345703125, 4.669189453125, 5.1231689453125, 4.4592132568359375, -0.4838409423828125, 2.0419921875, 9.479583740234375, 10.912673950195312, -4.4833221435546875, 4.55059814453125, 0.14543914794921875, 0.37677001953125, 10.564315795898438, 6.1876220703125, -1.043182373046875, 6.07421875, 9.8060302734375, -15.7637939453125, 2.3356246948242188, 3.3587265014648438, 3.2804393768310547, 15.755950927734375, -0.29901123046875, -7.66156005859375, 4.589397430419922, 0.87335205078125, 2.687856674194336, -4.943107604980469, 0.48187255859375, 2.01568603515625, 4.752357482910156, 2.5487213134765625, -4.606597900390625, 5.411590576171875, 3.5486679077148438, -2.7691802978515625, 8.260040283203125, 5.01446533203125, 5.18414306640625, 3.274169921875, -4.490287780761719, -9.299163818359375, -1.64154052734375, 6.565704345703125, -10.305511474609375, -2.5164794921875, -0.0855712890625, -6.27044677734375, 10.76165771484375, 8.816375732421875, 0.5155277252197266, 4.466026306152344, 2.416259765625, 9.75555419921875, 2.0884246826171875, 4.940460205078125, 15.4439697265625, -12.7818603515625, -0.8411865234375, 9.6610107421875, 4.220218658447266, -5.031005859375, -9.484817504882812, 1.3700180053710938, 4.564300537109375, 2.8642330169677734, 9.83294677734375, 2.229145050048828, -1.6897735595703125, 3.7321014404296875, -6.173309326171875, 4.148628234863281, -0.16923904418945312, 4.545806884765625, 1.1021442413330078, -1.19598388671875, -0.5381317138671875, 1.427764892578125, -6.7719879150390625, 2.0535736083984375, 8.53289794921875, 4.2995452880859375, -6.136993408203125, -7.4071044921875, -1.2066497802734375, 7.1703033447265625, -3.225067138671875, 2.824798583984375, 6.256843566894531, 0.9797592163085938, 14.914932250976562, -7.37896728515625, -1.4344482421875, 3.9212493896484375, -4.4814300537109375, 2.2938232421875, 3.435943603515625, -2.160665512084961, 2.74908447265625, 5.129968643188477], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000085.npy"}
{"epoch": 0.17801047120418848, "step": 86, "batch_size": 128, "mean": 3.743913412094116, "std": 7.646315097808838, "min": -16.883148193359375, "p10": -3.528268814086913, "median": 3.623427391052246, "p90": 13.174136352539062, "max": 25.005950927734375, "pos_frac": 0.6875, "sample": [4.6270294189453125, -10.915771484375, 3.690704345703125, -7.650146484375, 5.25830078125, 3.594879150390625, 4.542022705078125, 3.40924072265625, 7.22222900390625, 4.213226318359375, 25.005950927734375, -4.931854248046875, 0.7902259826660156, 1.94342041015625, 6.531040191650391, -11.071249008178711, -9.8341064453125, -3.1464996337890625, 1.0368576049804688, -0.42121124267578125, 1.8537750244140625, 5.713106155395508, -1.9708328247070312, -0.2637939453125, 16.274749755859375, -0.982940673828125, -15.41143798828125, 1.17596435546875, 16.794677734375, 12.294357299804688, -8.57171630859375, 11.306396484375, -0.6236572265625, 9.084716796875, 21.7911376953125, 11.733856201171875, 14.613067626953125, -4.141754150390625, 2.50299072265625, 12.50775146484375, 4.168190002441406, -1.95391845703125, 9.052696228027344, -1.7250900268554688, 3.4559707641601562, 1.0933151245117188, 8.800674438476562, 11.178802490234375, 11.442413330078125, 5.3745269775390625, -16.883148193359375, 11.6854248046875, 0.96673583984375, 7.623291015625, 5.09515380859375, -2.3548736572265625, 12.3729248046875, -16.24871826171875, 1.99713134765625, 9.998069763183594, 21.176971435546875, 4.532440185546875, 2.74688720703125, 11.902313232421875, 7.111724853515625, -1.561279296875, 0.0, -1.193511962890625, 3.1139984130859375, 6.2847900390625, 5.4418487548828125, 13.166290283203125, -12.094024658203125, 4.86151123046875, 2.1178970336914062, 1.6170654296875, -0.06956100463867188, 0.579498291015625, 10.12213134765625, 8.7811279296875, -2.0016403198242188, 8.062286376953125, 13.19244384765625, -0.15380859375, 4.987274169921875, -8.311996459960938, -11.222137451171875, 8.676712036132812, -3.2646484375, 0.584136962890625, -0.7949905395507812, 5.4673004150390625, 2.1012115478515625, 1.3066482543945312, 1.8809814453125, -2.05865478515625, 5.7177734375, 4.479736328125, 3.818634033203125, 3.651975631713867, -2.9693603515625, -2.76727294921875, -3.2653465270996094, 13.680145263671875, 15.15032958984375, 8.141265869140625, -1.2066917419433594, 22.966064453125, 2.1214752197265625, 4.46282958984375, -2.361431121826172, 6.482086181640625, 14.018081665039062, 5.504180908203125, 12.560882568359375, -1.527801513671875, 8.018997192382812, 15.63079833984375, -0.3119239807128906, 4.308563232421875, 6.204132080078125, 9.663532257080078, 9.908493041992188, 13.857742309570312, -1.566650390625, -0.43731689453125, 1.3271484375, 2.1563339233398438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000086.npy"}
{"epoch": 0.18010471204188483, "step": 87, "batch_size": 128, "mean": 3.9988515377044678, "std": 6.571066856384277, "min": -18.681854248046875, "p10": -2.308730316162109, "median": 3.460376739501953, "p90": 13.658328247070312, "max": 24.544952392578125, "pos_frac": 0.7109375, "sample": [5.211601257324219, 8.263206481933594, 6.484649658203125, 8.392822265625, 15.914230346679688, 0.0, -0.050201416015625, -2.5537185668945312, 9.036331176757812, 15.0076904296875, -4.553314208984375, 7.2408905029296875, -2.2037353515625, 9.283472061157227, 24.544952392578125, -0.197998046875, 0.5966911315917969, 7.733184814453125, 15.363082885742188, 4.6881866455078125, -1.8067626953125, -7.7823638916015625, 0.059844970703125, 4.918975830078125, 4.221923828125, -2.0962295532226562, -1.38592529296875, 6.08544921875, 16.801055908203125, 0.8666744232177734, 9.329498291015625, -7.132568359375, 6.592041015625, 15.2230224609375, 5.1291656494140625, 4.040283203125, 4.565879821777344, 3.9241943359375, 4.736724853515625, 2.1871337890625, -0.8563117980957031, -0.44110107421875, 7.06884765625, -1.628936767578125, -2.1411666870117188, -0.86553955078125, 9.840621948242188, -4.388763427734375, 10.45184326171875, 2.533355712890625, 7.523307800292969, 0.12596893310546875, 14.4158935546875, 15.44403076171875, 11.55828857421875, 7.85809326171875, 5.99169921875, 7.47027587890625, 12.2373046875, 18.464996337890625, 0.23687744140625, 0.6850433349609375, 2.091585159301758, 10.024787902832031, 3.413970947265625, 21.4302978515625, -0.592620849609375, 0.0, 4.2668609619140625, 3.556732177734375, 4.125373840332031, 5.56048583984375, 13.20086669921875, 4.486297607421875, 13.600128173828125, 3.2467041015625, 3.9326324462890625, -0.8473892211914062, 9.055450439453125, -0.24898910522460938, -6.06060791015625, -2.642242431640625, -1.7156982421875, 0.723419189453125, -1.0301589965820312, 9.8275146484375, -2.78448486328125, 6.1668548583984375, 3.5067825317382812, 7.7265625, 4.4979095458984375, 1.4715499877929688, 2.7459259033203125, -1.65966796875, 0.36578369140625, -9.922119140625, 1.4434432983398438, 1.2233142852783203, -6.631988525390625, 13.79412841796875, -0.594757080078125, 3.579315185546875, -0.10022354125976562, -0.913604736328125, 12.61297607421875, -18.681854248046875, 4.3692474365234375, -5.6613006591796875, 0.24537277221679688, 2.1734695434570312, 11.66314697265625, 15.477294921875, -0.6931304931640625, 8.20147705078125, -0.9993896484375, 0.121917724609375, 10.010009765625, 2.0581893920898438, 2.0010986328125, 0.6138534545898438, 3.961883544921875, 0.4161529541015625, -2.8024444580078125, 2.9250564575195312, 6.069976806640625, 0.2005767822265625, 14.627853393554688, 1.3567543029785156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000087.npy"}
{"epoch": 0.18219895287958116, "step": 88, "batch_size": 128, "mean": 3.2320239543914795, "std": 7.375482559204102, "min": -28.01806640625, "p10": -3.9646903991699216, "median": 2.160125732421875, "p90": 12.58150177001953, "max": 26.796905517578125, "pos_frac": 0.6640625, "sample": [9.276947021484375, 13.700653076171875, 11.19427490234375, 0.3178138732910156, 17.574951171875, 2.2804718017578125, -3.188507080078125, -0.144256591796875, 8.671768188476562, 2.6067276000976562, 1.7475948333740234, -3.9062957763671875, 1.71575927734375, 21.68206787109375, -0.5832405090332031, -0.62908935546875, -0.14389419555664062, 8.9605712890625, 0.5760116577148438, -4.03466796875, 12.895477294921875, -4.603118896484375, 8.8343505859375, -4.530113220214844, -0.4766845703125, 6.1360321044921875, -5.346160888671875, 1.6301651000976562, 15.26019287109375, -7.132080078125, -5.3458251953125, 9.637664794921875, 5.0978851318359375, 6.7655029296875, 2.7453079223632812, -1.4636421203613281, 18.220001220703125, 0.0, 0.5266170501708984, 1.29705810546875, 5.954071044921875, 5.838165283203125, 10.49407958984375, -1.597320556640625, 12.741195678710938, 14.305755615234375, 2.916015625, -6.3806610107421875, 2.5754127502441406, 3.3254852294921875, 0.0, 3.054443359375, 2.10791015625, 6.77728271484375, 11.992523193359375, 0.4857025146484375, -0.9186820983886719, 0.69110107421875, -1.6421661376953125, 4.678741455078125, 10.191062927246094, -0.7731170654296875, 0.638916015625, -6.552742004394531, 5.263145446777344, 2.029296875, -3.9347000122070312, 10.2227783203125, -0.15027618408203125, 10.91241455078125, 3.2856101989746094, -4.525726318359375, 13.244140625, -0.16953277587890625, -1.65985107421875, -20.608428955078125, 3.859283447265625, 1.7145805358886719, 0.802093505859375, -2.6578216552734375, 7.5966339111328125, 6.969573974609375, 7.479248046875, -1.702463150024414, 0.673370361328125, -3.06201171875, -2.2973556518554688, -7.2139434814453125, 6.525459289550781, 7.036949157714844, 0.2995777130126953, 5.539306640625, 2.4709320068359375, 3.3287353515625, 4.978240966796875, 15.303375244140625, 1.61602783203125, -1.9147796630859375, -2.1858062744140625, 20.1763916015625, 2.503082275390625, 3.54669189453125, -1.5628433227539062, 9.897216796875, -1.409942626953125, 26.796905517578125, 12.5130615234375, -0.3192605972290039, 9.7518310546875, -8.042877197265625, 8.99029541015625, 0.950897216796875, 1.89251708984375, 2.21234130859375, 22.119140625, 4.127702713012695, -1.3155384063720703, 0.19131088256835938, 2.3526458740234375, 1.08837890625, -28.01806640625, 7.320220947265625, -2.0999755859375, 7.803688049316406, 2.5035171508789062, 8.301727294921875, 4.016815185546875, -0.38434600830078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000088.npy"}
{"epoch": 0.18429319371727748, "step": 89, "batch_size": 128, "mean": 3.875415325164795, "std": 6.576765537261963, "min": -18.15350341796875, "p10": -2.9933494567871093, "median": 3.295318603515625, "p90": 12.145733642578124, "max": 24.86920166015625, "pos_frac": 0.7890625, "sample": [0.12181472778320312, -11.406341552734375, 12.29241943359375, 1.9771728515625, 5.985809326171875, 15.932403564453125, -1.598928451538086, 1.7044677734375, 2.10076904296875, 8.576019287109375, 1.46142578125, 4.390350341796875, 2.95928955078125, 11.95257568359375, 3.065399169921875, 11.850372314453125, 0.9864826202392578, -3.12628173828125, 5.1387939453125, 3.22332763671875, 5.5032958984375, 13.602935791015625, -2.9363784790039062, 0.6144542694091797, 1.3338699340820312, 1.870269775390625, 4.0213623046875, -0.91961669921875, -1.1986083984375, 3.5121383666992188, 3.5965423583984375, 0.6415023803710938, -0.5513916015625, 12.08477783203125, 14.176971435546875, 23.39666748046875, 12.2879638671875, -7.3819580078125, 1.5587005615234375, 3.3673095703125, -18.15350341796875, -2.649078369140625, 5.1329345703125, -9.339691162109375, 0.8681640625, 4.623138427734375, 7.819732666015625, -11.861419677734375, 0.603302001953125, -2.4847412109375, -0.6018714904785156, 8.096237182617188, 3.159423828125, 0.0980377197265625, 9.4835205078125, 10.69189453125, 5.3366851806640625, -3.5844650268554688, 6.873748779296875, 10.371124267578125, 1.8400535583496094, 14.863677978515625, 8.434295654296875, 7.586761474609375, 5.72119140625, 1.594757080078125, -6.3277130126953125, 2.225830078125, 1.1655845642089844, -3.79327392578125, 4.418724060058594, 6.65618896484375, 4.1167449951171875, 5.9212188720703125, -5.1566314697265625, -0.1323699951171875, 3.521961212158203, 6.856353759765625, 19.60699462890625, -3.870086669921875, 1.82080078125, 5.6883544921875, 2.5938262939453125, 10.283493041992188, 1.3380279541015625, 11.475250244140625, 2.904693603515625, 15.32708740234375, 9.582107543945312, 20.11419677734375, 5.681541442871094, -0.7734375, 12.5675048828125, 2.2172622680664062, 2.6713218688964844, 3.0955123901367188, 24.86920166015625, -1.56866455078125, 2.0264968872070312, 3.906291961669922, 7.89447021484375, 2.08172607421875, 3.5047454833984375, -0.4114990234375, 3.676410675048828, 4.76025390625, -1.9456634521484375, -1.4418792724609375, 1.624267578125, 4.77410888671875, 4.1423492431640625, 2.969493865966797, 10.060081481933594, 11.820098876953125, 5.016845703125, 4.136344909667969, 4.854347229003906, 3.654052734375, 8.27752685546875, -12.852691650390625, 1.2216377258300781, 12.354339599609375, 5.6492156982421875, 3.0906829833984375, 1.3803291320800781, 4.22698974609375, 3.0207977294921875, -3.20867919921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000089.npy"}
{"epoch": 0.18638743455497384, "step": 90, "batch_size": 128, "mean": 3.417588710784912, "std": 7.220695495605469, "min": -19.91851806640625, "p10": -4.009095001220703, "median": 2.4347457885742188, "p90": 12.3794189453125, "max": 23.549652099609375, "pos_frac": 0.7109375, "sample": [8.189216613769531, 4.873748779296875, 2.33453369140625, 18.242034912109375, -10.411651611328125, -1.05548095703125, -3.41229248046875, 1.3539886474609375, -1.376953125, 11.22296142578125, 1.9622554779052734, 2.0983734130859375, -7.86761474609375, -0.48163604736328125, -1.3144073486328125, 6.590728759765625, 22.319915771484375, 18.646942138671875, 13.7574462890625, 1.0333023071289062, 7.06884765625, -0.382598876953125, 0.9975128173828125, 2.8051834106445312, -12.924163818359375, -3.791229248046875, 11.007083892822266, 6.856292724609375, 6.1951904296875, 1.7922210693359375, -1.337127685546875, 6.787117004394531, 6.990447998046875, 2.86187744140625, 7.511077880859375, 1.3002548217773438, -4.37103271484375, -4.278564453125, 1.59515380859375, 5.823997497558594, -5.38957405090332, 2.0349273681640625, -0.0903778076171875, -9.20947265625, 7.3856201171875, 11.552001953125, 4.597114562988281, 8.968109130859375, 6.125, 2.47991943359375, 8.117965698242188, -3.8936080932617188, 0.536407470703125, 1.7257251739501953, 1.797271728515625, 1.2904434204101562, -11.126678466796875, 0.13709259033203125, 0.2813720703125, 5.717887878417969, 0.982940673828125, 0.0, 12.329925537109375, 6.8024444580078125, 9.377578735351562, 0.106842041015625, 7.642303466796875, 3.5456008911132812, -1.83209228515625, 12.80426025390625, 1.820892333984375, 1.36627197265625, 6.8638916015625, 17.39691162109375, 10.993621826171875, 0.51763916015625, 0.4741678237915039, 3.544921875, 2.9038009643554688, 8.2088623046875, 8.83544921875, -19.91851806640625, 10.24371337890625, -10.506317138671875, 0.1831207275390625, 3.57080078125, 4.367454528808594, 2.3895721435546875, -0.8792762756347656, -2.599761962890625, -0.5186767578125, -3.433837890625, 11.251007080078125, 12.494903564453125, -2.753509521484375, 2.88922119140625, -3.241973876953125, 5.941558837890625, 8.980194091796875, 10.153305053710938, -1.4854736328125, 12.044204711914062, 8.2430419921875, 0.4093017578125, -2.2183074951171875, 23.549652099609375, -2.5437164306640625, 4.7332305908203125, -5.48248291015625, -12.83538818359375, 5.3004608154296875, 0.68865966796875, -7.296112060546875, 12.21923828125, 2.702117919921875, -0.5042152404785156, 13.47454833984375, 1.853424072265625, 15.113754272460938, 2.529998779296875, -0.9140777587890625, -2.552675247192383, 19.884765625, 13.836944580078125, 3.4189224243164062, 13.067550659179688, 2.53155517578125, 12.13311767578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000090.npy"}
{"epoch": 0.18848167539267016, "step": 91, "batch_size": 128, "mean": 3.350696086883545, "std": 7.6048383712768555, "min": -12.979537963867188, "p10": -6.133451080322265, "median": 3.1015052795410156, "p90": 13.5757080078125, "max": 26.729949951171875, "pos_frac": 0.625, "sample": [2.9588623046875, 1.1939697265625, -0.6389808654785156, 16.215713500976562, 9.2666015625, 14.547149658203125, -0.050212860107421875, 3.2441482543945312, 6.0301055908203125, -3.5834274291992188, 0.7463836669921875, 5.8009185791015625, 6.6695098876953125, 6.102195739746094, 6.192253112792969, 8.743377685546875, 1.657440185546875, -12.979537963867188, -3.601654052734375, -0.25530242919921875, 8.410110473632812, 4.6060333251953125, 5.863250732421875, 7.06146240234375, -1.8189315795898438, 6.3422698974609375, 0.01739501953125, -0.373992919921875, -0.8230857849121094, -7.2996063232421875, 14.013519287109375, -0.8349838256835938, -1.69842529296875, -6.54718017578125, -3.252532958984375, 9.672943115234375, -5.412117004394531, 6.8525390625, 5.342506408691406, 6.6064453125, -8.804046630859375, -1.0828399658203125, 3.964202880859375, -2.580535888671875, 4.49871826171875, 17.63800048828125, -6.71575927734375, -4.8016357421875, 0.0, -10.78778076171875, -5.956138610839844, -4.635887145996094, -11.7353515625, 2.2308349609375, 6.1363677978515625, 5.598480224609375, -0.591217041015625, -5.8167724609375, 14.44757080078125, 5.452362060546875, 10.81280517578125, -9.124542236328125, 4.111328125, 0.6338481903076172, -2.749624252319336, -0.569061279296875, 22.472900390625, -1.468017578125, -0.81341552734375, 10.96893310546875, -7.8065185546875, -2.796539306640625, 26.729949951171875, 3.826292037963867, 13.519195556640625, 1.0985107421875, 4.49267578125, 0.6258468627929688, 3.575531005859375, 13.530059814453125, -9.385528564453125, -5.517059326171875, 21.193206787109375, 10.0943603515625, 13.08074951171875, 0.0, 9.416465759277344, 5.3199005126953125, 6.91552734375, 12.3880615234375, 3.6336669921875, 10.181610107421875, 4.08563232421875, -6.73828125, -0.422882080078125, 13.682220458984375, -8.801795959472656, 2.272350311279297, 3.5233154296875, 20.86737060546875, 7.3299560546875, 9.1695556640625, 7.971282958984375, 4.504486083984375, 4.985809326171875, -9.042953491210938, -1.8072509765625, 11.955879211425781, -0.344970703125, 15.094482421875, -3.841217041015625, 18.246047973632812, 19.60186767578125, -0.28878211975097656, -1.659393310546875, 9.254806518554688, 1.72662353515625, -3.074127197265625, 7.170501708984375, 0.8233184814453125, 8.547225952148438, -4.0185546875, 1.91107177734375, 1.741323471069336, 5.06341552734375, 1.852264404296875, 6.7645263671875, 0.947113037109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000091.npy"}
{"epoch": 0.1905759162303665, "step": 92, "batch_size": 128, "mean": 3.6277565956115723, "std": 7.883093357086182, "min": -28.476806640625, "p10": -5.510736083984375, "median": 3.5069198608398438, "p90": 13.476356506347654, "max": 21.813446044921875, "pos_frac": 0.7265625, "sample": [-6.382293701171875, 0.10434722900390625, -3.332916259765625, 3.75653076171875, 2.8835983276367188, 5.843986511230469, 5.002166748046875, 5.159332275390625, 12.654754638671875, 7.98370361328125, 1.612335205078125, 2.097047805786133, 3.9912109375, -0.350677490234375, 18.279327392578125, 19.871826171875, 19.109954833984375, -8.3243408203125, 7.44677734375, 1.5820388793945312, -6.0631866455078125, 6.605194091796875, 14.065902709960938, 12.2254638671875, 3.3604736328125, -2.6151123046875, 3.0750274658203125, 11.877838134765625, 2.12078857421875, -2.2220916748046875, 2.4033203125, 0.9235000610351562, 12.25872802734375, 7.5723114013671875, 0.8587799072265625, 0.3601531982421875, 0.580352783203125, 4.0309295654296875, 14.166961669921875, 9.60174560546875, 1.0381011962890625, 1.12603759765625, 0.09926414489746094, -28.476806640625, 10.737457275390625, 2.5573883056640625, 4.38616943359375, 2.74334716796875, 0.789337158203125, 1.5308456420898438, -6.032318115234375, -8.306640625, -9.7684326171875, -1.9405517578125, 18.00653076171875, -5.05743408203125, -10.869964599609375, 0.23726272583007812, -7.618316650390625, 8.34051513671875, 5.679542541503906, 5.507179260253906, 5.96282958984375, 5.5639801025390625, 6.26654052734375, 4.525733947753906, 9.169475555419922, 0.0, -1.92572021484375, 11.2701416015625, -2.353607177734375, 3.78741455078125, 1.5074462890625, -0.9469165802001953, -2.1097259521484375, 7.9578704833984375, -1.2316246032714844, 0.6897048950195312, 10.175857543945312, -1.1656227111816406, -7.88922119140625, 13.22369384765625, 5.7767486572265625, -20.995269775390625, -2.237579345703125, 1.00445556640625, 5.9212646484375, 20.3031005859375, 20.2774658203125, 5.268043518066406, 7.719062805175781, 6.3618621826171875, -0.8757171630859375, 0.1348114013671875, -0.14901256561279297, 21.813446044921875, -0.512939453125, -5.287200927734375, 19.683135986328125, 4.683479309082031, 9.48529052734375, 9.544891357421875, -11.2799072265625, 14.107192993164062, 7.95391845703125, 5.1906890869140625, 1.855682373046875, 3.6533660888671875, 2.6746673583984375, 4.625152587890625, 8.495025634765625, 4.4317474365234375, 9.621940612792969, -11.42724609375, 4.7481689453125, 9.1400146484375, -0.1194610595703125, 1.12890625, -1.45281982421875, 7.617279052734375, -2.8533935546875, 9.060422897338867, 3.787443161010742, 2.3770694732666016, 16.2127685546875, 17.352203369140625, 9.693145751953125, -1.52301025390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000092.npy"}
{"epoch": 0.19267015706806281, "step": 93, "batch_size": 128, "mean": 3.441769599914551, "std": 7.469954967498779, "min": -17.487274169921875, "p10": -5.078096771240234, "median": 2.4392852783203125, "p90": 14.633045196533203, "max": 24.3834228515625, "pos_frac": 0.6875, "sample": [0.8782081604003906, 13.7847900390625, 1.9539947509765625, 23.10101318359375, -5.564178466796875, 8.697265625, -3.1672210693359375, 11.57977294921875, 2.859375, 5.4275665283203125, 3.734527587890625, -8.611892700195312, 0.24835205078125, 0.407562255859375, 3.638214111328125, 11.88507080078125, 1.4332275390625, 15.861618041992188, -2.8977432250976562, 4.698326110839844, 5.468536376953125, 5.2924957275390625, 3.1567935943603516, 3.4238433837890625, 14.609092712402344, 1.509979248046875, -0.0797271728515625, 2.42193603515625, 13.30914306640625, 10.775970458984375, 2.3260421752929688, -7.940643310546875, 0.16533660888671875, -0.9774627685546875, 2.5402145385742188, 5.9934844970703125, 4.764385223388672, 4.0589599609375, -3.451507568359375, 14.688934326171875, 4.0740509033203125, 6.485206604003906, 2.512969970703125, -5.0493927001953125, -0.7244949340820312, 3.9645118713378906, 0.628448486328125, 14.9849853515625, 2.456634521484375, -0.7774124145507812, 1.8293304443359375, 5.81207275390625, 17.271331787109375, -7.394622802734375, 7.58203125, 1.042755126953125, -0.9398574829101562, -0.926513671875, 9.199493408203125, 0.6434097290039062, 7.5134429931640625, 4.1214599609375, 15.866455078125, -2.34619140625, -2.9327239990234375, -7.14849853515625, 9.922943115234375, 16.52081298828125, -7.0750579833984375, 0.8684234619140625, 3.4114818572998047, 3.0180435180664062, -4.107414245605469, -1.6062240600585938, 7.46563720703125, -2.970367431640625, 13.964752197265625, -0.4644927978515625, 19.20904541015625, -17.487274169921875, 20.83740234375, 7.42132568359375, 15.3297119140625, -6.967041015625, 5.653564453125, 1.1382598876953125, 0.0, -10.13751220703125, 2.037914276123047, 4.839927673339844, 15.724090576171875, 12.54498291015625, -3.841888427734375, 2.2317047119140625, 6.184574127197266, -15.89703369140625, 7.22015380859375, -0.9800491333007812, -1.20703125, -6.5051116943359375, 9.381927490234375, 6.9073944091796875, 2.679229736328125, 2.9410400390625, 4.2750396728515625, 0.0, 1.4540824890136719, -1.7722244262695312, -5.145072937011719, -0.5012321472167969, -0.0429840087890625, 0.9970321655273438, 5.437126159667969, 2.2596588134765625, -10.9459228515625, 21.90155029296875, 24.3834228515625, -1.5337333679199219, -0.9750137329101562, -1.658233642578125, 5.12213134765625, 4.534210205078125, 1.16546630859375, 14.16766357421875, 0.39870262145996094, 2.2073974609375, 0.49077606201171875, 6.366302490234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000093.npy"}
{"epoch": 0.19476439790575917, "step": 94, "batch_size": 128, "mean": 4.082524299621582, "std": 9.121895790100098, "min": -20.28363037109375, "p10": -7.644261169433594, "median": 2.744771957397461, "p90": 15.17275695800781, "max": 30.26300048828125, "pos_frac": 0.7109375, "sample": [-1.6839752197265625, 10.435691833496094, -7.5614471435546875, 17.92779541015625, 4.962127685546875, 10.626434326171875, 20.502105712890625, 2.936016082763672, 5.326873779296875, 9.555862426757812, 0.35955810546875, -5.394866943359375, -14.1309814453125, -0.4390106201171875, 11.958770751953125, -0.12237548828125, 13.40667724609375, 0.4063148498535156, 0.50140380859375, 0.4220123291015625, 14.544158935546875, 14.49066162109375, 4.178466796875, -2.7776412963867188, 19.151840209960938, 14.76153564453125, 2.4993858337402344, 4.979705810546875, -9.205291748046875, -1.88250732421875, 12.22357177734375, 23.7364501953125, -10.218048095703125, 22.95648193359375, 7.383678436279297, 6.787384033203125, 8.3472900390625, 1.872457504272461, -6.648658752441406, 20.958251953125, 5.085685729980469, 30.26300048828125, 14.9515380859375, -0.5306549072265625, 13.16607666015625, 3.06695556640625, 14.045181274414062, -6.127288818359375, 7.908088684082031, -2.438488006591797, 1.4851455688476562, 0.3476104736328125, -0.9618377685546875, 5.9222412109375, 3.0699386596679688, -6.39117431640625, 7.1698455810546875, -1.717498779296875, 8.3475341796875, -8.93206787109375, 0.54351806640625, -5.28076171875, 1.8653640747070312, -3.6737060546875, 2.55352783203125, 12.903411865234375, 15.688934326171875, 1.1171035766601562, -6.680931091308594, 13.427337646484375, 0.15093994140625, 0.8462486267089844, 7.516471862792969, 12.58612060546875, 18.00042724609375, -6.533905029296875, 5.59088134765625, -1.6360282897949219, 4.00750732421875, 4.2667236328125, 3.1436767578125, 1.985626220703125, -5.5312042236328125, -4.703887939453125, -12.161399841308594, 22.475738525390625, 12.94427490234375, -9.04559326171875, 16.586883544921875, -10.22418212890625, -20.28363037109375, 3.986083984375, 2.518911361694336, 8.869430541992188, 4.2920989990234375, 13.916816711425781, 10.1444091796875, 7.9751739501953125, 18.32501220703125, -7.837493896484375, 7.086029052734375, -8.1138916015625, 12.54974365234375, 1.5564117431640625, 2.24017333984375, 7.1827392578125, 2.4894676208496094, 9.919692993164062, 1.0406494140625, 1.4612579345703125, 3.95745849609375, 5.4927825927734375, 0.82061767578125, -17.1793212890625, 1.4293212890625, -1.1793212890625, 8.957061767578125, 1.715850830078125, 0.6912002563476562, -8.521881103515625, 9.288909912109375, 1.499267578125, 26.416259765625, 11.949089050292969, -10.329147338867188, 1.22283935546875, -0.1566448211669922, -1.423431396484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000094.npy"}
{"epoch": 0.1968586387434555, "step": 95, "batch_size": 128, "mean": 4.461293697357178, "std": 9.053359031677246, "min": -23.30303955078125, "p10": -6.2954666137695305, "median": 3.955760955810547, "p90": 16.243751525878903, "max": 26.882965087890625, "pos_frac": 0.703125, "sample": [-3.1443023681640625, 0.50634765625, 3.3492279052734375, 23.800628662109375, 9.72412109375, 1.971435546875, 1.1650848388671875, 5.33843994140625, 25.329071044921875, 7.038787841796875, -0.568267822265625, -0.094085693359375, 3.1668701171875, 3.0221633911132812, -8.817642211914062, 6.63232421875, 3.2707977294921875, 6.4160003662109375, 12.037277221679688, 1.7606735229492188, 13.447662353515625, 7.293235778808594, -9.85968017578125, 7.54107666015625, -1.1015625, -0.0767364501953125, 26.882965087890625, -4.302642822265625, 15.6981201171875, -23.30303955078125, 20.26776123046875, -1.0211181640625, 10.064346313476562, -6.8703155517578125, 25.59869384765625, 10.51007080078125, 1.25897216796875, 7.6483612060546875, 3.7981719970703125, 9.744873046875, 3.5601654052734375, 6.32879638671875, -3.27777099609375, -0.2511005401611328, -0.3466033935546875, 8.697525024414062, 4.4459075927734375, 9.171951293945312, 22.49676513671875, 2.8540802001953125, -1.1827926635742188, 9.34423828125, 5.00311279296875, -0.3557395935058594, 6.79010009765625, 6.999603271484375, 4.006477355957031, 3.7411575317382812, 5.22222900390625, 10.28692626953125, 7.74810791015625, -2.3321533203125, 2.5452880859375, 1.280609130859375, 3.678863525390625, 7.797119140625, -5.432830810546875, 10.497039794921875, 5.587303161621094, 3.9037952423095703, -6.049102783203125, 18.6708984375, 8.51776123046875, 17.516891479492188, -0.13189697265625, -0.9949417114257812, 0.12486839294433594, 15.225372314453125, 2.844940185546875, 6.044281005859375, -8.752151489257812, 9.880413055419922, 4.9827117919921875, 1.6225090026855469, 4.92193603515625, 24.1636962890625, 6.7090911865234375, 9.050888061523438, -0.867218017578125, -15.185821533203125, 4.8376922607421875, 8.0902099609375, -3.003997802734375, 1.6954269409179688, -13.99127197265625, 2.573822021484375, 8.3514404296875, 8.142852783203125, -9.511871337890625, 5.03436279296875, -3.4849700927734375, -2.9092540740966797, -5.755096435546875, 1.6294174194335938, 11.0892333984375, -8.833915710449219, 5.4067840576171875, -0.7614669799804688, 24.0003662109375, -10.116973876953125, 3.9050445556640625, 4.61492919921875, -0.906951904296875, 10.4896240234375, 0.572265625, 14.44866943359375, 0.378631591796875, 9.92706298828125, 6.24053955078125, -14.3189697265625, -8.124969482421875, -0.4187049865722656, 6.4907989501953125, 20.850067138671875, 4.0112762451171875, 26.341552734375, -12.2276611328125, 24.0621337890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000095.npy"}
{"epoch": 0.19895287958115182, "step": 96, "batch_size": 128, "mean": 4.274379253387451, "std": 8.278307914733887, "min": -19.3035888671875, "p10": -5.002455139160156, "median": 3.8083839416503906, "p90": 15.781875610351559, "max": 27.24078369140625, "pos_frac": 0.703125, "sample": [-0.44427490234375, -8.361724853515625, -8.4857177734375, 1.9512863159179688, 16.6031494140625, 8.135459899902344, -13.286794662475586, -6.6319580078125, 1.429168701171875, 0.8425445556640625, 3.606203079223633, -4.847991943359375, -5.3628692626953125, -0.61773681640625, 7.9351959228515625, 1.8435707092285156, 4.574920654296875, -5.7713470458984375, 1.1998138427734375, 4.4467926025390625, 11.35357666015625, 1.794036865234375, 9.1729736328125, 6.874053955078125, -3.254364013671875, 5.01690673828125, 18.220718383789062, 10.92059326171875, 2.4965896606445312, 4.1268310546875, 26.721298217773438, 5.30499267578125, 11.09197998046875, 11.228202819824219, 6.8343505859375, 4.229240417480469, 3.7471084594726562, 2.9818267822265625, 1.5944061279296875, 1.7964630126953125, 6.791015625, 4.440948486328125, -2.676788330078125, 19.87969970703125, -6.610107421875, 5.547080993652344, -7.800270080566406, -0.41463279724121094, 4.366004943847656, 0.289398193359375, 5.136749267578125, -2.297576904296875, 8.99658203125, -3.544281005859375, 3.9874267578125, -1.75860595703125, 3.9505844116210938, 13.342605590820312, 4.088226318359375, 12.8724365234375, -1.0260086059570312, 3.869659423828125, 11.448974609375, 9.832305908203125, 5.918731689453125, 20.3599853515625, 27.24078369140625, 3.49359130859375, -2.1949081420898438, 15.429901123046875, 4.909576416015625, 15.313240051269531, 2.2986373901367188, -11.806961059570312, 23.0068359375, 5.14190673828125, 8.967864990234375, -2.34075927734375, 18.66510009765625, 0.474761962890625, 8.325225830078125, 1.15087890625, 5.508880615234375, -6.9864501953125, 0.2415008544921875, -1.8778877258300781, 0.6451416015625, 11.519866943359375, 17.695220947265625, -0.358551025390625, -0.9857444763183594, -0.8898239135742188, 4.707237243652344, 18.136871337890625, 19.5072021484375, -6.433010101318359, 3.6419677734375, 7.34039306640625, 10.593667984008789, 14.405120849609375, -15.628997802734375, 20.61541748046875, 2.260467529296875, -1.9140033721923828, -0.8337173461914062, 10.79144287109375, -0.46099853515625, 26.41455078125, 0.5715484619140625, 4.0250244140625, 9.68292236328125, 0.0, 1.5102386474609375, 2.305278778076172, 0.7744426727294922, 4.288490295410156, -3.354705810546875, 8.4365234375, 5.1217041015625, 8.8201904296875, -3.5390987396240234, 7.732975006103516, -2.9995460510253906, -1.786407470703125, 2.4698486328125, 7.974822998046875, -1.3411712646484375, -19.3035888671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000096.npy"}
{"epoch": 0.20104712041884817, "step": 97, "batch_size": 128, "mean": 4.458014488220215, "std": 9.960394859313965, "min": -17.22552490234375, "p10": -7.711051177978516, "median": 3.6914215087890625, "p90": 16.98713684082031, "max": 30.485595703125, "pos_frac": 0.6796875, "sample": [5.310737609863281, 3.60052490234375, 6.9696197509765625, -6.2799224853515625, 7.2698974609375, 2.9778976440429688, -1.4095840454101562, 5.474212646484375, 7.558349609375, 8.580657958984375, -1.1861743927001953, 30.485595703125, 8.311279296875, 15.924407958984375, 0.39337921142578125, 4.445556640625, 16.502777099609375, -0.813323974609375, -4.984619140625, 12.798507690429688, 3.7955322265625, 2.5444717407226562, -2.901702880859375, -0.602264404296875, 8.720062255859375, -7.8089599609375, 3.782318115234375, -5.446266174316406, 8.21868896484375, 9.198966979980469, -7.625572204589844, 0.5152816772460938, -4.1039886474609375, -6.5770721435546875, 7.921478271484375, 25.414840698242188, 10.524681091308594, -4.1633453369140625, -6.7022552490234375, -10.27020263671875, 1.50030517578125, 1.883056640625, 3.3858795166015625, 23.3658447265625, -8.368133544921875, 5.743896484375, 4.57635498046875, 9.073558807373047, 1.253082275390625, 9.108154296875, 12.080718994140625, -6.710357666015625, 26.64190673828125, -0.24798583984375, 8.653182983398438, 11.695526123046875, -8.116943359375, 8.55572509765625, 0.921173095703125, -2.0501708984375, 9.787635803222656, 6.244354248046875, -1.1673812866210938, 7.779455184936523, 4.991851806640625, 0.0, 27.7896728515625, 8.35009765625, 8.93572998046875, 30.111968994140625, 0.6690044403076172, 12.388656616210938, 9.498428344726562, 6.892578125, -0.7021312713623047, 26.9095458984375, 2.8565673828125, -14.423599243164062, -7.669090270996094, 18.5390625, 8.189453125, 0.8281784057617188, -6.114532470703125, 3.589324951171875, 11.951171875, 5.810474395751953, -0.15319252014160156, 29.301788330078125, 4.683967590332031, -4.2783660888671875, 7.415760040283203, -10.08941650390625, 6.7294921875, 1.1044540405273438, 6.091400146484375, -17.22552490234375, 0.0, 7.259521484375, 12.617835998535156, 0.0720977783203125, 2.336639404296875, 29.92205810546875, -10.0123291015625, 2.2503509521484375, 3.1008758544921875, 2.12872314453125, -2.5486602783203125, -8.7462158203125, 26.9859619140625, 5.629638671875, -5.3160400390625, 24.850067138671875, -7.909511566162109, -11.326202392578125, 0.4288330078125, 11.178741455078125, 3.1570587158203125, 6.101593017578125, -0.6446990966796875, -13.07672119140625, 9.856430053710938, -4.4518890380859375, 1.369415283203125, 13.48681640625, 18.1173095703125, -13.64031982421875, 3.999969482421875, 8.5224609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000097.npy"}
{"epoch": 0.2031413612565445, "step": 98, "batch_size": 128, "mean": 4.760381698608398, "std": 10.292726516723633, "min": -27.0728759765625, "p10": -5.7648773193359375, "median": 3.7085723876953125, "p90": 17.140380859375, "max": 33.04150390625, "pos_frac": 0.671875, "sample": [12.4180908203125, 12.321243286132812, 23.81903076171875, 3.052888870239258, 32.704498291015625, 17.02117919921875, 15.314041137695312, -3.920166015625, 3.280242919921875, -1.4480743408203125, 15.200958251953125, -12.261932373046875, -3.190826416015625, 5.6822052001953125, 7.0737762451171875, 9.03857421875, 1.6639862060546875, -13.3746337890625, -6.421333312988281, 1.8367233276367188, 12.8035888671875, -4.975921630859375, -6.291046142578125, 3.702056884765625, 2.5391616821289062, 3.9005489349365234, -0.1994171142578125, -4.17547607421875, -6.2332763671875, -18.47943115234375, 2.022409439086914, -27.0728759765625, -4.949432373046875, -6.52337646484375, 3.6682968139648438, -1.0985107421875, -5.7299041748046875, 14.655609130859375, 3.4131317138671875, 3.2108688354492188, 2.5216064453125, 7.5460205078125, 1.10552978515625, 7.2252197265625, -1.3377227783203125, 15.86981201171875, 1.240142822265625, 17.617645263671875, -5.8464813232421875, 13.748027801513672, 31.84002685546875, 16.46612548828125, -1.3693389892578125, 3.715087890625, -0.3367424011230469, -0.932861328125, 29.18505859375, 13.779510498046875, -2.33074951171875, 24.296539306640625, -16.4842529296875, -4.187309265136719, -1.532012939453125, -5.30120849609375, 25.174896240234375, -2.111846923828125, 10.272064208984375, 3.765106201171875, 6.343757629394531, -2.2083740234375, 8.321983337402344, 15.94927978515625, -14.26971435546875, 2.5602264404296875, 20.102951049804688, 14.079635620117188, 6.20465087890625, 12.84490966796875, 0.882537841796875, 15.874359130859375, 5.752349853515625, 18.272491455078125, -9.778656005859375, -1.2249641418457031, 16.3404541015625, -4.3999481201171875, 0.34001922607421875, 1.7434921264648438, 3.4207763671875, -0.03546142578125, 21.942138671875, 7.9124755859375, 5.130592346191406, -1.6946868896484375, 7.649391174316406, 8.068084716796875, 0.237457275390625, 7.661155700683594, 2.9805374145507812, 5.564544677734375, 1.194610595703125, 16.307846069335938, 33.04150390625, 4.550018310546875, 6.640106201171875, -5.435646057128906, -3.3540802001953125, 5.099273681640625, 5.220527648925781, 5.7182464599609375, -1.3414306640625, 4.467498779296875, 4.380584716796875, 6.102121353149414, -1.2748298645019531, 8.346664428710938, 1.794189453125, -2.423919677734375, 7.251701354980469, 12.108978271484375, 12.503959655761719, -4.257598876953125, 17.41851806640625, 7.056549072265625, 20.183441162109375, 4.0150604248046875, -17.662567138671875, 11.513702392578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000098.npy"}
{"epoch": 0.20523560209424083, "step": 99, "batch_size": 128, "mean": 5.110473155975342, "std": 10.606465339660645, "min": -23.80401611328125, "p10": -7.761041259765625, "median": 4.64224910736084, "p90": 18.60758361816406, "max": 27.13739013671875, "pos_frac": 0.7265625, "sample": [-17.697265625, 0.59527587890625, -11.965301513671875, 27.13739013671875, 3.8049087524414062, 12.3095703125, -0.6575698852539062, 17.131378173828125, 0.77777099609375, -4.44586181640625, 6.118438720703125, -8.260452270507812, 0.0169677734375, 6.6330413818359375, -7.7521514892578125, 15.13232421875, 16.06622314453125, 8.572463989257812, 8.087860107421875, -0.8833160400390625, 5.199233055114746, 3.2348709106445312, 8.35223388671875, -19.45977783203125, 26.341339111328125, 7.876861572265625, 4.625730514526367, -10.210113525390625, 8.10137939453125, 18.52349853515625, 14.674041748046875, -5.21270751953125, 22.16656494140625, 2.6431961059570312, 6.13446044921875, 0.746917724609375, 6.125213623046875, 6.79339599609375, 23.3985595703125, 4.812400817871094, 10.54461669921875, 18.56646728515625, 6.039344787597656, 18.3924560546875, 6.505615234375, -5.0320892333984375, 12.533103942871094, 2.1323623657226562, -0.4190826416015625, -0.264556884765625, -13.767486572265625, -21.3323974609375, -19.5267333984375, 11.280799865722656, 10.672775268554688, -4.176795959472656, 14.350151062011719, -19.057098388671875, 14.991119384765625, 8.097137451171875, 4.518707275390625, 1.212249755859375, 19.2913818359375, 1.2871971130371094, -4.764190673828125, 16.64996337890625, -23.80401611328125, 6.415802001953125, 2.9417572021484375, 19.358734130859375, 1.812835693359375, 15.918670654296875, -1.304443359375, 0.6847076416015625, -0.7623291015625, 9.142974853515625, 5.306955337524414, 19.84918212890625, 6.0705108642578125, 3.4929885864257812, 9.05828857421875, 0.6287269592285156, -1.6820068359375, 3.6776962280273438, 17.466506958007812, 4.449592590332031, 13.436553955078125, -3.751129150390625, 7.1068572998046875, 3.052001953125, -7.7817840576171875, 2.445148468017578, 2.0399017333984375, 7.476768493652344, 19.4351806640625, 0.7727813720703125, 15.34063720703125, -11.2196044921875, 1.4971332550048828, -3.861358642578125, 18.703521728515625, -0.0707244873046875, 0.0, 4.6587677001953125, -0.7399749755859375, 16.2230224609375, 2.8607940673828125, -3.36248779296875, 17.343963623046875, 23.972564697265625, 24.131805419921875, 11.289566040039062, -22.605255126953125, 15.65045166015625, 2.4767074584960938, 7.709716796875, 12.65777587890625, 24.402191162109375, -1.91064453125, 2.2191505432128906, 3.0362319946289062, -2.2734375, 13.632049560546875, -0.6235198974609375, 13.879669189453125, 6.161872863769531, 22.0482177734375, 3.642303466796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000099.npy"}
{"epoch": 0.20732984293193718, "step": 100, "batch_size": 128, "mean": 4.444305419921875, "std": 9.659112930297852, "min": -17.9765625, "p10": -6.868327331542968, "median": 3.335735321044922, "p90": 18.52652740478515, "max": 28.47491455078125, "pos_frac": 0.6875, "sample": [7.57080078125, 2.96453857421875, 10.930496215820312, 3.8947906494140625, 14.00787353515625, 3.555419921875, -10.922286987304688, 17.604156494140625, -0.0113525390625, 8.780364990234375, -5.95068359375, 1.065338134765625, 0.840728759765625, 5.747802734375, 13.82421875, 26.069915771484375, 0.3524761199951172, 1.30047607421875, -8.689743041992188, 8.203262329101562, 1.652862548828125, 6.181060791015625, -9.27178955078125, -1.7287883758544922, 0.9764404296875, 11.506622314453125, 4.032867431640625, -8.444206237792969, -0.01209259033203125, 13.71124267578125, -13.533355712890625, 0.427734375, -1.8224639892578125, 5.134605407714844, 2.32781982421875, 4.696510314941406, -0.766632080078125, 20.448440551757812, 3.2097396850585938, 14.957412719726562, -1.858306884765625, 11.2523193359375, 8.025604248046875, 2.38275146484375, -1.4265403747558594, 5.6756591796875, 7.7814483642578125, 22.371337890625, 23.5615234375, 13.417144775390625, -0.9169082641601562, -6.6015472412109375, -5.4972686767578125, 5.546550750732422, 3.6482391357421875, 10.46978759765625, -13.6207275390625, -1.01702880859375, -17.9765625, 1.8510055541992188, 0.174102783203125, 27.318359375, -5.497039794921875, 4.782722473144531, -1.419586181640625, 6.13165283203125, -4.910308837890625, -12.993743896484375, 5.514251708984375, 4.8514404296875, 28.47491455078125, 9.045989990234375, 4.0547637939453125, 5.69279670715332, -0.9546432495117188, -7.490814208984375, -5.2996673583984375, 7.172943115234375, 1.2637176513671875, 11.642316818237305, -12.294113159179688, 4.7050933837890625, 5.784759521484375, 1.79071044921875, 6.54791259765625, 2.873353958129883, -4.58880615234375, -0.45391082763671875, 17.702850341796875, 25.53350830078125, 0.334991455078125, 3.663848876953125, -5.5842132568359375, 3.46173095703125, 5.7557373046875, 0.9993667602539062, 4.4351806640625, 2.4656238555908203, -0.02637481689453125, 11.580093383789062, 23.574310302734375, 0.26055908203125, 24.600936889648438, 15.463104248046875, 5.6803741455078125, -4.036529541015625, 16.773849487304688, 24.819412231445312, 0.9229145050048828, 2.0800323486328125, 24.746490478515625, 5.195465087890625, -0.376708984375, -3.5235671997070312, -11.414031982421875, 16.154617309570312, -0.1190948486328125, 2.860931396484375, 21.827362060546875, -9.719024658203125, 0.8140449523925781, 10.253082275390625, 14.766326904296875, -4.111003875732422, 9.425472259521484, -0.9369964599609375, 20.548004150390625, -11.7918701171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000100.npy"}
{"epoch": 0.2094240837696335, "step": 101, "batch_size": 128, "mean": 4.617471218109131, "std": 10.421812057495117, "min": -23.76544189453125, "p10": -5.783164978027341, "median": 4.67015266418457, "p90": 16.522279357910154, "max": 37.31256103515625, "pos_frac": 0.7265625, "sample": [5.752372741699219, -23.76544189453125, 5.5277252197265625, 16.417510986328125, 3.7283248901367188, 1.477081298828125, 2.8793258666992188, 5.39593505859375, -4.32879638671875, 1.225189208984375, 18.329315185546875, 11.137161254882812, -3.032073974609375, 21.733489990234375, 35.0672607421875, 26.4208984375, 2.6078128814697266, 4.886051177978516, -1.1935043334960938, 10.07464599609375, 7.4476470947265625, 6.172615051269531, 11.141571044921875, 8.621612548828125, 1.463541030883789, 20.455612182617188, 5.430816650390625, 2.863616943359375, 1.2655181884765625, 27.604248046875, -2.4368515014648438, 7.0716552734375, 3.7319602966308594, 7.5486602783203125, 15.21002197265625, 9.0889892578125, 20.814117431640625, 14.2183837890625, 12.416961669921875, 1.76873779296875, -8.237579345703125, 3.6016006469726562, 12.21575927734375, 10.533340454101562, 18.93238067626953, 16.766738891601562, 6.5843505859375, -5.0373077392578125, -0.2593269348144531, -11.463043212890625, -2.147430419921875, 7.1126251220703125, -15.972946166992188, 11.293815612792969, 20.6116943359375, 6.7091064453125, 7.4635162353515625, -17.424468994140625, -1.284860610961914, 3.1959304809570312, -4.543113708496094, 8.415023803710938, 4.454254150390625, 11.596946716308594, -1.093231201171875, -4.299966812133789, 6.976020812988281, 11.2423095703125, -1.0008468627929688, 14.895004272460938, 0.599365234375, 4.167230606079102, -0.6611328125, 5.691314697265625, 0.9813766479492188, 2.3429527282714844, 19.5672607421875, 1.346160888671875, 3.68280029296875, -21.628143310546875, -21.278472900390625, 7.604217529296875, -10.030036926269531, 13.568695068359375, 3.1727752685546875, 2.2804718017578125, 13.64691162109375, 14.281574249267578, 1.5499763488769531, -2.4991607666015625, 8.644458770751953, 5.034418106079102, 8.2008056640625, 4.960296630859375, 8.20428466796875, 3.3463287353515625, -1.730316162109375, 30.264129638671875, 13.35333251953125, 5.0560455322265625, -17.15765380859375, -21.8472900390625, 9.864496231079102, 1.76629638671875, 5.022796630859375, -4.0927734375, 7.40936279296875, 4.0086822509765625, 37.31256103515625, -7.484832763671875, 9.570953369140625, -0.5084991455078125, 11.498565673828125, -1.99676513671875, 1.2004852294921875, 6.269081115722656, -1.769561767578125, -5.0538787841796875, -4.87164306640625, 6.659889221191406, -9.146316528320312, -1.324615478515625, 3.610198974609375, 0.4791297912597656, -13.752307891845703, 1.3050565719604492, 7.451416015625, 10.817523956298828], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000101.npy"}
{"epoch": 0.21151832460732983, "step": 102, "batch_size": 128, "mean": 6.372309684753418, "std": 9.50105094909668, "min": -18.89666748046875, "p10": -4.835313415527343, "median": 5.748958587646484, "p90": 18.19741897583008, "max": 32.8017578125, "pos_frac": 0.7421875, "sample": [-4.87255859375, -1.2982177734375, 12.865653991699219, 6.464572906494141, 1.05535888671875, 18.169158935546875, 8.1075439453125, 0.1264781951904297, 2.5549774169921875, 14.5032958984375, 5.36578369140625, 6.597660064697266, 11.571090698242188, 6.96905517578125, 0.564910888671875, 15.68255615234375, 9.029937744140625, -10.30621337890625, -3.3515777587890625, -0.8786888122558594, -3.620391845703125, 12.618209838867188, 12.497879028320312, 22.67584228515625, 16.645767211914062, -3.590179443359375, 5.380367279052734, -7.2406005859375, 5.722801208496094, -7.5074462890625, -5.7129669189453125, 6.936454772949219, 3.1119728088378906, 9.2822265625, -8.4423828125, 19.827392578125, 15.58941650390625, -1.7248458862304688, 0.5997238159179688, 0.6112594604492188, -0.6416015625, 1.4898452758789062, -1.8204669952392578, 16.676597595214844, 8.344390869140625, 4.80267333984375, 3.2950363159179688, -4.68255615234375, -8.66741943359375, -6.64093017578125, 17.29962158203125, -4.0355682373046875, 5.689971923828125, 6.29449462890625, 5.775115966796875, 17.07171630859375, 7.7296142578125, 32.8017578125, 9.24713134765625, 15.955322265625, 1.6217384338378906, -1.868316650390625, 17.82220458984375, 10.790435791015625, 7.6796875, -4.8193511962890625, -8.806930541992188, 0.7762603759765625, -4.213569641113281, 6.069488525390625, 11.00003433227539, 16.747344970703125, 21.96019744873047, 15.6719970703125, 5.565410614013672, 18.871597290039062, 19.678058624267578, 5.4610595703125, 7.93341064453125, -13.015533447265625, 1.2598075866699219, 7.2344970703125, 0.0, 6.42962646484375, -0.6979522705078125, 18.47064208984375, 0.0, 5.279327392578125, 14.63140869140625, -0.028619766235351562, 12.590744018554688, 12.513092041015625, -1.3665008544921875, 2.3695068359375, 17.48828125, 4.98529052734375, 13.091156005859375, 0.5233230590820312, 11.42596435546875, 15.932777404785156, 18.7789306640625, 4.743110656738281, 2.7278213500976562, 25.12969970703125, 6.36163330078125, 9.32244873046875, 2.794656753540039, 18.26335906982422, 0.8391456604003906, 18.33380126953125, -18.89666748046875, 10.944000244140625, -2.49432373046875, 3.0254440307617188, 13.62896728515625, -15.322113037109375, 15.460060119628906, 3.519134521484375, 31.08477783203125, 4.1533050537109375, 16.83160400390625, 3.788726806640625, 11.282669067382812, 9.092529296875, 9.960418701171875, 29.458099365234375, -0.724822998046875, -12.05438232421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000102.npy"}
{"epoch": 0.2136125654450262, "step": 103, "batch_size": 128, "mean": 4.784704208374023, "std": 10.421130180358887, "min": -23.111419677734375, "p10": -6.565217590332028, "median": 2.7330589294433594, "p90": 18.591925048828124, "max": 46.500701904296875, "pos_frac": 0.7109375, "sample": [-16.4671630859375, -8.543441772460938, 7.70037841796875, 9.929107666015625, 20.89007568359375, 15.70688247680664, -0.242340087890625, -3.0496063232421875, 23.164306640625, 1.599212646484375, -1.3018798828125, 15.6400146484375, 18.90625, 4.568634033203125, 0.0, -1.684814453125, -15.27032470703125, -9.859161376953125, -3.009979248046875, 7.9749755859375, -0.7877731323242188, 1.3347244262695312, 9.869186401367188, -23.111419677734375, 2.9952163696289062, 5.1126708984375, 1.1097412109375, 2.66607666015625, 5.740386962890625, 1.0941009521484375, 11.7030029296875, 9.163093566894531, 0.3491363525390625, 5.643463134765625, -1.95611572265625, 7.213237762451172, -2.6104507446289062, 15.426483154296875, 8.58331298828125, 8.376476287841797, -1.7636489868164062, 14.2801513671875, 0.707000732421875, 29.5057373046875, -1.6525726318359375, 12.943435668945312, 2.6623916625976562, 8.451263427734375, 11.89532470703125, 46.500701904296875, 17.801544189453125, 13.782867431640625, 7.92059326171875, -9.86175537109375, 1.637420654296875, 0.1972503662109375, -10.46978759765625, 10.9324951171875, 10.412109375, 10.287506103515625, 17.542266845703125, 7.82049560546875, 1.0076751708984375, 11.08258056640625, 4.638801574707031, 20.4854736328125, 10.0042724609375, 11.742599487304688, 3.3563079833984375, 6.8195037841796875, 1.40771484375, 0.6057891845703125, 0.982879638671875, 2.92156982421875, 0.39849853515625, 1.1224708557128906, -8.632564544677734, -18.20074462890625, 0.6434326171875, 2.47882080078125, 21.419036865234375, 13.016365051269531, -0.1394519805908203, 10.8013916015625, -1.5462112426757812, 8.396820068359375, 5.2034149169921875, 1.4121780395507812, 25.69000244140625, -3.037109375, 2.8000411987304688, 2.37225341796875, 15.682647705078125, 8.198486328125, -2.260162353515625, 1.5361289978027344, 23.364288330078125, -1.39935302734375, -5.7174072265625, -12.86431884765625, 0.0, 2.9951934814453125, 3.04656982421875, 16.25823974609375, -1.087554931640625, 19.60089111328125, -11.01422119140625, 2.252593994140625, 1.2078170776367188, 11.726818084716797, 12.589950561523438, -4.8235931396484375, -3.6596641540527344, 18.45721435546875, 0.802398681640625, 0.5434894561767578, 20.31195068359375, 24.3814697265625, 10.841644287109375, 19.97509765625, -0.42706298828125, 6.113349914550781, -1.72650146484375, -14.962677001953125, 1.1021652221679688, -19.25677490234375, -5.650581359863281, 0.95135498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000103.npy"}
{"epoch": 0.2157068062827225, "step": 104, "batch_size": 128, "mean": 4.471951484680176, "std": 11.478793144226074, "min": -26.76025390625, "p10": -8.683087921142578, "median": 3.5751876831054688, "p90": 18.96433410644531, "max": 48.59619140625, "pos_frac": 0.6796875, "sample": [3.00628662109375, -3.538177490234375, 8.105026245117188, 17.5283203125, -10.43829345703125, 6.187889099121094, 4.612495422363281, -0.8216190338134766, 18.6800537109375, -5.93798828125, 18.27252197265625, -8.244930267333984, 5.089080810546875, -17.4332275390625, 7.4879150390625, 8.008235931396484, 28.509124755859375, 20.967437744140625, -0.01580810546875, -2.8889999389648438, 3.3054351806640625, -2.5506649017333984, 6.3079833984375, -22.24761962890625, 25.3656005859375, -3.4818572998046875, 3.7672119140625, -1.6906185150146484, 2.715608596801758, 24.1815185546875, 6.5446624755859375, 15.057373046875, -14.547515869140625, 0.0935211181640625, 3.006927490234375, -8.941246032714844, 1.076944351196289, -8.324424743652344, 20.35052490234375, -1.99346923828125, 12.347808837890625, 1.53839111328125, 1.0755958557128906, -10.758941650390625, -9.324066162109375, 10.1827392578125, 0.86590576171875, 9.929412841796875, -1.1030731201171875, -2.14166259765625, -4.689727783203125, 9.534088134765625, 9.545120239257812, 3.3831634521484375, 25.5672607421875, 5.4317169189453125, 19.627655029296875, -5.7738800048828125, -2.6968002319335938, 2.5763206481933594, 4.537319183349609, 9.52471923828125, 5.4478759765625, 6.638637542724609, 5.39190673828125, 3.2886276245117188, -0.8882369995117188, 15.88006591796875, 5.02557373046875, -14.580886840820312, 10.560897827148438, 18.206634521484375, 0.5991668701171875, -25.01483154296875, 48.59619140625, 5.9694976806640625, 22.10174560546875, 9.862663269042969, 11.250389099121094, 22.464447021484375, 6.006439208984375, -26.76025390625, -10.517120361328125, 11.507675170898438, 2.673675537109375, -0.2765693664550781, 1.4175910949707031, 7.471099853515625, -8.57244873046875, 17.409469604492188, 24.511138916015625, 0.2865447998046875, 4.9600830078125, 9.62530517578125, 1.168487548828125, 0.43022918701171875, -6.934661865234375, 0.7461013793945312, 24.326904296875, 12.842330932617188, 16.17766571044922, 29.1043701171875, -5.070579528808594, 16.38714599609375, -19.13385009765625, 0.7700347900390625, -4.66485595703125, 3.8770065307617188, 5.351615905761719, -14.19403076171875, 0.0, 9.323623657226562, -1.225494384765625, 14.039306640625, 3.8464202880859375, 4.92327880859375, 11.254791259765625, -7.018028259277344, 8.442131042480469, -0.1949310302734375, -0.361724853515625, 15.199203491210938, 2.4217662811279297, 7.1298065185546875, 17.97698974609375, -2.11199951171875, 0.36859130859375, 2.3588638305664062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000104.npy"}
{"epoch": 0.21780104712041884, "step": 105, "batch_size": 128, "mean": 8.068690299987793, "std": 11.162578582763672, "min": -23.83758544921875, "p10": -5.561851501464844, "median": 7.068836212158203, "p90": 22.574047851562497, "max": 34.1983642578125, "pos_frac": 0.7890625, "sample": [4.41973876953125, 3.897552490234375, 14.35662841796875, 1.6042633056640625, -6.470703125, 17.72393798828125, 8.977241516113281, 15.520469665527344, 11.219268798828125, 2.9809532165527344, 15.106842041015625, 5.490997314453125, 21.435455322265625, 19.096954345703125, -9.11865234375, 21.54437255859375, 6.313087463378906, -0.6874313354492188, 27.425430297851562, 8.97027587890625, -0.9703407287597656, -0.8818359375, 7.6940155029296875, 7.116432189941406, 23.9539794921875, 29.747772216796875, -22.0875244140625, 0.49480438232421875, -5.5500640869140625, 28.03289794921875, 5.51953125, 13.607772827148438, 0.8600387573242188, 18.26409912109375, -2.547119140625, 20.670745849609375, 23.263595581054688, 4.4506378173828125, 19.662109375, -10.95660400390625, 2.915252685546875, 22.86090087890625, 15.194305419921875, 15.67462158203125, 3.0202789306640625, 18.111114501953125, 11.083648681640625, -1.93902587890625, 20.87322998046875, 19.97747802734375, 3.213226318359375, 14.051910400390625, -0.816497802734375, 14.166595458984375, -2.3905029296875, 13.321273803710938, 10.34722900390625, 23.6365966796875, 8.323013305664062, -10.0557861328125, 4.08209228515625, -11.730072021484375, -0.978057861328125, 4.485198974609375, 4.944610595703125, 7.021240234375, 13.93023681640625, 0.8411865234375, 2.685148239135742, 3.464202880859375, 21.124114990234375, -18.5841064453125, 17.103118896484375, 21.17022705078125, 8.563720703125, -5.100128173828125, 12.1041259765625, 17.6007080078125, 26.23284912109375, 2.3398361206054688, 7.875244140625, 3.296356201171875, 7.721221923828125, 0.4064006805419922, 6.7916259765625, 19.12347412109375, -13.02764892578125, 23.817291259765625, 7.377384185791016, -23.83758544921875, 3.3730621337890625, 11.121124267578125, 24.839553833007812, -12.69805908203125, 6.139961242675781, 3.3371124267578125, 9.390581130981445, -7.075592041015625, 12.8447265625, 0.0, 4.5005340576171875, 6.709197998046875, 0.4042167663574219, -0.6203231811523438, 15.425628662109375, -3.3947601318359375, 32.59912109375, 26.96600341796875, 5.9959716796875, 5.141792297363281, 7.2972869873046875, 22.45111083984375, 0.4460258483886719, 4.224784851074219, 17.95068359375, 11.78021240234375, -5.58935546875, 11.21331787109375, 3.3214187622070312, -7.952667236328125, 7.931121826171875, 21.257911682128906, 4.269401550292969, 4.544200897216797, -1.708831787109375, 34.1983642578125, 19.780059814453125, 9.81298828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000105.npy"}
{"epoch": 0.2198952879581152, "step": 106, "batch_size": 128, "mean": 5.48937463760376, "std": 11.643394470214844, "min": -21.015731811523438, "p10": -8.859976196289061, "median": 5.053184509277344, "p90": 21.048883056640623, "max": 36.747039794921875, "pos_frac": 0.7109375, "sample": [-6.934959411621094, -15.0220947265625, 14.12408447265625, -3.388214111328125, -2.325103759765625, 20.8636474609375, -6.222808837890625, 7.8865814208984375, -13.299713134765625, 15.449714660644531, -12.52386474609375, 0.385101318359375, 5.4326019287109375, 19.395263671875, 14.45556640625, 31.810211181640625, 26.598175048828125, -8.528564453125, -3.9101104736328125, -2.8268508911132812, 13.764976501464844, 3.236114501953125, 4.840667724609375, 4.555564880371094, 11.99322509765625, 0.8361968994140625, 3.3323659896850586, 7.368736267089844, 12.200981140136719, 24.56512451171875, 2.19708251953125, 9.468353271484375, 16.070281982421875, 0.7883758544921875, 0.20733642578125, -6.410499572753906, 10.702880859375, 2.6941986083984375, 6.6646728515625, 8.362045288085938, 21.177825927734375, 0.464752197265625, 7.7999267578125, -17.17864990234375, -3.1627197265625, 14.10504150390625, -7.73974609375, 0.1588134765625, 13.24932861328125, 5.182304382324219, 6.31951904296875, 7.890571594238281, -4.691610336303711, 16.742156982421875, -17.834136962890625, -21.015731811523438, 10.04888916015625, 31.43988037109375, 27.549942016601562, 20.978790283203125, 7.441162109375, 7.593841552734375, -10.4720458984375, -7.6761474609375, 10.954833984375, -0.070068359375, -5.977386474609375, -5.1160430908203125, 23.278564453125, -11.885345458984375, 11.82012939453125, 6.520172119140625, -1.36480712890625, 5.3977813720703125, -12.814849853515625, 1.5366153717041016, -3.176025390625, 1.885101318359375, 17.6051025390625, 2.8213043212890625, 0.5576019287109375, 36.747039794921875, 7.658660888671875, 1.759185791015625, -5.438079833984375, 2.6070556640625, 22.713653564453125, -0.6576309204101562, 14.101409912109375, -9.828643798828125, 10.090492248535156, 4.147735595703125, 16.40869140625, 0.352935791015625, 1.0785064697265625, 4.924064636230469, 13.608901977539062, 20.579132080078125, 5.739349365234375, 8.962959289550781, 29.831207275390625, -17.176055908203125, 1.6099166870117188, -2.602142333984375, 6.4228668212890625, 12.336063385009766, 13.029891967773438, -13.567550659179688, 3.3094482421875, -5.79132080078125, -9.633270263671875, -3.90234375, 32.37176513671875, 7.550510406494141, -5.19610595703125, 6.0725250244140625, 11.82421875, -5.840080261230469, 20.155288696289062, 3.94390869140625, 22.34234619140625, 5.65484619140625, 8.590164184570312, 23.677001953125, 20.993621826171875, 0.6406631469726562, 9.638702392578125, 1.59649658203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000106.npy"}
{"epoch": 0.22198952879581152, "step": 107, "batch_size": 128, "mean": 7.324915885925293, "std": 12.289602279663086, "min": -56.1845703125, "p10": -4.995127487182617, "median": 5.470947265625, "p90": 23.624589538574217, "max": 34.37969970703125, "pos_frac": 0.734375, "sample": [1.1455230712890625, -3.52410888671875, 4.814453125, 4.9463958740234375, 10.5562744140625, -5.9761962890625, 14.369384765625, 27.130340576171875, 6.486480712890625, 5.027923583984375, -4.727783203125, 4.4328155517578125, 1.46771240234375, 13.435348510742188, 12.411029815673828, -1.250732421875, 12.743377685546875, -9.162628173828125, 11.0670166015625, 19.352783203125, -2.91815185546875, 21.722320556640625, 7.637901306152344, 13.64892578125, -11.606414794921875, 28.143280029296875, 8.749359130859375, -12.1517333984375, 15.98236083984375, -6.786674499511719, 18.347503662109375, 1.2234649658203125, 1.604644775390625, 18.65857696533203, 22.126708984375, 23.3970947265625, 11.1900634765625, 6.464324951171875, 34.37969970703125, 0.310211181640625, 26.443832397460938, 7.1689453125, 7.3325958251953125, 13.074111938476562, 3.62188720703125, 1.1147842407226562, -2.03564453125, 5.30108642578125, 13.760885238647461, 7.20428466796875, 8.271675109863281, -0.9994964599609375, 3.3196487426757812, 3.7012176513671875, 22.2825927734375, -5.080600738525391, 2.2686767578125, 0.6915397644042969, 20.504150390625, 9.340286254882812, 5.64080810546875, 2.329986572265625, -4.95849609375, 33.2852783203125, -0.7459259033203125, -5.748725891113281, 3.9883041381835938, 31.99468994140625, 22.461685180664062, 17.37109375, -12.284332275390625, 24.155410766601562, -3.8311309814453125, 15.620834350585938, 20.71630859375, 1.8855743408203125, 2.608184814453125, 19.2294921875, 13.916305541992188, 11.92837905883789, 27.623321533203125, 14.057052612304688, -2.517578125, -11.593307495117188, 0.344512939453125, 4.8387451171875, 7.1192779541015625, -1.1289901733398438, 33.823150634765625, -6.604164123535156, 28.810943603515625, 11.173904418945312, -0.9553909301757812, 7.330047607421875, 4.315032958984375, 0.8059253692626953, 17.275177001953125, 3.01300048828125, 0.0636138916015625, 0.4677886962890625, -56.1845703125, -8.505096435546875, 12.84942626953125, 27.558074951171875, 17.806640625, -3.0374679565429688, 25.28936767578125, -1.86968994140625, 8.292587280273438, 1.5589256286621094, 20.026412963867188, -1.2165679931640625, 10.724868774414062, 3.23455810546875, 5.72027587890625, 16.412353515625, -3.3861846923828125, -1.16693115234375, -5.40594482421875, -0.77960205078125, -0.30230712890625, 8.791091918945312, -4.2677001953125, 7.33489990234375, -1.122293472290039, 29.320510864257812, 5.0553131103515625, 14.8770751953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000107.npy"}
{"epoch": 0.22408376963350785, "step": 108, "batch_size": 128, "mean": 8.088191032409668, "std": 10.69871997833252, "min": -19.223251342773438, "p10": -4.7253990173339835, "median": 6.935230255126953, "p90": 24.418408203125, "max": 33.31805419921875, "pos_frac": 0.7734375, "sample": [24.560760498046875, 16.732986450195312, -1.8543624877929688, 6.4844207763671875, 14.262054443359375, -6.9576873779296875, 18.863773345947266, 9.01040267944336, 3.625957489013672, 12.558242797851562, 2.6769180297851562, -6.227508544921875, -4.568351745605469, 10.62994384765625, 4.8665924072265625, -6.8297119140625, 26.557861328125, 4.839080810546875, -2.1671295166015625, 9.17303466796875, 12.2718505859375, 5.7586669921875, 10.25830078125, -6.595733642578125, -0.8498382568359375, 8.456314086914062, 26.817138671875, -2.1640625, 0.0, -18.03668212890625, 13.727245330810547, 4.407989501953125, 10.767059326171875, 5.079021453857422, 17.30975341796875, 3.001068115234375, 26.46240234375, 6.996856689453125, 16.450057983398438, -0.9574966430664062, 3.3842201232910156, 9.988616943359375, 2.1514892578125, -4.000762939453125, 6.5146484375, 26.44207763671875, 11.81207275390625, -15.02874755859375, 5.71533203125, 26.14605712890625, 14.618934631347656, 4.187294006347656, 9.043426513671875, 33.31805419921875, 6.4007568359375, 2.4603118896484375, 0.0, -5.0918426513671875, 1.5514755249023438, 0.26253700256347656, 2.0420265197753906, 22.9912109375, 15.439117431640625, 8.7099609375, -19.223251342773438, 8.474838256835938, 18.5733642578125, 11.853458404541016, -1.0660018920898438, 10.849090576171875, 10.02239990234375, 5.83453369140625, -2.271148681640625, 3.10931396484375, -4.08099365234375, 24.365234375, 20.37432861328125, 2.3438644409179688, 12.72076416015625, 18.065185546875, 12.24951171875, -5.646339416503906, 24.54248046875, 11.670196533203125, 3.97509765625, 1.668243408203125, 22.286834716796875, 11.05245590209961, 8.939529418945312, 28.54901123046875, 8.5426025390625, 20.018310546875, 3.7424774169921875, 9.379928588867188, 9.217126846313477, -2.8780670166015625, 28.9334716796875, 15.874679565429688, 26.79730224609375, 20.06622314453125, 16.563400268554688, 3.827545166015625, 7.1640777587890625, 0.0, -1.54766845703125, 10.061874389648438, 6.873603820800781, 33.288787841796875, 5.445240020751953, 22.86602783203125, 6.520843505859375, -8.863006591796875, 4.252197265625, 1.2515983581542969, 16.7579345703125, 26.96563720703125, 21.9517822265625, 1.731109619140625, 2.3530197143554688, 10.779998779296875, -10.038803100585938, -11.472991943359375, -10.674346923828125, 5.8872222900390625, -0.24997711181640625, 10.52490234375, 2.895294189453125, 16.791641235351562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000108.npy"}
{"epoch": 0.2261780104712042, "step": 109, "batch_size": 128, "mean": 8.24761962890625, "std": 13.462215423583984, "min": -27.27117919921875, "p10": -6.243008232116699, "median": 6.79429817199707, "p90": 23.10716857910156, "max": 58.2725830078125, "pos_frac": 0.734375, "sample": [3.4164886474609375, 11.498062133789062, -10.464271545410156, 21.731307983398438, 4.9174346923828125, 7.954170227050781, 14.508453369140625, 16.8634033203125, -10.544845581054688, 6.5995330810546875, 7.16912841796875, 19.102294921875, 0.4470329284667969, -3.0397796630859375, -5.85809326171875, -4.72381591796875, -0.77734375, 30.23388671875, 41.389251708984375, 1.6996536254882812, 1.0726089477539062, -0.064544677734375, -3.5853729248046875, -12.411224365234375, -6.672525405883789, 10.705558776855469, 1.35076904296875, -1.7528572082519531, -0.09609222412109375, 10.881591796875, 5.8956451416015625, -0.7863311767578125, 14.1124267578125, 34.614654541015625, 23.4581298828125, 9.94668960571289, 13.226852416992188, 20.377899169921875, 1.6533584594726562, 19.9716796875, 12.4544677734375, 18.380828857421875, 5.401123046875, -0.160400390625, 12.208770751953125, 33.573333740234375, -0.505035400390625, 28.0970458984375, 7.0630645751953125, 6.760715484619141, 12.716384887695312, 4.72662353515625, 9.71234130859375, 6.563262939453125, -22.1427001953125, 6.56915283203125, 8.065185546875, 7.617895126342773, 12.755752563476562, 8.4334716796875, 16.111785888671875, 3.456890106201172, 0.66131591796875, 22.956756591796875, 6.827880859375, 11.393577575683594, -6.856170654296875, 9.962066650390625, 9.370285034179688, 12.013435363769531, 16.528900146484375, 3.769317626953125, -8.130661010742188, 17.95159912109375, 10.698051452636719, 22.79058837890625, 47.45849609375, -3.794281005859375, 1.350341796875, 31.15777587890625, 4.8658447265625, 5.690055847167969, 17.367088317871094, 14.248245239257812, -0.0053863525390625, 1.2851028442382812, 12.4522705078125, 0.7459869384765625, -10.278884887695312, 41.71875, 22.55291748046875, 17.769927978515625, 6.221282958984375, 58.2725830078125, 9.033355712890625, -1.1634521484375, 1.6126632690429688, 35.115966796875, -11.286651611328125, 4.2857513427734375, -6.058929443359375, 18.975997924804688, 0.0, 0.9852218627929688, 31.62677001953125, 8.597991943359375, 13.75, 8.396774291992188, -0.8204879760742188, 10.353515625, -0.1220703125, 6.5976409912109375, 1.34454345703125, -2.3837852478027344, 20.18463134765625, -1.8848876953125, 12.651573181152344, 20.205322265625, -13.346771240234375, -27.27117919921875, -26.539527893066406, 24.09197998046875, 4.74139404296875, 15.89495849609375, 4.9144287109375, -8.77972412109375, -4.570648193359375, 15.667022705078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000109.npy"}
{"epoch": 0.22827225130890053, "step": 110, "batch_size": 128, "mean": 7.896410942077637, "std": 13.432673454284668, "min": -39.536773681640625, "p10": -6.19917984008789, "median": 5.306194305419922, "p90": 26.800865173339844, "max": 43.62469482421875, "pos_frac": 0.7421875, "sample": [-0.425628662109375, -39.536773681640625, -11.483078002929688, 27.381317138671875, -2.2529983520507812, 1.3261985778808594, -2.5016555786132812, 1.123046875, -3.44171142578125, 28.22784423828125, 3.494903564453125, 3.5606689453125, 2.435321807861328, 22.961151123046875, -5.141429901123047, 0.044464111328125, 11.37359619140625, 5.272972106933594, 17.57763671875, 4.18267822265625, -7.7206878662109375, 20.477996826171875, 2.1971282958984375, 24.193450927734375, 7.07830810546875, 3.859283447265625, -6.023368835449219, 13.547019958496094, 1.895263671875, -3.30816650390625, 27.85107421875, 13.344573974609375, 27.1180419921875, 2.317190170288086, 35.4512939453125, -1.96649169921875, 16.86627197265625, 18.004928588867188, 10.689971923828125, -3.003662109375, 7.19696044921875, 5.70135498046875, 12.023635864257812, 28.3153076171875, -9.4034423828125, 0.07248306274414062, 14.446701049804688, -8.3892822265625, 11.710990905761719, 0.5584030151367188, 21.345611572265625, -20.4776611328125, 11.29315185546875, 1.2959442138671875, 31.46563720703125, 10.955169677734375, 4.4812774658203125, -30.347900390625, 16.071670532226562, 17.689208984375, 4.113487243652344, 5.94488525390625, 3.5130157470703125, -0.81707763671875, 1.595245361328125, 28.955902099609375, -6.724388122558594, 25.9571533203125, 19.199188232421875, -14.53533935546875, 0.1774120330810547, 15.64837646484375, 4.9271697998046875, 2.3752899169921875, 11.75921630859375, 7.7388763427734375, 0.7331809997558594, -3.5082931518554688, -1.8866424560546875, 11.565109252929688, 0.0, 4.74285888671875, 19.98284912109375, 14.288459777832031, -11.410945892333984, 14.5579833984375, 0.5086536407470703, -6.609405517578125, 5.33941650390625, 26.664932250976562, 14.6695556640625, 36.7960205078125, 15.4283447265625, 24.031509399414062, 5.8728790283203125, 19.67510986328125, 26.03997802734375, -1.96734619140625, -2.0002098083496094, 20.925048828125, 27.7457275390625, 7.039447784423828, 1.337386131286621, 22.580902099609375, -16.58740997314453, 8.586700439453125, 4.72314453125, -5.9456024169921875, 24.918121337890625, 19.62933349609375, -2.7298583984375, 30.108428955078125, -3.0698928833007812, 8.799896240234375, 10.89825439453125, 28.079345703125, 43.62469482421875, -2.415008544921875, 19.602294921875, 22.265960693359375, 0.8471832275390625, 16.776458740234375, 4.5750732421875, 7.743015289306641, -13.22052001953125, 2.6818695068359375, -0.402740478515625, 5.227783203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000110.npy"}
{"epoch": 0.23036649214659685, "step": 111, "batch_size": 128, "mean": 8.652054786682129, "std": 13.628812789916992, "min": -22.2723388671875, "p10": -8.421959686279294, "median": 7.591541290283203, "p90": 26.05764770507812, "max": 47.49751281738281, "pos_frac": 0.734375, "sample": [4.0135345458984375, -10.20465087890625, 12.55828857421875, -22.2723388671875, 46.65960693359375, -7.63818359375, 10.818557739257812, 27.138580322265625, 0.9958267211914062, -0.295166015625, 28.72686767578125, 5.869384765625, 6.3865509033203125, 5.545318603515625, 7.325286865234375, 15.051765441894531, 7.915931701660156, 35.547576904296875, -17.7652587890625, -13.756912231445312, 12.244064331054688, 5.760398864746094, 15.367935180664062, 3.75531005859375, 16.6793212890625, 2.46148681640625, 4.1875, 25.594390869140625, 2.0387191772460938, 11.669448852539062, 15.420112609863281, 20.2774658203125, -2.786846160888672, 32.30023193359375, -2.5311279296875, 5.6278533935546875, -6.581024169921875, 17.12298583984375, -0.4490203857421875, 10.375778198242188, 2.1346664428710938, 10.635330200195312, -7.824836730957031, 22.575836181640625, 11.55535888671875, 3.0784912109375, 15.40460205078125, -1.414093017578125, 31.060455322265625, 7.857795715332031, 0.5825119018554688, 47.49751281738281, -12.309906005859375, -2.7895355224609375, -1.29534912109375, 17.344970703125, 13.267578125, 23.357818603515625, 4.8877716064453125, 29.147918701171875, 6.8853759765625, 10.911605834960938, 5.4903717041015625, 9.332763671875, 3.7448577880859375, -5.50421142578125, 3.2386322021484375, 12.7635498046875, 41.02520751953125, 10.88232421875, 6.018890380859375, -1.62353515625, 8.0072021484375, -0.085723876953125, 3.830780029296875, 0.0543060302734375, -0.40533447265625, 25.082672119140625, 10.168746948242188, -18.193450927734375, 1.4243240356445312, 9.989990234375, -9.81524658203125, 4.765625, -3.634307861328125, -7.1738739013671875, -0.04644012451171875, 2.50927734375, 17.067237854003906, 27.57806396484375, 16.553070068359375, 30.9520263671875, 18.13433074951172, -2.3566551208496094, 16.467391967773438, 11.339981079101562, 16.020023345947266, -2.776580810546875, 43.74188232421875, 2.26654052734375, 21.292137145996094, 9.878364562988281, -13.857269287109375, -10.73602294921875, -4.17633056640625, 25.09503173828125, -10.883026123046875, 19.37445068359375, -1.4427642822265625, 19.890586853027344, 15.285446166992188, 8.45672607421875, 19.1549072265625, 35.61480712890625, 21.811798095703125, 19.225921630859375, 24.49908447265625, -16.559783935546875, 11.5897216796875, 14.501869201660156, 8.3619384765625, -11.31146240234375, 12.953216552734375, 6.69091796875, 19.724700927734375, 1.4093475341796875, -11.75982666015625, 2.836395263671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000111.npy"}
{"epoch": 0.2324607329842932, "step": 112, "batch_size": 128, "mean": 4.401219367980957, "std": 15.210238456726074, "min": -49.505126953125, "p10": -13.194552612304687, "median": 3.650543212890625, "p90": 26.010450744628905, "max": 46.516845703125, "pos_frac": 0.6171875, "sample": [16.92229461669922, 4.5820465087890625, -3.02593994140625, 11.6480712890625, 0.7789688110351562, 7.689872741699219, -0.8532180786132812, -29.6700439453125, -4.0146331787109375, 2.4051971435546875, -2.342000961303711, 6.387901306152344, -11.124267578125, 19.192474365234375, -2.181884765625, 3.7010269165039062, 13.011627197265625, 16.212081909179688, 5.791839599609375, 26.89117431640625, -7.74884033203125, 19.1710205078125, 12.857162475585938, 13.131134033203125, -49.505126953125, -13.956298828125, -9.214561462402344, 13.16864013671875, 9.472900390625, 9.813858032226562, -9.544471740722656, 5.6148834228515625, -26.456512451171875, 5.02252197265625, 4.7937469482421875, -19.354095458984375, 5.6462249755859375, 6.176736831665039, 0.5430755615234375, 3.6000595092773438, 16.455322265625, -2.83819580078125, 6.52349853515625, 14.964950561523438, 12.266197204589844, -2.62646484375, 13.02203369140625, -0.5915069580078125, 17.647811889648438, -4.952545166015625, 9.743621826171875, 4.1655426025390625, 0.73516845703125, 16.0992431640625, -16.871673583984375, -3.482473373413086, -15.034286499023438, -7.9623870849609375, -35.44921875, -8.162704467773438, -1.8769283294677734, -0.3975067138671875, 17.713577270507812, 35.308624267578125, -5.0408935546875, 3.0172882080078125, 30.204254150390625, 15.5172119140625, 0.1342315673828125, 7.151954650878906, 7.702480316162109, -6.08966064453125, 20.110870361328125, 14.88214111328125, 6.2874755859375, 36.639434814453125, -3.2800865173339844, 1.4703369140625, 27.504150390625, -17.80249786376953, 0.8829345703125, 8.810195922851562, 0.42952728271484375, -0.7466526031494141, 21.030670166015625, 25.027923583984375, 11.709228515625, -1.197723388671875, -6.023895263671875, 2.2076568603515625, 25.93280029296875, -12.94427490234375, 28.5916748046875, 4.2802886962890625, -5.38922119140625, 15.781829833984375, -1.8965835571289062, 0.5472202301025391, -38.20037841796875, -7.0306243896484375, -7.7005767822265625, 25.5107421875, -2.0984344482421875, -1.26519775390625, 27.269760131835938, 16.283004760742188, -13.778533935546875, 26.93011474609375, 4.6354522705078125, 18.522186279296875, 34.77508544921875, -12.257904052734375, 2.167755126953125, 27.729248046875, 6.1688690185546875, 0.606964111328125, -16.204925537109375, -2.3955841064453125, 1.6112213134765625, -0.06866455078125, 8.75732421875, -2.612518310546875, 26.7481689453125, 26.191635131835938, 46.516845703125, -15.6505126953125, -3.6566696166992188, 10.775619506835938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000112.npy"}
{"epoch": 0.23455497382198953, "step": 113, "batch_size": 128, "mean": 10.12381362915039, "std": 14.786063194274902, "min": -27.375152587890625, "p10": -7.768981933593749, "median": 8.303421020507812, "p90": 30.021734619140624, "max": 57.1114501953125, "pos_frac": 0.7421875, "sample": [5.881538391113281, 16.402610778808594, 7.5859375, 0.21595001220703125, 17.172607421875, 1.4198150634765625, -4.906951904296875, 14.823760986328125, 27.519073486328125, 5.28497314453125, 7.45367431640625, 7.447265625, 19.20623779296875, -10.88522720336914, -0.8182830810546875, 3.839998245239258, 3.3686389923095703, 2.0520172119140625, 4.85125732421875, 13.226432800292969, 57.1114501953125, -4.21246337890625, -2.0897598266601562, 42.54057312011719, 10.98321533203125, -27.375152587890625, 5.2732391357421875, 23.62030029296875, 13.012786865234375, -2.412841796875, -2.711395263671875, -10.29840087890625, 24.28924560546875, 31.75250244140625, 38.5499267578125, -0.73504638671875, 16.037033081054688, -6.8641357421875, 29.828933715820312, -10.5433349609375, 12.477157592773438, 30.471603393554688, -2.39605712890625, 13.44329833984375, 10.59149169921875, 4.733699798583984, -10.56854248046875, 1.07232666015625, 0.6016521453857422, 29.19403076171875, 4.7253265380859375, 18.387950897216797, -24.557876586914062, 25.863037109375, 33.17828369140625, 23.647064208984375, 11.69753646850586, 33.623077392578125, 6.2434844970703125, -11.8480224609375, 9.251304626464844, -1.1510162353515625, -5.2120208740234375, 25.358932495117188, -0.6778640747070312, 7.685302734375, 5.958255767822266, 6.531425476074219, -14.80938720703125, 16.771148681640625, 5.786285400390625, 31.8944091796875, -1.6836891174316406, 11.471260070800781, -8.611053466796875, 4.275810241699219, 13.15863037109375, 29.60260009765625, 15.240997314453125, 15.5557861328125, 14.419715881347656, -9.773681640625, 17.497146606445312, 28.156600952148438, 35.271087646484375, 13.531105041503906, 12.807395935058594, -3.8394393920898438, -3.7794952392578125, 6.449798583984375, 36.712982177734375, 26.258026123046875, 18.446075439453125, 13.743377685546875, -7.578460693359375, 27.355133056640625, 0.47739410400390625, 13.3853759765625, 15.091644287109375, 8.921539306640625, 5.473602294921875, -0.6661376953125, 12.575439453125, 3.2211837768554688, 23.18646240234375, 37.9012451171875, 24.737060546875, 39.04779052734375, -8.2579345703125, 1.59429931640625, 22.902511596679688, 17.719284057617188, 11.680389404296875, 43.38970947265625, -6.30340576171875, -8.213531494140625, 2.691253662109375, 3.4306106567382812, 19.566497802734375, 16.86590576171875, 0.32916259765625, 16.73931884765625, -17.391815185546875, 9.67535400390625, 4.418609619140625, -2.3753509521484375, 19.33648681640625, -4.854835510253906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000113.npy"}
{"epoch": 0.23664921465968586, "step": 114, "batch_size": 128, "mean": 5.156494140625, "std": 13.803410530090332, "min": -27.4796142578125, "p10": -12.687077331542968, "median": 4.0279693603515625, "p90": 22.85740051269531, "max": 41.1226806640625, "pos_frac": 0.6796875, "sample": [3.7295780181884766, 19.83563232421875, 3.0671234130859375, 26.551815032958984, -0.5894775390625, 7.016815185546875, 1.547882080078125, 4.824592590332031, -2.186614990234375, -5.507904052734375, -0.7438507080078125, 22.7613525390625, -0.12701416015625, 28.772918701171875, 15.063522338867188, 37.28363037109375, -11.339435577392578, 1.657958984375, -21.51483154296875, 1.4481658935546875, 1.5023193359375, -16.202911376953125, 4.218353271484375, 14.770477294921875, 2.2228164672851562, 17.61334228515625, 9.838897705078125, 9.18499755859375, 11.142120361328125, 15.5206298828125, 2.8396453857421875, 15.01171875, 4.9771881103515625, 7.370735168457031, 17.64154052734375, -12.32891845703125, 9.700447082519531, -25.1168212890625, 27.810333251953125, 6.155914306640625, -6.023963928222656, -4.7694091796875, 2.3928756713867188, -27.4796142578125, 5.8271484375, 1.5691909790039062, -0.0303497314453125, 1.526214599609375, 8.066070556640625, -8.425811767578125, 0.4731330871582031, -13.522781372070312, 10.810653686523438, -8.762374877929688, -16.874832153320312, -19.474761962890625, 11.737457275390625, 41.1226806640625, 23.081512451171875, -8.219894409179688, 25.6290283203125, -23.62017822265625, 20.135345458984375, -7.26202392578125, -3.50872802734375, 19.802490234375, 37.76476287841797, -9.692535400390625, 0.08102798461914062, 14.0701904296875, -23.579010009765625, 1.0532569885253906, -16.316192626953125, 10.226852416992188, 2.72174072265625, 35.48175048828125, 0.622802734375, -0.696807861328125, -2.4213714599609375, -16.86224365234375, 2.7857589721679688, 9.374336242675781, 7.294769287109375, -17.30859375, 1.6048297882080078, -0.9839687347412109, -19.3226318359375, -5.064971923828125, 0.0, 24.890289306640625, 11.578903198242188, -2.446788787841797, 12.546356201171875, 6.1266326904296875, -0.917327880859375, 22.681976318359375, -2.58856201171875, 18.2294921875, 2.47259521484375, 5.769020080566406, 10.365028381347656, 33.02716064453125, 12.036773681640625, 10.754669189453125, 36.8446044921875, 15.783729553222656, 4.971466064453125, 5.054222106933594, -7.9956817626953125, 0.0, 35.5390625, -1.7315826416015625, -5.8043975830078125, 14.3441162109375, 6.98187255859375, 20.347427368164062, 7.2683258056640625, 3.83758544921875, 5.558296203613281, 6.225654602050781, 3.2766265869140625, 8.222259521484375, 6.198284149169922, 0.579376220703125, 13.12652587890625, 8.080184936523438, 3.123411178588867, 15.216156005859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000114.npy"}
{"epoch": 0.2387434554973822, "step": 115, "batch_size": 128, "mean": 5.749182224273682, "std": 15.222766876220703, "min": -33.288482666015625, "p10": -11.807217788696288, "median": 5.067538261413574, "p90": 27.339093017578126, "max": 58.130615234375, "pos_frac": 0.6640625, "sample": [31.6959228515625, -15.927093505859375, 11.63360595703125, -10.7012939453125, -1.38262939453125, -13.209564208984375, -8.961883544921875, 2.603546142578125, 2.093841552734375, 20.341337203979492, -5.325996398925781, 9.628585815429688, 8.072235107421875, 8.116863250732422, 3.09735107421875, -9.624137878417969, 8.97726821899414, -6.218475341796875, 15.272430419921875, 2.0374298095703125, -11.73672103881836, 1.1593170166015625, 5.9083251953125, 6.494659423828125, 4.124101638793945, -3.56195068359375, 2.23211669921875, 12.744171142578125, 13.275726318359375, 15.55743408203125, -4.251678466796875, 13.797943115234375, 11.0743408203125, 3.7018814086914062, 3.1880149841308594, 8.944366455078125, -5.9891357421875, -20.041610717773438, -0.5636062622070312, 32.72711181640625, 6.6375732421875, 30.91436767578125, 37.9561767578125, 7.98529052734375, -19.69915771484375, -19.550704956054688, 23.58380889892578, -11.44671630859375, 26.487655639648438, 41.9144287109375, 6.0701751708984375, 58.130615234375, -11.971710205078125, -2.57366943359375, -13.269805908203125, 14.830490112304688, 27.371307373046875, 24.012420654296875, 12.915298461914062, -23.3985595703125, -16.375213623046875, -7.320220947265625, 27.325286865234375, 2.32977294921875, -3.8318634033203125, 12.53472900390625, 5.817047119140625, 0.778076171875, 5.792091369628906, 31.17822265625, 1.4250106811523438, 8.9259033203125, 14.790019989013672, 8.593788146972656, -3.2927684783935547, 7.312040328979492, 45.40631103515625, 1.793792724609375, -17.80816650390625, -8.381404876708984, -0.572509765625, 2.32562255859375, -11.62554931640625, 37.12933349609375, 20.96979522705078, 4.767852783203125, -25.733978271484375, 9.840736389160156, 1.6672248840332031, 32.720611572265625, -3.8793182373046875, 4.942466735839844, -5.085601806640625, 33.676116943359375, 13.595489501953125, -3.1419677734375, 2.415252685546875, 7.74169921875, -8.699478149414062, 7.003875732421875, 10.312530517578125, 5.384067535400391, 8.608955383300781, 8.300268173217773, 8.75048828125, 2.967132568359375, 39.5650634765625, 5.1273345947265625, -8.084625244140625, -12.441078186035156, 11.427505493164062, 5.007741928100586, 16.436309814453125, 6.966911315917969, 13.366180419921875, -10.997146606445312, -6.362251281738281, 13.551036834716797, -0.540252685546875, 21.922531127929688, -33.288482666015625, -2.7219772338867188, 10.415603637695312, -0.3532562255859375, 13.466323852539062, 3.209259033203125, -1.169342041015625, 12.114959716796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000115.npy"}
{"epoch": 0.24083769633507854, "step": 116, "batch_size": 128, "mean": 8.057055473327637, "std": 14.603961944580078, "min": -42.504180908203125, "p10": -7.570672607421875, "median": 7.206987380981445, "p90": 25.198344421386718, "max": 55.12017822265625, "pos_frac": 0.7265625, "sample": [1.47698974609375, -3.970531463623047, 5.0672454833984375, 2.6379776000976562, 7.3178558349609375, 11.881561279296875, 9.6571044921875, 25.587234497070312, 43.764495849609375, 7.162384033203125, 19.195281982421875, 10.893310546875, 21.238418579101562, 2.086986541748047, 16.91705322265625, 11.941123962402344, -7.61566162109375, 13.518310546875, 3.953031539916992, -10.0692138671875, -19.993896484375, -6.148065567016602, 22.232666015625, 23.915985107421875, 0.23426246643066406, -8.40716552734375, 8.700775146484375, -0.3935737609863281, 5.6045989990234375, 0.201263427734375, 2.107421875, 25.03167724609375, -9.77362060546875, 17.339569091796875, -5.560234069824219, 1.88818359375, 22.057418823242188, 13.95654296875, -6.9926910400390625, -4.2764739990234375, -3.2090530395507812, 19.4727783203125, 17.471282958984375, -9.918258666992188, 8.998672485351562, 3.4034347534179688, -2.98687744140625, 6.14727783203125, 21.930755615234375, 32.04669189453125, -42.504180908203125, 3.2603302001953125, 12.524032592773438, -17.550994873046875, -1.1669692993164062, 26.9033203125, 10.23187255859375, 24.241485595703125, -2.784515380859375, 0.7843170166015625, 9.894302368164062, 11.908843994140625, 4.845001220703125, 3.832427978515625, 9.360740661621094, 0.0, 8.496772766113281, 34.233489990234375, -1.9027786254882812, -7.5513916015625, 18.77655029296875, 33.07025146484375, -0.231658935546875, -6.296722412109375, -7.721595764160156, 2.2827682495117188, 2.3554611206054688, 0.6656684875488281, 6.70941162109375, 21.585174560546875, 3.8613204956054688, 38.404541015625, 14.49810791015625, 0.0, 11.806259155273438, 7.251590728759766, 20.229293823242188, 4.0628204345703125, 34.6146240234375, -1.44488525390625, -1.8820419311523438, 55.12017822265625, 12.943817138671875, -35.61968994140625, 9.02423095703125, 29.579986572265625, 19.652191162109375, 4.662422180175781, 9.56985092163086, 10.28839111328125, 10.963493347167969, -1.42333984375, 7.39105224609375, 29.432830810546875, 21.13446044921875, -21.880294799804688, 10.765228271484375, 13.619857788085938, 1.6788330078125, -13.992935180664062, 10.816741943359375, 7.5396575927734375, 44.438201904296875, -1.4579010009765625, 24.92816162109375, 16.809585571289062, 5.8049468994140625, 17.3289794921875, 21.969192504882812, 13.299003601074219, -7.4831695556640625, 6.5832061767578125, 1.2504730224609375, 28.359649658203125, -12.88812255859375, 6.163066864013672, -1.304229736328125, 18.85968017578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000116.npy"}
{"epoch": 0.24293193717277486, "step": 117, "batch_size": 128, "mean": 8.411303520202637, "std": 13.807329177856445, "min": -24.27362060546875, "p10": -7.075924682617187, "median": 7.167995452880859, "p90": 28.592860412597652, "max": 42.692352294921875, "pos_frac": 0.71875, "sample": [3.1250076293945312, 8.168827056884766, 11.73739242553711, 17.07843017578125, 4.39990234375, 13.472854614257812, 0.8491973876953125, -1.911590576171875, 10.07147216796875, -3.0285797119140625, 14.297866821289062, -6.9806365966796875, -1.339569091796875, 15.205535888671875, 18.479320526123047, 3.68682861328125, 8.951835632324219, 10.110809326171875, 25.652496337890625, 6.02069091796875, 18.11431884765625, -9.300186157226562, 8.262386322021484, 1.9769363403320312, 13.776317596435547, 28.165679931640625, 0.41180419921875, -18.650596618652344, 11.62255859375, 42.692352294921875, 16.598115921020508, 30.031524658203125, 34.5841064453125, 14.262792587280273, 9.28776741027832, -11.075714111328125, 25.255233764648438, 0.1199188232421875, 1.2468338012695312, -9.03643798828125, 34.175689697265625, 7.034477233886719, 4.60968017578125, 30.7021484375, 0.0, 22.3878173828125, -4.534637451171875, 5.15081787109375, 38.25897216796875, 24.941680908203125, 8.652801513671875, 34.018951416015625, -22.255767822265625, 17.83441162109375, 24.375, 23.240753173828125, -5.72540283203125, 2.319082260131836, 4.413482666015625, 7.415435791015625, 17.776611328125, 15.611328125, 14.132049560546875, 12.080093383789062, 6.076751708984375, 12.813751220703125, 13.367462158203125, -1.0989990234375, -7.2982635498046875, -8.29400634765625, 7.301513671875, -19.864410400390625, -0.1419239044189453, 9.864128112792969, 27.690338134765625, -13.695587158203125, 10.661651611328125, 11.275543212890625, 20.9066162109375, 1.1889381408691406, -2.0530624389648438, -0.7373046875, 6.1862030029296875, 31.872222900390625, 5.051116943359375, -6.567352294921875, -5.5697021484375, -13.193634033203125, -0.69122314453125, 4.417102813720703, -24.27362060546875, 13.8880615234375, 29.589614868164062, 29.5919189453125, 9.03103256225586, 4.4647216796875, -10.064498901367188, -2.816497802734375, -0.7736244201660156, 29.7879638671875, 4.1725921630859375, -4.6324462890625, 2.172384262084961, -4.515655517578125, -6.8595733642578125, 7.3350677490234375, 28.006118774414062, 6.384498596191406, 26.844970703125, -5.434017181396484, 16.416015625, 17.733642578125, 22.0604248046875, -0.4634552001953125, 0.025577545166015625, -3.339630126953125, 6.032131195068359, 0.048618316650390625, 4.8758697509765625, -3.3653182983398438, -20.75519561767578, 36.522857666015625, 12.595947265625, 14.52346420288086, 1.5909271240234375, 17.027400970458984, 38.57232666015625, 12.16912841796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000117.npy"}
{"epoch": 0.2450261780104712, "step": 118, "batch_size": 128, "mean": 8.955621719360352, "std": 15.163122177124023, "min": -25.351837158203125, "p10": -9.862434387207031, "median": 7.458168029785156, "p90": 26.148419189453122, "max": 49.2916259765625, "pos_frac": 0.7421875, "sample": [6.5037078857421875, 9.47613525390625, 19.027862548828125, 4.1771240234375, 28.715072631835938, 24.92169189453125, -21.82769775390625, -14.039234161376953, 18.68499755859375, 27.05328369140625, 15.6141357421875, 6.7824859619140625, 5.771778106689453, 6.105712890625, -13.576171875, 33.98455810546875, 18.563934326171875, 19.71563720703125, 22.54998016357422, 5.7957611083984375, 10.5848388671875, 20.33258056640625, 32.510948181152344, 2.3101806640625, -25.351837158203125, -25.109779357910156, 17.520477294921875, 1.83843994140625, -19.279800415039062, 22.110382080078125, 18.918426513671875, 24.221717834472656, 44.7410888671875, 15.219818115234375, 3.3808746337890625, 6.2136383056640625, 2.3603973388671875, 5.793785095214844, 17.992477416992188, 20.12578582763672, -3.420623779296875, 21.67657470703125, 30.860061645507812, -3.6214828491210938, 1.8570098876953125, -4.4394989013671875, 19.272659301757812, 18.21136474609375, 11.985107421875, 0.15571212768554688, 1.6330718994140625, 3.676849365234375, 25.655975341796875, 14.79437255859375, 27.447219848632812, 40.30096435546875, -8.603042602539062, 11.3056640625, -0.19752883911132812, 22.458236694335938, -6.1429901123046875, 0.0, -7.087371826171875, 9.670852661132812, -11.54498291015625, -7.469913482666016, -20.436767578125, 7.271965026855469, 15.871337890625, -10.0013427734375, -0.26458740234375, 32.051361083984375, 0.3207855224609375, 17.5267333984375, 6.447113037109375, 25.68115234375, 25.7606201171875, 16.254409790039062, -9.802902221679688, 25.5740966796875, 4.6533050537109375, 22.8717041015625, 9.325103759765625, -5.4090423583984375, 13.156517028808594, 25.2554931640625, -20.872528076171875, 1.684417724609375, -10.647216796875, -7.417549133300781, 49.0596923828125, -5.306915283203125, 25.141082763671875, 7.4682159423828125, -1.6749801635742188, 10.283073425292969, 1.4290084838867188, -13.457359313964844, 9.6728515625, -8.604843139648438, 5.111076354980469, 36.69123840332031, 8.25836181640625, 13.227081298828125, -0.1452789306640625, 10.00830078125, 4.528312683105469, 3.7976226806640625, 9.881431579589844, 33.561126708984375, -6.4280548095703125, -7.66473388671875, 17.12991714477539, 13.03271484375, 7.4481201171875, 1.8536224365234375, -22.27484130859375, 11.33843994140625, 6.1340179443359375, 21.11810302734375, 2.1384506225585938, 15.19793701171875, 4.189304351806641, 22.23455810546875, 0.0, 4.0844268798828125, 14.843170166015625, 49.2916259765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000118.npy"}
{"epoch": 0.24712041884816754, "step": 119, "batch_size": 128, "mean": 7.644491195678711, "std": 18.35783576965332, "min": -39.09455871582031, "p10": -15.69329833984375, "median": 5.071739196777344, "p90": 31.375390624999998, "max": 67.92462158203125, "pos_frac": 0.6875, "sample": [-39.09455871582031, -19.70476531982422, 24.738861083984375, 3.3838882446289062, 2.389678955078125, 2.6667709350585938, -9.180412292480469, -7.3607330322265625, 10.0753173828125, -4.4391632080078125, 22.0069580078125, -11.689247131347656, -12.28253173828125, 32.303741455078125, -15.7486572265625, 10.440582275390625, -15.466156005859375, 26.8524169921875, 25.94275665283203, -1.4864501953125, 7.9864501953125, 16.92181396484375, -1.4015560150146484, -8.343345642089844, 6.4338226318359375, 14.77117919921875, 21.616485595703125, 5.76397705078125, 33.92741394042969, 4.6720733642578125, 4.234853744506836, 2.2772064208984375, -2.79461669921875, 1.774658203125, 13.22772216796875, 9.777984619140625, 11.767791748046875, 3.4845619201660156, 9.141067504882812, 2.980377197265625, 23.368255615234375, -9.056068420410156, -1.378204345703125, 1.6622467041015625, 33.138397216796875, 0.8209629058837891, 5.07122802734375, -33.8106689453125, -15.72662353515625, -19.265472412109375, 50.642608642578125, 11.5152587890625, 4.08154296875, 29.350189208984375, -3.7455673217773438, -2.1924591064453125, 14.390472412109375, 1.9026832580566406, 15.419059753417969, 22.72303009033203, 4.2154541015625, 0.7230224609375, 23.899505615234375, 14.8770751953125, 19.1396484375, -9.35064697265625, 40.418731689453125, 12.148548126220703, 31.0343017578125, 18.845626831054688, 0.24579238891601562, 6.35638427734375, 13.451980590820312, 0.0, 37.546356201171875, -0.858489990234375, -2.1202316284179688, -26.96343994140625, 20.799560546875, -22.240921020507812, 21.358917236328125, 12.256149291992188, 4.030670166015625, -16.26275634765625, 13.081573486328125, 0.0956573486328125, 32.1712646484375, -7.616781234741211, 17.116973876953125, 6.8610382080078125, 23.637939453125, -15.988525390625, -2.49188232421875, 59.401123046875, -4.311607360839844, 67.92462158203125, 7.8909912109375, -0.656707763671875, 9.565277099609375, -27.0013427734375, 23.85247802734375, 32.43609619140625, 2.2705078125, 19.296920776367188, -9.352859497070312, 25.0804443359375, -7.810657501220703, 5.34521484375, 4.449378967285156, -15.67901611328125, 3.0688934326171875, -23.11761474609375, 47.339447021484375, 14.652946472167969, 58.1182861328125, 6.3063507080078125, -1.31671142578125, 13.65087890625, 0.200225830078125, 19.7811279296875, 6.8825225830078125, 5.0722503662109375, 28.09271240234375, -6.1065216064453125, 40.053131103515625, 4.4699249267578125, 6.167388916015625, -19.4188232421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000119.npy"}
{"epoch": 0.24921465968586387, "step": 120, "batch_size": 128, "mean": 9.025940895080566, "std": 17.009815216064453, "min": -30.72137451171875, "p10": -13.266122436523435, "median": 8.881050109863281, "p90": 32.032884216308595, "max": 57.746795654296875, "pos_frac": 0.6640625, "sample": [-12.710845947265625, 25.233001708984375, -5.820365905761719, 11.607101440429688, -4.629852294921875, 11.962310791015625, 23.10638427734375, -11.370956420898438, 7.278961181640625, 15.502716064453125, -3.268218994140625, -3.3836822509765625, 20.274532318115234, 3.5510406494140625, -15.991058349609375, -1.29241943359375, 33.16998291015625, -3.4501399993896484, 8.662261962890625, 17.098493576049805, -7.71966552734375, 32.02864074707031, 32.04278564453125, 24.75079345703125, 28.220748901367188, 12.201065063476562, 6.337371826171875, 0.352813720703125, -5.1477508544921875, 31.17462158203125, 26.72930908203125, -0.124603271484375, 4.70281982421875, 5.51458740234375, -24.7056884765625, 6.1362762451171875, -0.743316650390625, 15.818618774414062, 22.569091796875, 19.5650634765625, 0.22197723388671875, -1.5231895446777344, 16.711181640625, -25.700286865234375, 0.0, -2.14556884765625, -20.128204345703125, -2.3367462158203125, -6.4159393310546875, 4.9968414306640625, 11.78546142578125, 12.05239486694336, 50.3089599609375, 19.69189453125, 15.66851806640625, 9.099838256835938, -14.8770751953125, 14.391937255859375, 8.545623779296875, 3.4424667358398438, -2.8761367797851562, 0.0, 12.446319580078125, 20.326644897460938, 57.746795654296875, 14.776275634765625, 16.633941650390625, 39.11199951171875, -14.561767578125, 19.472557067871094, 21.25177001953125, 8.0762939453125, 3.6775283813476562, 13.98590087890625, 13.56878662109375, -7.8280792236328125, 7.769561767578125, 34.256988525390625, 17.976806640625, 18.98565673828125, 11.727691650390625, -17.153411865234375, -17.62786865234375, -7.49224853515625, -10.298835754394531, 36.335296630859375, 9.866752624511719, 37.292266845703125, 43.631072998046875, 16.142242431640625, -0.11285400390625, 26.37164306640625, -29.02362060546875, 32.448272705078125, 37.08348846435547, 5.57708740234375, -1.6693878173828125, 16.666534423828125, 11.875190734863281, 18.530624389648438, -21.521286010742188, 3.0834102630615234, 0.5623779296875, 18.960845947265625, 25.956649780273438, -0.9085006713867188, 11.632720947265625, -1.7217559814453125, 42.16162109375, -30.72137451171875, 22.6007080078125, 21.450714111328125, 20.22454071044922, -17.518218994140625, -22.51348876953125, 15.676544189453125, 14.207466125488281, 21.2015380859375, 7.287086486816406, -0.0090179443359375, -3.7515602111816406, 14.152030944824219, -3.4931793212890625, 49.11761474609375, 2.1175575256347656, 2.2933349609375, -4.1948089599609375, 14.996139526367188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000120.npy"}
{"epoch": 0.2513089005235602, "step": 121, "batch_size": 128, "mean": 5.410823822021484, "std": 15.32364273071289, "min": -35.439453125, "p10": -12.987806701660155, "median": 4.2655534744262695, "p90": 23.863360595703114, "max": 48.263214111328125, "pos_frac": 0.640625, "sample": [-0.9290008544921875, 0.9027099609375, 2.801055908203125, 6.568931579589844, 0.8489837646484375, 4.390678405761719, -3.4212646484375, 6.742755889892578, 8.313079833984375, 27.28460693359375, 3.2174110412597656, -1.2298736572265625, 33.71696472167969, 4.252140045166016, 17.40235137939453, 0.03900146484375, -2.705413818359375, 4.030220031738281, 0.5724067687988281, 21.834259033203125, 18.855422973632812, -16.300765991210938, 19.75933837890625, 16.16937255859375, -27.795623779296875, 0.0, 8.330802917480469, 3.711456298828125, 0.8451614379882812, 17.97003173828125, 9.954883575439453, 9.405410766601562, -35.439453125, 27.538726806640625, 5.732250213623047, -33.75921630859375, -2.9016571044921875, 5.765022277832031, 22.617401123046875, 13.0491943359375, -3.6536712646484375, -9.685638427734375, 7.582389831542969, 2.3616838455200195, 6.6983489990234375, 1.2133407592773438, 16.760406494140625, -12.595947265625, -8.031463623046875, 0.0, 7.7178955078125, -4.140542984008789, -0.8572921752929688, -2.11767578125, 15.535491943359375, -8.044334411621094, -0.557342529296875, 5.4748382568359375, 19.6201171875, -14.456039428710938, -2.461517333984375, -17.09478759765625, -2.3414955139160156, 46.569488525390625, 13.78997802734375, 9.104755401611328, 42.31581115722656, -7.8004150390625, -5.062896728515625, -1.4968109130859375, -5.083351135253906, 12.76910400390625, -2.91094970703125, 14.643089294433594, -9.301338195800781, -0.0152587890625, 4.278966903686523, 7.9313201904296875, -1.6494903564453125, 16.481433868408203, 15.19873046875, -1.3075847625732422, -4.6178741455078125, 28.910415649414062, 48.263214111328125, 6.582695007324219, -25.01227569580078, -1.282379150390625, -21.72760009765625, 46.407867431640625, 17.565811157226562, 18.793289184570312, 16.65357208251953, 2.6291351318359375, 33.480682373046875, 2.6476058959960938, -22.1339111328125, 4.5346527099609375, 12.900772094726562, 1.1581878662109375, 3.843830108642578, 7.1088714599609375, 13.453163146972656, 9.075286865234375, 4.90594482421875, 16.349365234375, -0.51287841796875, 4.461639404296875, -3.9542617797851562, -3.010913848876953, 5.99591064453125, 34.33404541015625, 9.934478759765625, 3.845703125, 8.2821044921875, 9.3853759765625, 28.72271728515625, -25.763336181640625, 26.770599365234375, 13.434600830078125, -0.8832645416259766, -13.902145385742188, 40.91754150390625, 13.679450988769531, 0.9481105804443359, -17.562088012695312, -22.487091064453125, 15.937713623046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000121.npy"}
{"epoch": 0.2534031413612565, "step": 122, "batch_size": 128, "mean": 7.446322441101074, "std": 18.371248245239258, "min": -37.261871337890625, "p10": -16.68065948486328, "median": 6.3603057861328125, "p90": 30.08456649780273, "max": 60.84100341796875, "pos_frac": 0.6484375, "sample": [22.491455078125, -15.934326171875, 25.6201171875, -17.6854248046875, 14.157989501953125, 22.6512451171875, 7.7547607421875, 3.8041229248046875, -34.75628662109375, 28.660491943359375, 24.721725463867188, 0.0, 8.94708251953125, -25.671051025390625, 0.3311805725097656, -37.261871337890625, 10.9273681640625, 17.288116455078125, -11.6693115234375, 13.548088073730469, -2.8284912109375, 22.97540283203125, -22.918006896972656, -31.292236328125, 6.281005859375, 34.06109619140625, -12.75250244140625, 10.657175064086914, 48.233367919921875, 2.5273590087890625, 9.626007080078125, -2.5541229248046875, 8.808197021484375, 7.536460876464844, 18.464263916015625, -4.775505065917969, 4.4618988037109375, -22.031951904296875, 0.376495361328125, 10.646385192871094, -27.567047119140625, 12.60919189453125, 6.439605712890625, 44.24676513671875, -2.818531036376953, -22.92327880859375, -10.775466918945312, 8.595680236816406, 39.676513671875, -8.23394775390625, -6.5384368896484375, 8.28570556640625, 13.115447998046875, -21.854019165039062, 10.09515380859375, 14.172218322753906, -0.8402462005615234, 22.871002197265625, -28.335113525390625, 11.46710205078125, 15.180908203125, 21.222442626953125, 19.2806396484375, 0.0318450927734375, -8.342567443847656, 32.947601318359375, 14.90350341796875, 26.10272216796875, 19.8560791015625, 29.03570556640625, 45.767974853515625, 12.860183715820312, 1.5001144409179688, 37.77801513671875, 60.84100341796875, 7.703819274902344, 5.825897216796875, -1.895721435546875, 31.58404541015625, 24.986083984375, -5.121681213378906, -16.250045776367188, 5.04449462890625, -10.459060668945312, -5.09161376953125, -2.89642333984375, -1.5863418579101562, 12.183860778808594, 5.2936248779296875, -5.52178955078125, 15.95391845703125, -1.2655181884765625, 2.3632774353027344, -18.961639404296875, -3.5065383911132812, -18.561767578125, -1.1137847900390625, 27.876800537109375, 23.9942626953125, -2.20599365234375, 5.844940185546875, -5.626312255859375, 22.548370361328125, 18.96624755859375, 2.2864608764648438, 25.08294677734375, 23.3828125, 29.441932678222656, 5.522407531738281, 1.6260604858398438, -1.056671142578125, -0.33481597900390625, -7.194427490234375, 14.878921508789062, 52.38311767578125, 0.28302001953125, 34.64788818359375, -2.3488616943359375, 2.5903167724609375, 25.892868041992188, 11.58050537109375, 22.736083984375, -4.078678131103516, 34.3031005859375, -5.54791259765625, 2.411865234375, 32.94476318359375, 7.5079345703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000122.npy"}
{"epoch": 0.2554973821989529, "step": 123, "batch_size": 128, "mean": 10.518993377685547, "std": 18.543909072875977, "min": -47.44842529296875, "p10": -12.566343688964844, "median": 9.701957702636719, "p90": 33.99500732421875, "max": 67.85397338867188, "pos_frac": 0.7265625, "sample": [24.898574829101562, -11.995574951171875, 29.330886840820312, -12.8037109375, 8.755996704101562, 5.846187591552734, 9.91644287109375, 3.35137939453125, 13.224853515625, 10.5489501953125, -0.8556785583496094, 4.34637451171875, -1.778167724609375, 2.44293212890625, 5.998865127563477, -7.219268798828125, -12.464614868164062, 16.3260498046875, 43.924774169921875, 8.794235229492188, -23.17010498046875, -23.8648681640625, 9.736465454101562, 16.973052978515625, 49.889923095703125, 2.912872314453125, 19.1600341796875, 29.583282470703125, 28.070068359375, 1.0873641967773438, 27.10943603515625, 21.572906494140625, 18.436309814453125, 24.088104248046875, 9.667449951171875, 25.35064697265625, 23.6307373046875, -23.423187255859375, 1.076568603515625, 28.9334716796875, 18.922653198242188, 4.710693359375, 0.66058349609375, 34.5732421875, 12.060836791992188, 55.97393798828125, 0.19745826721191406, 2.8106689453125, 17.1568603515625, 28.9866943359375, 3.2961654663085938, 18.5399169921875, 21.1649169921875, 67.85397338867188, 44.813507080078125, 12.988349914550781, -7.198673248291016, 0.1678619384765625, -0.5262451171875, 46.2918701171875, -3.567230224609375, 5.676704406738281, -47.44842529296875, 33.47242736816406, 10.409858703613281, -18.46099853515625, 22.817611694335938, 5.892547607421875, 36.8653564453125, 7.702150344848633, 12.52490234375, -0.651611328125, 9.039003372192383, -18.05634307861328, 12.010673522949219, 26.297653198242188, 23.603912353515625, -1.8865966796875, -16.746673583984375, 40.094451904296875, -2.38134765625, 10.652069091796875, 20.354595184326172, -4.01922607421875, 8.3924560546875, 19.459197998046875, -17.3660888671875, 17.62317657470703, 18.712677001953125, -7.239227294921875, 0.0, -5.381690979003906, 37.6351318359375, 11.817413330078125, 9.738372802734375, 0.0, 13.932159423828125, 8.890625, 0.4752063751220703, 22.016433715820312, -6.162849426269531, -16.614501953125, 6.2516021728515625, 27.2708740234375, -1.8725128173828125, 12.162025451660156, 8.304229736328125, 34.6500244140625, 16.593658447265625, -16.01776123046875, -2.468170166015625, -4.58258056640625, 32.39703369140625, 8.942245483398438, -32.197723388671875, 48.5655517578125, 14.324737548828125, -6.5108642578125, 13.6964111328125, -2.4819717407226562, 0.17968368530273438, 12.5391845703125, 6.488685607910156, 33.7471923828125, 45.408477783203125, -19.34161376953125, 17.564376831054688, 21.8389892578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000123.npy"}
{"epoch": 0.25759162303664923, "step": 124, "batch_size": 128, "mean": 9.586933135986328, "std": 17.50368881225586, "min": -51.185943603515625, "p10": -7.38549938201904, "median": 7.513836860656738, "p90": 33.25875549316406, "max": 57.868988037109375, "pos_frac": 0.7734375, "sample": [-1.087860107421875, 29.278717041015625, -51.185943603515625, 16.55743408203125, -4.282386779785156, 1.995391845703125, 25.976470947265625, 7.3974761962890625, 34.049224853515625, 46.05865478515625, 4.11383056640625, -28.14569091796875, 7.5816650390625, -6.59716796875, 22.99798583984375, 5.0901031494140625, 5.631988525390625, 19.9117431640625, -6.162349700927734, 12.24847412109375, 5.28399658203125, 5.728363037109375, 0.1975555419921875, 10.323089599609375, 20.78472900390625, 3.7947998046875, -1.6583786010742188, -5.292865753173828, 4.071510314941406, 7.9725799560546875, 29.7569580078125, -14.045166015625, 4.239341735839844, 23.606475830078125, 44.273406982421875, 38.05242919921875, 19.644763946533203, 12.545944213867188, 4.572235107421875, 24.36981201171875, 21.2301025390625, 18.377235412597656, 38.73004150390625, 10.8482666015625, 3.0559654235839844, -2.041534423828125, 0.4058990478515625, 13.780044555664062, 45.450531005859375, -9.224939346313477, 9.88848876953125, -17.620899200439453, 6.955497741699219, 0.0, 2.703828811645508, -2.647308349609375, 26.5653076171875, 5.430755615234375, 3.4746551513671875, -12.153236389160156, 32.91998291015625, -38.39288330078125, -27.9246826171875, 39.68524169921875, 12.95001220703125, 26.092132568359375, 8.605133056640625, 32.8089599609375, 24.57390785217285, -0.8238372802734375, 10.76220703125, 5.702056884765625, -20.159271240234375, -29.073974609375, 22.576316833496094, 1.8508567810058594, 4.017597198486328, 11.477066040039062, 34.945037841796875, 20.698776245117188, 10.157806396484375, 31.12139892578125, 7.843635559082031, -16.5545654296875, -5.685577392578125, 6.455806732177734, 3.0498809814453125, 14.774551391601562, 35.53216552734375, 16.055988311767578, 2.410400390625, 13.448158264160156, -4.2398681640625, 12.79278564453125, 11.178466796875, 7.458730697631836, -0.70166015625, 57.868988037109375, 11.98042106628418, 19.62701416015625, -3.6979217529296875, 2.846893310546875, 17.204742431640625, 1.2497634887695312, 1.733551025390625, 54.87298583984375, -10.336959838867188, 0.0377197265625, 7.568943023681641, 20.034423828125, -2.2281036376953125, 0.5142669677734375, 6.320404052734375, 12.776809692382812, 7.63311767578125, 21.830013275146484, 9.256942749023438, 35.21893310546875, 0.12948226928710938, -2.2657318115234375, 5.6056365966796875, 15.180206298828125, 25.179168701171875, 44.226043701171875, 4.686187744140625, 10.090667724609375, 4.44195556640625, -13.705856323242188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000124.npy"}
{"epoch": 0.25968586387434556, "step": 125, "batch_size": 128, "mean": 9.355475425720215, "std": 18.971237182617188, "min": -33.890892028808594, "p10": -11.743881225585938, "median": 6.3267822265625, "p90": 35.2380126953125, "max": 64.532958984375, "pos_frac": 0.6484375, "sample": [5.5932159423828125, -8.66389274597168, -33.890892028808594, 36.68475341796875, 1.9825897216796875, 14.104385375976562, 16.637496948242188, -10.457916259765625, 3.308701515197754, -15.144256591796875, 3.877105712890625, -3.5356521606445312, 34.282012939453125, -1.3896484375, 10.93658447265625, 25.345870971679688, -28.887908935546875, 15.115463256835938, 13.705474853515625, 14.620346069335938, -8.92108154296875, -2.9104957580566406, 20.6383056640625, -5.942268371582031, 16.053375244140625, 13.634803771972656, 11.124458312988281, 15.853546142578125, -4.408355712890625, 44.981689453125, 4.196746826171875, -7.068023681640625, 10.563522338867188, -1.3603572845458984, 7.2949066162109375, 32.535400390625, 21.355712890625, 36.765655517578125, 14.539871215820312, -4.141685485839844, 0.48300933837890625, -23.70733642578125, -12.208297729492188, -6.936576843261719, -5.43060302734375, 43.7537841796875, 2.1060791015625, -7.96893310546875, -31.796417236328125, 14.4642333984375, 17.6156005859375, -17.108306884765625, 24.99713134765625, 64.532958984375, 5.5994110107421875, 13.912521362304688, 9.871116638183594, 21.217315673828125, 21.23736572265625, 28.167877197265625, -2.7034912109375, 35.737762451171875, 15.026336669921875, -1.014719009399414, 0.270904541015625, 14.62152099609375, 4.0017547607421875, 1.3828315734863281, 5.84613037109375, 5.230377197265625, 55.780242919921875, 10.495712280273438, 29.174217224121094, 41.670654296875, -12.532699584960938, -19.495742797851562, 6.9771728515625, 22.434112548828125, 35.023834228515625, 24.906330108642578, 32.241851806640625, 4.0852203369140625, -1.0705718994140625, 44.1024169921875, -0.4591064453125, 14.369155883789062, -15.012802124023438, -7.19573974609375, 21.329744338989258, 32.04754638671875, 28.177780151367188, 2.61163330078125, -5.0753936767578125, 13.2178955078125, -9.1519775390625, -1.5491256713867188, -10.0389404296875, 27.961334228515625, 25.809417724609375, -0.43526458740234375, -8.175628662109375, 0.0, 3.458688735961914, -3.749298095703125, 39.89910888671875, -9.096359252929688, 0.8428955078125, 4.5145416259765625, 6.80743408203125, 45.7830810546875, -7.367218017578125, -11.544845581054688, 20.688262939453125, 31.42242431640625, 0.6896457672119141, 25.746337890625, 50.86907958984375, 11.722869873046875, -14.6539306640625, 17.89361572265625, 25.90765380859375, -30.60809326171875, -6.882537841796875, 47.384765625, 13.987930297851562, 18.419952392578125, -2.3704452514648438, -14.6988525390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000125.npy"}
{"epoch": 0.2617801047120419, "step": 126, "batch_size": 128, "mean": 10.690484046936035, "std": 19.144243240356445, "min": -43.6553955078125, "p10": -11.027044677734374, "median": 9.236248016357422, "p90": 34.385256958007815, "max": 53.31964111328125, "pos_frac": 0.7109375, "sample": [22.172637939453125, 51.372589111328125, 14.151962280273438, 18.113616943359375, 27.91607666015625, 41.523284912109375, 12.524421691894531, 39.896568298339844, 4.102756500244141, -1.6904525756835938, -42.80206298828125, 24.646873474121094, 2.5892410278320312, 19.4281005859375, 27.735843658447266, 4.19732666015625, 27.818572998046875, -4.4567108154296875, -21.16864013671875, -8.142791748046875, -14.60196304321289, 28.886993408203125, -11.4415283203125, -6.674814224243164, 5.305419921875, -9.027130126953125, -3.9288330078125, -3.6102142333984375, 4.92034912109375, -6.855712890625, 8.41497802734375, -3.532318115234375, 16.62683868408203, 12.954757690429688, 31.658660888671875, -10.872283935546875, -8.737495422363281, 0.377655029296875, -2.1077423095703125, 4.973150253295898, 7.617984771728516, 23.290279388427734, 32.68682861328125, -0.7278709411621094, 14.569000244140625, 24.516357421875, 8.441551208496094, 17.505767822265625, -13.622543334960938, -3.9526824951171875, 34.380706787109375, 0.0155792236328125, 13.880599975585938, 49.73284912109375, 3.9540252685546875, 37.85784149169922, 16.300559997558594, 6.7460784912109375, 4.282569885253906, 21.25419807434082, 12.013126373291016, 29.364013671875, 18.2120361328125, 24.4244384765625, -14.083160400390625, 16.7125244140625, 5.5157470703125, 24.1510009765625, 21.329360961914062, 48.648193359375, 41.64085388183594, -4.9195556640625, 27.108612060546875, 28.257339477539062, -2.858306884765625, 6.4054718017578125, 10.80078125, -7.9285125732421875, 0.04926872253417969, -24.34393310546875, 27.87218475341797, 11.929840087890625, -2.9253292083740234, 45.9425048828125, 21.385833740234375, 3.899005889892578, -28.04638671875, 6.314788818359375, 21.3424072265625, -36.23736572265625, 13.3363037109375, -9.95907211303711, 24.3883056640625, 30.996429443359375, -43.6553955078125, -9.9056396484375, 4.345827102661133, 10.03094482421875, 6.323577880859375, -16.321197509765625, 24.46209716796875, 27.087364196777344, 34.3958740234375, 43.57568359375, 7.72528076171875, 19.9049072265625, -7.1822509765625, 46.342926025390625, 22.767730712890625, 38.323028564453125, -10.6846923828125, 19.677032470703125, 17.23786163330078, 33.786163330078125, 20.566932678222656, 6.924774169921875, 3.7521438598632812, -1.054962158203125, 10.45404052734375, -1.019287109375, 14.594024658203125, -11.388153076171875, 1.3231201171875, 4.652862548828125, 29.532638549804688, 53.31964111328125, 4.7836151123046875, -18.4229736328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000126.npy"}
{"epoch": 0.2638743455497382, "step": 127, "batch_size": 128, "mean": 12.326555252075195, "std": 19.33078384399414, "min": -43.036102294921875, "p10": -10.013831329345702, "median": 8.876304626464844, "p90": 37.80677795410156, "max": 55.7584228515625, "pos_frac": 0.734375, "sample": [-0.78826904296875, -1.4198760986328125, -9.2860107421875, 25.826171875, 9.8509521484375, 4.345184326171875, 18.791534423828125, 12.047298431396484, -9.985649108886719, -43.036102294921875, 29.582275390625, 1.8220062255859375, 19.629913330078125, 18.102705001831055, 10.49261474609375, -34.9407958984375, -10.07958984375, 13.495681762695312, -17.503326416015625, 8.23956298828125, 7.231170654296875, 6.504913330078125, 21.588958740234375, 3.93878173828125, 19.535781860351562, 49.6619873046875, 23.6376953125, 35.240386962890625, 11.238609313964844, 24.819290161132812, 37.801422119140625, 15.293146133422852, 29.9215087890625, 4.123271942138672, 9.7108154296875, 46.270751953125, -8.361778259277344, -18.615524291992188, -0.5423355102539062, 9.399600982666016, 47.4720458984375, 7.3546295166015625, 7.6005401611328125, 7.18975830078125, -15.12457275390625, -24.133697509765625, 36.20948791503906, 36.30786895751953, 11.08612060546875, -22.54571533203125, 18.82952880859375, 35.981231689453125, 20.281524658203125, -8.72259521484375, 29.025848388671875, 39.8060302734375, 24.699859619140625, 14.982376098632812, -4.52435302734375, 3.0514144897460938, 2.648895263671875, 4.250091552734375, 36.795928955078125, 23.261962890625, 8.804718017578125, 34.34423828125, -6.8505401611328125, 5.127960205078125, 29.95269775390625, 42.96832275390625, 25.636062622070312, -2.29498291015625, -11.433578491210938, -2.4000587463378906, 9.539337158203125, -10.31805419921875, -0.98919677734375, -15.13443374633789, 7.2068634033203125, -3.2472801208496094, 4.28851318359375, 21.790634155273438, 55.027740478515625, -1.9765205383300781, 39.54339599609375, 17.741989135742188, 4.3695831298828125, 55.7584228515625, 31.21539306640625, 21.2991943359375, 46.565521240234375, -7.50604248046875, 7.8060302734375, 6.4991455078125, 4.141880035400391, -2.109708786010742, -20.585844039916992, 7.901947021484375, 8.393890380859375, 31.164108276367188, -6.17840576171875, 8.947891235351562, -9.824966430664062, 30.498497009277344, 23.008529663085938, 4.006797790527344, 39.034393310546875, 48.415252685546875, 2.758636474609375, -3.6313629150390625, 6.2475128173828125, 2.763092041015625, 23.97247314453125, 35.61553955078125, -16.715003967285156, 5.063312530517578, 36.59161376953125, 37.81927490234375, -1.9633216857910156, 11.006500244140625, 31.404541015625, 23.16094970703125, 3.54498291015625, 31.1724853515625, 1.92474365234375, -8.89923095703125, 22.592086791992188, 47.8519287109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000127.npy"}
{"epoch": 0.26596858638743454, "step": 128, "batch_size": 128, "mean": 13.686408996582031, "std": 20.826194763183594, "min": -42.233642578125, "p10": -10.018447875976562, "median": 12.690363883972168, "p90": 38.13626785278319, "max": 74.34564208984375, "pos_frac": 0.7265625, "sample": [26.702957153320312, 3.884857177734375, 20.053024291992188, 23.499053955078125, -33.190399169921875, -32.580841064453125, -20.08465576171875, 8.025588989257812, 25.384490966796875, 27.207473754882812, 23.82085418701172, 2.8633880615234375, 35.14727783203125, 0.403167724609375, 40.04975891113281, 0.7207603454589844, 3.9011573791503906, 23.9437255859375, 26.930130004882812, 25.092758178710938, -3.1624755859375, -1.4463119506835938, 36.851341247558594, 67.30316162109375, 4.416900634765625, 7.8224334716796875, -3.564208984375, 4.846099853515625, 28.2418212890625, 22.0150146484375, 8.94134521484375, 11.859556198120117, -19.484039306640625, 34.80485534667969, -7.68586540222168, 31.513137817382812, -16.546401977539062, 7.4616241455078125, 1.0755462646484375, -21.597076416015625, 25.03784942626953, 2.487030029296875, -3.2970809936523438, 18.273040771484375, 33.24688720703125, -2.4426422119140625, 0.6107597351074219, 24.14456558227539, 19.876693725585938, 37.316200256347656, 36.78950500488281, 26.53314208984375, -4.471672058105469, 35.00823974609375, -2.837982177734375, -6.35614013671875, 3.80157470703125, 25.342540740966797, 18.976409912109375, 41.943695068359375, 26.96158218383789, 4.350799560546875, 14.20987319946289, -0.05486297607421875, 1.1935081481933594, 7.389923095703125, 41.023223876953125, 29.195587158203125, 30.133636474609375, 13.521171569824219, 14.042755126953125, 28.820755004882812, 50.521484375, 16.81396484375, 47.309783935546875, 34.936492919921875, 11.03900146484375, -5.815948486328125, -0.37841796875, -2.1478919982910156, 40.71311950683594, 31.583984375, -13.548553466796875, 19.885772705078125, 29.868988037109375, -42.233642578125, 20.65960693359375, -20.992996215820312, 19.101097106933594, 74.34564208984375, 9.66326904296875, -9.81982421875, 35.489837646484375, 16.728652954101562, -5.777557373046875, 0.8754825592041016, -15.62969970703125, -7.83038330078125, 7.7120361328125, 11.687347412109375, 36.739044189453125, -5.60552978515625, 4.722412109375, 45.777000427246094, 23.34832763671875, 35.961578369140625, -7.521461486816406, -5.943115234375, 55.56451416015625, 1.1865081787109375, 8.265228271484375, 0.0, -22.664840698242188, -8.073898315429688, 27.093994140625, -11.48565673828125, -10.481903076171875, 13.8809814453125, 23.842514038085938, 42.634490966796875, 4.481536865234375, 19.0244140625, 11.2457275390625, 40.399200439453125, 61.843048095703125, 27.000335693359375, 20.68084716796875, -1.02716064453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000128.npy"}
{"epoch": 0.2680628272251309, "step": 129, "batch_size": 128, "mean": 12.588574409484863, "std": 21.516963958740234, "min": -46.897003173828125, "p10": -12.846661376953124, "median": 11.115862846374512, "p90": 42.3399658203125, "max": 79.72299194335938, "pos_frac": 0.7578125, "sample": [-46.897003173828125, 28.8759765625, -22.365219116210938, -14.081787109375, 12.296226501464844, 44.97334671020508, 28.536651611328125, 36.64519500732422, 31.25146484375, 2.4937591552734375, 31.372039794921875, 2.381467819213867, 15.4400634765625, -0.196533203125, -6.613945007324219, -10.876983642578125, 28.336151123046875, 35.834747314453125, 13.55950927734375, -10.860198974609375, 13.126312255859375, 12.294998168945312, 1.2078437805175781, 19.254058837890625, 43.01226806640625, -5.5364837646484375, 16.27587890625, -2.3023910522460938, 79.72299194335938, 59.750244140625, 5.38287353515625, 22.287094116210938, 52.27183532714844, 14.003387451171875, 1.1248016357421875, -43.920501708984375, 34.118499755859375, 4.319122314453125, 64.50201416015625, 10.533618927001953, -4.533092498779297, 14.1151123046875, 40.0697021484375, 43.069610595703125, 10.732742309570312, 42.13580322265625, -24.721771240234375, 10.823431015014648, 0.9610939025878906, 4.484504699707031, 26.195327758789062, 17.866195678710938, 10.63275146484375, 25.862060546875, -12.31732177734375, 8.707799911499023, 4.98541259765625, 44.97235107421875, 2.0469512939453125, 28.538551330566406, 1.9918899536132812, 4.17901611328125, 42.81634521484375, 22.602157592773438, 59.277679443359375, -6.28143310546875, 15.652099609375, 17.46044921875, 27.326171875, 3.7548980712890625, 5.069183349609375, 6.473293304443359, 3.0288238525390625, -4.899501800537109, 4.67437744140625, -15.080230712890625, 27.022178649902344, 20.498809814453125, -2.8429603576660156, 60.3502197265625, 27.901016235351562, 10.139114379882812, 20.383056640625, 3.8555831909179688, 2.5678367614746094, 2.495391845703125, 30.229690551757812, 3.146209716796875, 19.816009521484375, 39.36396789550781, -9.771636962890625, -28.33160400390625, -2.793792724609375, -0.68524169921875, -7.306358337402344, 13.760498046875, -10.20306396484375, -31.063751220703125, 16.361953735351562, 11.408294677734375, 45.4814453125, -17.52459716796875, 14.48419189453125, -2.7906742095947266, -20.504409790039062, -21.387451171875, 12.498397827148438, 4.11553955078125, 15.230430603027344, 21.902923583984375, 27.306503295898438, 6.570850372314453, 16.8143310546875, 8.6993408203125, 10.106719970703125, 45.663177490234375, 10.741924285888672, 31.040634155273438, -19.3658447265625, 20.41424560546875, 12.037422180175781, -0.3852519989013672, 2.0712738037109375, 30.309463500976562, 12.843681335449219, -14.086097717285156, 26.4267578125, 32.11529541015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000129.npy"}
{"epoch": 0.27015706806282724, "step": 130, "batch_size": 128, "mean": 11.678691864013672, "std": 19.64021110534668, "min": -38.734130859375, "p10": -10.870077514648434, "median": 9.692193984985352, "p90": 37.7603889465332, "max": 57.278228759765625, "pos_frac": 0.71875, "sample": [8.26434326171875, -3.7061920166015625, 31.356719970703125, 4.7141876220703125, 5.46728515625, 42.38154602050781, 9.238983154296875, 28.80047607421875, 38.63121795654297, -8.933258056640625, -2.9794044494628906, 9.87774658203125, 22.87755584716797, 40.009185791015625, -1.762969970703125, 6.5128173828125, -2.607666015625, -17.3258056640625, 8.045646667480469, 20.24395751953125, 16.46172332763672, 40.27001953125, -1.670013427734375, 10.557861328125, -2.168916702270508, 57.278228759765625, 1.647735595703125, 21.06658935546875, 46.115966796875, 49.40562438964844, -29.9637451171875, -0.411407470703125, 0.0, -10.151901245117188, 25.767547607421875, 13.212875366210938, 24.10406494140625, -38.733673095703125, 9.59555435180664, 21.82806396484375, 26.10186004638672, 14.924789428710938, 48.69903564453125, -9.715370178222656, -38.734130859375, -36.62644958496094, -4.513702392578125, 35.18000793457031, 2.565521240234375, -0.169403076171875, 35.00482177734375, 1.3960113525390625, 11.471427917480469, 33.276832580566406, -15.85894775390625, -1.90521240234375, 12.848562240600586, 20.18707275390625, 37.387176513671875, -3.00347900390625, 9.788833618164062, 1.4596977233886719, 23.692901611328125, 17.93682861328125, 6.1883544921875, 43.35246276855469, 26.6845703125, -18.426544189453125, -12.545822143554688, -0.9063720703125, 41.67724609375, 6.893989562988281, 30.21929931640625, -22.8614501953125, 5.145408630371094, 17.83575439453125, -14.866439819335938, 29.110870361328125, 9.120162963867188, 15.311651229858398, 28.075027465820312, 18.016021728515625, 31.7237548828125, 17.80755615234375, 31.6182861328125, 1.2778339385986328, 33.19384765625, 8.841873168945312, 28.35589599609375, -4.174560546875, 11.42352294921875, 7.3359375, 7.186279296875, 3.5893783569335938, 23.225906372070312, -28.770294189453125, -5.6804962158203125, 5.008392333984375, 2.824371337890625, 46.55633544921875, 43.4014892578125, 34.6832275390625, 21.543807983398438, 1.0179519653320312, -25.52154541015625, 26.776031494140625, 6.112983703613281, 3.9150390625, 19.531494140625, -5.3017120361328125, 34.450653076171875, 8.274188995361328, 28.32708740234375, -5.8113861083984375, 19.43603515625, 18.216411590576172, 39.7637939453125, 20.68408203125, -0.7236480712890625, 17.841903686523438, 15.566818237304688, -7.079803466796875, -1.702911376953125, 1.732421875, 29.898193359375, -25.188507080078125, 4.596109390258789, 26.35107421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000130.npy"}
{"epoch": 0.27225130890052357, "step": 131, "batch_size": 128, "mean": 9.653390884399414, "std": 22.120092391967773, "min": -46.29978942871094, "p10": -18.138761901855467, "median": 8.476333618164062, "p90": 39.82104492187499, "max": 70.8197021484375, "pos_frac": 0.625, "sample": [31.749618530273438, 8.932807922363281, 37.456024169921875, -22.79583740234375, 13.565643310546875, 2.3388214111328125, 2.541088104248047, 24.4571533203125, -17.893386840820312, 41.27020263671875, 29.397308349609375, 30.3359375, 41.206695556640625, 18.3692626953125, -15.80810546875, -0.042816162109375, 0.506591796875, -15.1199951171875, 9.601799011230469, 48.06085205078125, 32.249114990234375, 11.404266357421875, -6.462158203125, -5.856130599975586, 30.47388458251953, -6.0829010009765625, -9.980224609375, 44.62310791015625, 10.232635498046875, 14.9583740234375, -1.7159194946289062, 59.935791015625, -7.098236083984375, 0.1091156005859375, 20.19647216796875, -2.459575653076172, -23.38855743408203, 43.2515869140625, -3.37933349609375, 30.23883056640625, 7.49578857421875, 3.22357177734375, -2.330280303955078, -15.649658203125, 26.64544677734375, 7.0052032470703125, 56.877685546875, -17.473388671875, 19.193222045898438, 25.6080322265625, -0.423858642578125, -2.60687255859375, -15.50323486328125, 12.8721923828125, 13.9937744140625, 39.23291015625, -23.339492797851562, -4.1591796875, 25.795154571533203, 35.01042175292969, -8.369232177734375, -18.7113037109375, -19.21527099609375, 29.919212341308594, 7.235076904296875, 5.76922607421875, 16.81341552734375, 28.69818115234375, 70.8197021484375, 8.48065185546875, -46.29978942871094, 3.1119613647460938, 9.909881591796875, -9.944931030273438, -3.8854026794433594, 19.772979736328125, -28.002243041992188, 62.618072509765625, -7.10386848449707, 35.5789794921875, 11.533447265625, -23.217681884765625, 1.8557395935058594, -13.157318115234375, 12.619384765625, 19.057098388671875, -15.0943603515625, 38.69549560546875, 42.3304443359375, -25.892623901367188, -0.6703414916992188, -3.9582138061523438, 44.4554443359375, -22.284725189208984, 27.6549072265625, 7.87469482421875, 1.8401012420654297, 7.679046630859375, -7.21136474609375, 17.176353454589844, 11.422615051269531, -22.71575927734375, 29.644287109375, -3.7905654907226562, 10.65936279296875, 34.04119110107422, -2.97412109375, -33.810699462890625, 52.44049072265625, 8.472015380859375, 31.7559814453125, 41.193359375, 13.922332763671875, 20.029205322265625, -12.786163330078125, -3.4957733154296875, 20.288375854492188, -9.74945068359375, 27.846689224243164, 9.063232421875, 21.666412353515625, 15.078643798828125, 23.1505126953125, 17.795684814453125, -5.557647705078125, -21.102134704589844, -8.44329833984375, 2.2611732482910156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000131.npy"}
{"epoch": 0.2743455497382199, "step": 132, "batch_size": 128, "mean": 13.011011123657227, "std": 24.8736572265625, "min": -46.674560546875, "p10": -17.257180786132814, "median": 10.327079772949219, "p90": 45.15086822509765, "max": 96.55438232421875, "pos_frac": 0.703125, "sample": [-8.19818115234375, 27.44854736328125, 37.56193542480469, 13.47705078125, 0.14962005615234375, -9.42041015625, 27.440902709960938, 11.80755615234375, -3.47100830078125, 38.3951416015625, 3.417072296142578, -10.986824035644531, -20.873077392578125, 36.90472412109375, 22.478973388671875, 9.190811157226562, 3.05908203125, 45.21186828613281, 42.86244201660156, 60.495819091796875, 35.329071044921875, 31.354873657226562, 71.06375122070312, 49.3052978515625, 15.053375244140625, 2.1522293090820312, -0.9317626953125, -32.84910583496094, 14.030364990234375, 21.00634765625, -1.4632568359375, 46.167510986328125, 2.7291717529296875, 19.138031005859375, -21.72930908203125, -17.470611572265625, -16.272369384765625, 10.134658813476562, 4.6299285888671875, 2.27410888671875, 41.818328857421875, 16.400497436523438, -12.35693359375, -15.424331665039062, -23.482158660888672, 0.6398849487304688, 5.94696044921875, -17.16571044921875, 13.845672607421875, -13.250652313232422, 15.083587646484375, -10.05974006652832, 16.35600471496582, 30.806396484375, 3.57427978515625, -0.02899169921875, 3.9970703125, 17.55169677734375, 96.55438232421875, -0.040771484375, 25.482574462890625, 29.41644287109375, 21.373977661132812, -24.651611328125, 35.73816680908203, -2.1556396484375, -2.23431396484375, 11.188720703125, -19.7379150390625, 0.21600341796875, 17.39739990234375, 4.855003356933594, 13.7872314453125, 37.61994934082031, -5.55499267578125, -1.6336822509765625, 0.0, 37.503173828125, 13.023269653320312, -19.637451171875, 8.687850952148438, 36.890838623046875, 50.06965637207031, -33.213470458984375, -46.674560546875, 56.86798095703125, 13.466156005859375, 19.301300048828125, 3.791351318359375, 8.66094970703125, 21.901123046875, 12.493446350097656, 33.56752014160156, 25.623458862304688, 53.58012390136719, 90.81903076171875, -15.015228271484375, 27.376937866210938, 0.034576416015625, 15.0230712890625, 25.295166015625, 40.429962158203125, 2.454803466796875, 0.9833717346191406, 10.519500732421875, 51.63714599609375, 4.9493408203125, 6.3265380859375, 0.0, -17.76116943359375, 31.750228881835938, 40.66290283203125, -10.090888977050781, 1.3491859436035156, -8.700164794921875, 9.3662109375, -2.576751708984375, 55.392913818359375, -27.406646728515625, 45.124725341796875, 34.72645568847656, 28.04486083984375, -24.054351806640625, 14.29263687133789, -16.40826416015625, 17.245880126953125, 2.118743896484375, 63.118804931640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000132.npy"}
{"epoch": 0.2764397905759162, "step": 133, "batch_size": 128, "mean": 13.570047378540039, "std": 22.3725528717041, "min": -65.3787841796875, "p10": -12.663335418701168, "median": 11.523174285888672, "p90": 42.98012084960938, "max": 69.8863525390625, "pos_frac": 0.7578125, "sample": [21.214996337890625, 1.5207042694091797, -33.659149169921875, 0.0, 0.8732757568359375, 32.360565185546875, -0.9401378631591797, 13.98508071899414, 4.338401794433594, 29.245941162109375, 17.3211669921875, 17.5838623046875, 4.90728759765625, 25.37030029296875, 37.5751953125, 50.50543212890625, -9.45526123046875, -1.0640869140625, -1.8209800720214844, -16.431610107421875, 33.67730712890625, 50.159942626953125, 69.8863525390625, 5.424957275390625, 25.310882568359375, 44.81280517578125, 23.24349594116211, 53.358306884765625, 0.0, -1.205291748046875, 22.07269287109375, 23.9820556640625, -4.123298645019531, 2.64141845703125, 28.282058715820312, 32.26469421386719, 23.781829833984375, 27.150115966796875, -27.23419189453125, 33.686279296875, 43.670166015625, 7.487518310546875, -5.95806884765625, -6.524688720703125, 10.641799926757812, -1.187530517578125, -17.221084594726562, 7.799690246582031, 4.1363677978515625, 8.64752197265625, 16.360595703125, 39.89190673828125, 2.71429443359375, 29.613990783691406, 47.096466064453125, 9.851577758789062, 5.0257415771484375, 41.915313720703125, -4.317596435546875, 30.351226806640625, -20.22760009765625, 11.239509582519531, 35.61065673828125, 0.68206787109375, 18.41444969177246, 8.736259460449219, 24.937484741210938, 44.9981689453125, 31.080078125, 27.651443481445312, -37.22357177734375, 0.4470062255859375, 11.806838989257812, 7.6579437255859375, 50.719482421875, 9.24798583984375, -6.665046691894531, -19.863983154296875, 3.435199737548828, 43.01690673828125, -6.602878570556641, 2.553985595703125, 10.49371337890625, -9.035400390625, 20.475830078125, -11.558845520019531, 6.312042236328125, 26.442344665527344, 25.59051513671875, 41.40821838378906, -5.6069183349609375, 45.825775146484375, 28.899795532226562, 12.858963012695312, 16.204833984375, 4.400146484375, 7.179473876953125, 42.96435546875, 24.89666748046875, 45.46974182128906, 9.742912292480469, 29.562591552734375, 20.772377014160156, 24.114452362060547, -29.219558715820312, 21.681201934814453, 23.005035400390625, 28.3037109375, -26.324058532714844, 49.733642578125, 21.95166015625, 7.980010986328125, 4.9566650390625, -15.240478515625, -49.571197509765625, 17.671279907226562, -65.3787841796875, 4.654050827026367, 2.5985107421875, 38.92572021484375, 6.309844970703125, 25.85699462890625, 36.385982513427734, -4.6627044677734375, -22.837860107421875, 1.9914875030517578, 28.112396240234375, 40.417938232421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000133.npy"}
{"epoch": 0.27853403141361255, "step": 134, "batch_size": 128, "mean": 9.352567672729492, "std": 21.233110427856445, "min": -45.3677978515625, "p10": -15.9775146484375, "median": 7.536694526672363, "p90": 34.372805786132815, "max": 90.56903076171875, "pos_frac": 0.6796875, "sample": [40.07325744628906, 26.291366577148438, -19.79949951171875, -5.890106201171875, -25.64154052734375, -15.92584228515625, 7.800048828125, -17.174606323242188, -4.945587158203125, 24.02496337890625, -39.490478515625, -6.6561279296875, 11.13116455078125, 15.56195068359375, 27.053268432617188, -0.2552337646484375, -5.991512298583984, 8.783184051513672, 24.408111572265625, 8.085308074951172, 14.93115234375, 59.54693603515625, 0.0, 21.0693359375, 21.7977294921875, 21.80609130859375, 34.323028564453125, -3.6393203735351562, 6.16705322265625, 45.64360046386719, -28.820619583129883, -5.125274658203125, 17.098175048828125, 2.9074478149414062, 28.905956268310547, 2.5294036865234375, 26.573410034179688, 14.814186096191406, 13.964080810546875, 4.9385986328125, -9.894577026367188, 25.67359161376953, 14.431777954101562, 25.4873046875, 90.56903076171875, 35.3599853515625, 3.925567626953125, -13.13726806640625, 69.88998413085938, 27.594940185546875, -23.9560546875, 33.99321365356445, 10.8897705078125, -12.700759887695312, 0.3726043701171875, 2.380340576171875, 31.08154296875, 22.255523681640625, 38.533599853515625, 8.643852233886719, -10.594398498535156, -0.4874420166015625, 10.24884033203125, 6.17266845703125, -27.76531982421875, 40.67265319824219, -2.169647216796875, 15.635040283203125, 24.4461669921875, 9.088737487792969, 2.4539222717285156, 16.428558349609375, -17.836963653564453, -4.2890777587890625, 2.305419921875, 41.688568115234375, 10.513153076171875, 26.87646484375, 7.273340225219727, -34.198753356933594, 16.8699951171875, 32.57270812988281, 29.37622833251953, -10.77105712890625, -0.5087890625, -2.3062667846679688, 8.273651123046875, 12.958993911743164, 42.767578125, 0.19420623779296875, -0.42923736572265625, -6.457916259765625, 9.685871124267578, 0.68817138671875, -45.3677978515625, 27.369903564453125, 21.2891845703125, 11.6453857421875, 6.841667175292969, -8.647201538085938, 4.707309722900391, 13.24505615234375, -16.09808349609375, -41.75201416015625, -15.111740112304688, 21.2630615234375, -16.357574462890625, 1.352294921875, 35.97444152832031, 1.607269287109375, 6.9356231689453125, 31.68242645263672, 0.6773948669433594, 22.89898681640625, 39.1064453125, 26.701900482177734, 34.17326354980469, 5.7574005126953125, 5.378448486328125, -8.518325805664062, -1.39495849609375, 0.223297119140625, -3.2214279174804688, -6.21337890625, 34.48895263671875, 19.099987030029297, 5.7484588623046875, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000134.npy"}
{"epoch": 0.2806282722513089, "step": 135, "batch_size": 128, "mean": 13.225326538085938, "std": 20.82369041442871, "min": -51.728546142578125, "p10": -9.572761535644531, "median": 10.1134033203125, "p90": 45.19345397949218, "max": 58.9241943359375, "pos_frac": 0.765625, "sample": [0.43030548095703125, -25.6573486328125, 11.60201644897461, 23.5543212890625, -9.353286743164062, 41.856781005859375, 0.8721065521240234, 12.9246826171875, 29.917312622070312, 3.439138412475586, 52.9727783203125, 17.730865478515625, -6.189849853515625, -2.719329833984375, 11.7623291015625, 6.5277099609375, 8.323139190673828, 24.9832763671875, 25.0291748046875, 1.9930191040039062, 2.4910125732421875, 7.172054290771484, -0.4208831787109375, 2.9709243774414062, -27.619384765625, 0.9878482818603516, 34.124542236328125, 26.182273864746094, 18.860801696777344, 46.16400146484375, 34.382843017578125, 0.410858154296875, 32.472930908203125, 4.3585968017578125, -16.589752197265625, 39.6409912109375, 4.602386474609375, 16.90185546875, 28.4239501953125, 8.871322631835938, 2.851806640625, 14.686920166015625, 13.27197265625, 37.0543212890625, 57.00634765625, 17.971355438232422, 44.991607666015625, 10.837356567382812, -0.32568359375, -2.56597900390625, 19.715797424316406, 18.757583618164062, 2.4870834350585938, -10.084869384765625, -17.124008178710938, -8.933456420898438, 43.046539306640625, 5.444061279296875, 39.78509521484375, 17.085784912109375, 22.86029815673828, 16.840576171875, 33.47898864746094, -5.745025634765625, -2.9839324951171875, 10.354141235351562, -4.1878662109375, 20.922882080078125, 18.716217041015625, 3.569427490234375, 17.138916015625, 7.062343597412109, 51.06219482421875, 18.421234130859375, 33.766502380371094, 30.522315979003906, 14.695175170898438, 6.12835693359375, -6.1109161376953125, 48.88482666015625, 21.606414794921875, -17.593551635742188, -0.9871482849121094, 1.57562255859375, 8.267822265625, 49.36662292480469, 1.414703369140625, 52.03558349609375, 36.8865966796875, 49.79541015625, 20.3253173828125, 41.24566650390625, 4.81304931640625, -40.913177490234375, 45.92315673828125, 5.5281829833984375, 6.945518493652344, 7.08537483215332, 0.34429931640625, 25.189453125, 8.783470153808594, 10.26898193359375, 30.34618377685547, 58.9241943359375, 2.691028594970703, -0.71551513671875, 5.058906555175781, 14.583099365234375, 48.36138916015625, 58.063507080078125, -14.31817626953125, -15.594223022460938, 4.011993408203125, 9.95782470703125, 0.0, 21.361968994140625, -20.498931884765625, -51.728546142578125, -0.7997589111328125, 11.470634460449219, -6.544921875, 20.874771118164062, 5.065765380859375, 45.6644287109375, 25.70684814453125, -12.766281127929688, -19.3228759765625, -0.6635894775390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000135.npy"}
{"epoch": 0.28272251308900526, "step": 136, "batch_size": 128, "mean": 12.555157661437988, "std": 22.964813232421875, "min": -53.687713623046875, "p10": -9.96062545776367, "median": 10.219756126403809, "p90": 41.09892272949219, "max": 82.02072143554688, "pos_frac": 0.7578125, "sample": [43.831390380859375, 16.980247497558594, 3.0563812255859375, 16.006290435791016, 34.095863342285156, 37.012908935546875, 46.86370849609375, 11.035430908203125, 33.208892822265625, 59.86767578125, 6.26263427734375, 3.960416793823242, 41.81402587890625, 6.331520080566406, 26.055652618408203, 38.69319152832031, 14.1689453125, 38.29217529296875, 10.505157470703125, -0.4795265197753906, -43.612457275390625, 35.476646423339844, 7.0181884765625, 25.904403686523438, -20.478363037109375, 4.613853454589844, -7.856231689453125, -41.12213134765625, 7.0536651611328125, 12.942581176757812, 4.880638122558594, -0.16942977905273438, -7.1221923828125, 8.60479736328125, 24.335723876953125, -28.86865234375, -8.65234375, 28.766769409179688, 15.104240417480469, -4.3925018310546875, 26.17010498046875, -19.5816650390625, 58.08746337890625, -53.687713623046875, 11.5933837890625, -5.284904479980469, 3.473541259765625, 4.623783111572266, 17.086090087890625, -4.785118103027344, 16.77972412109375, -34.84515380859375, 43.2969970703125, 17.96466064453125, 30.15557861328125, 71.33990478515625, -9.662811279296875, 7.4352874755859375, 9.33160400390625, 31.613204956054688, 45.157958984375, 20.69000244140625, 37.238739013671875, -13.645309448242188, 11.137371063232422, 0.5085601806640625, -1.5008697509765625, 11.09075927734375, 33.99559020996094, -2.73822021484375, 4.9093780517578125, 15.7353515625, 7.526512145996094, 9.850982666015625, 4.26922607421875, 52.306915283203125, -1.3066940307617188, 13.166275024414062, 13.089485168457031, -7.75238037109375, 1.599517822265625, -25.463104248046875, 22.0426025390625, 57.91741943359375, 36.261962890625, 23.525238037109375, 1.8119888305664062, 17.606048583984375, -6.409730911254883, 45.908905029296875, 17.160472869873047, 7.86029052734375, -10.655525207519531, 13.928855895996094, 4.974517822265625, 0.1187286376953125, 58.88427734375, 13.804008483886719, 28.468215942382812, 23.137741088867188, -0.23708534240722656, 5.4566497802734375, 34.99139404296875, 0.7391815185546875, 32.1053466796875, -13.170745849609375, -7.3486785888671875, 1.7432098388671875, 34.517333984375, 40.792449951171875, 36.954559326171875, 9.934354782104492, -20.779022216796875, -6.23291015625, 4.708953857421875, 7.672119140625, 82.02072143554688, 5.266651153564453, -0.13824462890625, 9.422271728515625, 12.802688598632812, 13.716415405273438, 12.119827270507812, 2.73638916015625, 15.000232696533203, -50.786712646484375, 31.292633056640625, 4.453956604003906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000136.npy"}
{"epoch": 0.2848167539267016, "step": 137, "batch_size": 128, "mean": 11.287605285644531, "std": 21.92133903503418, "min": -47.85150146484375, "p10": -15.384351348876951, "median": 8.542205810546875, "p90": 37.950181579589845, "max": 74.68084716796875, "pos_frac": 0.765625, "sample": [-31.0472412109375, 4.024169921875, 37.9993896484375, -6.150421142578125, 11.86444091796875, 19.446304321289062, 39.42890930175781, 10.21649169921875, 4.098876953125, 8.033248901367188, 17.758522033691406, 15.684341430664062, 24.1131591796875, 55.915863037109375, 17.0072021484375, 4.05487060546875, 0.417205810546875, 31.61383056640625, 1.99822998046875, -14.1376953125, 29.68218231201172, 20.015762329101562, 7.350372314453125, 10.48809814453125, -43.46173095703125, 22.880599975585938, 5.328704833984375, -16.60333251953125, 30.168930053710938, 7.554862976074219, 49.322662353515625, 16.16510009765625, -2.518524169921875, 39.524169921875, -18.303115844726562, -1.2176513671875, 3.367523193359375, 7.400543212890625, 1.1094818115234375, -8.916671752929688, 7.8031005859375, -4.216852188110352, -19.80987548828125, 1.870269775390625, -47.85150146484375, 6.8050537109375, 6.790596008300781, 16.927764892578125, -45.788909912109375, 15.924232482910156, 7.4193115234375, 43.470489501953125, 7.31817626953125, 9.11676025390625, 16.21484375, 35.309326171875, -16.552490234375, 40.0924072265625, -16.782867431640625, 1.667388916015625, 5.6104736328125, -0.6380386352539062, 24.60003662109375, -0.0222320556640625, 11.458251953125, 57.24853515625, -27.973480224609375, 26.878551483154297, 10.435760498046875, -1.5318603515625, 1.8535003662109375, 23.834197998046875, 29.244720458984375, 12.283897399902344, 34.537654876708984, 1.9353294372558594, 1.2505931854248047, 74.68084716796875, 28.817230224609375, 14.786819458007812, 8.155609130859375, 58.135498046875, -1.3513107299804688, 35.688018798828125, 15.484222412109375, 20.5108642578125, 57.134765625, 18.630722045898438, -5.278327941894531, 35.16571044921875, 1.410797119140625, -14.883720397949219, -0.44254302978515625, 7.22454833984375, 2.6223487854003906, 64.5499267578125, 28.68353271484375, 3.0216712951660156, 18.311485290527344, 15.72015380859375, -36.978424072265625, -30.791580200195312, 8.928802490234375, -32.334442138671875, -9.856407165527344, 52.74497985839844, -14.720882415771484, 30.118194580078125, 37.92909240722656, 33.13414764404297, 25.254852294921875, 31.860260009765625, 0.262664794921875, 21.207733154296875, 15.12823486328125, 23.777400970458984, -1.9915771484375, 6.056358337402344, 0.471954345703125, 5.935699462890625, 7.558235168457031, 20.209373474121094, 5.65667724609375, 13.121902465820312, 14.170196533203125, 18.552978515625, 28.62371826171875, -4.406257629394531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000137.npy"}
{"epoch": 0.2869109947643979, "step": 138, "batch_size": 128, "mean": 11.22681999206543, "std": 24.602928161621094, "min": -45.789093017578125, "p10": -17.433428955078124, "median": 8.976593017578125, "p90": 43.210632705688475, "max": 83.78665161132812, "pos_frac": 0.671875, "sample": [69.5445556640625, 42.8939208984375, 45.04151916503906, 16.734840393066406, -11.462699890136719, 7.707794189453125, -7.175201416015625, 66.429443359375, 7.91937255859375, 21.447418212890625, 10.956695556640625, 10.87872314453125, 4.0180511474609375, 13.324356079101562, 12.387908935546875, -30.274192810058594, -18.046783447265625, 30.11529541015625, -12.622955322265625, -28.152374267578125, 21.879150390625, 4.519439697265625, -13.070510864257812, -9.181259155273438, -45.789093017578125, 26.13031005859375, -5.03656005859375, 35.282447814941406, 2.8172607421875, 83.78665161132812, -5.855377197265625, 6.317901611328125, 4.74542236328125, 43.94962692260742, 1.0329017639160156, 25.710494995117188, -26.945159912109375, 11.7821044921875, 40.79888916015625, -1.2978973388671875, 11.573822021484375, -1.907318115234375, 29.784881591796875, 36.8494873046875, 13.817474365234375, 33.2882080078125, 6.3310699462890625, 4.476837158203125, 16.401107788085938, -15.121292114257812, 12.069290161132812, 0.5827560424804688, 37.213897705078125, -16.448989868164062, 44.46002197265625, 11.7528076171875, 33.0841064453125, -1.63433837890625, 18.759723663330078, -19.31024169921875, -18.998886108398438, 1.34698486328125, -2.413818359375, -1.5258331298828125, 21.840911865234375, -29.039108276367188, 67.72283935546875, 30.730567932128906, 28.56475830078125, -2.712890625, -27.0810546875, 17.08599853515625, -6.83111572265625, 53.716888427734375, 15.816619873046875, -12.00701904296875, -3.559539794921875, 21.376083374023438, 12.98785400390625, 1.245574951171875, 16.425491333007812, 34.6025390625, 29.023223876953125, 33.205078125, -41.25543212890625, -11.0958251953125, 13.324127197265625, -0.3366241455078125, -14.036163330078125, -17.170562744140625, 27.641983032226562, 54.397003173828125, 44.93891906738281, 5.3944091796875, -2.4396820068359375, 28.37640380859375, 27.0096435546875, 26.30047607421875, 0.7887115478515625, -4.2664794921875, 29.085662841796875, -15.9466552734375, 73.80828857421875, -5.054817199707031, 8.221572875976562, 14.15948486328125, -7.409233093261719, -15.03564453125, 5.4751739501953125, 39.256797790527344, -20.738250732421875, 2.6324462890625, 2.6965255737304688, 19.92095947265625, 28.72772216796875, 1.8783721923828125, 56.850494384765625, -33.879974365234375, 32.9755859375, -37.765380859375, 4.820518493652344, 23.317153930664062, 5.3120574951171875, 21.408172607421875, 48.46856689453125, 28.930130004882812, 9.731613159179688, -13.173202514648438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000138.npy"}
{"epoch": 0.28900523560209423, "step": 139, "batch_size": 128, "mean": 8.096758842468262, "std": 24.27231216430664, "min": -48.42572021484375, "p10": -21.943605041503904, "median": 6.3019256591796875, "p90": 40.94730834960937, "max": 67.3377685546875, "pos_frac": 0.640625, "sample": [-37.502220153808594, -4.1416168212890625, -30.6522216796875, 6.53125, -32.873992919921875, 14.463165283203125, 31.231422424316406, 20.0283203125, -3.5263824462890625, -11.406097412109375, 50.775146484375, 3.1869354248046875, 20.294952392578125, 10.410614013671875, -28.3006591796875, 46.1529541015625, 52.488006591796875, 11.15386962890625, -14.77630615234375, 10.2623291015625, 5.4472808837890625, 24.59307861328125, 32.96665954589844, -14.35589599609375, 13.94354248046875, -15.145401000976562, 5.704048156738281, 35.44575500488281, -6.212989807128906, -48.42572021484375, 0.6944236755371094, -7.296012878417969, 36.09416198730469, 38.507904052734375, 25.789520263671875, 0.13667869567871094, 12.217338562011719, -9.480621337890625, 3.28515625, 29.200958251953125, -2.7271881103515625, -19.742279052734375, 1.577911376953125, 20.802291870117188, 16.819435119628906, -6.31396484375, -41.821624755859375, 0.7015380859375, -12.213920593261719, 39.7530517578125, -8.78814697265625, 0.590179443359375, -22.954330444335938, -21.51043701171875, 7.3123321533203125, 40.6387939453125, -1.121002197265625, -0.4940948486328125, -25.26715087890625, -19.792205810546875, 41.66717529296875, -13.959104537963867, 42.65290832519531, 23.822355270385742, 26.0626220703125, 15.088714599609375, 0.3907794952392578, 0.4567718505859375, 42.1383056640625, 21.86041259765625, -17.593673706054688, 19.4818115234375, 14.32513427734375, 42.79795837402344, -4.36614990234375, -12.25341796875, 4.77178955078125, 67.3377685546875, -9.1993408203125, 40.350250244140625, 24.599159240722656, 22.3779296875, 15.50726318359375, 51.40519714355469, 6.072601318359375, -15.499725341796875, 54.43206787109375, -37.477081298828125, 20.308319091796875, -4.609432220458984, 26.059814453125, 46.414306640625, 18.113525390625, 15.5845947265625, -19.182098388671875, 13.427902221679688, 0.3053436279296875, -37.950775146484375, -6.693389892578125, 2.0888538360595703, 23.839202880859375, 7.00592041015625, -8.35491943359375, 3.3164596557617188, 13.175796508789062, -11.640655517578125, 28.272483825683594, 66.5997314453125, 10.459808349609375, 59.489166259765625, 39.518646240234375, 19.216278076171875, 3.515838623046875, -34.47943115234375, 7.69384765625, -32.086822509765625, 0.0, 6.8378753662109375, 13.733413696289062, 33.03037643432617, -8.126304626464844, -33.687347412109375, 20.27642822265625, -5.0561676025390625, 21.363494873046875, -13.331298828125, 37.24884033203125, 5.078559875488281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000139.npy"}
{"epoch": 0.29109947643979056, "step": 140, "batch_size": 128, "mean": 14.924320220947266, "std": 22.66273307800293, "min": -42.10406494140625, "p10": -10.799784088134766, "median": 10.774024963378906, "p90": 47.30848693847656, "max": 80.060791015625, "pos_frac": 0.703125, "sample": [-7.41180419921875, 25.19891357421875, 32.59437561035156, -10.67877197265625, -11.082145690917969, 49.1904296875, -5.599002838134766, 38.79753112792969, -2.749176025390625, 0.1419219970703125, -16.3299560546875, 9.39398193359375, 29.9212646484375, 21.680084228515625, 5.9556427001953125, 7.925029754638672, 0.548187255859375, 20.751419067382812, 47.30029296875, 10.63140869140625, 7.957511901855469, -18.493850708007812, 29.04589080810547, -27.71209716796875, 20.648895263671875, 40.46087646484375, -2.49737548828125, 6.0082244873046875, 45.1431884765625, 41.42695617675781, 31.82745361328125, 31.882476806640625, 63.4698486328125, 38.82427215576172, -8.906906127929688, 55.74676513671875, -3.11138916015625, 4.667694091796875, 49.565185546875, 47.94075012207031, -2.292217254638672, -25.787429809570312, 11.802322387695312, -0.8597679138183594, -1.204376220703125, 36.31854248046875, 51.565673828125, 11.797027587890625, -7.73284912109375, 80.060791015625, 9.531532287597656, 22.670303344726562, 30.00152587890625, 10.916641235351562, 2.6436920166015625, 22.52984619140625, 21.523040771484375, 6.488080978393555, -8.53082275390625, 22.579483032226562, 26.779052734375, 1.894317626953125, 3.884601593017578, 22.6951904296875, 5.4187469482421875, 0.2866668701171875, -4.035015106201172, -12.250732421875, 0.0, -42.10406494140625, -1.59527587890625, -29.7259521484375, 67.15742492675781, 1.6496810913085938, 6.1341705322265625, 13.867271423339844, 23.148666381835938, 33.410400390625, 2.79644775390625, 47.327606201171875, 11.608657836914062, 3.8014602661132812, -0.10020256042480469, 41.89263916015625, 50.458221435546875, 35.57646179199219, 30.08343505859375, -3.4556541442871094, 33.88313293457031, 38.054351806640625, 6.556543350219727, 6.5941314697265625, 17.852462768554688, -3.3361358642578125, 30.06108856201172, 30.705581665039062, 0.0, -6.194578170776367, 18.04412841796875, -1.1109085083007812, -7.83935546875, 49.219940185546875, -17.258209228515625, 46.022705078125, 9.509063720703125, -18.6051025390625, 44.14361572265625, 22.283523559570312, 22.25884246826172, 22.933616638183594, 43.39569091796875, 21.635589599609375, 51.87554931640625, 21.007980346679688, 26.316497802734375, -10.453582763671875, -3.4271087646484375, -15.76416015625, 8.97027587890625, -25.434661865234375, 12.337677001953125, 55.33306884765625, 9.9527587890625, -2.011077880859375, 9.0501708984375, -17.3017578125, 27.378662109375, 22.9737548828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000140.npy"}
{"epoch": 0.2931937172774869, "step": 141, "batch_size": 128, "mean": 14.27104377746582, "std": 25.449798583984375, "min": -74.71243286132812, "p10": -10.927154541015623, "median": 12.66986083984375, "p90": 47.93629760742187, "max": 89.9302978515625, "pos_frac": 0.7421875, "sample": [40.693450927734375, 17.79107666015625, -24.982833862304688, -10.58282470703125, 7.03033447265625, 14.263442993164062, -7.815826416015625, 21.722442626953125, -47.56793212890625, 89.9302978515625, 5.0029144287109375, 10.915924072265625, 30.728363037109375, 26.975921630859375, 13.43902587890625, 44.488922119140625, 23.309768676757812, 17.530059814453125, 19.254802703857422, 21.849761962890625, 58.098114013671875, -5.4478759765625, 0.6139755249023438, 42.950286865234375, 0.14028549194335938, -17.060487747192383, 38.782466888427734, 12.61865234375, 1.4518280029296875, 50.218170166015625, 33.31170654296875, -17.696746826171875, -9.4559326171875, 69.8734130859375, 2.433349609375, 18.564239501953125, -25.85882568359375, -10.459367752075195, 10.939712524414062, -18.841644287109375, 43.2972412109375, 37.19146728515625, 14.48124885559082, 18.8984375, -6.681697845458984, 7.697509765625, 3.285003662109375, -7.264930725097656, 60.263885498046875, 7.871002197265625, 10.403457641601562, 27.294891357421875, 4.201574325561523, -7.6963348388671875, 29.2900390625, 21.202423095703125, 1.3661117553710938, 1.189697265625, 26.12908935546875, 1.50396728515625, 33.941986083984375, 33.91804504394531, -15.90924072265625, -44.40913391113281, -7.950225830078125, 70.4637451171875, -1.7034282684326172, 26.381744384765625, -4.7497100830078125, 38.11042785644531, 5.0831298828125, 14.785362243652344, 7.184226989746094, 23.43529510498047, 29.54327392578125, 36.77033233642578, 32.44158935546875, -7.107452392578125, 0.0, -17.59937858581543, 43.00518798828125, -3.19012451171875, -0.74676513671875, 59.990966796875, 6.9568939208984375, 53.919921875, -74.71243286132812, 17.52984619140625, 18.96893310546875, -2.42205810546875, 23.699600219726562, 19.02837371826172, 32.994728088378906, -44.813720703125, -6.567535400390625, 16.889175415039062, 0.040496826171875, 56.222137451171875, -1.1698455810546875, 31.672393798828125, -7.353729248046875, 11.72198486328125, 18.7901611328125, 39.82257080078125, 14.565155029296875, 3.8573741912841797, 23.85772705078125, 48.29052734375, 40.308013916015625, -11.7305908203125, 8.817296981811523, 12.7210693359375, 57.8275146484375, -37.56181335449219, 1.4738235473632812, 9.876937866210938, 11.485809326171875, 12.17529296875, 21.51751708984375, 19.709854125976562, -9.036216735839844, 48.385223388671875, 47.78448486328125, 7.302337646484375, 11.401378631591797, 57.216064453125, 17.285675048828125, 3.1789016723632812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000141.npy"}
{"epoch": 0.29528795811518327, "step": 142, "batch_size": 128, "mean": 14.104639053344727, "std": 25.519163131713867, "min": -67.17483520507812, "p10": -13.85699615478515, "median": 10.061027526855469, "p90": 49.261872863769526, "max": 79.48175048828125, "pos_frac": 0.6640625, "sample": [-3.773406982421875, -2.5950164794921875, 10.35715103149414, 26.276458740234375, 1.9141082763671875, -17.0059814453125, 10.72735595703125, -0.909881591796875, 74.26571655273438, 41.02020263671875, 46.355987548828125, -3.170989990234375, -8.392608642578125, 33.43145751953125, 20.009368896484375, 24.075286865234375, -0.0442352294921875, -3.5428009033203125, 53.860107421875, 26.891563415527344, -1.6649322509765625, 25.387130737304688, 0.0, 21.551071166992188, 19.64727783203125, 50.788116455078125, 9.79742431640625, 2.9935455322265625, -3.110370635986328, 0.441009521484375, -0.649017333984375, 29.849365234375, -2.9341087341308594, -25.740966796875, 2.8897247314453125, 2.2032127380371094, 24.31243896484375, 14.477989196777344, -20.849395751953125, 30.166748046875, 18.25537109375, 50.6068115234375, 44.1717529296875, 17.586196899414062, 16.790252685546875, -9.927490234375, -1.0685863494873047, 45.461700439453125, 18.75530242919922, 68.47784423828125, 13.740966796875, -6.461950302124023, 4.475982666015625, 13.845085144042969, -12.507431030273438, 2.7680282592773438, -0.4737396240234375, 33.993255615234375, -19.40948486328125, -19.232696533203125, 22.4400634765625, 31.4239501953125, -11.813140869140625, -9.04498291015625, 48.68547058105469, 37.368865966796875, 17.72705078125, 1.2250480651855469, 46.75743865966797, -3.9803466796875, 8.115371704101562, 21.74274444580078, 2.2088623046875, 6.1262664794921875, -23.300689697265625, 34.12181091308594, -39.017913818359375, 27.019195556640625, 44.104248046875, -26.086990356445312, -4.875, 71.07476806640625, 79.48175048828125, -3.293121337890625, -1.7663955688476562, 15.618797302246094, 10.324630737304688, 51.627685546875, 8.632514953613281, 5.518272399902344, -2.830301284790039, -3.6700668334960938, 33.11865234375, 42.23445129394531, 57.48004150390625, 5.547157287597656, 36.602325439453125, 26.287979125976562, 21.059600830078125, 54.3138427734375, -67.17483520507812, 54.131011962890625, -28.77154541015625, 1.1791629791259766, -0.8556785583496094, -27.629486083984375, 35.58055114746094, 15.695526123046875, 28.0416259765625, -3.8578414916992188, 37.371063232421875, -36.879638671875, 2.63494873046875, 6.2979736328125, 20.57579803466797, -6.503562927246094, 29.9725341796875, -1.5825843811035156, 5.82037353515625, 2.883270263671875, -20.158645629882812, 63.00628662109375, 0.42525291442871094, -1.498565673828125, 31.394378662109375, 28.98468780517578, 67.16065979003906, 41.687835693359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000142.npy"}
{"epoch": 0.2973821989528796, "step": 143, "batch_size": 128, "mean": 16.217622756958008, "std": 27.831613540649414, "min": -80.73004150390625, "p10": -11.58538703918457, "median": 13.279151916503906, "p90": 49.202030944824216, "max": 97.37802124023438, "pos_frac": 0.75, "sample": [27.574378967285156, 49.35638427734375, 56.13726806640625, 29.34930419921875, 40.575904846191406, 17.1793212890625, 31.7088623046875, 9.37646484375, -18.254898071289062, -8.01626968383789, 25.88812255859375, -0.0943756103515625, 20.701507568359375, 15.876449584960938, 2.0868263244628906, -12.766799926757812, 47.79559326171875, 4.89117431640625, -20.30975341796875, 97.37802124023438, -16.99871826171875, -10.337081909179688, 43.993255615234375, 46.5794677734375, 23.46369171142578, 45.65338134765625, 36.1312255859375, 63.24871826171875, 91.14080810546875, 1.98443603515625, -1.73004150390625, 42.026634216308594, 28.179000854492188, 31.883975982666016, 25.870391845703125, -9.417816162109375, -80.73004150390625, -1.5291290283203125, 28.72570037841797, 10.789398193359375, 9.003387451171875, -32.557106018066406, 10.460151672363281, 40.334197998046875, 7.050453186035156, 5.76153564453125, 9.401191711425781, 21.08111572265625, 17.492591857910156, -67.22991943359375, 5.64581298828125, 28.774658203125, 3.2951202392578125, 39.14604949951172, 19.5587158203125, 23.89166259765625, 0.431304931640625, 23.049842834472656, -13.255477905273438, -2.6908798217773438, -9.8031005859375, 1.1372699737548828, 69.70364379882812, 20.484649658203125, 5.484785079956055, -2.7470626831054688, 58.4910888671875, 32.212249755859375, 40.80889892578125, 3.401611328125, 32.605133056640625, 34.763092041015625, -11.07906723022461, 4.008333206176758, 32.39982604980469, -8.31231689453125, -41.105224609375, 0.5401763916015625, 9.355087280273438, 12.952613830566406, 3.92462158203125, -1.4619140625, 7.9942169189453125, 17.942367553710938, 14.46417236328125, -4.425384521484375, -18.116622924804688, 3.404834747314453, -0.24884033203125, 13.31219482421875, 35.08225631713867, 13.246109008789062, -2.7570037841796875, 22.671890258789062, 20.21405029296875, 25.811187744140625, 0.5921039581298828, 18.00225830078125, -4.712677001953125, 1.9336395263671875, 83.69171142578125, 29.38311767578125, -4.668060302734375, 12.473686218261719, -31.7044677734375, 76.533203125, 36.226280212402344, 3.388153076171875, 2.527833938598633, -1.6226348876953125, 27.075714111328125, 56.66851806640625, 28.551712036132812, 68.01589965820312, 24.0474853515625, 61.933441162109375, 42.97589111328125, 43.9183349609375, -40.67864990234375, 26.467388153076172, -15.61834716796875, 51.44073486328125, -7.5216064453125, 12.827972412109375, 49.13587951660156, 7.736602783203125, 15.01922607421875, 1.4504241943359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000143.npy"}
{"epoch": 0.2994764397905759, "step": 144, "batch_size": 128, "mean": 13.35289478302002, "std": 25.439943313598633, "min": -67.38496398925781, "p10": -15.592764282226561, "median": 7.698341369628906, "p90": 45.175811767578125, "max": 91.42803955078125, "pos_frac": 0.734375, "sample": [-1.9859199523925781, 16.37018585205078, 43.890411376953125, 5.509269714355469, 6.1797027587890625, 12.8062744140625, -18.383468627929688, -10.498771667480469, 1.2970142364501953, 91.42803955078125, 2.5466995239257812, -16.174705505371094, 7.6895599365234375, 58.5313720703125, 5.88128662109375, 34.704833984375, 2.6540145874023438, 43.86517333984375, 9.242286682128906, -1.0703125, 0.095428466796875, 21.547607421875, -15.343360900878906, 2.8925323486328125, 2.2076416015625, -5.8360595703125, -21.82415771484375, -3.5055160522460938, 28.811965942382812, 5.342550277709961, -40.4996337890625, 6.127658843994141, 0.425079345703125, 23.849578857421875, 51.27632141113281, 40.406585693359375, 31.835540771484375, 73.62353515625, 8.536430358886719, 59.81706237792969, -0.96502685546875, -7.9033660888671875, 5.098960876464844, 18.271697998046875, -0.7729701995849609, -17.648277282714844, 7.79931640625, 36.37774658203125, -17.800155639648438, 20.171142578125, 12.918312072753906, 56.0096435546875, 37.7196044921875, -21.31170654296875, -67.38496398925781, -27.20819091796875, 15.221343994140625, -14.109046936035156, -11.520370483398438, 4.29656982421875, 33.959381103515625, 0.2657470703125, -7.633087158203125, 5.38836669921875, 24.88231658935547, 12.131103515625, 66.1741943359375, 64.99481201171875, 37.931732177734375, 37.15203857421875, 0.14763641357421875, 18.46460723876953, 49.097259521484375, -26.188690185546875, 22.547775268554688, 32.929351806640625, 5.009124755859375, 10.236629486083984, 44.87437438964844, 44.07098388671875, 45.87916564941406, 37.230655670166016, 5.4102935791015625, -9.243255615234375, 14.433120727539062, 22.96971893310547, -0.83306884765625, 7.707122802734375, 2.3755836486816406, 33.10894775390625, 70.5775146484375, 33.8392333984375, 13.534957885742188, 0.3430328369140625, 11.749923706054688, 46.838134765625, 2.1011428833007812, -4.14947509765625, 7.45513916015625, 35.77008056640625, 8.171600341796875, 0.4023017883300781, 0.0, 3.2515716552734375, -13.65765380859375, -11.391143798828125, 5.365966796875, -35.549774169921875, 24.62542724609375, 23.494415283203125, 16.048126220703125, -2.862335205078125, 44.848785400390625, -7.811737060546875, 28.85882568359375, 3.643280029296875, 17.001617431640625, 4.525032043457031, 19.9490966796875, -24.549407958984375, 13.45697021484375, 39.31434631347656, 7.779380798339844, 44.26153564453125, -21.95794677734375, -0.309600830078125, 72.09176635742188, 5.1064453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000144.npy"}
{"epoch": 0.30157068062827225, "step": 145, "batch_size": 128, "mean": 17.44386100769043, "std": 26.350801467895508, "min": -54.58720397949219, "p10": -13.015023040771485, "median": 16.718063354492188, "p90": 52.5658721923828, "max": 84.5185546875, "pos_frac": 0.703125, "sample": [-12.981109619140625, 38.92308807373047, -42.75823974609375, 36.0028076171875, 13.742652893066406, -27.61309814453125, 42.817352294921875, 18.562408447265625, 82.21389770507812, 22.178619384765625, 18.167068481445312, 11.45721435546875, 24.20269775390625, 32.968505859375, 30.16234588623047, 63.29248046875, 4.254730224609375, 30.213592529296875, 6.972663879394531, 7.0995330810546875, 6.145866394042969, -7.1085968017578125, 51.19134521484375, -10.803852081298828, -8.265335083007812, 68.76272583007812, -27.4156494140625, 50.876312255859375, -0.6920852661132812, 21.7857666015625, 1.665435791015625, 31.450729370117188, 8.45355224609375, 36.93073272705078, 55.845123291015625, 24.301300048828125, 75.80435180664062, -54.58720397949219, 8.598129272460938, 2.824615478515625, 19.78876495361328, 19.9476318359375, -15.777816772460938, 65.53375244140625, 69.37557983398438, 14.525634765625, 37.576171875, 32.7974853515625, 9.929039001464844, 20.99249267578125, 16.892196655273438, -7.6581878662109375, -6.881202697753906, 34.231109619140625, 47.10736083984375, 46.08062744140625, -0.8604736328125, -4.978096008300781, -15.083290100097656, 3.5743255615234375, 16.6270751953125, 8.007171630859375, -7.587127685546875, 42.909881591796875, 63.93670654296875, -5.778282165527344, 56.559722900390625, -9.397262573242188, 56.271209716796875, 31.174673080444336, 55.773101806640625, -10.314453125, 6.2141571044921875, 33.26234436035156, 47.5693359375, 10.872406005859375, 1.3863143920898438, 9.271163940429688, 27.66271209716797, -19.734481811523438, -21.160537719726562, 5.887115478515625, 9.659828186035156, 30.299591064453125, 84.5185546875, 21.880210876464844, 5.45233154296875, 24.033355712890625, -21.904373168945312, -13.094154357910156, -3.1150894165039062, 0.0, 7.500602722167969, 20.25909423828125, 9.87839126586914, 4.141883850097656, 41.62803649902344, 70.44580078125, 42.200897216796875, 24.90423583984375, -3.0087356567382812, -8.56903076171875, 25.91131591796875, 24.085678100585938, 16.809051513671875, -12.8690185546875, 24.844200134277344, 29.441070556640625, -2.7271461486816406, -1.1365718841552734, -14.466644287109375, 49.650482177734375, 26.907432556152344, 41.2646484375, -2.547740936279297, -1.3865966796875, 3.5967788696289062, 31.14040756225586, 43.291473388671875, 48.881866455078125, 34.40350341796875, -14.912826538085938, -1.3529682159423828, -2.813446044921875, 27.510589599609375, -22.310791015625, -8.178695678710938, 20.498199462890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000145.npy"}
{"epoch": 0.3036649214659686, "step": 146, "batch_size": 128, "mean": 14.125727653503418, "std": 25.281673431396484, "min": -61.695404052734375, "p10": -15.20751953125, "median": 8.188369750976562, "p90": 44.752130126953126, "max": 77.26397705078125, "pos_frac": 0.734375, "sample": [5.3447723388671875, 2.2998580932617188, 18.546184539794922, -33.89752197265625, 12.07489013671875, 41.82676696777344, -27.918441772460938, 11.8348388671875, 33.83177947998047, -23.890777587890625, 77.26397705078125, 74.06256103515625, 10.524444580078125, -7.80303955078125, 18.11090087890625, 3.9177932739257812, 42.014801025390625, 26.252166748046875, 3.3207244873046875, 3.218505859375, 30.350082397460938, 2.5152816772460938, 7.114048004150391, 16.317184448242188, -20.3414306640625, -7.595123291015625, 3.404296875, 13.819244384765625, 16.140380859375, 21.345123291015625, 7.39495849609375, 1.2041740417480469, 40.9656982421875, 1.420501708984375, 6.430145263671875, 19.213516235351562, 17.140777587890625, -22.424484252929688, 15.111114501953125, 3.869640350341797, 7.0038909912109375, 48.17195129394531, 0.8499412536621094, 35.95263671875, 0.0, 44.78814697265625, -2.4122314453125, -1.686532974243164, 26.346115112304688, 27.53021240234375, 11.462860107421875, -15.17144775390625, 9.572904586791992, -1.3520050048828125, 52.418792724609375, 62.143890380859375, -61.695404052734375, 29.317001342773438, 4.113883972167969, 35.11128234863281, 35.627288818359375, 5.38092041015625, 5.755714416503906, 48.11431884765625, 35.499237060546875, -14.712631225585938, -6.7960205078125, -10.824554443359375, 6.505439758300781, -19.838760375976562, 52.86683654785156, -23.781204223632812, 4.007843017578125, 41.3690185546875, -19.191329956054688, 62.583465576171875, 10.57269287109375, 8.7125244140625, -11.04791259765625, 69.81301879882812, 7.5224609375, 33.519287109375, 2.810598373413086, -9.57763671875, 35.7054443359375, 1.8674049377441406, -6.849761962890625, 4.1226806640625, 0.0, 38.478538513183594, -2.3953704833984375, -5.400306701660156, 26.813674926757812, -21.22979736328125, 2.56878662109375, 58.74261474609375, 21.430145263671875, 40.287994384765625, 7.044258117675781, 51.74293518066406, -21.385269165039062, -15.29168701171875, 7.664215087890625, 11.5966796875, 42.506805419921875, -9.4786376953125, 33.48590087890625, 6.3452606201171875, 51.7783203125, 44.238494873046875, 44.7366943359375, -59.653106689453125, 30.904598236083984, 44.686912536621094, 13.5582275390625, 37.94024658203125, -5.996696472167969, 43.436073303222656, 3.8346214294433594, 42.313018798828125, -1.34588623046875, 7.3670654296875, 23.4105224609375, 44.41387939453125, 30.973663330078125, -9.618873596191406, 30.189361572265625, -9.13336181640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000146.npy"}
{"epoch": 0.3057591623036649, "step": 147, "batch_size": 128, "mean": 11.819829940795898, "std": 28.745960235595703, "min": -77.148193359375, "p10": -18.955963134765625, "median": 9.859481811523438, "p90": 50.29435577392577, "max": 88.66107177734375, "pos_frac": 0.65625, "sample": [-0.0235137939453125, -1.942169189453125, 7.962333679199219, 29.9224853515625, -8.915397644042969, -3.09759521484375, 82.70867919921875, -13.938735961914062, 0.0, 70.79644775390625, -17.414398193359375, 18.303237915039062, 14.291046142578125, 5.3245849609375, 24.79216194152832, 4.862823486328125, -18.67803955078125, 47.6170654296875, 53.75537109375, 3.1801071166992188, -5.755035400390625, 20.0172119140625, 0.0221099853515625, 34.93547821044922, -1.8276023864746094, 51.966522216796875, 1.283111572265625, 44.7626953125, 10.247283935546875, -0.255615234375, 70.56277465820312, 35.902069091796875, 17.70892333984375, 22.886260986328125, 10.40008544921875, -28.50201416015625, -44.716156005859375, -6.18927001953125, -17.35089111328125, 18.030807495117188, 10.5108642578125, 34.234336853027344, 32.28984451293945, 6.1907501220703125, -31.37398910522461, 28.9012451171875, -29.469558715820312, -4.30120849609375, -0.5726776123046875, 1.4694404602050781, -3.180267333984375, 51.969573974609375, 10.895721435546875, 12.804061889648438, 27.080780029296875, 28.753143310546875, 77.90423583984375, 11.961868286132812, 5.222282409667969, -5.0720977783203125, 21.376708984375, 22.7078857421875, 60.38722229003906, 11.462818145751953, 55.55853271484375, 0.432861328125, 1.251708984375, 60.3770751953125, 40.788055419921875, 3.18853759765625, -43.496185302734375, 12.648162841796875, 3.090179443359375, 1.16363525390625, 30.86279296875, 5.0588836669921875, 14.187932968139648, 2.9325637817382812, -0.09418487548828125, -0.2395172119140625, 28.84375, -11.23590087890625, -0.344451904296875, 36.356849670410156, 18.54759979248047, 38.502105712890625, 10.00299072265625, 33.04986572265625, -12.1322021484375, 32.610816955566406, 14.266082763671875, 43.453369140625, -24.78955078125, -77.148193359375, -38.586524963378906, -1.6804046630859375, 10.618789672851562, 31.264480590820312, -18.786773681640625, -41.88508605957031, -9.495742797851562, 3.7231597900390625, 55.5513916015625, 13.53521728515625, -19.350738525390625, -7.75932502746582, 45.995025634765625, -4.026123046875, -40.518035888671875, -14.545501708984375, -66.0276107788086, 9.715972900390625, 7.7423553466796875, 39.712615966796875, 49.57771301269531, 35.565582275390625, 14.14453125, 54.818450927734375, -5.43914794921875, -7.365997314453125, 16.09839630126953, 43.04042053222656, 44.9342041015625, 16.65728759765625, 88.66107177734375, 9.125694274902344, -4.882568359375, -30.676788330078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000147.npy"}
{"epoch": 0.3078534031413613, "step": 148, "batch_size": 128, "mean": 12.455220222473145, "std": 27.12109375, "min": -59.69450378417969, "p10": -24.25674057006836, "median": 10.939041137695312, "p90": 48.902444458007814, "max": 92.16928100585938, "pos_frac": 0.6484375, "sample": [14.964599609375, 13.557022094726562, 17.276748657226562, -14.847808837890625, -1.74725341796875, 70.15164184570312, -0.04076194763183594, 20.31195068359375, 72.17047119140625, 26.576339721679688, 5.339441299438477, 20.663192749023438, -4.2837982177734375, 56.02458190917969, 14.347381591796875, 37.76734924316406, 10.8878173828125, -20.474456787109375, 16.67095947265625, 25.85274887084961, -22.34912109375, 43.633544921875, -47.530906677246094, 0.4184532165527344, 8.16961669921875, -25.646484375, 13.444793701171875, -5.5946044921875, 15.235748291015625, -8.976594924926758, 20.12946319580078, 16.346054077148438, -9.24017333984375, -59.69450378417969, -9.254669189453125, -16.7435302734375, 61.665069580078125, 6.950286865234375, -27.593978881835938, 20.67333984375, 58.59698486328125, 1.6243972778320312, -18.428680419921875, -31.507904052734375, 1.307281494140625, 6.773052215576172, 1.6442642211914062, -16.148895263671875, 21.36114501953125, 33.600311279296875, 13.794090270996094, 8.557655334472656, 6.816967010498047, 20.15374755859375, -32.38987731933594, 71.22174072265625, 9.19158935546875, 37.63932800292969, 10.599235534667969, 48.99560546875, 25.758544921875, -7.9311981201171875, -0.36841583251953125, -0.7068023681640625, 0.0, 7.536441802978516, 54.975677490234375, 24.17603302001953, 37.82963562011719, 39.08306884765625, -3.682220458984375, -29.868484497070312, -2.94183349609375, -35.290924072265625, 38.83274841308594, 9.838104248046875, -29.41736602783203, 31.443374633789062, -2.6250152587890625, 29.912673950195312, -0.5095195770263672, 20.517730712890625, -27.14154052734375, 29.455535888671875, 56.91485595703125, 19.529712677001953, 2.393634796142578, 17.261474609375, -9.933578491210938, -15.535888671875, 27.796653747558594, 19.945125579833984, -0.7559757232666016, 29.227859497070312, 25.701461791992188, 48.12437438964844, -27.219635009765625, 5.375556945800781, -11.952499389648438, 62.3370361328125, 3.82550048828125, 27.368194580078125, 92.16928100585938, 38.42681884765625, 1.8570632934570312, -8.660764694213867, 31.142135620117188, -28.46307373046875, 30.816253662109375, 21.283653259277344, 48.862518310546875, 45.641761779785156, 16.619537353515625, 28.99981689453125, -25.24274444580078, -1.6026687622070312, -23.83416748046875, 17.629531860351562, 60.8392333984375, 0.0, -7.9801025390625, 13.001861572265625, 34.698280334472656, -1.0684165954589844, 10.990264892578125, 44.393585205078125, 60.26292419433594, -4.40655517578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000148.npy"}
{"epoch": 0.3099476439790576, "step": 149, "batch_size": 128, "mean": 15.689586639404297, "std": 26.402050018310547, "min": -42.338897705078125, "p10": -19.08871154785156, "median": 12.928115844726562, "p90": 48.032083129882814, "max": 98.1683349609375, "pos_frac": 0.765625, "sample": [48.207061767578125, -21.70501708984375, 25.690093994140625, 34.83113098144531, 44.73541259765625, 47.52789306640625, 23.419570922851562, 5.488094329833984, 7.1004638671875, 7.7138824462890625, 20.1405029296875, 2.738555908203125, 44.459259033203125, 16.584762573242188, 28.90692138671875, -3.05487060546875, 45.5516357421875, 10.640632629394531, -2.51776123046875, -8.646438598632812, 16.06267547607422, 4.2558746337890625, 19.842041015625, 4.099250793457031, 28.48000717163086, 14.249053955078125, -41.39250183105469, 36.033477783203125, 49.1470947265625, 16.510025024414062, 6.065948486328125, 18.39306640625, 34.2554931640625, 12.588623046875, 1.8488569259643555, 0.0349273681640625, 49.90638732910156, 2.725828170776367, -17.967437744140625, 0.48709869384765625, 34.80230712890625, 71.76177978515625, -10.459640502929688, 3.2660446166992188, 75.58915710449219, 47.95709228515625, 10.40057373046875, -24.786087036132812, 38.937957763671875, 51.505828857421875, 22.147201538085938, -4.0209808349609375, 9.4046630859375, 24.190155029296875, 5.3066558837890625, 78.27587890625, -26.968505859375, 16.23681640625, 19.630645751953125, 13.4381103515625, -7.0075836181640625, 23.677589416503906, -24.75445556640625, 40.579376220703125, -27.520660400390625, 2.22808837890625, 8.968856811523438, -6.089385986328125, 23.920669555664062, 23.112506866455078, 40.705413818359375, 16.874801635742188, 33.8718147277832, 13.267608642578125, 39.848846435546875, 14.45953369140625, 51.7294921875, 8.925384521484375, 57.553497314453125, 0.26775360107421875, 42.022918701171875, 0.041988372802734375, 32.596778869628906, 26.600341796875, 11.5238037109375, -12.2613525390625, 56.2928466796875, 42.67236328125, -40.96630859375, 3.8176345825195312, 11.768218994140625, 34.07844543457031, -4.66082763671875, -28.54443359375, -2.4166412353515625, -5.2275390625, 66.09515380859375, 5.193328857421875, 43.0550537109375, 5.9091033935546875, -4.685272216796875, 3.0182647705078125, 33.16477966308594, 0.2588043212890625, 28.792022705078125, -0.7554931640625, -28.165496826171875, -24.911666870117188, 2.3761749267578125, 98.1683349609375, -2.412628173828125, -2.73736572265625, 7.3502197265625, 26.190109252929688, 5.1995849609375, -12.978744506835938, 41.71863555908203, 28.421096801757812, -41.329368591308594, -42.338897705078125, 20.4814453125, 37.12841796875, -24.161834716796875, 74.77517700195312, 40.588294982910156, 22.111846923828125, 3.960296630859375, 6.7771148681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000149.npy"}
{"epoch": 0.31204188481675393, "step": 150, "batch_size": 128, "mean": 14.854695320129395, "std": 25.497289657592773, "min": -51.8594970703125, "p10": -12.206530761718748, "median": 13.416543960571289, "p90": 46.21770248413086, "max": 83.44277954101562, "pos_frac": 0.7734375, "sample": [22.767425537109375, 33.07292938232422, -4.332691192626953, 34.171844482421875, 10.508537292480469, -29.388031005859375, -2.6777496337890625, 19.389572143554688, -32.1005859375, 30.6614990234375, 26.070762634277344, 22.446258544921875, 47.132080078125, 14.75799560546875, -14.326568603515625, 3.36700439453125, 7.7743988037109375, 29.631996154785156, 5.946533203125, 39.822174072265625, 26.010421752929688, -14.494140625, -1.665252685546875, 3.019439697265625, -5.806547164916992, 14.879119873046875, 9.503814697265625, 6.786346435546875, 66.27365112304688, 33.29297637939453, 69.85189819335938, 0.3212547302246094, 20.3890380859375, 8.944055557250977, 17.058059692382812, 27.010498046875, 21.73138427734375, -38.455322265625, 39.41431427001953, 10.078720092773438, 25.042388916015625, 43.41667175292969, 17.84796142578125, 39.708526611328125, -9.469924926757812, 4.0701751708984375, 2.0201263427734375, 5.13250732421875, 13.605072021484375, 62.2265625, 18.32316017150879, 77.1046142578125, 0.0849609375, 32.279693603515625, 45.539398193359375, 61.861236572265625, 53.081451416015625, 5.846410751342773, -0.241424560546875, 0.7995681762695312, -0.1552581787109375, -1.386810302734375, -31.946990966796875, 46.455299377441406, 5.020263671875, 5.08294677734375, 6.1367034912109375, 0.357666015625, 43.81743621826172, 46.115875244140625, 21.471282958984375, 39.29693603515625, 2.7483367919921875, 46.08650207519531, 21.711881637573242, -10.2607421875, -43.13873291015625, 6.722137451171875, 2.290925979614258, 24.714385986328125, 13.60501480102539, -51.8594970703125, -1.1627349853515625, -21.297683715820312, -5.151939392089844, 17.82672119140625, -22.78143310546875, 83.44277954101562, 35.60400390625, 13.488433837890625, 15.64569091796875, -11.297943115234375, 6.4396209716796875, 24.788925170898438, -48.328857421875, 9.034141540527344, 41.02735900878906, -0.78564453125, -26.518463134765625, 1.5224323272705078, 18.224578857421875, 32.16771697998047, 8.67041015625, 53.88481140136719, 10.778280258178711, 18.44000244140625, 35.0853271484375, 10.365814208984375, 43.998443603515625, 31.525390625, 0.258819580078125, 15.924957275390625, 5.252111434936523, 51.99737548828125, 7.83087158203125, 0.42435455322265625, 15.570335388183594, -11.10479736328125, 55.5828857421875, -8.107086181640625, 0.0, 26.644256591796875, 79.68115234375, 0.557861328125, 13.344654083251953, 16.697891235351562, 17.832496643066406, -31.62713623046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000150.npy"}
{"epoch": 0.31413612565445026, "step": 151, "batch_size": 128, "mean": 22.405221939086914, "std": 29.35654067993164, "min": -56.99365234375, "p10": -16.504222106933593, "median": 24.564971923828125, "p90": 57.23146057128905, "max": 84.42578125, "pos_frac": 0.75, "sample": [40.634979248046875, 19.657806396484375, 1.3672103881835938, -24.729248046875, 0.41962432861328125, 84.42578125, 53.592681884765625, -27.9193115234375, 3.6371612548828125, 41.627899169921875, -1.415802001953125, -8.365081787109375, 26.424346923828125, 37.65338134765625, -7.7366943359375, -15.009056091308594, -3.439178466796875, 39.91389465332031, 37.7435302734375, 23.503639221191406, -56.99365234375, 2.997528076171875, -12.101959228515625, 24.152557373046875, 3.826904296875, 41.4041748046875, 5.0814361572265625, -4.060493469238281, 26.332427978515625, 44.942535400390625, 51.109130859375, 5.4040985107421875, -4.0774383544921875, 21.419082641601562, 39.796348571777344, 12.91009521484375, 19.800750732421875, 17.76915740966797, -4.869651794433594, 78.66403198242188, 27.5523681640625, 32.70469665527344, 13.398994445800781, 18.77899169921875, 45.04095458984375, -22.195281982421875, 24.465728759765625, 8.790374755859375, 46.80731201171875, 44.90216064453125, 44.27854919433594, -10.683502197265625, 4.490177154541016, -0.41376495361328125, 38.7249755859375, 45.81878662109375, 17.01513671875, 28.140625, -29.04962158203125, 21.513633728027344, -3.91815185546875, 83.393310546875, 50.83624267578125, -11.593841552734375, 32.144317626953125, 43.59397888183594, 29.64727783203125, -24.665985107421875, 42.228179931640625, 61.00885009765625, -36.77215576171875, -29.74114990234375, 45.2694091796875, 33.62834930419922, 32.49768829345703, 24.067779541015625, 55.612579345703125, 39.68584442138672, 77.69268798828125, -21.19329833984375, -26.7723388671875, 46.119842529296875, 71.4254150390625, 48.199737548828125, 80.41265869140625, 7.6151123046875, 37.89605712890625, -0.6531486511230469, 0.0, 12.606048583984375, 29.301116943359375, 41.83746337890625, 33.73603820800781, -11.762405395507812, 35.61662292480469, 32.65745544433594, 52.2198486328125, 70.46600341796875, -14.577308654785156, 73.650634765625, -33.404327392578125, 37.81219482421875, 2.659149169921875, 36.254615783691406, 62.354248046875, 48.03802490234375, 14.281013488769531, -17.346099853515625, 31.685829162597656, 33.366973876953125, 18.262115478515625, 67.296630859375, 14.89361572265625, -16.143417358398438, 76.56686401367188, 19.74090576171875, 52.803497314453125, 36.57249450683594, 67.6917724609375, 24.664215087890625, 39.9857177734375, 10.15313720703125, 13.669921875, 33.26629638671875, 39.43741989135742, -0.1861572265625, 9.57342529296875, -35.074440002441406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000151.npy"}
{"epoch": 0.3162303664921466, "step": 152, "batch_size": 128, "mean": 17.0307674407959, "std": 29.262435913085938, "min": -57.62200927734375, "p10": -13.116577148437498, "median": 16.177261352539062, "p90": 57.11192321777344, "max": 100.54534912109375, "pos_frac": 0.703125, "sample": [34.71076965332031, 47.8173828125, -4.4789581298828125, 16.278030395507812, 6.180145263671875, 36.382080078125, 12.47021484375, 34.517486572265625, 57.19512939453125, 30.671981811523438, 19.299423217773438, 39.24871826171875, 69.51051330566406, 6.866607666015625, 35.8895263671875, -12.409652709960938, 2.07757568359375, 3.6385650634765625, 2.35516357421875, -22.660400390625, -57.62200927734375, 29.02410888671875, -57.3656005859375, 21.155517578125, -10.202415466308594, 40.35015869140625, 18.205039978027344, -0.5882377624511719, 52.901214599609375, -1.916595458984375, -1.0120258331298828, 46.89625549316406, 57.243194580078125, 30.26336669921875, 28.96258544921875, 36.37188720703125, 51.1956787109375, 38.87031555175781, 58.65374755859375, 30.63202667236328, -41.24517822265625, 100.54534912109375, -29.78497314453125, 6.71160888671875, 34.28401184082031, -2.4211597442626953, -12.26055908203125, 23.3514404296875, -9.062986373901367, 0.35784912109375, 4.357269287109375, -6.990509033203125, 16.683349609375, -7.489166259765625, -7.539897918701172, 28.20477294921875, 12.906646728515625, -25.74517822265625, -9.120101928710938, -1.06378173828125, 61.99383544921875, 16.076492309570312, -18.0906982421875, 19.610809326171875, 69.98977661132812, -22.8907470703125, 16.35321044921875, -14.751251220703125, -8.673904418945312, 43.4910888671875, 7.302066802978516, -10.389190673828125, -5.590976715087891, 9.44427490234375, 1.8050079345703125, 65.69580078125, 41.03456115722656, 20.28143310546875, 75.74995422363281, 33.26898193359375, 57.076263427734375, -26.464096069335938, 1.9448184967041016, -7.611663818359375, 0.814666748046875, 3.5074615478515625, 42.45550537109375, 39.469024658203125, 71.36236572265625, 12.832647323608398, 8.007438659667969, -55.9696044921875, -10.405715942382812, 63.93304443359375, 34.999420166015625, 30.65813446044922, -13.947998046875, 5.06414794921875, 23.6973876953125, -12.174224853515625, 20.565032958984375, 46.4034423828125, 27.48272705078125, 22.963134765625, -12.76025390625, 62.0718994140625, 4.304431915283203, 10.281379699707031, -7.8101806640625, -5.619621276855469, 42.48042678833008, 9.189506530761719, -43.804443359375, 28.940338134765625, -1.6723175048828125, 12.996383666992188, 34.605979919433594, 1.6332855224609375, 44.717803955078125, 60.9581298828125, 45.638763427734375, 54.3988037109375, 47.94267272949219, 37.49150848388672, 5.731048583984375, 24.21514892578125, -10.216842651367188, 43.59319305419922], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000152.npy"}
{"epoch": 0.3183246073298429, "step": 153, "batch_size": 128, "mean": 24.618873596191406, "std": 27.884153366088867, "min": -60.364410400390625, "p10": -3.9107910156249996, "median": 20.17603302001953, "p90": 63.47185668945312, "max": 92.14340209960938, "pos_frac": 0.8515625, "sample": [15.806671142578125, 0.2001190185546875, 76.16973876953125, 17.998779296875, 62.581634521484375, 47.63585662841797, 48.207000732421875, 0.4139404296875, 92.14340209960938, 6.6488037109375, 46.279632568359375, 4.551822662353516, -17.703155517578125, 24.841812133789062, 35.79124450683594, 5.133247375488281, -7.823883056640625, 2.7757415771484375, 31.596107482910156, 69.89125061035156, 64.79132080078125, 73.58958435058594, 36.01028823852539, 3.5374908447265625, 5.29742431640625, -4.3665924072265625, 37.183860778808594, -5.174985885620117, 1.4079055786132812, 2.507049560546875, -14.502471923828125, 4.521171569824219, 49.94232177734375, 4.8886566162109375, 1.841949462890625, 14.756454467773438, 54.534912109375, 45.30384826660156, 29.524648666381836, 0.0, 62.8746337890625, 41.44132995605469, 57.60575866699219, 20.968643188476562, 47.266845703125, 44.555755615234375, 46.100067138671875, -1.9468841552734375, 18.2393798828125, 17.51251983642578, 17.42791748046875, 54.12158203125, -60.364410400390625, 3.8879165649414062, -3.74822998046875, 36.23248291015625, 52.76275634765625, 38.6905517578125, -0.376068115234375, 72.9749755859375, 17.384857177734375, 8.09283447265625, 21.686721801757812, 60.83526611328125, -4.29010009765625, 52.41845703125, 7.7635498046875, 10.88232421875, 19.234405517578125, 17.61376953125, -6.935699462890625, 20.199249267578125, 3.9834136962890625, 41.12652587890625, 0.5306396484375, -0.36492156982421875, 8.852386474609375, 32.900726318359375, 15.699859619140625, 40.97503662109375, 38.30253601074219, -14.03192138671875, 51.991943359375, 73.112060546875, 25.8173828125, 22.334686279296875, 6.1583709716796875, 14.689666748046875, 41.09442138671875, -0.2935943603515625, 3.1743927001953125, 21.846649169921875, 67.02536010742188, 48.74249267578125, 55.97602844238281, 11.221744537353516, 6.530961990356445, 20.152816772460938, 10.30804443359375, 7.294586181640625, 46.368499755859375, 45.809967041015625, 40.64833068847656, 7.260200500488281, 3.2655487060546875, 4.139625549316406, 29.875457763671875, 80.2884521484375, 29.416366577148438, 73.49139404296875, 65.66368103027344, 25.246978759765625, 6.360015869140625, 65.09835815429688, -24.28591537475586, 28.101776123046875, 10.748844146728516, 62.9063720703125, 31.061843872070312, 6.19140625, -20.22698974609375, 69.54951477050781, 59.19268798828125, 13.222343444824219, -48.92095184326172, -40.18254089355469, 33.46076965332031, 32.463836669921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000153.npy"}
{"epoch": 0.3204188481675393, "step": 154, "batch_size": 128, "mean": 17.83034896850586, "std": 32.73481369018555, "min": -78.86904907226562, "p10": -23.318249511718747, "median": 18.886703491210938, "p90": 62.23740539550781, "max": 88.33489990234375, "pos_frac": 0.7109375, "sample": [26.093673706054688, 51.05015563964844, -28.47021484375, 64.49298095703125, 32.37800598144531, 7.355827331542969, -9.342369079589844, 60.14179992675781, 7.511680603027344, 15.885406494140625, -52.17250061035156, 38.853721618652344, 22.93585968017578, 14.4158935546875, 35.761444091796875, -78.86904907226562, 37.58757019042969, -26.347137451171875, 35.29920959472656, 1.60589599609375, 12.241424560546875, -0.751434326171875, 8.428009033203125, -28.090866088867188, -5.47113037109375, 7.47833251953125, 34.81378173828125, 8.75164794921875, 3.267822265625, 74.85562133789062, 12.99774169921875, -6.822282791137695, 59.61358642578125, 70.00784301757812, 3.8515777587890625, 50.50357437133789, 29.10894775390625, 70.76751708984375, 42.4837646484375, 68.90606689453125, 3.561187744140625, 55.96673583984375, 69.64437866210938, 30.4781494140625, -54.38410949707031, 19.330490112304688, 80.50619506835938, -1.9505691528320312, -9.63299560546875, -27.022369384765625, -5.427680969238281, -10.657707214355469, 20.35186767578125, -22.77984619140625, 27.77056884765625, 43.44697570800781, 37.2652587890625, -5.874603271484375, 3.4071502685546875, -21.716461181640625, -2.5980148315429688, 10.40887451171875, 48.0682373046875, 25.50049591064453, 26.214691162109375, -53.83380126953125, -18.88727378845215, -0.0432281494140625, -4.0919952392578125, -8.845794677734375, -11.604015350341797, -46.798370361328125, 54.32182312011719, 68.95361328125, -4.95489501953125, -10.035987854003906, 11.174480438232422, -5.131988525390625, 31.724700927734375, 8.816024780273438, -36.526458740234375, 12.01885986328125, -0.6926651000976562, 22.241058349609375, 38.564697265625, 10.351226806640625, 61.6290283203125, 34.83148193359375, -6.8248291015625, 31.70354461669922, 23.45928955078125, 21.40643310546875, 63.656951904296875, 49.14995574951172, 37.73809814453125, -20.5987548828125, 22.778091430664062, 3.57965087890625, 88.33489990234375, 15.755813598632812, 20.825286865234375, 50.91827392578125, -24.57452392578125, -64.90919494628906, 42.17926025390625, 30.787628173828125, 35.55229187011719, 31.014251708984375, 18.20867919921875, 27.918746948242188, 68.9891357421875, 32.876983642578125, 34.76844787597656, 58.24259948730469, 3.51708984375, 6.176897048950195, 83.4046630859375, 18.442916870117188, 17.83868408203125, 23.483306884765625, -43.129150390625, -12.550739288330078, 60.906890869140625, 34.53523254394531, 3.5836143493652344, 23.07470703125, 38.85429382324219, 65.04242706298828], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000154.npy"}
{"epoch": 0.3225130890052356, "step": 155, "batch_size": 128, "mean": 13.301179885864258, "std": 29.772275924682617, "min": -61.139404296875, "p10": -22.066686248779295, "median": 12.278924942016602, "p90": 50.123944091796865, "max": 87.75537109375, "pos_frac": 0.6796875, "sample": [69.63079833984375, 25.675811767578125, 16.85791015625, -6.6189727783203125, 37.194332122802734, -0.5112895965576172, 11.174057006835938, -2.397003173828125, 7.30242919921875, 49.27796936035156, 44.419307708740234, 3.986013412475586, 32.5948486328125, -22.75604248046875, 24.340789794921875, 31.06427001953125, -16.50238037109375, 75.0595703125, 87.75537109375, -6.611572265625, 14.382049560546875, 28.5439453125, 48.670806884765625, 86.40069580078125, 4.8904876708984375, 41.84368896484375, -43.90525817871094, 0.0, -31.518646240234375, -7.9591064453125, -21.592140197753906, 57.7403564453125, 24.171676635742188, 48.49463653564453, 3.417999267578125, -15.462738037109375, -21.77124786376953, 42.36753845214844, -8.20208740234375, 1.077606201171875, 7.897216796875, 29.50335693359375, 17.2047119140625, 1.69537353515625, 38.363037109375, 54.40380859375, 34.832000732421875, 15.223968505859375, 4.4441375732421875, 36.720123291015625, -8.205894470214844, -53.858642578125, 8.737947463989258, 6.638763427734375, 3.3788280487060547, -36.353668212890625, -0.88287353515625, 39.18475341796875, 75.34088134765625, 52.09788513183594, 3.5377731323242188, 3.377532958984375, 19.077728271484375, -8.381004333496094, -43.638084411621094, -8.564760208129883, 45.62176513671875, 59.625640869140625, 14.42087173461914, 16.779678344726562, -3.3104095458984375, -30.810386657714844, 37.072021484375, 46.32666015625, -13.464248657226562, 37.803955078125, 8.161590576171875, -6.958747863769531, 42.606842041015625, -28.31829833984375, 34.643035888671875, 54.82902526855469, -28.81329345703125, -13.710357666015625, 0.39654541015625, 8.22698974609375, 41.204132080078125, -43.699615478515625, 2.073944091796875, 45.146209716796875, 16.916046142578125, 27.939956665039062, 23.22039794921875, -0.8033447265625, -44.57489013671875, 12.999786376953125, 12.834365844726562, 64.69863891601562, 13.954437255859375, -15.235206604003906, -16.272140502929688, -16.06531524658203, -3.5642242431640625, 11.72348403930664, 19.464080810546875, 41.45269775390625, 4.870719909667969, -5.369354248046875, -61.139404296875, 1.7095565795898438, -44.87646484375, 14.285675048828125, 38.28266143798828, -11.94854736328125, 56.316009521484375, 33.48321533203125, 38.655860900878906, 26.62946319580078, 14.71923828125, 61.88958740234375, -19.850189208984375, 0.696380615234375, -1.0385513305664062, 16.469940185546875, 18.656585693359375, 26.338409423828125, 14.879623413085938, 0.046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000155.npy"}
{"epoch": 0.32460732984293195, "step": 156, "batch_size": 128, "mean": 22.198448181152344, "std": 33.907596588134766, "min": -79.9735107421875, "p10": -22.302615356445312, "median": 24.905197143554688, "p90": 61.714779663085935, "max": 98.244140625, "pos_frac": 0.765625, "sample": [38.955360412597656, 41.57356262207031, 30.592185974121094, 19.898193359375, -2.411346435546875, 5.479133605957031, 48.2596435546875, 47.71638488769531, 34.200523376464844, 30.363248825073242, 44.54493713378906, 28.04266357421875, 53.919677734375, 33.278472900390625, 3.7273712158203125, -22.240753173828125, -6.669898986816406, 34.064361572265625, 7.8540191650390625, 43.12769317626953, 2.74334716796875, 44.550537109375, -11.55523681640625, -29.783668518066406, 28.634521484375, 10.188446044921875, -3.3987350463867188, 23.452789306640625, 0.8277664184570312, 44.69781494140625, 70.50860595703125, -5.21575927734375, 73.345947265625, -8.577468872070312, 13.522966384887695, -25.86865234375, -37.520233154296875, 6.939453125, 49.718475341796875, 3.3084259033203125, 24.41070556640625, 61.694976806640625, 23.9947509765625, 11.758331298828125, 3.435150146484375, 3.6952667236328125, 46.631011962890625, -38.683837890625, 47.52122116088867, 33.543792724609375, 3.5834426879882812, 37.343711853027344, 23.0394287109375, 25.399688720703125, 83.694580078125, -27.0367431640625, -39.73057556152344, 89.47747802734375, -0.993560791015625, 10.49053955078125, -8.24267578125, 87.3330078125, 36.43605041503906, 53.13044738769531, 33.18403625488281, 59.13104248046875, 51.98956298828125, 15.502677917480469, 61.760986328125, 5.8549652099609375, 25.4368896484375, 0.13437652587890625, -6.953453063964844, 67.58206176757812, 41.272003173828125, 33.646697998046875, 98.244140625, 16.400054931640625, 0.58837890625, -3.7381744384765625, 84.8294677734375, 31.979110717773438, 74.25967407226562, 1.33978271484375, 10.800590515136719, 30.273963928222656, 2.608182907104492, 38.10064697265625, 32.310302734375, -29.950439453125, 46.7442626953125, 52.320159912109375, 37.8076171875, 48.0927734375, 26.699462890625, 13.20172119140625, -20.837738037109375, 60.927330017089844, 27.226669311523438, -22.44696044921875, 61.68939208984375, 13.31890869140625, 17.439865112304688, -26.37933349609375, -79.9735107421875, 55.50457763671875, -3.582447052001953, 87.98516845703125, -2.9676361083984375, -40.795166015625, 6.07208251953125, 31.573089599609375, 89.27130126953125, 12.89190673828125, -78.38119506835938, -38.531795501708984, 1.08489990234375, 46.071449279785156, 54.93121337890625, 58.99847412109375, -0.9153404235839844, 27.081390380859375, 44.01371765136719, 63.5142822265625, -20.2806396484375, 60.4703369140625, -7.830329895019531, 6.0829620361328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000156.npy"}
{"epoch": 0.3267015706806283, "step": 157, "batch_size": 128, "mean": 15.276664733886719, "std": 30.77654266357422, "min": -66.36323547363281, "p10": -23.134693908691403, "median": 11.860990524291992, "p90": 55.76744155883789, "max": 98.8560791015625, "pos_frac": 0.6953125, "sample": [29.984222412109375, -0.6710968017578125, 25.334228515625, 65.49244689941406, -3.84326171875, -24.302047729492188, 13.2117919921875, 18.699615478515625, 56.241668701171875, -7.820556640625, 23.6864013671875, -0.640533447265625, -47.27581787109375, 25.671356201171875, 1.0210037231445312, 36.35369873046875, 11.654842376708984, -0.8654861450195312, 3.402240753173828, -13.625198364257812, -17.80450439453125, 67.85882568359375, 29.525497436523438, -10.926956176757812, 49.43648147583008, 1.5438365936279297, -33.70947265625, -1.4536590576171875, 5.0170745849609375, 17.88629150390625, 8.819671630859375, -7.251556396484375, 10.854568481445312, -54.03924560546875, 34.729148864746094, 56.52618408203125, 21.727020263671875, 33.173828125, -46.58990478515625, -66.36323547363281, 84.08917236328125, 15.246513366699219, 34.95184326171875, 22.2237548828125, 83.28329467773438, 41.01496887207031, 14.265892028808594, -8.182472229003906, 20.546966552734375, 0.0, 19.768325805664062, 50.52850341796875, 4.6819305419921875, -5.967559814453125, 69.07940673828125, 12.067138671875, 41.09254455566406, -8.131011962890625, -5.16015625, 1.5061264038085938, 52.17822265625, 24.87799072265625, 55.56420135498047, 11.417327880859375, 32.70757293701172, -3.213886260986328, 52.463623046875, -4.85443115234375, -52.38653564453125, 40.927490234375, 3.0264892578125, -15.682403564453125, -46.418243408203125, 9.063385009765625, -11.047119140625, 16.260589599609375, -35.74017333984375, 46.60662841796875, 7.6179962158203125, 57.45684814453125, 21.0047607421875, 37.6204833984375, -0.050140380859375, -6.8274078369140625, 2.2499237060546875, 10.031600952148438, 5.12310791015625, 14.444480895996094, 26.99148178100586, 0.0, -27.82293701171875, 62.78839111328125, 22.801834106445312, 16.023971557617188, 5.725372314453125, -5.060272216796875, 29.0798397064209, 57.83087921142578, 98.8560791015625, 54.52358627319336, 12.12530517578125, -22.6343994140625, 39.47467041015625, 7.245454788208008, 5.36712646484375, 4.334831237792969, 36.36961364746094, 36.27754592895508, 28.6983642578125, -9.94744873046875, -29.133529663085938, 42.378204345703125, 4.293212890625, 43.543212890625, 34.01805877685547, 39.46734619140625, -5.7166748046875, 1.4461126327514648, -29.406494140625, 47.56011962890625, -27.95538330078125, 7.833465576171875, 6.132177352905273, 34.83930969238281, 6.9668121337890625, 18.50977325439453, 98.61907958984375, 60.971923828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000157.npy"}
{"epoch": 0.3287958115183246, "step": 158, "batch_size": 128, "mean": 16.917186737060547, "std": 30.586597442626953, "min": -57.7410888671875, "p10": -25.48094787597656, "median": 16.751747131347656, "p90": 54.781665039062496, "max": 85.16690063476562, "pos_frac": 0.71875, "sample": [5.143610000610352, -2.2200698852539062, 22.37591552734375, 57.1715087890625, -34.33758544921875, 5.0706329345703125, -3.8760833740234375, 25.911529541015625, -14.187652587890625, 56.827239990234375, -17.49310302734375, 0.63616943359375, -42.873565673828125, -40.8077392578125, 23.896148681640625, 20.705535888671875, 35.055240631103516, 31.592483520507812, 9.290390014648438, 30.356796264648438, 19.203872680664062, 63.23284912109375, 39.502349853515625, 4.829994201660156, 16.10906982421875, 42.07814025878906, 11.829910278320312, -57.7410888671875, 43.369110107421875, 19.836929321289062, -9.64410400390625, 34.83500671386719, -13.042572021484375, 56.042724609375, 22.18853759765625, 6.0541839599609375, 40.941650390625, -36.441349029541016, 20.340187072753906, -44.051025390625, 38.99244689941406, 8.9193115234375, 11.599973678588867, -27.985671997070312, 13.1658935546875, -2.0204086303710938, -7.1880950927734375, 17.846969604492188, 35.30561828613281, 22.428688049316406, 69.721923828125, -4.587762832641602, 0.9059295654296875, 8.955902099609375, 36.337684631347656, 48.922393798828125, -38.39221954345703, -22.462646484375, 64.43026733398438, 19.6298828125, 63.745208740234375, 8.690101623535156, -6.26385498046875, 13.3553466796875, 35.8948974609375, 14.96209716796875, -22.097808837890625, 0.0, 32.00048065185547, 13.96441650390625, 71.33843994140625, -8.069839477539062, -3.16986083984375, 50.65966796875, 25.398895263671875, 54.2412109375, 4.728515625, -25.250885009765625, -11.54522705078125, -2.2544803619384766, 53.72450256347656, 9.6082763671875, -50.88934326171875, 15.744735717773438, 32.9659423828125, 37.7564697265625, 35.28985595703125, -0.080535888671875, 45.180419921875, 40.253173828125, 48.50433349609375, 29.622711181640625, 12.619117736816406, 19.570953369140625, 74.0592041015625, 46.13385009765625, 82.86932373046875, -6.511486053466797, -29.476806640625, -26.01776123046875, 17.394424438476562, 85.16690063476562, -18.379180908203125, 33.57939147949219, 49.648895263671875, 47.86273193359375, 56.69309997558594, 38.0228271484375, 38.937164306640625, 1.6323585510253906, 15.906005859375, 35.12750244140625, 38.161865234375, 1.1862030029296875, 29.592269897460938, -16.260040283203125, 54.092437744140625, 3.5491714477539062, -33.23332977294922, -6.014495849609375, 29.551513671875, 49.325714111328125, -49.42767333984375, 10.453399658203125, 84.96078491210938, 15.900985717773438, 27.453582763671875, 1.0232696533203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000158.npy"}
{"epoch": 0.3308900523560209, "step": 159, "batch_size": 128, "mean": 13.413917541503906, "std": 35.92433547973633, "min": -91.16546630859375, "p10": -31.27299346923828, "median": 9.134428024291992, "p90": 63.155218505859374, "max": 100.69007873535156, "pos_frac": 0.6171875, "sample": [-21.906646728515625, 59.31761169433594, 28.0791015625, 33.514892578125, -3.8999099731445312, -17.690277099609375, -38.29473876953125, 46.658203125, -4.040252685546875, 63.987396240234375, -55.57316589355469, 6.677947998046875, -91.16546630859375, -18.49694061279297, 4.584625244140625, 42.136268615722656, 12.113838195800781, -6.4431915283203125, -35.221588134765625, 47.07566833496094, 29.373977661132812, 4.42083740234375, -13.648727416992188, -29.210845947265625, 21.348846435546875, 63.220672607421875, 21.770782470703125, 62.10670471191406, 35.798370361328125, -8.616195678710938, 21.77399444580078, 2.48626708984375, 42.271331787109375, 17.876853942871094, 19.94091796875, -20.236907958984375, -0.878143310546875, -30.95184326171875, -4.014240264892578, -35.3497314453125, 48.83186340332031, 15.454620361328125, 26.508132934570312, 43.27386474609375, -8.982940673828125, 19.52252197265625, -31.122467041015625, 3.624460220336914, -45.143218994140625, -18.485130310058594, 1.3850326538085938, -7.8664703369140625, 52.39356994628906, 9.781341552734375, 17.77880859375, 52.40141296386719, 1.5283164978027344, 51.20448303222656, -12.564910888671875, -3.04510498046875, 100.69007873535156, 8.836181640625, -29.998008728027344, 12.944305419921875, 79.58026123046875, 14.196319580078125, 54.814605712890625, 0.825103759765625, -8.952789306640625, 12.116737365722656, 2.4802703857421875, 37.4898681640625, -6.582111358642578, 49.476722717285156, 39.5452880859375, -33.287841796875, 2.8053741455078125, 64.91293334960938, -18.268829345703125, 17.503692626953125, -20.0850830078125, -40.458282470703125, 33.979103088378906, 76.42013549804688, 82.11624145507812, -9.61417007446289, -14.515579223632812, -31.624221801757812, -8.753616333007812, 63.127166748046875, 27.99358367919922, 16.899169921875, -20.674774169921875, 27.74969482421875, 12.24700927734375, 33.90814208984375, -36.495880126953125, 63.281005859375, -17.289283752441406, 39.69669723510742, 34.109657287597656, -29.1096134185791, -40.165496826171875, -1.5981292724609375, 94.92819213867188, 82.50994873046875, 22.31268310546875, -56.29582214355469, 50.39942932128906, 9.301483154296875, -3.1492462158203125, -13.6627197265625, 65.25685119628906, -48.872650146484375, 80.37847900390625, -7.8292236328125, 53.3941650390625, 38.10943603515625, 56.277130126953125, 49.901702880859375, 8.95318603515625, 8.96737289428711, -4.5649566650390625, 74.835693359375, 3.8100967407226562, 8.485122680664062, -11.5751953125, 45.46416473388672], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000159.npy"}
{"epoch": 0.33298429319371725, "step": 160, "batch_size": 128, "mean": 20.686126708984375, "std": 31.442041397094727, "min": -49.6551513671875, "p10": -14.066198730468749, "median": 15.284378051757812, "p90": 61.706078338623044, "max": 111.40411376953125, "pos_frac": 0.75, "sample": [-6.240932464599609, 46.03572082519531, 0.0, 90.87716674804688, 26.769012451171875, 26.59368896484375, 29.519744873046875, 15.255523681640625, 82.41171264648438, -3.472900390625, 55.86097717285156, 70.62283325195312, 24.458942413330078, 71.9398193359375, 16.921531677246094, -3.4393310546875, 74.28790283203125, 26.397552490234375, 8.309310913085938, 17.556381225585938, 10.4637451171875, 3.222259521484375, 36.09832000732422, 35.6278076171875, 0.456329345703125, -8.80059814453125, -6.04266357421875, 82.44134521484375, 0.8282394409179688, -49.27838134765625, 11.793182373046875, 108.5205078125, -40.758567810058594, 61.67823028564453, 19.471702575683594, 2.5742530822753906, 12.348159790039062, 4.034656524658203, 34.98136901855469, 8.379228591918945, 36.723609924316406, 52.67530822753906, 10.26922607421875, 0.0, 38.157470703125, -13.640045166015625, 35.279884338378906, -24.7960205078125, -17.378875732421875, 14.001068115234375, 26.076568603515625, 35.84437561035156, 111.40411376953125, 73.18806457519531, -23.542648315429688, 19.542617797851562, 27.50567626953125, -14.525146484375, 11.078437805175781, -13.243980407714844, -11.56573486328125, 55.70628356933594, 7.961273193359375, -43.1258544921875, -10.890609741210938, 58.690338134765625, 61.57965087890625, 57.90318298339844, 5.141290664672852, 15.1376953125, 40.681373596191406, 2.4358367919921875, -0.07805633544921875, 9.249664306640625, 31.141265869140625, 32.77838134765625, 71.07940673828125, 53.258331298828125, 34.87933349609375, 19.205596923828125, -27.0819091796875, -5.3919677734375, 44.28589630126953, -30.535503387451172, -13.137542724609375, 62.34281921386719, 30.125717163085938, 13.33197021484375, 43.766326904296875, -13.8695068359375, 8.455734252929688, 6.593193054199219, 6.445777893066406, -29.459381103515625, 35.25115966796875, -49.6551513671875, -1.21124267578125, 28.77191162109375, 65.39456176757812, -0.326385498046875, 14.696548461914062, 9.445320129394531, -14.747014999389648, 23.81072998046875, 18.66229248046875, 5.885276794433594, 6.176427841186523, -8.320816040039062, 42.134613037109375, 35.15739440917969, 59.3690185546875, 1.0282440185546875, 15.313232421875, 45.919029235839844, -4.12359619140625, 26.974594116210938, 56.384552001953125, 11.672515869140625, 38.42108154296875, 52.92732238769531, 61.77105712890625, 32.54071044921875, -37.26800537109375, 37.3453369140625, 8.979652404785156, 51.648712158203125, 2.8004913330078125, 4.63093376159668], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000160.npy"}
{"epoch": 0.33507853403141363, "step": 161, "batch_size": 128, "mean": 21.655437469482422, "std": 34.0655517578125, "min": -71.77627563476562, "p10": -16.166745758056642, "median": 14.735977172851562, "p90": 71.11555480957031, "max": 93.17303466796875, "pos_frac": 0.71875, "sample": [-3.7876968383789062, 1.1013221740722656, -3.216583251953125, -69.66864013671875, 1.1884174346923828, 21.859649658203125, 2.3243560791015625, 11.28363037109375, 7.2835693359375, 2.77105712890625, 4.285755157470703, 82.36932373046875, 32.46153259277344, -34.19963073730469, 4.301483154296875, 2.8192138671875, 77.31912231445312, 14.2152099609375, 29.207775115966797, -5.63226318359375, 48.81640625, 13.142913818359375, 60.80436706542969, 80.4091796875, -16.11888885498047, 1.48944091796875, 7.2021331787109375, 77.79739379882812, 6.760810852050781, 32.54888153076172, 6.87530517578125, 22.8106746673584, -7.0924835205078125, -15.868698120117188, 22.05548095703125, 23.905532836914062, 34.537109375, 53.216827392578125, -1.2236595153808594, 12.845039367675781, -5.5946044921875, -44.90547180175781, -31.24603271484375, -14.411308288574219, 14.65887451171875, 46.826751708984375, -16.69586181640625, -2.177703857421875, 0.0, 41.945030212402344, 14.592315673828125, 39.627044677734375, 8.570198059082031, 0.6373863220214844, 10.71893310546875, -5.08294677734375, 55.198036193847656, 57.77826690673828, 44.016258239746094, 23.933395385742188, 58.066864013671875, 67.6539306640625, -20.71124267578125, 20.0159912109375, -12.628692626953125, 69.00888061523438, 59.6077880859375, 77.1767578125, -17.750717163085938, 14.813079833984375, 81.9769287109375, 29.771602630615234, 41.5255126953125, 72.64013671875, -32.15643310546875, 72.15771484375, 36.421112060546875, -2.1758270263671875, 1.374359130859375, 41.62200927734375, -16.947731018066406, -0.4009246826171875, 93.17303466796875, 70.66891479492188, 30.926864624023438, 75.60601806640625, 2.8772430419921875, 54.385162353515625, 86.09225463867188, -1.797922134399414, 0.481475830078125, 40.249053955078125, -17.7335205078125, 42.1602783203125, -7.51031494140625, 31.15753173828125, 69.40435791015625, -3.5167007446289062, 39.561614990234375, -8.263809204101562, 24.136810302734375, 63.328765869140625, 23.761444091796875, -14.82763671875, 10.333358764648438, 44.22569274902344, -15.224761962890625, 28.05987548828125, 40.59099578857422, 53.911041259765625, 62.759765625, 69.85894775390625, -16.278411865234375, 40.01995849609375, 28.387876510620117, -54.78369140625, 74.80952453613281, 9.12957763671875, 11.418769836425781, 0.0, -5.933380126953125, 72.9111328125, 61.199127197265625, -71.77627563476562, 33.587921142578125, 20.75592041015625, 2.1349945068359375, 54.8270263671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000161.npy"}
{"epoch": 0.33717277486910996, "step": 162, "batch_size": 128, "mean": 19.552875518798828, "std": 32.82073974609375, "min": -78.44917297363281, "p10": -23.01132354736328, "median": 18.69073486328125, "p90": 62.600039672851565, "max": 92.5682373046875, "pos_frac": 0.765625, "sample": [-7.437255859375, 6.939239501953125, 67.47198486328125, 44.378761291503906, 4.500823974609375, 18.13079833984375, -30.126983642578125, 22.557159423828125, 5.812690734863281, 30.48468017578125, 55.699066162109375, 81.93197631835938, 21.97705078125, 26.439205169677734, -4.7021484375, -5.2407073974609375, 26.033935546875, 30.32342529296875, 1.6671791076660156, 5.039947509765625, -32.5872802734375, 33.704559326171875, -35.863990783691406, 28.968536376953125, 16.143707275390625, 36.11940002441406, 62.6845703125, 11.15069580078125, -44.37669372558594, 6.977020263671875, -0.07483673095703125, 20.8380126953125, 48.13233947753906, -36.68585205078125, 75.45330810546875, 11.629104614257812, 16.20806884765625, -63.29241943359375, -8.139511108398438, 40.25062561035156, 3.8485069274902344, 58.02398681640625, 2.9318370819091797, 67.46185302734375, 66.50689697265625, -78.44917297363281, 62.2540283203125, 27.915245056152344, 28.9620361328125, 68.27178955078125, 3.238218307495117, 33.951171875, 44.080841064453125, 7.1212158203125, 29.274459838867188, -3.912799835205078, -19.57403564453125, 44.98576354980469, 74.2952880859375, 28.992584228515625, 68.33441162109375, 92.5682373046875, -4.59283447265625, -1.0171051025390625, 44.08964538574219, 58.3028564453125, 60.52140808105469, 71.60391998291016, -19.136688232421875, 51.59832763671875, 0.11721038818359375, 8.64400863647461, -28.4215087890625, 22.041290283203125, 90.98135375976562, -30.89667510986328, -29.263961791992188, 3.43505859375, 17.35498046875, 56.192169189453125, -5.872854232788086, 25.0203857421875, 45.26331329345703, 24.400955200195312, 62.97100830078125, 60.876800537109375, 6.663759231567383, -1.90716552734375, 29.352920532226562, 22.954818725585938, 11.129928588867188, 11.686920166015625, 7.277347564697266, -59.446624755859375, 0.0443115234375, 24.264511108398438, 60.28864288330078, 39.330230712890625, 19.25067138671875, 61.82391357421875, 44.99110412597656, -41.78570556640625, 49.04425048828125, 7.8589019775390625, 1.0594329833984375, 60.79290771484375, -11.592071533203125, 3.2117919921875, -4.266365051269531, 49.89320373535156, -25.303359985351562, 4.455116271972656, -8.86444091796875, 40.63262939453125, 29.925216674804688, 38.15483093261719, 11.962493896484375, -1.7081947326660156, 26.92852783203125, 20.977813720703125, 6.974634170532227, -22.029022216796875, 2.3212661743164062, 0.624786376953125, 5.1630096435546875, 62.563812255859375, 27.387252807617188, 6.264404296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000162.npy"}
{"epoch": 0.3392670157068063, "step": 163, "batch_size": 128, "mean": 23.38294219970703, "std": 28.851999282836914, "min": -50.575439453125, "p10": -12.15524444580078, "median": 21.357635498046875, "p90": 62.76804809570312, "max": 89.12969970703125, "pos_frac": 0.75, "sample": [21.42822265625, -2.5346755981445312, 42.21472930908203, -13.781845092773438, 21.07879638671875, -14.274154663085938, 25.543792724609375, 18.271865844726562, 35.322113037109375, 17.957366943359375, 5.7529144287109375, 66.24522399902344, 45.25852966308594, 13.9830322265625, 72.96127319335938, 21.28704833984375, 12.849441528320312, -6.432891845703125, 8.8021240234375, -5.447227478027344, -16.397796630859375, 38.71310806274414, 26.86053466796875, -0.48787689208984375, 16.87567138671875, 52.06352233886719, -9.354818344116211, 14.788223266601562, 51.476837158203125, -3.60284423828125, -6.44677734375, 63.605316162109375, -2.2053375244140625, 46.505645751953125, 30.254913330078125, 26.77655029296875, 10.274383544921875, 50.367401123046875, -7.38787841796875, -7.424018859863281, 45.761474609375, 32.62345886230469, -18.030746459960938, 22.367050170898438, 3.99261474609375, 21.86474609375, -50.575439453125, 64.85787963867188, 47.075836181640625, -33.33905029296875, 37.65388488769531, 7.90447998046875, 45.025421142578125, 41.916961669921875, -1.149627685546875, 61.01069641113281, 18.82550048828125, 63.41204833984375, 75.12771606445312, 43.02996826171875, 2.5234375, -2.172393798828125, -11.4581298828125, 27.288665771484375, -10.345626831054688, -10.143280029296875, 8.134260177612305, 51.70365905761719, 6.085113525390625, 40.67195129394531, -43.10980224609375, 62.59649658203125, 19.27770233154297, 30.146942138671875, 12.728179931640625, -17.587982177734375, 76.78509521484375, -15.177200317382812, 67.06735229492188, 63.1683349609375, 36.32860565185547, 48.15354919433594, 23.01654052734375, 19.78194808959961, 46.610206604003906, 41.395469665527344, 37.40760803222656, 2.723236083984375, 65.9730224609375, 51.424591064453125, 8.41618537902832, 7.5054931640625, 21.96788787841797, 89.12969970703125, 38.72191619873047, -24.580078125, -27.39654541015625, 19.035112380981445, 79.41351318359375, 33.61802673339844, 55.765098571777344, 51.58160400390625, -19.0794677734375, 57.323486328125, -6.987640380859375, 20.4266357421875, 60.05052185058594, 14.55926513671875, 31.10479736328125, 32.20757293701172, 60.598846435546875, 39.31855773925781, 17.18768310546875, -0.42496299743652344, 9.247322082519531, 19.347747802734375, 35.96478271484375, 59.43531799316406, 30.480316162109375, 67.90862274169922, -3.560760498046875, 6.385650634765625, -44.381195068359375, 54.12774658203125, 48.990386962890625, 20.081146240234375, 9.045433044433594, -1.6120376586914062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000163.npy"}
{"epoch": 0.3413612565445026, "step": 164, "batch_size": 128, "mean": 20.032072067260742, "std": 32.03227233886719, "min": -59.92120361328125, "p10": -14.055184936523437, "median": 16.195388793945312, "p90": 63.369439697265626, "max": 102.40542602539062, "pos_frac": 0.734375, "sample": [-14.232086181640625, 12.328887939453125, -2.387317657470703, 56.5462646484375, 0.14482879638671875, -1.04534912109375, -6.528604507446289, -6.7703399658203125, 36.62864685058594, 34.068603515625, 44.87120056152344, -13.054290771484375, 45.73558044433594, 37.88165283203125, 28.112213134765625, -59.92120361328125, -2.4286117553710938, 56.534637451171875, 6.816423416137695, 67.41876220703125, 0.857666015625, -48.6153564453125, 50.133331298828125, 12.4111328125, 11.612396240234375, -17.53104019165039, 28.564483642578125, 3.185333251953125, -21.865371704101562, 6.7349090576171875, 37.4171142578125, -0.203704833984375, 6.95684814453125, 63.40257263183594, -5.4901275634765625, 51.612213134765625, -40.737579345703125, 3.233978271484375, 1.7251129150390625, -44.751129150390625, -9.072906494140625, 72.38154602050781, 41.698760986328125, 25.4454345703125, 79.84866333007812, 9.842514038085938, 56.49542236328125, 1.452911376953125, 1.17218017578125, 3.457489013671875, -26.661056518554688, 12.048263549804688, 18.904296875, 40.24656677246094, -14.547531127929688, 65.86164855957031, -1.6958770751953125, 43.668548583984375, 37.892120361328125, 33.209014892578125, 1.201019287109375, 9.096160888671875, 64.04653930664062, 63.35523986816406, -13.528854370117188, 66.66683959960938, 28.62194061279297, -1.583251953125, 47.1580810546875, 15.99212646484375, 85.60055541992188, 31.755287170410156, 70.3692626953125, 2.655364990234375, -7.791055679321289, -0.08888626098632812, 22.77313232421875, 4.380344390869141, -49.764373779296875, 16.398651123046875, 31.0345458984375, 57.32398986816406, -9.765151977539062, 3.34075927734375, 1.314605712890625, 42.98243713378906, 56.850830078125, 51.754608154296875, 40.45904541015625, 30.847869873046875, -6.641971588134766, 48.98499298095703, -0.443939208984375, 43.59503173828125, 11.08660888671875, 53.029266357421875, 51.50708770751953, 65.15887451171875, -6.703094482421875, 1.5578460693359375, -36.579437255859375, 28.561981201171875, 0.8411483764648438, 29.859603881835938, 16.464920043945312, -8.0853271484375, 56.977264404296875, -13.9793701171875, 49.140342712402344, 10.203475952148438, 21.655136108398438, 54.740814208984375, 11.9263916015625, 16.838134765625, -55.3687744140625, -24.546783447265625, 7.6898651123046875, -9.247894287109375, 101.5672607421875, 18.366455078125, 16.446582794189453, 48.19879150390625, 26.283447265625, 33.702301025390625, 45.804595947265625, 64.73124694824219, 102.40542602539062, 13.898591995239258], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000164.npy"}
{"epoch": 0.34345549738219894, "step": 165, "batch_size": 128, "mean": 17.288249969482422, "std": 30.315807342529297, "min": -64.40272521972656, "p10": -14.081188964843749, "median": 8.412208557128906, "p90": 57.75978469848633, "max": 93.22119140625, "pos_frac": 0.6875, "sample": [5.2064971923828125, 57.683349609375, -8.017120361328125, 48.18572998046875, -13.54718017578125, 4.089744567871094, 30.722213745117188, 3.874725341796875, 59.00335693359375, 45.729583740234375, 68.23553466796875, 4.7955169677734375, -62.29046630859375, 50.586151123046875, 61.142547607421875, 14.284912109375, 3.06353759765625, 69.0618896484375, -5.992767333984375, 44.8365478515625, 39.749908447265625, -47.028656005859375, 56.44818115234375, 37.579620361328125, 8.264694213867188, -27.684818267822266, -24.28607177734375, -11.208526611328125, 18.154815673828125, 34.508392333984375, 4.1273193359375, -7.6397705078125, -1.283203125, 1.1806488037109375, 0.38782501220703125, 13.674934387207031, 31.383132934570312, -5.7192840576171875, -5.861249923706055, -11.327156066894531, 16.49407958984375, 14.849571228027344, 57.187835693359375, -20.996841430664062, 12.94268798828125, 57.0001220703125, -1.0242881774902344, 0.200592041015625, -12.246034622192383, 10.570114135742188, 87.83392333984375, 6.7456512451171875, 5.000518798828125, 3.92767333984375, 64.009521484375, 32.74383544921875, -7.252593994140625, 2.88458251953125, -5.020694732666016, 7.094024658203125, -5.382781982421875, 68.0853271484375, -12.737045288085938, 47.0921630859375, 28.019073486328125, -15.32720947265625, 25.37457275390625, -4.699676513671875, -3.378936767578125, -7.3470611572265625, 5.81829833984375, -1.6879196166992188, 21.4296875, 10.15880012512207, 30.278732299804688, 4.5273590087890625, 41.71685791015625, -26.851806640625, 52.139984130859375, 15.04608154296875, -29.97308349609375, 36.86127471923828, 68.99740600585938, -16.89288330078125, 7.523571014404297, 41.041473388671875, -0.834686279296875, 64.90826416015625, 54.944244384765625, 21.46533203125, 13.930999755859375, 1.243988037109375, 40.638671875, 45.5228385925293, 54.12092590332031, -18.80921173095703, -8.80804443359375, 22.802703857421875, 47.40027618408203, 5.4097137451171875, 93.22119140625, -3.5053558349609375, 40.29206848144531, 43.50749206542969, 64.74485778808594, -20.054855346679688, -3.129016876220703, 16.31695556640625, 57.13203430175781, 4.219757080078125, 0.62628173828125, 5.0483551025390625, 68.49411010742188, 31.81402587890625, 52.846923828125, 49.549285888671875, 3.989532470703125, 15.410888671875, 8.559722900390625, 49.39679718017578, -1.1223716735839844, 49.92967224121094, -64.40272521972656, -28.747222900390625, -3.5979995727539062, 57.938133239746094, -9.041275024414062, -7.329010009765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000165.npy"}
{"epoch": 0.34554973821989526, "step": 166, "batch_size": 128, "mean": 15.422724723815918, "std": 36.769859313964844, "min": -79.50958251953125, "p10": -28.507083129882812, "median": 11.957176208496094, "p90": 65.17687683105467, "max": 109.36581420898438, "pos_frac": 0.6640625, "sample": [15.10443115234375, 3.7290191650390625, -29.38330078125, 15.728408813476562, -2.481292724609375, 11.011802673339844, 67.7720947265625, 62.06854248046875, -3.7762889862060547, 3.1449337005615234, 12.337104797363281, -4.894737243652344, 106.18606567382812, -14.564701080322266, 3.0016632080078125, -9.22802734375, 24.450653076171875, 24.967193603515625, 5.778839111328125, 1.1678619384765625, -1.4986343383789062, -33.744903564453125, 24.096084594726562, 67.71125793457031, -58.90911865234375, -28.131561279296875, 1.7212238311767578, 34.05915832519531, -58.55145263671875, -19.645599365234375, -10.14447021484375, 19.88983154296875, 29.963180541992188, 77.33856201171875, 64.09071350097656, 80.4805908203125, 3.73675537109375, -10.1348876953125, -16.374107360839844, -76.41213989257812, 5.203559875488281, -31.853485107421875, 5.219970703125, -2.7259750366210938, -13.067420959472656, 71.63589477539062, 36.830810546875, 20.29974365234375, -17.79962158203125, 37.717803955078125, -2.6739349365234375, 91.9356689453125, -47.905426025390625, 72.92120361328125, 96.76846313476562, -53.74656677246094, -35.779083251953125, 94.90072631835938, 34.583770751953125, 43.95457458496094, -31.759361267089844, -9.59356689453125, -19.22088623046875, 38.06414794921875, 52.511138916015625, -23.42669677734375, 53.97674560546875, 36.373779296875, 5.498321533203125, -20.964889526367188, 20.185592651367188, 24.53729248046875, 75.13800048828125, -7.944660186767578, -15.567535400390625, 23.9072265625, 16.040504455566406, -1.1539382934570312, 13.34139633178711, 11.604736328125, -4.564105987548828, 49.154632568359375, 19.303985595703125, 8.488845825195312, 12.906097412109375, 75.552734375, 2.29962158203125, 42.99340057373047, 17.893951416015625, -41.8291015625, 7.09466552734375, -11.089393615722656, 11.016471862792969, 58.820953369140625, 50.950836181640625, 28.371200561523438, 17.611236572265625, 53.388404846191406, 28.939987182617188, 6.91912841796875, 25.789215087890625, 1.1887130737304688, -0.754180908203125, 50.06544494628906, 1.8021354675292969, 33.73760986328125, 54.18180847167969, 48.61041259765625, -10.328533172607422, 62.93286895751953, -45.7508544921875, -79.50958251953125, 15.1981201171875, 43.430511474609375, 58.657684326171875, -21.945953369140625, 13.0887451171875, -1.3848876953125, 19.7493896484375, -1.4697113037109375, 62.658660888671875, -1.506011962890625, 109.36581420898438, 12.760643005371094, 1.585662841796875, 35.38893127441406, 12.309616088867188, 10.404098510742188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000166.npy"}
{"epoch": 0.34764397905759165, "step": 167, "batch_size": 128, "mean": 18.556583404541016, "std": 33.45942306518555, "min": -63.68060302734375, "p10": -19.460024261474608, "median": 13.376800537109375, "p90": 65.08629608154297, "max": 101.56997680664062, "pos_frac": 0.71875, "sample": [46.94677734375, 6.123115539550781, 15.958526611328125, 44.70770263671875, 13.928482055664062, 57.99298095703125, 45.67303466796875, 75.516357421875, 33.0308837890625, 16.144485473632812, 61.15492248535156, -46.646270751953125, 69.02099609375, 1.3068618774414062, 13.6959228515625, 79.53561401367188, 16.730865478515625, 71.60687255859375, 0.32550048828125, 13.05767822265625, 3.136199951171875, 29.31182861328125, -58.54508972167969, -63.60554504394531, -1.5261459350585938, 0.37841796875, -27.813980102539062, 54.94468688964844, 33.8472900390625, 58.35877990722656, 31.770660400390625, 65.85774230957031, 19.097084045410156, 64.75567626953125, 45.39506530761719, 8.608078002929688, 16.46851348876953, -36.08485412597656, 18.2725830078125, -9.034751892089844, 72.12802124023438, -5.38409423828125, 8.496124267578125, 3.163238525390625, 0.3049774169921875, -63.68060302734375, 0.0, -0.030487060546875, 2.122467041015625, 34.85035705566406, 11.313308715820312, 75.38360595703125, -32.950904846191406, 67.48532104492188, -12.001861572265625, 16.26922607421875, 47.51519012451172, 12.428741455078125, 35.15789794921875, 1.96624755859375, -33.56170654296875, 2.5772705078125, 88.1229248046875, 0.229644775390625, 27.9471435546875, 11.89306640625, 18.5960693359375, -14.852813720703125, 38.22825622558594, 3.5390281677246094, -0.0343780517578125, 1.3573989868164062, 40.86254119873047, -37.378753662109375, 49.18898010253906, 42.392662048339844, 59.90867614746094, -0.7337646484375, 27.40618896484375, 3.6299591064453125, -2.2470245361328125, 46.85585021972656, 42.49847412109375, -26.492691040039062, 58.412750244140625, -6.835746765136719, 17.577713012695312, -4.267425537109375, 32.9827880859375, 41.43004608154297, 36.155609130859375, 9.202201843261719, 39.63702392578125, 18.847900390625, 38.10479736328125, 83.00128173828125, -5.7738494873046875, -19.629165649414062, -19.387535095214844, -11.255050659179688, -5.19134521484375, 5.614479064941406, -11.701446533203125, 69.05221557617188, -8.42352294921875, 9.4600830078125, 101.56997680664062, 23.346527099609375, 46.72509765625, -12.3929443359375, 6.87286376953125, 46.3050537109375, -33.038299560546875, -4.258308410644531, -2.0521697998046875, 2.9378509521484375, 49.5994873046875, 5.5255126953125, -3.10748291015625, 53.06013488769531, 46.652008056640625, 34.48340606689453, 10.316818237304688, -2.26837158203125, -57.09625244140625, 55.451873779296875, 5.604156494140625, 80.11857604980469], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000167.npy"}
{"epoch": 0.34973821989528797, "step": 168, "batch_size": 128, "mean": 23.916889190673828, "std": 33.85984802246094, "min": -68.74713134765625, "p10": -11.784298706054686, "median": 19.3778076171875, "p90": 66.77171936035157, "max": 100.16192626953125, "pos_frac": 0.75, "sample": [29.47234344482422, 17.144866943359375, 77.12734985351562, 65.36322021484375, 50.160491943359375, 27.32811164855957, 47.305694580078125, 25.80425262451172, 50.53271484375, -8.966293334960938, 48.028564453125, -11.65325927734375, 3.697509765625, 8.279838562011719, 0.1757049560546875, 5.18701171875, 26.994949340820312, 32.876495361328125, -6.347728729248047, 55.22937774658203, 16.742042541503906, 62.781341552734375, 18.445388793945312, 14.39971923828125, 62.535308837890625, -1.9202880859375, 22.736221313476562, 48.47282791137695, -2.0565109252929688, 1.8954238891601562, 0.0, 14.981842041015625, 77.02996826171875, 15.898422241210938, 67.38778686523438, 29.062225341796875, 17.699920654296875, 12.130035400390625, 33.347808837890625, 1.76416015625, 70.18316650390625, 20.1019287109375, -0.80950927734375, 43.969390869140625, -5.58380126953125, 99.64974975585938, 18.6536865234375, 40.48602294921875, 82.57421875, -14.489715576171875, 54.64019775390625, -8.676406860351562, 36.030033111572266, 15.4783935546875, -51.071563720703125, 36.86540222167969, 52.31829833984375, 62.638031005859375, 48.80773162841797, 6.575927734375, -23.357421875, -25.116615295410156, 1.56622314453125, 0.4864044189453125, 47.428070068359375, -0.8492717742919922, 60.6195068359375, 60.3472900390625, 100.16192626953125, 28.112823486328125, 77.45721435546875, 9.712055206298828, -68.74713134765625, 45.585601806640625, 67.04954528808594, 0.4293479919433594, 3.1877059936523438, 52.4835205078125, 5.4656524658203125, -2.881378173828125, -1.040435791015625, 6.63677978515625, 42.59385681152344, 21.16363525390625, 67.01885986328125, 54.80010986328125, 68.93135070800781, 43.9034423828125, 81.8123779296875, 55.928314208984375, -27.855026245117188, -6.34039306640625, 64.76519775390625, -12.090057373046875, 7.923408508300781, 45.84136962890625, -12.983016967773438, 15.970550537109375, 33.31239318847656, -57.902191162109375, 18.265518188476562, 8.664627075195312, 54.83820343017578, 30.498992919921875, 28.783355712890625, 79.7860107421875, -2.8687820434570312, -5.335845947265625, 4.88104248046875, -1.838592529296875, 6.188507080078125, -0.14617919921875, -31.45745849609375, -39.279327392578125, 56.41077423095703, 20.14544677734375, 49.85797119140625, 66.66580200195312, 65.95330810546875, 48.86480712890625, -1.292388916015625, 37.574249267578125, 9.046966552734375, -10.30072021484375, -37.45208740234375, 11.5120849609375, 64.03353881835938, -67.58547973632812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000168.npy"}
{"epoch": 0.3518324607329843, "step": 169, "batch_size": 128, "mean": 27.732311248779297, "std": 35.301143646240234, "min": -58.649169921875, "p10": -8.40015182495117, "median": 18.321372985839844, "p90": 76.87588195800781, "max": 112.595947265625, "pos_frac": 0.7890625, "sample": [-2.8497772216796875, 46.970550537109375, 89.94645690917969, 44.8177490234375, -21.5521240234375, 13.19512939453125, 1.9432086944580078, 5.796600341796875, 2.467041015625, 17.633712768554688, 20.329727172851562, 60.18534851074219, 59.225276947021484, 64.01504516601562, -46.95550537109375, -2.4177780151367188, 56.133758544921875, 10.8778076171875, 5.323492050170898, -3.6309051513671875, -0.956085205078125, 18.963851928710938, 51.519683837890625, 16.259727478027344, 53.80284118652344, 68.96115112304688, 28.916107177734375, -20.01116943359375, 39.107269287109375, 50.363311767578125, 64.06991577148438, 67.39950561523438, 56.1488037109375, -6.655052185058594, 3.9416275024414062, 9.736518859863281, -14.477193832397461, -58.649169921875, 29.93426513671875, -1.592529296875, -9.418434143066406, 68.1214599609375, -12.435157775878906, 15.265571594238281, 0.0, -45.864356994628906, 9.680927276611328, 37.03460693359375, 21.37615966796875, 26.7235107421875, 3.4285888671875, 10.3392333984375, 112.595947265625, 64.94049072265625, 44.121299743652344, 54.26179504394531, -52.67626190185547, 53.96336364746094, 0.43236732482910156, 0.33159637451171875, 72.5855712890625, 50.505645751953125, 30.64208984375, -10.069480895996094, 41.6429443359375, 77.27508544921875, 64.19593811035156, 88.54888916015625, 17.67889404296875, 76.75088500976562, 1.0137939453125, 92.71636962890625, 7.3145294189453125, 0.181793212890625, 41.9515380859375, -34.85237121582031, 14.955413818359375, 79.03970336914062, 48.973419189453125, -6.625701904296875, 48.38069152832031, 10.418365478515625, 16.787248611450195, 53.0050048828125, 27.658721923828125, 57.37550354003906, 66.78976440429688, -7.9637451171875, 42.729400634765625, 32.23463439941406, -0.9162368774414062, 41.20159912109375, 6.154243469238281, 103.58328247070312, 33.495269775390625, 9.863292694091797, 14.700775146484375, 6.09564208984375, 85.62161254882812, 36.80415344238281, 6.922601699829102, 53.14037322998047, 95.48507690429688, 16.67108154296875, -32.362152099609375, -29.250442504882812, 7.935638427734375, -2.3422317504882812, 96.70770263671875, 17.214019775390625, 46.7786865234375, -0.024190902709960938, 69.28118896484375, 54.75677490234375, 52.669830322265625, 0.0, 15.35546875, 3.6281280517578125, 28.0140380859375, 41.689247131347656, 8.330406188964844, 0.01839447021484375, 12.386688232421875, 13.815105438232422, 91.69192504882812, 90.52789306640625, 77.16754150390625, -5.377723693847656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000169.npy"}
{"epoch": 0.3539267015706806, "step": 170, "batch_size": 128, "mean": 24.251705169677734, "std": 33.54823303222656, "min": -55.788787841796875, "p10": -9.703698730468748, "median": 20.649517059326172, "p90": 67.85545501708984, "max": 116.73541259765625, "pos_frac": 0.75, "sample": [4.962005615234375, 14.294097900390625, 54.17308044433594, -8.314956665039062, 50.85638427734375, 10.45330810546875, 0.6131668090820312, 106.996826171875, 4.422599792480469, 12.160552978515625, 32.502532958984375, 94.25262451171875, 46.541534423828125, 26.63243865966797, -39.3397216796875, 105.98104858398438, -24.65814208984375, 32.0931396484375, -2.01031494140625, 15.39248275756836, -31.918487548828125, 6.11529541015625, 31.886932373046875, 38.97093200683594, 6.699211120605469, -33.22002410888672, -19.88580322265625, 3.205617904663086, 71.0943603515625, -8.72186279296875, 27.735931396484375, 23.77998161315918, 46.69624328613281, 11.288789749145508, 15.444007873535156, -3.25494384765625, -0.0675048828125, 4.4502105712890625, 54.8720703125, 60.57080078125, 1.6829376220703125, 37.55719757080078, -0.8588104248046875, 65.53048706054688, 50.42449188232422, 23.638702392578125, 33.7314453125, -1.6513671875, -23.53619384765625, 69.50784301757812, 68.55973815917969, 54.589210510253906, 53.90374755859375, -48.717742919921875, 34.089111328125, -9.045333862304688, 72.25051879882812, -10.882064819335938, 66.81373596191406, 47.43109893798828, 17.0531005859375, 78.03591918945312, 58.667510986328125, -20.20538330078125, 46.57246780395508, 50.85474395751953, -4.2158203125, 15.6358642578125, 61.82774353027344, 3.58837890625, 33.39872741699219, 56.480804443359375, 7.674121856689453, 21.2293701171875, 47.622886657714844, 51.444244384765625, 0.0, 77.5712890625, 23.124542236328125, 48.65321350097656, -31.40869140625, 47.599365234375, 3.8560791015625, 52.941009521484375, 57.7637939453125, 34.59880065917969, -28.127593994140625, 11.713134765625, 41.64141845703125, -55.788787841796875, 27.2547607421875, 23.3521728515625, -5.766845703125, 45.73834228515625, 1.3807830810546875, -3.040468215942383, 76.66140747070312, -1.0888671875, 30.53094482421875, 116.73541259765625, -1.932861328125, -9.198684692382812, 20.069664001464844, -2.127349853515625, -6.98046875, 10.50543212890625, 26.362060546875, 12.490936279296875, 1.5784912109375, 17.968475341796875, -3.5138397216796875, 6.85638427734375, 88.015380859375, 16.8076171875, -33.226051330566406, 22.373367309570312, 7.7071685791015625, 63.60064697265625, 31.741607666015625, 2.3463573455810547, 0.9073295593261719, 39.3907470703125, 74.30682373046875, 62.53144073486328, 6.0147552490234375, 40.007110595703125, -8.262847900390625, 67.55361938476562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000170.npy"}
{"epoch": 0.35602094240837695, "step": 171, "batch_size": 128, "mean": 24.558422088623047, "std": 34.36756134033203, "min": -37.784454345703125, "p10": -19.664871215820312, "median": 20.196243286132812, "p90": 73.17113189697264, "max": 96.41633605957031, "pos_frac": 0.7421875, "sample": [91.5731201171875, 17.691162109375, 31.59051513671875, 17.25701904296875, 65.44476318359375, 67.431884765625, 65.82544708251953, -26.739898681640625, -11.578790664672852, 11.94708251953125, -21.817581176757812, 0.9918746948242188, 80.07089233398438, -28.616897583007812, 3.0184326171875, -1.795318603515625, -13.635032653808594, 19.284591674804688, 32.5091552734375, 80.100341796875, 55.824737548828125, -5.45147705078125, 19.763702392578125, 83.84701538085938, 65.67044067382812, 27.94287109375, -26.446319580078125, 11.671730041503906, -27.28424072265625, 17.170166015625, 9.718437194824219, 12.634475708007812, -13.196075439453125, 54.38249588012695, 57.49444580078125, 75.94927978515625, 57.01966857910156, 0.847808837890625, 27.883880615234375, -5.184074401855469, 15.62005615234375, -13.819610595703125, -6.575870513916016, -7.2769775390625, -16.70763397216797, 93.7200927734375, 86.57586669921875, 12.576623916625977, -37.784454345703125, 4.8573150634765625, 66.89413452148438, 27.812255859375, 9.12548828125, 68.04690551757812, 31.54278564453125, 26.362014770507812, 42.37518310546875, 75.88327026367188, 80.59619140625, -19.200653076171875, 53.97320556640625, 32.06483459472656, -9.80694580078125, 58.73377227783203, 33.52025604248047, 64.09263610839844, 30.831329345703125, 12.458105087280273, 57.194854736328125, 43.72410583496094, 28.55034637451172, 79.35479736328125, -5.21893310546875, 3.74114990234375, -28.68914794921875, -16.507827758789062, 75.80094909667969, 3.5375709533691406, -29.236358642578125, 10.048141479492188, -0.3738727569580078, 1.9244232177734375, -11.96612548828125, -26.486297607421875, 17.389892578125, -35.37675476074219, 51.81467819213867, 50.421051025390625, -4.03314208984375, 25.933990478515625, 50.09130096435547, 96.41633605957031, -37.73404312133789, 11.929183959960938, 29.72869873046875, 61.917694091796875, 55.9530029296875, 2.9430503845214844, 72.0440673828125, 27.645111083984375, 22.564346313476562, 49.9775390625, -28.425430297851562, 11.216232299804688, 24.2857666015625, 17.5797119140625, 31.656780242919922, 60.936676025390625, 21.14563751220703, 1.4997100830078125, 20.6287841796875, -20.748046875, 60.867919921875, 57.759063720703125, 54.33415222167969, 16.543731689453125, 24.74310302734375, 4.789497375488281, 21.963119506835938, 8.036880493164062, 95.38262939453125, 51.267478942871094, -10.082347869873047, 0.8577938079833984, -3.2941741943359375, 71.97027587890625, 27.242324829101562, -11.0050048828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000171.npy"}
{"epoch": 0.3581151832460733, "step": 172, "batch_size": 128, "mean": 16.640243530273438, "std": 35.42803192138672, "min": -86.55230712890625, "p10": -18.05816898345947, "median": 15.20101547241211, "p90": 64.00453186035155, "max": 88.373291015625, "pos_frac": 0.6796875, "sample": [16.011611938476562, -10.801956176757812, 2.54815673828125, 24.85369873046875, 34.26329040527344, 17.743675231933594, -2.48406982421875, -8.701812744140625, 5.527763366699219, -0.90509033203125, -3.5307083129882812, -1.3746719360351562, -63.242218017578125, 5.966014862060547, 48.70204162597656, 12.856847763061523, 38.086830139160156, 71.45098876953125, 44.082611083984375, 11.796318054199219, -9.67132568359375, 9.040618896484375, 23.687713623046875, 80.58786010742188, 6.073516845703125, 46.076934814453125, 41.238311767578125, -8.210540771484375, -8.934181213378906, 62.1380615234375, 3.8135986328125, 28.360595703125, -68.6436767578125, 7.92877197265625, 17.8223876953125, 50.807342529296875, -16.350799560546875, 76.17709350585938, -20.209762573242188, -2.5682830810546875, -6.937955856323242, 15.292312622070312, -3.0135498046875, -1.1015281677246094, 77.13676452636719, -86.55230712890625, -8.058670043945312, 38.9521484375, 76.7103271484375, 59.0777587890625, -6.2682952880859375, -26.715911865234375, 21.3984375, -0.3459014892578125, 31.58392333984375, 15.243263244628906, 63.050048828125, 69.86262512207031, -17.17041015625, -3.5071563720703125, -7.47784423828125, 18.3173828125, 25.513519287109375, -2.2393035888671875, 60.73797607421875, 47.404693603515625, -5.647705078125, 7.8231964111328125, 21.289588928222656, 9.8070068359375, 88.373291015625, -61.839996337890625, 2.127838134765625, 41.9722900390625, 70.8681640625, 57.81292724609375, 0.839263916015625, 1.8134002685546875, 20.43096923828125, -69.70950317382812, 27.91156005859375, 15.158767700195312, 24.85881805419922, -60.3780517578125, 17.480712890625, 50.6387939453125, 9.343460083007812, 67.3740234375, 17.040802001953125, -46.80352783203125, 76.8814697265625, -6.30877685546875, 67.84979248046875, -36.000244140625, -42.3883171081543, 23.402755737304688, 25.4947509765625, 17.170196533203125, -20.129606246948242, 14.016807556152344, -5.780975341796875, 88.15478515625, 52.4840087890625, 44.562740325927734, 28.7520751953125, 26.601348876953125, -10.053428649902344, 20.256057739257812, 38.762813568115234, 7.146392822265625, 14.3880615234375, 53.55805206298828, 36.07234191894531, 62.69606018066406, 60.113372802734375, 41.54808044433594, -8.342529296875, 9.52825927734375, 7.848249435424805, 53.9268798828125, -76.84597778320312, 66.23165893554688, 0.161407470703125, 10.89666748046875, 0.0, 39.2042236328125, 35.86920166015625, -9.269287109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000172.npy"}
{"epoch": 0.36020942408376966, "step": 173, "batch_size": 128, "mean": 20.525074005126953, "std": 34.26640701293945, "min": -70.1417236328125, "p10": -17.25451965332031, "median": 18.838748931884766, "p90": 64.67021179199219, "max": 122.57144165039062, "pos_frac": 0.6953125, "sample": [-2.289703369140625, -70.1417236328125, -48.583473205566406, 21.344924926757812, 28.42828369140625, -15.529327392578125, 48.407470703125, -40.759368896484375, 46.66424560546875, 111.66595458984375, -11.69149398803711, 24.677825927734375, -13.741706848144531, 39.65069580078125, 50.036468505859375, 63.164093017578125, 12.552032470703125, 4.580577850341797, 20.8265380859375, 79.639892578125, 66.35771179199219, 12.159652709960938, 7.294464111328125, 36.112091064453125, 24.85558319091797, 16.55426025390625, -3.2028656005859375, 7.031768798828125, 40.973388671875, 53.29322814941406, -0.1361083984375, 47.78224182128906, -3.27215576171875, 5.694683074951172, 6.0753936767578125, 32.64750671386719, -34.977813720703125, 29.594482421875, 23.33626937866211, 40.233001708984375, -9.45953369140625, 25.763916015625, 40.538116455078125, 44.05865478515625, 54.619384765625, 36.18231201171875, -2.7035675048828125, -8.65057373046875, 3.82763671875, 11.3187255859375, -13.200965881347656, -42.647369384765625, -29.54412841796875, 122.57144165039062, 14.819091796875, -29.364776611328125, 10.457351684570312, -21.27996826171875, 28.211036682128906, 99.86569213867188, 62.17381286621094, 65.76788330078125, 2.9217567443847656, 56.38409423828125, 61.596405029296875, 64.2706298828125, -50.413238525390625, 34.394317626953125, 25.740325927734375, 21.653335571289062, 25.672325134277344, -2.6723403930664062, 70.95458984375, 49.32518768310547, 3.230133056640625, 49.978515625, 18.23847198486328, 70.05364990234375, -3.5308151245117188, 12.999717712402344, 24.49847412109375, -26.20660400390625, 3.158243179321289, 3.82342529296875, 57.223548889160156, 6.510467529296875, 38.542938232421875, 65.60256958007812, -3.57275390625, 6.58734130859375, 20.069305419921875, -3.87127685546875, 54.1221923828125, 30.923583984375, 68.52044677734375, 28.148956298828125, 58.877105712890625, 33.348052978515625, 41.690399169921875, 58.69569396972656, 69.98094177246094, 16.420867919921875, -2.750843048095703, -12.059234619140625, -4.858730316162109, 62.242225646972656, -31.92584228515625, 53.105743408203125, 42.62248992919922, 17.9774169921875, -4.6085357666015625, 3.246368408203125, -14.059494018554688, 12.454437255859375, 0.27341651916503906, -8.935150146484375, -4.068603515625, -37.36576843261719, 81.04180908203125, -5.461585998535156, 67.8392333984375, -9.955474853515625, 0.0, 19.43902587890625, 33.9097900390625, -39.20848083496094, -5.7628326416015625, 29.554046630859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000173.npy"}
{"epoch": 0.362303664921466, "step": 174, "batch_size": 128, "mean": 26.14525604248047, "std": 33.898780822753906, "min": -57.451171875, "p10": -14.01653366088867, "median": 20.28052520751953, "p90": 74.30059204101562, "max": 118.425048828125, "pos_frac": 0.7734375, "sample": [54.77222442626953, 1.933115005493164, 5.387353897094727, 0.8262405395507812, 90.46099853515625, 63.43572998046875, 75.972900390625, -30.91082763671875, 8.273895263671875, -0.2336273193359375, 69.21426391601562, 56.810585021972656, 2.23712158203125, 51.7545166015625, 13.2840576171875, 25.71826171875, 68.6025390625, 100.33120727539062, 12.510116577148438, 14.224922180175781, -7.3076171875, 73.65911865234375, 1.283050537109375, 82.35275268554688, 81.1302490234375, -4.07470703125, 46.34153747558594, 8.564971923828125, 80.36991882324219, 51.970306396484375, -2.2627182006835938, 54.66099548339844, 118.425048828125, -24.723388671875, -22.868942260742188, -9.791244506835938, 61.72705078125, 77.3592529296875, 90.31973266601562, 4.00714111328125, 15.628631591796875, 36.8948974609375, 39.309471130371094, 33.396881103515625, -2.5485687255859375, 77.62335205078125, 51.0255126953125, 38.97601318359375, -9.473388671875, 107.3973388671875, -10.421585083007812, -24.004638671875, 60.968666076660156, 35.6600341796875, -14.380531311035156, 3.77435302734375, 11.291725158691406, -1.4234886169433594, 25.486572265625, 21.6055908203125, 22.209823608398438, 5.6215362548828125, 75.79736328125, 68.138916015625, 21.74615478515625, -17.3948974609375, -4.3969879150390625, 42.235137939453125, 12.230178833007812, 34.71672058105469, 23.98394775390625, 51.20710754394531, 61.5299072265625, 61.89093017578125, 48.398529052734375, 32.97803497314453, 10.658355712890625, 15.226837158203125, 60.954071044921875, -18.256927490234375, -39.087677001953125, 45.09471130371094, 1.30816650390625, 43.6263427734375, 20.935684204101562, 1.7471446990966797, 65.99575805664062, 9.614675521850586, -14.49090576171875, 8.142024993896484, 8.210159301757812, 17.422378540039062, 42.095672607421875, 41.85539245605469, 47.01611328125, 48.034423828125, 6.71331787109375, 86.3792724609375, -1.223337173461914, 16.95793914794922, 57.94038391113281, -16.52178955078125, 46.15765380859375, 22.12158203125, 10.7261962890625, 13.204681396484375, -12.437431335449219, 0.8875446319580078, -3.072021484375, -27.2716064453125, 46.30158996582031, -2.85748291015625, 3.4943161010742188, 35.929115295410156, 34.696258544921875, -17.13323974609375, 69.572021484375, -13.86053466796875, 6.1405029296875, -57.451171875, 9.9212646484375, 14.915679931640625, 42.712860107421875, 23.19830322265625, 43.626495361328125, 0.5764617919921875, 19.6253662109375, -12.907073974609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000174.npy"}
{"epoch": 0.3643979057591623, "step": 175, "batch_size": 128, "mean": 21.093074798583984, "std": 36.59870910644531, "min": -96.70848083496094, "p10": -19.282141113281245, "median": 18.345921516418457, "p90": 67.13979492187498, "max": 122.88525390625, "pos_frac": 0.765625, "sample": [57.82672119140625, 26.106964111328125, 42.132904052734375, -2.298311233520508, 1.0848236083984375, -46.15864562988281, 24.412322998046875, 4.802070617675781, 93.446533203125, -50.171875, 16.0125732421875, 9.054588317871094, 30.48516845703125, 29.04621124267578, 50.59348678588867, 122.88525390625, 6.385503768920898, 51.28765869140625, 29.738052368164062, 74.02545166015625, -32.5762939453125, 59.79644012451172, 60.1312255859375, 47.834197998046875, 42.2884521484375, -2.005706787109375, -2.4085845947265625, -36.60345458984375, 21.014358520507812, -36.7286376953125, -4.208320617675781, 1.01910400390625, 22.99267578125, 52.41046142578125, -3.78863525390625, 55.69007873535156, 10.0714111328125, -4.1873321533203125, 78.559814453125, 40.97381591796875, 2.784088134765625, 10.202484130859375, 13.791778564453125, 19.6490478515625, 8.763427734375, 71.99671936035156, 54.6781005859375, 55.2889404296875, -14.5128173828125, 86.3154296875, 34.14149475097656, 11.173660278320312, 2.55999755859375, 10.196136474609375, -18.0963134765625, -22.049072265625, 20.0623779296875, -1.416778564453125, -1.0457916259765625, 73.43380737304688, 18.729188919067383, 70.4906005859375, 9.212722778320312, 41.14775848388672, 51.91046142578125, 9.288993835449219, 3.0136947631835938, 4.274431228637695, 0.13519287109375, 6.701812744140625, 59.66748046875, -2.4778823852539062, 15.701522827148438, 65.7037353515625, 51.4302978515625, -38.526214599609375, 0.4218292236328125, 65.31268310546875, 0.0, -96.70848083496094, 17.96265411376953, 44.09136962890625, 80.21853637695312, 4.72320556640625, 25.052001953125, -1.1627197265625, 25.777862548828125, -16.411026000976562, -35.9742431640625, 84.95315551757812, 43.87408447265625, 4.2666473388671875, 55.185577392578125, 53.186126708984375, 48.98674011230469, 32.752349853515625, -17.3035888671875, 11.601276397705078, -26.401992797851562, 7.817291259765625, 51.4803466796875, 55.94387435913086, 9.751800537109375, 32.868377685546875, 4.02972412109375, -76.1148681640625, 19.851409912109375, 22.343215942382812, -7.59539794921875, 24.961334228515625, 33.85410690307617, 20.932052612304688, 27.141403198242188, 34.978355407714844, 78.314453125, 8.0225830078125, 44.567989349365234, 87.04315185546875, 60.93304443359375, 11.330474853515625, 13.080810546875, 17.245574951171875, 22.737274169921875, -72.39445495605469, -57.5863037109375, 0.0, 4.609039306640625, 90.06954956054688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000175.npy"}
{"epoch": 0.36649214659685864, "step": 176, "batch_size": 128, "mean": 15.690946578979492, "std": 38.62723922729492, "min": -114.03622436523438, "p10": -30.343058776855468, "median": 10.708789825439453, "p90": 62.21864700317383, "max": 108.88992309570312, "pos_frac": 0.734375, "sample": [-30.011688232421875, 13.069374084472656, 3.20355224609375, -23.014785766601562, 20.17584228515625, 51.222389221191406, -54.56208801269531, -57.86920166015625, 14.411598205566406, 18.12359619140625, -85.16168212890625, 4.3184967041015625, 30.455265045166016, 62.315338134765625, 10.35107421875, 44.656829833984375, 45.803802490234375, 4.959278106689453, 1.99981689453125, -9.010284423828125, -41.514495849609375, 4.14501953125, 6.990839004516602, -74.39599609375, 15.868438720703125, 77.07577514648438, 32.19622039794922, 100.40237426757812, 39.99774169921875, 2.4107666015625, 95.38186645507812, -32.96430969238281, -4.563789367675781, 19.000717163085938, 15.494918823242188, 3.063323974609375, -4.316619873046875, 25.329971313476562, 14.656082153320312, 8.305038452148438, -51.625823974609375, 11.066505432128906, 47.630889892578125, -4.29315185546875, 7.077392578125, 62.177207946777344, 7.075572967529297, 2.9494476318359375, 41.12657165527344, 80.56109619140625, -34.86798858642578, 54.56820297241211, 48.156829833984375, -23.938690185546875, 14.029541015625, 51.01141357421875, 1.43756103515625, 47.99480438232422, 39.812408447265625, 4.810283660888672, 4.4647216796875, 1.717742919921875, -5.715087890625, -18.301231384277344, 47.695037841796875, 62.1373291015625, 38.05802917480469, -12.48394775390625, 69.67718505859375, 4.21539306640625, -12.670242309570312, 75.94281005859375, 18.519027709960938, 8.7359619140625, 108.88992309570312, 60.034576416015625, 1.4635162353515625, -15.141754150390625, 16.073760986328125, 9.97869873046875, 36.19872283935547, 34.872161865234375, 54.30335998535156, 47.188819885253906, 45.09008026123047, 58.10296630859375, -5.629974365234375, 17.18600082397461, 10.349853515625, 84.15594482421875, 30.99212646484375, -0.5108184814453125, -31.116256713867188, 35.0084228515625, -16.078704833984375, 6.204624176025391, 46.610809326171875, -69.916748046875, -22.859619140625, 32.753448486328125, -23.851791381835938, 55.191497802734375, 2.9678726196289062, -17.423492431640625, 7.133934020996094, -69.66583251953125, 26.0450439453125, 15.6871337890625, -17.668731689453125, 3.0431785583496094, 66.25981140136719, -7.428947448730469, 83.56930541992188, 57.256591796875, 87.11395263671875, 38.95288848876953, 2.91705322265625, 22.7899169921875, -31.157135009765625, -114.03622436523438, 67.21856689453125, 1.594207763671875, 0.6712379455566406, 34.555747985839844, 38.1939697265625, 8.842071533203125, 42.18242645263672, -3.4701385498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000176.npy"}
{"epoch": 0.36858638743455496, "step": 177, "batch_size": 128, "mean": 15.236845016479492, "std": 42.240882873535156, "min": -111.21678161621094, "p10": -35.2026252746582, "median": 9.939956665039062, "p90": 70.88319091796875, "max": 106.9796142578125, "pos_frac": 0.6796875, "sample": [51.533355712890625, -0.7276115417480469, 1.654266357421875, 86.32862854003906, -18.86602783203125, 70.00082397460938, 12.667781829833984, -27.07421875, 33.131103515625, 7.350898742675781, 68.83599853515625, -47.502769470214844, -91.780029296875, -12.46626091003418, 3.6049880981445312, -21.603164672851562, 45.88920974731445, 4.54608154296875, 52.46937561035156, 8.732177734375, 48.311981201171875, 29.850418090820312, 61.07684326171875, 65.390625, -25.35919189453125, 30.336334228515625, -37.55987548828125, 85.98336791992188, 53.896697998046875, 98.01617431640625, 8.016433715820312, 30.943923950195312, -68.54559326171875, 92.37252807617188, 1.307037353515625, 29.181411743164062, 73.9683837890625, 2.8554458618164062, -9.240608215332031, 6.42645263671875, 42.890838623046875, -23.054489135742188, -3.9041385650634766, 2.08209228515625, -7.21771240234375, -34.19237518310547, 4.602809906005859, 15.097732543945312, 51.616172790527344, 52.8187255859375, -11.48834228515625, -3.3910369873046875, 91.96926879882812, 26.420974731445312, -4.7150421142578125, 63.2515869140625, 23.187286376953125, -4.7789306640625, 81.00010681152344, 6.230865478515625, -47.783660888671875, 81.16021728515625, 0.35382080078125, -32.038818359375, 75.45379638671875, -3.9587554931640625, -47.496368408203125, -10.303670883178711, 5.967327117919922, 68.01156616210938, -15.23492431640625, 97.023193359375, -6.4185943603515625, 7.612918853759766, -5.072074890136719, 52.31355285644531, 50.70166015625, -88.84860229492188, 23.141754150390625, 8.29010009765625, 21.6656494140625, -57.325714111328125, -51.95215606689453, 20.980712890625, -75.06689453125, 9.84765625, 36.74732971191406, 0.4249267578125, -20.48211669921875, 10.54376220703125, 27.725982666015625, 63.77618408203125, 42.45433807373047, 36.42351531982422, -12.8831787109375, -19.185394287109375, 12.707595825195312, 16.103759765625, 59.97174072265625, 10.032257080078125, -25.119537353515625, -3.4610443115234375, 14.744171142578125, 43.05255126953125, 8.020278930664062, -82.1600341796875, 7.3464202880859375, 0.246551513671875, -43.790924072265625, -111.21678161621094, 29.869369506835938, 47.28228759765625, 72.94204711914062, 14.494384765625, 16.3642578125, 76.45263671875, 38.1744384765625, 41.86116027832031, 106.9796142578125, 11.616958618164062, 52.90081787109375, 1.0591201782226562, -1.49761962890625, -5.333290100097656, 18.865982055664062, 5.6876678466796875, 35.124420166015625, 62.046226501464844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000177.npy"}
{"epoch": 0.3706806282722513, "step": 178, "batch_size": 128, "mean": 24.93454360961914, "std": 39.89390182495117, "min": -93.98300170898438, "p10": -15.476158142089844, "median": 20.697731018066406, "p90": 81.29218749999998, "max": 146.45257568359375, "pos_frac": 0.734375, "sample": [62.63929748535156, 16.46405792236328, 10.343711853027344, -8.18292236328125, -7.663948059082031, 7.647764205932617, 41.409271240234375, 52.48558044433594, 33.171077728271484, 40.87025833129883, 146.45257568359375, 61.393890380859375, 20.953994750976562, 27.120819091796875, -12.16644287109375, 86.15219116210938, 0.0, 62.231719970703125, 60.639373779296875, 37.867698669433594, 12.77435302734375, 93.93417358398438, 8.597366333007812, 78.40142822265625, 54.785980224609375, -15.295089721679688, -8.338958740234375, 98.80792236328125, -25.52484130859375, 44.878814697265625, 5.093145370483398, -5.989288330078125, -0.23577880859375, 38.97877502441406, -66.25320434570312, 14.69140625, 27.76239013671875, 71.80885314941406, -31.166259765625, 57.700706481933594, 80.0693359375, 38.665679931640625, 44.05570983886719, 43.87518310546875, 79.56427001953125, 58.555572509765625, 10.687820434570312, -49.9674072265625, -8.659515380859375, 92.52761840820312, 61.706878662109375, 93.92044067382812, -56.377838134765625, -93.98300170898438, -6.23333740234375, 11.953712463378906, 15.72381591796875, 37.544349670410156, -36.01701354980469, 24.242950439453125, 0.6724586486816406, 101.93997192382812, 9.37176513671875, 40.661590576171875, 66.75872802734375, 28.270408630371094, 28.43060302734375, 32.18133544921875, 40.72747802734375, 3.4634246826171875, 96.80947875976562, 24.30554962158203, -13.225383758544922, -4.131317138671875, 13.201925277709961, 84.1455078125, -48.6104736328125, 16.378387451171875, -6.621395111083984, 2.434906005859375, 44.864837646484375, 23.435089111328125, 53.67555236816406, -15.898651123046875, -44.55841064453125, -4.151222229003906, 85.65737915039062, -2.8270263671875, 26.778411865234375, 33.110321044921875, 16.283370971679688, 4.5623931884765625, 0.200408935546875, 19.734283447265625, 16.06243133544922, 98.05728149414062, 13.398162841796875, 10.604026794433594, 0.7945404052734375, 67.9130859375, 10.170738220214844, -13.268041610717773, -22.3504638671875, 96.842041015625, 22.249862670898438, -14.74237060546875, 4.894996643066406, 68.83065795898438, 70.41416931152344, -15.002151489257812, -4.039436340332031, 102.96990966796875, -35.180694580078125, 5.09332275390625, 43.19769287109375, 20.44146728515625, -23.94403839111328, 19.05144500732422, 50.349510192871094, 3.0467529296875, 28.80792236328125, 33.04791259765625, 24.36273193359375, -6.5037841796875, -5.806610107421875, 27.61614990234375, 56.388057708740234, 42.72560119628906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000178.npy"}
{"epoch": 0.37277486910994767, "step": 179, "batch_size": 128, "mean": 18.463321685791016, "std": 38.249656677246094, "min": -66.40380859375, "p10": -32.887641906738274, "median": 14.794933319091797, "p90": 73.34733428955077, "max": 97.970703125, "pos_frac": 0.6875, "sample": [44.18104553222656, 1.2002544403076172, 64.85577392578125, -8.8096923828125, 13.441650390625, 68.14692687988281, 8.278877258300781, 10.79885482788086, -58.76202392578125, 0.84539794921875, 13.851776123046875, -2.144805908203125, 23.986419677734375, 2.350311279296875, 74.17098999023438, -38.159645080566406, 46.39312744140625, 29.1871337890625, 6.45965576171875, -5.102323532104492, 25.285888671875, 46.36876678466797, 2.192718505859375, 28.915451049804688, 21.14111328125, -49.13462448120117, -43.37860107421875, -6.994815826416016, -5.486541748046875, -56.894264221191406, 51.92076110839844, 6.5519866943359375, -50.0439453125, 29.264129638671875, -10.799530029296875, 28.936630249023438, 49.629791259765625, 77.23521423339844, 22.369781494140625, -6.91015625, -30.3306884765625, -53.0081787109375, -6.347564697265625, 24.0679931640625, -19.284088134765625, 36.38526916503906, 6.25738525390625, -10.28736686706543, 69.2591552734375, -43.398284912109375, -30.13189697265625, -28.828933715820312, 29.99444580078125, 19.74896240234375, 4.5182342529296875, 81.94772338867188, 72.98600006103516, 65.6263656616211, 17.6082763671875, -4.5699462890625, -66.40380859375, -11.498672485351562, 95.4183349609375, 14.67340087890625, -13.23270034790039, 36.157989501953125, 72.99433898925781, -9.16876220703125, -1.1416778564453125, 12.09967041015625, 15.324871063232422, 42.973724365234375, 47.45344543457031, 34.651702880859375, 1.1118011474609375, 7.2451171875, 80.42964172363281, 11.085235595703125, 9.348918914794922, 55.871421813964844, 11.645172119140625, 8.088668823242188, 93.6907958984375, -16.59638214111328, 14.916465759277344, -40.84722900390625, 83.23758697509766, 78.0465087890625, 0.9414215087890625, -43.85597229003906, 51.95159912109375, 12.657562255859375, 83.4635009765625, 35.409576416015625, 33.75694274902344, 29.030120849609375, 46.149993896484375, 43.12110900878906, 16.04718780517578, 46.01554870605469, 86.96829223632812, 41.86100769042969, 69.32965087890625, 20.24921417236328, -38.73175048828125, 33.210296630859375, -56.98054504394531, 4.7844696044921875, 29.83667755126953, 65.3722152709961, 20.62030792236328, 44.23126220703125, -30.628211975097656, 97.970703125, -16.767852783203125, -19.92249298095703, 44.142704010009766, 51.28572082519531, -2.357421875, -10.617996215820312, 82.3779296875, 55.76197814941406, -2.140960693359375, 13.097393035888672, -9.450729370117188, -21.159515380859375, 75.53097534179688, 55.615455627441406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000179.npy"}
{"epoch": 0.374869109947644, "step": 180, "batch_size": 128, "mean": 23.881372451782227, "std": 37.832027435302734, "min": -66.25631713867188, "p10": -18.29076690673828, "median": 17.10162353515625, "p90": 77.14398803710937, "max": 104.70620727539062, "pos_frac": 0.7421875, "sample": [-1.0628204345703125, -4.252780914306641, -40.333709716796875, -11.109878540039062, 1.2640380859375, 36.02606201171875, 68.65255737304688, 7.9825439453125, -19.265838623046875, 76.49980163574219, -2.229736328125, 58.358642578125, 101.20932006835938, -62.053009033203125, 8.241561889648438, 69.39724731445312, -7.352752685546875, 26.359130859375, 17.816368103027344, -66.25631713867188, 56.016876220703125, 104.70620727539062, 42.672821044921875, 37.24755859375, 22.847747802734375, 52.14494323730469, 19.451141357421875, 67.27635192871094, 97.23269653320312, -6.156696319580078, -57.65742492675781, 4.8736724853515625, 78.982666015625, 100.97161865234375, 78.4901123046875, 12.876251220703125, 16.65411376953125, 10.192718505859375, 9.913230895996094, 15.595321655273438, 69.8721694946289, -6.5274505615234375, 50.02781677246094, 32.996803283691406, 91.34939575195312, 17.042022705078125, -32.730499267578125, -2.2790908813476562, 8.08587646484375, 0.62139892578125, -38.355735778808594, -5.26116943359375, 5.3739471435546875, 30.73907470703125, 83.0850830078125, 1.6707763671875, 17.626041412353516, 54.09761047363281, 11.1494140625, 12.447265625, 93.38177490234375, 46.139076232910156, -7.91302490234375, 96.24932861328125, 53.347206115722656, -16.88555908203125, 43.74330139160156, 34.408905029296875, -43.400177001953125, 19.3621826171875, 34.00067138671875, 3.3443984985351562, 19.410301208496094, 76.56707763671875, 74.31280517578125, 74.51596069335938, -0.07546234130859375, 0.23329925537109375, 74.09072875976562, 47.217254638671875, 100.69345092773438, 16.658226013183594, -14.841644287109375, 56.12506103515625, 59.02157974243164, 49.3259391784668, 35.241058349609375, -39.50074005126953, 2.70806884765625, 34.439453125, -0.2872161865234375, 17.161224365234375, 4.990224838256836, 32.49969482421875, 56.55787658691406, 12.648162841796875, 26.752593994140625, 29.758453369140625, 1.1628379821777344, -5.64495849609375, 5.0979461669921875, 21.107147216796875, 68.92861938476562, 0.616302490234375, 8.868444442749023, -21.753311157226562, 4.392192840576172, -20.5264892578125, 79.48599243164062, 7.0880889892578125, -15.958648681640625, 43.11581802368164, -6.598178863525391, -24.27716064453125, 52.92719268798828, 33.586883544921875, 45.04774475097656, -24.412384033203125, -12.818817138671875, 55.16007995605469, 1.001108169555664, 37.168853759765625, 3.8705215454101562, 3.706310272216797, -12.724212646484375, 92.42343139648438, -17.872879028320312, 29.392578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000180.npy"}
{"epoch": 0.3769633507853403, "step": 181, "batch_size": 128, "mean": 22.855342864990234, "std": 40.870330810546875, "min": -81.1500244140625, "p10": -30.297531127929688, "median": 20.654098510742188, "p90": 76.58374938964843, "max": 133.31170654296875, "pos_frac": 0.7265625, "sample": [0.8652992248535156, 5.805389404296875, 2.9768905639648438, 34.37432861328125, 0.0, 44.77734375, 77.70614624023438, 32.456787109375, 8.087608337402344, 37.968788146972656, 62.83026123046875, -11.725894927978516, 52.673065185546875, -8.830902099609375, 88.8798828125, 32.234649658203125, 0.0, 73.64228820800781, 44.556636810302734, -33.94578552246094, 52.24595642089844, 6.563262939453125, 78.030029296875, 13.652740478515625, -73.55467224121094, 95.19749450683594, -30.7900390625, 37.526065826416016, 6.844108581542969, 28.044036865234375, 65.04063415527344, 4.001380920410156, 13.389644622802734, 65.37350463867188, 28.929931640625, 47.043060302734375, 60.084228515625, 18.660720825195312, 20.11029052734375, 33.27690124511719, 18.767715454101562, 39.686248779296875, -27.04656982421875, -5.459959030151367, 10.940216064453125, 14.455924987792969, 9.575057983398438, 8.066497802734375, 98.45819854736328, -41.75846862792969, 33.31695556640625, -6.176338195800781, 54.632347106933594, 64.945556640625, -20.454238891601562, -64.94900512695312, -2.5812644958496094, 15.286346435546875, -2.84722900390625, -3.716583251953125, 38.4091796875, -4.442314147949219, -23.66058349609375, 81.42837524414062, 47.722442626953125, 45.404048919677734, 8.309844970703125, 66.65678405761719, 6.482208251953125, 35.526214599609375, 21.197906494140625, 39.93467712402344, 36.121177673339844, 84.48976135253906, 111.75927734375, 133.31170654296875, -54.31219482421875, 13.215721130371094, -30.086456298828125, 67.08740234375, -34.96044921875, 29.3499755859375, -5.350015640258789, -72.15440368652344, -9.434768676757812, 71.28575134277344, 21.64398193359375, -11.022575378417969, 21.803009033203125, 14.349365234375, -44.66893005371094, 36.9217529296875, 76.10272216796875, -29.35223388671875, 12.8782958984375, 28.281654357910156, 11.083770751953125, 45.750328063964844, 21.44268798828125, 88.87860107421875, 81.496826171875, -33.0732421875, 46.91192626953125, 13.057220458984375, -29.68133544921875, -12.68365478515625, 12.577789306640625, 88.85296630859375, 2.12567138671875, -81.1500244140625, 60.80006408691406, 28.8128662109375, 72.08241271972656, -17.3280029296875, 40.24803161621094, 84.12757873535156, -48.02992248535156, 17.054779052734375, 14.918128967285156, -20.918426513671875, 14.1634521484375, 31.167739868164062, 64.12652587890625, -45.27546691894531, 68.22637939453125, 55.471405029296875, 49.308860778808594, 62.566070556640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000181.npy"}
{"epoch": 0.37905759162303665, "step": 182, "batch_size": 128, "mean": 28.771465301513672, "std": 34.87516784667969, "min": -70.05517578125, "p10": -10.336610794067383, "median": 27.23236083984375, "p90": 74.21409149169922, "max": 107.32656860351562, "pos_frac": 0.8046875, "sample": [47.10693359375, 64.70541381835938, 10.26385498046875, -2.1672611236572266, 22.698593139648438, 17.349288940429688, 71.93696594238281, 33.708465576171875, 24.073150634765625, 5.84588623046875, 66.0841064453125, 27.344482421875, 27.1202392578125, 54.62896728515625, 54.84075927734375, 2.766815185546875, 42.47221374511719, 55.848175048828125, 74.56282043457031, 49.46794891357422, 46.58094787597656, 22.226543426513672, 49.90717315673828, 86.8746337890625, -18.68994140625, 78.35543823242188, -7.099769592285156, 49.8778076171875, 97.8096923828125, 35.650054931640625, 54.82548522949219, 40.406036376953125, 40.85711669921875, 36.78358459472656, 26.656509399414062, -4.755565643310547, -10.897293090820312, 12.0977783203125, 2.050689697265625, 9.649703979492188, -55.63690185546875, 25.185287475585938, -70.05517578125, 45.359046936035156, 66.55868530273438, -10.248748779296875, 86.82965087890625, -51.73924255371094, 65.75460815429688, 107.32656860351562, 31.5006103515625, 64.08151245117188, 67.82821655273438, 78.04055786132812, 46.265716552734375, -6.022857666015625, 21.467039108276367, 7.304046630859375, 6.66180419921875, 41.09130859375, 35.0045166015625, 72.240966796875, 27.864349365234375, 18.8262939453125, 40.661529541015625, 13.098358154296875, 37.779624938964844, 45.224884033203125, 99.35641479492188, 26.827075958251953, -37.16566467285156, 9.524084091186523, 50.171478271484375, 15.838088989257812, 2.378631591796875, 30.21856689453125, 32.39178466796875, 51.487159729003906, 7.252250671386719, 8.4765625, 9.307418823242188, -4.0763397216796875, 19.559967041015625, 58.92332458496094, -13.3883056640625, -0.6615371704101562, 40.65956115722656, 7.5793609619140625, 14.90850830078125, -5.4514617919921875, 81.99432373046875, 54.9022216796875, 82.4022216796875, 74.06463623046875, -44.38774108886719, -10.541622161865234, 82.7977294921875, 49.24900817871094, 26.37890625, 41.707557678222656, -36.37428283691406, -0.50665283203125, 16.020835876464844, 13.067659378051758, -3.16064453125, 49.87448501586914, 49.494537353515625, 26.72747802734375, 79.46731567382812, 7.361083984375, -3.34600830078125, 59.949249267578125, 70.7220458984375, 2.102130889892578, 1.6663131713867188, 7.59942626953125, 23.76830291748047, 66.32806396484375, 12.349449157714844, -38.65550231933594, -22.312713623046875, 69.967041015625, 42.0084228515625, 0.1026611328125, 81.46206665039062, -9.049209594726562, 32.82269287109375, -37.46949768066406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000182.npy"}
{"epoch": 0.381151832460733, "step": 183, "batch_size": 128, "mean": 24.2431583404541, "std": 39.822357177734375, "min": -68.13087463378906, "p10": -26.315582275390625, "median": 21.281707763671875, "p90": 74.42483978271484, "max": 112.64447021484375, "pos_frac": 0.7265625, "sample": [-26.19610595703125, 43.54176330566406, -29.399200439453125, 57.56964874267578, -50.7120361328125, 75.3203125, -35.416961669921875, 12.285873413085938, 62.59535217285156, 18.550140380859375, 3.945556640625, 22.425209045410156, 51.740234375, 2.492401123046875, 8.05157470703125, -62.910003662109375, 1.9084739685058594, 69.291015625, 18.264556884765625, -0.2781829833984375, 22.989349365234375, 8.605400085449219, 58.68266296386719, -16.695449829101562, 12.526939392089844, -26.5943603515625, 72.94120788574219, 41.175811767578125, 9.913864135742188, 3.07708740234375, -7.010284423828125, -57.456817626953125, 16.346237182617188, 18.343536376953125, -0.01926422119140625, 61.9410400390625, 14.275848388671875, 76.918701171875, 31.26513671875, -36.700897216796875, 95.92706298828125, -6.821769714355469, 48.96034240722656, -20.6566162109375, 73.54510498046875, 31.376922607421875, 61.41835021972656, -12.061767578125, 85.94525146484375, 46.1212043762207, 8.39215087890625, 68.57763671875, 60.28741455078125, -49.5887451171875, 0.0, 0.19129180908203125, -2.6873779296875, 8.711685180664062, -19.30072021484375, 59.27117919921875, 12.555137634277344, 76.15737915039062, 0.567840576171875, 91.2679443359375, 74.07540893554688, -39.265380859375, 33.3037109375, 90.1162109375, 59.61860656738281, -18.690261840820312, 44.658836364746094, 100.51657104492188, 112.64447021484375, 52.01118469238281, -5.2813873291015625, 28.43939971923828, -58.59587097167969, 27.307952880859375, -3.485595703125, 32.9053955078125, 33.70123291015625, 30.236785888671875, 15.118400573730469, 21.931215286254883, -3.7557754516601562, 6.921417236328125, 43.4488525390625, 55.832916259765625, 12.846649169921875, 13.484039306640625, -8.362281799316406, 35.831939697265625, 4.57928466796875, 3.717578887939453, -22.661468505859375, 74.16885375976562, 75.02214050292969, 51.34747314453125, 42.81329345703125, -0.30322265625, -68.13087463378906, 3.734039306640625, -6.970184326171875, 21.052093505859375, 2.221790313720703, 59.10261535644531, 0.18245315551757812, 72.98110961914062, -5.4622344970703125, 92.91165161132812, 38.647674560546875, 90.82321166992188, 51.476348876953125, 57.35302734375, 44.359222412109375, -53.187896728515625, 21.511322021484375, 47.52520751953125, -47.735504150390625, 106.54864501953125, -7.171051025390625, 58.74982452392578, 69.7305908203125, 33.76995849609375, -6.632476806640625, 35.11126708984375, 51.43095397949219, 57.24066162109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000183.npy"}
{"epoch": 0.3832460732984293, "step": 184, "batch_size": 128, "mean": 36.264495849609375, "std": 44.371124267578125, "min": -83.6290283203125, "p10": -24.5119384765625, "median": 40.47712707519531, "p90": 87.43255920410157, "max": 120.72824096679688, "pos_frac": 0.7890625, "sample": [63.665679931640625, 120.72824096679688, 84.20550537109375, 6.735664367675781, 20.4107666015625, -13.502792358398438, 39.29632568359375, -7.681413650512695, 72.10595703125, 48.537567138671875, 25.776931762695312, 9.496826171875, -58.79339599609375, 24.230255126953125, -34.260589599609375, -1.563812255859375, 101.747802734375, 78.65023803710938, 49.720680236816406, 56.666404724121094, 64.37576293945312, 89.6041259765625, -83.6290283203125, 86.16522216796875, -6.589935302734375, 87.57113647460938, 73.65057373046875, -4.1418609619140625, -49.32758331298828, 102.23867797851562, 34.42582702636719, 94.35400390625, 55.2254638671875, -37.547630310058594, 35.4908447265625, 85.94970703125, 6.57733154296875, 85.62086486816406, 61.981689453125, 105.26473999023438, 27.238433837890625, 97.73614501953125, -57.413055419921875, 49.29856491088867, 44.4801025390625, 109.40701293945312, -51.66778564453125, 84.9976806640625, -25.236419677734375, 44.55712890625, 26.228076934814453, -15.98699951171875, 44.065948486328125, 6.60565185546875, -24.201446533203125, 37.3468017578125, 70.37982177734375, 42.511138916015625, -8.743011474609375, 107.45147705078125, 8.471221923828125, 5.436256408691406, 45.197845458984375, -56.62324523925781, 72.572998046875, 53.565826416015625, 71.6796875, 85.62211608886719, 25.0804443359375, 71.65451049804688, 12.075439453125, 30.88201904296875, 70.47140502929688, 36.36625671386719, 43.2197265625, 22.13341522216797, 26.096038818359375, 67.34197998046875, 36.14762878417969, 81.01950073242188, 85.4310302734375, 63.358123779296875, 1.32415771484375, 78.96868896484375, 15.515609741210938, -46.05162048339844, 6.32659912109375, 82.43168640136719, 4.453948974609375, 15.58251953125, 15.7689208984375, 87.3731689453125, 41.657928466796875, 24.169334411621094, 68.23443603515625, -7.563323974609375, -27.482454299926758, 21.161758422851562, 1.9004631042480469, 85.9639892578125, 57.616355895996094, 60.9879150390625, 49.9232177734375, 75.62274169921875, 0.7482433319091797, 3.758241653442383, 70.56631469726562, 45.480438232421875, -14.373741149902344, 93.86076354980469, 28.314319610595703, -44.72698974609375, 65.70751953125, 14.927459716796875, 73.5709228515625, -8.08837890625, 95.46925354003906, 8.071578979492188, -8.506622314453125, 68.42581176757812, -8.3519287109375, -29.77020263671875, 83.80569458007812, 16.56207275390625, 108.396240234375, 80.31439208984375, -10.633468627929688, 54.75331497192383], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000184.npy"}
{"epoch": 0.38534031413612563, "step": 185, "batch_size": 128, "mean": 23.638671875, "std": 42.59965896606445, "min": -73.45697021484375, "p10": -27.120741271972655, "median": 21.356582641601562, "p90": 82.11810150146484, "max": 116.641357421875, "pos_frac": 0.6796875, "sample": [31.80970001220703, -60.841278076171875, 59.31475830078125, -17.1671142578125, -27.817184448242188, 80.3077392578125, 30.74072265625, 83.18400573730469, 85.32437133789062, -7.992279052734375, 21.55438232421875, 45.981956481933594, 75.42276000976562, 57.04323959350586, 32.341827392578125, 98.52975463867188, 48.76611328125, -14.57513427734375, 10.266654968261719, 70.66729736328125, 40.44508361816406, 7.3978271484375, 53.40445327758789, 35.841888427734375, 29.5755615234375, 19.70611572265625, 81.66128540039062, 8.752700805664062, 1.5624313354492188, 88.53668212890625, -73.45697021484375, 56.523712158203125, 1.7526893615722656, -9.01861572265625, 24.20581817626953, 70.13534545898438, 26.723739624023438, 49.68169403076172, 35.177154541015625, 111.94747924804688, -11.8631591796875, 32.218017578125, -5.90673828125, 24.533294677734375, 6.159881591796875, 84.32984924316406, -8.201934814453125, 38.155113220214844, 28.155029296875, 63.4549560546875, -49.4427490234375, -25.816238403320312, -47.30426025390625, -16.979217529296875, 10.24212646484375, 0.3488426208496094, 12.75250244140625, 0.0, 12.171188354492188, 64.97270965576172, -43.15974426269531, -26.822265625, 11.81585693359375, 1.2799301147460938, 34.360870361328125, 21.158782958984375, 7.256156921386719, 16.5277099609375, 41.2305908203125, -5.120306015014648, -12.045150756835938, 54.307525634765625, 57.67724609375, 76.81231689453125, -16.25127410888672, -1.621185302734375, 8.501861572265625, 16.584259033203125, 27.998279571533203, 70.93549346923828, 116.641357421875, -13.415359497070312, 11.7406005859375, 0.0, -22.739089965820312, 64.2492904663086, 70.71121215820312, -22.0198974609375, 89.99642944335938, -17.576507568359375, 50.31132507324219, 52.1458740234375, -16.574996948242188, -5.0857696533203125, 51.54193115234375, 92.46090698242188, -50.046630859375, 69.67245483398438, 16.550888061523438, 14.59454345703125, -20.89214324951172, 44.39088439941406, 52.586517333984375, -39.0006103515625, 94.81399536132812, -57.11091613769531, -13.914146423339844, 17.620193481445312, 111.99745178222656, 112.50537109375, 61.50885009765625, 34.837890625, 13.562370300292969, -2.5311279296875, 78.286376953125, -68.4498291015625, 45.889312744140625, 92.1839599609375, 56.9212646484375, -30.32025146484375, 31.08202362060547, -18.43646240234375, -10.877975463867188, -18.46576690673828, -42.063941955566406, -34.63269805908203, 37.771060943603516, 26.505218505859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000185.npy"}
{"epoch": 0.387434554973822, "step": 186, "batch_size": 128, "mean": 23.03746223449707, "std": 39.170352935791016, "min": -81.02630615234375, "p10": -19.154293823242185, "median": 19.477981567382812, "p90": 75.74997253417968, "max": 129.8616943359375, "pos_frac": 0.703125, "sample": [27.204360961914062, -1.8156814575195312, 6.782196044921875, 50.454620361328125, 19.3013916015625, 78.6767578125, 27.63037109375, -3.181243896484375, 14.6761474609375, 14.777175903320312, -3.0140762329101562, 9.62261962890625, 10.2669677734375, 52.296142578125, -18.7835693359375, 65.15069580078125, 60.338768005371094, 2.8721256256103516, 72.99655151367188, 12.423507690429688, 10.8857421875, -8.903282165527344, -20.019317626953125, 4.578948974609375, 0.9872589111328125, 57.43006896972656, 41.62571716308594, 27.737884521484375, 3.028656005859375, -0.57275390625, 8.406742095947266, 37.8092041015625, 95.92507934570312, 19.654571533203125, -7.1172027587890625, -50.57366943359375, 34.066375732421875, 92.38644409179688, 30.59063720703125, 31.978271484375, 3.2970199584960938, 69.75933837890625, 61.83317565917969, 52.72348403930664, 60.546268463134766, 42.2366943359375, -53.10760498046875, 4.961881637573242, 53.71453857421875, 30.415863037109375, 27.479827880859375, 42.893341064453125, 79.026611328125, -5.58709716796875, -13.216552734375, -15.2841796875, 66.77197265625, 25.18365478515625, -0.23028564453125, 5.59320068359375, -9.109542846679688, -10.497833251953125, 87.94915771484375, -34.452545166015625, 74.49563598632812, 1.1668701171875, 33.745452880859375, -3.072265625, 2.7159423828125, 79.38999938964844, -7.546302795410156, 20.633056640625, 101.3079833984375, 27.782928466796875, -28.27569580078125, 29.629913330078125, 62.23774719238281, -75.762451171875, -0.7952632904052734, 83.44586181640625, 59.4610595703125, 5.6358642578125, 24.8704833984375, 124.90667724609375, 63.74224853515625, -63.82310485839844, 24.244720458984375, -21.358726501464844, 5.8513641357421875, -43.443603515625, 5.96063232421875, 20.546844482421875, 32.577552795410156, -8.795440673828125, 98.93157958984375, 96.05081176757812, 8.149436950683594, 12.931747436523438, -81.02630615234375, 33.903526306152344, 38.09328842163086, 70.96073913574219, 9.911170959472656, -7.381317138671875, 43.311798095703125, 62.61651611328125, 25.40081787109375, -1.1306571960449219, -6.86334228515625, 49.888797760009766, 52.546417236328125, -14.128776550292969, 32.79742431640625, 3.8895492553710938, 129.8616943359375, 0.0, 88.23712158203125, 59.3416862487793, -21.204940795898438, 31.1119384765625, -22.505680084228516, -33.20591354370117, -0.323486328125, 0.32837677001953125, 32.482879638671875, -11.623687744140625, 57.994140625, -3.5096435546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000186.npy"}
{"epoch": 0.38952879581151834, "step": 187, "batch_size": 128, "mean": 23.179241180419922, "std": 45.95780944824219, "min": -88.742919921875, "p10": -36.0076416015625, "median": 16.619029998779297, "p90": 90.447705078125, "max": 120.45062255859375, "pos_frac": 0.6875, "sample": [49.3070068359375, 74.30023193359375, 17.4622802734375, 38.5789794921875, -55.55322265625, 9.805267333984375, 95.49563598632812, 0.6320896148681641, -39.934417724609375, 120.45062255859375, 81.8665771484375, 1.9760589599609375, -34.324737548828125, -1.1283416748046875, 75.87345886230469, 115.0072021484375, 50.33317565917969, 46.479942321777344, 90.03445434570312, 13.158111572265625, 65.929443359375, -33.97144317626953, 40.456268310546875, -5.1142425537109375, -28.4862060546875, -16.799041748046875, 70.92680358886719, -50.54766845703125, -49.919158935546875, -66.834228515625, -10.495712280273438, 80.81828308105469, 41.94329833984375, 0.4583587646484375, -25.086631774902344, 45.278778076171875, -26.31109619140625, -12.88958740234375, 43.89268493652344, 41.304443359375, 14.32122802734375, -7.691925048828125, 104.3876953125, 14.92694091796875, -13.8900146484375, 48.055511474609375, 91.41195678710938, 5.2777099609375, -19.63140869140625, 48.23876953125, 0.129119873046875, 35.996253967285156, 29.15826416015625, 67.03079223632812, -88.742919921875, -10.535736083984375, 61.830902099609375, -40.773529052734375, 60.45999526977539, 92.07209777832031, 60.17366027832031, -7.900382995605469, 40.201385498046875, -44.007904052734375, 110.54644775390625, 73.23268127441406, -60.26203155517578, -67.3975830078125, 14.6365966796875, 29.55645751953125, 91.70693969726562, -46.4135627746582, -6.217624664306641, -28.322784423828125, -9.133880615234375, -10.79827880859375, 70.11763000488281, 25.639869689941406, 81.28121948242188, 16.30682373046875, 0.635406494140625, 11.421279907226562, -47.24507141113281, 2.998931884765625, 7.0322723388671875, -5.628997802734375, 20.326446533203125, 9.618453979492188, -62.462646484375, 63.28692626953125, 57.63618469238281, 75.87399291992188, 73.78044128417969, 23.537765502929688, 30.097091674804688, 94.3563232421875, -0.048736572265625, 64.17160034179688, 61.66148376464844, 16.931236267089844, -26.23944091796875, 29.787139892578125, 106.1427001953125, 10.43353271484375, 0.8729820251464844, 95.50625610351562, 0.76068115234375, 24.958587646484375, -24.25982666015625, 95.08197021484375, 57.170379638671875, 32.943115234375, 1.924530029296875, 17.55548095703125, 10.77285385131836, 78.2567138671875, 91.87126159667969, 1.1763153076171875, 8.257543563842773, 11.749755859375, -10.503814697265625, 48.742286682128906, 32.149986267089844, -6.067192077636719, 0.0, 82.56536865234375, -13.637908935546875, 31.639984130859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000187.npy"}
{"epoch": 0.39162303664921466, "step": 188, "batch_size": 128, "mean": 24.640993118286133, "std": 45.25967788696289, "min": -95.92424011230469, "p10": -28.46469573974609, "median": 19.86806869506836, "p90": 83.18899230957031, "max": 112.785400390625, "pos_frac": 0.703125, "sample": [14.075340270996094, 48.100555419921875, 83.27920532226562, 0.0, 38.83415222167969, 13.965972900390625, 5.6357421875, 90.7235107421875, -8.266826629638672, 45.51044464111328, 58.69024658203125, -8.519790649414062, 100.01165771484375, -11.729400634765625, 27.694732666015625, -3.0937042236328125, 91.0533447265625, 51.25537109375, 18.044769287109375, 112.785400390625, 73.1221923828125, 63.510650634765625, -18.6693115234375, 2.093505859375, -18.794479370117188, -22.39483642578125, -32.190704345703125, 15.965499877929688, -27.515609741210938, 64.86198425292969, 79.2972412109375, -13.218460083007812, 88.91348266601562, 52.363380432128906, 0.1439361572265625, 29.590866088867188, 80.9542236328125, 67.78799438476562, 68.56858825683594, 26.222763061523438, 19.4783935546875, 38.819854736328125, 7.587646484375, 39.632720947265625, 100.33062744140625, -83.0377197265625, 42.74066162109375, 60.242919921875, 63.322601318359375, 11.840896606445312, 10.561859130859375, 69.854736328125, -20.94915771484375, -3.324432373046875, 38.136505126953125, 90.87835693359375, 107.90606689453125, 8.807262420654297, 62.463348388671875, 97.41900634765625, 45.14739990234375, 32.97820281982422, 55.192955017089844, 76.20303344726562, 4.851207733154297, 3.470062255859375, 9.699539184570312, 79.49723815917969, 53.26832580566406, 15.376495361328125, -1.308807373046875, 84.88394165039062, 75.89202880859375, -0.7916603088378906, -30.679229736328125, 2.1612548828125, 71.08241271972656, 93.5582275390625, -4.315673828125, 5.424766540527344, -64.71636962890625, 62.51325988769531, 20.25774383544922, 78.81967163085938, -0.4838714599609375, -6.099720001220703, 40.48030090332031, 67.30118560791016, 3.74609375, 0.23162841796875, -95.92424011230469, 30.732086181640625, -74.61735534667969, -52.58148193359375, 39.79927062988281, -7.703826904296875, -13.353996276855469, -36.635833740234375, 33.124290466308594, 65.42369842529297, 24.112518310546875, 2.0991592407226562, -54.29664611816406, -93.84750366210938, 50.56048583984375, 12.221229553222656, 47.68207550048828, 74.65319061279297, 82.97897338867188, -10.874004364013672, -57.737518310546875, 1.5307083129882812, 54.51484680175781, 34.50587463378906, 83.15032958984375, -14.79443359375, 91.85467529296875, -5.327617645263672, 0.3317413330078125, -44.772857666015625, -22.455352783203125, -2.546905517578125, 0.0, 16.681915283203125, 22.64202880859375, 40.085872650146484, 3.5228500366210938, -55.70843505859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000188.npy"}
{"epoch": 0.393717277486911, "step": 189, "batch_size": 128, "mean": 29.504898071289062, "std": 38.26386260986328, "min": -73.49143981933594, "p10": -14.729759216308592, "median": 26.148372650146484, "p90": 80.28130493164062, "max": 112.68438720703125, "pos_frac": 0.7890625, "sample": [80.05319213867188, 20.423858642578125, -20.9373779296875, 11.058036804199219, 10.14129638671875, 49.416717529296875, 59.28435516357422, 101.3692626953125, 53.28089904785156, 23.362762451171875, 49.20824432373047, 60.402557373046875, 99.7039794921875, 43.290191650390625, 67.08222961425781, 36.9739990234375, 76.00825500488281, -1.8980541229248047, -8.741363525390625, 8.739013671875, 59.03167724609375, 80.91888427734375, -15.899612426757812, 96.60881042480469, 38.227996826171875, 38.214813232421875, 37.000396728515625, 69.7979736328125, 9.63405990600586, -4.757942199707031, -27.494415283203125, 6.179229736328125, 22.12055206298828, 26.06395721435547, 1.5222320556640625, 25.180885314941406, 26.244403839111328, 3.177448272705078, 16.614288330078125, 19.483291625976562, -2.677793502807617, 85.11468505859375, 70.71817016601562, 33.42108154296875, 59.67431640625, 48.583984375, 31.052490234375, -17.24505615234375, 76.83856201171875, -73.49143981933594, -31.7142333984375, 13.25823974609375, -10.344482421875, 76.9251708984375, 2.19525146484375, 25.37774658203125, 30.857696533203125, 2.6244049072265625, 30.073421478271484, 14.133392333984375, 75.4827880859375, 53.25675964355469, 77.40560913085938, -5.0150604248046875, -14.2283935546875, 7.159511566162109, 7.527677536010742, 4.745460510253906, 4.569004058837891, 7.184898376464844, 48.34800720214844, 3.4427566528320312, -38.04133987426758, 55.70689392089844, 3.911651611328125, 2.1706275939941406, 25.692657470703125, 6.283428192138672, 3.51318359375, 27.351043701171875, 4.9198150634765625, 70.92474365234375, 27.85174560546875, -12.850128173828125, 47.05925750732422, 48.738624572753906, 110.23332214355469, -29.318130493164062, -2.833904266357422, -0.6326236724853516, 13.287322998046875, 86.83074951171875, 28.243560791015625, -56.90380859375, 2.366943359375, 0.0, 112.68438720703125, 21.8480224609375, 32.69972229003906, 83.67230224609375, 73.30081176757812, 36.85459899902344, 78.28367614746094, 40.67506408691406, 24.360641479492188, 0.0, 28.56964111328125, 102.33447265625, -24.0390625, 69.16510009765625, 75.90107727050781, 15.848770141601562, 39.87579345703125, 13.111270904541016, 26.2327880859375, -56.565948486328125, 53.507843017578125, 80.81356811523438, 65.53633117675781, 36.2154541015625, 38.2606201171875, 99.40298461914062, -34.139793395996094, 111.57546997070312, 59.815673828125, -19.814697265625, -5.4744873046875, -9.736129760742188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000189.npy"}
{"epoch": 0.3958115183246073, "step": 190, "batch_size": 128, "mean": 28.43511962890625, "std": 44.021968841552734, "min": -89.25912475585938, "p10": -24.16014518737793, "median": 30.28997802734375, "p90": 82.89212341308594, "max": 131.621826171875, "pos_frac": 0.75, "sample": [-39.63489532470703, 14.625, 101.490966796875, 44.164886474609375, -0.4742584228515625, 40.424896240234375, -67.10081481933594, -3.556642532348633, -23.010910034179688, 31.121673583984375, 48.91316223144531, -35.54975128173828, -19.524009704589844, 82.99566650390625, 48.94873046875, 29.941551208496094, 7.96539306640625, -17.801956176757812, 23.034576416015625, -85.45843505859375, -24.124225616455078, -24.24395751953125, -40.23509216308594, 73.62794494628906, 75.47066497802734, 7.191253662109375, 38.06134033203125, 1.6388397216796875, 51.0146484375, 94.60562133789062, 118.84033203125, -31.141464233398438, 13.845016479492188, 48.71246337890625, 51.71771240234375, -5.10870361328125, 76.11697387695312, 4.967464447021484, 59.11742401123047, 27.621673583984375, 82.84774780273438, -84.13333129882812, 4.8217010498046875, 102.58226013183594, 68.19624328613281, -0.289276123046875, 62.638710021972656, 38.03990936279297, 76.49832153320312, 104.1619873046875, 16.474807739257812, 11.422176361083984, 57.140785217285156, -1.1040573120117188, 45.48406982421875, 49.07708740234375, 5.753124237060547, -2.18310546875, 30.225189208984375, -11.938697814941406, -3.01690673828125, 75.34756469726562, 36.591644287109375, 10.580368041992188, 42.732147216796875, 34.62371826171875, 30.354766845703125, -14.015045166015625, 48.833099365234375, 60.0882568359375, 59.47355651855469, 21.392292022705078, 57.13128662109375, 0.0, 71.27261352539062, 81.06349182128906, 45.49125289916992, 41.729530334472656, 114.736083984375, 48.49551773071289, 1.44281005859375, 32.227264404296875, 30.500106811523438, 51.136375427246094, 76.48748779296875, 15.73529052734375, 14.203514099121094, 12.740997314453125, 61.4012451171875, -6.88250732421875, 17.991012573242188, 131.621826171875, 47.61711120605469, 96.09255981445312, -0.1070556640625, -89.25912475585938, 16.474700927734375, -0.236572265625, 75.70537567138672, -37.928466796875, 50.2197265625, 21.282745361328125, 68.11041259765625, 84.80575561523438, 42.6209716796875, 47.958740234375, 23.078720092773438, 7.152435302734375, -0.4576263427734375, 30.029464721679688, 4.956525802612305, 116.38076782226562, 17.01685905456543, 45.8170166015625, 6.215335845947266, 11.6197509765625, -13.407791137695312, 55.81570816040039, -69.251953125, 52.50677490234375, 37.779052734375, 0.9716072082519531, 10.052860260009766, 52.78623580932617, -74.55644226074219, 88.89300537109375, -34.567962646484375, 91.07090759277344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000190.npy"}
{"epoch": 0.39790575916230364, "step": 191, "batch_size": 128, "mean": 26.24307632446289, "std": 46.36713790893555, "min": -115.06268310546875, "p10": -32.85548553466796, "median": 15.817699432373047, "p90": 89.5921630859375, "max": 118.6302490234375, "pos_frac": 0.75, "sample": [98.28964233398438, 64.68423461914062, -9.130584716796875, -4.238861083984375, 31.885021209716797, 99.84658813476562, 14.427001953125, 15.657981872558594, 2.3459320068359375, -57.44854736328125, 6.4013214111328125, 55.633331298828125, -3.512319564819336, 8.556953430175781, 2.021697998046875, -16.599884033203125, 60.06747817993164, -1.009002685546875, 1.3637256622314453, 47.126861572265625, 44.70851135253906, 89.68679809570312, 11.435043334960938, 4.546600341796875, 0.0213623046875, 17.638412475585938, -7.3460845947265625, -80.72480773925781, 66.09930419921875, 54.38459777832031, 10.059982299804688, 7.1230010986328125, -18.454559326171875, 63.778228759765625, 5.46063232421875, -37.914794921875, 39.81013488769531, -7.573387145996094, -115.06268310546875, -31.516677856445312, 41.430755615234375, 78.208984375, 107.9449462890625, 15.579849243164062, 22.328594207763672, 107.38261413574219, 15.9774169921875, 7.26263427734375, 5.898342132568359, 11.87677001953125, -1.4687347412109375, 73.79052734375, 107.89691162109375, 87.61479187011719, 12.076126098632812, -2.834808349609375, 59.05055236816406, 91.10525512695312, 21.88140869140625, 27.4154052734375, 7.411844253540039, -12.41162109375, 44.95952606201172, -13.338920593261719, 78.01324462890625, 7.749834060668945, 110.16549682617188, 64.80188751220703, 25.743316650390625, 4.033952713012695, 2.773040771484375, 64.92625427246094, 8.129951477050781, -35.9793701171875, 13.272911071777344, -66.37783813476562, 56.85174560546875, 25.800048828125, 63.268585205078125, -10.70159912109375, 1.173675537109375, 43.15911865234375, 2.1504173278808594, 89.55160522460938, -49.81886291503906, 118.6302490234375, 86.19244384765625, -62.896484375, 81.01864624023438, 92.8380126953125, 37.19769287109375, 75.80950927734375, 14.478729248046875, 1.9847049713134766, -1.6010894775390625, -42.9053955078125, 65.10012817382812, 34.397369384765625, 105.25384521484375, 33.63398742675781, 27.729736328125, 11.416671752929688, 72.49037170410156, 52.91450500488281, 117.46780395507812, 84.0892333984375, 3.8433990478515625, 4.1174774169921875, -54.690948486328125, 18.203887939453125, -0.840545654296875, -70.61465454101562, 58.93265914916992, 61.06962585449219, 61.03010559082031, 53.96881103515625, -6.25433349609375, 2.8039398193359375, -17.911117553710938, 75.27130126953125, 44.61016845703125, 43.272308349609375, 25.61632537841797, -41.508514404296875, 0.0, 79.73330688476562, -46.595428466796875, 111.56051635742188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000191.npy"}
{"epoch": 0.4, "step": 192, "batch_size": 128, "mean": 21.726247787475586, "std": 48.3000602722168, "min": -115.56002807617188, "p10": -37.06542358398437, "median": 16.11336040496826, "p90": 87.9399200439453, "max": 112.41107177734375, "pos_frac": 0.6953125, "sample": [8.170852661132812, 30.476318359375, 89.39505004882812, -80.19326782226562, 80.36956787109375, 20.139617919921875, -62.777069091796875, 15.140090942382812, -23.14638900756836, 88.08349609375, 36.41436767578125, 26.6912841796875, 64.35919952392578, 86.0841064453125, 88.07281494140625, -7.5687713623046875, -54.9029541015625, 56.50482177734375, -115.56002807617188, 112.41107177734375, -81.09039306640625, 106.075927734375, -13.36090087890625, -69.12753295898438, 11.7208251953125, 83.14056396484375, 26.52178192138672, 12.196517944335938, -12.559440612792969, 64.60169982910156, -92.58877563476562, 52.148155212402344, 58.54022216796875, 60.590423583984375, 1.281097412109375, 2.8116416931152344, -27.943771362304688, 64.60897064208984, 23.578964233398438, 17.280609130859375, -0.935882568359375, -11.26727294921875, -24.262252807617188, 65.41128540039062, 87.88296508789062, -28.5274658203125, -2.17193603515625, -67.21884155273438, 0.0, 0.0, 1.9202804565429688, -4.59027099609375, -69.2690658569336, 34.7745361328125, -65.3720703125, 65.6384506225586, 1.0616912841796875, 50.033935546875, -5.509864807128906, -21.2449951171875, 50.44397735595703, 44.50912857055664, -14.654876708984375, 2.9305877685546875, 74.9326171875, 89.48300170898438, 76.65696716308594, 84.22947692871094, 0.47252655029296875, 73.49703979492188, -39.5230712890625, 5.6043701171875, 20.85760498046875, -0.018798828125, 36.488258361816406, 56.253623962402344, 44.51513671875, 88.23458862304688, 89.93307495117188, 16.439712524414062, 15.093727111816406, 80.68942260742188, -7.970306396484375, -36.01214599609375, -6.72662353515625, 19.2421875, 8.421634674072266, 0.761016845703125, 64.4740982055664, 71.01736450195312, 1.23992919921875, 20.175537109375, -75.35879516601562, 109.454345703125, 15.787008285522461, 9.47320556640625, -1.7502212524414062, -9.499038696289062, 93.75271606445312, 51.216094970703125, 12.249603271484375, 21.7154541015625, 10.143798828125, -23.382572174072266, 10.607433319091797, 30.64190673828125, 40.033294677734375, 13.603317260742188, 93.277587890625, 51.828330993652344, 59.27880859375, -12.76068115234375, 67.8755111694336, 11.78582763671875, 47.226654052734375, 14.219741821289062, 74.73019409179688, -9.21048355102539, -76.9329833984375, 1.629150390625, 4.131843566894531, 91.21588134765625, 53.6199951171875, 35.712738037109375, 19.002365112304688, 91.189208984375, -15.880924224853516, 85.69886779785156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000192.npy"}
{"epoch": 0.40209424083769635, "step": 193, "batch_size": 128, "mean": 23.96562385559082, "std": 43.70136642456055, "min": -75.430419921875, "p10": -27.694287109374997, "median": 19.720722198486328, "p90": 85.63825073242187, "max": 111.73382568359375, "pos_frac": 0.6875, "sample": [79.01282501220703, -4.756019592285156, -5.31683349609375, -9.215377807617188, -13.38433837890625, 62.722869873046875, 19.337806701660156, 43.23797607421875, 43.42027282714844, 34.65032958984375, 23.904434204101562, 2.8789920806884766, 24.0843505859375, 79.31352233886719, 6.746337890625, 66.33560180664062, -25.30364990234375, 39.272369384765625, -0.3440093994140625, 3.8258609771728516, 71.78387451171875, -20.230682373046875, 64.88035583496094, 10.798553466796875, 28.571243286132812, 35.40716552734375, -2.086761474609375, -54.68684387207031, 63.99920654296875, -5.42572021484375, -44.75269317626953, -1.4223785400390625, 24.19989013671875, 27.210723876953125, 61.25297546386719, -19.3314208984375, 93.808837890625, -58.60504150390625, 3.9552173614501953, 35.576316833496094, -75.430419921875, -22.459136962890625, 0.928314208984375, 41.84112548828125, 18.600723266601562, 45.125732421875, 85.44512939453125, 33.652984619140625, 87.8052978515625, 96.64398193359375, -0.6748046875, 10.680328369140625, 43.21057891845703, 6.398689270019531, 52.02146911621094, 3.1558761596679688, 100.63568878173828, 27.514083862304688, -17.0938720703125, 37.2701416015625, -9.269088745117188, 57.041351318359375, 62.80693054199219, 5.660064697265625, 100.97235107421875, -26.91949462890625, 58.09077835083008, 40.3978271484375, 22.271026611328125, -7.9349365234375, -48.090240478515625, -29.905807495117188, 4.441925048828125, 104.37063598632812, -43.379669189453125, 59.287353515625, 73.85293579101562, 91.54412841796875, 10.1871337890625, 96.8646240234375, 74.78152465820312, 31.6336669921875, 75.59161376953125, -24.502059936523438, 99.53939819335938, 103.78692626953125, 10.752388000488281, 1.5594406127929688, -8.854766845703125, 20.1036376953125, 58.9930419921875, -6.3229827880859375, 111.73382568359375, 67.36415100097656, 13.84033203125, 18.987396240234375, 33.12303161621094, 5.159431457519531, 27.9818115234375, 68.821533203125, 55.4244384765625, 86.0888671875, 28.967100143432617, -41.81773376464844, 50.56451416015625, -32.59620666503906, 101.74566650390625, -25.826324462890625, -68.45814514160156, -5.27838134765625, -2.66302490234375, -4.863189697265625, -74.65362548828125, 54.37530517578125, 50.23931884765625, 19.089351654052734, 79.8253173828125, -25.69879150390625, 17.32958984375, -3.864501953125, 2.299652099609375, -4.769378662109375, -68.9549560546875, 3.225128173828125, 71.51483154296875, -29.50213623046875, 3.198577880859375, 65.69743347167969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000193.npy"}
{"epoch": 0.4041884816753927, "step": 194, "batch_size": 128, "mean": 18.382205963134766, "std": 43.289573669433594, "min": -94.27847290039062, "p10": -23.945980834960935, "median": 10.112468719482422, "p90": 78.49981384277343, "max": 120.75729370117188, "pos_frac": 0.640625, "sample": [116.40664672851562, 23.240493774414062, 22.604217529296875, -48.46906280517578, 23.47735595703125, 112.8763427734375, 45.393524169921875, -9.959701538085938, 2.629547119140625, -2.1072349548339844, 76.36663055419922, -22.295562744140625, -3.591684341430664, 4.8351898193359375, 15.627166748046875, 2.9524364471435547, -10.340324401855469, 52.89952850341797, 91.49337768554688, -13.521774291992188, 18.333831787109375, -4.42921257019043, 38.53824234008789, 6.9780731201171875, 13.303497314453125, -18.708881378173828, 2.7841949462890625, 28.62152099609375, 3.4129791259765625, -85.53289794921875, -10.87939453125, 25.899009704589844, -8.463790893554688, 11.5244140625, 20.254837036132812, 15.990692138671875, 120.75729370117188, 58.08557891845703, 14.772796630859375, 67.48143005371094, -18.148681640625, 106.412353515625, -7.8634033203125, -7.357757568359375, -13.798812866210938, 91.47506713867188, -31.902305603027344, 62.05683135986328, 66.37104797363281, 9.411865234375, -10.411521911621094, 31.459716796875, 29.304977416992188, 37.01611328125, 113.582763671875, 49.05232238769531, 42.73487854003906, 31.03460693359375, 35.555213928222656, -14.10040283203125, 17.3902587890625, 12.3875732421875, -1.6506805419921875, 65.630859375, -47.80792236328125, -26.3856201171875, 8.348892211914062, -94.27847290039062, 33.737060546875, -15.98077392578125, 106.9710693359375, -14.014617919921875, -29.618255615234375, 88.91400146484375, -1.71099853515625, 83.95932006835938, -2.713287353515625, -19.531517028808594, 36.066253662109375, 69.06613159179688, -22.900421142578125, 5.3914794921875, 71.83001708984375, 80.24496459960938, -5.750570297241211, -64.50518798828125, 55.82025146484375, -6.418739318847656, 66.11648559570312, 77.75189208984375, 46.3494873046875, -68.92324829101562, 17.287582397460938, -40.92302703857422, 5.253116607666016, -0.3935546875, 7.398113250732422, -74.09303283691406, 111.8741455078125, 6.9641571044921875, -5.796051025390625, 45.871337890625, 1.9162063598632812, 8.182563781738281, -49.515533447265625, 14.715133666992188, -13.053253173828125, 29.3985595703125, -5.81475830078125, -14.39312744140625, 5.2283782958984375, 93.65328979492188, 62.525909423828125, 10.813072204589844, 39.08927917480469, 32.940185546875, 64.09540557861328, 2.309539794921875, -33.45054626464844, 3.7473526000976562, 11.40301513671875, -4.6517333984375, 59.22711181640625, -21.55426025390625, 3.5435943603515625, -21.678802490234375, 41.149017333984375, 54.742034912109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000194.npy"}
{"epoch": 0.406282722513089, "step": 195, "batch_size": 128, "mean": 17.847671508789062, "std": 44.795326232910156, "min": -88.23529052734375, "p10": -31.26360168457031, "median": 9.400598526000977, "p90": 76.70082244873046, "max": 129.37908935546875, "pos_frac": 0.6328125, "sample": [-9.465408325195312, -4.359291076660156, -1.6856689453125, 69.10736083984375, 76.78645324707031, 70.10401916503906, -41.794464111328125, 1.198150634765625, 59.714141845703125, 9.587360382080078, 40.11328125, 9.213836669921875, 3.1734161376953125, -9.42702865600586, -6.7417144775390625, 20.39764404296875, -11.049072265625, -16.4100341796875, 19.748809814453125, 30.12530517578125, -14.556961059570312, 16.969406127929688, 77.743896484375, -11.817161560058594, -16.483154296875, -4.3633575439453125, 0.376708984375, -0.1007080078125, -30.8360595703125, 10.868865966796875, -13.2823486328125, 66.06640625, -20.415328979492188, -14.20574951171875, 7.0931396484375, 2.82720947265625, 92.11347961425781, 94.88748168945312, -3.860504150390625, 0.351043701171875, -22.890762329101562, -9.24298095703125, 15.150245666503906, 7.156768798828125, 71.66036987304688, 5.993110656738281, 76.66412353515625, -70.96453857421875, -21.489105224609375, 44.366180419921875, -41.42024230957031, 36.4283447265625, 33.44496154785156, 36.700927734375, -86.24908447265625, 24.07672119140625, -2.6763916015625, -73.88809204101562, 78.12665557861328, -7.775001525878906, 26.822677612304688, -88.23529052734375, 4.762542724609375, 59.87712860107422, -16.658363342285156, -77.22433471679688, -67.10101318359375, -12.8372802734375, -1.913604736328125, 49.96868896484375, 97.16484069824219, 36.84039306640625, 65.3974609375, -12.169952392578125, 31.88677978515625, 93.022216796875, 11.575439453125, 1.78387451171875, 35.768585205078125, 16.8031005859375, 40.72981262207031, 58.459991455078125, 33.39306640625, 48.837188720703125, 7.535545349121094, -2.4687652587890625, -0.28009033203125, -57.32403564453125, -0.942967414855957, 21.095794677734375, -32.261199951171875, 121.50131225585938, 70.93826293945312, 0.4801483154296875, 69.0310287475586, 56.145782470703125, 129.37908935546875, 3.51910400390625, 68.55149841308594, 98.83511352539062, -32.32447052001953, 53.1064453125, 9.850341796875, 67.66581726074219, 6.56744384765625, 10.410171508789062, -13.64569091796875, 68.66119384765625, 52.45196533203125, 77.24508666992188, 107.51992797851562, 8.3201904296875, -82.763916015625, 24.341705322265625, -29.879844665527344, 7.889183044433594, 88.82672119140625, -65.50335693359375, 11.278640747070312, 66.20233154296875, 9.94561767578125, -4.904533386230469, -1.9492950439453125, 53.128753662109375, 76.53738403320312, 49.959381103515625, 47.63720703125, -13.6478271484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000195.npy"}
{"epoch": 0.4083769633507853, "step": 196, "batch_size": 128, "mean": 30.164718627929688, "std": 40.97949981689453, "min": -69.64434814453125, "p10": -14.420933532714843, "median": 19.270278930664062, "p90": 81.49465637207031, "max": 125.45953369140625, "pos_frac": 0.765625, "sample": [0.05165672302246094, 11.20794677734375, 24.82373809814453, 1.56976318359375, 18.044891357421875, 9.578369140625, -21.293670654296875, 5.382667541503906, 66.56767272949219, 0.5126495361328125, 4.234107971191406, 84.59158325195312, 45.98406982421875, 47.666961669921875, 105.1204833984375, -2.291534423828125, 1.8753662109375, 2.322998046875, 18.235000610351562, 24.079498291015625, 49.97552490234375, 6.803165435791016, 56.218170166015625, 79.70409393310547, -0.1261138916015625, 63.9776611328125, -11.718109130859375, -1.3183460235595703, -56.97334289550781, 65.4061279296875, 108.27841186523438, 87.0038070678711, 59.320037841796875, 58.627197265625, 1.551910400390625, 54.864532470703125, 9.384902954101562, -36.76593017578125, 72.09600830078125, 51.73460388183594, -25.149673461914062, 96.0743408203125, 38.34405517578125, -8.972564697265625, 13.850399017333984, 10.731163024902344, 18.95562744140625, 97.17138671875, 21.700927734375, 57.47021484375, -32.44378662109375, 0.8440437316894531, 0.2756805419921875, 6.512554168701172, -4.265159606933594, 82.63104248046875, 80.10220336914062, 3.4554834365844727, 4.41424560546875, 52.2935791015625, 62.928375244140625, 11.829254150390625, 81.4854736328125, -8.921127319335938, 21.8577880859375, 88.44094848632812, -38.50762939453125, 35.785736083984375, -11.93218994140625, -17.569580078125, 17.584129333496094, 19.584930419921875, 1.0697803497314453, 64.495361328125, 46.67784118652344, 74.31719970703125, 1.41632080078125, 20.3433837890625, 63.092864990234375, 48.168975830078125, 13.95611572265625, -48.71647644042969, 59.54283142089844, -3.6815948486328125, -14.73486328125, 69.5595703125, 63.9920654296875, 77.11726379394531, 72.83114624023438, 4.16156005859375, 77.57855224609375, -69.64434814453125, 31.9779052734375, 108.19973754882812, 54.64373779296875, -5.027374267578125, -12.358970642089844, -14.286392211914062, 73.42919921875, 76.90409851074219, 64.58317565917969, -19.004608154296875, 7.9212646484375, 125.45953369140625, 90.32162475585938, 16.611894607543945, -27.775623321533203, 70.73116302490234, 80.87838745117188, 81.51608276367188, 67.81167602539062, 106.512451171875, 77.33834838867188, 77.357421875, -5.73114013671875, -5.8944091796875, -6.56743049621582, 65.14866638183594, 62.902061462402344, 37.923095703125, 0.124237060546875, -8.764593124389648, -41.952423095703125, 55.5465087890625, 12.19216537475586, 2.849609375, -9.31866455078125, 4.46759033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000196.npy"}
{"epoch": 0.41047120418848165, "step": 197, "batch_size": 128, "mean": 38.507484436035156, "std": 43.9466552734375, "min": -74.7935791015625, "p10": -10.489158630371092, "median": 41.97639465332031, "p90": 92.1558609008789, "max": 146.91925048828125, "pos_frac": 0.8046875, "sample": [-69.66603088378906, 5.944671630859375, 3.0799560546875, 75.80859375, 56.12580871582031, -10.061676025390625, 68.66600036621094, 23.306808471679688, 18.861408233642578, 89.34844970703125, 97.32778930664062, 78.23129272460938, -9.72137451171875, 55.153411865234375, 51.34626770019531, -31.607715606689453, 126.03790283203125, 101.07500457763672, 61.42063903808594, 104.8046875, 1.3576107025146484, 61.382789611816406, 92.32585144042969, 24.695068359375, 22.294525146484375, 31.112136840820312, 66.53390502929688, 91.05902099609375, 81.29611206054688, 83.39393615722656, 73.3271484375, 52.89513397216797, -53.58807373046875, 56.05982208251953, 30.0556640625, 41.128929138183594, 69.36241149902344, -1.9145889282226562, 37.087799072265625, 92.65020751953125, 45.48883819580078, 55.44329833984375, -74.7935791015625, 92.0830078125, -48.578521728515625, 22.06353759765625, 100.81854248046875, 0.5977935791015625, -11.486618041992188, -6.713836669921875, 56.38385009765625, 54.086639404296875, 23.814208984375, -16.979888916015625, 5.62506103515625, -0.7573661804199219, 71.05830383300781, 0.4561767578125, 85.12973022460938, 71.15325927734375, 64.28533935546875, 146.91925048828125, 5.310310363769531, 96.37384033203125, -33.975250244140625, 2.134113311767578, 62.74757385253906, 78.59320831298828, 34.00238800048828, 45.7344970703125, 18.434005737304688, 38.92333984375, -1.934814453125, 74.18603515625, 2.02239990234375, 81.72335815429688, -0.612518310546875, 42.251739501953125, -44.09368133544922, 5.09307861328125, -1.51702880859375, 9.583755493164062, 52.263458251953125, 53.437286376953125, -28.0865478515625, 74.51651000976562, 3.377227783203125, 56.87549591064453, 68.50579833984375, 22.078765869140625, 34.027313232421875, -0.419830322265625, 87.8861083984375, 43.9617919921875, 59.993804931640625, 41.648193359375, 26.52886962890625, -40.208892822265625, 19.048614501953125, -55.40655517578125, 13.2037353515625, 55.41007995605469, -4.537849426269531, 73.12210083007812, 82.09466552734375, 59.468017578125, 37.04345703125, 36.46162414550781, 91.07098388671875, 41.7010498046875, 113.37432861328125, 104.31619262695312, 10.4544677734375, 6.7291412353515625, 61.667686462402344, 18.3350830078125, 96.81988525390625, 1.4905624389648438, 73.31971740722656, 15.610031127929688, 0.0, 130.9449462890625, 55.94441223144531, 58.1563720703125, 66.61911010742188, -41.731048583984375, 55.89959716796875, -3.1324996948242188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000197.npy"}
{"epoch": 0.41256544502617803, "step": 198, "batch_size": 128, "mean": 28.713790893554688, "std": 50.69441604614258, "min": -107.89923095703125, "p10": -28.746279907226562, "median": 23.148635864257812, "p90": 93.24703369140624, "max": 127.623291015625, "pos_frac": 0.7421875, "sample": [78.12689208984375, 22.39404296875, 9.74676513671875, -28.416030883789062, 62.667236328125, -0.51165771484375, 76.86203002929688, 94.49835205078125, 35.196075439453125, 14.77777099609375, -13.78607177734375, -8.231224060058594, 46.153472900390625, 37.684722900390625, 9.47677230834961, -10.378555297851562, 102.66505432128906, 62.20355224609375, 43.8365478515625, 9.653533935546875, 89.77490234375, -90.0474853515625, -75.8160400390625, 92.71075439453125, 127.623291015625, 118.01336669921875, -70.30279541015625, 11.7666015625, 21.982452392578125, 23.4217529296875, 88.64398193359375, 76.65106964111328, 79.86117553710938, 44.54217529296875, 0.409912109375, 56.39629364013672, -34.6790771484375, 126.9478759765625, -45.70182800292969, 118.79074096679688, -37.75140380859375, 3.128326416015625, 86.37518310546875, 74.67704772949219, -20.622894287109375, 0.3215293884277344, 110.47589111328125, -33.390350341796875, -29.516860961914062, 4.822513580322266, 50.57099914550781, 45.981292724609375, -3.0024337768554688, 72.67469787597656, 27.230789184570312, 76.16270446777344, 101.60952758789062, 2.1476516723632812, 24.3812255859375, 48.396484375, 13.545127868652344, 26.55047607421875, -5.904071807861328, 91.22418212890625, 65.06024169921875, 30.203125, 1.6750564575195312, 100.3814697265625, 25.2838134765625, 35.52232360839844, 17.408119201660156, -78.0390625, -17.760986328125, 39.38238525390625, -107.84989929199219, -18.97882080078125, 19.39215087890625, 81.99676513671875, -16.449501037597656, 55.0087890625, 22.875518798828125, 79.87342834472656, 63.48661804199219, -12.106842041015625, 74.0274429321289, 1.2288284301757812, 56.03190612792969, 91.20085144042969, -70.01762390136719, 13.9725341796875, 0.0, 86.86968994140625, 11.911460876464844, -16.920684814453125, -58.3482666015625, 12.04888916015625, -0.12433815002441406, 64.06912994384766, 102.17852783203125, 35.794281005859375, 104.0550537109375, 2.7650146484375, 76.6832275390625, 55.651641845703125, 27.981292724609375, 57.163970947265625, 19.364013671875, 4.111911773681641, 0.95965576171875, -107.89923095703125, 106.86248779296875, 2.639923095703125, 2.388608932495117, -1.4321060180664062, -6.028951644897461, -6.372314453125, 111.9989013671875, 27.53277587890625, 90.87979125976562, 13.248321533203125, 29.116546630859375, -28.23663330078125, 7.48492431640625, -20.244853973388672, 18.140323638916016, 83.89581298828125, 75.48046875, 1.1915283203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000198.npy"}
{"epoch": 0.41465968586387436, "step": 199, "batch_size": 128, "mean": 27.93408203125, "std": 46.027687072753906, "min": -85.25717163085938, "p10": -26.77552547454833, "median": 22.05693817138672, "p90": 89.07094955444336, "max": 124.66876220703125, "pos_frac": 0.7265625, "sample": [17.307159423828125, 43.52325439453125, 83.51882934570312, 73.92154693603516, 96.79650115966797, 70.7441177368164, 70.43070983886719, 5.6097259521484375, 88.42068481445312, 20.6710205078125, 118.72503662109375, 78.69866180419922, 37.243804931640625, 60.369659423828125, 108.15945434570312, -57.406768798828125, 71.76162719726562, 61.685516357421875, 8.094650268554688, -3.01470947265625, -2.9695053100585938, 1.5128097534179688, -1.8237667083740234, 3.0655593872070312, 0.9039306640625, 39.56956481933594, 54.791168212890625, 1.6264419555664062, 23.146392822265625, -0.286834716796875, 47.51233673095703, 49.99774169921875, 63.768028259277344, 21.983200073242188, 30.332237243652344, 58.11878967285156, -45.913124084472656, -14.404136657714844, 74.37543487548828, 11.371612548828125, 11.517127990722656, 76.85562133789062, 11.756195068359375, 3.9625186920166016, 13.475349426269531, 76.08770751953125, -11.388916015625, -6.752174377441406, -4.3265838623046875, -61.30438232421875, 69.13506317138672, 110.25283813476562, 13.55926513671875, 96.2530517578125, -13.120712280273438, 27.550613403320312, 18.992767333984375, 22.684585571289062, 93.50321960449219, -40.218048095703125, -84.60247802734375, 74.7049560546875, 82.78948974609375, 28.146148681640625, -18.91878890991211, -48.627925872802734, 57.261512756347656, 12.703582763671875, -18.17864227294922, 30.480682373046875, 106.77755737304688, -0.354217529296875, 4.75755500793457, 113.79147338867188, 22.192367553710938, -3.0348358154296875, 49.61456298828125, 26.31182861328125, -1.2230358123779297, 35.24470520019531, 100.18765258789062, 6.825508117675781, 76.57901000976562, -11.489471435546875, 73.35162353515625, 92.3082275390625, -70.93787384033203, 66.99978637695312, 78.81365966796875, 36.392608642578125, -31.3853759765625, -19.73468017578125, 56.905303955078125, 96.5755615234375, -7.0265350341796875, 50.174774169921875, -0.060787200927734375, 9.6197509765625, 22.822433471679688, 39.17558288574219, 22.13067626953125, -24.799875259399414, 18.767059326171875, 10.366973876953125, 66.24026489257812, -85.25717163085938, -63.721290588378906, 0.9669876098632812, 12.67608642578125, 54.723968505859375, 14.380470275878906, 11.09405517578125, 78.65194702148438, -3.15093994140625, 70.28797912597656, 16.657562255859375, 90.5882339477539, 3.66583251953125, -6.420372009277344, 58.167945861816406, 8.49261474609375, -70.05807495117188, -11.318450927734375, -32.010887145996094, 124.66876220703125, 82.2293701171875, 44.24415588378906, -33.048065185546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000199.npy"}
{"epoch": 0.4167539267015707, "step": 200, "batch_size": 128, "mean": 24.25342559814453, "std": 54.099674224853516, "min": -102.86869812011719, "p10": -48.843258666992185, "median": 14.673370361328125, "p90": 93.84683837890623, "max": 156.1068115234375, "pos_frac": 0.6796875, "sample": [73.47732543945312, -65.40757751464844, 114.4688720703125, 16.396041870117188, -50.72064208984375, 14.239219665527344, -44.34588623046875, 65.58566284179688, -7.792755126953125, 7.424943923950195, -3.33111572265625, 79.97506713867188, 23.953216552734375, 3.3380966186523438, -90.22628784179688, 80.887451171875, 69.32830810546875, -13.974967956542969, 26.774444580078125, 124.16458129882812, -85.9024658203125, -42.06768798828125, 109.65107727050781, 42.21575164794922, 99.2293701171875, 71.275146484375, -20.91614532470703, -12.235145568847656, 57.111724853515625, -6.4832763671875, -37.691619873046875, -93.84417724609375, 12.445146560668945, -57.66046142578125, 14.78533935546875, 43.202423095703125, -58.315330505371094, 80.29055786132812, 69.76587677001953, -12.5518798828125, 0.0, 55.31571960449219, 90.71463012695312, 14.84234619140625, -3.8328094482421875, -83.31832885742188, 84.3089599609375, 8.50537109375, 122.52020263671875, -8.198074340820312, 17.63568115234375, 13.808837890625, 91.7626953125, 3.7890777587890625, -25.706130981445312, -21.169204711914062, 17.81147003173828, -76.04252624511719, 102.5364990234375, -4.5514678955078125, 55.71751403808594, 4.7606201171875, 1.6022453308105469, 71.0218505859375, 79.05429077148438, 68.69126892089844, 12.434600830078125, 98.7098388671875, -48.038665771484375, 62.199676513671875, 105.14009094238281, 51.75445556640625, -8.180023193359375, 44.96691131591797, 45.95832824707031, 23.34576416015625, -6.545440673828125, -11.272026062011719, 10.105712890625, 65.7061767578125, -83.1226806640625, 102.00390625, 156.1068115234375, 0.4954681396484375, 15.307266235351562, 50.85514831542969, -9.231674194335938, 69.46809387207031, -102.86869812011719, 90.56298828125, 10.838272094726562, 121.77362060546875, -57.54631805419922, 90.58718872070312, 60.320831298828125, 87.43643188476562, 7.150543212890625, 104.11044311523438, -29.68511962890625, -39.22746276855469, 14.5614013671875, 48.24365234375, 76.22677612304688, 34.09794616699219, -15.17364501953125, 83.06620788574219, 12.965225219726562, -56.143646240234375, 5.07208251953125, 0.912353515625, 11.028022766113281, 71.07160186767578, 64.39382934570312, 117.8394775390625, 35.147125244140625, 42.14208984375, -5.5357208251953125, -20.859413146972656, 63.611541748046875, 40.422550201416016, 8.5689697265625, -2.120149612426758, 60.07677459716797, 5.4484405517578125, -10.568817138671875, 13.3189697265625, 33.937225341796875, 2.9682159423828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000200.npy"}
{"epoch": 0.418848167539267, "step": 201, "batch_size": 128, "mean": 29.130123138427734, "std": 45.392799377441406, "min": -83.3218994140625, "p10": -22.313464355468742, "median": 23.782669067382812, "p90": 88.98226394653321, "max": 128.80172729492188, "pos_frac": 0.7109375, "sample": [-58.932159423828125, -15.007354736328125, 59.067962646484375, 78.033203125, 82.87332153320312, 50.2078857421875, 15.545074462890625, 19.199371337890625, 26.81463623046875, 64.80472564697266, 81.80119323730469, 1.5856332778930664, -5.056114196777344, 24.706817626953125, -55.542694091796875, -38.262451171875, 79.56842041015625, 98.3695068359375, 63.68480682373047, 34.28550720214844, 14.598846435546875, 77.33332824707031, 89.17159271240234, 0.0, 68.27704620361328, 40.71533203125, 76.90066528320312, 75.66668701171875, 1.5884056091308594, 91.89419555664062, 99.1895751953125, 52.106781005859375, 18.798324584960938, -8.040901184082031, 0.0, -60.5784912109375, 8.470352172851562, -1.4891357421875, 8.404830932617188, 6.115997314453125, 37.181190490722656, 42.14717102050781, 81.18547058105469, 78.9012451171875, -45.261688232421875, 14.8760986328125, 53.349449157714844, 63.2401123046875, -18.819496154785156, 52.29273986816406, -2.20947265625, 22.12889862060547, 22.8585205078125, 66.147216796875, -2.533720016479492, 26.5186767578125, 27.359519958496094, 85.36572265625, 60.07783508300781, 80.88812255859375, -5.36175537109375, 51.519012451171875, 104.11334228515625, 11.636688232421875, 93.79013061523438, 71.22834777832031, 32.837432861328125, 0.0, 10.167861938476562, -83.3218994140625, 19.515625, 9.782928466796875, -2.020111083984375, 118.0068359375, 43.23583984375, 1.9512176513671875, 0.6474075317382812, 9.070632934570312, -11.822021484375, 35.191619873046875, 24.8214111328125, 27.627197265625, -14.718952178955078, -14.315849304199219, -2.916717529296875, 94.84022521972656, -7.106803894042969, 88.901123046875, -26.78314208984375, 128.80172729492188, 51.67303466796875, 64.8724365234375, 60.13458251953125, -6.9808349609375, -33.0206298828125, -20.39788818359375, 10.235626220703125, 76.48770141601562, 7.690959930419922, 90.74456787109375, 77.57620239257812, 72.50444030761719, -62.064117431640625, 70.9271240234375, 18.51934814453125, -7.455841064453125, -52.27955627441406, 33.063232421875, 89.937255859375, -53.856414794921875, 21.17291259765625, 25.4359130859375, 17.71240234375, -0.350677490234375, -4.475654602050781, 109.4183349609375, 10.688186645507812, 56.78211212158203, 54.11155700683594, 88.75753784179688, 8.440719604492188, -57.4459228515625, 10.6947021484375, 99.30789184570312, -4.679229736328125, 66.92477416992188, -18.024459838867188, -66.00985717773438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000201.npy"}
{"epoch": 0.42094240837696334, "step": 202, "batch_size": 128, "mean": 28.813377380371094, "std": 51.55903244018555, "min": -107.31680297851562, "p10": -23.906183624267577, "median": 15.801481246948242, "p90": 101.16786651611328, "max": 118.18209838867188, "pos_frac": 0.7265625, "sample": [107.71905517578125, 8.792442321777344, 5.696441650390625, 15.186294555664062, 94.97813415527344, 0.6664810180664062, -14.431144714355469, 105.47952270507812, -57.26829528808594, -6.86322021484375, -6.364013671875, 105.65719604492188, 26.106185913085938, 66.43472290039062, 70.6180648803711, -2.0497894287109375, -21.542266845703125, 72.6623306274414, 4.623176574707031, 112.1796875, 15.267940521240234, 79.37846374511719, -10.8470458984375, 17.3656005859375, -23.12065887451172, 7.948089599609375, 107.89105224609375, 1.0607986450195312, 51.78947448730469, 10.043869018554688, -44.779014587402344, 29.632781982421875, -19.5396728515625, 102.49554443359375, 42.579376220703125, 77.31941223144531, 68.42062377929688, 109.4818115234375, 10.490211486816406, -71.27986145019531, 1.8336410522460938, 109.985595703125, 65.64176940917969, 86.03140258789062, -0.6130828857421875, 57.525421142578125, 47.70782470703125, 37.287391662597656, 82.2900390625, -5.5337066650390625, -79.56721496582031, 67.07136535644531, 87.81993865966797, 1.1555328369140625, -79.07696533203125, 61.43550109863281, 50.54473876953125, -8.76422119140625, 27.756103515625, -22.059036254882812, 45.37400817871094, 69.89936828613281, 103.8193359375, -25.73907470703125, -9.09967041015625, -15.694427490234375, 108.30584716796875, 0.9363632202148438, 11.556793212890625, 10.20477294921875, 1.8323173522949219, 87.84326171875, 111.60025024414062, 2.2574634552001953, 69.38491821289062, 99.47015380859375, 71.60028839111328, 53.89898681640625, 92.62586975097656, 80.13182067871094, 0.0, 118.18209838867188, 1.4441070556640625, 16.33502197265625, 18.2647705078125, 8.850698471069336, 0.935882568359375, -19.910263061523438, 4.59942626953125, -47.026397705078125, 60.1524658203125, 78.7789306640625, 2.3289031982421875, 17.304183959960938, 99.08450317382812, 60.539039611816406, -35.84342956542969, -14.677032470703125, -19.49688720703125, -30.349609375, -107.31680297851562, 97.22262573242188, 44.813499450683594, 88.05978393554688, 35.84808349609375, 7.8812255859375, -21.759979248046875, 15.1937255859375, 30.751819610595703, 0.09374237060546875, 63.995391845703125, 76.30741882324219, -56.26954650878906, 10.8603515625, 66.46759033203125, -103.36859130859375, 19.66668701171875, 72.819580078125, -69.1468505859375, -8.751266479492188, 4.852807998657227, 88.56593322753906, 0.68994140625, -2.5378990173339844, 3.937664031982422, 100.59886169433594, -12.475433349609375, 113.05508422851562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000202.npy"}
{"epoch": 0.42303664921465967, "step": 203, "batch_size": 128, "mean": 22.56171226501465, "std": 39.70869827270508, "min": -96.72044372558594, "p10": -23.783096313476562, "median": 13.529510498046875, "p90": 79.14384765625, "max": 122.42115783691406, "pos_frac": 0.7109375, "sample": [2.6857357025146484, 8.954269409179688, 6.58721923828125, 73.43854522705078, 66.43386840820312, 20.123031616210938, 10.456430435180664, 14.507736206054688, 45.988037109375, 70.81205749511719, 74.697509765625, 30.64105224609375, 4.283367156982422, 38.691680908203125, 29.275833129882812, 37.3834228515625, -11.90997314453125, -52.88622283935547, -8.408668518066406, 10.423324584960938, 17.587020874023438, 0.0, -40.037353515625, 57.655601501464844, -26.056686401367188, 11.146621704101562, -23.45355224609375, 68.96337890625, 0.8671875, -3.7115402221679688, 25.1552734375, 45.822021484375, 27.5814208984375, 79.1072998046875, 58.00292205810547, -6.933134078979492, 30.18035888671875, 113.09738159179688, 16.41229248046875, -12.638336181640625, 45.87684631347656, 87.05345153808594, 28.528961181640625, 67.06798553466797, 2.220836639404297, -4.823944091796875, 11.105316162109375, 66.31718444824219, -5.342987060546875, 62.96147918701172, 0.5469570159912109, 34.460472106933594, 25.50103759765625, 42.08668518066406, 122.42115783691406, -21.3521728515625, 49.24456787109375, -1.7242908477783203, 82.8330078125, 118.65719604492188, 5.08160400390625, 47.7442626953125, 47.538970947265625, -2.034160614013672, -59.09306335449219, 87.13642883300781, -47.336761474609375, 80.646484375, 11.583755493164062, -2.1523895263671875, 37.905487060546875, 3.3243026733398438, -96.72044372558594, 68.79412841796875, -4.69586181640625, -2.758901596069336, -26.578765869140625, 1.29241943359375, -27.569183349609375, 1.83660888671875, 31.70501708984375, 12.707626342773438, -47.09735107421875, 49.3267822265625, 81.2671127319336, 19.7740478515625, 64.52372741699219, 25.50421142578125, 95.880859375, 79.2291259765625, 13.20648193359375, -4.0330810546875, -9.019378662109375, 62.4332275390625, 7.832082748413086, -8.739578247070312, 38.677001953125, -14.338424682617188, 76.51447296142578, 11.656356811523438, 74.99540710449219, 3.432403564453125, -2.9779052734375, 11.5625, 12.931724548339844, -61.135406494140625, -5.07470703125, -34.002479553222656, 17.8729248046875, 103.99075317382812, 0.0, 39.396636962890625, 23.951622009277344, 31.323410034179688, 0.141876220703125, -3.93853759765625, 28.58753204345703, 84.51284790039062, 10.213897705078125, -2.396575927734375, 35.118316650390625, -35.2142333984375, 9.492538452148438, 85.059814453125, 13.8525390625, -24.552032470703125, 10.671792984008789, 24.561052322387695], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000203.npy"}
{"epoch": 0.42513089005235605, "step": 204, "batch_size": 128, "mean": 33.18299102783203, "std": 46.04639434814453, "min": -89.72470092773438, "p10": -19.913607788085933, "median": 25.322879791259766, "p90": 93.11761322021485, "max": 142.4130859375, "pos_frac": 0.7734375, "sample": [112.71127319335938, 72.88066101074219, 109.7415771484375, 15.563003540039062, 14.305824279785156, 92.5264892578125, 5.540851593017578, 58.542198181152344, 4.779460906982422, 103.43502807617188, 12.40618896484375, 8.52886962890625, 72.347900390625, -19.032073974609375, -42.853271484375, -89.72470092773438, 15.520713806152344, -32.118621826171875, 6.717414855957031, 22.52764892578125, 88.77691650390625, 5.589820861816406, 47.04046630859375, 68.97235107421875, 25.51031494140625, 8.52081298828125, 12.844490051269531, -56.881072998046875, 16.059219360351562, 21.542724609375, 32.51116943359375, 142.4130859375, 4.4187469482421875, -30.507644653320312, 42.16925048828125, -14.69134521484375, 11.498662948608398, 120.5250244140625, -3.2115936279296875, 80.24447631835938, 102.68400573730469, 11.416633605957031, 49.6824951171875, -15.92724609375, 76.23721313476562, 76.7403564453125, 15.6190185546875, 115.3472900390625, 90.17044067382812, 107.52362060546875, 77.71268463134766, 61.6739501953125, -1.95208740234375, 65.88379669189453, 35.91515350341797, -3.2665481567382812, 73.95913696289062, 34.85486602783203, -1.1947860717773438, -29.26226806640625, 31.222991943359375, 74.43746948242188, 32.824798583984375, -21.97052001953125, -9.688262939453125, 106.19561767578125, 65.80279541015625, 74.3245849609375, -74.24261474609375, 84.15554809570312, -7.65812873840332, 76.68487548828125, -1.610595703125, -11.998031616210938, 5.891476631164551, 6.77490234375, 29.963008880615234, 86.47071075439453, -2.6508636474609375, 19.20086669921875, 8.884010314941406, -5.195404052734375, -41.62865447998047, 11.39654541015625, 71.02545166015625, -5.993408203125, 56.998687744140625, 14.991241455078125, 81.170654296875, 73.16690063476562, 93.08148193359375, 22.136871337890625, -38.337554931640625, 98.97235107421875, 5.063262939453125, 0.69403076171875, 84.05250549316406, -39.59991455078125, 25.451980590820312, 90.953125, 9.366744995117188, -0.787567138671875, 9.33685302734375, 39.16773986816406, 69.66459655761719, 103.9783935546875, 25.60992431640625, 25.19377899169922, 82.42721557617188, 39.70416259765625, 69.09480285644531, 0.7079620361328125, 52.898193359375, 81.86923217773438, 14.771202087402344, 27.747161865234375, 70.87223815917969, 3.425567626953125, -7.296573638916016, 30.646636962890625, -36.832237243652344, 29.307830810546875, 93.20191955566406, 53.483360290527344, -44.684051513671875, 116.98358154296875, 0.14470672607421875, 4.494873046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000204.npy"}
{"epoch": 0.4272251308900524, "step": 205, "batch_size": 128, "mean": 31.51034164428711, "std": 48.70752716064453, "min": -81.79322814941406, "p10": -28.349230957031242, "median": 28.551742553710938, "p90": 91.64468383789062, "max": 134.0887451171875, "pos_frac": 0.7734375, "sample": [69.63003540039062, 0.0, -4.754486083984375, 3.2150726318359375, 9.490346908569336, 82.0396728515625, -25.750579833984375, -25.247955322265625, -34.52394104003906, 0.8074951171875, 134.0887451171875, 68.2720947265625, 5.704196929931641, 11.252212524414062, -12.239990234375, 8.641937255859375, 112.95771789550781, 103.6202392578125, 78.32760620117188, 11.550689697265625, 11.301406860351562, 37.669105529785156, -16.9488525390625, 95.27494812011719, 57.1090087890625, 12.9920654296875, 65.26678466796875, 16.112449645996094, 105.2989501953125, 47.14070129394531, 69.38774108886719, 54.58220291137695, 2.35650634765625, 7.053985595703125, 10.980987548828125, 6.485040664672852, 67.18970489501953, 12.443115234375, 55.10674285888672, 9.746650695800781, 28.91607666015625, 1.225921630859375, -46.9114990234375, 82.09745788574219, 13.803985595703125, 65.584716796875, 0.0, 29.98565673828125, -71.37420654296875, 30.915496826171875, 2.4764175415039062, 21.889007568359375, 91.05838012695312, 94.64523315429688, -60.181671142578125, 8.194625854492188, 82.9849624633789, 58.65928649902344, 1.25811767578125, 57.096710205078125, -74.72936248779297, -3.38739013671875, 6.69000244140625, 48.28700256347656, -34.412750244140625, 2.9097900390625, 93.01272583007812, -5.254669189453125, -68.61410522460938, 129.97869873046875, 80.37826538085938, 87.86883544921875, -60.60540771484375, 51.06256103515625, -1.1804428100585938, 73.19017028808594, 5.9268646240234375, 72.52177429199219, 69.58787536621094, -6.28546142578125, 32.93980407714844, 112.71563720703125, 67.43267822265625, 87.60708618164062, 55.04536437988281, 10.355598449707031, -77.90191650390625, 2.0100555419921875, 71.10449981689453, 72.57484436035156, 115.33612060546875, 74.98837280273438, 116.39971923828125, -24.023529052734375, 35.977447509765625, 99.58135986328125, 84.80755615234375, -1.2578506469726562, 15.362833023071289, 1.982086181640625, 114.511474609375, 72.41107177734375, -4.86273193359375, -43.579742431640625, 46.253082275390625, 57.9285888671875, 3.473907470703125, 72.614990234375, 67.26264953613281, 73.03364562988281, 28.187408447265625, 15.33538818359375, -81.79322814941406, 7.030427932739258, 41.9249267578125, -45.97352600097656, -65.81594848632812, 88.18528747558594, -9.239494323730469, 68.82952880859375, 53.289794921875, 59.6536865234375, 0.9238128662109375, 51.245452880859375, 9.780044555664062, 85.296875, 0.0, 31.47870635986328], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000205.npy"}
{"epoch": 0.4293193717277487, "step": 206, "batch_size": 128, "mean": 29.675325393676758, "std": 46.0465087890625, "min": -116.66180419921875, "p10": -24.070028686523436, "median": 26.90437126159668, "p90": 92.03455047607422, "max": 113.3426513671875, "pos_frac": 0.7578125, "sample": [101.41726684570312, 24.988765716552734, -72.6087646484375, 75.91143798828125, 111.8201904296875, -0.8739433288574219, -20.958251953125, 31.199798583984375, 14.29400634765625, 87.18992614746094, 39.39680480957031, 92.12513732910156, 97.12411499023438, -18.46841049194336, -35.58856201171875, 35.5914306640625, 37.611083984375, 5.4874420166015625, -3.862823486328125, 91.9957275390625, 0.9465808868408203, 78.76608276367188, -8.801612854003906, -27.086517333984375, -3.35931396484375, 7.9130706787109375, -43.78752136230469, 48.73968505859375, 61.45824432373047, 8.295799255371094, 5.58221435546875, -8.07098388671875, -59.06145477294922, 74.35145568847656, 73.67507934570312, -13.793342590332031, 110.57359313964844, 90.22463989257812, 98.62548828125, 42.58930206298828, 15.611480712890625, 36.41154479980469, -5.338134765625, 28.819976806640625, 49.255462646484375, 32.270904541015625, -25.9725341796875, 90.76419067382812, 74.38442993164062, 82.15629577636719, 52.78106689453125, 64.2076644897461, 67.24716186523438, 43.11273193359375, 3.376955032348633, 43.89753723144531, 51.95738220214844, 107.07440185546875, -116.66180419921875, 22.8511962890625, 59.43792724609375, -39.622581481933594, -21.029006958007812, 73.77976989746094, 30.999481201171875, 12.0386962890625, 29.8853759765625, 18.52947998046875, -4.552717208862305, 44.462005615234375, 9.757537841796875, 73.838134765625, 50.789459228515625, -72.0157470703125, 104.782958984375, 3.492218017578125, -12.009246826171875, 45.50373077392578, 11.283416748046875, 1.5108509063720703, -23.254669189453125, -32.470245361328125, 97.135498046875, 70.26290893554688, 75.30012512207031, 11.12359619140625, 54.39117431640625, -39.44573974609375, 10.092878341674805, 0.4396095275878906, 66.41098022460938, 74.55583190917969, 69.11544799804688, 6.991123199462891, -20.684173583984375, 49.5731201171875, 20.7572021484375, -12.265029907226562, 4.67236328125, -4.402891159057617, 75.33819580078125, 31.0216064453125, 103.74847412109375, 93.18405151367188, 2.279998779296875, 79.05206298828125, -37.28668212890625, -14.853256225585938, -47.547393798828125, 17.4095458984375, 2.9240264892578125, 113.3426513671875, 76.82069396972656, 5.77117919921875, 1.4906196594238281, 100.93930053710938, 65.78524017333984, 5.60870361328125, 91.46173095703125, 37.1195068359375, 1.8873291015625, 12.082275390625, 2.730194091796875, 76.71842956542969, 50.54045104980469, 7.8946380615234375, -22.5057373046875, 16.5421142578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000206.npy"}
{"epoch": 0.431413612565445, "step": 207, "batch_size": 128, "mean": 25.478565216064453, "std": 55.33804702758789, "min": -106.8033447265625, "p10": -38.71475830078124, "median": 13.795703887939453, "p90": 98.56165771484375, "max": 149.04376220703125, "pos_frac": 0.6875, "sample": [78.80712890625, 47.08404541015625, 16.0673828125, -9.788825988769531, 17.123199462890625, 6.486846923828125, 29.527374267578125, 123.8995361328125, 48.122344970703125, -4.429759979248047, 28.671621322631836, 129.18328857421875, -53.849365234375, 5.2360076904296875, -7.66156005859375, 89.8097152709961, 1.6122817993164062, 54.73002624511719, 135.53692626953125, -10.94354248046875, 85.62631225585938, -12.49322509765625, -101.63702392578125, -28.826496124267578, 73.291015625, 30.892105102539062, -11.828475952148438, 13.300674438476562, -79.23648071289062, 66.52024841308594, 81.95703125, 16.744247436523438, 10.746337890625, -30.739410400390625, 76.13137817382812, 78.45919799804688, 16.2259521484375, -2.934955596923828, 21.73653793334961, 100.41140747070312, -12.419677734375, 8.175765991210938, 50.182861328125, 7.984344482421875, 6.708251953125, -22.686798095703125, 4.451993942260742, 4.790924072265625, -91.20831298828125, 6.201568603515625, -106.8033447265625, -10.168212890625, -50.4913330078125, 28.286376953125, 91.4730224609375, 0.0, 10.699657440185547, -31.40869140625, 73.06295776367188, -16.41345977783203, 115.13507080078125, -10.44305419921875, -57.31439971923828, 28.511322021484375, 15.167388916015625, -93.453857421875, 100.38362121582031, -72.04965209960938, 7.376129150390625, -2.30780029296875, 48.018829345703125, -13.04412841796875, 52.46771240234375, 90.24917602539062, 104.54937744140625, 97.06831359863281, 68.8694839477539, -5.6843719482421875, 3.0020751953125, 98.00883483886719, -0.1806011199951172, -71.89028930664062, -6.712158203125, 34.56113052368164, 9.836494445800781, 149.04376220703125, 14.290733337402344, 82.20018005371094, -44.716796875, 1.5819931030273438, 8.12164306640625, 7.5184783935546875, -30.83984375, 108.19683837890625, 19.99224853515625, 71.58615112304688, 48.22520446777344, 3.4935455322265625, 58.04248046875, -83.07846069335938, 109.5594482421875, 91.95419311523438, -17.2064208984375, 85.09144592285156, 6.936103820800781, 119.03094482421875, 9.9779052734375, 3.371612548828125, 9.392576217651367, 84.14495849609375, 9.873794555664062, 70.31155395507812, -2.48724365234375, 99.85157775878906, 90.74310302734375, -6.537528991699219, 90.45724487304688, 86.69706726074219, 116.7100830078125, -21.9755859375, 15.470893859863281, -75.79013061523438, 85.11029052734375, 87.74488830566406, -36.1424560546875, 25.924850463867188, 65.44893646240234, 23.820556640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000207.npy"}
{"epoch": 0.43350785340314135, "step": 208, "batch_size": 128, "mean": 29.482038497924805, "std": 48.4234619140625, "min": -119.37838745117188, "p10": -14.271372604370116, "median": 19.110610961914062, "p90": 92.00130615234374, "max": 146.67822265625, "pos_frac": 0.7421875, "sample": [-6.0242919921875, -10.363174438476562, -28.764892578125, 19.208770751953125, -93.76345825195312, 4.666770935058594, -6.335660934448242, 26.661529541015625, 6.0515289306640625, 84.50277709960938, -11.01723861694336, 10.406097412109375, 56.3919677734375, 49.884674072265625, 65.6361083984375, 125.65234375, 17.23773193359375, 11.652488708496094, 19.012451171875, 136.6226806640625, -86.8021240234375, 83.77737426757812, 3.121917724609375, 50.87873840332031, 11.94708251953125, 75.83468627929688, 117.8402099609375, 74.84341430664062, 16.469223022460938, -13.28240966796875, 146.67822265625, 62.470069885253906, 60.39161682128906, 74.66748046875, -13.762489318847656, 110.75277709960938, -119.37838745117188, 15.533889770507812, 14.522613525390625, 8.885574340820312, 39.253936767578125, -103.22967529296875, 82.41969299316406, -15.45876693725586, 85.8167724609375, 36.42315673828125, 68.39445495605469, 109.83538818359375, 0.0, 134.4754638671875, -35.14268493652344, 1.9989595413208008, 27.66595458984375, -26.502578735351562, 2.6400604248046875, 130.904296875, 10.49249267578125, 29.292800903320312, 29.67235565185547, 99.8038330078125, 42.34189987182617, 37.21195983886719, -9.272819519042969, 51.8392333984375, -5.8232269287109375, 4.831207275390625, 5.7901611328125, -1.3184356689453125, 24.87146759033203, 62.076812744140625, 129.547119140625, 0.0, 27.2440185546875, 57.01513671875, 33.37480163574219, 91.34445190429688, -7.192436218261719, 88.4541015625, 18.810791015625, 67.18701934814453, -3.52655029296875, 44.6798095703125, 81.98324584960938, 96.9591064453125, -8.639419555664062, 57.56056213378906, -10.367462158203125, -1.1058502197265625, 1.3112716674804688, 42.558502197265625, -12.470184326171875, 26.19189453125, 53.592987060546875, 14.380340576171875, 76.91949462890625, 32.382476806640625, 77.70880889892578, 28.12799072265625, 103.51904296875, -16.510704040527344, 3.9761505126953125, 3.1214046478271484, 6.718292236328125, 15.472160339355469, 18.60222625732422, 10.000038146972656, -1.5719223022460938, 78.72210693359375, 17.97808074951172, 28.58758544921875, 22.264892578125, -25.824462890625, 70.14412689208984, 9.430206298828125, 7.25787353515625, -59.5784912109375, 17.278526306152344, 0.9346675872802734, -27.120620727539062, 52.115814208984375, 64.24905395507812, 23.03399658203125, -10.777153015136719, 93.53396606445312, 59.86058044433594, -38.9835205078125, -2.76580810546875, 51.9881591796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000208.npy"}
{"epoch": 0.4356020942408377, "step": 209, "batch_size": 128, "mean": 24.003206253051758, "std": 49.841217041015625, "min": -106.16116333007812, "p10": -33.662115478515624, "median": 15.933788299560547, "p90": 91.85896377563476, "max": 128.9181365966797, "pos_frac": 0.625, "sample": [122.03596496582031, -10.959136962890625, 78.09667205810547, 77.86185455322266, -4.2879638671875, -4.579193115234375, 28.62298583984375, -31.3876953125, -41.8580322265625, 20.6373291015625, 85.69448852539062, -12.210853576660156, -11.164886474609375, 7.368303298950195, 115.2196044921875, 91.93334197998047, 25.199569702148438, 5.108911514282227, 7.97088623046875, 99.13839721679688, 97.22164916992188, -13.85540771484375, 6.402923583984375, -71.6016616821289, 24.47320556640625, 7.885936737060547, 11.549495697021484, 10.61529541015625, -8.818115234375, 79.57695007324219, 83.12112426757812, -52.250274658203125, -3.0523681640625, -60.73077392578125, -46.491790771484375, -57.71101379394531, 120.03408813476562, 1.497528076171875, 0.0, 9.634735107421875, -33.05059814453125, -48.88226318359375, 63.68395233154297, -32.60223388671875, 7.3309326171875, 59.044281005859375, 35.9752311706543, 4.60468864440918, -19.8668212890625, 105.70368957519531, 41.33380126953125, 34.92372131347656, 63.839599609375, -35.0889892578125, 41.38641357421875, 0.48860931396484375, -17.13397216796875, 16.22039031982422, 72.26194763183594, -14.554485321044922, 19.984283447265625, -25.92950439453125, -6.338409423828125, 90.90875244140625, 91.82708740234375, 26.24010467529297, -12.0108642578125, -22.117919921875, 106.7598876953125, -15.9163818359375, 20.03839111328125, 101.057373046875, 15.647186279296875, 22.111846923828125, -2.1314849853515625, 0.0, 62.5699462890625, 80.36384582519531, -20.661338806152344, 82.141357421875, -16.785491943359375, -5.427886962890625, 61.02555847167969, -1.70062255859375, 32.60357666015625, 92.05935668945312, 80.4375, -50.621551513671875, 8.243621826171875, 9.03436279296875, 33.196136474609375, -68.0025634765625, -21.38714599609375, 67.02230834960938, -2.5678234100341797, -3.58868408203125, 28.29400634765625, 51.452369689941406, 128.9181365966797, -106.16116333007812, -11.387847900390625, 5.216835021972656, -44.488922119140625, -6.745849609375, -3.2425384521484375, -79.96292877197266, -5.332466125488281, 35.92628479003906, 76.378662109375, 79.3856201171875, 43.44282531738281, 81.90223693847656, 76.80281066894531, 19.960617065429688, 62.57757568359375, 47.3837890625, 36.64884948730469, 83.44528198242188, 16.392547607421875, 119.57867431640625, 37.0875244140625, -30.74127197265625, 62.02586364746094, 41.880218505859375, 102.50155639648438, 84.34054565429688, -10.459823608398438, 89.74551391601562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000209.npy"}
{"epoch": 0.437696335078534, "step": 210, "batch_size": 128, "mean": 30.796512603759766, "std": 50.78741455078125, "min": -108.47000122070312, "p10": -26.807884216308594, "median": 24.77862548828125, "p90": 91.13731994628905, "max": 172.84521484375, "pos_frac": 0.6875, "sample": [-7.532783508300781, -5.066566467285156, 45.09130859375, -7.8289794921875, 79.24835205078125, 21.119461059570312, 110.1458740234375, 115.66604614257812, 90.81341552734375, 76.35377502441406, 90.16854858398438, 69.97145080566406, 3.0660247802734375, 69.4709701538086, 75.22573852539062, -2.42413330078125, 71.30323791503906, -4.716648101806641, 66.16741943359375, 3.9164886474609375, 28.890199661254883, -0.3625946044921875, -64.0802001953125, 30.978515625, 68.15096282958984, 56.646888732910156, 7.04400634765625, 172.84521484375, -55.47039794921875, -5.9882965087890625, 54.685546875, 28.405319213867188, 16.998931884765625, -0.4258575439453125, 80.06787109375, -15.203826904296875, 5.498497009277344, 54.11474609375, 1.1636028289794922, 31.38116455078125, -0.171783447265625, 81.73368835449219, 18.463382720947266, -8.793136596679688, 12.966484069824219, 66.4835205078125, 31.106842041015625, -5.217010498046875, 78.56182861328125, 13.612874984741211, 25.530792236328125, -1.8126220703125, 71.79861450195312, 61.8760986328125, 50.934173583984375, 67.84629821777344, 14.474273681640625, -59.45643615722656, 104.22015380859375, -7.569671630859375, 56.53985595703125, 66.51858520507812, 86.11061096191406, 72.87005615234375, 25.7183837890625, 49.7210693359375, -92.203369140625, -9.114013671875, 8.3309326171875, 84.8436279296875, -74.42860412597656, -31.267913818359375, 91.89309692382812, 89.70718383789062, 99.62924194335938, 94.05888366699219, 0.28893280029296875, 9.532638549804688, -9.395355224609375, -108.47000122070312, 75.54344940185547, 9.752609252929688, -1.500762939453125, 87.18682861328125, 56.485862731933594, 85.21536254882812, 123.32623291015625, 58.785308837890625, -3.47955322265625, -38.0650634765625, -20.836151123046875, 13.379962921142578, 3.1218624114990234, -30.423416137695312, 23.81757354736328, 75.259521484375, -28.8553466796875, 118.95590209960938, -18.5364990234375, 57.78801727294922, 60.9495849609375, 126.08111572265625, 7.965457916259766, -6.83209228515625, 104.966552734375, 24.026458740234375, 98.96540069580078, -27.507553100585938, 10.168212890625, -16.62286376953125, 60.668739318847656, 25.910247802734375, -6.3606719970703125, 17.9844970703125, 17.4346923828125, -97.55564880371094, 14.7099609375, 92.49575805664062, 31.44683837890625, -75.67915344238281, 72.95752716064453, -1.525482177734375, 77.67686462402344, 61.7958984375, -26.508026123046875, -6.9810791015625, -7.83294677734375, 79.26206970214844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000210.npy"}
{"epoch": 0.4397905759162304, "step": 211, "batch_size": 128, "mean": 39.53279113769531, "std": 46.128746032714844, "min": -76.918701171875, "p10": -10.831109619140623, "median": 35.126861572265625, "p90": 98.44841613769532, "max": 145.63229370117188, "pos_frac": 0.8046875, "sample": [73.13538360595703, 63.03831481933594, 116.85162353515625, 21.956878662109375, 80.31946563720703, 98.19854736328125, 13.891597747802734, 6.384164810180664, -12.678115844726562, 65.03897094726562, 90.71722412109375, -75.06832885742188, 41.206634521484375, 46.50634765625, -11.92669677734375, 89.50192260742188, -0.134979248046875, -0.28314971923828125, 0.07373046875, -17.130966186523438, 10.516883850097656, 75.2559814453125, 19.06616973876953, 45.98081970214844, 78.89091491699219, 27.213104248046875, 8.543281555175781, -58.01300048828125, 78.39283752441406, -10.361572265625, 0.5979804992675781, 78.75875091552734, 1.706634521484375, 72.19367980957031, -7.065467834472656, 68.94052124023438, 24.11616325378418, -56.47386932373047, 17.536224365234375, 74.64974975585938, 5.08966064453125, 30.1346435546875, 8.379730224609375, -0.466644287109375, -3.377685546875, 125.6453857421875, 96.36859130859375, 54.4501953125, 27.92120361328125, 10.08376693725586, 68.33804321289062, -19.036941528320312, 3.494384765625, 4.5630340576171875, 145.63229370117188, 76.08071899414062, 98.57785034179688, -32.92768096923828, 21.4573974609375, 66.00830078125, -7.811492919921875, -21.926544189453125, 14.628158569335938, 133.08935546875, 16.19256591796875, 7.6813201904296875, -5.2054443359375, 65.64701843261719, 98.3929443359375, 16.742019653320312, -26.22882080078125, -1.40057373046875, -2.6746139526367188, 70.94495391845703, 0.9691982269287109, 11.671571731567383, 81.12040710449219, 67.25860595703125, 77.69510650634766, 36.00592041015625, 58.0067138671875, 62.52001953125, -0.8797683715820312, 77.3427734375, 67.97384643554688, 130.45147705078125, 22.443359375, 0.75885009765625, 73.00579071044922, 5.8000946044921875, -12.999526977539062, 126.42599487304688, 107.42092895507812, -33.79150390625, 103.43798828125, 24.478515625, 93.80328369140625, 50.778472900390625, 13.113271713256836, 12.507431030273438, 78.69097900390625, 112.9271240234375, 46.03765869140625, 6.662715911865234, -76.918701171875, 4.190887451171875, 44.94903564453125, 44.311187744140625, 106.309814453125, 48.42193603515625, 94.2852783203125, 12.856401443481445, 90.65467834472656, 39.364654541015625, 10.973684310913086, 34.247802734375, 112.05526733398438, 101.327880859375, 70.055908203125, 79.41835021972656, 24.00042724609375, 49.924842834472656, -3.0080928802490234, 89.95887756347656, 55.32294464111328, 48.04896545410156, 2.1812591552734375, 91.09520721435547], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000211.npy"}
{"epoch": 0.4418848167539267, "step": 212, "batch_size": 128, "mean": 29.3961181640625, "std": 50.84250259399414, "min": -133.18035888671875, "p10": -32.11551818847656, "median": 23.42066192626953, "p90": 92.24268493652343, "max": 131.50567626953125, "pos_frac": 0.7265625, "sample": [3.3132095336914062, -2.943756103515625, 35.3646240234375, 60.18012237548828, -16.524749755859375, -54.13841247558594, 1.8090057373046875, 15.524459838867188, 3.7636642456054688, -21.382659912109375, 106.31810760498047, -3.7752227783203125, 126.43197631835938, 2.8984432220458984, 40.10272216796875, -54.740989685058594, 11.349468231201172, 77.32070922851562, 74.14057922363281, 33.0511474609375, 91.00685119628906, 25.90228271484375, -37.1087646484375, 20.166259765625, 81.75521850585938, 16.375579833984375, -30.709442138671875, -77.02962493896484, 40.528621673583984, 66.0536117553711, -21.919811248779297, -63.4833984375, 4.2564544677734375, 37.488189697265625, 21.78610610961914, 61.3797607421875, 23.7257080078125, 97.74639892578125, 62.0704345703125, -6.3665313720703125, 81.72348022460938, -94.68804931640625, -3.2005615234375, -2.440340042114258, -9.335662841796875, 80.6695785522461, 81.14306640625, 79.09140014648438, -3.41949462890625, -56.562828063964844, 71.83529663085938, 35.47113037109375, -19.258209228515625, 108.80078125, 48.00390625, -7.448150634765625, -12.738876342773438, 75.77984619140625, 19.835357666015625, -3.2538528442382812, -8.071617126464844, 6.535331726074219, 79.61036682128906, 36.59820556640625, -9.322853088378906, 78.55166625976562, -11.50775146484375, 59.94496154785156, 71.7916259765625, 96.47755432128906, 52.454193115234375, 82.10359191894531, -64.98619079589844, 41.085540771484375, 85.28292846679688, 15.639976501464844, 59.252899169921875, 77.67745971679688, -0.8712482452392578, 105.09274291992188, 52.49542236328125, 111.12075805664062, 23.115615844726562, 59.47698974609375, -117.26260375976562, 22.120330810546875, -3.3126983642578125, -45.540069580078125, 27.762863159179688, 7.712779998779297, 82.93997192382812, 43.38983154296875, 13.17059326171875, 7.418666839599609, 122.41522216796875, 67.16056823730469, 21.37762451171875, 97.03211212158203, 63.61590576171875, 55.56150817871094, 42.70478820800781, 131.50567626953125, 57.20005798339844, 1.0089454650878906, 20.853179931640625, -133.18035888671875, 91.35745239257812, 112.11929321289062, -35.3963623046875, -12.626861572265625, 6.4804534912109375, -40.03662109375, 35.0631103515625, 7.633842468261719, 64.61111450195312, 2.3853607177734375, 37.7603759765625, 15.908416748046875, 69.25665283203125, 9.552680969238281, 94.3082275390625, 11.4683837890625, 63.42302703857422, 88.29578399658203, 5.949920654296875, -0.92730712890625, 111.16525268554688, 18.053680419921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000212.npy"}
{"epoch": 0.44397905759162304, "step": 213, "batch_size": 128, "mean": 28.352882385253906, "std": 50.27021408081055, "min": -83.38856506347656, "p10": -30.343092346191405, "median": 14.479456901550293, "p90": 103.64983978271485, "max": 144.7022705078125, "pos_frac": 0.671875, "sample": [109.54745483398438, 82.14137268066406, -38.02375030517578, 89.10446166992188, -35.4666748046875, 67.95369720458984, 81.0076904296875, 75.30419921875, -4.137939453125, 30.953399658203125, 20.89312744140625, 14.77923583984375, 21.501312255859375, -23.64191436767578, 14.179677963256836, 132.23236083984375, -4.707122802734375, 21.82666015625, 24.537200927734375, 52.1065673828125, 43.25714111328125, -9.36224365234375, 141.40072631835938, -6.94146728515625, 103.89030456542969, 7.7855224609375, -9.77923583984375, 15.16082763671875, 23.343170166015625, 15.420486450195312, 8.522369384765625, -30.187530517578125, -15.55926513671875, -4.3375091552734375, 95.25852966308594, 36.52642822265625, 82.37043762207031, -2.0496063232421875, 12.005538940429688, -25.0374755859375, 89.85890197753906, 47.01591491699219, 6.84832763671875, 76.50079345703125, -18.678955078125, 82.62957763671875, 46.6400146484375, 8.139801025390625, -45.186431884765625, 27.559814453125, 12.549575805664062, 92.35003662109375, 104.69061279296875, 2.851490020751953, 129.6832275390625, -43.294273376464844, -5.033164978027344, 71.48829650878906, -25.611236572265625, 104.64201354980469, 15.409088134765625, 17.9105224609375, 30.029991149902344, -64.23374938964844, 7.964111328125, 30.7322998046875, 110.6895751953125, -38.844329833984375, -17.489334106445312, -8.440811157226562, 77.92446899414062, -15.0968017578125, 30.76904296875, -83.38856506347656, 27.182952880859375, 10.849082946777344, 55.42500305175781, 96.30862426757812, 8.365734100341797, 28.09820556640625, -18.67218017578125, 8.54739761352539, -2.4647960662841797, -11.4122314453125, -2.25054931640625, 92.768798828125, -50.717437744140625, -6.438446044921875, 47.263580322265625, 127.59033203125, 72.73497772216797, 103.54678344726562, 5.454925537109375, 7.10430908203125, 43.01239013671875, -4.408866882324219, 107.458251953125, -30.706069946289062, -33.303955078125, 144.7022705078125, 1.5425682067871094, 9.536705017089844, -8.923065185546875, 96.45530700683594, -34.766937255859375, 118.44244384765625, 89.21449279785156, -6.905662536621094, 54.88848876953125, 0.0, 10.210014343261719, 49.95915222167969, 7.1836090087890625, 101.30926513671875, 10.889547348022461, -74.91864013671875, 90.13406372070312, 39.721431732177734, -4.040985107421875, 22.97406005859375, 3.600666046142578, 86.7283935546875, -30.9581298828125, 6.574867248535156, 116.38778686523438, -1.0652542114257812, -28.0379638671875, 3.6355361938476562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000213.npy"}
{"epoch": 0.44607329842931936, "step": 214, "batch_size": 128, "mean": 25.98303985595703, "std": 49.09097671508789, "min": -94.42276000976562, "p10": -25.895553588867184, "median": 19.172805786132812, "p90": 89.72681732177733, "max": 163.81539916992188, "pos_frac": 0.703125, "sample": [60.22584533691406, 72.2437744140625, -19.994888305664062, 4.51080322265625, 13.791053771972656, 94.56280517578125, 21.337387084960938, -12.754852294921875, -50.11578369140625, -18.4462890625, 26.638671875, 55.97587585449219, -27.851593017578125, 24.91766357421875, 53.202880859375, 68.92108154296875, -94.42276000976562, 81.9845199584961, 23.343612670898438, 40.97845458984375, -5.301521301269531, -18.7576904296875, 22.83501434326172, 70.19170379638672, 38.2850341796875, 17.466659545898438, 72.0066909790039, -25.0572509765625, 7.0369873046875, 5.829933166503906, 7.032691955566406, 12.565963745117188, 10.556331634521484, 4.263458251953125, 53.442138671875, 23.31146240234375, 163.81539916992188, 76.81721496582031, 104.7213134765625, -81.24087524414062, 33.22611999511719, -34.169891357421875, 3.4519405364990234, 41.231689453125, 7.8284759521484375, 2.4565162658691406, -12.617919921875, 91.29214477539062, 22.398345947265625, -7.6195220947265625, 70.3276138305664, 20.24755859375, -40.1693115234375, 25.811195373535156, 3.3733978271484375, 85.33824157714844, 61.58953857421875, 50.3128662109375, -13.09228515625, 85.3565673828125, 73.37112426757812, 105.432861328125, 65.56124114990234, 17.872406005859375, 11.72747802734375, -31.89557647705078, -5.62689208984375, -42.955604553222656, -0.554473876953125, 0.0, -18.812423706054688, 9.338836669921875, -13.791015625, 8.2152099609375, 70.53884887695312, 2.135040283203125, 52.97089385986328, -59.97978591918945, 5.600151062011719, 14.099845886230469, -17.973236083984375, -82.8848876953125, 82.5081787109375, -0.229156494140625, 78.60964965820312, -24.10699462890625, 93.61528015136719, 53.820884704589844, 22.143035888671875, 16.242706298828125, -86.23129272460938, 35.283294677734375, 27.330352783203125, 107.23855590820312, 91.23626708984375, 10.984466552734375, -14.333503723144531, 54.37799072265625, -4.677734375, 63.30335998535156, 89.07991027832031, 118.80731201171875, 15.0408935546875, 55.72227478027344, -10.76678466796875, 102.30068969726562, 88.4415283203125, 28.912017822265625, -23.35870361328125, 73.82498168945312, 143.6981201171875, 17.959915161132812, -21.292800903320312, -43.51177978515625, -3.86639404296875, 92.55747985839844, 25.542814254760742, -75.72525024414062, 7.25726318359375, -1.014251708984375, 18.098052978515625, 52.7078857421875, 20.808639526367188, 58.32945251464844, -20.931297302246094, 152.43524169921875, 55.958778381347656, 63.865478515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000214.npy"}
{"epoch": 0.4481675392670157, "step": 215, "batch_size": 128, "mean": 27.04085922241211, "std": 54.69382095336914, "min": -113.08230590820312, "p10": -40.79649353027343, "median": 19.733688354492188, "p90": 96.58652038574218, "max": 161.34881591796875, "pos_frac": 0.671875, "sample": [18.591447830200195, 16.947662353515625, 50.571624755859375, 86.67617797851562, 121.1971435546875, 79.00521850585938, 88.89349365234375, 59.689537048339844, 71.3165283203125, -9.235481262207031, 110.43524169921875, 87.4959716796875, 108.71026611328125, -28.909805297851562, 5.85943603515625, 30.19952392578125, -10.088714599609375, 138.49407958984375, 142.2371826171875, 77.86351013183594, 82.80621337890625, 0.533416748046875, 0.0, -28.0701904296875, -4.0806121826171875, -26.137191772460938, 43.127471923828125, 76.35064697265625, 134.15771484375, -10.447242736816406, 87.58280944824219, -0.606475830078125, -16.366973876953125, -49.39674377441406, 2.752460479736328, 10.072509765625, 89.46698760986328, -3.36077880859375, 21.406082153320312, 131.41305541992188, -23.064239501953125, 26.223175048828125, -29.124755859375, -32.527984619140625, -0.12195587158203125, -10.560821533203125, -76.20516967773438, 63.67218017578125, 82.05142211914062, 59.689552307128906, 38.69862365722656, 161.34881591796875, 17.895111083984375, 24.619384765625, -58.983489990234375, 0.8599586486816406, 6.5807647705078125, -43.279541015625, 0.0, 84.96859741210938, -2.4842987060546875, -3.1903152465820312, 50.497406005859375, -39.732330322265625, 57.023681640625, 47.83889389038086, -74.34678649902344, 3.0514373779296875, 99.72503662109375, 29.293212890625, 48.18696594238281, 8.613784790039062, -7.005523681640625, 80.95806884765625, -64.15835571289062, 78.07876586914062, 8.21942138671875, -113.08230590820312, 63.29204559326172, 45.801116943359375, 14.32855224609375, 51.532562255859375, 20.52593994140625, -21.430450439453125, -76.8739242553711, -1.9368114471435547, 81.82252502441406, 14.2308349609375, 99.7012939453125, -38.490386962890625, -13.312278747558594, 36.802093505859375, 21.464378356933594, -3.91400146484375, 6.8881988525390625, 16.886619567871094, 1.523834228515625, 27.477447509765625, 117.34886169433594, 152.869384765625, -33.63238525390625, 28.23479461669922, -76.35089111328125, 0.3743019104003906, -57.629638671875, -45.5986328125, 63.06822204589844, 93.00141143798828, -66.82276916503906, 95.25161743164062, 15.891693115234375, -8.180662155151367, 37.29636764526367, -13.4471435546875, 18.941436767578125, 32.88825988769531, 28.458419799804688, 90.42909240722656, -52.3258056640625, 29.98828125, 49.66020202636719, 2.008087158203125, 16.84649658203125, 100.8948974609375, 25.528045654296875, 65.6944351196289, 27.25762939453125, 89.58682250976562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000215.npy"}
{"epoch": 0.450261780104712, "step": 216, "batch_size": 128, "mean": 31.159339904785156, "std": 58.28197479248047, "min": -140.94000244140625, "p10": -42.538073730468746, "median": 25.509475708007812, "p90": 107.48550109863281, "max": 149.31561279296875, "pos_frac": 0.734375, "sample": [38.05461120605469, 57.63267517089844, 69.21742248535156, 32.13340759277344, 18.568359375, -41.47755432128906, -140.94000244140625, 35.22723388671875, 72.92251586914062, -6.085102081298828, 80.8101806640625, 98.74673461914062, 70.86639404296875, -11.129745483398438, 8.243423461914062, 58.35847473144531, 149.31561279296875, 9.04323959350586, 110.61541748046875, 122.71963500976562, 23.3104248046875, -90.433837890625, 73.24432373046875, 0.974151611328125, -7.478481292724609, 21.80535888671875, -95.0699462890625, 14.308486938476562, 59.566184997558594, 88.44744873046875, -56.05668640136719, -55.371307373046875, -111.23626708984375, 4.019004821777344, 72.76458740234375, 18.81891632080078, 89.18215942382812, -47.035400390625, 116.67510986328125, 81.68553924560547, -11.319110870361328, 91.80902099609375, 124.952880859375, 85.29654693603516, 88.70231628417969, 62.129730224609375, 37.22125244140625, 51.873565673828125, 6.008628845214844, 0.0582427978515625, -51.173553466796875, -5.540424346923828, 136.8441162109375, 49.06938171386719, 8.921859741210938, 52.896331787109375, 14.801239013671875, 110.02566528320312, -13.788726806640625, 4.610782623291016, 0.0, 66.82199096679688, -21.1383056640625, 74.96566772460938, 108.63720703125, 94.9788818359375, 102.00518798828125, -12.835296630859375, -4.82635498046875, 12.30450439453125, 59.22059631347656, 41.68914794921875, -121.995361328125, 16.12158203125, 117.06053161621094, 106.99191284179688, 8.274627685546875, 110.77023315429688, 70.11578369140625, 11.612274169921875, 12.171157836914062, 24.971466064453125, 96.86874389648438, 26.0474853515625, 116.85687255859375, 75.20970153808594, -45.01261901855469, -8.96014404296875, 111.86846923828125, 100.60675048828125, -40.6446533203125, -76.98989868164062, 3.9410400390625, 67.26318359375, -5.828865051269531, -18.6263427734375, 15.1290283203125, -18.069549560546875, 32.56510543823242, 20.373504638671875, 125.2725830078125, 12.90325927734375, 0.0, 71.04143524169922, 85.52748107910156, -80.76101684570312, 2.436717987060547, 71.45912170410156, 71.71589660644531, 60.131919860839844, -0.25612449645996094, 58.837371826171875, 89.25025939941406, -10.41424560546875, 65.2668228149414, -8.485443115234375, 32.715667724609375, 38.87957763671875, 6.3560791015625, 53.150665283203125, 2.9732666015625, -111.98135375976562, 61.87626647949219, 87.45262145996094, 5.6451416015625, 4.05755615234375, -21.67193603515625, 6.138153076171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000216.npy"}
{"epoch": 0.4523560209424084, "step": 217, "batch_size": 128, "mean": 41.4807243347168, "std": 56.20308303833008, "min": -93.05966186523438, "p10": -20.65669174194336, "median": 34.00887870788574, "p90": 111.2808609008789, "max": 221.373291015625, "pos_frac": 0.7578125, "sample": [10.339515686035156, -2.264068603515625, -7.01153564453125, 82.4796142578125, 110.58641052246094, -81.23526000976562, 1.2805023193359375, 102.29498291015625, 7.453422546386719, 46.43060302734375, 104.56623840332031, -6.083251953125, -0.7446746826171875, 33.534759521484375, -24.5849609375, 103.80619812011719, 80.05348205566406, 88.66140747070312, 78.2291259765625, 27.098114013671875, 61.09446716308594, 81.01998901367188, 6.622001647949219, 86.74603271484375, 168.359130859375, 71.93624877929688, 90.21200561523438, 94.28655242919922, 93.64627075195312, 103.58792877197266, 33.13885498046875, 78.86825561523438, -13.344635009765625, 127.90185546875, 59.15068054199219, 27.440643310546875, 221.373291015625, 85.27728271484375, 95.5797119140625, 51.775146484375, -93.05966186523438, 97.36264038085938, 40.019439697265625, 151.4619140625, -83.18025207519531, 41.59119415283203, 12.390884399414062, -23.36302947998047, 42.7125244140625, 22.44134521484375, 3.8482933044433594, 1.754669189453125, 11.165740966796875, 40.650299072265625, 59.7318115234375, -49.30352783203125, -5.293601989746094, 92.69242858886719, 65.7261734008789, -19.939056396484375, -3.5323638916015625, 41.78173828125, -0.1733551025390625, 126.37261962890625, 5.3986968994140625, 46.76786804199219, 11.36872673034668, 101.93504333496094, 79.54456329345703, 8.47418212890625, 77.9477767944336, 85.38675689697266, 43.919830322265625, -3.951812744140625, 103.86383056640625, -20.179763793945312, -29.85693359375, -21.76952362060547, 62.19219970703125, 162.73745727539062, 10.763885498046875, 115.92218017578125, 29.97955322265625, -10.568073272705078, -87.5498046875, 128.2509765625, 116.41921997070312, 112.9012451171875, 17.26287841796875, -71.45458984375, 17.48980712890625, 94.380126953125, -13.45831298828125, 84.80587768554688, -1.2264251708984375, 66.79095458984375, 0.5036697387695312, 15.869770050048828, 7.9885406494140625, 80.41644287109375, -9.92120361328125, 0.0, 16.419219970703125, 38.51179504394531, -31.249053955078125, 129.22543334960938, 86.17256164550781, -13.658355712890625, 50.436798095703125, 123.90501403808594, 64.54005432128906, 26.841567993164062, 23.9793701171875, 139.74737548828125, 85.13296508789062, 11.382125854492188, 4.831825256347656, 26.335906982421875, -29.63922119140625, 9.186561584472656, 62.956939697265625, 34.48299789428711, 53.03228759765625, 19.146347045898438, 25.450286865234375, -22.249679565429688, -3.272216796875, 5.116939544677734], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000217.npy"}
{"epoch": 0.4544502617801047, "step": 218, "batch_size": 128, "mean": 32.75338363647461, "std": 50.61418533325195, "min": -96.18354034423828, "p10": -24.52636795043945, "median": 22.10723876953125, "p90": 95.56424255371094, "max": 158.60272216796875, "pos_frac": 0.7421875, "sample": [40.12713623046875, 17.29230499267578, 4.6151123046875, 102.58309936523438, -8.301361083984375, 6.595184326171875, -10.21905517578125, 112.51629638671875, 21.566612243652344, 7.622314453125, -68.64532470703125, 13.445571899414062, 55.16595458984375, 58.35198974609375, 47.492584228515625, -90.28842163085938, -3.12249755859375, 83.69735717773438, 75.33850860595703, 77.92050170898438, -12.982765197753906, 56.11962890625, 88.02288818359375, 24.34246826171875, 38.326629638671875, 23.523193359375, 12.70526123046875, -13.91046142578125, -21.940536499023438, 68.25370788574219, 73.42560577392578, 123.662109375, 5.4260101318359375, 91.90326690673828, -25.4482421875, 13.385986328125, 81.45572662353516, -27.43255615234375, 82.68626403808594, 69.58590698242188, 147.37844848632812, 50.770835876464844, 21.421112060546875, 2.261087417602539, -30.396652221679688, 97.76829528808594, 14.539196014404297, -67.53872680664062, -46.13458251953125, 91.44505310058594, 38.33720397949219, -24.13127899169922, 81.59745025634766, 47.47084045410156, 98.831298828125, -1.9079723358154297, 95.201904296875, 26.0269775390625, 92.97418212890625, -19.83428955078125, 14.65606689453125, -8.514694213867188, -29.659133911132812, 10.601455688476562, 0.0, 68.26275634765625, 16.49542236328125, 1.2027587890625, -74.1422119140625, 91.90365600585938, 17.954635620117188, 38.98731994628906, -66.073486328125, 95.27520751953125, 34.866729736328125, 95.53570556640625, -2.4717330932617188, 14.066299438476562, 25.1961669921875, 6.3916473388671875, 158.60272216796875, 95.63082885742188, 119.2373046875, 4.56005859375, -6.537353515625, 21.38751220703125, -5.20318603515625, -0.9150314331054688, 117.12210083007812, 31.28863525390625, 20.738731384277344, 69.46466827392578, 19.00689697265625, -70.24905395507812, 86.2903060913086, 77.68172454833984, 95.86119079589844, 93.45916748046875, -42.79473876953125, 48.7506103515625, 4.282806396484375, 79.75396728515625, -8.283458709716797, 16.319610595703125, 52.40380859375, -9.885696411132812, -96.18354034423828, 60.61894226074219, 21.675537109375, 76.0216064453125, 10.7252197265625, 16.007171630859375, 46.86334228515625, 81.21127319335938, -1.9625244140625, 20.744384765625, 121.74307250976562, 3.444976806640625, -17.131454467773438, -15.827484130859375, 98.20709228515625, 68.17247009277344, 70.79196166992188, 22.5389404296875, 46.058013916015625, 82.9852294921875, 37.67780303955078, 8.597991943359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000218.npy"}
{"epoch": 0.45654450261780105, "step": 219, "batch_size": 128, "mean": 33.04231643676758, "std": 53.468788146972656, "min": -124.92926025390625, "p10": -22.475895690917966, "median": 26.354171752929688, "p90": 106.33465576171874, "max": 137.94955444335938, "pos_frac": 0.765625, "sample": [-124.92926025390625, 26.544189453125, 52.38972473144531, 11.517196655273438, 94.74504089355469, 46.596771240234375, -14.64794921875, 53.6217041015625, 91.62164306640625, 30.56097412109375, 75.08206176757812, -0.8838672637939453, 40.66802978515625, -58.1678466796875, -1.3304901123046875, 80.98934936523438, 61.91314697265625, -13.18780517578125, 17.701339721679688, 78.89669799804688, 45.33891296386719, 7.405057907104492, 44.773590087890625, 123.55499267578125, -3.3682174682617188, 26.164154052734375, 86.06607055664062, 4.170158386230469, 77.98797607421875, -24.641815185546875, -9.323413848876953, -4.383035659790039, -97.40884399414062, 43.89000701904297, 15.080780029296875, 97.08875274658203, -3.49041748046875, 35.70230484008789, 24.495147705078125, 92.77157592773438, 111.5185546875, 137.94772338867188, 7.397209167480469, 54.88987731933594, 8.315223693847656, -82.44583129882812, 12.780792236328125, 78.97503662109375, 10.318359375, 121.77525329589844, 0.0, 25.535537719726562, 18.913223266601562, 27.827239990234375, -39.57969665527344, 112.40765380859375, -23.42364501953125, 105.535400390625, 116.76828002929688, 17.094024658203125, 47.59547424316406, 47.51171875, 32.49430847167969, 6.01171875, 1.778855323791504, -11.5645751953125, 6.170928955078125, 5.238151550292969, 2.6722564697265625, 66.06694030761719, -11.608245849609375, 23.0496826171875, 13.147796630859375, 119.51138305664062, 19.53778076171875, -0.7053985595703125, -88.924072265625, 15.697769165039062, 57.32244873046875, 100.293212890625, 4.340215682983398, 100.81236267089844, 25.2313232421875, 27.60394287109375, -5.00373649597168, -36.02397155761719, 118.25103759765625, 4.11897087097168, 119.27980041503906, -3.611114501953125, 108.1995849609375, 67.35018920898438, 53.00004577636719, 73.4705810546875, 80.11835479736328, 91.08935546875, 33.233211517333984, 51.004180908203125, 27.282470703125, 96.02113342285156, -22.069717407226562, 92.0748291015625, 89.39556884765625, -61.32487487792969, 100.97320556640625, 65.01826477050781, 58.951995849609375, 1.3836517333984375, 24.670806884765625, 25.0224609375, 39.227020263671875, -4.0395355224609375, 42.647003173828125, -115.1953125, 135.82427978515625, 112.32672119140625, 46.94151306152344, -56.41754913330078, 12.655563354492188, 14.419158935546875, 53.17236328125, 18.703659057617188, 47.093017578125, 12.061920166015625, 137.94955444335938, -10.423164367675781, 3.2117996215820312, -76.02886962890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000219.npy"}
{"epoch": 0.4586387434554974, "step": 220, "batch_size": 128, "mean": 36.17693328857422, "std": 55.324951171875, "min": -130.73577880859375, "p10": -29.461550140380858, "median": 34.85313415527344, "p90": 111.8696792602539, "max": 145.8577880859375, "pos_frac": 0.71875, "sample": [113.48007202148438, 119.83258056640625, 63.94244384765625, 60.943115234375, 21.897659301757812, 73.6114501953125, 3.42694091796875, 91.3724365234375, 24.19573974609375, 74.50811767578125, -11.4927978515625, -0.7491455078125, 80.57195281982422, -62.41693115234375, 11.278701782226562, 33.596466064453125, 35.269989013671875, 107.89788818359375, -27.66436767578125, 10.313873291015625, 63.11302185058594, -10.285003662109375, 26.27942657470703, 2.3015260696411133, 0.5895175933837891, 145.8577880859375, -6.36383056640625, 102.97184753417969, -72.90542602539062, 19.722732543945312, 48.75982666015625, 4.6462249755859375, -8.7181396484375, -31.160552978515625, 30.394561767578125, 9.724334716796875, -24.28131103515625, 53.875518798828125, 1.8802013397216797, 71.85565185546875, 52.854835510253906, 32.5323486328125, 97.04489135742188, 46.24383544921875, 117.54469299316406, 85.23604583740234, -3.273448944091797, -130.73577880859375, 69.2978515625, 43.7747802734375, 55.04315185546875, 84.74290466308594, -11.727691650390625, 118.52349853515625, -28.73340606689453, 20.375690460205078, -7.936309814453125, -24.50023651123047, 6.486328125, 56.973045349121094, -33.109344482421875, 68.29969787597656, 57.4127197265625, 55.843040466308594, 10.655525207519531, -14.901756286621094, -9.982284545898438, 56.23345947265625, 2.5271453857421875, 90.32421875, 109.22651672363281, 118.181396484375, -6.2705230712890625, 23.338058471679688, 80.51736450195312, 76.16305541992188, -22.6905517578125, 67.2947998046875, 80.00656127929688, -12.68768310546875, 13.738662719726562, 15.464324951171875, 86.09982299804688, 63.539093017578125, -4.5893402099609375, 120.94945526123047, -87.92828369140625, 8.714752197265625, 133.78573608398438, 132.97647094726562, -34.19140625, 34.436279296875, -89.31369018554688, -1.9259300231933594, 64.57870483398438, 124.66000366210938, 112.288818359375, 17.29510498046875, 81.75982666015625, -17.80752182006836, 86.628173828125, 64.69610595703125, -36.933074951171875, -46.706329345703125, 1.8815231323242188, -19.670379638671875, 83.07089233398438, -32.056793212890625, 119.45486450195312, 111.69004821777344, 74.13737487792969, 31.36883544921875, 63.04144287109375, -35.45268249511719, 129.0443115234375, -26.757080078125, 100.6722412109375, 75.41172790527344, 11.943843841552734, 88.2215576171875, -94.16484069824219, 47.89019775390625, 77.45719146728516, 59.37487030029297, 87.0753173828125, -10.807273864746094, 71.53726196289062, 45.844573974609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000220.npy"}
{"epoch": 0.4607329842931937, "step": 221, "batch_size": 128, "mean": 28.033981323242188, "std": 54.39280700683594, "min": -119.22714233398438, "p10": -35.983839416503905, "median": 17.70873260498047, "p90": 100.24585571289062, "max": 134.470458984375, "pos_frac": 0.71875, "sample": [3.947662353515625, 74.81704711914062, -17.11065673828125, 21.6201171875, 51.07145690917969, 11.822479248046875, 20.834716796875, 57.797882080078125, -14.8193359375, -6.064184188842773, 69.38201141357422, 52.143341064453125, 1.447235107421875, 59.715087890625, -5.214179992675781, 86.6217041015625, 14.044281005859375, -71.12226867675781, 7.540103912353516, -12.776802062988281, 75.109619140625, 38.29620361328125, 134.470458984375, -35.02293395996094, 31.796844482421875, 93.30748748779297, 104.55377197265625, 72.87179565429688, 93.1845703125, -0.38818359375, 12.9635009765625, 12.503280639648438, 121.0672607421875, -119.22714233398438, -47.676605224609375, 15.444061279296875, 39.780975341796875, 84.3096923828125, 7.265960693359375, 115.90478515625, -58.06549072265625, 8.6142578125, 7.89556884765625, 35.09368896484375, -71.74714660644531, -4.910919189453125, 94.62992858886719, -73.749755859375, 82.46783447265625, 35.71630859375, -0.14184951782226562, 81.82273864746094, 22.41412353515625, -16.99603271484375, -5.781702041625977, 79.89764404296875, 61.056396484375, -65.16886901855469, 2.7415771484375, 56.857269287109375, 7.96563720703125, 8.995132446289062, 63.404449462890625, 11.841897964477539, 42.49957275390625, -0.08748626708984375, 64.6116943359375, 96.19309997558594, -2.5714187622070312, 78.23655700683594, 74.42974853515625, -11.109512329101562, 69.578369140625, 57.06947326660156, 105.21983337402344, 63.9808349609375, 110.5472412109375, 25.327072143554688, -57.41747283935547, -86.67208862304688, 99.48785400390625, 102.0145263671875, -16.47052001953125, 7.444732666015625, -8.890625, 96.16403198242188, 8.8629150390625, -11.15240478515625, 31.418731689453125, 76.84530639648438, 120.732666015625, 104.14974975585938, 0.25555419921875, 8.169912338256836, -38.2259521484375, 5.935890197753906, -18.70587158203125, 121.9979248046875, -105.44769287109375, 54.931640625, 4.4307098388671875, -81.3365478515625, -26.484603881835938, 120.07406616210938, 16.14984130859375, 17.61785888671875, 17.799606323242188, 102.92341613769531, 96.8729248046875, 10.462196350097656, 57.616912841796875, 46.73345947265625, -34.118927001953125, -19.659015655517578, 6.100563049316406, 4.636650085449219, 83.56405639648438, 28.672088623046875, 81.8912353515625, -7.26568603515625, 60.04026794433594, 85.69784545898438, -25.52349853515625, 103.43753051757812, 30.03082275390625, 9.23405647277832, -100.69735717773438, 9.029510498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000221.npy"}
{"epoch": 0.46282722513089003, "step": 222, "batch_size": 128, "mean": 35.9345588684082, "std": 51.26780319213867, "min": -159.864990234375, "p10": -19.518463134765625, "median": 22.31437587738037, "p90": 107.78757171630858, "max": 136.1854248046875, "pos_frac": 0.78125, "sample": [66.01288604736328, 13.7640380859375, 90.99205017089844, 80.80752563476562, -27.082717895507812, -27.021408081054688, 77.65878295898438, 30.834871292114258, 19.119537353515625, 10.259841918945312, -34.581939697265625, 20.8028564453125, 92.56491088867188, 50.765411376953125, 25.065460205078125, 23.999183654785156, 81.01724243164062, 124.40875244140625, 62.5279541015625, 28.844947814941406, 8.347915649414062, 0.71624755859375, -27.167510986328125, -54.15643310546875, 64.17181396484375, 130.90728759765625, -14.11798095703125, 0.72735595703125, -75.13700103759766, 6.533805847167969, 12.8873291015625, 41.15447235107422, 134.74835205078125, 5.400062561035156, 18.235546112060547, 10.424053192138672, 16.9334716796875, 67.05615234375, 80.97120666503906, 35.064117431640625, 3.7294769287109375, -39.44415283203125, 76.603515625, 2.679840087890625, 93.44895935058594, 8.72812271118164, -31.864463806152344, 42.264434814453125, -159.864990234375, 41.9818115234375, 101.2252197265625, 73.44215393066406, -0.38654327392578125, 106.66389465332031, -11.031482696533203, 136.1854248046875, 20.76514434814453, 105.424560546875, 0.3812370300292969, 5.019256591796875, 27.83746337890625, 24.298568725585938, 25.58587646484375, 110.40948486328125, -8.475311279296875, 15.9278564453125, 89.34130859375, 133.32275390625, 60.0916748046875, 71.46870422363281, 101.59127807617188, 88.86167907714844, 0.8268585205078125, 93.11822509765625, -22.052398681640625, 14.13580322265625, 16.64409637451172, 2.7380638122558594, 118.59951782226562, -9.206298828125, 8.591049194335938, 23.825895309448242, -8.260820388793945, 68.26107788085938, -19.628692626953125, -1.2195587158203125, 111.72328186035156, -19.471221923828125, -7.4721221923828125, 62.79066467285156, 92.31007385253906, 88.42323303222656, 18.309722900390625, 27.960098266601562, 6.353893280029297, -16.266021728515625, 5.0701751708984375, 78.61836242675781, 25.43707275390625, 114.55014038085938, 123.25674438476562, -7.728294372558594, 2.46551513671875, 38.97816467285156, 113.927001953125, -5.741905212402344, 97.53565979003906, 38.19026184082031, 16.824462890625, 97.9644775390625, 20.28662109375, 47.84413146972656, -35.099822998046875, 18.747909545898438, -3.588348388671875, -8.946731567382812, -15.970458984375, 102.8883056640625, -20.906768798828125, 82.2703857421875, 56.20916748046875, 9.67388916015625, 122.0059814453125, 87.33029174804688, 20.662246704101562, 1.3822498321533203, 14.299911499023438, 119.47895812988281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000222.npy"}
{"epoch": 0.4649214659685864, "step": 223, "batch_size": 128, "mean": 39.759456634521484, "std": 57.442787170410156, "min": -102.1331787109375, "p10": -35.12673950195312, "median": 40.70157241821289, "p90": 111.80096893310545, "max": 153.18994140625, "pos_frac": 0.7734375, "sample": [10.801933288574219, -16.285964965820312, 5.600028991699219, 89.70904541015625, 11.195304870605469, 14.825019836425781, 1.844970703125, -11.40709114074707, 23.378925323486328, 104.88728332519531, 34.0450439453125, 107.30696105957031, 72.81304931640625, 8.46173095703125, 47.59417724609375, 146.214111328125, 109.31819152832031, -102.1331787109375, 88.73922729492188, -39.393646240234375, 29.008636474609375, 127.1322021484375, -4.999176025390625, 56.1444091796875, 43.39955139160156, -39.80218505859375, 17.915985107421875, 11.6517333984375, 119.86468505859375, -15.865020751953125, 22.309967041015625, 72.8602294921875, 110.13255310058594, 102.67764282226562, 94.31558227539062, 99.71380615234375, 145.81396484375, -100.49029541015625, 65.8851318359375, -78.69952392578125, 14.278594970703125, 56.74609375, 7.608612060546875, 77.01177978515625, 107.39602661132812, 91.78167724609375, 119.56179809570312, 17.723907470703125, 126.44647216796875, 27.965301513671875, 0.62176513671875, 22.947845458984375, 132.30075073242188, 134.45919799804688, 87.045166015625, 54.709564208984375, 0.783111572265625, 117.728515625, 21.755889892578125, 49.84649658203125, 41.5113525390625, 20.16180419921875, 74.2008056640625, 153.18994140625, 41.77033233642578, 46.20292663574219, 27.454559326171875, 68.7939453125, 82.12519836425781, -9.276443481445312, -14.033416748046875, 51.11735534667969, 7.4966583251953125, 101.05479431152344, -44.75120544433594, 25.84068489074707, -55.77685546875, -31.651123046875, 98.02749633789062, -13.073814392089844, 39.89179229736328, -5.99560546875, 0.16483306884765625, 30.50395965576172, 96.80123901367188, 69.3343505859375, -71.66085815429688, -15.472084045410156, 76.14694213867188, -15.265491485595703, -85.65522003173828, 49.68487548828125, 45.41615295410156, 83.83935546875, 89.908447265625, 128.96649169921875, 78.70138549804688, 83.86055755615234, 70.93601989746094, 9.805084228515625, -1.977691650390625, 69.033935546875, 30.560791015625, 75.43528747558594, 57.618011474609375, -16.20654296875, 96.1365966796875, 140.86099243164062, 0.3333549499511719, 19.811065673828125, 115.69393920898438, -66.77855682373047, -8.7713623046875, 8.104759216308594, 103.39404296875, 66.38389587402344, -68.06986236572266, -3.3625335693359375, 72.12309265136719, 5.6055908203125, -62.5224609375, 95.30474853515625, 60.754974365234375, 45.288543701171875, 31.96624755859375, 22.288002014160156, -33.298065185546875, -52.001068115234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000223.npy"}
{"epoch": 0.46701570680628274, "step": 224, "batch_size": 128, "mean": 40.759185791015625, "std": 53.28098678588867, "min": -103.8021240234375, "p10": -22.614141845703124, "median": 33.9688720703125, "p90": 110.44888000488281, "max": 152.82781982421875, "pos_frac": 0.796875, "sample": [129.50836181640625, 152.82781982421875, 57.05047607421875, 61.741355895996094, 2.432889938354492, 2.93939208984375, -28.06805419921875, 49.19036865234375, 87.85322570800781, 17.66748809814453, 4.7822418212890625, 80.7569351196289, -4.1122894287109375, 77.84127807617188, 29.922714233398438, 33.498748779296875, 9.473388671875, 85.535888671875, 22.552772521972656, -62.93211364746094, 78.5509033203125, 0.1084747314453125, 119.64891052246094, 123.95541381835938, 16.127761840820312, 76.01858520507812, 76.40967559814453, 34.438995361328125, 18.00775146484375, 65.71414184570312, 126.26644897460938, 90.57254028320312, -103.8021240234375, 71.86787414550781, 37.74873352050781, 21.393630981445312, 36.053375244140625, 82.6644515991211, 69.65377807617188, -1.2481689453125, 30.542694091796875, -20.110260009765625, 89.18201446533203, 73.72386169433594, 60.22918701171875, 28.68085479736328, 56.473388671875, -49.807281494140625, 108.67929077148438, 0.7130908966064453, 76.76771545410156, 58.70391845703125, 0.0, 115.07644653320312, 92.0225830078125, 48.372703552246094, 13.837127685546875, -60.84977722167969, 4.7262725830078125, 31.06610107421875, 68.64434814453125, 21.52020263671875, 1.9602622985839844, 96.1805419921875, -1.74896240234375, 114.94699096679688, -22.297607421875, -31.001495361328125, 56.649688720703125, -34.15818786621094, 6.046844482421875, 132.218994140625, 31.681167602539062, 112.07696533203125, 18.496826171875, -10.333984375, -13.922286987304688, 73.0576171875, 9.293060302734375, 118.04461669921875, -20.634353637695312, -15.032264709472656, -101.55169677734375, 9.909866333007812, 13.560111999511719, 99.76556396484375, 45.7869873046875, -23.35272216796875, 106.80325317382812, -88.6821060180664, 95.74600219726562, 90.0247802734375, 52.63922119140625, 103.4019775390625, 30.0316162109375, 83.99847412109375, 109.75112915039062, 60.998687744140625, 37.6688232421875, 6.974250793457031, -35.80982208251953, 27.595855712890625, 7.804290771484375, 18.2608642578125, 21.895797729492188, 57.438133239746094, 151.59060668945312, -23.78791046142578, 115.14285278320312, 80.71072387695312, 19.651596069335938, -50.364288330078125, 11.718149185180664, 2.191375732421875, -2.17236328125, 31.087844848632812, 98.6734619140625, 10.839309692382812, 67.88119506835938, -19.36798095703125, 7.251712799072266, 105.0675048828125, 76.5260009765625, 81.91287231445312, 101.49739074707031, 79.07967376708984, -2.987171173095703, 124.04104614257812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000224.npy"}
{"epoch": 0.46910994764397906, "step": 225, "batch_size": 128, "mean": 27.805259704589844, "std": 63.26692581176758, "min": -116.64437866210938, "p10": -57.447994995117185, "median": 24.085342407226562, "p90": 118.43734436035156, "max": 148.29437255859375, "pos_frac": 0.6328125, "sample": [121.7696533203125, -3.14825439453125, 34.357421875, 117.65234375, 106.973388671875, 6.090911865234375, -80.27740478515625, -75.56793212890625, 72.48452758789062, 112.84515380859375, 28.315719604492188, 72.10079956054688, 10.649139404296875, -10.064132690429688, 123.10476684570312, 31.63616943359375, 139.6796875, 85.73126220703125, 3.750774383544922, 43.04945373535156, -12.43792724609375, -13.852813720703125, 55.506248474121094, -44.34820556640625, 105.58865356445312, 45.0732421875, 119.26113891601562, -6.8908538818359375, 1.6075439453125, -59.97137451171875, -12.816734313964844, -116.64437866210938, 89.16045379638672, 110.32000732421875, 13.974258422851562, 7.632759094238281, 78.10494995117188, -2.042858123779297, -0.500152587890625, -56.366546630859375, 65.30166625976562, 127.01736450195312, -39.37671661376953, -73.83815002441406, -94.13864135742188, -71.78425598144531, 53.139923095703125, 24.925018310546875, -5.972080230712891, 49.35968017578125, -86.9207763671875, 21.54791259765625, 78.9507064819336, 94.497314453125, -10.75830078125, 16.283416748046875, -33.74114990234375, 35.1546630859375, 108.95639038085938, 82.99699401855469, 91.58154296875, 124.90843200683594, 132.35003662109375, -27.6217041015625, 28.73675537109375, -12.064102172851562, -42.006378173828125, 70.08944702148438, 58.28387451171875, 139.55703735351562, 1.330078125, 118.08428955078125, 28.84918212890625, 70.16407775878906, -7.045806884765625, 53.24247741699219, 103.60429382324219, 131.65789794921875, -25.0015869140625, 91.52447509765625, -46.338531494140625, -59.9918212890625, -51.88231658935547, 29.991626739501953, -91.85328674316406, -9.1798095703125, 10.983724594116211, -0.3258552551269531, 43.04060363769531, -109.96844482421875, -7.10662841796875, -36.19329833984375, -14.62310791015625, -14.0703125, 26.343231201171875, -28.56939697265625, 0.419342041015625, 44.915672302246094, 148.29437255859375, 67.75491333007812, 4.859992980957031, 66.60155487060547, 123.54812622070312, 113.50054931640625, -32.551666259765625, 12.571731567382812, -72.6871566772461, 49.363677978515625, 11.56610107421875, 136.96258544921875, -19.766220092773438, 63.8326416015625, 77.86288452148438, 142.4129180908203, 10.30596923828125, 72.37933349609375, 62.919586181640625, -28.036041259765625, 47.4743537902832, -60.79154968261719, 23.24566650390625, 2.5226516723632812, -7.824180603027344, -4.67742919921875, -21.042236328125, 31.74310302734375, 39.67088317871094, 98.14837646484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000225.npy"}
{"epoch": 0.4712041884816754, "step": 226, "batch_size": 128, "mean": 28.228778839111328, "std": 55.265533447265625, "min": -102.207763671875, "p10": -41.11267242431641, "median": 15.324333190917969, "p90": 107.79468231201172, "max": 138.75189208984375, "pos_frac": 0.7265625, "sample": [18.7139892578125, 138.75189208984375, -15.17169189453125, 71.79122924804688, 98.0496826171875, 114.71200561523438, 6.7471160888671875, -0.3555450439453125, 72.51219177246094, 100.421142578125, 115.6287841796875, 48.970401763916016, 10.80413818359375, 118.03691101074219, -6.93310546875, 12.939044952392578, 9.477081298828125, -74.0889892578125, 96.68223571777344, 108.49238586425781, 2.9930496215820312, 0.9549064636230469, 91.95191955566406, 71.84586334228516, 3.638702392578125, 22.559967041015625, -10.403207778930664, -47.94024658203125, 88.04800415039062, 68.27175903320312, 1.461141586303711, 78.86720275878906, 59.0797119140625, 118.99517822265625, 10.305404663085938, 15.72637939453125, 98.17589569091797, -41.17779541015625, -7.74609375, 63.75323486328125, -84.97331237792969, 63.84734344482422, -41.08476257324219, 0.13681411743164062, -16.124725341796875, 56.123138427734375, 14.922286987304688, -8.119537353515625, -102.207763671875, 55.00262451171875, 36.141510009765625, -52.207183837890625, 98.52627563476562, 2.02716064453125, 10.292312622070312, -77.37939453125, 2.8387374877929688, 24.7598876953125, 4.181854248046875, -0.1743621826171875, -5.9690093994140625, 17.368484497070312, 4.993133544921875, -99.59274291992188, 18.032577514648438, -6.908863067626953, 13.6295166015625, 101.86660766601562, 5.524936676025391, 9.97650146484375, -26.525497436523438, 6.927219390869141, 88.745849609375, -23.457229614257812, 67.5751953125, 1.856536865234375, 23.938751220703125, 107.49566650390625, 7.358955383300781, 118.91983032226562, -52.28608703613281, 96.60271453857422, 128.55096435546875, -82.61882781982422, 112.3336181640625, 6.9732666015625, 16.659194946289062, 3.3724441528320312, 34.123046875, 78.36273193359375, 68.984375, 31.56689453125, 93.31371307373047, 47.92431640625, 113.1470947265625, 14.218002319335938, -19.389862060546875, 0.4276580810546875, 79.85047149658203, 56.210205078125, 120.82011413574219, 17.60064697265625, 115.70022583007812, 0.4500732421875, 6.33233642578125, 22.427017211914062, 99.03114318847656, 36.930023193359375, 74.4326171875, -25.676376342773438, 45.54150390625, -40.65473937988281, -1.6689605712890625, 75.2152099609375, 53.86249542236328, -20.10748291015625, 70.49293518066406, -17.264402389526367, -23.6234130859375, 126.4564208984375, 73.22796630859375, -42.37360382080078, -62.98011779785156, 18.49618148803711, -54.20744323730469, -39.40997314453125, 13.989654541015625, -11.909446716308594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000226.npy"}
{"epoch": 0.4732984293193717, "step": 227, "batch_size": 128, "mean": 39.79815673828125, "std": 57.769569396972656, "min": -123.71295166015625, "p10": -29.634455871582023, "median": 33.83650588989258, "p90": 115.02428436279295, "max": 157.27716064453125, "pos_frac": 0.765625, "sample": [-77.12664794921875, 37.040679931640625, 21.054595947265625, 29.364242553710938, 75.6356201171875, 43.206993103027344, 126.98577880859375, 19.64604949951172, 10.677200317382812, 33.78815460205078, 113.43325805664062, 98.90922546386719, 12.991867065429688, -1.98333740234375, 24.384384155273438, 59.238746643066406, 0.0, 120.8936767578125, -111.48162841796875, -71.83566284179688, 73.93441772460938, -64.4876708984375, 91.31173706054688, 109.61155700683594, 144.58688354492188, 157.27716064453125, 112.02783203125, -96.78085327148438, 100.95243072509766, -17.439727783203125, -7.285896301269531, 98.58799743652344, 13.655082702636719, 14.419815063476562, 39.52044677734375, -12.38836669921875, 39.26532745361328, 78.6758041381836, 32.859130859375, 126.2559814453125, 129.41757202148438, 67.10733032226562, 15.049346923828125, 70.48541259765625, 39.0345458984375, -38.567626953125, 21.17535400390625, 65.1962890625, -19.0526123046875, 37.36572265625, 112.32501220703125, 44.82543182373047, -9.9249267578125, -27.234771728515625, 70.431396484375, 145.11672973632812, 8.2216796875, 80.90243530273438, 117.92039489746094, 26.465572357177734, 117.56324768066406, -44.15583801269531, -12.233673095703125, 88.22095489501953, -123.71295166015625, 24.519287109375, 24.673187255859375, 133.36968994140625, -44.170135498046875, 43.55426025390625, 81.43124389648438, 21.185226440429688, 2.4209232330322266, 88.06822967529297, 112.9771728515625, 61.0626220703125, 67.2886962890625, 68.8143310546875, -40.526123046875, 78.658203125, 127.209228515625, 7.607330322265625, 18.219886779785156, 113.9361572265625, 54.5567626953125, 29.831710815429688, 59.803741455078125, 32.451141357421875, 32.51393127441406, 70.47402954101562, -33.937347412109375, 70.63858032226562, 13.62371826171875, -66.80171203613281, 98.28463745117188, 43.535400390625, 93.11741638183594, 17.19402313232422, -17.879920959472656, 104.48587799072266, 2.166259765625, 129.5396728515625, 1.4932003021240234, 119.97088623046875, -56.15558624267578, 103.47402954101562, -4.145267486572266, -2.059234619140625, 3.7213287353515625, -0.8284912109375, 100.27166748046875, 25.596786499023438, 98.68878173828125, 102.02656555175781, 13.980255126953125, 31.24139404296875, 33.884857177734375, 52.84820556640625, 101.27716064453125, -3.876800537109375, -16.937957763671875, -27.790359497070312, 28.0428466796875, 16.518409729003906, -14.565597534179688, 5.677427291870117, 38.835716247558594, 37.752532958984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000227.npy"}
{"epoch": 0.47539267015706804, "step": 228, "batch_size": 128, "mean": 40.995811462402344, "std": 56.789642333984375, "min": -79.08382415771484, "p10": -24.047628784179683, "median": 29.748336791992188, "p90": 115.8361846923828, "max": 144.45437622070312, "pos_frac": 0.71875, "sample": [-70.643798828125, -13.109016418457031, 66.9932861328125, 101.46112060546875, 49.44561767578125, -4.865026473999023, 132.87815856933594, 31.88201904296875, 76.52880859375, 104.73782348632812, 3.1829833984375, -11.044654846191406, -12.485931396484375, 117.92962646484375, 57.725616455078125, 73.90533447265625, -6.3673553466796875, 89.3508071899414, -13.944686889648438, 11.727264404296875, 91.69925689697266, 26.609046936035156, -79.08382415771484, -14.940521240234375, 15.396270751953125, 94.49635314941406, 26.686737060546875, 3.1454238891601562, -18.62091064453125, 69.32492065429688, 12.802490234375, 26.885147094726562, -44.58168029785156, 111.002685546875, 31.393951416015625, 102.99447631835938, 68.973876953125, 84.97393798828125, 7.84979248046875, -7.2576904296875, 76.95108032226562, -8.29388427734375, -20.301681518554688, -27.87103271484375, 87.15335083007812, 9.0152587890625, -7.55950927734375, 16.10753631591797, 47.078277587890625, 87.02761840820312, 42.1712646484375, 130.946533203125, 67.25904846191406, 21.60193634033203, 96.71871948242188, 37.35600280761719, 92.555419921875, 66.84814453125, 122.673583984375, 123.884033203125, -1.17657470703125, -72.68067932128906, -43.666038513183594, 0.00757598876953125, 87.42715454101562, 29.659393310546875, 0.0, 23.223968505859375, 80.57745361328125, 19.501976013183594, 12.532859802246094, 144.45437622070312, 17.06468963623047, -10.311004638671875, -36.244873046875, 99.17106628417969, -18.575225830078125, 110.8558349609375, -16.654296875, 114.93899536132812, -16.43902587890625, -59.81452178955078, 18.48712158203125, 97.1685791015625, 107.94577026367188, 122.28793334960938, -2.4739990234375, 3.883392333984375, 126.33146667480469, 82.10357666015625, -22.409027099609375, 106.40252685546875, 13.3526611328125, 96.76751708984375, 47.61248016357422, 71.44257354736328, -73.09896850585938, 4.79681396484375, 20.51409149169922, 110.90827941894531, 111.45703125, -54.51786804199219, -41.39483642578125, 50.29474639892578, 85.80903625488281, 59.97113037109375, 101.99424743652344, -52.2606201171875, -18.074371337890625, -0.842498779296875, 15.401512145996094, 122.94195556640625, -1.9536895751953125, 92.85223388671875, 101.50494384765625, 125.04940795898438, 136.0926513671875, 72.33959197998047, 24.193702697753906, -47.32330322265625, 85.2772216796875, 88.63299560546875, 129.15261840820312, 26.109222412109375, 143.21356201171875, 5.4729766845703125, 3.9977874755859375, 29.8372802734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000228.npy"}
{"epoch": 0.4774869109947644, "step": 229, "batch_size": 128, "mean": 51.05020523071289, "std": 59.37055587768555, "min": -98.74226379394531, "p10": -19.398931884765624, "median": 50.094709396362305, "p90": 123.76640319824219, "max": 175.12994384765625, "pos_frac": 0.8046875, "sample": [83.59246826171875, -24.09173583984375, 100.38067626953125, 80.642822265625, -12.96142578125, -96.26728820800781, 66.17364501953125, 98.42280578613281, 39.64764404296875, -66.62413024902344, 155.759765625, 126.20326232910156, 92.30560302734375, -0.23194503784179688, 136.8580322265625, 109.22296142578125, 33.760894775390625, 9.4637451171875, 21.483455657958984, 46.81561279296875, 41.58287048339844, 23.1031494140625, 107.5845947265625, -10.14727783203125, 41.09013366699219, -72.12481689453125, 9.740921020507812, 24.988059997558594, 113.75326538085938, 86.73030090332031, 7.453643798828125, 50.7279052734375, 125.11724853515625, 73.32177734375, 19.315216064453125, 137.1007080078125, -3.908477783203125, 20.144378662109375, 150.23931884765625, -98.74226379394531, 8.35003662109375, -59.590240478515625, 64.33836364746094, 102.0772705078125, 71.34254455566406, 143.95306396484375, 175.12994384765625, 113.73085021972656, 67.72732543945312, 100.91065979003906, 40.2674560546875, -0.8498764038085938, 55.05247497558594, 7.3165283203125, 88.00550079345703, 113.34257507324219, 102.68988037109375, 15.970794677734375, 10.820693969726562, -19.61285400390625, 51.665863037109375, -74.77349853515625, 108.7213134765625, 87.48101806640625, -12.260604858398438, -7.3565673828125, 109.91552734375, 15.605537414550781, 80.05148315429688, -1.090423583984375, 1.3358840942382812, 62.29261779785156, 0.8370609283447266, -19.3072509765625, 107.68270874023438, 59.81403350830078, 33.075416564941406, 98.05939483642578, 119.30865478515625, 42.43914794921875, 116.54432678222656, 36.58592224121094, 39.981719970703125, 105.11564636230469, 49.46151351928711, 11.87225341796875, -36.04107666015625, 97.46090698242188, 137.3199462890625, 66.88059997558594, 147.05982971191406, 104.96464538574219, 13.531831741333008, 55.21809387207031, 14.257415771484375, 112.21676635742188, 46.38935852050781, 123.18746948242188, 46.8802490234375, 24.436019897460938, 94.23654174804688, 20.197471618652344, 47.480316162109375, 126.29010009765625, -2.3712310791015625, 13.457275390625, 61.220130920410156, -17.724761962890625, 9.728900909423828, 120.81854248046875, 72.2451171875, -8.371074676513672, 47.049468994140625, 48.821014404296875, 89.11077880859375, 94.38348388671875, 100.2366943359375, -51.3673095703125, 93.78569030761719, -80.14306640625, 135.98663330078125, 70.77618408203125, 81.05352783203125, 26.432815551757812, 77.5516357421875, -68.30441284179688, 145.93222045898438, -39.50749969482422], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000229.npy"}
{"epoch": 0.47958115183246075, "step": 230, "batch_size": 128, "mean": 38.576175689697266, "std": 60.0947151184082, "min": -178.0042724609375, "p10": -21.43107681274414, "median": 28.387771606445312, "p90": 117.98174438476562, "max": 160.65048217773438, "pos_frac": 0.75, "sample": [122.15032958984375, 56.162017822265625, 28.3466796875, 97.48422241210938, 160.65048217773438, 76.4222412109375, 121.37773132324219, -11.83294677734375, 118.39437866210938, 31.288894653320312, 16.085357666015625, 28.428863525390625, 13.84661865234375, -76.92681121826172, -178.0042724609375, 101.14111328125, 101.8089599609375, -2.074319839477539, -7.6143798828125, -28.0318603515625, 3.95245361328125, 12.639423370361328, 124.23110961914062, -21.201309204101562, 0.0, 107.431640625, 1.990875244140625, 32.27206802368164, 64.4873046875, 102.68603515625, 89.42890167236328, 99.8079833984375, 9.843063354492188, 40.063812255859375, -0.5809097290039062, -83.2596435546875, 66.09197998046875, 94.47227478027344, 77.129638671875, -24.818191528320312, 0.0602569580078125, -75.6734619140625, 5.230621337890625, 30.959014892578125, 37.746551513671875, 109.69964599609375, 78.76058197021484, 2.825469970703125, 96.85745239257812, -22.3983154296875, -6.32867431640625, 74.03128051757812, -37.0712890625, 59.1146240234375, 108.04473876953125, -10.887542724609375, 36.00160217285156, 24.012481689453125, 110.11190795898438, 28.106277465820312, -79.77281188964844, 84.97335815429688, 132.596923828125, 85.459228515625, 15.7200927734375, 5.182769775390625, -20.810455322265625, 97.93844604492188, -51.55339050292969, 98.61199951171875, 44.63433837890625, 6.76123046875, -11.869415283203125, -14.055442810058594, 60.78070068359375, 19.288238525390625, 10.91241455078125, 52.18349838256836, 134.97393798828125, -10.134384155273438, 2.3851795196533203, 60.925567626953125, 27.681610107421875, -19.374868392944336, 16.46453857421875, -131.17080688476562, -0.32373809814453125, -21.967201232910156, 138.72296142578125, 76.79376220703125, 150.80560302734375, 66.66098022460938, 10.408309936523438, 1.10272216796875, 117.80490112304688, 116.02508544921875, 1.8822917938232422, 30.567359924316406, 99.3462142944336, 83.50920104980469, 114.22611999511719, 3.322540283203125, 84.78900146484375, 7.265584945678711, -14.035362243652344, 90.04824829101562, -1.5181884765625, 130.0245361328125, 81.42987060546875, 5.91448974609375, 82.47380065917969, 122.95906066894531, 72.20054626464844, 2.5220298767089844, 58.05845642089844, 32.06298065185547, 4.01148796081543, 134.9249725341797, 159.63687133789062, -1.8152008056640625, -0.1146392822265625, 14.80767822265625, 1.8798751831054688, 10.039642333984375, 26.788665771484375, 36.82395935058594, -2.99517822265625, -23.027557373046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000230.npy"}
{"epoch": 0.4816753926701571, "step": 231, "batch_size": 128, "mean": 30.924819946289062, "std": 54.551212310791016, "min": -111.893310546875, "p10": -38.905087280273435, "median": 24.815521240234375, "p90": 101.2783920288086, "max": 148.51300048828125, "pos_frac": 0.7578125, "sample": [85.59718322753906, 14.593231201171875, -41.5150146484375, -62.92402648925781, 90.9681167602539, 61.81211853027344, 50.1597900390625, -3.4483489990234375, 80.0667724609375, -111.893310546875, 106.78977966308594, -36.733978271484375, 62.571502685546875, 74.5274658203125, -5.237998962402344, 107.98849487304688, 9.647781372070312, 8.985118865966797, 65.55254364013672, 2.13360595703125, -24.886199951171875, -73.95457458496094, -13.241134643554688, 48.957244873046875, 10.3118896484375, 32.86212158203125, 12.68792724609375, 91.66059112548828, -0.5054931640625, -54.67529296875, 24.74761962890625, 92.69692993164062, 101.42642211914062, 3.3412818908691406, 45.94842529296875, -11.777206420898438, 127.75732421875, 44.63751220703125, 17.542556762695312, 42.6004638671875, 2.7113037109375, 41.1219482421875, -2.057769775390625, 1.4058761596679688, 112.31356811523438, 94.50973510742188, -5.10687255859375, 55.64674377441406, 41.25921630859375, 33.86073303222656, 92.90093994140625, -60.1851806640625, 21.40912628173828, 14.34716796875, 5.969970703125, 13.149932861328125, 94.7003173828125, -22.281646728515625, 37.533172607421875, 148.51300048828125, -72.84487915039062, 5.105167388916016, 4.3100738525390625, 44.53968811035156, 2.660268783569336, 92.000732421875, 49.13671875, 24.53533935546875, 21.15777587890625, 53.50917053222656, 26.663406372070312, 56.4239501953125, 141.1416015625, -27.28277587890625, -97.46626281738281, -0.957000732421875, 54.091712951660156, -10.05169677734375, 90.9263916015625, 28.68670654296875, 16.75103759765625, 21.719024658203125, 36.0338134765625, 70.60601806640625, 11.710651397705078, 30.237319946289062, 9.23065185546875, 109.15142822265625, 49.50105285644531, 8.032302856445312, 49.5028076171875, 72.91549682617188, 71.89202880859375, 23.260452270507812, -69.90147399902344, 72.78103637695312, 13.575000762939453, 74.7314453125, 145.3817138671875, 0.93743896484375, -66.57476806640625, 24.8834228515625, 21.055770874023438, 139.7216033935547, 88.53176879882812, -38.340179443359375, -2.0012588500976562, 34.46844482421875, 18.278610229492188, 74.33551788330078, -1.3690185546875, 52.64649200439453, 32.60296630859375, -92.0169677734375, 86.97618103027344, 139.84310913085938, 13.148651123046875, 115.18539428710938, 11.697158813476562, 63.34327697753906, -17.70830535888672, 1.90777587890625, 101.21495056152344, -40.22320556640625, 64.21150207519531, 0.0, -92.17985534667969, 120.90191650390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000231.npy"}
{"epoch": 0.4837696335078534, "step": 232, "batch_size": 128, "mean": 36.68476867675781, "std": 58.08219909667969, "min": -95.76153564453125, "p10": -28.970092391967768, "median": 23.773967742919922, "p90": 112.14546966552734, "max": 174.9351806640625, "pos_frac": 0.6875, "sample": [-23.8375244140625, 108.43124389648438, -2.9330673217773438, 123.76535034179688, 31.319000244140625, 77.77810668945312, 98.48056030273438, 48.495025634765625, -5.17352294921875, -7.5261383056640625, 92.83821105957031, 102.37969970703125, 20.130401611328125, -25.49748992919922, -41.17474365234375, 23.972389221191406, 7.676460266113281, 4.624427795410156, 21.63787841796875, 21.900802612304688, 28.27691650390625, 10.731346130371094, -95.76153564453125, 77.7874755859375, 100.1287841796875, 109.4097900390625, 23.066383361816406, 42.8385009765625, -1.2334632873535156, 141.69122314453125, 17.22528076171875, -7.391075134277344, -41.1669921875, 78.26077270507812, 82.1416015625, 101.79580688476562, -63.944580078125, 39.037109375, 19.799530029296875, 109.023193359375, 75.79502868652344, -27.893714904785156, -1.0216636657714844, 24.694740295410156, 11.443008422851562, 1.1851387023925781, 83.99061584472656, 11.200286865234375, 113.40318298339844, 0.5173263549804688, 30.853424072265625, -24.92767333984375, 97.54635620117188, 23.186111450195312, 4.8745269775390625, 54.04132080078125, 18.841796875, 101.67525482177734, -1.479156494140625, 89.57221984863281, 18.36034393310547, 139.388427734375, 111.90699768066406, -24.67822265625, 95.14617919921875, 100.89503479003906, -7.95257568359375, 32.164886474609375, -6.1862030029296875, -13.192901611328125, 10.405227661132812, 115.89968872070312, 11.507080078125, -43.728607177734375, 132.93667602539062, 128.35873413085938, -2.34912109375, -40.078216552734375, 50.96002197265625, 66.1719970703125, 1.0225410461425781, 9.447586059570312, -17.514404296875, 92.6981430053711, 112.701904296875, -19.30588150024414, -6.298553466796875, 80.4698486328125, -5.836090087890625, -37.1141357421875, 95.37708282470703, 73.55029296875, 0.0, 122.29397583007812, 41.651275634765625, 8.008121490478516, 57.513397216796875, 31.60887908935547, 106.1008529663086, 65.94171142578125, 95.32793426513672, 23.575546264648438, 134.75454711914062, -89.89752197265625, 94.35777282714844, 174.9351806640625, 11.72216796875, -22.39013671875, 101.515625, 75.39657592773438, -6.0531005859375, 119.60781860351562, -85.6406478881836, -14.017303466796875, -19.616844177246094, -52.07757568359375, 38.14813995361328, -31.481639862060547, 54.562103271484375, -49.524993896484375, 109.02859497070312, -5.236328125, -88.64491271972656, 136.05789184570312, 68.8970947265625, 83.70423126220703, 34.625244140625, -20.73834228515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000232.npy"}
{"epoch": 0.48586387434554973, "step": 233, "batch_size": 128, "mean": 38.04512023925781, "std": 58.07225799560547, "min": -98.22166442871094, "p10": -30.960017395019523, "median": 27.44054412841797, "p90": 117.64497680664063, "max": 172.420654296875, "pos_frac": 0.7890625, "sample": [-49.6495361328125, 101.60417938232422, 0.91925048828125, -1.5228271484375, 12.981964111328125, 113.01152038574219, 45.21931457519531, 35.89422607421875, 1.3226318359375, 15.097648620605469, 16.61004638671875, 32.713138580322266, 126.99362182617188, 68.09097290039062, 17.591171264648438, -0.969573974609375, 25.774017333984375, 117.69287109375, 41.71342468261719, 161.30001831054688, 103.06680297851562, 95.69713592529297, 107.80352783203125, 100.71867370605469, 69.9910888671875, 112.21440124511719, 102.24066162109375, 9.687667846679688, 97.54905700683594, -23.878929138183594, 26.800140380859375, -12.415138244628906, 62.40446472167969, 3.2882919311523438, 119.06010437011719, 37.16192626953125, 25.01708984375, -83.10904693603516, 82.40281677246094, 132.8673858642578, -58.465782165527344, -5.178062438964844, 4.650508880615234, 23.824951171875, -35.89512634277344, 148.87625122070312, 63.3458251953125, 18.828834533691406, 31.918426513671875, 16.995868682861328, 116.91165161132812, 140.751953125, 14.19696044921875, -17.511642456054688, 39.29547119140625, -10.782623291015625, 154.756591796875, 32.07550048828125, 105.88011169433594, 8.765869140625, 52.76019287109375, 101.159423828125, -45.879791259765625, 42.600704193115234, 9.2490234375, -28.844970703125, -37.096893310546875, 65.75314331054688, -73.18841552734375, -81.60650634765625, 81.80963134765625, 114.4444580078125, 127.68511962890625, 63.71746826171875, 149.92071533203125, 53.474456787109375, 18.382553100585938, 13.75262451171875, 14.907508850097656, 3.641357421875, 1.9007148742675781, 5.497802734375, 13.08160400390625, 49.544158935546875, -54.26792907714844, 15.432825088500977, 172.420654296875, 61.224822998046875, -13.840576171875, 51.8382568359375, 21.22071075439453, 6.3135528564453125, -79.04364013671875, 137.55697631835938, 23.523841857910156, -25.13299560546875, 1.5803718566894531, 90.90225219726562, -6.89263916015625, 106.87637329101562, -13.93182373046875, 5.350929260253906, -10.542129516601562, -98.22166442871094, 74.06427001953125, 117.62445068359375, 33.623779296875, 84.23619842529297, 33.6993408203125, -82.2789077758789, 0.756011962890625, 98.55548095703125, 52.676055908203125, -6.3228912353515625, 32.882080078125, 28.080947875976562, -41.30708312988281, 118.38009643554688, 56.511314392089844, 43.298919677734375, 58.139503479003906, 4.332468032836914, 3.80572509765625, 72.60629272460938, 67.09173583984375, 19.022125244140625, 26.07000732421875, 20.995635986328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000233.npy"}
{"epoch": 0.48795811518324606, "step": 234, "batch_size": 128, "mean": 39.879486083984375, "std": 68.00126647949219, "min": -124.42294311523438, "p10": -50.336082458496094, "median": 35.513671875, "p90": 122.25628967285155, "max": 191.68051147460938, "pos_frac": 0.7578125, "sample": [-84.40144348144531, -105.34774780273438, -62.286773681640625, 0.8968791961669922, 52.189300537109375, 121.57058715820312, 123.85626220703125, 47.371368408203125, -29.388519287109375, 143.4613037109375, 111.91490173339844, -112.09197998046875, 106.8814697265625, 94.08662414550781, -26.1685791015625, 104.494384765625, 37.919349670410156, 52.686981201171875, 85.90721130371094, 131.20956420898438, 91.52415466308594, 67.360595703125, 74.45700073242188, 137.71142578125, 13.47869873046875, 5.885368347167969, 103.20355987548828, 90.07695007324219, 96.5338363647461, 52.48602294921875, 22.028472900390625, 53.98992919921875, 108.77725219726562, 140.09637451171875, 93.8665771484375, 23.96033477783203, 35.21246337890625, 12.137451171875, 1.395477294921875, 59.88409423828125, 64.18817138671875, 19.834564208984375, 6.169891357421875, -75.12069702148438, 24.842620849609375, -23.203720092773438, 93.45672607421875, 92.58670043945312, -1.3584213256835938, 115.03064727783203, 10.953788757324219, 2.12353515625, 138.76568603515625, 69.6015625, -19.387100219726562, 164.01800537109375, 96.72433471679688, 85.37110137939453, 106.92837524414062, 37.0489501953125, 10.210813522338867, 7.089679718017578, 128.69070434570312, 11.87628173828125, 7.066356658935547, -1.308349609375, -34.25862121582031, 0.0, 26.198974609375, -14.224227905273438, 15.14788818359375, 27.922195434570312, -99.78559875488281, -3.2184371948242188, 103.48652648925781, 146.5357666015625, 106.75344848632812, 28.7666015625, -87.29714965820312, 54.184783935546875, -1.3124027252197266, 94.38160705566406, 119.88829040527344, -19.069549560546875, 57.258514404296875, 34.94194030761719, 88.9991455078125, 47.92578125, -124.42294311523438, 138.43817138671875, -80.47598266601562, 156.18804931640625, 19.90380859375, 22.717010498046875, 94.67694854736328, 2.6795196533203125, 96.86134338378906, 35.81488037109375, 1.2385177612304688, 9.228271484375, 100.20608520507812, 0.0, 25.849502563476562, 60.70478820800781, 191.68051147460938, 139.30328369140625, -112.54835510253906, -47.073516845703125, 120.5543212890625, 77.75537109375, 84.63399505615234, 4.82574462890625, -52.07545471191406, 22.47772216796875, 62.32232666015625, 112.31123352050781, -32.9658203125, -35.332244873046875, -49.59063720703125, 0.67236328125, 49.9949951171875, 8.969696044921875, -73.3636474609375, 10.181205749511719, -97.63919067382812, 110.05203247070312, -10.852752685546875, 116.42034912109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000234.npy"}
{"epoch": 0.4900523560209424, "step": 235, "batch_size": 128, "mean": 37.37578582763672, "std": 64.4922866821289, "min": -129.63018798828125, "p10": -35.25825729370117, "median": 30.598438262939453, "p90": 112.70286407470702, "max": 157.70407104492188, "pos_frac": 0.703125, "sample": [21.883392333984375, 6.27435302734375, -8.67742919921875, 1.44317626953125, 8.69921875, -124.958740234375, -61.39350891113281, -4.8490142822265625, 101.10977172851562, 98.13333129882812, 11.056846618652344, 150.97683715820312, 83.2843017578125, 52.62554931640625, 81.12541198730469, 13.519490242004395, 99.22478485107422, 113.4652099609375, -1.702096939086914, -7.00823974609375, 90.7835693359375, 23.4573974609375, 36.56982421875, 99.333740234375, 140.34075927734375, 138.26583862304688, 73.86651611328125, -19.027725219726562, 96.49005126953125, -6.511146545410156, 119.8291015625, 1.70306396484375, 81.680419921875, 105.24897003173828, 46.451995849609375, 101.78900909423828, 135.5076904296875, 107.86727905273438, 7.836353302001953, 150.19921875, -12.798736572265625, 64.10076904296875, -34.944068908691406, -35.991363525390625, 60.968170166015625, 136.0416259765625, -117.79769897460938, -23.611854553222656, -97.4979248046875, 73.2890625, 57.490875244140625, -19.562164306640625, -0.8357734680175781, -22.458465576171875, 33.76007080078125, 95.71078491210938, -19.819808959960938, 97.22535705566406, -10.61407470703125, -67.71829986572266, 10.680419921875, 20.172393798828125, 111.55372619628906, -20.499755859375, 149.03036499023438, 75.99607849121094, -9.83502197265625, -96.68926239013672, 7.939277648925781, 42.232696533203125, -21.49029541015625, 48.051727294921875, -69.22119140625, 98.23625183105469, 7.742656707763672, 10.435943603515625, 5.364837646484375, 120.76155090332031, 7.06195068359375, 56.5565185546875, 63.748199462890625, 21.323944091796875, -129.63018798828125, 31.881202697753906, 46.72943115234375, -11.482318878173828, 83.01103973388672, 106.69789123535156, 68.1512451171875, 4.4489593505859375, 151.017333984375, 105.80224609375, 74.69033813476562, 157.70407104492188, 114.68280029296875, 96.84600830078125, 37.70203399658203, 19.53057861328125, 106.24711608886719, -8.335420608520508, -1.2503585815429688, 107.96066284179688, -74.34819030761719, 65.58676147460938, -3.47509765625, 89.90087127685547, -4.517921447753906, 58.88134765625, 96.43023681640625, -7.366668701171875, -2.249176025390625, 28.26476287841797, -0.0579986572265625, -36.70849609375, 29.000083923339844, 15.224895477294922, 105.58663940429688, -127.33697509765625, -49.361328125, 4.49932861328125, 88.24287414550781, 85.72562408447266, 24.90679931640625, 112.37614440917969, 92.35440063476562, 40.41493225097656, 29.315673828125, 0.4024925231933594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000235.npy"}
{"epoch": 0.49214659685863876, "step": 236, "batch_size": 128, "mean": 42.724327087402344, "std": 59.30649948120117, "min": -126.9608154296875, "p10": -25.01144409179687, "median": 33.29423522949219, "p90": 113.21519927978515, "max": 162.08013916015625, "pos_frac": 0.78125, "sample": [62.90843200683594, 22.156982421875, -73.5618896484375, 66.57655334472656, 77.66647338867188, 95.4307861328125, 8.86444091796875, 108.82821655273438, -14.552536010742188, -40.425506591796875, -87.7987060546875, 25.944252014160156, -24.119903564453125, 67.06929016113281, 76.39923095703125, 110.04147338867188, -0.9359130859375, 95.68135070800781, 71.01768493652344, 17.511016845703125, 26.021942138671875, 89.04114532470703, 40.23944091796875, -58.822383880615234, 40.15582275390625, 97.54541015625, 96.6458740234375, 4.565282821655273, 70.91677856445312, 112.34982299804688, 18.66126251220703, 32.28399658203125, 73.70540618896484, 19.909339904785156, 14.728721618652344, 18.42491912841797, 14.00537109375, 139.45013427734375, 105.9052734375, 3.518035888671875, 8.318466186523438, 104.14240264892578, 110.89886474609375, 33.51762390136719, 144.33377075195312, 1.5678625106811523, -45.24296569824219, 4.9259033203125, 24.673202514648438, 0.0, -4.9306640625, 20.852752685546875, 98.0489501953125, 102.50570678710938, 11.317140579223633, -27.091705322265625, -106.68606567382812, 107.19303894042969, 11.206512451171875, -43.02178955078125, 84.32377624511719, 122.74749755859375, 54.743408203125, 72.92768096923828, 5.47601318359375, -35.74566650390625, 93.3883056640625, -126.9608154296875, 75.5556640625, 6.442222595214844, -16.582725524902344, 162.08013916015625, 59.91720199584961, 146.5899658203125, -0.041534423828125, -5.310981750488281, -47.504974365234375, -3.6236572265625, 114.94352722167969, 76.03370666503906, 58.64375305175781, 67.599609375, -49.11962890625, 88.68496704101562, 105.14697265625, 35.51165771484375, 121.04013061523438, 142.07733154296875, -6.9878387451171875, 82.59205627441406, 61.173858642578125, 111.41629028320312, 76.81005859375, 22.268585205078125, 2.401275634765625, 112.4744873046875, 105.38983154296875, 79.5496597290039, 92.96443939208984, 121.69146728515625, 106.64094543457031, 32.8994140625, 4.19134521484375, 2.756866455078125, 0.0, 44.99176025390625, 7.7587738037109375, 28.362625122070312, -104.36337280273438, 107.89329528808594, 6.9640960693359375, 0.3766136169433594, -9.386863708496094, -19.854156494140625, 46.98699951171875, 109.44905853271484, 126.25059509277344, 12.828689575195312, 8.53961181640625, -21.056442260742188, 125.71141052246094, 95.26315307617188, 122.19241333007812, 17.79144287109375, -2.6790771484375, 126.0577392578125, 7.865922927856445, 33.07084655761719], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000236.npy"}
{"epoch": 0.4942408376963351, "step": 237, "batch_size": 128, "mean": 36.030120849609375, "std": 61.8300666809082, "min": -134.63946533203125, "p10": -26.725165557861317, "median": 26.156036376953125, "p90": 120.11110610961914, "max": 198.20071411132812, "pos_frac": 0.734375, "sample": [51.45733642578125, 13.46929931640625, 34.291805267333984, 125.54458618164062, -10.410675048828125, 29.77752685546875, 80.04613494873047, 92.13335418701172, 40.917327880859375, 33.84779357910156, -2.664276123046875, 57.39836120605469, 145.705810546875, -77.06993865966797, -33.68779754638672, -8.054878234863281, 45.35186767578125, 0.4094696044921875, 198.20071411132812, 92.93946075439453, 83.44511413574219, 32.43632507324219, 116.15252685546875, 112.23284149169922, 11.616153717041016, 4.29248046875, 83.29093933105469, 0.0, 78.98524475097656, 91.42697143554688, -2.9466629028320312, 13.815673828125, -56.01300048828125, 45.80054473876953, -15.553237915039062, 38.32623291015625, 0.6914215087890625, 1.2393951416015625, -134.63946533203125, -18.460838317871094, -4.4474639892578125, 0.9036865234375, 5.108726501464844, 99.72892761230469, 8.156463623046875, 30.704376220703125, 120.79049682617188, 30.264892578125, 38.25062561035156, 2.2563037872314453, 0.8285064697265625, 152.25299072265625, 68.88440704345703, 58.918785095214844, 13.492050170898438, 87.69436645507812, 113.23410034179688, 63.605377197265625, 69.4825439453125, 24.997802734375, 128.05120849609375, 37.40779113769531, 26.242279052734375, 15.453826904296875, 63.880455017089844, -8.75244140625, 127.48052978515625, 112.99662780761719, -60.60328674316406, -64.39532470703125, 0.6121807098388672, 137.56378173828125, -86.65451049804688, 13.7445068359375, 109.95431518554688, -3.2115249633789062, -79.03451538085938, -8.273529052734375, 152.45912170410156, -21.204833984375, 15.246612548828125, 26.514602661132812, 123.4119873046875, 29.939727783203125, 123.74786376953125, -70.22268676757812, -3.3544845581054688, 3.66693115234375, 119.81993865966797, 26.069793701171875, 9.285255432128906, 148.89483642578125, -4.393924713134766, -1.097930908203125, -38.751220703125, 136.95123291015625, -61.51702880859375, 3.135772705078125, 114.74512481689453, 94.66960144042969, 21.8779296875, 11.284584045410156, 102.439453125, -2.7764205932617188, 2.2654953002929688, 2.396188735961914, 101.3714599609375, -23.741180419921875, -16.605499267578125, -13.257049560546875, 3.3717041015625, 117.3377685546875, -48.015380859375, 0.39548492431640625, 74.2323989868164, 43.26983642578125, -131.19949340820312, 69.53022766113281, 91.563720703125, 112.10941314697266, 104.8583984375, -17.512298583984375, 24.701385498046875, 89.49261474609375, -1.3320198059082031, 34.92547607421875, 18.52801513671875, 69.01470947265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000237.npy"}
{"epoch": 0.4963350785340314, "step": 238, "batch_size": 128, "mean": 39.97496032714844, "std": 64.2522964477539, "min": -98.3853759765625, "p10": -40.22531433105468, "median": 27.22174072265625, "p90": 125.35440292358398, "max": 173.15847778320312, "pos_frac": 0.671875, "sample": [-59.45585632324219, 2.8360671997070312, 5.77093505859375, 14.239494323730469, -53.709808349609375, 152.43896484375, 106.06784057617188, 46.46087646484375, 102.39581298828125, -50.73321533203125, -3.6002197265625, 87.43780517578125, -4.929229736328125, 17.513824462890625, 3.4595565795898438, -57.535213470458984, 130.92132568359375, 109.63568115234375, -2.5123462677001953, 44.88031005859375, 109.92027282714844, 76.27999877929688, 39.27825927734375, 119.69803619384766, 152.6723175048828, 17.53978729248047, -13.221435546875, 1.3773727416992188, 14.740692138671875, -90.55784606933594, -96.77250671386719, -95.58158874511719, 135.93084716796875, -80.37722778320312, -8.42791748046875, 134.40054321289062, 34.607330322265625, 103.3353271484375, -0.8746261596679688, 105.4525146484375, 149.87420654296875, 38.8411865234375, 35.662841796875, 162.4056396484375, 87.11822509765625, 119.46188354492188, 70.97549438476562, 62.28839111328125, -11.14410400390625, 37.859100341796875, 122.19511413574219, 159.08001708984375, 81.54266357421875, -3.21710205078125, 21.04718017578125, 16.15142822265625, -10.931015014648438, 7.860860824584961, -9.813583374023438, 81.89411926269531, -7.642021179199219, -1.1094970703125, -39.470855712890625, -98.3853759765625, 103.87345886230469, 85.625244140625, 91.46902465820312, 111.6180191040039, 87.9788818359375, 115.73363494873047, -50.25579833984375, -41.9857177734375, 36.685821533203125, -6.39501953125, -1.2757110595703125, 81.13198852539062, 131.33062744140625, -14.0865478515625, 90.25125122070312, 104.9808349609375, 89.70332336425781, 128.8414306640625, -55.941593170166016, -9.584609985351562, 7.2513427734375, 26.7342529296875, 97.95929718017578, -4.67289924621582, 64.30729675292969, 37.730743408203125, 0.31586456298828125, -13.418319702148438, -25.255844116210938, 26.8680419921875, 173.15847778320312, -9.02618408203125, 32.392242431640625, 83.5181884765625, 17.284278869628906, -0.1643524169921875, 27.026947021484375, 130.54196166992188, 123.8599624633789, 28.3778076171875, 38.88978576660156, 7.3343963623046875, -14.097259521484375, 115.80345153808594, 23.618743896484375, -51.53668212890625, 27.416534423828125, 118.57235717773438, 16.72882843017578, 14.254653930664062, 113.09783935546875, -12.962936401367188, -3.7111358642578125, 115.7811279296875, -11.595977783203125, -10.66827392578125, 115.08039855957031, 148.7652587890625, 34.756500244140625, -28.243896484375, 39.098724365234375, 7.7113037109375, 88.41836547851562, -3.75244140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000238.npy"}
{"epoch": 0.49842931937172774, "step": 239, "batch_size": 128, "mean": 34.216766357421875, "std": 64.47785186767578, "min": -154.83563232421875, "p10": -28.807611083984362, "median": 20.069847106933594, "p90": 129.33909606933594, "max": 174.50543212890625, "pos_frac": 0.71875, "sample": [-66.33131408691406, -58.133544921875, 144.43426513671875, 79.72686767578125, -13.738861083984375, 0.9737548828125, -154.83563232421875, 80.41397094726562, 139.41452026367188, 1.69573974609375, 96.76068115234375, -130.66168212890625, -104.66818237304688, -3.835113525390625, 94.84940338134766, 84.503662109375, 102.82125854492188, 130.21099853515625, 35.64203643798828, -12.271209716796875, -0.715484619140625, 119.62225341796875, 1.4412345886230469, 28.328842163085938, 8.004997253417969, 89.01321411132812, 0.5266284942626953, 105.40775299072266, 20.579833984375, -17.730728149414062, -44.64385986328125, 99.96432495117188, 84.02388763427734, -6.02130126953125, 150.93612670898438, -15.66278076171875, 25.308700561523438, 103.6500244140625, 174.50543212890625, 118.48876953125, 90.3160171508789, -4.693603515625, 26.88629150390625, 0.395782470703125, 153.42257690429688, -11.791709899902344, -14.6173095703125, 78.41705322265625, -79.16804504394531, 99.18405151367188, -76.01244354248047, 42.796451568603516, 18.46770477294922, 67.54202270507812, 38.68048095703125, -18.497802734375, 47.821990966796875, 3.0955123901367188, 65.23494720458984, -38.299652099609375, 17.6314697265625, 122.78996276855469, 8.287408828735352, 85.548583984375, 51.31633758544922, 29.547119140625, 104.4649429321289, 1.6908340454101562, 8.277618408203125, 146.4139404296875, 1.4289875030517578, 6.03497314453125, 13.433670043945312, 39.920989990234375, 98.52828216552734, -8.452880859375, 102.04153442382812, 20.359329223632812, -73.55581665039062, 17.575881958007812, -2.2635498046875, 13.136932373046875, 148.21217346191406, -19.55523681640625, -2.822509765625, 140.8939208984375, -24.739593505859375, 30.472068786621094, 10.3238525390625, -47.9827880859375, 115.57551574707031, -4.013713836669922, -11.45281982421875, 113.01139068603516, 5.40277099609375, 29.043445587158203, 8.368537902832031, 2.45123291015625, 143.0272216796875, 35.225189208984375, 0.4654998779296875, 143.8729248046875, 17.110855102539062, -16.938079833984375, 52.091949462890625, -4.406867980957031, -24.050704956054688, 28.435333251953125, 35.996856689453125, 133.86715698242188, 9.306900024414062, 38.771270751953125, 54.968292236328125, 147.6409912109375, 128.96542358398438, 0.0, 22.346084594726562, 0.7055473327636719, 113.35191345214844, 11.868072509765625, -64.06448364257812, 84.33116912841797, 19.780364990234375, 25.112945556640625, 34.69989013671875, -84.66305541992188, -1.9959564208984375, 15.399185180664062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000239.npy"}
{"epoch": 0.5005235602094241, "step": 240, "batch_size": 128, "mean": 43.45766067504883, "std": 59.14866638183594, "min": -117.08866882324219, "p10": -25.78734436035156, "median": 34.9007568359375, "p90": 120.04489440917969, "max": 165.66616821289062, "pos_frac": 0.734375, "sample": [7.852060317993164, 119.21182250976562, 70.60565185546875, 115.52688598632812, -55.695220947265625, 7.4745330810546875, 73.33370971679688, 38.166015625, 0.0, 139.07354736328125, 33.1356201171875, -26.58306884765625, -54.77171325683594, 4.7620849609375, -20.562034606933594, 103.4174575805664, 57.72010803222656, 79.6048583984375, 97.38207244873047, 62.374122619628906, 17.156097412109375, -0.39501953125, 85.97250366210938, 110.2882080078125, 19.48870849609375, 109.18881225585938, 88.8826904296875, -55.137725830078125, 120.68057250976562, 118.02294921875, 41.309722900390625, 3.57861328125, 141.29217529296875, 74.35692596435547, -13.29248046875, -17.110626220703125, 131.0919189453125, 92.55393981933594, 99.9151382446289, 12.826728820800781, 26.685256958007812, 46.09552001953125, 126.87767028808594, -2.0414199829101562, 85.26123046875, 50.94049072265625, 96.31607055664062, 29.3846435546875, 98.3096923828125, 38.8297119140625, 110.80361938476562, 74.73159790039062, 0.0, 83.29029083251953, -117.08866882324219, 27.59600830078125, 0.38165283203125, -5.8951568603515625, -0.3994140625, -4.832305908203125, 91.19618225097656, 103.09664916992188, -7.75152587890625, -13.603363037109375, 122.1029052734375, 110.59622192382812, 48.63494873046875, 93.13919830322266, 117.46101379394531, 37.194549560546875, 144.43768310546875, -31.2484130859375, 0.0, 18.5174560546875, -17.976638793945312, -11.193267822265625, 148.46966552734375, -3.0790252685546875, 15.053382873535156, 29.104095458984375, -11.095695495605469, 37.361968994140625, 100.41525268554688, 22.360275268554688, 90.15400695800781, 165.66616821289062, -29.32568359375, 160.23947143554688, -107.16534423828125, 133.95266723632812, 6.130424499511719, -25.446319580078125, 6.824249267578125, 103.27006530761719, 105.24276733398438, 53.61946105957031, 43.7564697265625, 84.30648803710938, 16.79512596130371, -10.204339981079102, -20.83441162109375, 7.805055618286133, 1.4245872497558594, 10.003952026367188, 106.89054870605469, 36.6658935546875, 86.57681274414062, 119.7724609375, 53.824371337890625, 11.20587158203125, 120.70795440673828, 11.34344482421875, 29.3017578125, -61.395416259765625, 29.63818359375, 4.093730926513672, -49.5382080078125, 83.03524780273438, -31.92254638671875, 104.42938232421875, 0.0, 13.299324035644531, 66.43061065673828, 13.36083984375, 21.97552490234375, -41.54460144042969, -40.6160888671875, 137.69216918945312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000240.npy"}
{"epoch": 0.5026178010471204, "step": 241, "batch_size": 128, "mean": 33.738006591796875, "std": 63.22780227661133, "min": -117.32626342773438, "p10": -37.702145385742185, "median": 17.574670791625977, "p90": 118.66668243408203, "max": 177.98007202148438, "pos_frac": 0.703125, "sample": [2.041839599609375, 119.70947265625, 89.38711547851562, 59.50433349609375, 18.377643585205078, -33.505035400390625, 22.793975830078125, 1.3749637603759766, 75.52963256835938, 84.697265625, 112.26531982421875, 0.0, -11.7496337890625, 127.17587280273438, 97.46879577636719, 151.62661743164062, 7.127227783203125, -1.93243408203125, 15.39764404296875, 125.84576416015625, 37.12833786010742, 109.46831512451172, -13.664688110351562, -14.412078857421875, 99.56416320800781, 16.502731323242188, -41.2708740234375, 20.737030029296875, 106.41159057617188, -10.461395263671875, 127.42266082763672, -37.0655517578125, 109.1190185546875, 40.3548583984375, 10.961944580078125, 2.055633544921875, -8.45257568359375, -9.624710083007812, 5.58544921875, 111.96722412109375, 11.694549560546875, 20.913604736328125, 101.50858306884766, 149.487548828125, 13.2890625, 86.41079711914062, 58.8060302734375, -115.74749755859375, 12.94586181640625, 25.21978759765625, 20.435546875, 2.1511878967285156, 3.0839195251464844, 96.15481567382812, -24.181243896484375, -11.274505615234375, 114.751708984375, -3.91302490234375, 20.326873779296875, -74.21565246582031, 35.317291259765625, 88.91818237304688, 57.90496826171875, 107.03460693359375, -6.392677307128906, 24.607696533203125, -14.113037109375, 63.58198547363281, 1.609649658203125, 104.34329223632812, -39.42466735839844, 6.65765380859375, 135.3649139404297, 91.7747802734375, -23.6282958984375, -53.305908203125, 130.37734985351562, 88.64446258544922, 4.4752960205078125, -0.984649658203125, 2.141571044921875, 0.7908477783203125, 151.20590209960938, 30.264251708984375, 32.428619384765625, 30.246383666992188, -39.187530517578125, 107.75843811035156, -6.1801300048828125, 118.21977233886719, 6.85809326171875, -75.63276672363281, -1.8953933715820312, 2.0034637451171875, 110.71499633789062, 92.12081909179688, 3.4222679138183594, 19.851890563964844, 128.8067626953125, -23.20526123046875, 89.42233276367188, -52.08013916015625, 20.3240966796875, 9.211837768554688, 115.422607421875, 114.59945678710938, -83.86175537109375, 6.488868713378906, 132.53115844726562, 67.40460968017578, -22.163330078125, -116.95049285888672, -33.558990478515625, -7.946876525878906, 92.92675018310547, -0.41973876953125, 143.00540161132812, 53.9102783203125, -72.4383544921875, 71.80184936523438, 177.98007202148438, 4.1548004150390625, 16.771697998046875, -117.32626342773438, 10.91314697265625, -22.60003662109375, 48.55194091796875, -54.417205810546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000241.npy"}
{"epoch": 0.5047120418848168, "step": 242, "batch_size": 128, "mean": 40.18089294433594, "std": 58.061767578125, "min": -125.45869445800781, "p10": -20.767485046386714, "median": 41.21026611328125, "p90": 118.8853302001953, "max": 151.10986328125, "pos_frac": 0.765625, "sample": [102.87648010253906, 11.866050720214844, 75.17523193359375, 10.782184600830078, 17.283645629882812, 59.94508361816406, 33.68438720703125, 82.09420776367188, -27.53668212890625, 11.487548828125, 128.63638305664062, 95.22540283203125, 44.84004211425781, -16.253952026367188, -105.04984283447266, 76.37642669677734, 64.53111267089844, 107.20670318603516, -125.45869445800781, 44.52223587036133, 9.876394271850586, 140.72189331054688, 130.60931396484375, 39.44029998779297, 54.3330078125, 144.92333984375, 3.5851287841796875, 137.44256591796875, -3.009735107421875, 4.354867935180664, 86.0159912109375, 105.51116943359375, 1.0008544921875, 132.6893310546875, 40.87841796875, 147.68582153320312, 128.99403381347656, 1.3664741516113281, -99.86711883544922, -0.5778923034667969, 99.83843994140625, -18.80584716796875, -0.90228271484375, 57.74053955078125, -23.652389526367188, -3.127927780151367, 32.49530029296875, 91.30990600585938, 57.147335052490234, 2.38226318359375, -7.78759765625, -9.195751190185547, -2.2084197998046875, 57.08668518066406, -8.008949279785156, -2.1383705139160156, 107.03379821777344, 12.79425048828125, -14.856201171875, 101.13125610351562, 51.5802001953125, 0.32324981689453125, -27.83203125, 124.69607543945312, 7.552886962890625, -2.37548828125, 151.10986328125, 70.1845703125, 75.64352416992188, 23.922134399414062, 20.802108764648438, 58.1964111328125, -24.319854736328125, 15.033439636230469, 90.2244873046875, 10.580833435058594, 91.43999481201172, 16.064716339111328, 16.34326171875, -73.9239730834961, 65.0538101196289, 87.56571960449219, 24.3753662109375, 52.150115966796875, -70.75469970703125, 15.473922729492188, 3.8301219940185547, 51.771812438964844, 41.5421142578125, 90.83866882324219, 147.05886840820312, 4.367156982421875, 64.3111343383789, 29.694599151611328, -3.7134323120117188, 19.80938720703125, 60.374481201171875, 107.13815307617188, 79.17461395263672, 16.0645751953125, 0.9093399047851562, 43.433258056640625, -27.205005645751953, -16.066482543945312, 106.18991088867188, 55.21277618408203, 43.3354606628418, -118.22320556640625, 76.03176879882812, 49.85459899902344, 36.93328857421875, 8.860931396484375, 118.06185913085938, 132.75148010253906, 83.26492309570312, 87.75662231445312, 120.8067626953125, 60.35431671142578, -56.368629455566406, 93.60198974609375, -13.63311767578125, 55.139251708984375, 12.356246948242188, -49.295013427734375, 44.33259582519531, 101.66671752929688, -19.531097412109375, 104.76577758789062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000242.npy"}
{"epoch": 0.506806282722513, "step": 243, "batch_size": 128, "mean": 46.12397766113281, "std": 58.34880447387695, "min": -118.55059814453125, "p10": -19.086360168457027, "median": 28.806442260742188, "p90": 125.35938568115233, "max": 168.91943359375, "pos_frac": 0.75, "sample": [-48.431976318359375, -23.283660888671875, 28.5643310546875, 73.39793395996094, 115.06982421875, 78.409912109375, 26.31146240234375, 134.6042938232422, -75.13995361328125, 97.90658569335938, 25.5244140625, 146.8218994140625, 25.988195419311523, -26.078857421875, 55.876220703125, -12.471542358398438, -33.2608642578125, 85.44125366210938, 137.40679931640625, -47.90355682373047, -2.9135208129882812, 19.79888916015625, -21.151763916015625, 0.936553955078125, 69.51741027832031, 121.01104736328125, 168.91943359375, 108.9547119140625, -1.38214111328125, 51.88214111328125, 3.4693775177001953, 96.58879089355469, 6.636810302734375, 25.553024291992188, 112.044677734375, 13.5328369140625, 124.87864685058594, 15.096160888671875, 69.75633239746094, 132.69656372070312, -18.201187133789062, 103.91751098632812, 59.859375, 65.68267822265625, 82.57298278808594, 107.88986206054688, 128.523193359375, -53.584869384765625, 54.777099609375, 66.77127838134766, 72.14645385742188, -118.55059814453125, 3.395885467529297, 40.91668701171875, -12.84674072265625, 7.206298828125, 23.613021850585938, 21.22967529296875, 15.39801025390625, -2.38763427734375, 149.2166748046875, 55.646759033203125, 94.4344253540039, -12.574172973632812, -7.41168212890625, 94.38206481933594, 7.00762939453125, 107.94259643554688, 92.16261291503906, -7.120849609375, 27.054718017578125, -28.999542236328125, -12.9239501953125, 120.46559143066406, 91.47592163085938, 1.35052490234375, -6.4229736328125, 65.85044860839844, 11.305900573730469, 20.008132934570312, -10.797880172729492, 103.40753936767578, -26.37061309814453, 79.8043212890625, 10.33056640625, 0.0, 1.9988937377929688, 67.00532531738281, 103.34139251708984, 109.51278686523438, 27.950904846191406, 84.32876586914062, 68.81710815429688, -5.804466247558594, 119.28158569335938, 126.48110961914062, 130.31332397460938, 17.047943115234375, 3.4799652099609375, 35.85919189453125, 24.2666015625, -2.5223388671875, -4.6562042236328125, 29.048553466796875, -17.593475341796875, 123.822265625, 140.0555419921875, 98.22727966308594, 5.018714904785156, 30.079498291015625, 11.226253509521484, 9.213493347167969, 114.6501235961914, -14.777252197265625, -41.26646423339844, 159.75796508789062, 20.081703186035156, 119.10836791992188, -9.31378173828125, 26.941574096679688, 71.71307373046875, 95.38119506835938, 155.34823608398438, 101.00738525390625, 141.82310485839844, 49.24609375, 54.222503662109375, -23.987457275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000243.npy"}
{"epoch": 0.5089005235602094, "step": 244, "batch_size": 128, "mean": 47.40515899658203, "std": 63.5877685546875, "min": -139.574951171875, "p10": -20.440335083007806, "median": 41.349647521972656, "p90": 128.81532287597656, "max": 215.60311889648438, "pos_frac": 0.7421875, "sample": [-139.574951171875, 50.321807861328125, 41.935455322265625, -2.34820556640625, 28.432693481445312, 15.708393096923828, 118.88775634765625, -31.363525390625, 150.00958251953125, 11.966880798339844, 106.50616455078125, 52.55035400390625, 41.25627136230469, -6.9079742431640625, 11.278923034667969, 19.548141479492188, 92.23919677734375, 92.81886291503906, 93.59107971191406, 76.378173828125, 21.428115844726562, 108.6728744506836, -3.7507400512695312, -6.900177001953125, 114.98513793945312, 1.21923828125, -33.423492431640625, -66.06340026855469, 117.05419158935547, -24.694061279296875, 8.61810302734375, -17.757965087890625, -5.265800476074219, -6.064140319824219, -104.4677734375, 11.850227355957031, 151.48455810546875, -7.124603271484375, 95.67840576171875, 115.62409973144531, 4.1590728759765625, 54.686279296875, 134.97421264648438, 82.0478744506836, 141.728759765625, 47.749176025390625, 0.243255615234375, 80.15679931640625, 31.824417114257812, 84.57603454589844, 51.710205078125, 113.11187744140625, 0.0, 98.26885986328125, -18.6173095703125, -47.957611083984375, -56.514404296875, 102.0745849609375, -2.128173828125, -7.558280944824219, 128.44631958007812, -29.048828125, 4.575164794921875, 87.81658172607422, 5.723773956298828, 129.67633056640625, 123.03752136230469, 138.55191040039062, 215.60311889648438, 16.72003173828125, 55.15118408203125, 79.20295715332031, -39.5234375, 69.784912109375, 33.817596435546875, 0.6253204345703125, 134.73272705078125, -10.85888671875, 140.155517578125, 89.97659301757812, -1.670623779296875, 67.92520141601562, 12.463211059570312, -62.31373596191406, 22.79833984375, 85.44091796875, -3.401214599609375, 113.48178100585938, 104.18034362792969, 4.87530517578125, 4.6209259033203125, -2.021484375, 41.443023681640625, 72.9066162109375, 147.745849609375, 130.13192749023438, 120.41932678222656, 4.1323699951171875, 159.3958740234375, 96.86430358886719, -0.09963226318359375, 97.62704467773438, 39.86383056640625, 96.85906219482422, -11.134292602539062, 101.20516967773438, 24.818161010742188, 16.0093994140625, 127.9275131225586, 75.76081848144531, -24.94989013671875, 70.231689453125, 3.6220245361328125, 114.47409057617188, 68.85006713867188, 153.5369873046875, 0.07943534851074219, 86.5628890991211, 88.86528778076172, -10.715499877929688, 10.822677612304688, 16.830551147460938, 105.94434356689453, -110.55413818359375, 30.313934326171875, 89.15090942382812, 124.2642822265625, -0.76507568359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000244.npy"}
{"epoch": 0.5109947643979058, "step": 245, "batch_size": 128, "mean": 47.51504135131836, "std": 64.22496032714844, "min": -135.9349822998047, "p10": -23.865673828124994, "median": 46.10694122314453, "p90": 122.08076934814451, "max": 170.2431640625, "pos_frac": 0.78125, "sample": [13.2628173828125, -13.734832763671875, 35.09678649902344, 68.44507598876953, 3.429931640625, 28.350067138671875, 92.408203125, 4.541259765625, 14.51300048828125, 102.39205932617188, 38.61128234863281, -30.86419677734375, 16.28839111328125, 136.86563110351562, 89.71269226074219, 16.8233642578125, 157.0797882080078, 148.4700927734375, 18.878005981445312, 170.2431640625, -135.9349822998047, -28.79296875, -65.5914306640625, -27.93365478515625, 46.87461853027344, 83.51663208007812, 0.0, 35.1595458984375, 84.14531707763672, -6.797271728515625, 98.67086791992188, 101.02153015136719, 69.3907470703125, 117.1578369140625, 3.8927001953125, 150.79144287109375, 3.1567516326904297, 45.339263916015625, 97.95931243896484, 67.98153686523438, 138.18130493164062, 12.497894287109375, 77.64875793457031, 105.02365112304688, 11.2730712890625, 100.38192749023438, 31.24029541015625, 45.053436279296875, 53.04142379760742, -1.62078857421875, -22.12225341796875, 49.832252502441406, 8.983123779296875, 119.81735229492188, 16.855316162109375, -6.395660400390625, 92.02252197265625, 50.02325439453125, 58.3221435546875, 127.36207580566406, 152.28851318359375, 6.934772491455078, -5.4328460693359375, 75.28385925292969, 49.33912658691406, 105.6644515991211, 14.981315612792969, -42.92921447753906, 54.683349609375, -92.59890747070312, 72.03924560546875, 106.05449676513672, 87.0511474609375, 91.5191650390625, 106.26375579833984, -13.284576416015625, -71.728759765625, 33.64673614501953, 167.4039306640625, 119.81045532226562, 7.088075637817383, 112.40279388427734, -89.25338745117188, 7.20147705078125, 10.61968994140625, -59.03175354003906, 4.68927001953125, 118.14480590820312, 104.11611938476562, 109.30351257324219, 65.78376770019531, 23.121002197265625, -2.5528640747070312, 149.88137817382812, 62.18206787109375, 166.57589721679688, 32.001251220703125, 52.427490234375, 56.50891876220703, 90.48345947265625, -92.35627746582031, 117.16159057617188, 85.22030639648438, -18.59552001953125, -4.522401809692383, 85.51385498046875, 33.323455810546875, 111.20428466796875, 82.79010009765625, 113.9959716796875, 9.863067626953125, -15.8021240234375, 42.0367431640625, 41.10076904296875, 103.91261291503906, 21.25634765625, -2.358306884765625, -21.19622802734375, 146.30715942382812, 2.180042266845703, 117.62826538085938, 20.179489135742188, 76.40948486328125, 168.410888671875, -130.32986450195312, -45.203758239746094, 0.0, 76.87106323242188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000245.npy"}
{"epoch": 0.5130890052356021, "step": 246, "batch_size": 128, "mean": 31.371074676513672, "std": 66.22015380859375, "min": -131.48971557617188, "p10": -40.45327606201171, "median": 14.899314880371094, "p90": 130.1616241455078, "max": 167.35272216796875, "pos_frac": 0.640625, "sample": [80.24813079833984, 99.38958740234375, 89.99178314208984, -4.31268310546875, 75.286376953125, -5.389945983886719, 167.35272216796875, 11.088165283203125, 18.56155776977539, 56.65327453613281, -25.88375473022461, 89.49308776855469, 24.910552978515625, 19.549209594726562, -68.36609649658203, -29.259170532226562, -14.127639770507812, 140.51792907714844, 142.23489379882812, -59.307151794433594, 122.00126647949219, 151.787353515625, 21.49591064453125, 104.14183044433594, 42.69085693359375, -3.5376739501953125, 9.5106201171875, -61.77197265625, 24.825439453125, -6.0909423828125, 43.022705078125, -32.991912841796875, 129.56005859375, 45.573341369628906, 13.640533447265625, -10.646881103515625, -116.25965881347656, 126.51123046875, 4.3829345703125, 0.5813331604003906, 14.86773681640625, -76.9691162109375, -22.865631103515625, 128.15762329101562, -43.72355651855469, -19.91131591796875, -74.16456604003906, -21.186370849609375, 5.42889404296875, 92.79769897460938, 25.0179443359375, 149.49462890625, -35.80328369140625, 19.250732421875, 135.5157470703125, -14.463827133178711, 32.66387939453125, -24.896209716796875, -131.48971557617188, 27.629531860351562, 143.25564575195312, 77.63299560546875, 85.22512817382812, 62.177001953125, -39.051727294921875, 80.51138305664062, 144.90780639648438, 84.80081176757812, -15.020660400390625, 142.71722412109375, 138.6207275390625, 33.4036865234375, -57.96014404296875, -1.170074462890625, 21.99237060546875, 1.3702392578125, 112.12896728515625, 15.10589599609375, -30.3026123046875, -66.35028076171875, -49.216827392578125, -8.287200927734375, 104.7330322265625, 80.39288330078125, -27.921951293945312, -5.9141998291015625, -91.99746704101562, -75.6070556640625, 13.07794189453125, 7.735832214355469, 59.403411865234375, 128.41065979003906, 89.79293823242188, -38.90020751953125, 86.07381439208984, 27.68785858154297, 0.7592315673828125, -11.552337646484375, 0.0, 120.28375244140625, 133.3810577392578, 109.844482421875, -17.677780151367188, -14.217582702636719, 103.4949951171875, 0.6977691650390625, 2.77325439453125, 92.60067749023438, 4.607555389404297, 75.00813293457031, 25.80402374267578, 5.350189208984375, 123.4063720703125, 78.7257080078125, 14.930892944335938, 131.56527709960938, -5.816257476806641, -36.09600830078125, -22.047607421875, -20.75439453125, 133.1986083984375, -7.6771240234375, 1.8115653991699219, 44.07463073730469, 4.8961029052734375, 11.51800537109375, -5.3416900634765625, 122.07987976074219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000246.npy"}
{"epoch": 0.5151832460732985, "step": 247, "batch_size": 128, "mean": 41.71050262451172, "std": 61.60791778564453, "min": -137.180908203125, "p10": -25.709154510498045, "median": 29.070510864257812, "p90": 126.71336517333984, "max": 183.24542236328125, "pos_frac": 0.734375, "sample": [118.35714721679688, -28.95654296875, 87.81866455078125, 78.95005798339844, -16.009902954101562, 68.42499542236328, 124.72833251953125, 130.25257873535156, 109.31072998046875, 73.5978012084961, 0.0, 99.94293975830078, 126.33155822753906, 18.747283935546875, 7.52276611328125, 38.384857177734375, -25.587135314941406, 6.533782958984375, -29.958831787109375, 97.75, -10.471832275390625, 127.604248046875, -25.993865966796875, -90.75669860839844, 62.001190185546875, 98.82025146484375, 44.35797119140625, -4.12646484375, 102.17306518554688, 129.0616455078125, 73.96353149414062, 116.3141098022461, 14.722335815429688, 156.1907958984375, 80.52558898925781, -29.599609375, -0.8482666015625, 143.44461059570312, -7.7464599609375, 9.023880004882812, -10.304679870605469, -18.24298858642578, 26.78375244140625, -47.247528076171875, 11.379730224609375, 70.39739990234375, -2.0032615661621094, 37.384521484375, 3.1996688842773438, -68.14898681640625, 7.670879364013672, 95.258544921875, 67.87763214111328, 73.33572387695312, -33.69569396972656, 115.6968765258789, 160.824462890625, 18.353927612304688, 3.37872314453125, -14.325790405273438, -17.39427947998047, -7.604454040527344, -3.3239288330078125, 137.08889770507812, -0.3116455078125, 1.980133056640625, 77.17549133300781, -98.28579711914062, 14.740997314453125, -24.684356689453125, 27.562881469726562, 172.19393920898438, -137.180908203125, 2.1017684936523438, 115.15625, 12.991867065429688, 16.7852783203125, 4.6215057373046875, 34.516021728515625, 108.59440612792969, 38.823577880859375, 80.60661315917969, 183.24542236328125, 13.609466552734375, -28.6951904296875, -7.651947021484375, 3.2948074340820312, 42.396514892578125, 103.48619079589844, 30.749176025390625, 29.970794677734375, 95.80270385742188, 78.7210693359375, 26.610870361328125, 16.557327270507812, 104.07487487792969, 131.42230224609375, 97.32643127441406, 147.14608764648438, -17.097076416015625, 31.401382446289062, 14.901008605957031, 59.50860595703125, 43.318359375, 96.9561538696289, 91.07624816894531, 28.17022705078125, -31.78173828125, 137.6759033203125, 11.833740234375, -102.091552734375, 101.66776275634766, 88.84732818603516, -19.60986328125, 25.293277740478516, 30.819589614868164, 78.98233032226562, -11.37826919555664, 7.56854248046875, -3.36016845703125, 88.24678039550781, 38.0643310546875, 132.2826385498047, 4.114980697631836, 110.01919555664062, 7.9602203369140625, 26.0784912109375, 72.88092041015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000247.npy"}
{"epoch": 0.5172774869109947, "step": 248, "batch_size": 128, "mean": 32.29326629638672, "std": 65.02127075195312, "min": -156.06741333007812, "p10": -54.316244506835936, "median": 24.014625549316406, "p90": 126.45941162109375, "max": 184.5484619140625, "pos_frac": 0.734375, "sample": [12.736858367919922, 127.11907958984375, -68.00506591796875, 13.286949157714844, 148.11688232421875, -21.2342529296875, 23.6243896484375, 2.86029052734375, 75.30438232421875, 11.470649719238281, 27.05910873413086, 41.72010803222656, 82.6259765625, 25.41574478149414, 0.6477508544921875, 34.500335693359375, 141.99102783203125, 25.441314697265625, -1.51995849609375, 97.06265258789062, -7.309089660644531, 23.195953369140625, 88.74017333984375, 42.572471618652344, 138.03396606445312, 38.986083984375, 51.793155670166016, 67.0723876953125, 2.00213623046875, 94.99395751953125, 21.815528869628906, -20.216285705566406, 150.35549926757812, -16.230804443359375, 6.263572692871094, 36.971343994140625, 128.84121704101562, -156.06741333007812, 44.896034240722656, -3.664327621459961, -34.00927734375, 3.1180572509765625, 24.281166076660156, -27.715728759765625, -11.160816192626953, 39.077423095703125, -0.5578880310058594, -117.94696044921875, 28.0260009765625, 54.54840087890625, 17.929046630859375, 145.7498321533203, 34.80570983886719, 23.101999282836914, 162.4210205078125, -60.699249267578125, 129.3231201171875, 0.6394405364990234, -35.044525146484375, 137.32110595703125, 46.095123291015625, 104.24913787841797, -22.1177978515625, -85.07080078125, 17.251602172851562, 12.73651123046875, 184.5484619140625, 126.35943603515625, 84.69291687011719, 93.36453247070312, 90.63436126708984, -56.249298095703125, 1.7159194946289062, 72.44400024414062, 1.5537109375, -104.64164733886719, 0.0, 29.441009521484375, 105.85693359375, 124.493896484375, 114.8460693359375, 79.77780151367188, 88.00419616699219, 78.5322265625, 7.626066207885742, 68.14602661132812, -26.94464111328125, -103.284912109375, -9.207672119140625, 10.819137573242188, -68.35871887207031, 13.72772216796875, 126.69268798828125, 104.16256713867188, 119.54289245605469, -91.62661743164062, 36.162689208984375, 3.2107315063476562, 7.4275970458984375, 111.82656860351562, 1.4511260986328125, -53.48779296875, 40.085845947265625, -15.0743408203125, 65.79251098632812, 30.1224365234375, 101.18731689453125, 8.499557495117188, 8.848457336425781, -74.73756408691406, 23.96881103515625, 23.19317626953125, 22.15447235107422, -0.7451934814453125, 153.83255004882812, 89.58819580078125, 76.17691040039062, 57.62104797363281, 63.92991638183594, -81.1473388671875, -27.276039123535156, -5.687248229980469, 24.060440063476562, 24.6287841796875, 96.63653564453125, -71.76971435546875, -9.651168823242188, 16.418182373046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000248.npy"}
{"epoch": 0.5193717277486911, "step": 249, "batch_size": 128, "mean": 39.986228942871094, "std": 65.00186157226562, "min": -141.49880981445312, "p10": -32.155838012695305, "median": 27.36370086669922, "p90": 119.90782318115234, "max": 174.12969970703125, "pos_frac": 0.703125, "sample": [-14.055755615234375, 6.154628753662109, -58.93736267089844, -93.209228515625, 37.054847717285156, 4.73736572265625, -15.971572875976562, 16.75537109375, -5.75628662109375, 26.86865234375, -1.8537826538085938, -70.15613555908203, 103.52169799804688, 149.28863525390625, -11.593490600585938, 93.11471557617188, 41.849700927734375, 119.2947998046875, 24.201812744140625, 108.87002563476562, 15.706130981445312, -141.49880981445312, 117.47962188720703, 17.48828125, -35.686920166015625, 154.8465576171875, 82.6219482421875, -15.1676025390625, 38.825439453125, 72.30319213867188, 21.217430114746094, -7.705902099609375, 39.25335693359375, -14.211387634277344, -7.68994140625, 35.63079833984375, 133.39013671875, 63.84392547607422, 96.77432250976562, -76.46421813964844, 121.33821105957031, 112.316162109375, -11.780517578125, 119.13356018066406, 159.7579345703125, 36.5289306640625, -99.55583190917969, 102.55618286132812, 60.106964111328125, 5.7469482421875, -13.961883544921875, 15.035255432128906, 173.120849609375, 86.02722930908203, 140.365966796875, -1.96966552734375, 14.5523681640625, 78.41414642333984, -29.737060546875, 90.91204833984375, 36.853790283203125, 0.57061767578125, -0.753387451171875, 3.176361083984375, 104.91851806640625, 9.497230529785156, 99.74838256835938, 3.991283416748047, 85.58517456054688, 99.54551696777344, -21.428306579589844, -56.109153747558594, 27.76214599609375, 106.2523193359375, 45.706573486328125, 89.83612060546875, 174.12969970703125, 143.483154296875, 112.69259643554688, 111.36150360107422, 153.15863037109375, 86.97330474853516, 19.92858123779297, -68.79498291015625, 96.85403442382812, 144.31854248046875, -6.0970458984375, -30.64251708984375, -20.890880584716797, 99.8916015625, 22.698532104492188, -56.92822265625, 81.04547119140625, 98.31234741210938, 26.965255737304688, 103.06375122070312, 40.971343994140625, 26.207382202148438, -5.2677154541015625, -42.97125244140625, 128.41790771484375, 3.033966064453125, 24.532970428466797, 87.64935302734375, 64.81874084472656, 5.59857177734375, -5.717277526855469, -82.54254150390625, -40.30126953125, -12.508544921875, -28.828460693359375, 12.861160278320312, 54.402435302734375, 2.0448150634765625, 17.25787353515625, -4.70343017578125, 87.90621948242188, 38.153839111328125, 108.79512023925781, 2.566650390625, 112.37513732910156, 104.04763793945312, 106.0169677734375, -28.872787475585938, -3.9884872436523438, 40.361846923828125, 28.52140235900879, 140.6784210205078], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000249.npy"}
{"epoch": 0.5214659685863874, "step": 250, "batch_size": 128, "mean": 41.47142028808594, "std": 70.37757110595703, "min": -123.40924072265625, "p10": -32.683859252929686, "median": 36.76850128173828, "p90": 134.38750610351562, "max": 193.5834197998047, "pos_frac": 0.7265625, "sample": [48.522369384765625, 14.061431884765625, -63.27690124511719, 127.3682861328125, 104.875, 174.29470825195312, 162.580078125, 94.2642822265625, 41.993988037109375, -100.14974212646484, 100.20507049560547, 22.79441261291504, 138.6689453125, 32.641571044921875, 20.7908935546875, 139.55126953125, 90.21035766601562, -11.18194580078125, 132.96307373046875, 109.347900390625, 131.8421173095703, 45.0882568359375, 39.87615966796875, 108.93324279785156, -32.269683837890625, 1.41522216796875, -123.40924072265625, 22.901153564453125, -7.1967926025390625, -0.15740966796875, 137.711181640625, -98.60113525390625, 50.24127197265625, 169.75033569335938, -9.221794128417969, 8.950927734375, 64.34796142578125, 2.0035877227783203, 140.21728515625, -115.84890747070312, -2.3118038177490234, 69.97103881835938, 127.22955322265625, 31.08777618408203, 100.81201171875, 83.32125854492188, -31.89105224609375, -97.11773681640625, -1.6680583953857422, 80.91122436523438, 98.20304870605469, 11.506736755371094, 49.868927001953125, 6.25714111328125, 35.89605712890625, 12.3270263671875, -8.678634643554688, -33.6502685546875, 68.3102035522461, 14.639022827148438, 115.87689208984375, 140.6717071533203, 97.556884765625, -0.8794002532958984, 30.542015075683594, 41.90887451171875, 160.17837524414062, 109.65074157714844, -4.399261474609375, 45.04315185546875, 91.75162506103516, 105.97914123535156, -15.822418212890625, 127.69371032714844, 37.11749267578125, -25.635787963867188, 68.87521362304688, 118.21675109863281, 141.00570678710938, 36.41950988769531, 27.374357223510742, -25.618011474609375, 39.23828125, -27.0513916015625, 119.85482788085938, 23.494251251220703, -2.9777679443359375, -116.23564147949219, 105.93728637695312, 147.9136962890625, -4.37628173828125, 193.5834197998047, -75.06417846679688, -66.6300277709961, -113.10772705078125, -0.2531280517578125, 7.553558349609375, -74.08973693847656, -101.3988037109375, 109.66389465332031, 81.67448425292969, 108.3140640258789, 66.659912109375, 91.92019653320312, 96.4274673461914, 45.493309020996094, -28.19842529296875, 44.78277587890625, 17.073741912841797, -2.541563034057617, 5.211639404296875, 86.52894592285156, 159.09898376464844, -13.384597778320312, 42.20208740234375, 91.13165283203125, 109.76155090332031, 6.4925079345703125, 7.24725341796875, 2.3938064575195312, -27.388038635253906, 98.55903625488281, 44.149139404296875, 2.4350738525390625, 8.779693603515625, 25.102874755859375, 13.439544677734375, 3.290760040283203], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000250.npy"}
{"epoch": 0.5235602094240838, "step": 251, "batch_size": 128, "mean": 41.48546600341797, "std": 65.26004791259766, "min": -137.62515258789062, "p10": -28.129083251953123, "median": 28.229034423828125, "p90": 129.02965087890624, "max": 255.7861328125, "pos_frac": 0.734375, "sample": [86.86117553710938, 73.17227172851562, 24.953399658203125, -15.460105895996094, -56.063568115234375, -46.6817626953125, -73.47296142578125, 59.046241760253906, 4.6127777099609375, 21.844329833984375, 9.182435989379883, 62.330963134765625, 115.21615600585938, 88.79639434814453, 5.305450439453125, -4.451404571533203, 33.817047119140625, 27.23406982421875, 5.576988220214844, 147.2906494140625, -1.2921295166015625, 6.7906494140625, -30.529052734375, 121.543212890625, 98.63232421875, 133.53067016601562, -61.46551513671875, 119.0684814453125, 144.32470703125, 23.017745971679688, 112.3572998046875, 1.7223052978515625, -24.35253143310547, 46.131591796875, 29.913665771484375, -73.08938598632812, 65.59375, -1.3149871826171875, 0.516754150390625, -20.930824279785156, 99.23304748535156, 38.989044189453125, 111.48199462890625, 133.789306640625, 133.80189514160156, -17.918212890625, -0.897705078125, -48.841064453125, 3.612943649291992, 5.4474334716796875, -32.05974578857422, 17.8914794921875, 255.7861328125, 35.17059326171875, 85.9678726196289, 106.14200592041016, 28.3145751953125, 121.27134704589844, 0.5309371948242188, 78.15909576416016, 93.9320297241211, -137.62515258789062, 24.025554656982422, 98.11712646484375, 32.2008056640625, -15.4884033203125, 9.66363525390625, -4.100589752197266, 24.131591796875, -1.0573196411132812, 112.63531494140625, 96.1092529296875, 12.19952392578125, 4.575267791748047, 51.773101806640625, 33.419830322265625, 0.35831451416015625, 28.14349365234375, -7.351280212402344, 89.05656433105469, 2.252288818359375, 68.68145751953125, 2.6330881118774414, -0.3204345703125, 87.44425964355469, 99.99100494384766, 131.4710693359375, -96.20271301269531, -1.86431884765625, 4.7669830322265625, 72.31412506103516, 130.46588134765625, 115.62799072265625, 69.69729614257812, 155.08169555664062, -9.936702728271484, 203.207275390625, 128.41412353515625, 96.90178680419922, -4.639942169189453, 71.43685913085938, 33.31439208984375, 72.63348388671875, 40.980438232421875, 85.8656005859375, 34.0076904296875, 122.67756652832031, 4.05401611328125, 87.47511291503906, -57.3763427734375, -0.306304931640625, 33.82855224609375, 13.612396240234375, -23.974449157714844, 4.8475799560546875, -76.99072265625, -27.10052490234375, -21.2236328125, 4.0677032470703125, 155.5408935546875, 119.162109375, -37.289215087890625, 135.5630340576172, 143.37875366210938, -13.77105712890625, 2.6523170471191406, 90.88056945800781, 96.33573913574219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000251.npy"}
{"epoch": 0.5256544502617801, "step": 252, "batch_size": 128, "mean": 44.56657409667969, "std": 73.0477294921875, "min": -152.14053344726562, "p10": -37.27190399169922, "median": 35.300437927246094, "p90": 137.30088958740234, "max": 218.252197265625, "pos_frac": 0.703125, "sample": [181.67703247070312, 141.88836669921875, 140.311767578125, 57.98561096191406, -87.44683837890625, 79.03388214111328, -88.1693115234375, 10.014251708984375, 5.072360992431641, -20.927032470703125, -5.313804626464844, -38.27449035644531, 14.30633544921875, 31.156463623046875, 218.252197265625, 32.670013427734375, 131.12747192382812, 85.17865753173828, 124.08502197265625, 114.7318115234375, 3.0751094818115234, 55.4102783203125, -18.81494140625, 137.20875549316406, -152.14053344726562, 8.043693542480469, 3.5117034912109375, 79.75301361083984, -36.3740234375, 57.29841613769531, 1.3691635131835938, 131.89736938476562, 122.01910400390625, 106.43170166015625, 90.13871002197266, 24.613861083984375, -52.327178955078125, -10.490930557250977, 83.79248046875, 206.8748779296875, 105.57211303710938, -2.8929367065429688, 93.50949096679688, 30.013626098632812, 27.963348388671875, 9.213287353515625, 24.930526733398438, -8.093093872070312, -0.04620361328125, 14.511856079101562, 119.67662048339844, 3.950439453125, 137.515869140625, 95.23820495605469, -21.4564208984375, -7.1180419921875, 104.24015808105469, -46.835357666015625, 35.00028991699219, 38.63787078857422, 96.494140625, -19.713157653808594, 34.92242431640625, 146.8447265625, 37.7047119140625, 78.61090087890625, 127.38687133789062, 36.97552490234375, 45.75, 130.56085205078125, -0.49237060546875, 112.99893951416016, 146.04205322265625, -7.476493835449219, -41.22978210449219, 115.28765869140625, -105.79696655273438, 6.254478454589844, 98.71319580078125, 96.8541259765625, 89.61875915527344, 1.42236328125, 45.558349609375, 106.82042694091797, -5.260711669921875, -1.435638427734375, 0.0, 102.19767761230469, -109.63784790039062, 120.64156341552734, 0.09262275695800781, 111.79191589355469, -12.0950927734375, -21.8624267578125, 35.6005859375, 146.9361572265625, -1.7547607421875, 143.08712768554688, 47.634796142578125, 123.85963439941406, -2.12274169921875, 100.22083282470703, 69.31597900390625, -1.5709991455078125, 141.62301635742188, 0.74114990234375, 104.71780395507812, -124.75405883789062, 19.20599365234375, 4.9896392822265625, -20.32696533203125, -23.990386962890625, 39.06707763671875, -118.95587158203125, -110.72958374023438, -0.261962890625, 110.75872802734375, 151.94512939453125, -53.68280029296875, 113.26007080078125, 28.2138671875, 149.00711059570312, 96.22482299804688, -36.84222412109375, 55.73150634765625, 99.05322265625, 11.817035675048828, 119.77659606933594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000252.npy"}
{"epoch": 0.5277486910994764, "step": 253, "batch_size": 128, "mean": 34.897972106933594, "std": 68.12645721435547, "min": -146.44561767578125, "p10": -39.51597442626953, "median": 18.789688110351562, "p90": 131.3455078125, "max": 166.03668212890625, "pos_frac": 0.734375, "sample": [158.725830078125, -5.78387451171875, 10.027633666992188, 75.55413055419922, 18.23809814453125, 24.24169921875, 2.4400463104248047, 81.65589904785156, 166.03668212890625, 26.934356689453125, 40.37841796875, 131.0869140625, 7.177734375, -65.78146362304688, 64.90243530273438, -109.3428955078125, 81.80374145507812, -75.27098083496094, 115.2947998046875, -32.106201171875, 2.616943359375, 44.34722900390625, 36.06571960449219, 10.567548751831055, -1.53155517578125, -34.5625, 34.604339599609375, 139.77532958984375, 108.54155731201172, 98.56830596923828, -2.5268020629882812, -13.539737701416016, -112.48001098632812, 26.75274658203125, 164.76321411132812, 8.326202392578125, -10.632431030273438, 137.28530883789062, -146.44561767578125, -15.3809814453125, 129.32733154296875, 47.06939697265625, 136.30618286132812, 26.014625549316406, 20.23388671875, -39.90544128417969, 115.00306701660156, 3.863494873046875, -11.059463500976562, 16.436798095703125, 131.275634765625, 36.30413818359375, 129.33737182617188, 10.913681030273438, -46.5350341796875, 118.2100830078125, 6.549766540527344, 10.864433288574219, -36.960418701171875, 90.32142639160156, 28.914596557617188, -33.2806396484375, 6.576629638671875, 112.07500457763672, 19.341278076171875, 126.42216491699219, -39.23614501953125, -18.806907653808594, 96.62493896484375, -29.205596923828125, 25.278724670410156, -12.915481567382812, -113.8569107055664, -1.74163818359375, -117.4947509765625, 60.919525146484375, -91.189697265625, 4.128366470336914, -2.2254638671875, -30.5667724609375, 82.83694458007812, 0.829681396484375, 54.021728515625, 106.05609130859375, 35.53009033203125, -3.114105224609375, 29.877037048339844, 7.45367431640625, -47.625762939453125, 148.91702270507812, -64.37599182128906, 7.928245544433594, 111.65185546875, 1.066741943359375, 6.02618408203125, 116.92060852050781, -49.85505676269531, 156.53680419921875, 8.555305480957031, 82.79458618164062, 97.96763610839844, 13.833503723144531, -39.34906005859375, 2.2955322265625, 37.0545654296875, 131.508544921875, 14.614593505859375, 141.92636108398438, 15.219970703125, -4.908306121826172, 19.41558837890625, 133.7349853515625, 99.86992645263672, 3.670185089111328, 100.40983581542969, 13.548004150390625, 11.25787353515625, 14.119583129882812, 35.20709228515625, 162.653076171875, 7.139820098876953, 44.8656005859375, 139.88720703125, 104.31495666503906, 11.080352783203125, 100.76081848144531, 77.9227294921875, 110.22772216796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000253.npy"}
{"epoch": 0.5298429319371728, "step": 254, "batch_size": 128, "mean": 44.04839324951172, "std": 62.52665710449219, "min": -148.73223876953125, "p10": -26.517162322998047, "median": 29.90936279296875, "p90": 127.36439743041991, "max": 166.54615783691406, "pos_frac": 0.75, "sample": [155.4349365234375, 0.2475433349609375, -89.34004211425781, 28.148231506347656, -11.412834167480469, -3.859548568725586, 18.2066650390625, 17.649337768554688, -0.22484779357910156, 19.8292236328125, -4.0899658203125, -20.4716796875, -9.40948486328125, -3.56903076171875, 8.387985229492188, 138.08523559570312, 5.5098876953125, 116.18820190429688, 50.427490234375, -34.491302490234375, -32.693939208984375, 136.66778564453125, 17.678382873535156, 8.92041015625, -7.68756103515625, 51.58705139160156, 104.6112060546875, -68.18009948730469, 105.90377807617188, 84.82977294921875, 111.10467529296875, 148.45278930664062, -96.0411376953125, 85.4012451171875, 1.4334487915039062, 19.65985107421875, -27.129005432128906, 72.49905395507812, 6.021903991699219, 13.466262817382812, 72.58944702148438, 58.519317626953125, 93.8265380859375, 46.83308410644531, -52.504608154296875, 20.659133911132812, 112.86248779296875, 51.40716552734375, 67.466796875, 107.52464294433594, 73.61686706542969, 4.0224151611328125, 44.538429260253906, 122.60183715820312, 90.54296875, 29.8096923828125, 68.90255737304688, -72.2506103515625, 16.37933349609375, 87.45478820800781, 1.8490066528320312, 20.81629180908203, 141.55795288085938, 6.719963073730469, 56.595367431640625, 126.17423248291016, 47.08355712890625, 124.2237548828125, 130.14144897460938, 34.86756896972656, 11.465538024902344, 38.09613037109375, 139.14093017578125, 5.297027587890625, 48.58360290527344, -26.25494384765625, -3.380096435546875, 30.009033203125, 1.5606536865234375, 3.2305145263671875, -3.157135009765625, -9.656951904296875, 140.26190185546875, 122.37295532226562, 89.22602081298828, 148.39044189453125, 69.66888427734375, 109.42710876464844, 17.030609130859375, 119.26480102539062, 23.75959014892578, 122.19120788574219, 121.40341186523438, 100.20822143554688, 114.4941635131836, 25.21185302734375, -0.3278045654296875, 119.73213958740234, -3.8867950439453125, -7.0956878662109375, -16.903961181640625, 53.744476318359375, 45.67572021484375, -2.7814083099365234, 0.0, -29.299026489257812, 132.97723388671875, 92.30638122558594, 36.288421630859375, 23.7982177734375, 10.996231079101562, 24.35260772705078, 38.04058074951172, 125.77774047851562, 89.71182250976562, -0.6046600341796875, -148.73223876953125, -69.03347778320312, 151.03826904296875, 23.06463623046875, -44.977783203125, -36.38604736328125, 5.611061096191406, 166.54615783691406, 156.84388732910156, 103.97354125976562, 102.1837158203125, 85.13139343261719], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000254.npy"}
{"epoch": 0.5319371727748691, "step": 255, "batch_size": 128, "mean": 47.90528869628906, "std": 72.31405639648438, "min": -127.06089782714844, "p10": -24.46545791625976, "median": 31.322792053222656, "p90": 140.25442810058593, "max": 188.87103271484375, "pos_frac": 0.765625, "sample": [22.475914001464844, -62.74554443359375, 150.1226806640625, 52.79615783691406, 55.23042297363281, -117.87127685546875, 3.756134033203125, 133.60623168945312, -11.36383056640625, 113.2943115234375, -7.6729278564453125, 9.180740356445312, 10.736072540283203, 11.33453369140625, 160.78402709960938, 133.5511474609375, 3.84710693359375, 16.76611328125, 152.43609619140625, 11.002166748046875, 124.42950439453125, -29.5037841796875, -10.796722412109375, 116.71864318847656, 43.59771728515625, 140.54937744140625, 59.269447326660156, 114.94955444335938, -18.342681884765625, 31.144805908203125, 86.76687622070312, 7.307891845703125, 20.132278442382812, 28.15924072265625, 52.95751953125, 33.64817810058594, 133.14710998535156, 64.48869323730469, -80.8851318359375, -80.29388427734375, 188.87103271484375, 159.6668701171875, 139.95909118652344, 31.676956176757812, -92.64402770996094, -22.306175231933594, -126.5272216796875, 142.01638793945312, 102.48274993896484, 95.36944580078125, 128.4818115234375, 124.8984375, 104.01419067382812, 100.22677612304688, 157.30355834960938, -2.74365234375, 52.0323486328125, -100.2618408203125, -65.51718139648438, 119.15380859375, 25.6529541015625, 127.69757080078125, 114.98501586914062, 112.83551025390625, 30.58526611328125, 79.82528686523438, 8.758827209472656, -108.44522857666016, 134.66226196289062, 19.180953979492188, 154.0030517578125, 22.061859130859375, 28.616397857666016, -14.568038940429688, 1.3934326171875, 15.381997108459473, 5.961210250854492, 55.452911376953125, 10.800567626953125, 106.57022857666016, -7.629150390625, 44.61041259765625, -127.06089782714844, 154.312255859375, 128.86660766601562, 105.88438415527344, -4.50994873046875, 31.500778198242188, 16.871231079101562, 18.059974670410156, 140.12802124023438, 127.03936767578125, 83.76359558105469, 134.65444946289062, -17.15704345703125, 163.402099609375, 123.81875610351562, 4.933135986328125, 6.911285400390625, -11.025672912597656, 3.022019386291504, -38.7657470703125, 90.18185424804688, 6.614410400390625, 141.57760620117188, 84.35595703125, 9.529876708984375, 114.00753784179688, -16.959007263183594, 101.53738403320312, 4.44366455078125, 11.469451904296875, -12.264022827148438, 128.37277221679688, 23.281692504882812, 34.38203430175781, -3.5145263671875, -11.19146728515625, 116.68980407714844, 54.33659362792969, 12.408027648925781, 120.21158599853516, -6.2322235107421875, 85.54827880859375, -53.580322265625, 167.0286865234375, 21.870223999023438, -20.135169982910156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000255.npy"}
{"epoch": 0.5340314136125655, "step": 256, "batch_size": 128, "mean": 48.449710845947266, "std": 75.93132781982422, "min": -142.02621459960938, "p10": -48.67302093505859, "median": 49.164825439453125, "p90": 139.72991638183595, "max": 174.21310424804688, "pos_frac": 0.7265625, "sample": [-4.681854248046875, 131.50091552734375, 139.57516479492188, -40.156982421875, 20.661178588867188, 97.44390869140625, 130.53680419921875, 96.88217163085938, 121.09609985351562, -0.8192138671875, 112.26579284667969, -51.61244201660156, -32.453277587890625, -115.91261291503906, 69.96537780761719, 135.8953399658203, 140.09100341796875, -17.771759033203125, 100.94313049316406, 135.57025146484375, 2.2257308959960938, 120.82920837402344, -15.23040771484375, 174.21310424804688, 16.56201171875, 135.1255340576172, 119.970458984375, -63.67083740234375, 151.74786376953125, 16.056396484375, 146.63058471679688, 41.81547927856445, 63.206298828125, 38.34674072265625, 37.376953125, 11.915802001953125, -41.803375244140625, -131.39202880859375, 97.34603881835938, 67.29792785644531, 149.28900146484375, 54.86224365234375, 92.66317749023438, 139.5426483154297, 60.75096130371094, 56.68951416015625, -75.0565185546875, 42.8734130859375, -47.41326904296875, 164.20333862304688, 4.549648284912109, -59.56311798095703, 145.41278076171875, 146.07473754882812, 26.47882080078125, 3.104990005493164, -140.4491729736328, -32.33599853515625, 121.1640625, 93.32615661621094, 107.79086303710938, 18.54595947265625, 113.9554443359375, 149.60452270507812, -4.722857475280762, 101.30412292480469, 48.58795166015625, -109.75032806396484, 111.80862426757812, -6.1296234130859375, -33.53839111328125, -103.55314636230469, 93.24850463867188, 43.097808837890625, 35.374267578125, 122.00360107421875, -15.567596435546875, 124.41729736328125, 27.913360595703125, 149.52084350585938, 157.1064453125, 102.71891021728516, 101.40717315673828, 0.16656494140625, -0.70513916015625, 66.13833618164062, -100.59642028808594, 6.205768585205078, 5.51898193359375, 138.48553466796875, -75.3035888671875, 131.12432861328125, 110.83857727050781, 140.9395751953125, 12.828125, 113.9920883178711, 56.178619384765625, -12.13250732421875, 17.98621368408203, 24.92926025390625, 41.5523681640625, 51.67931365966797, 137.44873046875, 88.73638916015625, 34.509979248046875, 49.74169921875, 24.638961791992188, 130.61068725585938, -64.7652587890625, -142.02621459960938, 6.552337646484375, 138.93324279785156, 140.19796752929688, 15.375251770019531, -14.271499633789062, -1.0766143798828125, 137.96517944335938, -31.07196044921875, -9.206161499023438, -40.742919921875, 108.24775695800781, -6.96728515625, -1.0624008178710938, 67.81116485595703, 17.83465576171875, 51.42373275756836, 74.7530517578125, 119.2471923828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000256.npy"}
{"epoch": 0.5361256544502618, "step": 257, "batch_size": 128, "mean": 43.611209869384766, "std": 69.00592803955078, "min": -122.00140380859375, "p10": -26.929141235351562, "median": 25.7664794921875, "p90": 136.47390441894532, "max": 196.1267852783203, "pos_frac": 0.671875, "sample": [86.9080810546875, 56.424346923828125, 196.1267852783203, -21.64397430419922, -1.7385940551757812, 186.25238037109375, 24.62347412109375, 135.79103088378906, 75.27017211914062, 118.62613677978516, 79.50225830078125, 154.96560668945312, 121.43478393554688, 71.78564453125, 0.07425880432128906, 26.651222229003906, 37.154632568359375, -73.54165649414062, -25.864707946777344, 103.6370849609375, 136.1790771484375, -37.22663116455078, 91.15550231933594, 168.93502807617188, -17.54107666015625, -72.246826171875, 19.855712890625, 20.795852661132812, -5.9805908203125, -38.42887878417969, -21.172996520996094, -34.8507080078125, 127.206298828125, 22.594268798828125, 93.2222900390625, 133.01889038085938, 184.99298095703125, -0.7968292236328125, 145.29949951171875, 104.0274887084961, -23.33306884765625, 9.295665740966797, -21.323463439941406, 117.46871948242188, 139.28274536132812, -26.779296875, -2.7523269653320312, 27.9210205078125, 13.6893310546875, -122.00140380859375, 97.44815063476562, 169.64163208007812, 69.27802276611328, 61.74365997314453, 127.9063720703125, 26.567108154296875, 52.278533935546875, -20.833587646484375, 79.40664672851562, 17.0714111328125, -2.1047821044921875, 140.919677734375, 116.173583984375, 16.285858154296875, -70.4708251953125, -26.6822509765625, 77.162841796875, 3.186664581298828, 15.579620361328125, 66.84793090820312, 2.302154541015625, 111.88153076171875, 83.4310073852539, 40.2425537109375, 22.442710876464844, 7.628662109375, -106.89799499511719, -1.2646102905273438, 113.90321350097656, 137.16183471679688, 1.5383987426757812, 2.042673110961914, 78.66168212890625, 31.197357177734375, 23.157108306884766, -16.285430908203125, 3.36517333984375, 135.88003540039062, 60.937103271484375, -4.787689208984375, -17.3985595703125, 76.31478881835938, -34.539520263671875, 133.917724609375, 68.29322814941406, 119.9937744140625, 146.17996215820312, 36.13775634765625, -13.75152587890625, 7.670036315917969, 24.965850830078125, 12.757614135742188, 124.73812866210938, 34.13189697265625, -5.9910430908203125, -14.194976806640625, -4.438140869140625, -110.21966552734375, -32.462860107421875, 137.2266845703125, 23.6087646484375, 125.11865234375, -20.274887084960938, 176.37930297851562, -48.6710205078125, -2.3539581298828125, 70.89791870117188, -27.278778076171875, 103.06309509277344, 35.668212890625, 103.72557067871094, -5.008270263671875, -10.840263366699219, 67.83482360839844, -9.962074279785156, -14.549102783203125, -21.458587646484375, 120.11514282226562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000257.npy"}
{"epoch": 0.5382198952879581, "step": 258, "batch_size": 128, "mean": 49.342437744140625, "std": 63.70368957519531, "min": -155.18313598632812, "p10": -21.151481628417965, "median": 35.92991638183594, "p90": 131.44857177734374, "max": 164.419189453125, "pos_frac": 0.765625, "sample": [115.22250366210938, 23.623537063598633, 17.37078094482422, 5.643333435058594, 85.59516906738281, 16.604354858398438, 87.65725708007812, -24.713592529296875, 42.08296585083008, 121.91448974609375, -46.0078125, 36.029205322265625, 58.217193603515625, 8.188125610351562, 124.72982788085938, 89.54698944091797, -24.497726440429688, 17.867889404296875, 134.202880859375, 104.55807495117188, 136.93551635742188, 71.57551574707031, -44.88951110839844, 11.87322998046875, 162.08822631835938, -19.717376708984375, -155.18313598632812, 124.42144775390625, 133.5028076171875, 46.628997802734375, 120.71649169921875, 113.41046142578125, -36.141571044921875, 71.93489837646484, 89.835693359375, 33.870391845703125, 3.4812583923339844, -27.204620361328125, 75.50154876708984, 111.93803405761719, 131.11160278320312, 98.60044860839844, 6.919769287109375, 38.87034606933594, 14.162246704101562, 21.125640869140625, 26.35906982421875, 26.175537109375, 17.93231201171875, 14.847625732421875, 79.11698150634766, 7.474363327026367, 115.93222045898438, 27.6787109375, -136.3997039794922, 89.38135528564453, 33.63359069824219, -7.54754638671875, -54.0555419921875, 4.234710693359375, -18.587478637695312, 28.87989044189453, 35.83062744140625, 37.79498291015625, 128.96054077148438, 3.9612655639648438, 163.70648193359375, 121.91033935546875, -4.710113525390625, 107.17607116699219, 65.60263061523438, 67.67840576171875, -4.232963562011719, -4.4516754150390625, -0.7629384994506836, 73.31390380859375, 142.0447998046875, -57.97129821777344, 95.8448715209961, 27.69842529296875, -8.75029182434082, -25.9210205078125, 106.3257064819336, 116.44363403320312, -2.5674591064453125, -1.8554916381835938, 106.91873168945312, 33.080841064453125, 132.23483276367188, -43.912109375, 53.22746658325195, 4.745574951171875, 20.24407958984375, 14.798843383789062, -5.8862152099609375, 70.6114501953125, 107.62637329101562, 121.24806213378906, 9.33541488647461, -15.210418701171875, 11.02496337890625, -2.9957275390625, 71.94403076171875, -8.344131469726562, 164.419189453125, -0.1069793701171875, 115.0859375, 140.08656311035156, 130.19970703125, 115.86216735839844, 45.2398681640625, -15.831268310546875, 161.1812744140625, 94.8074951171875, 153.56524658203125, 141.42999267578125, 90.60173034667969, 9.608810424804688, 80.61373901367188, 103.86567687988281, -14.226104736328125, 111.01936340332031, 93.58598327636719, 155.77334594726562, 8.53460693359375, 27.228271484375, 21.2034912109375, -65.93716430664062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000258.npy"}
{"epoch": 0.5403141361256545, "step": 259, "batch_size": 128, "mean": 44.7153205871582, "std": 68.41375732421875, "min": -128.04168701171875, "p10": -32.14271697998046, "median": 42.401939392089844, "p90": 135.1135498046875, "max": 174.853271484375, "pos_frac": 0.734375, "sample": [104.61764526367188, -20.776107788085938, 135.45376586914062, 10.90250015258789, -29.678497314453125, 14.6280517578125, 13.590057373046875, 150.80165100097656, 127.37570190429688, -12.60711669921875, 57.23828125, 15.928009033203125, 40.42622375488281, 111.87734985351562, 11.405647277832031, 85.67405700683594, 108.985107421875, -121.19734191894531, 90.02085876464844, 125.14485168457031, 112.9075927734375, 159.29461669921875, -75.88009643554688, 35.724613189697266, 144.50149536132812, 24.9862060546875, 85.00687408447266, 95.31751251220703, -65.5802001953125, -7.80487060546875, 11.856842041015625, 36.16078186035156, -23.300994873046875, 92.92022705078125, 69.18276977539062, 46.6650390625, 75.97137451171875, 14.936386108398438, 17.158172607421875, 102.27566528320312, -42.092193603515625, 20.48052978515625, 139.24017333984375, 12.278589248657227, 2.66375732421875, -66.22866821289062, 90.33512115478516, 143.7989501953125, -2.2237091064453125, 93.18035888671875, 98.22166442871094, 87.5174560546875, 132.86407470703125, 3.3663330078125, 17.84809112548828, 9.88873291015625, -2.617919921875, 149.60205078125, 44.377655029296875, 13.651771545410156, 174.853271484375, 80.8678207397461, 19.40570068359375, 1.2558326721191406, -6.209075927734375, -110.0404052734375, 52.367034912109375, -37.89256286621094, -87.44715881347656, 144.11849975585938, 62.562347412109375, -10.921875, -5.651397705078125, 75.03378295898438, 137.88587951660156, -16.250564575195312, 46.570098876953125, 0.0, -11.00958251953125, 125.44961547851562, 124.76651000976562, 123.94064331054688, -14.550064086914062, 19.6549072265625, 52.81011962890625, 3.2605133056640625, 138.38809204101562, -96.11856842041016, -1.1082382202148438, 51.370208740234375, -128.04168701171875, -18.266754150390625, 139.4886016845703, 134.96774291992188, 26.77728271484375, 67.889404296875, 90.6355209350586, -22.707611083984375, 117.51985931396484, -16.40576171875, 77.38719177246094, 72.40042114257812, -17.102218627929688, 125.6380615234375, 7.96044921875, 79.91763305664062, 119.31746673583984, -103.41717529296875, 58.60015869140625, 119.05511474609375, 37.82293701171875, 127.09906005859375, 12.916336059570312, -58.92340087890625, 12.85723876953125, 55.762664794921875, 147.3938751220703, 3.900146484375, 57.20252227783203, -19.32781982421875, 110.63052368164062, -51.438201904296875, -24.156219482421875, 116.8614501953125, 3.49346923828125, 81.05654907226562, 119.45925903320312, 103.71177673339844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000259.npy"}
{"epoch": 0.5424083769633508, "step": 260, "batch_size": 128, "mean": 20.4696044921875, "std": 68.1166763305664, "min": -159.277099609375, "p10": -63.47238082885742, "median": 10.505232810974121, "p90": 118.96266021728515, "max": 167.2877197265625, "pos_frac": 0.6328125, "sample": [134.41082763671875, -4.2064361572265625, 108.70122528076172, 10.519149780273438, -130.34014892578125, 116.36257934570312, 48.214447021484375, 136.457275390625, 18.188766479492188, 98.11907958984375, -19.891571044921875, 17.68280029296875, 0.0, -18.700042724609375, 14.778450012207031, 4.079154968261719, 2.6043701171875, 167.2877197265625, 25.637062072753906, 46.6894416809082, 112.70613098144531, -159.277099609375, -0.7011871337890625, 18.388671875, 161.59442138671875, -53.054351806640625, 18.932537078857422, 104.31950378417969, 32.77587890625, 15.668548583984375, 36.281829833984375, -52.31092834472656, 13.249038696289062, 0.8380260467529297, 4.732595443725586, 152.96939086914062, -39.01971435546875, -74.41848754882812, -10.952651977539062, 1.0431976318359375, -0.61370849609375, 18.39752197265625, 33.84478759765625, -39.6064453125, -31.992401123046875, 106.17887878417969, -95.6004867553711, -24.395191192626953, -69.25277709960938, 10.491315841674805, 2.0947265625, -77.81201171875, 127.6962890625, 112.6116943359375, 3.225299835205078, 13.531158447265625, 126.3582763671875, -78.59970092773438, 118.62802124023438, 96.83006286621094, -32.658447265625, -3.4946975708007812, 125.62959289550781, -4.817047119140625, 7.093101501464844, -100.09087371826172, 74.23878479003906, 119.74348449707031, -4.07391357421875, 135.18637084960938, 17.75750732421875, 77.381591796875, -36.781585693359375, 9.06805419921875, 3.5351409912109375, -22.769851684570312, -19.103805541992188, 33.231536865234375, 14.94757080078125, 15.072662353515625, -108.996826171875, -61.59455871582031, 152.00091552734375, 1.6268463134765625, 0.6956787109375, 32.68017578125, 26.2188720703125, -17.928726196289062, -3.5775146484375, -122.62522888183594, -29.182769775390625, 3.67779541015625, -117.02552795410156, 144.06353759765625, 21.145736694335938, 13.18768310546875, -8.349281311035156, 63.759307861328125, -30.3056640625, 120.65341186523438, 79.41522216796875, 77.68331909179688, 19.503143310546875, 59.47636413574219, 57.7633056640625, 4.5048828125, -5.00958251953125, 95.68598175048828, -67.85396575927734, 78.07647705078125, -7.520843505859375, 79.41862487792969, 116.84442138671875, 16.606719970703125, -49.953826904296875, 4.3018341064453125, -15.071929931640625, -124.58706665039062, -11.231033325195312, 49.014434814453125, 104.4603271484375, -46.63787841796875, 26.05426025390625, 8.331380844116211, 103.6502685546875, -1.9359893798828125, 79.77709197998047, -12.25030517578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000260.npy"}
{"epoch": 0.5445026178010471, "step": 261, "batch_size": 128, "mean": 38.865570068359375, "std": 74.16605377197266, "min": -131.3951873779297, "p10": -65.02316436767578, "median": 32.1768798828125, "p90": 132.79217376708985, "max": 202.0584716796875, "pos_frac": 0.7109375, "sample": [-14.113906860351562, -130.3363037109375, 69.62150573730469, -31.77344512939453, 47.910308837890625, 16.099365234375, 117.90576171875, -21.582550048828125, 16.08880615234375, 123.14469146728516, 141.14263916015625, 0.0, -22.589454650878906, -25.465850830078125, 9.213876724243164, 44.02685546875, -21.365570068359375, 42.30535888671875, 132.43995666503906, 36.33758544921875, 84.33309936523438, -99.14651489257812, 119.73841094970703, -112.24029541015625, 92.7635498046875, -47.55923843383789, -19.264663696289062, 136.29354858398438, -68.3435287475586, -101.33853149414062, 98.88864135742188, 131.1930389404297, 133.614013671875, 90.75968933105469, 139.07315063476562, 55.603179931640625, -17.067626953125, 16.058238983154297, -50.607513427734375, 89.33419036865234, 15.0557861328125, 13.010467529296875, 140.66473388671875, -28.381622314453125, 1.1588897705078125, 23.768524169921875, -0.3973846435546875, 126.5838394165039, -95.7688217163086, 104.23945617675781, 26.355499267578125, -28.419708251953125, 41.425567626953125, 150.848388671875, 126.23641967773438, 52.05743408203125, 90.18878936767578, -41.77964782714844, 104.66380310058594, 124.60069274902344, 124.81793212890625, 129.75511169433594, 85.4091796875, 15.778549194335938, 97.01988220214844, -1.1091880798339844, 69.1541976928711, -68.01487731933594, 28.122100830078125, -1.1059799194335938, 115.16241455078125, 202.0584716796875, 0.2918701171875, 159.4664306640625, 42.33345031738281, -0.15936279296875, 9.97064208984375, 2.8294200897216797, 6.072113037109375, 13.683860778808594, 33.54931640625, -7.554168701171875, 131.58290100097656, -65.13677978515625, 159.688720703125, 97.59140014648438, 37.47410583496094, 82.87618255615234, 12.496910095214844, -102.10443115234375, 24.545326232910156, 12.116241455078125, -102.99566650390625, 4.79150390625, -113.81787109375, 141.72850036621094, 120.35992431640625, 121.57382202148438, 124.69148254394531, 44.37171173095703, -64.97447204589844, 68.0380859375, 0.61572265625, 108.54714965820312, 15.570465087890625, 52.076171875, -68.91157531738281, 158.258544921875, 135.41534423828125, -37.51799774169922, 130.15371704101562, -7.918724060058594, 36.328712463378906, -131.3951873779297, 60.91162109375, 10.147247314453125, 10.018524169921875, 30.804443359375, 97.7384033203125, 37.59820556640625, 89.96196746826172, 162.52227783203125, 11.586822509765625, -34.466064453125, -2.9103260040283203, 65.45333862304688, 3.1820526123046875, 101.38758087158203], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000261.npy"}
{"epoch": 0.5465968586387434, "step": 262, "batch_size": 128, "mean": 49.015560150146484, "std": 67.1991195678711, "min": -127.31208801269531, "p10": -27.839245605468747, "median": 32.78285598754883, "p90": 137.59140319824218, "max": 174.5909423828125, "pos_frac": 0.75, "sample": [123.1954345703125, 24.04010009765625, 80.1988525390625, 32.788536071777344, 109.23129272460938, 4.765983581542969, 164.49761962890625, 143.80093383789062, 28.13116455078125, 148.57763671875, 174.5909423828125, -21.01165771484375, 78.81841278076172, 12.865402221679688, 11.168914794921875, 4.054773330688477, 109.5255355834961, 34.775238037109375, 2.5549240112304688, 93.12660217285156, 137.90269470214844, -7.1767578125, -79.74429321289062, 133.6319580078125, 84.13214111328125, 147.7447509765625, 14.464698791503906, 20.6815185546875, 117.28271484375, 154.49737548828125, 83.927001953125, -127.31208801269531, 30.011962890625, -18.911880493164062, -0.6167449951171875, -112.41358947753906, -32.8360595703125, 45.73699951171875, 1.7293853759765625, 130.76763916015625, 31.11224365234375, -6.0641326904296875, -10.052642822265625, -33.455078125, 36.440521240234375, 72.31707000732422, 78.78227233886719, -12.13873291015625, 64.88339233398438, -18.3385009765625, 63.899688720703125, 168.5147705078125, 102.07411193847656, 102.76443481445312, -30.19696044921875, -0.5507698059082031, 92.43453979492188, 43.65625, -26.82879638671875, 16.39715576171875, 72.31735229492188, -16.431365966796875, -9.379104614257812, 78.75775146484375, -1.0275039672851562, 16.78363037109375, 142.46798706054688, 14.231765747070312, 141.03683471679688, 32.77717590332031, -90.41461181640625, -3.2705612182617188, 138.65457153320312, 5.830047607421875, 79.09860229492188, -53.2877197265625, -34.07176208496094, 130.63006591796875, 0.94671630859375, -33.799102783203125, -109.09217834472656, -19.70757293701172, 0.0, 135.027099609375, 161.12448120117188, 40.56956481933594, 16.303192138671875, 104.59440612792969, 25.298416137695312, 24.829238891601562, 116.33612823486328, 127.68792724609375, 118.83308410644531, 21.30889892578125, 137.45799255371094, 80.64935302734375, -7.960113525390625, 84.27899169921875, 66.42556762695312, 136.5904998779297, 4.307468414306641, 3.31182861328125, 117.02899169921875, 126.11019897460938, -33.153045654296875, 16.05760955810547, 27.4136962890625, 28.49847412109375, 137.3245849609375, 61.623443603515625, -2.590343475341797, 24.106773376464844, 156.26602172851562, 101.53582763671875, 122.87002563476562, 21.39556884765625, 134.55996704101562, 80.41682434082031, 131.4057159423828, 120.79286193847656, 26.359954833984375, -51.22106170654297, 41.03375244140625, 90.74273681640625, 10.984222412109375, 102.69731140136719, 5.0944366455078125, -22.2349853515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000262.npy"}
{"epoch": 0.5486910994764398, "step": 263, "batch_size": 128, "mean": 39.7635612487793, "std": 70.62631225585938, "min": -157.4957275390625, "p10": -28.630783081054684, "median": 19.265274047851562, "p90": 143.68575744628907, "max": 201.0517578125, "pos_frac": 0.6953125, "sample": [147.48464965820312, 64.77995300292969, 144.51174926757812, 30.71270751953125, -5.494087219238281, 95.93452453613281, -30.406097412109375, 0.915771484375, 89.00142669677734, 60.366943359375, -18.458940505981445, 98.14671325683594, 28.145355224609375, 25.50982666015625, -45.556640625, 3.438121795654297, 0.11825942993164062, -93.074951171875, 48.795867919921875, 1.72711181640625, 58.40936279296875, -21.038864135742188, -9.350250244140625, 96.73896789550781, -10.93414306640625, 130.88949584960938, 40.09806823730469, 84.03617095947266, 1.018798828125, -27.86993408203125, -19.034942626953125, 47.677764892578125, 148.19244384765625, 142.047607421875, 152.94244384765625, -13.64788818359375, -11.127838134765625, -13.309341430664062, -49.71783447265625, 132.2382354736328, 7.710784912109375, -6.379791259765625, -7.0845794677734375, 155.4456787109375, 56.6387939453125, 21.309484481811523, 107.93307495117188, 2.9651031494140625, 129.58267211914062, 27.720191955566406, 70.47381591796875, -22.884765625, -20.220901489257812, 101.96510314941406, 147.35397338867188, 93.1590576171875, -108.54031372070312, 9.5347900390625, 121.72967529296875, 127.35746765136719, 0.0, 38.56671142578125, -157.4957275390625, 11.710687637329102, 80.63287353515625, 8.81640625, -2.3365020751953125, 1.5575485229492188, 15.07696533203125, 26.875991821289062, 153.8516387939453, 4.4453277587890625, -42.136077880859375, -36.506431579589844, 20.24054718017578, 12.76885986328125, 99.86309814453125, 64.16656494140625, 71.32757568359375, 4.324066162109375, -12.425338745117188, 8.802658081054688, -53.4547119140625, 6.8710784912109375, 15.574386596679688, 79.89060974121094, -103.10356140136719, 151.5968475341797, 144.33160400390625, 17.63928985595703, -20.5484619140625, 152.67633056640625, 11.743515014648438, -126.48248291015625, -3.483966827392578, -8.357513427734375, -99.09440612792969, 143.40896606445312, 19.217681884765625, -44.552764892578125, 108.58998107910156, 5.80816650390625, 0.0, 10.148941040039062, 100.52340698242188, 94.25912475585938, 201.0517578125, 37.337646484375, 184.35372924804688, -10.256996154785156, 58.709747314453125, 1.1737442016601562, 137.23953247070312, 0.0, 111.96923828125, 112.32064056396484, 19.3128662109375, 26.410507202148438, 117.77096557617188, 4.90216064453125, 132.95660400390625, 125.37983703613281, 74.73878479003906, -0.80694580078125, 183.66555786132812, -9.302696228027344, 105.818115234375, -20.964202880859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000263.npy"}
{"epoch": 0.5507853403141362, "step": 264, "batch_size": 128, "mean": 50.30201721191406, "std": 67.60765075683594, "min": -132.11416625976562, "p10": -18.90445861816406, "median": 35.17832946777344, "p90": 143.66976928710938, "max": 209.54971313476562, "pos_frac": 0.734375, "sample": [-6.438597679138184, -10.8094482421875, 167.17816162109375, 81.76715087890625, 155.91705322265625, 118.73169708251953, 98.772216796875, 103.29180908203125, 1.5332279205322266, 12.19549560546875, -1.4757118225097656, 150.04721069335938, 39.5562744140625, 63.43304443359375, -27.89862060546875, 114.16394805908203, 95.71986389160156, 129.36692810058594, -18.482986450195312, 1.281219482421875, 135.15689086914062, 126.55612182617188, -18.263702392578125, 87.19715881347656, -71.66132354736328, 180.15899658203125, 132.16476440429688, -0.015804290771484375, 68.06280517578125, 59.045745849609375, -29.04718017578125, 115.10345458984375, 140.520751953125, 132.61412048339844, 158.65203857421875, 0.0, -5.659423828125, 116.27105712890625, 76.3761215209961, 14.060295104980469, -24.314788818359375, 143.18875122070312, -19.887893676757812, 65.50213623046875, 92.1205825805664, 118.9580078125, -35.82640075683594, 164.43228149414062, -63.94610595703125, 37.88560485839844, 3.802978515625, 19.98736572265625, 133.7783203125, 62.77198791503906, 118.65875244140625, -104.01296997070312, 133.01092529296875, 106.80802917480469, 73.09359741210938, 18.805458068847656, 53.5206298828125, 5.850475311279297, 51.02748107910156, 50.26429748535156, 39.9576416015625, -25.97515869140625, -5.93994140625, 24.294143676757812, 20.354446411132812, 129.40426635742188, -2.021820068359375, 11.062835693359375, -132.11416625976562, -17.46039581298828, 27.917125701904297, -3.3084640502929688, 24.215862274169922, 11.946014404296875, 144.3519287109375, 143.83148193359375, 104.8255615234375, 3.9439773559570312, -0.2581977844238281, -21.095932006835938, -2.41363525390625, 209.54971313476562, -6.720458984375, 0.0, 46.79901123046875, 29.158252716064453, -45.88555908203125, 36.571319580078125, 120.78278350830078, 143.6004638671875, 20.97479248046875, 117.20956420898438, -4.267980575561523, 44.23162841796875, 101.93575286865234, 17.042343139648438, 145.16680908203125, 79.20114135742188, 12.510353088378906, -15.7386474609375, 0.310333251953125, 1.528289794921875, 31.29852294921875, 8.440643310546875, -14.755462646484375, 132.54550170898438, 2.8755950927734375, 46.24559020996094, 46.251617431640625, 185.06573486328125, 33.78533935546875, 156.84658813476562, -6.6953125, 22.197593688964844, 150.38232421875, -6.174110412597656, -77.6507797241211, 12.3460693359375, 97.3304214477539, 6.8209381103515625, 127.735595703125, 2.367666244506836, 42.98565673828125, 18.31219482421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000264.npy"}
{"epoch": 0.5528795811518324, "step": 265, "batch_size": 128, "mean": 48.01148223876953, "std": 73.29334259033203, "min": -125.29359436035156, "p10": -41.89317932128906, "median": 45.40176010131836, "p90": 145.01305847167967, "max": 195.48614501953125, "pos_frac": 0.71875, "sample": [2.9536209106445312, 160.70654296875, 129.44924926757812, -106.20535278320312, 47.639404296875, 10.45623779296875, 127.86980438232422, 88.02159118652344, 97.62472534179688, -90.83448028564453, 56.88459777832031, 43.338958740234375, 107.60791015625, -14.38079833984375, 145.94219970703125, 114.17179870605469, 38.1661376953125, 145.7347412109375, 118.54322814941406, 101.39641571044922, 195.48614501953125, 158.0775146484375, 159.2843017578125, 14.618881225585938, 171.25393676757812, 112.243408203125, 121.50738525390625, 148.63999938964844, 144.70376586914062, -6.0939788818359375, -46.527549743652344, 118.67881774902344, 165.17788696289062, 6.819183349609375, 127.34196472167969, 144.40713500976562, 86.38619995117188, 122.48332214355469, 50.237335205078125, 138.93609619140625, -4.811126708984375, 0.344818115234375, -46.34466552734375, 53.34490203857422, -12.73199462890625, -12.848403930664062, -19.376953125, -41.414398193359375, -3.8035545349121094, 109.14089965820312, 1.170501708984375, -19.111221313476562, -38.66316223144531, 3.7664031982421875, -41.854095458984375, 94.985107421875, 127.6256103515625, 19.521438598632812, 94.70867919921875, -125.29359436035156, 22.71621322631836, 117.49951171875, 2.99853515625, 110.55436706542969, 26.592315673828125, -34.67333984375, 142.00437927246094, 17.64666748046875, 90.64921569824219, -24.376251220703125, -41.984375, 131.495849609375, 149.84532165527344, 13.768043518066406, 0.0, -5.822536468505859, -109.01553344726562, -98.380126953125, 101.0726318359375, 127.19245910644531, 41.012939453125, 7.77191162109375, 98.17835235595703, 2.5057373046875, 119.94271850585938, 27.75042724609375, 0.0, -4.807228088378906, -10.634613037109375, 90.221923828125, 76.24072265625, -26.49072265625, -1.306915283203125, 1.3910903930664062, 50.35319519042969, 4.8037872314453125, 63.4700927734375, 180.10113525390625, 20.887298583984375, -54.593780517578125, 18.877723693847656, 47.51141357421875, 86.81251525878906, 76.24125671386719, 76.2158203125, 82.42013549804688, -116.56304931640625, 157.09368896484375, 123.02323913574219, -43.01373291015625, -5.605567932128906, 55.61712646484375, 7.1568603515625, -11.1976318359375, 76.57415771484375, -46.63018798828125, 0.995269775390625, 147.06622314453125, 96.32188415527344, 101.567626953125, 116.0843505859375, -15.210693359375, 47.464561462402344, 30.40521240234375, 23.1605167388916, -99.7177734375, 28.810379028320312, 60.303436279296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000265.npy"}
{"epoch": 0.5549738219895288, "step": 266, "batch_size": 128, "mean": 47.413230895996094, "std": 67.43380737304688, "min": -160.73374938964844, "p10": -11.117470550537108, "median": 34.22911071777344, "p90": 132.68739776611326, "max": 181.39996337890625, "pos_frac": 0.828125, "sample": [46.54345703125, 109.59439849853516, 181.39996337890625, 21.144989013671875, -11.689910888671875, -12.303817749023438, 111.74659729003906, 7.25518798828125, -17.116119384765625, 13.91302490234375, 35.069244384765625, -0.6612701416015625, -5.4611663818359375, 57.2554931640625, 84.46449279785156, -10.269287109375, -10.4158935546875, 30.267623901367188, 35.06940841674805, -102.2337646484375, 127.54725646972656, 15.992645263671875, 3.30859375, 32.53265380859375, 139.55014038085938, 69.18646240234375, 0.32965087890625, 50.077392578125, 149.53228759765625, 132.1500701904297, 2.8060989379882812, 164.095458984375, -7.16387939453125, 46.61785888671875, 8.061580657958984, 116.427734375, -45.750732421875, 9.96145248413086, 91.40689086914062, 73.47079467773438, 3.0593338012695312, 28.337158203125, 16.891952514648438, -0.17893218994140625, 124.62381744384766, 115.78616333007812, 95.76539611816406, 131.83041381835938, 9.026145935058594, 30.083328247070312, 9.30621337890625, 113.69171142578125, -125.65815734863281, -123.49421691894531, 0.01483154296875, -53.14909362792969, 5.765546798706055, 42.80781555175781, 32.033935546875, 40.42230224609375, 134.39511108398438, 102.01986694335938, 15.504989624023438, 17.555206298828125, 20.859716415405273, 91.70407104492188, 110.62779235839844, 120.45327758789062, 10.205078125, 32.59172058105469, 33.38897705078125, 148.52804565429688, 68.4566650390625, 43.57470703125, 129.67971801757812, 176.40789794921875, 40.7718505859375, -3.458648681640625, -84.23397827148438, -110.22147369384766, -160.73374938964844, 146.41966247558594, 17.322601318359375, 129.9014434814453, -15.275222778320312, 42.7655029296875, 42.056549072265625, 16.302001953125, 143.9757080078125, 35.701324462890625, -86.80970764160156, 6.51947021484375, 0.8764991760253906, 27.98675537109375, 124.86135864257812, 64.29512023925781, 112.76869201660156, 99.93597412109375, 80.953125, 6.998931884765625, 18.504150390625, 53.02947998046875, 30.01007080078125, 152.08416748046875, 22.813539505004883, 17.513229370117188, -10.872138977050781, 1.7733230590820312, 40.16596221923828, -5.9609222412109375, 123.03048706054688, 150.6492919921875, 155.4640655517578, 5.303474426269531, 129.8892059326172, 11.7010498046875, 124.105224609375, 94.69490814208984, 14.24191665649414, 37.23260498046875, 131.74169921875, 16.820030212402344, 130.40675354003906, 9.900535583496094, 126.41145324707031, 111.03269958496094, 66.95465087890625, 133.941162109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000266.npy"}
{"epoch": 0.5570680628272251, "step": 267, "batch_size": 128, "mean": 40.86311340332031, "std": 77.84436798095703, "min": -160.31057739257812, "p10": -50.08448181152343, "median": 31.3480224609375, "p90": 136.8916030883789, "max": 234.4107666015625, "pos_frac": 0.6953125, "sample": [3.341339111328125, 125.99746704101562, 115.37913513183594, 102.39715576171875, 173.9974365234375, 7.2469635009765625, -0.6303787231445312, 32.256256103515625, -69.96832275390625, -3.031524658203125, -0.5317802429199219, 27.022003173828125, -65.96514892578125, 81.10661315917969, 28.02879524230957, -28.546966552734375, 0.2618865966796875, 156.148681640625, 154.7736358642578, 34.83319091796875, -91.9853515625, 186.72833251953125, 129.73684692382812, 6.413774490356445, 65.289794921875, -15.362579345703125, 66.44987487792969, 116.03019714355469, 74.39111328125, -135.652587890625, 113.70416259765625, 24.195968627929688, 28.954742431640625, -20.881500244140625, 14.325927734375, 103.71305847167969, -11.544633865356445, 153.15847778320312, 132.9368896484375, 86.4247055053711, -96.17027282714844, 136.58990478515625, 81.14493560791016, 109.50070190429688, 14.123985290527344, 195.822265625, 14.302978515625, -24.93939208984375, 122.90585327148438, 140.18975830078125, -118.4804458618164, -104.48580932617188, 118.75900268554688, 119.55872344970703, 10.161087036132812, 185.5562744140625, 111.77135467529297, 107.83963012695312, 152.19485473632812, 35.22608947753906, 158.468994140625, 80.92304229736328, -114.89776611328125, 13.062744140625, 51.776519775390625, 3.10601806640625, 80.76229858398438, -73.67507934570312, 54.7337646484375, 110.36172485351562, 11.0474853515625, 86.08746337890625, 77.13589477539062, -160.31057739257812, 69.24191284179688, 86.08599853515625, -30.430435180664062, 64.20416259765625, 3.994029998779297, -0.5877285003662109, -41.137901306152344, 98.17242431640625, -9.974853515625, 44.36285400390625, -10.668815612792969, 57.65361022949219, -52.87335205078125, 55.47279357910156, 13.236602783203125, 100.37240600585938, -24.540264129638672, -28.006057739257812, 67.00669860839844, -15.11993408203125, 4.0088043212890625, 114.82901763916016, 51.23430633544922, 86.0576171875, 45.0093994140625, -13.505157470703125, 130.08197021484375, 30.2572021484375, -29.2601318359375, 77.861083984375, -11.695137023925781, 25.33880615234375, 26.99700927734375, 132.97288513183594, -6.2198333740234375, 234.4107666015625, -24.306045532226562, -48.889251708984375, 111.15463256835938, 30.439788818359375, 99.73799133300781, 1.9639720916748047, 138.03323364257812, 119.10464477539062, -9.778656005859375, -44.56329345703125, 137.59556579589844, -155.30577087402344, -4.2974853515625, 2.961759567260742, 58.968894958496094, -91.14813232421875, -36.89048767089844, 9.55706787109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000267.npy"}
{"epoch": 0.5591623036649215, "step": 268, "batch_size": 128, "mean": 44.36071014404297, "std": 69.67716217041016, "min": -118.29032897949219, "p10": -25.11529083251953, "median": 29.661605834960938, "p90": 139.8276580810547, "max": 187.5614013671875, "pos_frac": 0.75, "sample": [31.3277587890625, 10.645797729492188, 102.2763442993164, 126.03640747070312, 127.4658203125, 11.037506103515625, -2.3901596069335938, -90.60482025146484, -28.39484405517578, -67.34657287597656, -8.538742065429688, 173.02011108398438, 17.15057373046875, 137.32427978515625, 2.64971923828125, 100.307373046875, 187.5614013671875, -5.655912399291992, -1.01593017578125, 113.00997924804688, 122.98812103271484, 131.05662536621094, 47.17591094970703, -118.29032897949219, 59.5242919921875, 34.55094909667969, -8.347919464111328, 41.2724609375, 28.038543701171875, 40.19036865234375, 9.928346633911133, 111.31037902832031, 116.43246459960938, 102.52085876464844, -24.309005737304688, 0.5963897705078125, -116.93736267089844, -87.53326416015625, -24.624618530273438, 185.5191650390625, 125.48623657226562, 35.1661376953125, 25.267578125, 140.38262939453125, -7.677734375, 14.449447631835938, 57.03892517089844, -112.46939086914062, -48.868560791015625, -13.169132232666016, 31.173797607421875, 28.1494140625, 54.768829345703125, 138.54122924804688, 5.076881408691406, 41.189208984375, 168.65345764160156, -3.9134902954101562, 85.25494384765625, 10.822296142578125, 179.626953125, 14.057075500488281, 131.35499572753906, -6.47320556640625, -26.26019287109375, 61.972320556640625, 140.65560913085938, 41.81855773925781, 1.7184906005859375, -9.76800537109375, 109.55619812011719, 19.4847412109375, 121.40194702148438, 125.06301879882812, 16.75115966796875, -8.486991882324219, 45.86712646484375, 149.3375244140625, 135.64569091796875, 60.35882568359375, 36.533966064453125, 76.10163879394531, 69.94091796875, -2.4310531616210938, 78.95309448242188, 118.45915222167969, -79.89356994628906, 51.88751220703125, 55.36341857910156, 8.283782958984375, 31.30029296875, 142.05296325683594, 164.2567138671875, 91.18441772460938, 3.0625, 137.90554809570312, 18.677200317382812, 7.749114990234375, 34.900726318359375, 46.77813720703125, -10.333511352539062, 25.698043823242188, 68.073974609375, 0.1130828857421875, 5.871671676635742, 94.92904663085938, 12.724517822265625, -36.8975830078125, 1.11602783203125, 45.144798278808594, 176.14920043945312, 12.917533874511719, 110.9327392578125, 139.58981323242188, -1.34130859375, 159.962890625, -86.17959594726562, -6.3365936279296875, 184.14874267578125, 13.708938598632812, 0.1678466796875, -41.52679443359375, 20.305465698242188, 17.393692016601562, 8.5203857421875, 14.318107604980469, -7.99945068359375, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000268.npy"}
{"epoch": 0.5612565445026177, "step": 269, "batch_size": 128, "mean": 43.119911193847656, "std": 76.53092956542969, "min": -197.47933959960938, "p10": -30.769151306152345, "median": 32.08555603027344, "p90": 140.22691650390624, "max": 184.21719360351562, "pos_frac": 0.7109375, "sample": [-117.17861938476562, 62.43241882324219, 118.3160400390625, -71.04664611816406, 67.05133056640625, -5.591766357421875, 6.180763244628906, 130.69290161132812, 14.195526123046875, -30.910003662109375, 111.23347473144531, -7.375770568847656, -8.646194458007812, 82.67559051513672, -30.63275146484375, 122.82110595703125, 6.4940948486328125, -3.8477783203125, 132.66757202148438, 107.79845428466797, 184.21719360351562, 169.51577758789062, 29.461563110351562, 152.42091369628906, 171.09548950195312, -23.731170654296875, 31.502777099609375, 133.45591735839844, -12.1158447265625, 81.99038696289062, 3.911754608154297, 151.2810516357422, 111.2774658203125, -98.7591323852539, 111.69979858398438, 0.0, 34.990509033203125, -69.83621215820312, 99.4053955078125, 114.17214965820312, -51.40520477294922, 19.298431396484375, 146.15237426757812, 70.5238037109375, 7.962821960449219, 99.30393981933594, 154.68919372558594, 0.0, 104.01020812988281, 67.07879638671875, 116.38028717041016, -19.673583984375, 135.47178649902344, -134.98577880859375, -86.28013610839844, 134.59178161621094, 111.8851089477539, -16.76067352294922, -119.17816162109375, -2.2447261810302734, 134.352783203125, 31.280288696289062, 63.396759033203125, 4.5263671875, 115.4963607788086, -21.80230712890625, 13.139007568359375, 109.5010986328125, 127.72389221191406, 23.26007843017578, 52.721099853515625, 124.67019653320312, 0.0, 146.16749572753906, 15.385459899902344, 18.4581298828125, -140.32290649414062, 2.3893089294433594, 50.102386474609375, 4.623283386230469, -6.269744873046875, -29.35333251953125, 143.9747314453125, 22.985374450683594, 109.36268615722656, 33.7880859375, 32.6683349609375, 60.194549560546875, 61.39508056640625, 140.55123901367188, 50.889801025390625, 27.87054443359375, 15.49591064453125, -20.234344482421875, 129.93524169921875, 33.792327880859375, -17.443939208984375, 122.755859375, 9.38397216796875, 107.54411315917969, 41.230255126953125, 38.08360290527344, 2.4795188903808594, -24.493072509765625, 5.436759948730469, -122.87765502929688, 97.526611328125, 143.47406005859375, 17.345672607421875, 92.9374008178711, -24.89251708984375, 13.149833679199219, 98.03341674804688, 1.702484130859375, 115.87153625488281, -81.28814697265625, -30.708786010742188, -5.3185882568359375, 9.419357299804688, -12.228195190429688, 95.08364868164062, 141.21913146972656, 84.86575317382812, -197.47933959960938, -16.892364501953125, 29.778411865234375, 140.08792114257812, 163.34011840820312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000269.npy"}
{"epoch": 0.5633507853403141, "step": 270, "batch_size": 128, "mean": 52.699180603027344, "std": 80.70469665527344, "min": -186.60733032226562, "p10": -37.902777099609374, "median": 49.021873474121094, "p90": 145.52012939453124, "max": 234.82012939453125, "pos_frac": 0.7421875, "sample": [-153.19818115234375, 50.49845886230469, 114.993896484375, 106.3934555053711, -34.743194580078125, 145.61203002929688, -13.379585266113281, 30.899261474609375, 88.3656005859375, -51.89659118652344, 134.85641479492188, 29.647247314453125, 20.55950927734375, 47.5452880859375, 22.719635009765625, 122.05801391601562, -16.613067626953125, 148.11659240722656, 98.87850189208984, -86.76606750488281, -38.41278076171875, 41.84553527832031, 3.529510498046875, 38.52911376953125, 36.66554260253906, -5.183624267578125, 126.74026489257812, 121.63945770263672, -19.98602294921875, 0.9807815551757812, 43.035308837890625, 133.1593780517578, 66.78411865234375, 3.2458572387695312, 8.452705383300781, 11.033645629882812, -46.0048828125, 42.63731384277344, -160.529052734375, 104.63946533203125, -66.007568359375, 145.48074340820312, 29.2156982421875, 12.326980590820312, 124.6605224609375, 14.644393920898438, -5.0466766357421875, 143.39712524414062, 100.30325317382812, -112.0570068359375, -125.3184814453125, 140.03500366210938, 144.64520263671875, 143.17784118652344, 71.12059783935547, 106.20468139648438, 159.15524291992188, -36.89501953125, 138.32745361328125, 5.225921630859375, 29.849609375, -19.561172485351562, 16.16314697265625, 29.168209075927734, 140.00653076171875, 128.9088592529297, 153.39752197265625, 9.935859680175781, 136.43792724609375, 90.58908081054688, 101.27555847167969, -8.543609619140625, -147.77215576171875, 129.60235595703125, -60.12994384765625, 147.6607666015625, 70.66587829589844, 55.82871627807617, 119.41275024414062, -29.179718017578125, -14.094268798828125, 11.410751342773438, 89.82980346679688, 133.77413940429688, 22.11068344116211, -5.851280212402344, 1.236175537109375, -7.68603515625, 0.0886383056640625, -4.4453887939453125, -37.6842041015625, -48.26463317871094, 104.06793212890625, 174.8385009765625, -8.0606689453125, 91.66506958007812, 90.57070922851562, 151.03875732421875, 61.9464111328125, -4.52593994140625, 129.32568359375, 171.8697509765625, 33.615997314453125, 104.9112777709961, 100.71580505371094, -186.60733032226562, 12.393585205078125, -6.9437255859375, 161.068359375, 6.888519287109375, 63.073760986328125, 123.59210205078125, 73.54689025878906, 131.3424072265625, 132.33273315429688, 153.67431640625, 152.37945556640625, 70.26178741455078, 10.04248046875, 160.5592041015625, -17.549652099609375, 234.82012939453125, 131.78378295898438, 134.82420349121094, 110.95092010498047, 114.79998779296875, 122.9139404296875, -30.7176513671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000270.npy"}
{"epoch": 0.5654450261780105, "step": 271, "batch_size": 128, "mean": 48.79624938964844, "std": 70.7165756225586, "min": -122.551025390625, "p10": -18.301233673095698, "median": 32.20060729980469, "p90": 135.76224212646486, "max": 222.96810913085938, "pos_frac": 0.765625, "sample": [115.3801498413086, -92.9886245727539, -10.38043212890625, 32.6383056640625, 16.926376342773438, -0.751800537109375, 10.561637878417969, -10.75992202758789, 36.71923828125, -6.141834259033203, 13.7904052734375, 16.591415405273438, 133.07327270507812, 0.5673675537109375, 13.747879028320312, 103.1279296875, 98.51046752929688, 34.64464569091797, 31.859527587890625, 89.70867156982422, 46.968746185302734, 10.140365600585938, 128.49453735351562, 115.54826354980469, 2.7422103881835938, 3.899433135986328, -33.88067626953125, 119.63597106933594, 33.156742095947266, -8.014434814453125, 117.69706726074219, 138.35302734375, -2.87353515625, 43.15985107421875, 1.4189910888671875, 27.971649169921875, 30.7220458984375, 52.63792419433594, 104.21185302734375, -87.17486572265625, 125.90386962890625, 36.631011962890625, -1.565155029296875, 130.1727294921875, 115.96699523925781, 25.841583251953125, 47.5892333984375, 0.1462554931640625, 135.47727966308594, 10.117950439453125, -5.537200927734375, 59.224334716796875, 143.61851501464844, 70.22010803222656, 222.96810913085938, 164.64166259765625, 0.1928730010986328, 143.18408203125, 29.008392333984375, 126.31019592285156, 107.85920715332031, 12.6571044921875, -5.291267395019531, 200.45919799804688, 119.21574401855469, -57.9842529296875, -22.611083984375, 135.06179809570312, -7.16815185546875, 32.54168701171875, 131.30230712890625, 179.57260131835938, -33.47894287109375, 101.84417724609375, -15.700897216796875, 23.3048095703125, -94.47114562988281, 30.293914794921875, 41.16326904296875, 105.41765594482422, -24.7552490234375, -122.551025390625, 146.8487548828125, -16.45415496826172, 173.51324462890625, 101.99665832519531, 1.2798004150390625, -37.761627197265625, 52.11784744262695, 28.954559326171875, 8.374561309814453, 148.01385498046875, 120.25277709960938, -13.913925170898438, 2.848236083984375, 115.10408020019531, 7.749668121337891, 110.99549102783203, 136.42715454101562, 54.91065216064453, 9.223297119140625, 24.974624633789062, -2.00067138671875, 110.32136535644531, 164.77804565429688, -8.184188842773438, -104.28518676757812, 12.00494384765625, 17.277267456054688, 106.287109375, 38.2122802734375, 120.2855224609375, 116.98470306396484, -9.934799194335938, 35.76226806640625, 17.31536865234375, 130.49069213867188, 134.6297607421875, 1.4886016845703125, -8.25146484375, 130.9110107421875, -104.85136413574219, 134.95852661132812, 165.21929931640625, 23.111709594726562, 0.6872711181640625, 38.87736511230469, -80.06510925292969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000271.npy"}
{"epoch": 0.5675392670157068, "step": 272, "batch_size": 128, "mean": 46.14250946044922, "std": 71.19876098632812, "min": -150.4344940185547, "p10": -26.32872772216797, "median": 32.223854064941406, "p90": 141.91563110351564, "max": 199.40231323242188, "pos_frac": 0.7578125, "sample": [90.74734497070312, 0.0, -4.0032958984375, 33.39759826660156, -93.594482421875, 58.22998046875, 174.16534423828125, 76.1666259765625, 130.53616333007812, 5.6622161865234375, -23.686447143554688, 22.558319091796875, 51.622344970703125, 128.06915283203125, 57.0213623046875, 158.38369750976562, 118.79718017578125, 116.63671875, 39.76416015625, 21.318588256835938, 122.51715087890625, 10.768424987792969, -8.2694091796875, 0.985809326171875, 124.56866455078125, -26.691726684570312, 4.78666877746582, 174.12106323242188, 0.8784961700439453, 16.20123291015625, 121.73764038085938, 50.01519012451172, 72.57138061523438, 1.7971916198730469, 142.22918701171875, 122.14682006835938, 166.86624145507812, 141.78125, 15.441436767578125, -1.144439697265625, 94.18910217285156, -44.93865966796875, -76.1378173828125, 84.85653686523438, -6.220733642578125, 30.668380737304688, 128.2430419921875, -45.899871826171875, -3.9422988891601562, 1.2925834655761719, 147.1328125, 121.25091552734375, -16.03032875061035, 27.644058227539062, 132.36502075195312, 164.15521240234375, 38.810455322265625, 48.191802978515625, 106.72119140625, 1.3351936340332031, -0.7567195892333984, 10.404693603515625, 9.870590209960938, 15.649795532226562, 105.9875717163086, 63.59092712402344, 199.40231323242188, 74.77487182617188, 156.17816162109375, 8.6524658203125, 138.44046020507812, -31.261627197265625, 75.30764770507812, 12.199775695800781, -68.84063720703125, -101.79399108886719, -10.8690185546875, 55.93519592285156, 5.866851806640625, 131.38394165039062, 107.97280883789062, 53.932952880859375, -8.742340087890625, 171.05799865722656, 3.4005126953125, -150.4344940185547, 34.758811950683594, 13.628528594970703, 31.05010986328125, 4.110261917114258, 39.143951416015625, 96.3253173828125, 5.6941986083984375, -78.70864868164062, 91.20573425292969, 67.8016357421875, 52.05047607421875, 119.8864974975586, -26.17315673828125, 7.566839218139648, 0.0, -25.246063232421875, 9.645408630371094, -118.82086181640625, -21.1195068359375, 5.48419189453125, 7.200653076171875, -1.5904083251953125, -2.6159439086914062, 29.661102294921875, 111.75874328613281, 197.6434326171875, -53.09600830078125, 67.31287384033203, 136.84600830078125, -100.2161865234375, 12.473983764648438, 3.8503780364990234, 7.485149383544922, -9.53857421875, 154.65411376953125, 85.45807647705078, 59.76665496826172, 63.6624755859375, 100.9873046875, 148.66241455078125, 115.52415466308594, 85.96923828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000272.npy"}
{"epoch": 0.5696335078534032, "step": 273, "batch_size": 128, "mean": 50.30378723144531, "std": 78.1760025024414, "min": -154.96490478515625, "p10": -42.60020446777342, "median": 41.06462097167969, "p90": 148.74648132324216, "max": 211.376220703125, "pos_frac": 0.6953125, "sample": [16.3006591796875, 171.73773193359375, 162.068359375, 9.919561386108398, 168.54290771484375, -8.372795104980469, 190.09832763671875, 18.59912109375, 69.78865051269531, 7.5396270751953125, -27.68389892578125, 57.40375518798828, 1.5124320983886719, 95.12481689453125, 211.376220703125, -154.96490478515625, 71.80889892578125, 187.70712280273438, 16.05126953125, 68.66217041015625, -59.80632019042969, 84.99614715576172, 132.65432739257812, 25.27484703063965, 152.390869140625, 25.139495849609375, -124.92678833007812, 35.88533020019531, 129.31201171875, -23.32257080078125, 18.051467895507812, -39.615814208984375, -27.709999084472656, 130.30438232421875, -21.987167358398438, -0.301025390625, 9.306488037109375, -4.488037109375, -6.797943115234375, 141.04507446289062, -7.2867431640625, -3.1337890625, 1.84564208984375, 47.3656005859375, 98.27197265625, 26.001495361328125, 116.96678924560547, -0.5546283721923828, -91.02439880371094, 4.00274658203125, 107.15478515625, -26.3555908203125, 156.67752075195312, -22.700775146484375, 100.26486206054688, -7.888427734375, 1.96868896484375, 67.39193725585938, 57.75245666503906, 96.0184555053711, -50.861083984375, 119.62300872802734, 10.748367309570312, 112.29277038574219, -22.813262939453125, 163.04705810546875, 103.68597412109375, -0.1990966796875, 108.87339782714844, 5.104095458984375, 126.49179077148438, 22.248886108398438, -49.56378173828125, -16.88189697265625, 66.18536376953125, 145.66888427734375, 60.801513671875, 137.64801025390625, 123.69450378417969, -78.80218505859375, 109.69144439697266, 124.44355010986328, 72.72862243652344, -51.355369567871094, 61.06573486328125, 120.01892852783203, 31.231781005859375, 68.1855239868164, 153.51611328125, 110.32984924316406, 111.59475708007812, -92.07049560546875, -3.5805816650390625, 125.59246826171875, -147.11318969726562, 38.1629638671875, -105.67828369140625, 129.3914337158203, 144.78353881835938, -10.794509887695312, 162.46044921875, 110.89947509765625, 156.239013671875, 136.7098388671875, 141.82400512695312, -6.139636993408203, -21.003707885742188, 7.32806396484375, 130.98983764648438, 113.64920043945312, 22.33270263671875, 42.0069580078125, -2.0581512451171875, 40.122283935546875, 27.282440185546875, 66.50442504882812, -72.06341552734375, 147.18460083007812, 27.017852783203125, -7.462699890136719, -12.508243560791016, 141.18789672851562, 47.4622802734375, 127.45191192626953, 159.46554565429688, -57.231781005859375, 131.0604248046875, -30.326248168945312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000273.npy"}
{"epoch": 0.5717277486910994, "step": 274, "batch_size": 128, "mean": 53.984901428222656, "std": 75.17115020751953, "min": -152.65090942382812, "p10": -24.56255340576171, "median": 43.416603088378906, "p90": 143.4785919189453, "max": 208.79290771484375, "pos_frac": 0.78125, "sample": [126.93317413330078, 130.18798828125, -50.203369140625, 126.59097290039062, -0.591583251953125, 68.57080078125, 69.53915405273438, 28.827102661132812, 112.47269439697266, 12.489472389221191, 19.972412109375, 119.39177703857422, 177.2825164794922, 84.8172836303711, 29.84869384765625, 101.27961730957031, 187.60186767578125, 158.55506896972656, -64.86541748046875, -31.438419342041016, 15.77227783203125, 166.94784545898438, 131.39968872070312, 43.34100341796875, 33.99217224121094, -4.9053955078125, -124.63489532470703, -22.988372802734375, 14.157615661621094, 87.033203125, 14.8543701171875, 35.898956298828125, 115.509765625, 70.64874267578125, 127.02203369140625, 82.9954833984375, 43.49220275878906, 119.69540405273438, 4.3082122802734375, 154.0836181640625, 157.62100219726562, -17.30691146850586, -23.086883544921875, 20.81029510498047, -29.24456787109375, 134.92405700683594, 13.02154541015625, 40.72351837158203, 22.07977294921875, 162.654052734375, -9.575126647949219, 9.888702392578125, 123.06777954101562, 67.12019348144531, -137.322265625, 27.708099365234375, -6.284454345703125, -28.005783081054688, 0.8819694519042969, 34.08319091796875, 19.334732055664062, 4.465175628662109, -2.2136306762695312, 113.91409301757812, -96.2998046875, -1.925638198852539, 138.08213806152344, -0.27852821350097656, 58.405303955078125, 127.40339660644531, 16.801513671875, -152.65090942382812, 142.54736328125, 108.07707214355469, 61.00421142578125, 53.371337890625, 59.415618896484375, 50.15284729003906, 208.79290771484375, 9.00628662109375, 194.59051513671875, 112.00027465820312, 23.72119140625, 141.42367553710938, -2.3582916259765625, 65.458984375, 124.44479370117188, 133.374755859375, 75.30810546875, 140.268310546875, 134.0756072998047, 32.65782165527344, -68.73640441894531, 96.97154235839844, 103.31491088867188, 118.42880249023438, 19.332290649414062, 145.65145874023438, 164.70947265625, -8.491226196289062, -100.27265930175781, 17.205902099609375, 5.925872802734375, -14.953353881835938, 17.454574584960938, 126.49232482910156, 13.081855773925781, 139.9015655517578, 79.06216430664062, 55.8712158203125, 145.72970581054688, 108.05236053466797, -40.0628662109375, 60.3721923828125, 0.7148590087890625, -22.1666259765625, 2.4135894775390625, 6.590911865234375, 38.612762451171875, 71.42922973632812, 14.360006332397461, -149.01055908203125, 131.51287841796875, -0.8339347839355469, 27.211963653564453, 114.6842041015625, 127.41845703125, 190.07293701171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000274.npy"}
{"epoch": 0.5738219895287958, "step": 275, "batch_size": 128, "mean": 60.7177619934082, "std": 74.70948028564453, "min": -126.96859741210938, "p10": -15.054299926757812, "median": 46.70817947387695, "p90": 151.1356628417969, "max": 183.93017578125, "pos_frac": 0.7890625, "sample": [183.93017578125, -12.1461181640625, -2.3115005493164062, -113.0274658203125, 144.651611328125, 107.37368774414062, 141.54067993164062, 4.0194091796875, 17.5819091796875, -114.53445434570312, 2.3375396728515625, 162.14666748046875, -37.0369873046875, 151.38385009765625, 0.0, 10.332366943359375, 19.757415771484375, 109.599853515625, 3.2180747985839844, 5.2118072509765625, 0.0, 0.0, 126.34962463378906, 132.43032836914062, 111.05661010742188, 4.8989410400390625, 156.4996337890625, 115.27644348144531, 19.6689453125, 70.44825744628906, 59.73443603515625, 7.219718933105469, 140.59896850585938, 139.1693115234375, 168.67205810546875, 165.57403564453125, 0.4510612487792969, -2.44744873046875, -126.96859741210938, 48.87091827392578, 134.8323974609375, 31.552398681640625, 66.59772491455078, 24.63922119140625, 15.9022216796875, -4.893278121948242, -71.61470031738281, 137.38706970214844, 145.7200927734375, 21.3892822265625, 147.77304077148438, -121.37368774414062, 2.115264892578125, 95.96824645996094, 109.09980773925781, 21.096527099609375, 105.44900512695312, 37.1259765625, 144.281494140625, 0.0, 24.70587158203125, 44.545440673828125, 25.008819580078125, 151.029296875, -6.366905212402344, 174.37582397460938, 172.94979858398438, 72.9315185546875, 23.666091918945312, 0.0, 11.702285766601562, 154.79861450195312, 80.14488220214844, 120.14157104492188, -22.04241180419922, 125.87481689453125, 144.01071166992188, 183.71771240234375, 87.76560974121094, 7.801456451416016, 124.31902313232422, 133.59481811523438, 93.782470703125, 111.43655395507812, 90.17281341552734, 2.848949432373047, 23.527252197265625, 122.830322265625, 41.024391174316406, 35.883636474609375, -14.724884033203125, 110.23247528076172, 8.78033447265625, 169.390869140625, -0.37811279296875, 138.68585205078125, -15.82293701171875, 26.55864715576172, 179.38262939453125, 4.7306976318359375, 92.99887084960938, 31.754730224609375, 146.22967529296875, -0.43195343017578125, 141.07752990722656, 130.44195556640625, 12.086502075195312, -49.91778564453125, 1.740570068359375, -19.05010986328125, 144.04083251953125, -104.5531005859375, 122.00199890136719, 1.43798828125, 51.812469482421875, 7.331365585327148, 15.5338134765625, 136.14166259765625, 178.769775390625, 125.29789733886719, -38.540283203125, 136.47897338867188, 68.88423156738281, -19.551681518554688, 108.93942260742188, -12.4130859375, 131.77597045898438, 103.95852661132812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000275.npy"}
{"epoch": 0.5759162303664922, "step": 276, "batch_size": 128, "mean": 59.16699981689453, "std": 76.91448974609375, "min": -120.811767578125, "p10": -25.437736892700187, "median": 43.15990447998047, "p90": 153.27262573242186, "max": 251.46661376953125, "pos_frac": 0.7421875, "sample": [251.46661376953125, 18.58843231201172, 123.57096099853516, -3.7959365844726562, 20.73065185546875, -0.79791259765625, 37.74591064453125, 67.6484375, 75.74862670898438, 91.81372833251953, 115.11677551269531, 7.270660400390625, -11.1116943359375, 1.725433349609375, -6.053314208984375, 89.3800277709961, 79.48342895507812, 35.873779296875, -50.8148193359375, 105.96589660644531, 155.18963623046875, 96.02940368652344, 7.272539138793945, 152.4510498046875, 92.95370483398438, -22.845947265625, -14.3905029296875, -31.0594482421875, -10.0028076171875, 134.01641845703125, 124.02886962890625, -6.1962738037109375, 94.287353515625, 132.50100708007812, 64.83389282226562, 18.00323486328125, 41.3173828125, -1.2364063262939453, 127.69688415527344, 85.29127502441406, -0.53179931640625, 24.709503173828125, 135.08511352539062, 54.626434326171875, 6.045688629150391, 170.64340209960938, 114.74142456054688, 149.361572265625, 10.54351806640625, -9.8609619140625, 9.093391418457031, 141.24282836914062, 121.91781616210938, 33.68243408203125, 146.10910034179688, 11.978652954101562, 45.00242614746094, 30.089874267578125, 144.3114013671875, -36.963623046875, 0.0, -98.4989013671875, 75.56355285644531, -112.34928894042969, 8.924560546875, -49.010101318359375, 67.00042724609375, 182.91644287109375, -1.1872138977050781, -19.29974365234375, 80.58871459960938, -19.745574951171875, 110.70855712890625, 138.6011962890625, 123.22228240966797, 148.65093994140625, 141.3001708984375, -23.374656677246094, 33.41278839111328, 127.94297790527344, 147.6731719970703, 124.06021118164062, 123.77120971679688, 33.87286376953125, 13.144615173339844, 101.53118896484375, -50.538818359375, 127.1413803100586, -2.80792236328125, 51.287841796875, -7.52186393737793, 201.0516357421875, 19.64666748046875, 19.58551025390625, 11.992111206054688, 9.288162231445312, -1.4508209228515625, 161.84677124023438, 13.156051635742188, 143.234130859375, 142.06063842773438, 71.47309875488281, 157.43701171875, 3.85107421875, -1.2254486083984375, 107.13691711425781, -47.00177001953125, -30.251590728759766, 29.807830810546875, 157.6041259765625, 5.778472900390625, 133.20504760742188, 17.753341674804688, 171.807861328125, 144.27711486816406, -112.61932373046875, 182.78567504882812, 28.54132080078125, 122.78221130371094, -51.240081787109375, 167.3253173828125, 8.898162841796875, 174.2986602783203, -120.811767578125, -103.68119812011719, 150.63836669921875, 177.94091796875, 137.9475860595703], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000276.npy"}
{"epoch": 0.5780104712041885, "step": 277, "batch_size": 128, "mean": 37.82111358642578, "std": 82.72518920898438, "min": -178.23269653320312, "p10": -47.31019058227539, "median": 27.188613891601562, "p90": 142.8836700439453, "max": 230.05242919921875, "pos_frac": 0.6875, "sample": [6.760711669921875, 5.557891845703125, -5.792863845825195, 118.517333984375, 29.86883544921875, -24.5264892578125, -133.85195922851562, -8.9083251953125, 124.24063110351562, 5.5667724609375, 19.97540283203125, 105.05061340332031, 134.2359161376953, 64.25794982910156, -8.158355712890625, 151.84017944335938, 107.13502502441406, -170.9583740234375, 89.10269165039062, -0.18377685546875, 31.961029052734375, 11.02737808227539, 103.2021713256836, 53.121795654296875, 117.99467468261719, 164.49032592773438, -1.3350868225097656, 3.6168365478515625, 146.20169067382812, -10.758003234863281, 95.24816131591797, -151.850341796875, -5.465110778808594, -47.249488830566406, -55.44483184814453, 9.757125854492188, 120.21986389160156, 7.180736541748047, 93.11542510986328, 81.94647216796875, -178.23269653320312, 2.9453582763671875, -19.172142028808594, -28.301361083984375, 9.445953369140625, 98.98193359375, 39.789154052734375, 68.46852111816406, 19.941162109375, 47.489013671875, 86.0426025390625, -19.448963165283203, 1.8646202087402344, 230.05242919921875, 45.864044189453125, 0.0, 78.94314575195312, 89.35997772216797, 122.82406616210938, 28.64874267578125, 169.63119506835938, 103.08303833007812, -134.02935791015625, -5.950016021728516, 2.304931640625, 2.8531494140625, 106.07738494873047, -110.4825439453125, 203.2303466796875, 192.820556640625, -27.46835708618164, 4.240386962890625, 171.1358642578125, 59.44831848144531, 10.266754150390625, -171.11135864257812, 15.978240966796875, -99.64044189453125, -47.45182800292969, 69.81674194335938, 19.3707275390625, 116.76065826416016, 57.68109130859375, -26.949371337890625, 91.2166748046875, 120.64646911621094, 116.57376098632812, 106.26289367675781, -110.45077514648438, 163.229736328125, 2.3316478729248047, -0.0559539794921875, -15.605926513671875, 102.40192413330078, 59.21258544921875, -23.338623046875, 37.2340087890625, 112.27017211914062, 39.68501281738281, 130.04617309570312, 1.2039947509765625, 14.943511962890625, 128.73468017578125, 151.55145263671875, 44.233734130859375, 131.4820098876953, 29.09356689453125, 145.235595703125, -25.603302001953125, 0.0, 147.14797973632812, 182.0714111328125, -24.3629150390625, 71.29638671875, -32.535247802734375, -11.798660278320312, -132.50625610351562, 141.87570190429688, 18.932144165039062, 25.728485107421875, 113.695556640625, -108.82562255859375, -0.8318367004394531, -1.9079055786132812, 9.977924346923828, 116.70462036132812, 35.35657501220703, -45.2830810546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000277.npy"}
{"epoch": 0.5801047120418849, "step": 278, "batch_size": 128, "mean": 34.62403106689453, "std": 85.15026092529297, "min": -196.93594360351562, "p10": -86.04334106445313, "median": 23.83424186706543, "p90": 140.26858825683593, "max": 224.83511352539062, "pos_frac": 0.71875, "sample": [139.21905517578125, 110.7679443359375, -39.963348388671875, 23.721939086914062, 19.3050537109375, 44.398773193359375, 101.90280151367188, 22.48992919921875, 23.864925384521484, 146.66505432128906, 72.00146484375, 126.92731475830078, 110.1759033203125, 27.135604858398438, 113.45564270019531, 125.40121459960938, -126.70292663574219, -20.13831329345703, 26.194808959960938, -5.27850341796875, 43.790870666503906, 79.43313598632812, 21.009796142578125, -29.395660400390625, 12.493682861328125, 196.40628051757812, 162.2130126953125, 34.626556396484375, 11.590316772460938, -42.290771484375, -184.5098876953125, 200.4801025390625, -110.4782943725586, 5.917938232421875, 99.0633544921875, 115.95361328125, 100.90451049804688, -85.81552124023438, 166.65029907226562, 22.334739685058594, 142.71749877929688, 164.9254150390625, 92.10299682617188, 12.972015380859375, -36.11732482910156, 102.55810546875, -0.8611907958984375, 163.44021606445312, 18.17784881591797, 34.41338348388672, 8.65163803100586, -1.32513427734375, 2.6417884826660156, -105.59555053710938, -20.115997314453125, -196.93594360351562, 2.4958267211914062, 170.60787963867188, -86.57492065429688, -112.59980773925781, 52.43678283691406, 121.83319091796875, 28.162506103515625, 9.095703125, 13.79290771484375, 144.3931884765625, -115.49636840820312, 130.2659454345703, 130.0386505126953, 39.93119812011719, 172.265380859375, 224.83511352539062, -56.972373962402344, -113.17530822753906, 76.50140380859375, -114.67361450195312, 113.61476135253906, 69.19581604003906, 45.704010009765625, 44.924530029296875, -148.45005798339844, 17.961769104003906, 3.0895843505859375, 52.21075439453125, 117.34492492675781, -18.162208557128906, 43.800201416015625, 48.045265197753906, 11.3770751953125, 122.01663208007812, -0.2587738037109375, -5.58892822265625, 127.39198303222656, 186.095458984375, -29.408966064453125, 136.40972900390625, -48.396240234375, 42.0577392578125, 113.62750244140625, 2.2413482666015625, -11.383056640625, 8.504989624023438, 22.5472412109375, -105.29569244384766, -15.02423095703125, 30.977386474609375, 73.54672241210938, -35.30400848388672, 23.803558349609375, 8.186901092529297, 124.44558715820312, 3.479391098022461, 15.205841064453125, 82.57366943359375, 1.078378677368164, -27.17560577392578, 37.74083709716797, -74.12197875976562, 107.7064208984375, -33.949493408203125, 5.156822204589844, 33.10221862792969, 13.5052490234375, 138.42034912109375, -134.40521240234375, 79.98509216308594, -80.63163757324219, 131.62094116210938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000278.npy"}
{"epoch": 0.5821989528795811, "step": 279, "batch_size": 128, "mean": 57.213043212890625, "std": 76.16563415527344, "min": -135.168701171875, "p10": -23.762976837158202, "median": 39.180084228515625, "p90": 158.21787872314454, "max": 236.2691650390625, "pos_frac": 0.78125, "sample": [44.97930908203125, 126.71330261230469, 35.49183654785156, -23.037940979003906, 18.35614013671875, -9.260894775390625, 82.22740173339844, 105.7098388671875, 6.193733215332031, 140.02362060546875, 96.73785400390625, 175.95416259765625, 39.36480712890625, -40.41925048828125, -3.374267578125, 38.995361328125, 160.84762573242188, 118.82146453857422, 36.6080322265625, 130.45724487304688, 19.805892944335938, 34.66320037841797, -9.9686279296875, 35.185791015625, 78.86666870117188, 26.359039306640625, 20.75253677368164, 8.231353759765625, -127.55796813964844, -27.379013061523438, 107.10614013671875, 178.04147338867188, 139.997802734375, 135.90435791015625, 82.05929565429688, -39.555389404296875, 5.30194091796875, 121.79977416992188, -31.396377563476562, 11.619705200195312, 123.7575454711914, 126.97871398925781, 135.7643280029297, 160.39100646972656, 4.438621520996094, 158.74609375, -0.5086517333984375, 10.986526489257812, -37.145538330078125, 111.3984146118164, 22.649078369140625, 27.557952880859375, 11.752334594726562, 52.308170318603516, 32.65983581542969, -5.8468475341796875, 236.2691650390625, -135.168701171875, 123.17654418945312, 175.7537841796875, -15.014595031738281, -7.57135009765625, 24.51214599609375, 21.47314453125, -94.76864624023438, -10.099227905273438, -21.678024291992188, 139.4810028076172, -2.1996421813964844, 200.39129638671875, -37.2493896484375, 202.8206787109375, 112.77630615234375, 97.1568832397461, 10.434562683105469, -125.82337951660156, 9.7120361328125, 148.64630126953125, 185.8150634765625, -117.59097290039062, 9.756072998046875, 25.32928466796875, 75.55941772460938, 92.97589111328125, 134.27389526367188, 40.065643310546875, 74.86553955078125, 9.955558776855469, 155.27674865722656, 33.2767333984375, 71.14788818359375, -24.928146362304688, 125.01234436035156, 50.317626953125, 22.212141036987305, 17.633163452148438, 153.116943359375, 2.8359203338623047, 49.9678955078125, 130.3845672607422, 74.21359252929688, 157.20205688476562, -12.194252014160156, 143.79953002929688, 172.41351318359375, 155.22857666015625, 14.5604248046875, 10.25238037109375, 17.655853271484375, 157.9915008544922, -2.273101806640625, 37.618743896484375, 43.295166015625, 184.66131591796875, 126.90370178222656, 53.38458251953125, 18.458587646484375, 45.62139892578125, -52.369476318359375, -23.26361846923828, 167.81027221679688, -8.174285888671875, 92.70318603515625, 2.8971405029296875, 142.22552490234375, 93.63139343261719, 40.388916015625, 111.22250366210938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000279.npy"}
{"epoch": 0.5842931937172775, "step": 280, "batch_size": 128, "mean": 50.03667068481445, "std": 81.02944946289062, "min": -166.81842041015625, "p10": -33.315118408203126, "median": 28.83316993713379, "p90": 157.89238586425782, "max": 197.97412109375, "pos_frac": 0.734375, "sample": [125.75407409667969, 129.73291015625, 0.9354400634765625, 137.56594848632812, -4.70268440246582, -166.81842041015625, 141.20834350585938, 133.41558837890625, 173.15768432617188, 8.841354370117188, -17.403793334960938, 32.53437042236328, 48.277442932128906, 83.43017578125, 19.36163330078125, -44.853858947753906, 131.91864013671875, 84.86062622070312, 167.29937744140625, -6.95208740234375, -1.988861083984375, 157.59063720703125, -83.84819030761719, 159.37254333496094, -33.7001953125, 123.39277648925781, 27.074548721313477, 21.2862548828125, -10.49395751953125, 128.4029541015625, 0.0, 28.12176513671875, 29.7724609375, -38.959320068359375, 47.962921142578125, 197.97412109375, -1.0182571411132812, 140.2451171875, 10.42972183227539, 4.089111328125, -6.640869140625, 116.65858459472656, 9.81768798828125, 23.989959716796875, 119.10381317138672, 105.86381530761719, 121.72335815429688, -129.80355834960938, 158.59646606445312, 18.875946044921875, 5.86376953125, 10.561790466308594, 5.3853759765625, 5.959001541137695, 120.07783508300781, 160.6182861328125, 60.48187255859375, -27.102294921875, 1.55059814453125, 132.98822021484375, 152.10968017578125, 59.269256591796875, 87.51244354248047, -8.200218200683594, -6.83416748046875, 51.56481170654297, -155.68179321289062, 151.36279296875, 7.9121551513671875, 103.62348937988281, 139.17568969726562, 9.905601501464844, 160.26824951171875, 163.1488037109375, 162.45364379882812, -25.81890869140625, 119.28488159179688, 97.25070190429688, 24.01336669921875, 50.7142333984375, 97.7752685546875, 156.54156494140625, -55.975311279296875, 126.14265441894531, 146.40322875976562, 8.204490661621094, 15.011932373046875, -5.6492919921875, -18.907516479492188, -0.3037109375, 159.31939697265625, 53.233154296875, 109.31352233886719, -131.73182678222656, 2.4436378479003906, 174.87322998046875, -8.357391357421875, 107.21676635742188, 155.79031372070312, 27.49774169921875, 145.17779541015625, 9.879173278808594, 131.41200256347656, -21.794723510742188, 16.413558959960938, 30.566802978515625, -0.9759025573730469, -7.5699462890625, 168.04873657226562, 15.5146484375, 23.705535888671875, 48.382972717285156, 4.861660003662109, 111.26318359375, -88.12576293945312, 121.14995574951172, 128.08212280273438, 191.443603515625, -86.63717651367188, -149.5768585205078, -30.44049072265625, 10.954517364501953, -83.16757202148438, 29.544574737548828, 99.59750366210938, 7.381683349609375, 53.037445068359375, -33.15008544921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000280.npy"}
{"epoch": 0.5863874345549738, "step": 281, "batch_size": 128, "mean": 52.9275016784668, "std": 74.73344421386719, "min": -132.25601196289062, "p10": -19.34268646240234, "median": 31.535018920898438, "p90": 155.28429412841797, "max": 270.7769775390625, "pos_frac": 0.71875, "sample": [124.80621337890625, 81.92496490478516, 3.123870849609375, 2.1709976196289062, 30.919219970703125, 10.478118896484375, 45.626861572265625, 121.67952728271484, 85.03207397460938, 169.58453369140625, 148.935546875, -116.83694458007812, -123.79428100585938, 160.18392944335938, 20.71053695678711, 24.449859619140625, 126.74285125732422, 107.4178466796875, 3.5342559814453125, 13.185722351074219, 154.36582946777344, 12.400382995605469, -8.070419311523438, -22.312347412109375, 144.01705932617188, -13.169891357421875, -78.50204467773438, 158.8212890625, 19.543792724609375, 7.2352294921875, 0.0, 55.34100341796875, -6.31683349609375, 47.8018798828125, 92.03215026855469, 143.9156951904297, 36.318878173828125, -4.07220458984375, 155.9864044189453, 113.09304809570312, -132.25601196289062, 15.728485107421875, 135.9527587890625, -18.689865112304688, 0.0, 33.5701904296875, 85.87045288085938, 133.00094604492188, -23.92779541015625, 44.24578857421875, 113.00192260742188, 8.21148681640625, 95.96675109863281, 40.68736267089844, -0.149993896484375, 188.7564697265625, -8.53271484375, 35.81866455078125, -7.41845703125, 22.760757446289062, 11.017120361328125, 50.77140808105469, -23.18255615234375, -1.9042034149169922, 123.52789306640625, 72.05926513671875, 25.713905334472656, -5.1139373779296875, -6.449310302734375, 5.275726318359375, 18.539306640625, 155.8406219482422, 3.774658203125, 151.93368530273438, 138.1313934326172, 42.5135498046875, 149.92938232421875, -0.93328857421875, 30.79052734375, 119.88485717773438, 44.16960906982422, 78.60665893554688, 270.7769775390625, 135.41769409179688, 144.65414428710938, -26.259033203125, -5.039093017578125, 116.34933471679688, 86.65473937988281, 3.813495635986328, 180.55224609375, 33.417144775390625, 116.69500732421875, -2.053943634033203, 17.88104248046875, -59.10546875, -9.246856689453125, 169.8406982421875, 133.5997314453125, -12.22186279296875, -27.09503173828125, 81.65982055664062, 0.0, 165.8477783203125, 4.395515441894531, -21.5035400390625, 117.24772644042969, -122.8359375, 179.72509765625, 164.58383178710938, 130.45028686523438, 158.89352416992188, 124.78085327148438, -3.9105224609375, 155.04586791992188, 15.644500732421875, -13.682815551757812, 77.5462646484375, 17.652969360351562, 14.650726318359375, 22.57244873046875, -20.865936279296875, -6.70965576171875, 150.29330444335938, 13.965362548828125, 77.36941528320312, 32.15081787109375, -4.676605224609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000281.npy"}
{"epoch": 0.5884816753926702, "step": 282, "batch_size": 128, "mean": 43.4375114440918, "std": 87.244873046875, "min": -177.91424560546875, "p10": -54.70844421386715, "median": 18.14112091064453, "p90": 156.6761474609375, "max": 221.38067626953125, "pos_frac": 0.71875, "sample": [148.83258056640625, -137.0475616455078, -1.9481201171875, 221.38067626953125, 44.429229736328125, 16.42070770263672, -177.91424560546875, 156.23248291015625, 119.80917358398438, 9.036056518554688, 10.460586547851562, 111.38506317138672, 20.283477783203125, -78.29928588867188, 13.5487060546875, 9.734626770019531, 176.36453247070312, -39.176513671875, 141.72723388671875, -13.604393005371094, 16.18243408203125, 19.49554443359375, -90.66462707519531, 40.94366455078125, 11.188896179199219, 219.20501708984375, 141.9514617919922, 44.73431396484375, -3.73193359375, 13.541961669921875, -44.59808349609375, 22.329391479492188, 113.7835693359375, 60.41203308105469, -4.0233154296875, 13.02606201171875, 114.12582397460938, 76.66563415527344, 49.64715576171875, 51.1898193359375, 17.543991088867188, 108.69410705566406, 3.202545166015625, -28.838775634765625, 135.69674682617188, -36.11163330078125, -133.67127990722656, 39.47911071777344, 164.56927490234375, 10.179588317871094, 24.39501953125, 8.93310546875, -25.086151123046875, 11.310005187988281, -37.582672119140625, 194.47146606445312, 36.430816650390625, -0.913055419921875, 8.194637298583984, -94.5928955078125, 7.109413146972656, -6.11346435546875, -31.151641845703125, 137.6334228515625, 147.0654296875, 115.37291717529297, 113.89238739013672, -10.220169067382812, -6.0592193603515625, 144.70968627929688, -103.0733642578125, -7.178497314453125, 140.04928588867188, 83.72979736328125, 16.58111572265625, 178.1845703125, 132.9520263671875, -97.06683349609375, 75.7410888671875, -27.906585693359375, -115.12298583984375, 17.449581146240234, -10.0404052734375, 180.11122131347656, 4.836273193359375, -91.29814147949219, 18.738250732421875, 14.685165405273438, 186.38421630859375, 157.0576171875, 184.98385620117188, 16.869384765625, 13.7032470703125, 85.962158203125, -132.9212646484375, 92.118896484375, 130.6974334716797, 156.6178741455078, 8.60251235961914, 1.760284423828125, 139.31100463867188, 215.7095947265625, 126.70240783691406, 9.745983123779297, 127.248291015625, 6.8340301513671875, -160.62570190429688, 51.21232604980469, 35.74041748046875, 28.106395721435547, 143.8332061767578, 30.943283081054688, 61.756805419921875, 119.40853118896484, 7.6835479736328125, 146.44610595703125, 163.2471466064453, -4.124847412109375, 156.81211853027344, 81.83609008789062, 9.715789794921875, -127.98197937011719, -3.44708251953125, -42.88360595703125, 147.00067138671875, -4.710845947265625, -2.6734161376953125, 148.34466552734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000282.npy"}
{"epoch": 0.5905759162303665, "step": 283, "batch_size": 128, "mean": 50.39990234375, "std": 77.1614990234375, "min": -122.09547424316406, "p10": -48.08570175170898, "median": 41.61109924316406, "p90": 144.61103515624998, "max": 213.59539794921875, "pos_frac": 0.7421875, "sample": [-92.60638427734375, 23.035140991210938, 143.31005859375, -26.0989990234375, 19.89080238342285, -55.079010009765625, 140.44342041015625, -37.3763427734375, 90.94720458984375, 7.02044677734375, -9.144561767578125, 144.13818359375, 80.18524169921875, 99.97712707519531, -1.78057861328125, -50.80543518066406, -19.770538330078125, 44.13677215576172, 1.003936767578125, -37.624176025390625, 188.08697509765625, 165.29391479492188, 33.765228271484375, -64.25421142578125, 90.86569213867188, 131.38900756835938, 103.23054504394531, 41.70648193359375, 17.202590942382812, 16.237380981445312, 43.71266174316406, 20.738067626953125, 16.758132934570312, 136.1678924560547, 122.72452545166016, 91.64505767822266, 15.029312133789062, 0.8894309997558594, 137.9420166015625, 91.92156982421875, -38.72406005859375, 127.32826232910156, -80.06495666503906, -100.40440368652344, 128.544921875, 109.8438720703125, 87.7589111328125, 20.778076171875, -112.2708740234375, 213.59539794921875, 110.1851806640625, 145.71435546875, 133.16525268554688, 123.18434143066406, 169.84510803222656, 4.891639709472656, 8.545827865600586, 28.389297485351562, 106.3095703125, 55.938446044921875, 87.253173828125, 42.65478515625, 153.50653076171875, 95.447509765625, 151.2608642578125, -4.4559326171875, 140.3883056640625, -1.22015380859375, 27.078231811523438, 40.497802734375, -48.777931213378906, 81.16671752929688, -97.74063110351562, 173.73635864257812, -5.53924560546875, 97.55746459960938, 154.9641876220703, 128.93983459472656, 0.0, 8.3756103515625, -47.789031982421875, 41.829925537109375, 117.00199890136719, -122.09547424316406, 140.5338134765625, 124.90768432617188, 71.78474426269531, 47.07038879394531, -83.17776489257812, 49.691741943359375, 41.515716552734375, -13.36065673828125, 11.037796020507812, -18.668701171875, 67.99179077148438, 138.23599243164062, 127.4254150390625, 94.60205078125, -114.32708740234375, 120.83673095703125, 0.39087867736816406, 31.21087646484375, 211.87091064453125, -3.1484832763671875, 33.183868408203125, 25.203567504882812, 130.63519287109375, -78.61953735351562, 18.96343994140625, 3.4774017333984375, 171.99508666992188, 22.44525146484375, -0.1743316650390625, 1.8189125061035156, 0.0, 111.32315063476562, 60.580116271972656, 211.20147705078125, 10.494697570800781, 5.874300003051758, 23.921951293945312, 126.0247573852539, 73.7003173828125, 42.22467041015625, -4.105865478515625, 156.35443115234375, -34.2537841796875, -28.95519256591797], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000283.npy"}
{"epoch": 0.5926701570680628, "step": 284, "batch_size": 128, "mean": 51.6978645324707, "std": 90.85394287109375, "min": -161.1236572265625, "p10": -65.74591369628907, "median": 39.730552673339844, "p90": 164.92191467285156, "max": 288.1304016113281, "pos_frac": 0.734375, "sample": [-117.93551635742188, 99.36654663085938, -91.5350112915039, -161.1236572265625, 130.53794860839844, 36.653594970703125, 165.03472900390625, 155.75592041015625, -62.15672302246094, 105.05532836914062, 120.68402099609375, 166.57965087890625, 50.02330780029297, 57.518714904785156, 29.36505126953125, 105.81889343261719, 145.33172607421875, 64.4361572265625, 170.4512939453125, 7.09716796875, 138.30487060546875, 254.2294921875, -20.737030029296875, 157.49298095703125, -7.515077590942383, 15.6800537109375, 32.31103515625, -86.29763793945312, 12.061752319335938, -43.802459716796875, -65.81121826171875, 40.62451171875, 116.226806640625, -42.505149841308594, 10.574478149414062, 41.47637939453125, 70.70266723632812, 112.10989379882812, 98.45944213867188, 13.309066772460938, -133.47943115234375, 10.709945678710938, 39.51051330566406, 38.57586669921875, -8.549339294433594, -28.932037353515625, -25.301513671875, 47.9742431640625, 11.894744873046875, 203.8782958984375, 154.0769500732422, -79.256103515625, 30.10675048828125, 17.64141845703125, 189.162841796875, 109.78970336914062, 18.08905029296875, -4.862693786621094, 162.32278442382812, 66.74002075195312, 123.31005859375, 173.4585723876953, 80.68092346191406, 6.233131408691406, 151.89292907714844, 55.5665283203125, 188.06781005859375, 28.222900390625, 105.70492553710938, 10.874521255493164, 66.40098571777344, 69.66024780273438, -25.00153923034668, 107.88510131835938, 1.3719520568847656, 288.1304016113281, 11.7032470703125, 20.692306518554688, 8.49700927734375, -1.0037994384765625, 188.41432189941406, 145.78778076171875, 117.46178436279297, 164.87356567382812, -105.71859741210938, -147.00271606445312, 9.362152099609375, -21.31134033203125, 61.13163757324219, -12.559165954589844, 115.94535827636719, 17.0762939453125, 78.48806762695312, 134.32025146484375, 0.6275138854980469, -1.6931304931640625, -8.540283203125, -119.20677185058594, -3.6692047119140625, 130.2715606689453, 160.70394897460938, 39.950592041015625, 126.45460510253906, 11.164886474609375, 0.8092155456542969, 148.85858154296875, -24.959213256835938, 115.89163208007812, 25.751220703125, 68.95033264160156, 148.75070190429688, 212.67144775390625, 119.87899780273438, -23.83740234375, -155.082275390625, -119.437255859375, 118.99118041992188, -65.71792602539062, 0.9371337890625, -40.044677734375, 189.8131103515625, 30.606613159179688, 122.795654296875, 208.93359375, 124.73681640625, -2.999835968017578, -98.68844604492188, 111.090087890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000284.npy"}
{"epoch": 0.5947643979057592, "step": 285, "batch_size": 128, "mean": 37.90556335449219, "std": 81.66541290283203, "min": -149.80709838867188, "p10": -55.76970214843749, "median": 20.66228675842285, "p90": 145.67670745849608, "max": 195.82803344726562, "pos_frac": 0.703125, "sample": [16.264678955078125, 91.9998779296875, 130.43551635742188, 30.94403076171875, 135.07394409179688, 146.04376220703125, 128.29083251953125, -145.14813232421875, 143.344482421875, 60.27873229980469, 33.52459716796875, -16.669387817382812, 10.117256164550781, -63.188232421875, -51.8465576171875, 45.60169982910156, 15.423431396484375, -50.6534423828125, -18.05999755859375, -17.896942138671875, -7.021087646484375, 140.692138671875, -120.55022430419922, 34.33393859863281, 18.92639923095703, 24.42242431640625, 165.12161254882812, 60.248779296875, 115.2467041015625, 35.841400146484375, 195.82803344726562, -96.33123779296875, 8.362411499023438, -15.920562744140625, 145.5379638671875, 147.1395721435547, 134.2850341796875, 9.90032958984375, -24.196945190429688, 0.690765380859375, -45.638580322265625, -31.48101806640625, -96.4852294921875, -7.925994873046875, 16.203523635864258, 121.89019775390625, 33.22674560546875, 114.63546752929688, 13.395618438720703, 10.213851928710938, 146.0004425048828, 2.9005165100097656, 13.500762939453125, -25.81214141845703, 166.87747192382812, 1.6461906433105469, -6.4569091796875, -2.4213333129882812, 136.13645935058594, 16.83142852783203, 86.7205810546875, 85.36395263671875, 138.7472686767578, 101.09989929199219, 50.932952880859375, 4.281005859375, -126.16607666015625, 30.612056732177734, -119.8523178100586, 94.34814453125, -149.1700439453125, 153.02984619140625, 35.45549011230469, 4.819953918457031, 109.68228149414062, -52.59033203125, 83.66897583007812, 22.398174285888672, 17.65020751953125, 17.545623779296875, 109.72742462158203, 127.9825439453125, 13.65777587890625, 1.7692108154296875, -17.005935668945312, 6.74066162109375, 36.9603271484375, 130.19898986816406, 158.78173828125, 124.28201293945312, 116.42631530761719, 133.7702178955078, -3.6769638061523438, -27.78765869140625, -26.183563232421875, -1.674835205078125, 41.9129638671875, 13.85797119140625, -113.96084594726562, 18.788818359375, 106.48361206054688, -81.08204650878906, 97.9063720703125, -35.520477294921875, -39.9412841796875, -26.883743286132812, 61.64013671875, -12.597450256347656, 155.81536865234375, 138.90643310546875, -149.80709838867188, 3.594207763671875, 29.12408447265625, 12.372772216796875, -9.861526489257812, -141.43121337890625, 138.9665069580078, 4.966522216796875, 133.84469604492188, 77.24017333984375, 49.07489013671875, 34.717437744140625, 118.55569458007812, 149.87615966796875, -84.00430297851562, 173.55697631835938, 175.4766845703125, 164.10256958007812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000285.npy"}
{"epoch": 0.5968586387434555, "step": 286, "batch_size": 128, "mean": 42.386383056640625, "std": 83.39368438720703, "min": -150.78878784179688, "p10": -59.0855354309082, "median": 28.92896270751953, "p90": 154.72183532714843, "max": 207.31591796875, "pos_frac": 0.7109375, "sample": [-13.537628173828125, -27.64422607421875, 128.36196899414062, 13.847396850585938, 36.50279998779297, 20.253524780273438, 9.28009033203125, 9.4033203125, 8.0101318359375, 102.44418334960938, 33.77455139160156, 18.11700439453125, -3.72064208984375, 58.09205627441406, -136.3999481201172, 167.7945098876953, 10.74810791015625, 166.13006591796875, 122.86849975585938, 158.03488159179688, -53.870849609375, 158.93289184570312, 9.448455810546875, 34.88383483886719, 17.53784942626953, 25.6024169921875, -20.86260986328125, -6.0031585693359375, 140.5540771484375, 0.3036346435546875, -150.78878784179688, 2.1806182861328125, 143.47293090820312, 134.3996124267578, -80.63800048828125, 12.099275588989258, 29.653594970703125, 100.27078247070312, 10.116573333740234, 57.936248779296875, -8.96783447265625, 150.24005126953125, 125.3419189453125, -106.88164520263672, 111.322265625, -48.1551513671875, 41.27459716796875, 207.24871826171875, -24.025238037109375, -5.04876708984375, 143.82919311523438, -149.07998657226562, 143.674560546875, 108.88571166992188, 36.96368408203125, 182.66845703125, 4.925811767578125, 91.8765869140625, -118.1163101196289, 170.0667724609375, -12.415946960449219, 0.0, -6.116455078125, -108.64468383789062, 105.05947875976562, 54.018310546875, -5.839115142822266, -49.712188720703125, -3.224029541015625, 78.35147857666016, -58.148101806640625, 132.139892578125, 132.2857208251953, 54.218292236328125, -65.290283203125, 66.14031982421875, 143.42715454101562, 106.83535766601562, -84.12012481689453, 102.33601379394531, 98.7064208984375, 96.05736541748047, 1.2163448333740234, 207.31591796875, 141.18963623046875, 37.95097351074219, 25.476715087890625, -22.992965698242188, 120.23951721191406, -120.63055419921875, 2.6249313354492188, 171.85556030273438, 30.19281005859375, 55.782440185546875, 153.31817626953125, 37.787750244140625, 28.204330444335938, 120.2348403930664, 111.96436309814453, 107.61991882324219, 176.7171630859375, 37.2669677734375, 157.99703979492188, 59.87464141845703, 67.3310546875, -50.01934814453125, 168.17288208007812, 14.922889709472656, -60.794952392578125, -8.943866729736328, -19.399765014648438, 132.2227783203125, -4.954376220703125, -58.352928161621094, -120.0961685180664, 197.11282348632812, -110.10568237304688, 0.731964111328125, 22.99488067626953, 13.892440795898438, 22.683181762695312, 128.99673461914062, 0.16529083251953125, 17.998489379882812, 48.71258544921875, -23.60784912109375, 111.56398010253906, 11.325035095214844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000286.npy"}
{"epoch": 0.5989528795811518, "step": 287, "batch_size": 128, "mean": 44.83531951904297, "std": 79.6950912475586, "min": -177.4283447265625, "p10": -45.92956542968749, "median": 46.994483947753906, "p90": 145.42029876708983, "max": 181.79403686523438, "pos_frac": 0.7109375, "sample": [91.65205383300781, -28.29613494873047, -135.10702514648438, -11.185277938842773, -36.49798583984375, -18.131088256835938, 145.1202392578125, 134.22372436523438, 49.49272918701172, 153.12936401367188, 126.41084289550781, 20.167694091796875, 47.62924575805664, 138.94412231445312, 24.18572998046875, 17.183502197265625, -25.193649291992188, -85.0186767578125, -71.32028198242188, 100.43982696533203, -35.48614501953125, 181.79403686523438, 137.06488037109375, -48.801513671875, 34.7781982421875, -24.176101684570312, -15.525672912597656, 7.178009033203125, 134.0513916015625, 43.553070068359375, 14.115463256835938, 148.03494262695312, 114.78459167480469, 138.63253784179688, -24.319793701171875, 3.8623046875, 85.83712768554688, 126.61529541015625, -1.6644725799560547, 127.41876220703125, 68.57727813720703, 134.2439727783203, 17.57525634765625, 121.00311279296875, 148.71763610839844, 1.3353137969970703, -2.632659912109375, 138.83999633789062, 140.5758056640625, 170.34872436523438, 114.48518371582031, 115.469482421875, 142.93197631835938, 110.72805786132812, 28.82257080078125, 46.373291015625, 89.14485931396484, -88.09503173828125, 51.059722900390625, -43.0147705078125, 128.419189453125, 56.11604309082031, 47.9556884765625, 41.89443588256836, 3.7083282470703125, 23.336029052734375, 72.08029174804688, 113.5848388671875, 49.350341796875, 83.705078125, -139.272216796875, -8.818635940551758, 81.8990478515625, 153.541748046875, 53.30010986328125, 156.953125, 141.35455322265625, 64.84832763671875, 14.648193359375, 113.62782287597656, 178.3315887451172, -72.46286010742188, -120.80668640136719, 47.61567687988281, -81.41726684570312, 99.87743377685547, -27.39093017578125, 32.24574279785156, -31.1065673828125, 6.7008819580078125, 61.99906921386719, 165.5282745361328, -177.4283447265625, -13.965667724609375, 74.71025848388672, -6.71795654296875, 64.40556335449219, 4.1813507080078125, -133.76414489746094, 39.15216064453125, 3.574432373046875, 13.070068359375, 7.238746643066406, 83.79931640625, -29.451141357421875, -53.6627197265625, 125.3863754272461, -143.40798950195312, 169.01251220703125, -16.15442657470703, 94.83389282226562, -9.204437255859375, 133.3543701171875, 47.72718811035156, 158.3922576904297, 8.52227783203125, 17.84162139892578, 150.8496856689453, 83.91989135742188, -21.73215103149414, 57.26280212402344, 7.618965148925781, -19.194000244140625, -44.69873046875, 17.68060302734375, 146.1204376220703, -11.307647705078125, 107.54353332519531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000287.npy"}
{"epoch": 0.6010471204188481, "step": 288, "batch_size": 128, "mean": 44.93536376953125, "std": 84.95999908447266, "min": -147.82138061523438, "p10": -45.55787506103515, "median": 25.642303466796875, "p90": 154.0107513427734, "max": 241.12353515625, "pos_frac": 0.7109375, "sample": [116.41879272460938, 3.2844009399414062, -26.435272216796875, -67.84970092773438, 104.29229736328125, 4.390869140625, -28.55633544921875, -16.4483642578125, 69.28533935546875, 112.09060668945312, 19.17144775390625, 175.16143798828125, -0.422515869140625, 41.622589111328125, 34.40740966796875, -7.628631591796875, 5.667076110839844, 50.58421325683594, 25.55291748046875, 0.0, 15.4761962890625, -9.18280029296875, -6.174774169921875, 8.99795913696289, 128.96676635742188, -15.01531982421875, -30.8179931640625, 8.244384765625, -43.93440246582031, 18.5361328125, 177.26052856445312, 131.552734375, 107.23970031738281, 99.75276184082031, 8.291526794433594, 0.0, -5.587486267089844, 82.13484191894531, 6.91455078125, 20.26727294921875, 0.1321258544921875, 81.49944305419922, -27.06634521484375, -108.38076782226562, 16.6964168548584, 127.48234558105469, 173.5018768310547, 165.38917541503906, -147.82138061523438, 0.8830699920654297, 143.6820068359375, -7.219490051269531, 179.32373046875, 3.08697509765625, 3.9505615234375, -143.66497802734375, -9.732677459716797, 43.04376220703125, 135.31053161621094, -135.45126342773438, 217.8697509765625, -78.59452819824219, -43.125030517578125, 103.81316375732422, 191.95806884765625, 18.462127685546875, -49.98600769042969, -49.345977783203125, 141.82144165039062, 10.140655517578125, 141.28695678710938, 30.819915771484375, -3.1036529541015625, 148.3972625732422, 10.96728515625, 22.525802612304688, 13.9346923828125, 109.47911071777344, -41.5189208984375, 55.28949737548828, 109.58064270019531, 35.06282043457031, -18.850181579589844, -120.24948120117188, -40.150421142578125, -126.5668716430664, 15.38125991821289, 164.96548461914062, 148.47325134277344, 75.29435729980469, 52.090179443359375, -1.8048858642578125, 113.58843994140625, -5.86372184753418, 160.46560668945312, 130.98727416992188, 111.65018463134766, -134.13241577148438, 151.244384765625, 25.731689453125, 112.06471252441406, 95.2718505859375, -122.313232421875, 207.64422607421875, 9.425434112548828, 39.7950439453125, 34.76862335205078, 40.5601806640625, 62.26678466796875, -14.63421630859375, -106.53363037109375, 98.31491088867188, 24.291702270507812, 146.4439697265625, 117.11895751953125, 133.68603515625, 27.30694580078125, 14.4388427734375, 13.12841796875, 133.2139129638672, 87.72552490234375, 217.63519287109375, 40.853912353515625, 183.15493774414062, 147.32363891601562, 29.62249755859375, 241.12353515625, 127.88223266601562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000288.npy"}
{"epoch": 0.6031413612565445, "step": 289, "batch_size": 128, "mean": 31.806005477905273, "std": 82.3008041381836, "min": -165.1478271484375, "p10": -65.75232543945312, "median": 14.590606689453125, "p90": 143.04779663085938, "max": 183.5059814453125, "pos_frac": 0.6171875, "sample": [-46.82861328125, -22.928619384765625, 14.61199951171875, -77.3120346069336, -5.13861083984375, -0.428192138671875, 13.2880859375, -26.08892822265625, 76.1484375, -49.50201416015625, 137.98861694335938, -5.3249359130859375, 1.260239601135254, 46.266357421875, 0.7614593505859375, 17.327239990234375, -30.438003540039062, 159.15789794921875, 19.666915893554688, -69.60382080078125, 179.08816528320312, 158.60769653320312, 168.10098266601562, 66.55403900146484, -5.822265625, -24.950347900390625, -24.969970703125, 3.5250244140625, 11.028533935546875, 73.39707946777344, -3.7505645751953125, 139.185302734375, -2.709442138671875, 17.72515869140625, 1.5058059692382812, 0.0, 8.758201599121094, 102.06390380859375, -28.92718505859375, 14.5692138671875, 11.406244277954102, -27.60040283203125, -122.79415893554688, -137.88949584960938, 79.56915283203125, -32.930755615234375, 54.686279296875, 133.17877197265625, -12.8389892578125, -2.913238525390625, -8.387664794921875, 144.69833374023438, -143.77099609375, -148.11648559570312, 14.50091552734375, 2.33306884765625, -14.021821975708008, -6.726905822753906, 111.46099090576172, 170.58218383789062, -102.127685546875, 70.15765380859375, 102.786865234375, 138.19046020507812, 42.560791015625, 127.36511993408203, -82.69734191894531, -125.69854736328125, -103.20185089111328, -3.12200927734375, 142.0516357421875, -50.1806640625, -5.725555419921875, 121.32395935058594, 18.042808532714844, 16.272781372070312, 93.16354370117188, 44.72607421875, 114.67578125, 5.960845947265625, 57.487152099609375, -22.383209228515625, -165.1478271484375, 152.535400390625, -16.402557373046875, 95.29882049560547, -7.053497314453125, 140.1898956298828, -44.44194030761719, 149.592529296875, -11.6085205078125, 168.83944702148438, 133.237060546875, 10.7393798828125, 37.74458312988281, 109.93170166015625, 141.84710693359375, 8.462738037109375, 106.66110229492188, 93.82258605957031, 146.53341674804688, 34.569725036621094, 172.01617431640625, 114.4608154296875, -23.18450927734375, 134.38082885742188, 24.082305908203125, 70.28254699707031, 52.458892822265625, 183.5059814453125, -120.5032958984375, 133.11892700195312, 142.6251983642578, -37.55989074707031, 144.0338592529297, -64.1016845703125, -131.94277954101562, 25.977500915527344, -34.48326110839844, 140.94049072265625, 32.629058837890625, 26.540924072265625, 119.82275390625, 4.286598205566406, 16.280487060546875, -6.505149841308594, -31.934173583984375, 28.67254638671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000289.npy"}
{"epoch": 0.6052356020942409, "step": 290, "batch_size": 128, "mean": 38.0391845703125, "std": 80.19786071777344, "min": -162.60757446289062, "p10": -52.332630920410146, "median": 20.945907592773438, "p90": 144.67974243164062, "max": 240.9862060546875, "pos_frac": 0.6484375, "sample": [155.26893615722656, 20.30706787109375, -9.2822265625, 116.70648193359375, 62.486053466796875, -162.60757446289062, -3.0100250244140625, 111.56109619140625, -16.5604248046875, -25.6607666015625, 177.8641357421875, 115.30239868164062, 46.30999755859375, -14.577438354492188, 134.48805236816406, 118.30339050292969, 10.162162780761719, 35.479522705078125, 150.56192016601562, 135.33056640625, 134.51953125, -29.645751953125, 2.6690673828125, -6.14141845703125, 193.69583129882812, 4.93426513671875, -101.95243835449219, -102.09259033203125, -22.678924560546875, 2.038360595703125, 92.9639892578125, 5.931339263916016, -2.237762451171875, 59.15289306640625, 116.69427490234375, 167.68719482421875, 157.0229949951172, 120.66244506835938, -1.64453125, -135.03204345703125, 117.09857940673828, 86.20013427734375, -1.3861083984375, 135.4376220703125, 60.003692626953125, 0.1455841064453125, 0.6259765625, -82.2025146484375, 81.75102233886719, 114.75592041015625, -16.2188720703125, 2.4376373291015625, 113.45719909667969, 29.480131149291992, 12.854347229003906, -30.173675537109375, 42.555816650390625, 120.30191040039062, -27.657928466796875, 21.584747314453125, 13.407196044921875, 15.0125732421875, -19.314910888671875, 118.06005859375, 17.578369140625, 26.362762451171875, 17.777488708496094, 72.9486083984375, 55.46125793457031, 100.1329345703125, -32.28790283203125, 41.5595703125, 3.4013671875, -15.90008544921875, -22.3431396484375, 176.998779296875, -7.004417419433594, -49.74932861328125, -3.707000732421875, 23.5135498046875, -1.39971923828125, -13.993354797363281, -58.36033630371094, 141.62799072265625, 136.15036010742188, 138.33721923828125, 148.4595184326172, -45.767486572265625, -152.96923828125, 7.6757659912109375, 52.31365966796875, -31.920852661132812, 123.72659301757812, -12.5172119140625, -96.50137329101562, 4.809822082519531, -82.81910705566406, -11.6341552734375, -72.43650817871094, 119.82247924804688, 8.815399169921875, -100.25597381591797, 43.3636474609375, 240.9862060546875, 71.890869140625, 42.258697509765625, 143.52520751953125, 147.3736572265625, 148.39727783203125, 103.1604232788086, -115.15591430664062, 22.07684326171875, 116.8016357421875, 48.4549560546875, 80.87957763671875, -61.58988952636719, -9.549369812011719, 1.7366943359375, -33.960357666015625, 123.60049438476562, -17.571044921875, -2.0716705322265625, 25.9017333984375, 125.49354553222656, -48.41290283203125, 32.752227783203125, 159.57772827148438, 149.9930419921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000290.npy"}
{"epoch": 0.6073298429319371, "step": 291, "batch_size": 128, "mean": 58.45766830444336, "std": 84.9409408569336, "min": -184.25277709960938, "p10": -46.836405181884764, "median": 49.752197265625, "p90": 160.89172058105467, "max": 207.95330810546875, "pos_frac": 0.7109375, "sample": [166.83694458007812, -0.058807373046875, 190.9852294921875, 17.9111328125, 184.514892578125, 99.32374572753906, 149.16415405273438, 138.727294921875, 105.38284301757812, -94.77015686035156, -8.25927734375, 115.596435546875, 32.04290008544922, -6.81732177734375, 162.79473876953125, 13.522689819335938, 159.08294677734375, -15.905471801757812, 124.7764892578125, 27.037321090698242, 116.95742797851562, -11.151580810546875, 120.95645141601562, 17.343109130859375, -11.722671508789062, 145.29034423828125, 41.26087951660156, 9.3890380859375, 121.80294799804688, -27.104461669921875, -145.14610290527344, -19.07876968383789, -78.93019104003906, 20.238658905029297, -184.25277709960938, 27.755096435546875, 120.2818603515625, 107.28369140625, 45.767822265625, 6.191131591796875, 160.07614135742188, -49.7159423828125, -100.41665649414062, 92.84036254882812, 154.46043395996094, 34.816253662109375, -4.174468994140625, 116.88459777832031, 118.16571044921875, 75.91940307617188, 154.78887939453125, 108.53507232666016, 108.26656341552734, -19.705963134765625, 207.95330810546875, 122.38191223144531, 93.13455200195312, -38.888824462890625, -59.73448181152344, 128.51840209960938, 165.50515747070312, 11.473846435546875, -14.23931884765625, -1.870147705078125, 154.58468627929688, -71.59585571289062, 8.583633422851562, -6.37451171875, -99.23501586914062, 158.7371063232422, -97.07777404785156, 16.832794189453125, 68.20292663574219, 199.306884765625, 118.71607971191406, 23.63018798828125, 124.0268783569336, 136.30548095703125, 155.498291015625, -8.0089111328125, 75.76861572265625, 0.6685390472412109, -13.351043701171875, -19.8248291015625, 22.260940551757812, 166.23419189453125, 104.99577331542969, -95.70211791992188, 0.0, -20.105052947998047, -75.43145751953125, 119.35629272460938, 25.237716674804688, -45.602317810058594, 197.8223876953125, 125.51631164550781, 154.51333618164062, 127.67727661132812, 163.39764404296875, -30.6927490234375, 76.21939086914062, 143.56826782226562, 32.598388671875, 124.60116577148438, 0.875823974609375, 67.197021484375, 16.4276123046875, 37.8486328125, 190.70526123046875, 53.736572265625, -94.20199584960938, 109.0677490234375, -7.119194030761719, 108.42188262939453, 136.82664489746094, 25.2449951171875, 134.52716064453125, -0.6518154144287109, 170.6912841796875, 149.06906127929688, 101.00637817382812, 28.764389038085938, 2.8822021484375, -10.867542266845703, 156.569580078125, 163.81268310546875, 125.16766357421875, 24.724349975585938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000291.npy"}
{"epoch": 0.6094240837696335, "step": 292, "batch_size": 128, "mean": 47.172149658203125, "std": 79.62665557861328, "min": -200.9761962890625, "p10": -40.44688644409179, "median": 35.00785446166992, "p90": 146.97350158691404, "max": 224.592529296875, "pos_frac": 0.75, "sample": [12.617401123046875, 4.9762420654296875, 126.97020721435547, -66.61956787109375, 125.48558044433594, 130.52618408203125, 143.3594207763672, 205.92822265625, 21.697158813476562, 149.34320068359375, 212.416748046875, 107.33351135253906, -30.4658203125, -11.292236328125, 57.046905517578125, 26.741851806640625, -38.72480010986328, -9.900800704956055, 25.3798828125, -5.21563720703125, 41.54052734375, 35.74818420410156, 39.03656005859375, 67.647216796875, 113.38298797607422, 109.16165161132812, 141.1974639892578, 28.294769287109375, 25.94525146484375, -5.681713104248047, 142.50442504882812, 120.87467956542969, 1.1803550720214844, 41.87245178222656, 121.51348876953125, 26.49209213256836, 45.184814453125, -75.75753784179688, -200.9761962890625, 10.666343688964844, 97.86241149902344, 65.1120376586914, 108.25607299804688, 34.26752471923828, -1.2733154296875, 7.859397888183594, 73.30687713623047, -107.39718627929688, 1.1636123657226562, 157.50604248046875, -109.6474609375, 164.88958740234375, -85.16650390625, -14.082260131835938, -3.08270263671875, 123.01628112792969, 0.42523193359375, -70.79910278320312, 73.408447265625, 13.828094482421875, 123.43743133544922, 15.8753662109375, 135.53248596191406, -70.42454528808594, -12.108154296875, 15.126502990722656, 11.33721923828125, 22.759857177734375, 224.592529296875, 56.713623046875, 7.3182373046875, 16.167694091796875, 102.30374145507812, 36.81231689453125, 127.3720474243164, 178.91522216796875, 26.82400131225586, 152.68650817871094, 29.222919464111328, -16.416839599609375, -57.579498291015625, -192.85968017578125, 182.22369384765625, 44.03080749511719, 145.95791625976562, 105.78543090820312, 65.23455810546875, -11.096649169921875, 8.237529754638672, 155.84243774414062, -37.17584228515625, 21.031639099121094, -44.465087890625, 19.0338134765625, 89.21206665039062, -27.02033233642578, -10.719619750976562, -25.68048095703125, 16.794605255126953, 159.716552734375, 104.87620544433594, 119.07784271240234, 19.01629638671875, 83.81072998046875, 26.685630798339844, 82.91796875, 50.005340576171875, -0.2025146484375, -7.9469757080078125, 6.1702728271484375, 12.117828369140625, 50.0887451171875, 44.934600830078125, 0.09515380859375, 140.08163452148438, 135.22970581054688, 191.28018188476562, -46.80596923828125, 100.13607788085938, 100.82244873046875, 157.9449462890625, -133.21588134765625, -20.208885192871094, 121.70001220703125, 46.19671630859375, 136.8802490234375, 122.22543334960938, 56.68428039550781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000292.npy"}
{"epoch": 0.6115183246073298, "step": 293, "batch_size": 128, "mean": 50.42085266113281, "std": 78.458740234375, "min": -155.13583374023438, "p10": -31.429324340820305, "median": 31.769058227539062, "p90": 158.792724609375, "max": 213.09136962890625, "pos_frac": 0.7578125, "sample": [156.31285095214844, -7.8226318359375, 140.43614196777344, 171.02459716796875, 25.992706298828125, -76.22100830078125, 17.014312744140625, 113.958740234375, 3.1660308837890625, 101.82463836669922, -6.3440093994140625, 124.91130828857422, 41.99554443359375, 21.153778076171875, 29.48952293395996, 16.325119018554688, -37.839637756347656, 0.45721435546875, -97.57490539550781, 14.127288818359375, 33.369056701660156, 1.71929931640625, 166.26144409179688, 128.93582153320312, 19.247833251953125, 0.0, 166.87664794921875, 37.64732360839844, 165.65646362304688, 19.02399444580078, 130.58795166015625, -29.18438720703125, -11.010452270507812, 4.591583251953125, 15.565185546875, 32.02703857421875, 18.9183349609375, 135.2071990966797, -2.23492431640625, 62.58283996582031, 154.32852172851562, -2.0419864654541016, 136.54266357421875, 148.2591552734375, 22.1805419921875, -155.13583374023438, 158.11871337890625, 149.0363006591797, 190.20086669921875, 35.167266845703125, 75.24082946777344, 44.26727294921875, 131.83460998535156, 104.50689697265625, 1.792633056640625, 118.78694152832031, 27.651092529296875, -10.52789306640625, 5.8139801025390625, 21.658905029296875, -9.209281921386719, 156.76075744628906, 61.1658935546875, -55.1326904296875, -136.6370849609375, 65.1346435546875, 14.425209045410156, -40.940948486328125, 121.15254211425781, -8.379493713378906, -147.8265380859375, -36.667510986328125, 119.62092590332031, 213.09136962890625, -79.6881103515625, 11.22406005859375, -8.70599365234375, 130.60848999023438, -46.037353515625, 40.959381103515625, 24.946380615234375, 57.59576416015625, 51.2216796875, 17.84808349609375, 160.36541748046875, 30.4332275390625, 51.087310791015625, 141.88348388671875, 110.61810302734375, -27.307052612304688, -17.01275634765625, 169.41473388671875, 57.27088928222656, 143.60568237304688, 50.194305419921875, 31.511077880859375, 91.15657043457031, -115.59100341796875, 41.81695556640625, 15.571109771728516, 46.35969543457031, 193.12026977539062, 208.5615234375, 189.06607055664062, 64.6094970703125, 135.54241943359375, 20.242820739746094, 28.8809814453125, -45.71150207519531, 0.1802043914794922, -20.2420654296875, 25.693267822265625, -3.618499755859375, 80.68408203125, 211.01437377929688, 203.1483154296875, 14.74908447265625, 32.30274963378906, 128.7218475341797, -15.40948486328125, 66.11648559570312, -10.661277770996094, 8.96783447265625, 114.97088623046875, 55.14118194580078, -3.373321533203125, 40.761749267578125, 26.644989013671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000293.npy"}
{"epoch": 0.6136125654450262, "step": 294, "batch_size": 128, "mean": 44.23646926879883, "std": 82.44789123535156, "min": -144.26040649414062, "p10": -47.349623107910155, "median": 26.00989532470703, "p90": 153.25788269042968, "max": 220.7156982421875, "pos_frac": 0.734375, "sample": [151.40628051757812, 121.7822265625, 144.61160278320312, 1.3598175048828125, -11.099609375, 9.953742980957031, 0.8693714141845703, 2.3568115234375, 16.69298553466797, 149.62301635742188, 46.20343017578125, 170.15179443359375, 134.234375, 4.5169525146484375, -3.632598876953125, 7.534229278564453, 140.62222290039062, 177.1884307861328, 150.61758422851562, 27.048583984375, 133.2127685546875, 137.03309631347656, 120.53173828125, -80.99889373779297, 22.77947235107422, 153.064697265625, 10.9464111328125, -26.451614379882812, 10.91302490234375, 19.322021484375, 176.47210693359375, 32.726837158203125, 12.445419311523438, 107.1846923828125, -7.754356384277344, 20.050506591796875, -35.4725341796875, -138.8460693359375, 44.529510498046875, 27.52972412109375, 86.89656066894531, 172.5184326171875, -49.1529541015625, -125.41230010986328, 154.95901489257812, 93.2638168334961, 116.40038299560547, -4.8144073486328125, -3.5536727905273438, 39.83306884765625, 152.40121459960938, -5.7498779296875, 163.93020629882812, 144.88299560546875, 58.64885330200195, 155.57200622558594, 156.21864318847656, -25.094879150390625, 19.64642333984375, 12.722389221191406, 116.01666259765625, 27.696868896484375, 29.742156982421875, 135.07913208007812, -133.62884521484375, 96.4931411743164, 115.23268127441406, 42.56951904296875, 5.480926513671875, 8.86431884765625, 15.463775634765625, 172.6960906982422, 158.66461181640625, 0.0, 147.27426147460938, 220.7156982421875, -116.58424377441406, -8.009201049804688, -16.1712646484375, 48.300201416015625, -4.761575698852539, 153.70864868164062, 7.335460662841797, -137.775146484375, -16.823394775390625, 44.863861083984375, -38.187957763671875, 5.60565185546875, 120.46113586425781, -46.57676696777344, 27.309066772460938, -2.3281631469726562, 32.49181365966797, -144.26040649414062, 21.57489776611328, 112.81523132324219, 24.971206665039062, 3.4592208862304688, 10.8258056640625, -16.836883544921875, 13.905723571777344, 128.55276489257812, 12.7547607421875, 28.470216751098633, 120.0233383178711, 61.47050476074219, -90.49212646484375, -127.401123046875, 38.7227783203125, -0.49941253662109375, -81.41416931152344, 95.65065002441406, 112.70780944824219, 15.54705810546875, 107.12660217285156, 90.2039794921875, 23.17577362060547, -27.839012145996094, 106.44010925292969, -16.496551513671875, 10.99615478515625, -67.80068969726562, 97.06192016601562, 2.824373245239258, 160.53570556640625, -140.798583984375, 122.50332641601562, 147.19229125976562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000294.npy"}
{"epoch": 0.6157068062827226, "step": 295, "batch_size": 128, "mean": 49.781620025634766, "std": 84.529052734375, "min": -142.38507080078125, "p10": -41.720105743408205, "median": 23.898731231689453, "p90": 154.78971405029296, "max": 237.982666015625, "pos_frac": 0.734375, "sample": [170.81060791015625, 21.787506103515625, -35.70562744140625, -126.79023742675781, -11.393447875976562, 63.68226623535156, -41.867820739746094, 146.73886108398438, -124.48187255859375, -13.72772216796875, -16.03392791748047, 22.896446228027344, 150.30780029296875, 18.34698486328125, -13.602081298828125, -22.119171142578125, -131.76730346679688, 125.99246215820312, -121.83975219726562, 28.337448120117188, 103.6021728515625, 189.8275909423828, -10.71929931640625, -142.38507080078125, -26.71044921875, 122.77758026123047, 21.580032348632812, -52.253265380859375, 120.6331787109375, -2.9103546142578125, 7.3985595703125, 25.2525634765625, 35.76019287109375, 112.97055053710938, 141.07664489746094, 9.4476318359375, 116.40139770507812, 12.581298828125, 70.1270751953125, 101.3486328125, 194.75782775878906, -10.956562042236328, 108.3726806640625, 117.83380126953125, -1.616302490234375, 80.19732666015625, 12.987388610839844, 179.00320434570312, 145.30914306640625, 129.938720703125, -130.24423217773438, -88.61474609375, 154.45655822753906, 237.982666015625, 144.6831817626953, 201.47296142578125, -23.02471923828125, 110.73783874511719, 55.258819580078125, 11.039497375488281, -17.275848388671875, 4.162567138671875, 39.432464599609375, 3.4083786010742188, 126.5833740234375, 78.75740814208984, 18.552207946777344, 79.26547241210938, 124.84185791015625, 19.959442138671875, 6.293975830078125, 8.46142578125, 1.6484508514404297, 154.43743896484375, -30.86065673828125, 131.16087341308594, 150.8152618408203, 134.14974975585938, 13.384048461914062, 107.265625, 9.249042510986328, 31.259735107421875, 72.11138153076172, 49.81273651123047, 7.3733673095703125, 21.420135498046875, 0.0, 98.29632568359375, 5.399162292480469, 15.92120361328125, 153.8131103515625, 183.04342651367188, 23.825439453125, 48.871429443359375, 150.44393920898438, 0.7279396057128906, -79.84957885742188, 174.361083984375, 35.3157844543457, 154.02345275878906, 116.70452880859375, -29.35504150390625, 9.54751968383789, 22.71435546875, 29.20068359375, -123.25973510742188, 143.76211547851562, 17.515308380126953, 2.0125579833984375, 103.77113342285156, -25.940380096435547, 147.2660675048828, 221.347900390625, 158.56405639648438, -1.946990966796875, 180.64883422851562, 0.27434539794921875, -7.927513122558594, 23.972023010253906, -48.7784423828125, -41.65679931640625, 8.5538330078125, 155.56707763671875, -7.16168212890625, 130.235595703125, 164.16639709472656, 90.27023315429688, -52.8365478515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000295.npy"}
{"epoch": 0.6178010471204188, "step": 296, "batch_size": 128, "mean": 65.61614990234375, "std": 85.37946319580078, "min": -160.84506225585938, "p10": -32.727404785156246, "median": 67.53662109375, "p90": 167.5974334716797, "max": 198.39593505859375, "pos_frac": 0.7734375, "sample": [124.56487274169922, 25.542293548583984, 18.56854248046875, 128.71826171875, 130.165283203125, -17.301963806152344, -0.7961578369140625, -1.3288898468017578, 87.67522430419922, 7.571056365966797, 106.3094482421875, -97.804443359375, 49.516357421875, 41.17108154296875, 9.572967529296875, 137.12841796875, 149.2822265625, 123.6982421875, 130.79200744628906, -76.913330078125, 142.34381103515625, 115.506103515625, -11.713623046875, 141.8155059814453, 62.585540771484375, 169.81784057617188, -10.746540069580078, 16.36540985107422, 156.32452392578125, -46.925140380859375, 14.92095947265625, 93.44352722167969, 189.192626953125, 6.889690399169922, -145.66006469726562, 179.6954345703125, 51.25311279296875, 141.5181884765625, 151.8684844970703, 45.20134735107422, -25.477584838867188, 20.36907958984375, 176.94390869140625, 141.5555419921875, 5.317176818847656, -31.2635498046875, 36.83782958984375, -59.256195068359375, 194.559326171875, 14.9107666015625, 172.71646118164062, 92.10719299316406, -14.034759521484375, 160.15069580078125, 166.986328125, 191.34494018554688, 12.551368713378906, 176.24676513671875, 76.58889770507812, 132.45201110839844, 13.59356689453125, -24.67046356201172, 134.9572296142578, -75.70651245117188, 52.37237548828125, 111.49436950683594, 29.6353759765625, 158.891357421875, -135.5074920654297, 92.07879638671875, 146.16305541992188, 69.87103271484375, -126.13394165039062, 139.004150390625, 111.44430541992188, 45.975677490234375, 178.97744750976562, 194.44332885742188, -2.7478408813476562, 129.40402221679688, 65.20220947265625, 149.06378173828125, 12.897750854492188, -160.84506225585938, 164.43826293945312, 122.47724914550781, -0.99664306640625, -95.48106384277344, 169.02334594726562, -7.902374267578125, 129.2928009033203, 6.52496337890625, 113.72683715820312, -44.207855224609375, 43.25445556640625, 0.0, 130.462890625, 137.6485595703125, 135.739990234375, 105.21038818359375, 23.020538330078125, 151.26498413085938, 76.50465393066406, 139.459228515625, 8.125831604003906, 41.362823486328125, 13.089385986328125, 14.234039306640625, 47.43743896484375, 136.65115356445312, 182.16864013671875, 103.81610107421875, 4.304080963134766, -36.14306640625, -7.961906433105469, 146.8192138671875, 142.72943115234375, 1.497711181640625, -3.72613525390625, 160.5489044189453, 6.295097351074219, 109.80624389648438, 22.6431884765625, 161.51007080078125, 107.5701904296875, 198.39593505859375, -144.6588134765625, -8.401992797851562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000296.npy"}
{"epoch": 0.6198952879581152, "step": 297, "batch_size": 128, "mean": 51.13225555419922, "std": 81.28225708007812, "min": -153.45993041992188, "p10": -34.59688110351563, "median": 30.479461669921875, "p90": 160.22325439453124, "max": 191.64007568359375, "pos_frac": 0.6796875, "sample": [185.91842651367188, 77.46939086914062, 141.0342254638672, 160.049560546875, -31.945602416992188, -1.2242355346679688, 143.28854370117188, 53.763214111328125, 160.6285400390625, -25.876907348632812, -2.2235260009765625, -2.24383544921875, -29.064987182617188, 6.2929229736328125, 169.400634765625, -97.6517333984375, -8.95801067352295, 32.613067626953125, 148.74058532714844, 108.66812133789062, 9.467658996582031, 36.03411865234375, -0.39006805419921875, 50.8497314453125, -85.96832275390625, -63.308837890625, -20.956756591796875, -153.45993041992188, 12.34228515625, -21.96658706665039, -57.78521728515625, 120.60260009765625, 28.505111694335938, 84.0025634765625, 3.1301422119140625, 8.621841430664062, 152.62496948242188, -0.256744384765625, 191.6282958984375, 24.160186767578125, -13.976051330566406, 11.863418579101562, 93.38044738769531, -34.4285888671875, 3.730743408203125, -17.39532470703125, -3.44976806640625, 23.7882080078125, -121.04632568359375, 147.14990234375, 37.216339111328125, -23.475738525390625, 0.31201171875, -34.98956298828125, 0.2679443359375, 191.64007568359375, 145.94342041015625, 114.46354675292969, -3.598499298095703, -25.333740234375, 92.50460815429688, 140.30831909179688, -143.772705078125, 25.969711303710938, 174.98435974121094, 34.244384765625, 15.75823974609375, -66.39279174804688, 168.1595458984375, -10.384185791015625, 31.350006103515625, 116.98553466796875, 171.6815185546875, 163.96681213378906, 185.8704833984375, -63.01043701171875, 30.4630126953125, 4.74444580078125, 40.019775390625, -7.693290710449219, -93.33462524414062, 140.00059509277344, 65.97439575195312, -8.508232116699219, 1.9090328216552734, 29.170547485351562, 39.98439025878906, -10.923271179199219, 155.00291442871094, 8.4793701171875, 134.3668212890625, -11.22113037109375, 85.49586486816406, 130.129638671875, 126.56298828125, 107.53831481933594, 7.9137420654296875, 156.66880798339844, -3.6522064208984375, 176.81585693359375, 118.74215698242188, 148.03717041015625, 97.02641296386719, 147.88717651367188, -9.173812866210938, 117.68405151367188, 91.94967651367188, 30.49591064453125, 132.99966430664062, 141.67742919921875, 175.6478271484375, 133.13421630859375, 2.624053955078125, 54.29530334472656, -3.2283935546875, -24.162796020507812, -37.2008056640625, 97.1712646484375, 135.2188720703125, 19.73468017578125, 17.071517944335938, 140.19766235351562, 176.85195922851562, -45.177886962890625, -3.839324951171875, 59.42181396484375, 129.15835571289062, 157.935791015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000297.npy"}
{"epoch": 0.6219895287958115, "step": 298, "batch_size": 128, "mean": 57.84134292602539, "std": 89.3624496459961, "min": -173.84356689453125, "p10": -43.979132080078124, "median": 35.82292175292969, "p90": 167.2574462890625, "max": 324.503173828125, "pos_frac": 0.71875, "sample": [220.3340301513672, 130.52316284179688, -19.294830322265625, 16.344223022460938, -43.54541015625, -44.99114990234375, -5.331207275390625, -11.304779052734375, -23.8978271484375, 168.07711791992188, 96.91989135742188, -11.914443969726562, 9.449203491210938, 12.618743896484375, 133.5092010498047, -96.51666259765625, 156.69082641601562, 108.31166076660156, 34.773956298828125, 124.1928482055664, 158.77841186523438, 152.87033081054688, 39.797882080078125, 210.463134765625, 9.978286743164062, -36.68328857421875, 138.20611572265625, -119.13226318359375, 144.39259338378906, 91.56602478027344, 166.90615844726562, -38.457366943359375, 136.1151123046875, 58.10084533691406, 32.35833740234375, -57.01751708984375, 7.2265777587890625, 71.86648559570312, 149.63705444335938, -64.9451904296875, 130.94259643554688, -15.196502685546875, 137.93765258789062, -20.686553955078125, 42.118072509765625, 176.48193359375, 43.2127685546875, 20.110687255859375, -122.86387634277344, -30.264785766601562, 33.26252746582031, 143.7587890625, 21.0941162109375, 157.94229125976562, -82.86685180664062, 55.41180419921875, 138.14891052246094, 164.8333740234375, 14.796661376953125, 139.805908203125, 197.93991088867188, -173.84356689453125, 139.5299072265625, -14.503921508789062, -12.735519409179688, 200.60903930664062, 171.2267303466797, 47.98291015625, -78.36358642578125, 0.0, 5.301116943359375, 8.105598449707031, 189.941162109375, 51.1212158203125, 24.52953338623047, -5.521793365478516, 132.11199951171875, 97.99964904785156, 9.609001159667969, 147.83721923828125, -56.7457275390625, -6.478981018066406, 15.049209594726562, 87.65249633789062, 137.67611694335938, 5.988151550292969, 157.4886016845703, 147.020751953125, 146.90228271484375, 12.62445068359375, 222.468994140625, 37.54254150390625, -52.90375518798828, -10.09368896484375, 15.591598510742188, 34.101348876953125, 92.14659118652344, 14.246337890625, -6.876808166503906, -7.2374267578125, -8.347625732421875, 141.72988891601562, 65.635498046875, 149.1097412109375, 206.77175903320312, 184.30819702148438, -22.176300048828125, 117.31759643554688, 324.503173828125, 2.1861572265625, 152.16836547851562, 35.07232666015625, 3.6605377197265625, 36.573516845703125, 165.05136108398438, 11.26373291015625, -13.58599853515625, 143.12677001953125, 40.4730224609375, 184.19207763671875, 5.046709060668945, 111.64291381835938, 17.272735595703125, 15.262603759765625, -91.28350830078125, -23.650634765625, 62.066436767578125, -91.69422912597656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000298.npy"}
{"epoch": 0.6240837696335079, "step": 299, "batch_size": 128, "mean": 47.15193557739258, "std": 76.36326599121094, "min": -163.31283569335938, "p10": -28.120922851562497, "median": 24.54034423828125, "p90": 147.79349060058593, "max": 193.80654907226562, "pos_frac": 0.703125, "sample": [120.07588195800781, 147.38336181640625, 20.05704116821289, 110.52268981933594, 22.6859130859375, 3.05987548828125, 152.0423583984375, -112.89273071289062, 132.66461181640625, -163.31283569335938, 168.90115356445312, 42.096893310546875, 22.14038848876953, 118.55892944335938, -27.2738037109375, 143.67559814453125, -5.626007080078125, 22.876564025878906, 13.73040771484375, -137.7401885986328, 140.89739990234375, -3.194488525390625, 51.67034912109375, -92.36790466308594, 150.2452392578125, -5.518035888671875, -54.0771484375, -34.372467041015625, 129.91554260253906, 23.532318115234375, 115.10516357421875, -6.612457275390625, 149.06468200683594, -2.0762176513671875, -106.7376708984375, 88.69296264648438, 123.16644287109375, -12.407958984375, -81.06742858886719, 131.21791076660156, 10.99078369140625, 178.07115173339844, 25.548370361328125, 11.829732894897461, -8.762657165527344, -0.12841796875, 152.72198486328125, -4.66789436340332, -5.44525146484375, -107.71139526367188, -108.66168212890625, 17.147186279296875, 10.210224151611328, 17.988265991210938, 123.2642822265625, -35.229736328125, 183.32769775390625, -7.735321044921875, 180.2691650390625, 55.188743591308594, 132.42306518554688, 44.857330322265625, 18.394729614257812, 125.21896362304688, -2.2157936096191406, 193.80654907226562, 3.5208740234375, 11.741069793701172, 133.17901611328125, 34.60382080078125, 122.4501953125, 9.64004898071289, 63.19903564453125, 8.34625244140625, 125.58155822753906, 125.04804992675781, -10.212425231933594, -6.097492218017578, -4.3936614990234375, 138.62188720703125, -5.198638916015625, 92.78359985351562, 52.879547119140625, 13.79644775390625, 76.56684875488281, -5.3593902587890625, 145.73663330078125, 44.0352783203125, 86.32093811035156, 114.67633056640625, 16.626495361328125, 112.53634643554688, 153.91490173339844, -13.689239501953125, 20.06540870666504, -9.47137451171875, 43.770225524902344, 8.232322692871094, 131.26773071289062, -15.029930114746094, -30.0975341796875, 38.17503356933594, 33.1937255859375, -22.602508544921875, 142.1238250732422, 188.90866088867188, 15.600296020507812, 148.75045776367188, 28.8748779296875, -0.8897705078125, 13.45172119140625, 2.208404541015625, 133.20376586914062, -38.5089111328125, 9.41156005859375, 12.24615478515625, -2.410797119140625, 35.3333740234375, 73.84489440917969, 38.07171630859375, -14.821784973144531, 112.37350463867188, 161.04945373535156, 129.49664306640625, 28.5128173828125, 138.99734497070312, 26.97467041015625, 114.88510131835938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000299.npy"}
{"epoch": 0.6261780104712041, "step": 300, "batch_size": 128, "mean": 48.82493591308594, "std": 87.54669189453125, "min": -142.7777099609375, "p10": -60.02279815673828, "median": 33.783002853393555, "p90": 158.20391845703125, "max": 277.905029296875, "pos_frac": 0.671875, "sample": [155.61923217773438, 140.82577514648438, 22.298091888427734, -102.83052062988281, 18.611648559570312, -28.14080810546875, 134.28915405273438, 83.94163513183594, 8.286344528198242, -52.2825927734375, 170.2633056640625, 148.00711059570312, 157.59536743164062, 8.072689056396484, -3.0337600708007812, 146.83074951171875, -125.30441284179688, 33.2418212890625, 89.03941345214844, -7.201629638671875, 166.82858276367188, 76.262939453125, 47.87782287597656, 121.57807922363281, 17.433767318725586, 7.7978515625, 41.79681396484375, -15.591766357421875, 48.5782470703125, -43.665428161621094, 134.30300903320312, 114.22219848632812, 145.68984985351562, 132.46417236328125, 169.14578247070312, -60.928955078125, -141.412841796875, 14.08172607421875, 4.082645416259766, 199.47467041015625, 66.25735473632812, -18.91351318359375, -41.406158447265625, -0.5716552734375, 118.0228271484375, 142.54324340820312, -78.16777038574219, 146.427490234375, 133.6195068359375, 186.00778198242188, 7.379558563232422, -126.40518188476562, -113.63108825683594, 26.009979248046875, 34.32418441772461, -2.0227928161621094, 50.044891357421875, 69.10862731933594, 159.62387084960938, -9.231666564941406, 136.21466064453125, 139.7264404296875, -43.1346435546875, 169.78250122070312, 40.46685791015625, -17.072967529296875, 2.82330322265625, 117.4976806640625, 115.84208679199219, 140.9767608642578, 54.83807373046875, 109.25518798828125, 0.9150924682617188, 17.04888916015625, -21.959121704101562, -113.24653625488281, 179.9133758544922, 55.866943359375, -66.21511840820312, 145.04229736328125, 167.85897827148438, 14.891315460205078, 49.952789306640625, -59.63444519042969, 125.14310455322266, 18.020416259765625, -7.90399169921875, 0.0, -17.072830200195312, 35.55987548828125, 184.85226440429688, -7.38201904296875, -142.7777099609375, 117.47689819335938, 2.0052337646484375, 149.98281860351562, 105.41677856445312, -3.143768310546875, 115.38470458984375, 142.5181884765625, 15.853179931640625, 277.905029296875, 7.0139312744140625, 82.67262268066406, -19.761444091796875, 111.9781494140625, 154.58035278320312, -63.9935302734375, -39.9530029296875, -16.328689575195312, 116.43243408203125, -8.215009689331055, 19.978363037109375, -22.438095092773438, 61.34613037109375, 128.5178680419922, 111.66317749023438, -2.4630889892578125, -108.15798950195312, -0.50335693359375, 189.7601318359375, 1.578033447265625, -107.26290893554688, -29.586944580078125, 140.81121826171875, 162.95828247070312, -6.829959869384766, 9.1390380859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000300.npy"}
{"epoch": 0.6282722513089005, "step": 301, "batch_size": 128, "mean": 48.00785827636719, "std": 81.6325912475586, "min": -198.36285400390625, "p10": -32.36062927246093, "median": 27.11932373046875, "p90": 165.20315704345703, "max": 224.18829345703125, "pos_frac": 0.71875, "sample": [14.514297485351562, 186.03814697265625, 141.43548583984375, 101.94561767578125, 21.430091857910156, 152.75906372070312, 29.177284240722656, -43.77720642089844, 106.76557922363281, 8.983217239379883, 3.4868640899658203, 0.51068115234375, 144.45455932617188, 7.189697265625, -0.0672607421875, 7.916038513183594, -30.90423583984375, -44.639678955078125, 131.91534423828125, 174.80316162109375, 14.335128784179688, 14.032600402832031, 90.8343505859375, 26.8310546875, 39.83970642089844, 17.05340576171875, -0.49468994140625, 53.14208984375, 8.397979736328125, 65.68463134765625, 121.57562255859375, -12.627716064453125, 0.0, 172.1360626220703, 145.57113647460938, 94.5585708618164, -15.017181396484375, 34.57958984375, -8.38836669921875, 137.94113159179688, -9.75271224975586, 141.48353576660156, 172.53402709960938, 23.068222045898438, -163.03594970703125, -0.321044921875, -54.223358154296875, 126.69683837890625, -30.89801025390625, 21.76446533203125, 66.8614730834961, -11.08819580078125, -18.474777221679688, 167.4883575439453, 191.53314208984375, 69.72032928466797, 42.617889404296875, 173.68377685546875, 0.9736328125, 224.18829345703125, 30.771728515625, 127.95071411132812, 21.17340087890625, 35.09660339355469, -106.18673706054688, -51.78554153442383, 79.67385864257812, -198.36285400390625, -144.72122192382812, 77.13357543945312, 94.8057861328125, 17.818695068359375, -20.59576416015625, 22.023874282836914, 11.7767333984375, 31.55224609375, 151.43739318847656, 25.890167236328125, 164.22378540039062, -80.6693115234375, 8.005050659179688, -24.702392578125, 190.79136657714844, 118.64950561523438, 172.49371337890625, 28.58251953125, -5.792560577392578, -24.365097045898438, 127.24807739257812, -134.15574645996094, 149.31982421875, 35.6380615234375, -21.73834228515625, 2.7376556396484375, 145.1832275390625, -39.3572998046875, 112.12428283691406, 27.4075927734375, 114.3945083618164, 121.86944580078125, 135.49790954589844, 149.8272705078125, 168.40609741210938, -9.850677490234375, 29.302490234375, 158.51547241210938, 32.23846435546875, -14.355430603027344, 171.9063262939453, -14.670082092285156, 20.3280029296875, 1.9015350341796875, -10.54638671875, 11.022964477539062, 6.5496368408203125, 118.61331939697266, -48.430511474609375, 31.872081756591797, 4.315708160400391, 145.37103271484375, 179.66961669921875, -35.758880615234375, -20.3319091796875, 79.08658599853516, 24.657485961914062, 75.59706115722656, 119.8070068359375, -7.6185302734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000301.npy"}
{"epoch": 0.6303664921465969, "step": 302, "batch_size": 128, "mean": 58.14426040649414, "std": 85.57516479492188, "min": -164.22119140625, "p10": -35.77343139648434, "median": 50.8044319152832, "p90": 158.36203155517578, "max": 195.17181396484375, "pos_frac": 0.734375, "sample": [162.8224639892578, 59.02569580078125, 136.75918579101562, 167.267822265625, 22.440780639648438, 127.06316375732422, 0.0, -6.7026519775390625, -0.050079345703125, -74.94609069824219, -145.24359130859375, 135.81716918945312, 51.18006134033203, 155.86880493164062, -7.831878662109375, 36.94895935058594, 74.38287353515625, 45.884429931640625, -19.030624389648438, -88.76151275634766, 173.23129272460938, 130.70440673828125, 178.05535888671875, -13.37811279296875, 132.82107543945312, 141.77011108398438, 195.17181396484375, 19.705116271972656, -117.3470458984375, 119.43872833251953, 41.52809143066406, 148.97882080078125, 134.16693115234375, 122.70272827148438, 63.89902877807617, 116.431396484375, 181.24630737304688, 86.74720764160156, 52.5379638671875, 48.551544189453125, 132.2783203125, -1.0856857299804688, -164.22119140625, -4.935211181640625, 106.00000762939453, 85.51010131835938, 192.05352783203125, 23.917434692382812, 175.82635498046875, 155.08045959472656, -13.278900146484375, 149.38328552246094, 15.296592712402344, 3.3674545288085938, -7.008216857910156, -88.34152221679688, 131.23182678222656, 49.776885986328125, 128.15084838867188, 40.3675537109375, 123.89425659179688, 110.98516082763672, 45.62457275390625, 1.6647872924804688, 38.354248046875, 12.78924560546875, 14.68768310546875, 12.986026763916016, -5.200824737548828, -104.50324249267578, 128.1363525390625, 164.32183837890625, -24.275970458984375, -55.971405029296875, 151.24462890625, 130.51190185546875, -24.702232360839844, 99.0867919921875, 127.91726684570312, 117.69068145751953, 24.297164916992188, 71.12799072265625, 143.26705932617188, -123.72189331054688, 158.80697631835938, -7.1448974609375, -161.2357177734375, 78.82011413574219, -27.117156982421875, -113.77302551269531, 4.544797897338867, -116.04867553710938, 3.002410888671875, 47.6767578125, -1.336374282836914, 50.428802490234375, 120.60342407226562, 157.45883178710938, 156.17257690429688, -56.515899658203125, 2.678863525390625, 25.778533935546875, 50.131935119628906, -11.805004119873047, 140.23501586914062, 0.5067806243896484, 35.45770263671875, 189.896728515625, 83.45664978027344, 120.43002319335938, -5.38800048828125, 7.238746643066406, 54.9417724609375, -16.520050048828125, 147.01986694335938, 114.94314575195312, 192.26858520507812, 2.9664840698242188, 136.169189453125, 59.20036315917969, 24.1954345703125, 193.3099365234375, 126.62760925292969, 149.84226989746094, 106.07853698730469, -12.38714599609375, 158.1713409423828, -6.7624664306640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000302.npy"}
{"epoch": 0.6324607329842932, "step": 303, "batch_size": 128, "mean": 52.056339263916016, "std": 86.83002471923828, "min": -174.31658935546875, "p10": -36.14376525878905, "median": 47.900306701660156, "p90": 156.95894012451168, "max": 210.68365478515625, "pos_frac": 0.765625, "sample": [14.499786376953125, 14.982475280761719, -145.22811889648438, -130.6361541748047, 127.34429931640625, -9.019683837890625, 129.283447265625, 54.95111083984375, 141.90460205078125, 122.81015014648438, -33.5870361328125, 140.6256103515625, 154.70135498046875, 146.84664916992188, 40.296173095703125, -2.076324462890625, -10.050933837890625, 135.98989868164062, 1.4359512329101562, 166.19580078125, 115.9981689453125, 12.515899658203125, 127.6927261352539, 32.499053955078125, 62.922027587890625, -42.41351318359375, 120.00679016113281, -11.65411376953125, 173.23397827148438, 4.966758728027344, 168.05474853515625, 142.60894775390625, -42.109466552734375, 143.4362030029297, -11.297576904296875, 13.29443359375, 13.509185791015625, 47.51605224609375, 0.0, 6.3773040771484375, -119.10541534423828, 66.43415832519531, -17.770973205566406, 16.394996643066406, 11.142959594726562, 5.328094482421875, 144.8357696533203, 48.301177978515625, 120.67591857910156, 125.07476806640625, 134.58514404296875, -136.4183807373047, 128.99810791015625, -173.87100219726562, 84.96359252929688, 148.1790313720703, 166.47470092773438, 96.68046569824219, 41.656951904296875, 132.49334716796875, 0.7262077331542969, -4.1701812744140625, 24.363906860351562, -9.45458984375, -69.42544555664062, 9.49371337890625, 119.4432373046875, 131.95401000976562, 150.8184356689453, 101.4483413696289, 178.88348388671875, -129.002197265625, 21.527984619140625, -3.009002685546875, 142.55221557617188, 121.67926025390625, 0.90399169921875, 98.63119506835938, 147.72845458984375, 209.19488525390625, 68.8533706665039, -174.31658935546875, 121.76280212402344, -136.4727020263672, 7.312585830688477, 14.888427734375, 128.03265380859375, 59.08982849121094, 82.89749145507812, 88.9755859375, 13.819122314453125, -3.767578125, 162.2266387939453, 5.3151397705078125, 163.6842041015625, 1.8428802490234375, -3.8698577880859375, 48.28456115722656, -33.259765625, 6.344207763671875, -29.295318603515625, -157.817626953125, 57.19207763671875, -1.261505126953125, 6.525672912597656, 66.28369140625, 164.47625732421875, 59.719329833984375, 145.08595275878906, 117.06764221191406, 39.45367431640625, 76.12857055664062, 80.07026672363281, 131.34796142578125, -28.054351806640625, 210.68365478515625, 122.41412353515625, -118.38751220703125, 167.13348388671875, 12.329383850097656, 165.40673828125, 20.2452392578125, 187.15435791015625, 29.163970947265625, 33.864166259765625, 8.422454833984375, 9.183380126953125, 59.26490783691406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000303.npy"}
{"epoch": 0.6345549738219896, "step": 304, "batch_size": 128, "mean": 45.689971923828125, "std": 84.37744140625, "min": -158.89697265625, "p10": -44.765875244140624, "median": 25.855133056640625, "p90": 155.57933044433594, "max": 201.9617919921875, "pos_frac": 0.71875, "sample": [0.0, 29.628921508789062, -13.059112548828125, 100.23281860351562, -43.77972412109375, 0.420257568359375, -148.2384033203125, 148.39605712890625, 145.990478515625, -23.7657470703125, 110.28312683105469, 145.38946533203125, 102.42570495605469, 109.33210754394531, -80.70481872558594, 43.77471923828125, 15.13153076171875, 160.58773803710938, 25.32470703125, 14.9466552734375, 13.34033203125, 12.718769073486328, -14.361282348632812, -15.1671142578125, 132.09375, 145.67587280273438, 134.88314819335938, 19.906986236572266, 6.27923583984375, 1.0022430419921875, -17.696533203125, 17.837724685668945, 2.2290191650390625, -3.84271240234375, 36.091522216796875, 26.954498291015625, 42.94544219970703, 8.457450866699219, 103.01359558105469, 116.90283203125, -19.814682006835938, 107.66806030273438, 133.5712127685547, -33.7530517578125, 186.00482177734375, 19.708343505859375, -6.693168640136719, 132.862548828125, -103.38455200195312, 41.5474853515625, 37.802886962890625, -89.06072998046875, 77.45722961425781, -158.89697265625, 0.0, 156.9970245361328, -90.9054183959961, 117.0935287475586, 6.0673828125, 2.30615234375, -57.212005615234375, 24.8251953125, -124.49583435058594, 98.70817565917969, 21.47735595703125, 71.04417419433594, -38.64239501953125, 48.4111328125, 195.51416015625, 138.9141845703125, -122.30563354492188, 22.397369384765625, 44.61912536621094, -121.9417724609375, -47.06689453125, 172.1712646484375, 162.9754638671875, 3.5053634643554688, 161.75811767578125, -118.03988647460938, 145.90121459960938, 122.53219604492188, -26.94963836669922, -5.645938873291016, 201.9617919921875, 20.062545776367188, 155.52810668945312, -6.811737060546875, 148.52197265625, -150.1656494140625, -0.12786865234375, 1.5381546020507812, 26.38555908203125, 155.6988525390625, 162.4035186767578, 9.278093338012695, -22.024139404296875, 136.4105224609375, 147.37860107421875, 161.548095703125, 33.774993896484375, 33.13507080078125, 71.20695495605469, 8.562042236328125, 42.0216064453125, 147.06393432617188, 28.4609375, 175.96923828125, 143.3347930908203, 147.16519165039062, 123.35116577148438, 50.317138671875, -43.14857482910156, 8.64227294921875, 127.58906555175781, 16.1536865234375, -9.0443115234375, -18.227325439453125, 132.67889404296875, -17.336502075195312, 15.218315124511719, 117.15555572509766, 1.938568115234375, 141.15814208984375, -20.628143310546875, 145.03701782226562, 22.913116455078125, 175.65362548828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000304.npy"}
{"epoch": 0.6366492146596858, "step": 305, "batch_size": 128, "mean": 67.430419921875, "std": 87.30249786376953, "min": -199.04171752929688, "p10": -19.45819244384765, "median": 51.35894775390625, "p90": 168.97730102539063, "max": 311.224853515625, "pos_frac": 0.8046875, "sample": [8.618934631347656, 18.291194915771484, -17.854202270507812, 209.0880126953125, 40.7845458984375, 4.31610107421875, 124.22007751464844, -60.588897705078125, 134.84585571289062, 20.33098602294922, 110.95458221435547, 26.272171020507812, -12.254669189453125, 74.27297973632812, 21.570343017578125, -89.20156860351562, -79.55511474609375, 27.17205810546875, 47.84605407714844, 0.6360149383544922, 61.11102294921875, 155.73309326171875, 0.53790283203125, 75.11436462402344, 5.1562957763671875, 18.654205322265625, -122.72537231445312, -4.587005615234375, 34.06678771972656, 132.58355712890625, 24.576263427734375, 73.613525390625, 136.08969116210938, 112.0245361328125, 169.24505615234375, 23.645095825195312, -0.9962921142578125, -4.46502685546875, 143.84674072265625, -2.525604248046875, -126.92515563964844, 76.54960632324219, -6.207386016845703, 123.9134292602539, 161.95904541015625, 150.5688018798828, 54.184814453125, 170.50039672851562, 151.58050537109375, 48.5330810546875, 0.14542293548583984, -122.06684875488281, 178.83538818359375, 168.862548828125, 41.81083679199219, 78.550537109375, 311.224853515625, 106.561767578125, 193.43536376953125, 173.62432861328125, 11.653274536132812, 1.1734199523925781, -8.96026611328125, 160.86026000976562, 38.849578857421875, 121.35738372802734, 162.49313354492188, 141.6418914794922, 136.390869140625, 150.40390014648438, -23.200836181640625, 134.31076049804688, -86.39249420166016, 40.11009216308594, 139.84597778320312, -199.04171752929688, 38.27812194824219, 136.0239715576172, -1.79559326171875, 107.1485595703125, 77.682373046875, 163.97381591796875, 172.49105834960938, 11.62445068359375, 126.99778747558594, 139.39657592773438, -34.838134765625, 222.80755615234375, 2.7513160705566406, 6.6394500732421875, 170.7196044921875, 116.26953125, 149.75238037109375, 40.04023742675781, 121.17489624023438, 142.70962524414062, 73.82858276367188, 162.99456787109375, 189.8907470703125, 37.07940673828125, 33.243133544921875, -3.6705322265625, 137.30245971679688, 139.319091796875, 211.11691284179688, 27.49420166015625, 34.57781982421875, 31.59283447265625, 47.53681945800781, 26.3084716796875, 165.5576171875, -0.4622230529785156, 95.42561340332031, 5.8192138671875, 10.0225830078125, -96.10620880126953, 166.75015258789062, 4.513519287109375, 10.622695922851562, -38.98637390136719, 140.16412353515625, 55.0743408203125, 201.3424072265625, 146.39959716796875, 157.33282470703125, -104.5897216796875, -3.37060546875, 159.5206756591797], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000305.npy"}
{"epoch": 0.6387434554973822, "step": 306, "batch_size": 128, "mean": 59.313499450683594, "std": 84.00444030761719, "min": -160.48471069335938, "p10": -22.151889038085937, "median": 59.903934478759766, "p90": 162.37554473876952, "max": 201.47689819335938, "pos_frac": 0.7265625, "sample": [96.9173583984375, -116.99642944335938, -120.14505767822266, 201.47689819335938, -160.48471069335938, 190.78216552734375, -0.47756385803222656, 67.12236022949219, -11.449310302734375, 149.33477783203125, 6.469017028808594, 28.132095336914062, 154.2970428466797, 107.53929138183594, -5.932914733886719, 9.4425048828125, -39.94427490234375, 137.2079315185547, -4.207359313964844, 24.09417724609375, 42.03302001953125, 60.140357971191406, -41.41949462890625, 55.21527099609375, -153.1734619140625, 179.1973876953125, 181.38580322265625, 139.01266479492188, 131.2685546875, -1.2765693664550781, 59.667510986328125, 111.6702880859375, 33.48353576660156, 142.75619506835938, 162.4630126953125, 137.46438598632812, 184.97125244140625, 139.25820922851562, 124.61550903320312, 121.40518188476562, 8.124435424804688, 125.8441390991211, -8.3302001953125, 8.940994262695312, 109.73654174804688, 108.22103881835938, 60.898956298828125, 107.90457153320312, 56.966400146484375, -85.60931396484375, -11.26214599609375, 162.3380584716797, -16.77782440185547, 195.43536376953125, 66.226318359375, -149.98455810546875, 92.00858306884766, 70.7747802734375, -14.1612548828125, 198.046875, -5.041770935058594, 80.05096435546875, 120.11917114257812, 43.483734130859375, 133.70855712890625, 4.352914810180664, -1.7105560302734375, 3.045116424560547, 122.5814208984375, -22.091583251953125, 135.2624053955078, -128.40097045898438, 122.75100708007812, 56.58441162109375, 191.9891357421875, -1.2910842895507812, -57.6488037109375, 21.422523498535156, 18.467193603515625, 31.45654296875, 10.068893432617188, 60.959136962890625, 137.12997436523438, 132.43353271484375, 159.383056640625, -13.97900390625, -22.2926025390625, 64.86602783203125, 99.41619873046875, 198.79019165039062, 69.5235595703125, 88.65994262695312, 23.43255615234375, -24.5030517578125, 59.604736328125, 133.08612060546875, 126.92489624023438, -10.148834228515625, 141.84478759765625, -18.951244354248047, 156.74862670898438, 100.75057983398438, 150.39747619628906, 157.7750244140625, 87.79450988769531, 21.64226531982422, 5.904792785644531, -14.330612182617188, -4.28753662109375, -4.00994873046875, -8.221302032470703, 4.8170166015625, 154.91188049316406, 103.27549743652344, 16.22833251953125, 187.5814208984375, 15.437414169311523, -3.7112884521484375, -17.77960205078125, 7.785186767578125, 92.60943603515625, 130.10865783691406, 0.6222305297851562, -122.7937240600586, 121.91580200195312, 21.216461181640625, 172.37167358398438, 163.3702392578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000306.npy"}
{"epoch": 0.6408376963350786, "step": 307, "batch_size": 128, "mean": 63.05182647705078, "std": 81.02584075927734, "min": -201.15560913085938, "p10": -18.699066162109375, "median": 65.6617431640625, "p90": 155.3752212524414, "max": 208.465087890625, "pos_frac": 0.796875, "sample": [153.86746215820312, 148.3553009033203, 49.75221252441406, 188.330078125, -91.53765869140625, -6.820930480957031, 106.43065643310547, 118.0496826171875, 122.1759033203125, 148.46966552734375, -173.7705078125, 2.793063163757324, 102.1595458984375, 5.7771453857421875, -12.316253662109375, 130.8188018798828, 148.21664428710938, 131.2374725341797, 10.680362701416016, 20.22125244140625, 61.6318359375, -44.856781005859375, 74.65313720703125, 9.553472518920898, 192.19189453125, -31.22705078125, 69.691650390625, 13.2073974609375, 138.4874725341797, 173.27450561523438, 118.50244140625, 24.81402587890625, 32.049652099609375, 108.20745086669922, 127.99604797363281, -11.562576293945312, 168.4696044921875, 70.99600219726562, 20.323272705078125, -8.396720886230469, 24.663238525390625, 114.72119903564453, 132.6122283935547, -13.00311279296875, 152.5435791015625, 107.03726196289062, 152.25352478027344, -10.4622802734375, 2.237457275390625, 26.695770263671875, 95.94911193847656, 0.12457275390625, 151.2173309326172, 8.290714263916016, 95.34274291992188, 141.24928283691406, 124.17742919921875, 158.89332580566406, 18.526344299316406, 3.4930038452148438, 17.1202392578125, 84.90496063232422, -71.59722900390625, 119.38568115234375, 0.5468177795410156, -3.6071014404296875, 141.01206970214844, 41.411277770996094, -7.9765167236328125, 5.200733184814453, -5.228410720825195, 147.85215759277344, -53.4134521484375, -52.70196533203125, 29.48297119140625, -18.2423095703125, 34.91912841796875, 143.77590942382812, 100.747314453125, 28.73895263671875, 153.3136444091797, -19.76483154296875, 25.847118377685547, 169.15774536132812, 175.20230102539062, 152.90554809570312, -53.323177337646484, 136.75088500976562, 5.403415679931641, 105.26663208007812, 18.788063049316406, -42.18162536621094, 101.56559753417969, 181.7066650390625, 111.26045227050781, 8.97149658203125, 149.04066467285156, -153.37841796875, -11.209701538085938, 139.33489990234375, 3.218109130859375, 189.68411254882812, 32.502716064453125, 177.742431640625, -201.15560913085938, 2.651275634765625, 3.119964599609375, 76.15513610839844, 29.58544921875, 99.67564392089844, -42.9329833984375, 27.2823486328125, 208.465087890625, 135.28067016601562, 21.966217041015625, 113.15875244140625, -0.76385498046875, -7.019775390625, 72.69773864746094, 146.82235717773438, 107.93426513671875, 5.223060607910156, 130.6072540283203, 152.08262634277344, 162.703857421875, 3.2696571350097656, 159.22061157226562, 127.00868225097656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000307.npy"}
{"epoch": 0.6429319371727749, "step": 308, "batch_size": 128, "mean": 42.07502746582031, "std": 83.6346206665039, "min": -184.77886962890625, "p10": -47.28709564208984, "median": 18.49178695678711, "p90": 157.5385986328125, "max": 218.68551635742188, "pos_frac": 0.71875, "sample": [160.2179718017578, 37.188819885253906, -46.6964111328125, 163.05978393554688, 74.91390991210938, -120.64067077636719, -49.068817138671875, 55.157562255859375, 8.3468017578125, 76.36418151855469, 3.63385009765625, 173.03463745117188, 0.42400360107421875, -79.01373291015625, -4.410064697265625, 3.56768798828125, 127.30043029785156, -10.29278564453125, 133.32611083984375, 100.19091796875, 140.91693115234375, 35.000213623046875, 0.693023681640625, -97.72571563720703, 14.770263671875, 126.48487854003906, -167.00558471679688, -30.34368133544922, 12.635208129882812, 157.62225341796875, 129.47584533691406, 26.78790283203125, 85.73333740234375, 116.28855895996094, -17.834075927734375, -6.47698974609375, 142.23294067382812, 8.346450805664062, 114.53199768066406, 12.855712890625, -4.1315765380859375, 39.94379425048828, 197.52435302734375, 1.17462158203125, -38.43309020996094, -1.2183380126953125, 0.762786865234375, 149.839599609375, -4.792823791503906, 131.60691833496094, 4.99078369140625, 12.93115234375, 3.5634307861328125, 127.31468963623047, -10.211952209472656, 30.874237060546875, 160.4775390625, 29.881027221679688, -184.77886962890625, 158.1927490234375, 101.15098571777344, 11.967498779296875, -8.78125, -19.818634033203125, 218.68551635742188, 128.52882385253906, -2.6912174224853516, -137.23440551757812, 11.51904296875, -3.2712249755859375, 113.53764343261719, -95.28974914550781, 158.75372314453125, 82.47811889648438, 9.4837646484375, 25.037109375, 176.74002075195312, 201.7755126953125, 157.50274658203125, -1.6871910095214844, 106.13360595703125, -8.5296630859375, 2.0783729553222656, 1.612579345703125, 74.42514038085938, -137.2344970703125, 28.178794860839844, 11.20245361328125, -60.213165283203125, 111.56076049804688, 21.138092041015625, 15.845481872558594, 32.87103271484375, 10.850204467773438, 131.42718505859375, 131.378173828125, 169.46957397460938, 143.78915405273438, 34.73857116699219, 140.3023681640625, 27.502960205078125, 77.868408203125, 192.27072143554688, -48.66535949707031, 34.94574737548828, 11.2506103515625, 14.25628662109375, -42.759098052978516, 123.92805480957031, 132.54800415039062, -147.1806640625, 13.080718994140625, -26.5601806640625, 2.4113845825195312, -3.3439407348632812, 109.971923828125, -36.20892333984375, 9.982086181640625, -5.2186279296875, 8.159942626953125, -43.588104248046875, 106.82501220703125, 70.14212036132812, 142.03677368164062, -105.686767578125, 82.185302734375, 95.61785888671875, 83.31517028808594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000308.npy"}
{"epoch": 0.6450261780104712, "step": 309, "batch_size": 128, "mean": 62.947906494140625, "std": 90.88772583007812, "min": -139.50643920898438, "p10": -31.441101074218743, "median": 46.242942810058594, "p90": 175.37186279296876, "max": 317.0699157714844, "pos_frac": 0.7421875, "sample": [64.92207336425781, 176.03396606445312, 147.98529052734375, 131.92221069335938, -2.351654052734375, -29.891204833984375, 2.492767333984375, -107.422607421875, 145.99325561523438, 46.941558837890625, 175.08810424804688, 12.843025207519531, -13.034698486328125, 6.71875, -98.19656372070312, 132.237548828125, 11.330322265625, 25.747947692871094, -8.472457885742188, 44.510650634765625, -11.08819580078125, 127.91459655761719, -4.381866455078125, -131.6688690185547, 10.07855224609375, 101.06048583984375, 1.107269287109375, 317.0699157714844, 137.5166778564453, 33.917633056640625, -7.3441925048828125, 184.8809814453125, 132.84202575683594, 3.2684326171875, -3.706024169921875, -12.791397094726562, 118.44410705566406, -85.39675903320312, 214.77340698242188, -87.45785522460938, 205.84988403320312, -86.03955078125, -4.541286468505859, 74.41220092773438, 139.01663208007812, -91.4347915649414, 2.721954345703125, -35.057525634765625, -21.621673583984375, 9.70068359375, 143.72265625, -1.421285629272461, 13.81390380859375, -97.70916748046875, 163.37527465820312, 133.34335327148438, 146.90823364257812, 16.47564697265625, 124.5054931640625, 16.66339111328125, 173.92425537109375, 9.9041748046875, 43.40350341796875, 248.177001953125, -14.343376159667969, 186.1097412109375, 11.088333129882812, 138.23631286621094, 24.92455291748047, 48.455413818359375, 45.54432678222656, 34.81256103515625, -3.1655654907226562, 191.6812744140625, 2.3910675048828125, 165.4776611328125, 158.20553588867188, 117.89842224121094, 13.042251586914062, 176.73480224609375, 139.17660522460938, 143.98626708984375, 7.484230041503906, 36.931297302246094, 135.2998046875, 157.74151611328125, -5.153617858886719, 99.5821304321289, 42.51190185546875, -19.882095336914062, 150.237548828125, -139.50643920898438, 66.98512268066406, -18.35045051574707, 177.67782592773438, 102.75900268554688, 20.085830688476562, 146.43052673339844, 36.5330810546875, 10.147529602050781, -125.33688354492188, 67.39549255371094, 147.8883056640625, 169.57904052734375, 154.93063354492188, -4.356109619140625, 202.81765747070312, 34.26812744140625, 64.78271484375, -129.1688690185547, 167.0478515625, 58.84332275390625, 57.19671630859375, 161.04794311523438, 89.47996520996094, 144.74270629882812, 55.91851806640625, 179.901611328125, -15.07122802734375, 158.59877014160156, 189.005126953125, 72.31066131591797, 123.78819274902344, -83.526123046875, 129.48309326171875, 41.28546142578125, -1.7303085327148438, 125.90878295898438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000309.npy"}
{"epoch": 0.6471204188481675, "step": 310, "batch_size": 128, "mean": 52.012542724609375, "std": 80.10633850097656, "min": -141.07293701171875, "p10": -35.724937438964844, "median": 25.177703857421875, "p90": 156.66666259765623, "max": 260.15606689453125, "pos_frac": 0.7265625, "sample": [260.15606689453125, 18.720306396484375, -2.2352218627929688, 201.49481201171875, 41.516021728515625, 37.406005859375, 25.77264404296875, 2.102022171020508, 137.25167846679688, 89.64767456054688, 123.61559295654297, 188.039794921875, 112.34556579589844, 136.4332275390625, 24.582763671875, 36.872589111328125, 6.2838134765625, -4.37083625793457, 42.269287109375, -64.92189025878906, -0.141326904296875, 14.77154541015625, 19.569473266601562, 59.48895263671875, 81.2854232788086, -102.2659683227539, 145.1759490966797, -40.6439208984375, 133.79299926757812, 1.0572509765625, -2.445596694946289, -5.111980438232422, -18.3677978515625, 152.91201782226562, 3.657806396484375, -4.6089324951171875, 66.84451293945312, 175.86557006835938, -53.8394775390625, -114.02349853515625, 127.02767944335938, -2.8498077392578125, 19.665252685546875, 19.08148193359375, 116.03080749511719, 35.306182861328125, 92.77890014648438, 47.06645965576172, 23.418212890625, -1.835479736328125, -9.8714599609375, 30.2197265625, 172.48953247070312, 129.27691650390625, 112.06488037109375, 151.95074462890625, 191.58782958984375, 18.3115234375, -3.7028274536132812, -141.07293701171875, 14.198022842407227, 172.70651245117188, 69.15064239501953, 22.7301025390625, -3.76837158203125, -119.28559112548828, 8.131790161132812, 19.36566162109375, 124.6143798828125, 3.9320716857910156, -19.113807678222656, 80.82461547851562, 24.37872314453125, 29.367340087890625, 38.133209228515625, 141.67318725585938, 142.18756103515625, 179.12005615234375, 56.64007568359375, -10.758255004882812, -21.98204803466797, 115.94526672363281, 122.17135620117188, -0.7802886962890625, -44.81951904296875, 138.85110473632812, 158.99951171875, 5.0818939208984375, 126.20519256591797, -6.69647216796875, 20.33734893798828, -60.5120849609375, 142.43463134765625, -98.2803955078125, 96.47502136230469, 3.0739517211914062, 10.8291015625, 148.75794982910156, 89.59738159179688, 15.858200073242188, 54.89086151123047, 166.08984375, 2.147796630859375, -29.02558135986328, 149.96954345703125, -11.5128173828125, -5.4022216796875, -34.7095947265625, -113.12109375, 115.45864868164062, -38.09407043457031, 162.947998046875, -8.583206176757812, 137.97848510742188, 9.0592041015625, -47.23748779296875, 18.4412841796875, 1.9173698425292969, 19.132423400878906, 115.63499450683594, 45.016082763671875, 6.378765106201172, 146.82144165039062, 164.00765991210938, 213.16058349609375, 131.74310302734375, 155.6668701171875, 140.1551055908203], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000310.npy"}
{"epoch": 0.6492146596858639, "step": 311, "batch_size": 128, "mean": 54.99326705932617, "std": 93.62425994873047, "min": -208.114990234375, "p10": -52.09420700073237, "median": 37.07876968383789, "p90": 165.30486755371092, "max": 223.2672119140625, "pos_frac": 0.7421875, "sample": [7.625083923339844, -1.6859283447265625, 135.7754364013672, 122.57462310791016, 159.72393798828125, 22.201921463012695, 30.903152465820312, -3.8086395263671875, 156.4530029296875, 128.539794921875, 150.5276641845703, 10.274417877197266, -9.580955505371094, -26.672103881835938, 37.01103210449219, -3.439697265625, 160.014404296875, 50.11176300048828, -101.59744262695312, 46.262603759765625, 190.508056640625, 149.92025756835938, 1.9762725830078125, 134.17088317871094, 0.12752532958984375, -19.460113525390625, 171.0960693359375, 126.29568481445312, 5.996097564697266, 60.328521728515625, -122.30048370361328, -208.114990234375, -204.01895141601562, 148.23960876464844, 32.582977294921875, 170.53143310546875, -139.9708251953125, -102.974365234375, 146.00375366210938, 53.81243896484375, -87.0311508178711, 129.24163818359375, 38.778717041015625, 6.892333984375, -1.7937984466552734, 54.58856201171875, -14.966102600097656, 133.8849639892578, -115.23161315917969, 33.14629364013672, 130.36288452148438, 18.613525390625, 2.25750732421875, 26.976699829101562, 135.8274383544922, -14.465011596679688, 110.19800567626953, -18.01580810546875, -22.91973876953125, 174.01885986328125, 170.545166015625, 148.28582763671875, 157.28353881835938, 134.7490997314453, 137.87518310546875, -108.48695373535156, 33.641937255859375, 15.29876708984375, 17.99200439453125, 34.697425842285156, -120.15959167480469, -13.375171661376953, 187.33673095703125, 101.46255493164062, 22.579986572265625, 36.364158630371094, 173.889404296875, 135.64047241210938, 8.553688049316406, 126.32711791992188, 223.2672119140625, 3.546875, -113.36007690429688, 37.146507263183594, -0.1074371337890625, -32.72857666015625, 177.01168823242188, 185.57611083984375, 151.55752563476562, 23.764739990234375, 129.9131622314453, 111.58782958984375, 119.11021423339844, 42.06103515625, 188.36572265625, -141.1708526611328, 183.1253662109375, -5.584686279296875, -140.7500457763672, 129.8806610107422, 7.140918731689453, 163.06491088867188, -22.7335205078125, 103.983154296875, 8.586456298828125, -5.161529541015625, 160.80470275878906, 157.94552612304688, 9.087244033813477, 43.30865478515625, 128.33084106445312, 117.84941101074219, 124.153076171875, 36.122589111328125, -37.12123107910156, 5.990043640136719, 136.40074157714844, 153.2538299560547, 151.7867889404297, 2.7049942016601562, 1.1091384887695312, -4.155731201171875, 5.695320129394531, 207.9158935546875, 118.98100280761719, 140.21060180664062, 82.565185546875, -17.723098754882812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000311.npy"}
{"epoch": 0.6513089005235602, "step": 312, "batch_size": 128, "mean": 55.28929901123047, "std": 87.85151672363281, "min": -183.7906494140625, "p10": -47.74467773437499, "median": 39.171539306640625, "p90": 160.28793029785155, "max": 246.48077392578125, "pos_frac": 0.7421875, "sample": [78.6019287109375, -3.6827239990234375, -19.287567138671875, 147.7088623046875, 151.5374298095703, -183.7906494140625, 150.66592407226562, 131.25558471679688, -39.575172424316406, -23.892425537109375, -27.118011474609375, 11.800392150878906, 95.13302612304688, 124.02911376953125, 102.57589721679688, 173.16366577148438, 126.63444519042969, 142.3636016845703, -20.230438232421875, -135.80670166015625, 5.81610107421875, 133.23582458496094, 147.64462280273438, -0.284912109375, 8.302635192871094, 101.06941986083984, 58.609375, 211.12960815429688, -43.6812744140625, -13.706615447998047, 168.5216064453125, -9.331687927246094, 1.0882568359375, 170.9498291015625, 90.95391845703125, 32.653350830078125, 92.12428283691406, 1.3317070007324219, 106.66035461425781, 122.78842163085938, 182.76971435546875, 165.25498962402344, 83.23368072509766, 146.53277587890625, 126.78089904785156, 123.73627471923828, 62.655792236328125, 129.58932495117188, 139.23828125, 27.031402587890625, 148.67416381835938, 149.9453125, 18.639633178710938, 2.7769088745117188, -57.2259521484375, 11.719207763671875, 35.218231201171875, 36.510528564453125, 0.8475799560546875, 246.48077392578125, 10.171363830566406, -34.28399658203125, 167.71188354492188, -123.39961242675781, 48.881805419921875, -36.04693603515625, 179.88565063476562, 144.70172119140625, -9.146240234375, 135.292724609375, 26.762542724609375, 136.26571655273438, 25.18252944946289, -5.459693908691406, 117.30120849609375, 86.86041259765625, 1.4990615844726562, 159.26113891601562, 1.5719680786132812, 124.29193115234375, 157.64341735839844, 162.68377685546875, -83.06771850585938, 134.45948791503906, -4.102182388305664, 0.0, 61.71771240234375, -91.7806396484375, -138.12542724609375, -101.92610168457031, 18.57958984375, 82.44381713867188, 3.8137245178222656, 38.42286682128906, 70.78231811523438, 0.7903366088867188, -141.8327178955078, 26.644657135009766, 131.681396484375, 3.243988037109375, 19.17047119140625, 13.049156188964844, 142.75827026367188, 123.25279235839844, -21.5147705078125, 123.7325439453125, 144.47137451171875, -3.08148193359375, 187.20391845703125, 39.76800537109375, 148.60556030273438, 17.934326171875, 38.5750732421875, -5.539571762084961, -83.25006103515625, 102.30473327636719, 84.765869140625, 4.154449462890625, -73.56486511230469, 187.22503662109375, -125.51040649414062, 166.83245849609375, -32.14251708984375, 144.2035369873047, 126.4787368774414, 24.261474609375, 5.979455947875977, -60.84156799316406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000312.npy"}
{"epoch": 0.6534031413612565, "step": 313, "batch_size": 128, "mean": 50.28001403808594, "std": 92.01277160644531, "min": -165.66046142578125, "p10": -70.99247131347654, "median": 32.932708740234375, "p90": 169.5498519897461, "max": 244.7464599609375, "pos_frac": 0.71875, "sample": [176.824951171875, 157.372314453125, -28.571044921875, -27.7877197265625, -67.1436767578125, -1.2945022583007812, 67.28097534179688, 146.12213134765625, 174.109619140625, -13.82232666015625, -11.025192260742188, 146.41845703125, -86.90478515625, 155.37623596191406, 110.35694122314453, 169.3921661376953, 155.0009765625, 42.894195556640625, -14.447052001953125, -28.80279541015625, 0.6438446044921875, 14.33978271484375, 3.8630638122558594, 153.3218536376953, 131.351806640625, -8.692596435546875, -4.60870361328125, 25.234649658203125, 122.66781616210938, 194.66836547851562, 182.22900390625, 220.593994140625, 45.388763427734375, 79.9193115234375, -46.026092529296875, 49.77960205078125, 113.02839660644531, 91.87966918945312, 191.33578491210938, 72.79204559326172, -79.97299194335938, 197.24554443359375, 159.0034637451172, 19.94049072265625, -115.77095031738281, 25.9666748046875, 16.06805419921875, -61.6767578125, 131.71493530273438, 166.27362060546875, 27.032180786132812, 94.79883575439453, -134.17575073242188, 20.511001586914062, 73.49923706054688, 0.9003486633300781, 169.91778564453125, 54.63639831542969, 14.909095764160156, 2.3060150146484375, 134.289794921875, 5.65606689453125, 136.99581909179688, 17.19855499267578, 45.27552032470703, 44.948394775390625, -84.06826782226562, 125.4046630859375, -121.98323059082031, -104.5333251953125, 176.5810089111328, -19.487152099609375, 23.62694549560547, 168.0159912109375, 17.3011474609375, -4.88031005859375, 22.629379272460938, 149.67385864257812, -119.92453002929688, 115.816650390625, 12.812362670898438, -117.33615112304688, 113.35842895507812, -47.87031555175781, 38.01416015625, 140.42080688476562, 98.60955810546875, -9.67523193359375, 4.021171569824219, 6.176506042480469, 32.213134765625, 244.7464599609375, 123.85169982910156, 54.18572998046875, 142.65081787109375, 168.0288848876953, 26.48040008544922, 179.3330078125, -43.71356201171875, -165.4739990234375, 62.35247802734375, 46.10735321044922, 7.020116806030273, 185.16949462890625, 45.07362365722656, -54.288787841796875, -0.8859424591064453, -2.424041748046875, 159.26556396484375, 135.50460815429688, 28.319915771484375, 94.85276794433594, 25.172576904296875, 4.692230224609375, -2.983287811279297, -11.615058898925781, 132.51315307617188, 33.65228271484375, -116.51177978515625, -144.54957580566406, -0.365814208984375, -165.66046142578125, 123.77723693847656, 104.8289566040039, 17.28369140625, 161.9985809326172, 26.024383544921875, 173.9595184326172], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000313.npy"}
{"epoch": 0.6554973821989529, "step": 314, "batch_size": 128, "mean": 54.02167510986328, "std": 90.63722229003906, "min": -186.70407104492188, "p10": -57.03547058105468, "median": 40.501556396484375, "p90": 160.74913024902344, "max": 223.10287475585938, "pos_frac": 0.703125, "sample": [30.899337768554688, 116.12258911132812, 129.1216583251953, 47.7774658203125, -15.249130249023438, 22.322052001953125, -28.193222045898438, 123.104248046875, -99.89859771728516, -144.65472412109375, -39.04457092285156, 82.50877380371094, 159.8702392578125, 170.50880432128906, 173.2181396484375, -69.11886596679688, -62.60400390625, 196.53372192382812, 144.02572631835938, 163.58944702148438, 125.70367431640625, -10.861663818359375, 52.23725891113281, 140.94288635253906, 7.513427734375, 11.531669616699219, 13.590141296386719, 181.2000732421875, 8.564910888671875, 185.85586547851562, 4.704418182373047, -20.568313598632812, 14.446746826171875, -75.1448974609375, 149.84417724609375, 57.080810546875, 18.05303955078125, 158.48931884765625, 41.77496337890625, 156.08984375, -16.08343505859375, 129.50193786621094, -87.27114868164062, 124.67601013183594, 5.6252899169921875, 83.55125427246094, 197.6224365234375, 91.05426025390625, 9.660003662109375, -4.2054443359375, 149.49014282226562, -0.29973602294921875, 17.99481201171875, 87.39971923828125, -0.6313972473144531, 114.0595703125, 39.2281494140625, 21.67138671875, -22.899185180664062, -130.9112548828125, 160.46920776367188, 157.93809509277344, 133.97494506835938, 0.0, -5.34197998046875, -71.90237426757812, -175.6427001953125, 136.57565307617188, 22.352203369140625, -20.874176025390625, -35.1544189453125, 20.756927490234375, 117.07958984375, 161.40228271484375, 172.54966735839844, -54.648956298828125, -19.38275909423828, 173.44677734375, -31.373291015625, -186.70407104492188, 145.17617797851562, 43.007293701171875, 146.82550048828125, 95.69583129882812, 0.0, 154.77609252929688, -3.6673583984375, 100.003173828125, 145.786865234375, 18.7459716796875, 20.884056091308594, 138.84814453125, 145.38075256347656, 35.106109619140625, 114.81271362304688, 158.02130126953125, 138.0849609375, 20.045989990234375, 47.0546875, 6.275634765625, 8.72390365600586, -112.82244873046875, -3.48077392578125, -32.42578125, 109.60113525390625, 2.730987548828125, 130.70376586914062, 131.9066925048828, 45.69622802734375, 121.91352844238281, 179.78671264648438, 39.15040588378906, -48.125946044921875, -7.3291168212890625, 3.055084228515625, 223.10287475585938, 151.71607971191406, -18.98029327392578, 22.972450256347656, 115.02281188964844, 142.39093017578125, 126.04788208007812, -90.309326171875, 100.59774780273438, -178.41018676757812, 180.2271728515625, 116.55422973632812, -2.7478981018066406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000314.npy"}
{"epoch": 0.6575916230366492, "step": 315, "batch_size": 128, "mean": 60.21345520019531, "std": 87.28994750976562, "min": -168.6333465576172, "p10": -26.486984252929684, "median": 30.027299880981445, "p90": 173.28522644042968, "max": 285.9542236328125, "pos_frac": 0.765625, "sample": [-0.248016357421875, 60.57524108886719, 2.8436222076416016, 4.6373748779296875, 12.469589233398438, -29.588134765625, -19.9423828125, 15.144912719726562, 177.77984619140625, -12.330108642578125, 32.01606750488281, -13.97113037109375, 135.91998291015625, 22.914093017578125, 134.31497192382812, -35.57423400878906, 107.18525695800781, 170.83013916015625, 52.483489990234375, -25.280364990234375, -46.551025390625, -17.621421813964844, 3.05078125, -2.8560256958007812, 189.22555541992188, 12.7711181640625, 129.12005615234375, 183.07176208496094, 210.0716552734375, 0.9834480285644531, 28.038532257080078, 43.1702880859375, 163.40989685058594, 168.956298828125, 33.83123779296875, 188.38848876953125, 166.034423828125, 157.5148162841797, 114.65501403808594, -3.266632080078125, -0.27178955078125, -37.36737060546875, 22.7091064453125, 170.38998413085938, -29.30242919921875, 10.656463623046875, 8.17291259765625, 6.321533203125, 13.2103271484375, 6.6563720703125, 156.02325439453125, -9.99142074584961, 159.18096923828125, 173.260498046875, 154.91604614257812, 155.6026153564453, 18.885223388671875, -8.517784118652344, 38.78973388671875, 65.24015808105469, 15.097549438476562, 10.525543212890625, -8.641101837158203, 76.455078125, 156.0635986328125, 156.5455780029297, 102.54486083984375, 132.89895629882812, 16.985107421875, 285.9542236328125, -111.53573608398438, 34.03947448730469, 141.67745971679688, 25.187042236328125, 173.34292602539062, -42.241302490234375, -166.0743865966797, 185.88690185546875, 3.803659439086914, 177.67828369140625, 159.32220458984375, 52.092620849609375, -0.8737030029296875, 150.32708740234375, 202.9932861328125, 15.005538940429688, 39.99981689453125, 121.931640625, 22.43695068359375, 188.013671875, 6.581024169921875, 52.326202392578125, -157.3695068359375, -46.48699951171875, 162.6497802734375, 180.83740234375, 64.23165893554688, 42.51808547973633, -168.6333465576172, 162.91714477539062, 27.977867126464844, -10.4415283203125, -3.1541748046875, 111.61602020263672, 10.0289306640625, -117.94137573242188, -52.359222412109375, 80.8349609375, 14.791177749633789, 19.08453369140625, 166.0836181640625, 14.852813720703125, 146.00025939941406, 220.6592254638672, 17.82763671875, 116.03720092773438, 168.3636474609375, 3.9118576049804688, -3.54852294921875, 14.863800048828125, 81.8980484008789, -2.7300643920898438, 146.06173706054688, 51.18110656738281, 145.99807739257812, 99.1009521484375, 9.379470825195312, 25.186771392822266], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000315.npy"}
{"epoch": 0.6596858638743456, "step": 316, "batch_size": 128, "mean": 62.193031311035156, "std": 90.71003723144531, "min": -155.60675048828125, "p10": -27.75478363037109, "median": 41.117191314697266, "p90": 177.87828369140624, "max": 306.5335693359375, "pos_frac": 0.7265625, "sample": [121.5845947265625, 105.02960205078125, 87.0931396484375, 38.247528076171875, -14.420654296875, -25.126235961914062, -8.09820556640625, 181.29824829101562, 77.63209533691406, 183.11944580078125, 25.361236572265625, -39.56585693359375, 155.8292694091797, 34.744659423828125, -98.99617004394531, 33.581214904785156, 134.49868774414062, -72.70362854003906, 122.53463745117188, 9.022430419921875, 169.92019653320312, 167.74887084960938, -25.564483642578125, 134.01365661621094, 144.34744262695312, 269.40826416015625, 39.17767333984375, 142.93731689453125, 49.013275146484375, 198.27906799316406, -9.79010009765625, 152.22471618652344, -24.0706787109375, 141.78334045410156, 179.63247680664062, 17.72967529296875, 118.02063751220703, 162.198974609375, 172.4935760498047, 23.226898193359375, -9.0404052734375, 102.86880493164062, 4.70953369140625, 67.62371826171875, 8.435745239257812, 68.16216278076172, -10.33010482788086, 167.93023681640625, 16.2059326171875, -0.75823974609375, 98.53225708007812, 4.41143798828125, -155.60675048828125, -104.55731201171875, -29.324111938476562, 197.358642578125, 123.93173217773438, 111.56910705566406, -3.2381591796875, 0.3840484619140625, -10.778266906738281, 105.53814697265625, 186.46719360351562, 152.09185791015625, 23.226837158203125, 25.67340087890625, -25.90869140625, 306.5335693359375, 40.93494415283203, 138.1783447265625, 5.0679931640625, -2.9237518310546875, -34.831146240234375, -69.05657958984375, 4.633270263671875, 161.43536376953125, 115.849853515625, 166.66854858398438, 4.722221374511719, 150.28170776367188, 137.15451049804688, 141.29051208496094, 177.2384033203125, -15.9317626953125, -5.057126998901367, 68.79595947265625, 30.789215087890625, 199.04608154296875, -93.9847640991211, 143.62371826171875, -122.61166381835938, 59.4190673828125, 0.0, -125.28765869140625, 18.173675537109375, 183.344482421875, 179.371337890625, 15.391593933105469, -1.071563720703125, 191.19256591796875, 113.21426391601562, 23.235336303710938, -139.68434143066406, 13.048904418945312, -17.36175537109375, 92.46554565429688, 162.23788452148438, -0.43230438232421875, 110.22858428955078, 68.42051696777344, 192.3424072265625, 14.27996826171875, 168.3299560546875, -27.08221435546875, 22.70599365234375, 70.502197265625, -13.246997833251953, -141.52613830566406, 41.2994384765625, 3.8828582763671875, 64.29318237304688, 131.83421325683594, 126.75535583496094, 39.037628173828125, 134.76136779785156, 12.722946166992188, 147.85601806640625, -10.7696533203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000316.npy"}
{"epoch": 0.6617801047120419, "step": 317, "batch_size": 128, "mean": 57.40822219848633, "std": 84.23892974853516, "min": -161.64439392089844, "p10": -27.334239196777343, "median": 36.959367752075195, "p90": 166.85116271972655, "max": 255.77008056640625, "pos_frac": 0.7265625, "sample": [144.04066467285156, -137.62176513671875, 0.0, 8.433815002441406, 205.4666748046875, -28.77123260498047, 172.80686950683594, 109.78970336914062, 11.077239990234375, 173.45794677734375, 17.678680419921875, 163.6995391845703, 81.62113952636719, -13.658203125, 168.25473022460938, -139.15304565429688, -30.538619995117188, 153.91287231445312, 93.06723022460938, 0.43491363525390625, 48.30902099609375, 13.748138427734375, 157.49160766601562, 125.60122680664062, -1.2540969848632812, 148.8924102783203, 9.751182556152344, 1.1180267333984375, -8.57354736328125, 33.57452392578125, 0.4629974365234375, -51.301116943359375, 36.201904296875, -7.1385040283203125, -27.899154663085938, 114.38923645019531, 130.46163940429688, 169.95677185058594, 152.1019287109375, -27.092132568359375, -22.919219970703125, 32.50743103027344, 175.3031005859375, -155.49468994140625, -141.47482299804688, 161.04522705078125, 181.63299560546875, 166.15255737304688, 9.033241271972656, 0.0, 94.05021667480469, 114.71588134765625, 35.002655029296875, 198.77975463867188, -63.949188232421875, -10.300140380859375, 35.498191833496094, 31.885848999023438, 5.1809539794921875, -3.4151611328125, 16.55560302734375, 166.35986328125, 161.1707305908203, 121.5447998046875, 0.0, 14.714508056640625, 18.728515625, -1.2530670166015625, 37.71683120727539, 97.60797119140625, 167.99752807617188, -6.4670562744140625, -15.150054931640625, -161.64439392089844, 178.30026245117188, 71.62181091308594, 166.083984375, -89.61981201171875, 54.64082336425781, -11.93438720703125, 99.41415405273438, 134.31689453125, 124.08872985839844, -12.934318542480469, 40.31646728515625, 111.41973876953125, 49.19264221191406, -28.49688720703125, 122.21812438964844, 160.25808715820312, -3.3486251831054688, 2.31097412109375, 174.0182647705078, 12.210723876953125, 5.1015167236328125, -2.0435237884521484, -2.1694793701171875, 142.22103881835938, 135.60272216796875, 255.77008056640625, 1.1898574829101562, 126.63917541503906, 95.98695373535156, 3.0633544921875, -36.28880310058594, 44.88037109375, 24.810821533203125, 39.103729248046875, 0.26964569091796875, 95.37469482421875, 101.53704071044922, 5.423826217651367, 147.37002563476562, 6.739837646484375, 107.14117431640625, 131.92300415039062, 129.60433959960938, 43.302452087402344, 175.15025329589844, 6.775379180908203, -7.292963027954102, 81.37301635742188, -7.094573974609375, 162.37538146972656, 70.57066345214844, 134.78396606445312, -3.4171142578125, 144.47930908203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000317.npy"}
{"epoch": 0.6638743455497382, "step": 318, "batch_size": 128, "mean": 59.482791900634766, "std": 82.80902862548828, "min": -172.981689453125, "p10": -30.05619888305664, "median": 61.156494140625, "p90": 160.47382507324218, "max": 203.01611328125, "pos_frac": 0.734375, "sample": [124.77671813964844, 24.052230834960938, 150.05191040039062, 144.28158569335938, 14.080230712890625, -10.58447265625, 32.85272216796875, 75.97879028320312, 159.53970336914062, 45.61802673339844, 11.32489013671875, -15.05352783203125, 158.3992919921875, 69.42364501953125, 184.009521484375, 0.074859619140625, 143.02923583984375, -18.868499755859375, 129.68536376953125, 102.85995483398438, -22.0838623046875, -19.842926025390625, 143.127197265625, 162.6534423828125, -19.23065185546875, 40.9970703125, 146.40985107421875, 152.1653289794922, -125.14802551269531, 33.17724609375, 0.369384765625, -30.208763122558594, 94.15020751953125, 84.06607818603516, 70.87617492675781, -29.990814208984375, 9.235228538513184, 71.03826904296875, 155.53125, 186.77841186523438, 102.3560791015625, 127.21060180664062, 116.35360717773438, 63.314697265625, 82.04629516601562, 8.206756591796875, -149.93963623046875, 126.5020980834961, -2.4708251953125, 94.70819091796875, -166.6319580078125, 13.630962371826172, 106.1784896850586, 174.76437377929688, 132.12872314453125, 47.46476745605469, -19.1143798828125, 130.90602111816406, 153.5440673828125, -19.23492431640625, 106.75051879882812, 128.778564453125, 184.00189208984375, 42.912139892578125, -100.90866088867188, 40.54901123046875, -46.86480712890625, 3.143890380859375, 169.352783203125, -3.1605758666992188, 142.22369384765625, 27.793853759765625, 5.86236572265625, 102.73889923095703, 130.9793701171875, 150.98117065429688, 89.48844909667969, 139.51953125, 85.48837280273438, -8.827239990234375, 55.25556182861328, 99.1424560546875, 31.10662841796875, 147.85708618164062, 105.91801452636719, -40.95037841796875, 83.3175048828125, 181.19699096679688, -43.271080017089844, -9.026718139648438, -9.199951171875, 53.7938232421875, 26.36468505859375, 134.7703857421875, 60.82080078125, 50.856842041015625, 115.44656372070312, -77.42863464355469, 133.14559936523438, -7.621917724609375, 119.54718780517578, 203.01611328125, 179.24261474609375, 36.556427001953125, -42.78575134277344, 144.62615966796875, 75.955078125, 9.000106811523438, -27.852828979492188, -31.244781494140625, 167.26776123046875, 36.507568359375, 61.4921875, 163.15249633789062, 2.117431640625, 163.17050170898438, 9.557743072509766, 154.15753173828125, -27.08306884765625, 128.69424438476562, -151.167236328125, -6.18060302734375, -2.3068466186523438, 32.06394958496094, 165.5113525390625, -10.060317993164062, -172.981689453125, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000318.npy"}
{"epoch": 0.6659685863874345, "step": 319, "batch_size": 128, "mean": 63.1138916015625, "std": 74.5772476196289, "min": -168.45619201660156, "p10": -15.446340942382811, "median": 46.77491760253906, "p90": 157.51759338378906, "max": 215.03753662109375, "pos_frac": 0.8203125, "sample": [119.90499114990234, 3.227140426635742, 63.45428466796875, 47.893707275390625, 30.870311737060547, 215.03753662109375, 140.54403686523438, -35.345703125, 115.64369201660156, 24.083160400390625, -10.244796752929688, 13.802034378051758, 119.0451431274414, 176.58270263671875, 85.59100341796875, 174.013427734375, 8.55023193359375, 0.0, 57.155975341796875, 77.55770874023438, 11.779056549072266, 10.510627746582031, 105.3616943359375, 12.651229858398438, 3.00994873046875, 141.25411987304688, 9.45452880859375, 83.75698852539062, -74.57452392578125, 118.96147155761719, -5.9923095703125, 10.01873779296875, 98.89361572265625, 5.351345062255859, 118.71065521240234, 67.18255615234375, 7.3250579833984375, 123.06439208984375, 152.4417724609375, 31.281265258789062, 4.200430870056152, 150.5875244140625, 44.9686279296875, -119.67605590820312, 2.5015792846679688, 125.98239135742188, 173.809814453125, 9.490966796875, -51.434600830078125, 155.1298828125, 110.42694091796875, 156.619140625, 11.111663818359375, 30.499908447265625, 158.61956787109375, 24.5233154296875, 29.70654296875, -64.2437973022461, 76.9891586303711, 146.23939514160156, -168.45619201660156, 84.2962875366211, 119.7424545288086, -88.01211547851562, 87.14909362792969, -23.461471557617188, 32.954803466796875, 24.102895736694336, 77.0140380859375, 138.13186645507812, 2.1713333129882812, 121.7263412475586, 121.06781005859375, 156.08984375, -4.507904052734375, -14.871856689453125, 127.719970703125, 126.81709289550781, 157.04531860351562, -25.452072143554688, 16.40343475341797, 160.33428955078125, 140.32662963867188, 31.649139404296875, 37.54437255859375, -1.250152587890625, 125.3812255859375, 211.81143188476562, 107.86825561523438, 19.292022705078125, -12.236541748046875, 72.48992919921875, 93.32241821289062, -16.78680419921875, 145.83395385742188, 45.6561279296875, 31.581863403320312, 170.10946655273438, -45.052734375, 172.4324951171875, 34.3497314453125, 6.7873687744140625, 180.04495239257812, 40.2598876953125, 11.484222412109375, 11.471832275390625, 141.5400390625, 59.31964111328125, 151.35902404785156, -7.042232513427734, 37.642181396484375, 14.561748504638672, -2.4506072998046875, 148.304443359375, -27.76983642578125, 131.3408203125, 132.8561248779297, 189.88800048828125, -34.613731384277344, 19.23980712890625, 175.568115234375, 11.840423583984375, 60.05962371826172, -2.0825881958007812, 155.77679443359375, 53.81494140625, 31.02398681640625, 162.16400146484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000319.npy"}
{"epoch": 0.6680628272251309, "step": 320, "batch_size": 128, "mean": 58.266090393066406, "std": 93.41276550292969, "min": -168.0660400390625, "p10": -48.33162231445312, "median": 55.16374588012695, "p90": 166.01384582519532, "max": 212.41036987304688, "pos_frac": 0.7265625, "sample": [126.95106506347656, 85.08132934570312, -93.16448974609375, -156.79061889648438, 163.3126220703125, 162.07345581054688, -80.63543701171875, 7.7374420166015625, 157.82955932617188, 162.27230834960938, 21.74658203125, 128.67279052734375, 21.62481689453125, 11.828506469726562, 175.38919067382812, 97.85885620117188, -17.54962158203125, 0.041015625, 17.984619140625, 37.82160949707031, 19.475143432617188, 153.05145263671875, 43.8895263671875, 1.81787109375, 0.937255859375, -82.39315795898438, 183.53314208984375, 118.08909606933594, 47.52716064453125, 58.54920959472656, 115.91966247558594, 142.0279541015625, 166.75701904296875, 18.394195556640625, 53.47430419921875, 151.01806640625, -11.2191162109375, 168.00802612304688, 8.97491455078125, -1.2563095092773438, 5.06641960144043, 69.54373168945312, 124.81222534179688, -27.376022338867188, 175.6277618408203, 191.3856201171875, 123.37615966796875, -20.85504150390625, 113.32672119140625, 109.44490051269531, -46.6732177734375, 11.5008544921875, -145.25100708007812, -28.49859619140625, 177.76589965820312, -20.4678955078125, 110.91259765625, -52.20123291015625, 123.28717041015625, 164.819091796875, 146.67169189453125, 162.407470703125, 9.55401611328125, 154.828125, 165.69534301757812, -4.9113311767578125, 158.03549194335938, -19.64508056640625, 176.23455810546875, 151.5057373046875, 13.015922546386719, -4.369873046875, 25.551071166992188, -14.1337890625, -168.0660400390625, 112.29762268066406, 11.627471923828125, 161.168212890625, 8.75958251953125, 142.0705108642578, -140.58612060546875, 37.468109130859375, 1.718759536743164, -1.52294921875, -21.150726318359375, 44.81224822998047, 128.68170166015625, 157.37664794921875, 111.1978759765625, -1.8896293640136719, -107.10737609863281, 48.01580810546875, 203.00079345703125, -11.48486328125, 136.66554260253906, -30.17135238647461, 11.770217895507812, 110.60474395751953, -24.26190185546875, 172.92300415039062, 89.23089599609375, -37.26251220703125, 89.13946533203125, 165.50271606445312, 152.11175537109375, 212.41036987304688, 144.47903442382812, -16.513702392578125, 117.07147216796875, -132.99337768554688, 79.76545715332031, 101.90127563476562, -146.66769409179688, -5.845001220703125, 175.77166748046875, 25.467071533203125, 158.46755981445312, 73.96794128417969, 153.7811737060547, 120.81044006347656, -26.09650993347168, 88.0052490234375, 26.472488403320312, -136.78115844726562, 193.50497436523438, 56.853187561035156, -161.9712677001953, 136.91212463378906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000320.npy"}
{"epoch": 0.6701570680628273, "step": 321, "batch_size": 128, "mean": 50.88386535644531, "std": 84.49235534667969, "min": -176.70513916015625, "p10": -43.79491539001463, "median": 30.523345947265625, "p90": 163.21630401611327, "max": 199.909912109375, "pos_frac": 0.734375, "sample": [158.08316040039062, -74.59468078613281, 120.150390625, -157.88873291015625, 27.685455322265625, 179.90673828125, 31.6759033203125, 12.3167724609375, -61.79475402832031, 182.96185302734375, 137.4622039794922, 16.505172729492188, -76.38908386230469, 81.03787994384766, 110.0611572265625, 8.5262451171875, 8.575279235839844, 152.23785400390625, 175.43267822265625, 3.4859771728515625, 102.45854949951172, -83.347900390625, 2.192821502685547, -10.089599609375, 172.5787353515625, 122.55256652832031, -1.0740203857421875, 6.11865234375, 121.2886962890625, 5.50732421875, 135.59445190429688, 80.50738525390625, 112.03340148925781, 14.582534790039062, 43.30108642578125, -58.4427490234375, -35.4881591796875, 112.37138366699219, 2.6583709716796875, 141.97760009765625, 80.52024841308594, 31.605804443359375, -7.89788818359375, 42.634124755859375, 10.375518798828125, 174.52964782714844, -23.017776489257812, 160.6023406982422, 199.909912109375, 121.8653564453125, 34.92521667480469, 19.530563354492188, -38.44838333129883, 40.225372314453125, 1.02484130859375, 10.536483764648438, 157.35116577148438, -9.9259033203125, 159.58514404296875, -23.80426025390625, 29.440887451171875, 110.99000549316406, 52.233238220214844, 146.82574462890625, 71.37234497070312, 165.76565551757812, -1.555877685546875, 69.95733642578125, -57.75255584716797, -5.7650299072265625, -13.271408081054688, 54.328826904296875, -35.97285461425781, 178.658935546875, -164.93301391601562, 131.34629821777344, 179.3062744140625, 194.46621704101562, 41.771759033203125, 135.14175415039062, 22.45166015625, -33.55145263671875, 120.890625, 119.84608459472656, 9.340972900390625, 162.70938110351562, -56.27015686035156, 164.3991241455078, -113.654296875, 138.342529296875, 13.598617553710938, 15.881441116333008, 117.92620849609375, -1.9017715454101562, -5.591342926025391, 193.9580078125, 7.596435546875, 91.38052368164062, 28.927719116210938, 120.88348388671875, 137.08169555664062, -3.404693603515625, 112.53115844726562, 15.330062866210938, -8.133056640625, 134.32073974609375, 5.869709014892578, 7.424839019775391, 1.092935562133789, 1.119476318359375, -30.477401733398438, -62.975738525390625, 10.419260025024414, -24.690155029296875, 91.0384292602539, 183.98239135742188, -176.70513916015625, 157.77392578125, 37.273529052734375, 88.7026138305664, 16.847076416015625, 66.21820068359375, 2.03082275390625, -3.6379318237304688, 154.79547119140625, 156.009521484375, 0.0, -121.06802368164062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000321.npy"}
{"epoch": 0.6722513089005235, "step": 322, "batch_size": 128, "mean": 60.523075103759766, "std": 83.66630554199219, "min": -281.154541015625, "p10": -33.923686218261714, "median": 45.49611854553223, "p90": 168.40679931640625, "max": 209.02828979492188, "pos_frac": 0.7578125, "sample": [-57.919464111328125, -35.172607421875, -56.35064697265625, 105.37390899658203, 34.56340026855469, 15.036270141601562, 41.59001159667969, -43.1942138671875, 150.08265686035156, 72.51947021484375, -15.777984619140625, 21.75140380859375, 157.22201538085938, 9.56561279296875, 158.2199249267578, 129.3900146484375, 33.9393310546875, 1.23779296875, 128.69044494628906, 75.141357421875, 53.2313232421875, -14.947540283203125, 177.82196044921875, 0.1228485107421875, 0.9492568969726562, 39.31585693359375, -31.915298461914062, 116.83465576171875, 5.928741455078125, 143.20013427734375, 84.23013305664062, 145.77650451660156, 0.8117485046386719, -9.307281494140625, 140.7605743408203, 33.826629638671875, 14.001495361328125, 149.04505920410156, -35.002410888671875, 135.79766845703125, 149.046875, 129.63507080078125, -56.25926208496094, 15.493408203125, -7.3607177734375, 6.1965484619140625, 108.3108139038086, -33.64894104003906, 133.95297241210938, -141.20281982421875, 137.7082977294922, 2.6568756103515625, 153.7958984375, 202.27969360351562, -7.541961669921875, 142.79312133789062, 120.54512023925781, -17.168502807617188, 52.16917419433594, 129.4528350830078, 173.2303466796875, 17.45220947265625, 205.77801513671875, 209.02828979492188, -45.81915283203125, 133.87420654296875, 101.57466125488281, 30.868133544921875, 122.67747497558594, 13.026412963867188, 185.5191650390625, 164.23382568359375, 122.61376953125, 62.808197021484375, 11.314361572265625, 0.0, -63.49664306640625, 47.44891357421875, 60.908905029296875, -34.56475830078125, 170.986083984375, 44.696842193603516, 147.88702392578125, 165.462646484375, -24.213607788085938, 65.08712768554688, 48.869903564453125, 10.32904052734375, 0.270111083984375, 142.39378356933594, 178.90240478515625, -3.405975341796875, -12.10821533203125, -43.0911865234375, -18.0372314453125, 0.2829170227050781, 106.8760986328125, 0.8965301513671875, -281.154541015625, 76.06298828125, 153.85614013671875, 82.73382568359375, 38.12322998046875, 184.98489379882812, 167.3013916015625, -36.49956512451172, -2.152008056640625, 202.04823303222656, 199.28936767578125, 6.67987060546875, -12.618949890136719, 148.2069549560547, 20.897216796875, 46.29539489746094, -10.89385986328125, 11.83310317993164, -16.24895477294922, 19.897064208984375, 23.32037353515625, 125.5450439453125, 84.60498046875, 188.77597045898438, 111.63388061523438, -5.80572509765625, 18.0185546875, 158.10498046875, 68.94537353515625, 177.36241149902344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000322.npy"}
{"epoch": 0.6743455497382199, "step": 323, "batch_size": 128, "mean": 39.196693420410156, "std": 95.51254272460938, "min": -248.73095703125, "p10": -80.63355255126953, "median": 20.842254638671875, "p90": 160.51932678222656, "max": 203.8092803955078, "pos_frac": 0.6640625, "sample": [14.829833984375, 6.180023193359375, -151.0244903564453, -128.226806640625, 138.214599609375, 150.3011016845703, -79.13877868652344, 94.82948303222656, 30.0843505859375, 194.56219482421875, -106.94010925292969, 145.23062133789062, 24.47174072265625, -20.5672607421875, 2.7508392333984375, 122.41699981689453, 169.94993591308594, -7.2068939208984375, 90.07177734375, -32.32568359375, 193.0770263671875, 116.2392578125, 22.98822021484375, 153.33355712890625, 24.03228759765625, 177.69703674316406, 8.4183349609375, 142.60791015625, -104.76060485839844, 155.9099578857422, -17.787078857421875, -193.51983642578125, -5.481193542480469, 9.53924560546875, 0.2719593048095703, 116.0545654296875, 23.448822021484375, -7.069374084472656, 69.82199096679688, 18.645776748657227, 28.072341918945312, 145.21060180664062, -29.337738037109375, 104.4637451171875, 147.88177490234375, -8.437286376953125, -22.30694580078125, -248.73095703125, 164.18084716796875, 9.867179870605469, 8.96090316772461, 0.5172233581542969, -80.12353515625, 9.865840911865234, 79.49819946289062, 70.18951416015625, 58.345977783203125, 156.6845703125, 203.8092803955078, -28.19196319580078, 82.67193603515625, 76.3094482421875, 138.48648071289062, -22.1866455078125, 168.727294921875, 24.47027587890625, 142.94300842285156, -18.448760986328125, -51.983642578125, 150.5001220703125, 37.55816650390625, 2.641632080078125, -81.82359313964844, -1.26434326171875, 137.73277282714844, 91.91954040527344, 185.896240234375, -76.32600402832031, 182.67166137695312, 4.799163818359375, 134.79476928710938, -156.92466735839844, 3.7316055297851562, 37.067047119140625, 20.72314453125, 185.34378051757812, -47.47599792480469, 155.53219604492188, -147.06475830078125, -4.706401824951172, 149.5294189453125, 41.383819580078125, -52.21333312988281, 128.13241577148438, -0.012359619140625, -8.941879272460938, -101.82259368896484, -13.857597351074219, -52.573577880859375, -7.199554443359375, 20.96136474609375, -113.49017333984375, -12.008941650390625, 12.202651977539062, 18.6175537109375, 56.0057373046875, 6.474983215332031, -6.343742370605469, 140.10540771484375, 185.1181640625, -2.127584457397461, 78.31344604492188, 165.58717346191406, 153.24278259277344, -140.54312133789062, -146.86380004882812, 74.05314636230469, 27.0384521484375, 125.13893127441406, 15.460369110107422, 18.36056137084961, -7.26617431640625, 114.06024169921875, 158.95010375976562, 134.22445678710938, -31.98944091796875, 3.599079132080078, 199.2041015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000323.npy"}
{"epoch": 0.6764397905759162, "step": 324, "batch_size": 128, "mean": 66.98324584960938, "std": 80.99124908447266, "min": -126.53997802734375, "p10": -10.005941772460934, "median": 42.07286071777344, "p90": 168.67036132812498, "max": 216.23907470703125, "pos_frac": 0.7890625, "sample": [116.41217041015625, 170.385498046875, 15.08062744140625, 139.14239501953125, 25.52410888671875, 137.5179443359375, 4.040130615234375, 9.10441780090332, 146.776123046875, 34.13629150390625, 35.6566162109375, 1.6290473937988281, -61.324066162109375, 176.22317504882812, 7.554107666015625, 181.30557250976562, 137.55270385742188, 11.74831771850586, -11.8067626953125, 0.0, 9.138916015625, 8.002281188964844, 142.39413452148438, -6.347015380859375, 4.150362014770508, 33.48114013671875, 149.78477478027344, -57.214447021484375, 144.25082397460938, -46.62274932861328, 6.412626266479492, 152.22860717773438, 160.24502563476562, 167.26235961914062, 141.73263549804688, 138.19720458984375, -20.337677001953125, 53.729522705078125, 80.86627197265625, 0.0, -6.70469856262207, 158.53314208984375, 173.45001220703125, 119.16925048828125, 47.292388916015625, 143.76217651367188, 0.0, -2.561767578125, 117.01730346679688, -9.234161376953125, -112.54673767089844, 45.48216247558594, 50.17913818359375, 21.7293701171875, 124.456787109375, 105.68006896972656, 10.410476684570312, 148.53309631347656, 174.46380615234375, 3.05718994140625, -59.525718688964844, 184.033935546875, 10.312667846679688, 143.92950439453125, 123.0559310913086, -2.1913795471191406, 144.343017578125, 148.61843872070312, 16.2423095703125, -121.57853698730469, 167.51683044433594, 87.20381927490234, -126.53997802734375, 140.62179565429688, -6.88775634765625, -7.9745330810546875, 9.576370239257812, 101.3511962890625, 12.12017822265625, 216.23907470703125, 197.37966918945312, 7.739278793334961, -19.355484008789062, 92.36947631835938, 194.620361328125, 162.36410522460938, 18.225982666015625, 21.181732177734375, -6.03411865234375, 4.485748291015625, 150.0104217529297, 39.602935791015625, 2.0433120727539062, 0.075469970703125, -2.8621177673339844, 59.0538330078125, 31.41292953491211, 174.4141845703125, 142.95742797851562, 2.306121826171875, -44.97216033935547, -4.623405456542969, 8.895370483398438, 37.030242919921875, -52.25921630859375, 209.1802978515625, 192.66802978515625, 176.40859985351562, 78.82577514648438, 8.14385986328125, 162.7789306640625, 119.12176513671875, -23.71143341064453, 132.01165771484375, 44.54278564453125, 141.26597595214844, 167.935302734375, 142.4477996826172, -0.18217086791992188, 160.7581024169922, 145.46734619140625, 0.913818359375, 146.04119873046875, 155.0605010986328, 152.0373992919922, 10.229873657226562, 3.6859130859375, 1.5129547119140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000324.npy"}
{"epoch": 0.6785340314136126, "step": 325, "batch_size": 128, "mean": 56.38215637207031, "std": 99.70594787597656, "min": -158.71804809570312, "p10": -58.11977386474608, "median": 30.54267120361328, "p90": 184.7837341308594, "max": 290.1083679199219, "pos_frac": 0.6640625, "sample": [73.89254760742188, 158.884521484375, -128.62677001953125, 11.45025634765625, 126.97906494140625, 22.38714599609375, 63.37139892578125, 160.59201049804688, 34.05364990234375, -2.379608154296875, 8.772216796875, 12.150299072265625, 155.50914001464844, 154.79183959960938, -7.530914306640625, 5.127790451049805, 172.8264617919922, 117.4119873046875, 132.58544921875, 190.96902465820312, 150.86187744140625, 186.51377868652344, 110.99560546875, -2.7400245666503906, 148.14944458007812, 2.3656082153320312, 13.539924621582031, 83.64730834960938, 142.2683868408203, 28.753509521484375, -124.06330871582031, 7.490234375, 97.55229187011719, 175.07676696777344, 177.5010986328125, 120.39967346191406, 22.141260147094727, 177.4490509033203, 37.92640686035156, -92.5814208984375, 137.03134155273438, -158.71804809570312, 169.42210388183594, -0.4311065673828125, -80.50486755371094, -1.360321044921875, 290.1083679199219, -138.25816345214844, 198.7719268798828, -3.064483642578125, -1.4337005615234375, 77.8358154296875, 173.7989501953125, 13.37982177734375, 9.345199584960938, -123.69318389892578, -9.4390869140625, -129.21917724609375, -8.392303466796875, -68.56471252441406, 80.18319702148438, 147.75033569335938, -7.510810852050781, -0.59124755859375, 85.68016052246094, 62.3018798828125, 2.6000518798828125, -131.03582763671875, 207.40573120117188, -9.875885009765625, 121.98312377929688, 21.45728302001953, 260.6893310546875, -9.657073974609375, -7.42303466796875, -2.175933837890625, -11.354339599609375, -8.996063232421875, 118.786865234375, 158.95733642578125, -147.27059936523438, -13.506195068359375, -53.64337158203125, 185.17376708984375, -116.82850646972656, 74.9151611328125, 138.95538330078125, 40.53265380859375, 46.231689453125, -28.63372802734375, -11.971038818359375, 136.77566528320312, 186.87432861328125, -33.05076599121094, 2.410747528076172, -155.42544555664062, -20.877212524414062, 189.90411376953125, 83.0599365234375, 154.04498291015625, 184.6165771484375, 32.33183288574219, 162.31370544433594, 213.80413818359375, 172.12203979492188, -38.492156982421875, 5.491123199462891, 21.34862518310547, 27.324691772460938, -0.98883056640625, -21.84466552734375, 167.061279296875, 55.184356689453125, 150.59188842773438, 8.445220947265625, 139.18771362304688, 147.3489990234375, 201.685302734375, 15.6734619140625, 1.1913986206054688, -40.07984924316406, -32.28857421875, 219.53594970703125, -25.595977783203125, -16.662612915039062, 218.6630859375, 74.1148681640625, 164.90611267089844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000325.npy"}
{"epoch": 0.680628272251309, "step": 326, "batch_size": 128, "mean": 56.132781982421875, "std": 96.03788757324219, "min": -258.0162353515625, "p10": -76.6211471557617, "median": 64.072509765625, "p90": 163.58357849121094, "max": 195.62693786621094, "pos_frac": 0.7109375, "sample": [158.00479125976562, 145.40371704101562, 140.42593383789062, 191.1464385986328, 136.52488708496094, 67.43109130859375, -51.384918212890625, -24.829742431640625, 146.10369873046875, -123.05516052246094, 167.6629638671875, -13.4775390625, 149.15541076660156, 154.36813354492188, 23.62090301513672, 3.3739356994628906, 5.829010009765625, 166.74163818359375, 195.62693786621094, 24.785629272460938, 67.53842163085938, 150.85316467285156, 6.0564727783203125, -26.31134033203125, 66.05914306640625, 121.48033142089844, 131.09083557128906, 93.11419677734375, 79.3521728515625, 18.796737670898438, -131.83746337890625, 164.50064086914062, 0.81475830078125, 79.68125915527344, -140.91510009765625, -113.06072998046875, 27.207443237304688, 117.82520294189453, -35.02093505859375, 150.7938232421875, 94.60372924804688, 124.24575805664062, -28.942108154296875, 153.16888427734375, 50.560028076171875, 103.02220153808594, -70.93501281738281, 152.3437957763672, -30.97088623046875, 178.8208465576172, 145.78802490234375, 105.05914306640625, 158.09933471679688, 18.60956573486328, 161.53561401367188, -138.02682495117188, -38.149749755859375, 21.8233642578125, 141.59909057617188, -10.819194793701172, 139.61195373535156, 126.03140258789062, 24.8421630859375, 37.509429931640625, 0.0, -258.0162353515625, 15.324256896972656, -2.79931640625, 144.95925903320312, -89.8887939453125, 20.62005615234375, 175.78387451171875, 190.79879760742188, 100.28561401367188, 39.952476501464844, -57.301788330078125, 34.48131561279297, -110.57432556152344, 32.527252197265625, 130.11947631835938, -3.355926513671875, 173.13742065429688, -5.921905517578125, 17.451141357421875, 12.362220764160156, 151.75192260742188, -36.35748291015625, -121.68246459960938, 124.47415161132812, 14.897476196289062, 148.57528686523438, 16.469482421875, 179.0013427734375, -9.439056396484375, 103.95791625976562, -16.465576171875, 130.76882934570312, 163.1905517578125, 24.880645751953125, 136.12432861328125, 172.5352020263672, 0.0, 146.91641235351562, 71.32972717285156, 62.08587646484375, 144.0167236328125, -44.18487548828125, 161.1160125732422, 117.65278625488281, 0.0, 146.80569458007812, -170.6142578125, -24.8880615234375, -98.45221710205078, 50.989654541015625, -135.79083251953125, 172.74725341796875, -3.47430419921875, 160.39480590820312, 26.682621002197266, 115.78096771240234, 24.62176513671875, -90.31777954101562, -54.17454528808594, 150.9134063720703, 166.63436889648438, 86.56179809570312, 148.10797119140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000326.npy"}
{"epoch": 0.6827225130890052, "step": 327, "batch_size": 128, "mean": 64.78064727783203, "std": 92.89358520507812, "min": -171.723388671875, "p10": -24.810342407226564, "median": 39.68790817260742, "p90": 178.9358947753906, "max": 235.06243896484375, "pos_frac": 0.7734375, "sample": [39.972984313964844, 147.1631622314453, 135.81210327148438, 8.601852416992188, 211.19802856445312, -160.30361938476562, 149.80215454101562, 155.39141845703125, 7.46240234375, 127.66700744628906, -1.815500259399414, 195.21481323242188, -5.33856201171875, -155.18905639648438, 65.4929428100586, -20.1177978515625, -60.34893798828125, 10.824951171875, 193.4349365234375, 184.25076293945312, 129.89456176757812, 117.56639862060547, 19.741851806640625, 78.89796447753906, 143.02218627929688, -89.584228515625, 11.596664428710938, 124.40272521972656, 10.729888916015625, 140.12625122070312, 42.01300048828125, 191.17440795898438, -4.896364212036133, -7.180747985839844, 67.70762634277344, 124.60311889648438, 184.88482666015625, 10.501815795898438, 235.06243896484375, 127.37237548828125, -24.1292724609375, -5.3442840576171875, -9.758934020996094, 125.48451232910156, 35.93543243408203, -12.977081298828125, 178.2540283203125, 39.40283203125, 132.41465759277344, 10.401626586914062, 160.76937866210938, 6.000396728515625, -12.828140258789062, -171.723388671875, 131.87588500976562, 17.844512939453125, 149.75271606445312, 133.95628356933594, 190.2113037109375, 143.7847900390625, 6.8764495849609375, 152.94869995117188, 127.36585998535156, 38.645660400390625, 178.62408447265625, 160.61785888671875, 188.8568115234375, 39.20989990234375, 99.52035522460938, -24.958404541015625, 32.643035888671875, 3.291229248046875, -13.208740234375, -65.5633773803711, 136.86477661132812, 9.976028442382812, 32.91932678222656, 62.08349609375, 185.70806884765625, 11.46893310546875, -24.74688720703125, 2.147308349609375, 22.48760986328125, 128.86602783203125, 25.096267700195312, 179.6634521484375, 22.454833984375, 0.0, 165.2327880859375, -5.703727722167969, 25.45294189453125, 173.367431640625, 198.94790649414062, 26.75921630859375, 149.15135192871094, 16.86005401611328, 153.45834350585938, 35.84259033203125, 153.34056091308594, 157.24136352539062, 66.66265869140625, 13.644302368164062, -161.8021240234375, -45.18311309814453, 122.02484130859375, 131.52993774414062, -110.00392150878906, 4.0391693115234375, 69.12744140625, 147.55189514160156, -150.89208984375, 133.500244140625, 184.55838012695312, 8.22943115234375, -40.701812744140625, 152.54855346679688, -5.128395080566406, 175.47824096679688, 1.7681026458740234, -5.2621612548828125, 173.754150390625, 163.03634643554688, 149.53390502929688, 7.539276123046875, -155.05477905273438, 162.25912475585938, 22.081893920898438, 1.134368896484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000327.npy"}
{"epoch": 0.6848167539267016, "step": 328, "batch_size": 128, "mean": 60.289451599121094, "std": 86.18779754638672, "min": -159.822998046875, "p10": -25.716674041748046, "median": 40.188079833984375, "p90": 167.71917114257812, "max": 257.00946044921875, "pos_frac": 0.765625, "sample": [40.33251953125, 146.32286071777344, 159.86264038085938, 41.04718017578125, 207.42596435546875, 30.49102783203125, 2.604949951171875, 140.1278839111328, 142.07101440429688, 139.98069763183594, 139.23858642578125, 89.65351867675781, 4.08709716796875, 128.932373046875, 74.8388442993164, -0.43167686462402344, 11.8468017578125, -8.295654296875, 49.39697265625, 14.566757202148438, 64.09724426269531, 108.118408203125, 114.4434814453125, -25.914451599121094, 193.1317138671875, 34.903594970703125, -5.1679840087890625, 45.63336181640625, 123.68975830078125, 132.85324096679688, -4.7993316650390625, 135.23019409179688, 34.742095947265625, 172.50872802734375, 211.9874267578125, 109.97161102294922, 187.8733367919922, 12.239471435546875, 188.8115234375, -0.3986644744873047, 4.3616943359375, 150.06346130371094, 90.6544189453125, -2.0896759033203125, 7.785858154296875, -6.79039192199707, -23.74761962890625, 33.30303955078125, -26.555908203125, 152.81578063964844, 55.0799560546875, 19.52886962890625, -26.217620849609375, 182.4197998046875, 15.689865112304688, 33.545654296875, 11.0341796875, 11.266067504882812, 100.8346176147461, -37.03173828125, 37.24866485595703, 52.21624755859375, 175.7784423828125, 165.66650390625, 142.3562469482422, 174.240478515625, 134.82568359375, -74.09588623046875, 150.14532470703125, -114.58274841308594, 14.237594604492188, 135.49053955078125, 135.444091796875, 145.9365692138672, 27.192413330078125, 243.56689453125, 192.61056518554688, 26.68963623046875, 3.64544677734375, 120.51898193359375, -159.49624633789062, 9.690544128417969, 156.16021728515625, 53.50700378417969, -127.16464233398438, 10.041961669921875, 137.58473205566406, 138.4625244140625, 135.58184814453125, -22.81622314453125, -15.469696044921875, 185.7080078125, 11.959095001220703, -0.29573822021484375, 7.03863525390625, -22.196853637695312, 8.934738159179688, 25.004150390625, -132.3245849609375, 33.751434326171875, 61.10096740722656, 35.83502197265625, 131.26654052734375, 3.295562744140625, -0.8268356323242188, 156.74676513671875, 156.693359375, -159.822998046875, -4.10858154296875, -25.631912231445312, 40.04364013671875, 20.239471435546875, 158.846435546875, -61.28955078125, 26.958831787109375, 26.301254272460938, -1.102996826171875, 134.01123046875, -5.71544075012207, -148.42767333984375, 54.919403076171875, 163.02450561523438, 47.23152160644531, -66.03311157226562, 67.77703857421875, 257.00946044921875, 135.08082580566406, 56.8291015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000328.npy"}
{"epoch": 0.6869109947643979, "step": 329, "batch_size": 128, "mean": 50.043365478515625, "std": 99.33186340332031, "min": -189.81118774414062, "p10": -63.687728881835916, "median": 27.223114013671875, "p90": 179.12743530273437, "max": 239.17491149902344, "pos_frac": 0.671875, "sample": [-28.381317138671875, 9.121826171875, -6.0301513671875, 43.8209228515625, 3.1257171630859375, -5.338047027587891, 6.6854400634765625, 27.748046875, 3.730987548828125, 32.975555419921875, 158.0482635498047, -18.124923706054688, -170.998779296875, 129.35208129882812, 120.71638488769531, 190.5377197265625, 6.754486083984375, -57.309661865234375, 160.1355438232422, 112.06619262695312, 135.74078369140625, 22.8922119140625, 169.82211303710938, -19.20343017578125, 138.08453369140625, 165.87159729003906, -147.1318817138672, 133.02816772460938, 188.59864807128906, 94.068115234375, 1.696197509765625, 24.03692626953125, -1.8482666015625, 170.11251831054688, -44.102783203125, 42.75848388671875, 116.30445861816406, 138.901611328125, 22.529220581054688, 119.64132690429688, -2.96966552734375, 184.99560546875, 0.747406005859375, -81.26214599609375, 153.236572265625, -105.1187744140625, -20.633193969726562, 169.35882568359375, 16.6846923828125, -185.49468994140625, -4.948163986206055, 183.4725341796875, -29.388153076171875, -22.35040283203125, 102.009033203125, 140.5222930908203, 76.549560546875, 31.561782836914062, 62.638641357421875, -157.10382080078125, 7.935680389404297, 165.90335083007812, -5.93206787109375, -115.99479675292969, -1.1984024047851562, -5.170648574829102, -0.4925727844238281, -19.630950927734375, 214.15386962890625, 35.27001190185547, 25.823074340820312, 1.706787109375, 155.10086059570312, 226.6934814453125, 14.535280227661133, 107.45211791992188, 72.96023559570312, -25.4188232421875, -123.19376373291016, 162.08282470703125, 148.39175415039062, -16.092132568359375, -189.81118774414062, 133.4071044921875, -13.035369873046875, 187.52259826660156, 48.2376708984375, 2.87255859375, 177.794921875, 3.764678955078125, 200.0550537109375, 2.1458053588867188, -54.87516784667969, 187.37515258789062, 143.69281005859375, -144.7581787109375, 239.17491149902344, 182.23663330078125, -2.5668373107910156, 142.73782348632812, -5.1815032958984375, 12.910308837890625, 130.30886840820312, 55.66358947753906, -2.494354248046875, 1.04229736328125, 223.36767578125, 62.52056884765625, 155.11497497558594, 225.583740234375, -16.9029541015625, 124.25370788574219, 145.70932006835938, -6.8624267578125, 35.470703125, -114.22992706298828, 26.69818115234375, -19.431732177734375, 81.87376403808594, -4.249267578125, 103.11216735839844, 145.8135986328125, 0.7336311340332031, 124.26409912109375, 150.45445251464844, 46.44035339355469, -78.56988525390625, -171.66357421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000329.npy"}
{"epoch": 0.6890052356020943, "step": 330, "batch_size": 128, "mean": 41.9713134765625, "std": 93.50900268554688, "min": -171.900390625, "p10": -87.92027587890624, "median": 25.468120574951172, "p90": 170.46470947265624, "max": 278.18621826171875, "pos_frac": 0.703125, "sample": [-15.544464111328125, 147.25265502929688, 20.693603515625, -144.47833251953125, 27.40435028076172, 114.54638671875, 10.4560546875, 192.3909912109375, 141.72914123535156, 96.61079406738281, -78.868896484375, 117.66162109375, -7.924072265625, -171.900390625, 117.20440673828125, -6.1944732666015625, 170.89990234375, 30.891143798828125, -1.20513916015625, -5.1876220703125, 162.668212890625, -147.56350708007812, -35.66856384277344, 7.49017333984375, 3.845947265625, 125.66790771484375, 26.16375732421875, -60.89942169189453, 7.6455841064453125, -0.5886993408203125, 64.63394165039062, 24.854156494140625, 184.8760986328125, 124.60493469238281, 14.330787658691406, -6.039276123046875, -1.71588134765625, -98.54635620117188, -98.15692138671875, 18.259979248046875, 43.03375244140625, 183.75653076171875, 38.6695556640625, 43.361480712890625, 18.84844970703125, 136.89828491210938, 18.022178649902344, 202.80291748046875, 15.0540771484375, 117.78866577148438, 1.6608543395996094, 170.2781982421875, 92.71353149414062, -140.91061401367188, 1.357818603515625, 26.331832885742188, -135.24765014648438, 33.70001220703125, 21.50958251953125, 109.44741821289062, 15.611740112304688, 186.1110382080078, 13.791015625, 87.25320434570312, -127.61470031738281, 7.286468505859375, 146.66163635253906, -83.53314208984375, 7.545017242431641, 12.581390380859375, 45.203338623046875, -138.5391082763672, 193.68972778320312, 124.84512329101562, -4.09478759765625, 126.04446411132812, -142.93978881835938, 188.33779907226562, -22.887664794921875, 2.886627197265625, 27.457763671875, 154.5780792236328, -10.227813720703125, -13.654754638671875, 27.91143798828125, 7.8502044677734375, 35.59027099609375, 165.95803833007812, 174.319580078125, 118.6283187866211, 186.1096649169922, 12.90594482421875, 122.18863677978516, 42.900970458984375, 10.8197021484375, -19.36273193359375, 19.8505859375, 14.789773941040039, 116.05171203613281, 150.33901977539062, 189.2877197265625, 36.283782958984375, -139.39950561523438, 107.5903549194336, -131.19537353515625, 161.4215545654297, -14.79901123046875, 26.08208465576172, -37.01226043701172, 90.49386596679688, 160.30172729492188, -122.52831268310547, -2.1770401000976562, -66.47402954101562, 0.0989990234375, 195.9102783203125, 36.866607666015625, 101.58326721191406, -1.55364990234375, 149.27272033691406, -26.280303955078125, 278.18621826171875, 66.92605590820312, -2.6619873046875, -29.097686767578125, 149.60894775390625, 103.84039306640625, 37.131256103515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000330.npy"}
{"epoch": 0.6910994764397905, "step": 331, "batch_size": 128, "mean": 52.833900451660156, "std": 90.26664733886719, "min": -186.5515899658203, "p10": -36.41537322998047, "median": 33.08610534667969, "p90": 177.77603759765626, "max": 206.5372314453125, "pos_frac": 0.765625, "sample": [0.9215545654296875, 202.15744018554688, 187.2601318359375, -112.80667114257812, 206.5372314453125, 48.60723876953125, 59.53594970703125, -11.013641357421875, 45.33062744140625, 130.77197265625, -4.795166015625, -47.184539794921875, 114.667236328125, -118.87445068359375, 198.43505859375, -27.032440185546875, 5.0354766845703125, 4.259798049926758, -9.576766967773438, -94.03460693359375, 15.056159973144531, 52.89324951171875, -28.852432250976562, 34.008453369140625, 18.569869995117188, 168.08164978027344, 142.06387329101562, 70.35711669921875, 81.14913940429688, 116.92640686035156, 14.9697265625, 46.88482666015625, -14.2904052734375, 47.446990966796875, 50.15699768066406, 178.48855590820312, 3.672088623046875, 119.17022705078125, 183.66339111328125, 138.6820068359375, -109.83948516845703, 12.9434814453125, -11.787849426269531, -173.33958435058594, 4.239831924438477, 9.058719635009766, 172.2040557861328, 77.621826171875, 101.1884765625, 0.0, 60.890472412109375, -1.7486724853515625, 140.60195922851562, 2.70745849609375, 10.09967041015625, 81.43426513671875, 10.947906494140625, 161.29867553710938, 40.699676513671875, 59.6365966796875, 28.162109375, 51.23687744140625, -35.6507568359375, -22.039596557617188, 193.05459594726562, 118.78752899169922, 177.47067260742188, 137.55992126464844, 12.760993957519531, 54.52826690673828, 4.873504638671875, 18.29949951171875, 166.45611572265625, -29.4766845703125, 37.18622970581055, 26.605113983154297, -59.1446533203125, 5.148284912109375, 7.963569641113281, -108.62958526611328, 97.81890869140625, 32.472869873046875, 131.62374877929688, 33.6993408203125, -186.5515899658203, 202.60943603515625, 7.4141845703125, 171.9561767578125, -38.19947814941406, 25.420196533203125, 193.57962036132812, 136.93582153320312, 122.3447036743164, 196.34490966796875, 21.345977783203125, 181.051513671875, 172.89329528808594, 5.949195861816406, -17.50518798828125, -5.59234619140625, -78.29229736328125, 0.74847412109375, 163.95286560058594, 147.59031677246094, -154.64796447753906, 27.968154907226562, 204.30059814453125, 3.2712650299072266, 149.04348754882812, -2.228792190551758, 3.6065673828125, 16.073585510253906, 155.20480346679688, 11.2615966796875, 133.74668884277344, 147.47802734375, -1.099609375, 67.061767578125, 19.869041442871094, 143.74559020996094, -165.80316162109375, 21.916236877441406, 193.13375854492188, 156.78468322753906, 129.0826416015625, -33.059722900390625, 149.21307373046875, 15.897102355957031], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000331.npy"}
{"epoch": 0.6931937172774869, "step": 332, "batch_size": 128, "mean": 66.28262329101562, "std": 96.4311294555664, "min": -199.00967407226562, "p10": -44.99922180175781, "median": 75.917236328125, "p90": 176.7306671142578, "max": 254.85043334960938, "pos_frac": 0.75, "sample": [116.57524108886719, 4.9954833984375, 11.627479553222656, 200.82403564453125, 139.6676483154297, 148.60287475585938, 98.60484313964844, 50.330406188964844, 202.5223388671875, 146.54458618164062, -3.7820587158203125, 8.246835708618164, 8.703657150268555, -5.339630126953125, 120.52476501464844, 121.12176513671875, -37.4317626953125, 78.77946472167969, 157.06118774414062, -39.12858581542969, 75.89468383789062, -5.570709228515625, 45.368743896484375, 118.7970199584961, 0.7205352783203125, -5.2283935546875, -129.16128540039062, 5.3398590087890625, 14.7864990234375, 12.50384521484375, 168.06207275390625, 184.75314331054688, 5.53021240234375, 3.0185546875, 169.84335327148438, 138.91552734375, 151.19595336914062, 190.781982421875, 108.587158203125, 125.45065307617188, 134.88406372070312, 5.456787109375, 144.0880126953125, 5.8746795654296875, 158.76083374023438, 84.55752563476562, 188.09716796875, 154.94580078125, -4.405731201171875, 24.053665161132812, 162.24026489257812, 196.1900634765625, -121.63177490234375, -57.83306884765625, 197.82598876953125, 140.92135620117188, -125.8294677734375, 70.30715942382812, 165.67193603515625, -199.00967407226562, 27.64525604248047, 174.98907470703125, 159.51327514648438, -2.519784927368164, 14.902366638183594, 8.758224487304688, 131.4859619140625, 19.963623046875, 13.469284057617188, -44.930145263671875, -36.194854736328125, 147.73324584960938, 103.63352966308594, 21.854736328125, -7.676910400390625, 184.11102294921875, 158.74179077148438, 145.16586303710938, -37.15202331542969, 136.5981903076172, -93.00100708007812, 132.81866455078125, -28.176223754882812, 3.19268798828125, 177.3626708984375, -120.8404541015625, 61.9208984375, 21.822479248046875, 5.9322509765625, 127.67349243164062, 170.69320678710938, 128.53070068359375, -110.77484893798828, 168.5487060546875, 92.40655517578125, -1.62628173828125, 153.015869140625, -30.564430236816406, -16.466690063476562, 117.9884033203125, 176.45980834960938, 136.87997436523438, -104.66650390625, -45.160400390625, 254.85043334960938, 201.4967041015625, 142.53311157226562, 18.561737060546875, 10.997528076171875, 34.66096496582031, 172.64500427246094, 143.58602905273438, 136.16989135742188, 191.86557006835938, 55.97503662109375, 148.462890625, -69.99046325683594, 161.1910400390625, -10.655437469482422, 131.90243530273438, 133.29397583007812, 188.21774291992188, -6.769279479980469, -122.16426086425781, 0.0, 10.53204345703125, -196.99278259277344, 75.93978881835938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000332.npy"}
{"epoch": 0.6952879581151833, "step": 333, "batch_size": 128, "mean": 47.911865234375, "std": 87.18694305419922, "min": -167.47695922851562, "p10": -58.65627288818358, "median": 38.87165832519531, "p90": 159.19737548828124, "max": 188.862060546875, "pos_frac": 0.7109375, "sample": [129.3177490234375, -167.47695922851562, -28.211544036865234, 173.33444213867188, 187.71206665039062, -41.13275146484375, 99.75772094726562, 114.63751220703125, 120.75576782226562, 68.049560546875, -55.45735168457031, 9.3995361328125, -19.981491088867188, 163.04000854492188, 155.34420776367188, 49.00885009765625, 152.18539428710938, 86.73727416992188, 75.06398010253906, 133.996826171875, -18.879898071289062, 41.959442138671875, 11.7703857421875, 161.068359375, 149.87123107910156, 52.70916748046875, 60.868927001953125, 42.2376708984375, 149.12149047851562, 25.649871826171875, 13.70355224609375, 188.862060546875, -12.87799072265625, 72.72698974609375, -155.53546142578125, 4.032501220703125, 55.317718505859375, -13.634765625, 149.66632080078125, 148.00704956054688, 115.93032836914062, 80.81535339355469, -6.89971923828125, 15.578826904296875, 119.159912109375, 153.4221954345703, 5.262697219848633, 180.10513305664062, 25.73876953125, -70.50108337402344, -66.12042236328125, 88.33694458007812, 0.0, -4.23626708984375, 151.56777954101562, 7.199005126953125, 167.45489501953125, 180.07568359375, 158.3370361328125, 57.416595458984375, -2.529052734375, 135.34913635253906, -43.80473327636719, 147.752685546875, 96.71490478515625, -94.04771423339844, 124.91226196289062, 187.42010498046875, 109.81076049804688, 35.78387451171875, -129.60203552246094, 81.92367553710938, 61.127105712890625, -72.35565185546875, 2.8301467895507812, 145.40296936035156, -1.53271484375, -3.84613037109375, 3.405487060546875, 143.86965942382812, 49.32560729980469, 72.61148834228516, 31.91131591796875, 175.76483154296875, -138.2489013671875, 59.13465881347656, 167.2476806640625, -5.45635986328125, -106.505615234375, 159.64859008789062, 52.21542739868164, 163.30343627929688, -5.758460998535156, 125.72171020507812, -26.152793884277344, -19.37890625, 24.62835693359375, 33.38330078125, -140.63168334960938, 128.81341552734375, -19.019989013671875, 134.71856689453125, -148.35501098632812, -32.5361328125, 71.89007568359375, 147.49436950683594, -1.062042236328125, 43.78558349609375, 159.00399780273438, 11.264617919921875, 144.225830078125, 3.7506752014160156, 137.00344848632812, 33.800506591796875, 14.757150650024414, -6.45294189453125, -144.1752166748047, 4.072364807128906, 7.271728515625, 25.316314697265625, -8.739471435546875, 5.30535888671875, -126.91731262207031, 15.734024047851562, -6.2017669677734375, 28.725921630859375, 14.20916748046875, 2.3184032440185547], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000333.npy"}
{"epoch": 0.6973821989528796, "step": 334, "batch_size": 128, "mean": 57.1641845703125, "std": 97.7696762084961, "min": -207.87449645996094, "p10": -45.298480224609364, "median": 40.10680389404297, "p90": 179.56860046386717, "max": 308.75665283203125, "pos_frac": 0.71875, "sample": [141.25216674804688, 157.19711303710938, -28.15325927734375, -13.47259521484375, 0.0, -8.58905029296875, 43.62071228027344, -106.13188171386719, 53.48625183105469, 2.1862754821777344, 34.565216064453125, 117.28732299804688, 13.7401123046875, 67.98455810546875, -41.133544921875, 145.24057006835938, -3.9549560546875, 179.21170043945312, 61.285125732421875, -15.59674072265625, 134.05421447753906, 69.92965698242188, 217.23182678222656, 176.29290771484375, -102.18463134765625, 36.21342468261719, 21.740875244140625, 34.82623291015625, -4.564689636230469, 16.000137329101562, 87.4781494140625, -10.622779846191406, -0.4206867218017578, 198.87747192382812, 148.04299926757812, 43.99249267578125, -4.315208435058594, 129.0709228515625, -184.84271240234375, 177.62799072265625, -1.4999580383300781, 6.521026611328125, 20.598663330078125, 16.7376708984375, 9.939338684082031, 138.305908203125, 11.0260009765625, -55.01666259765625, 147.0257110595703, -7.33349609375, 41.515350341796875, -95.30972290039062, 171.5044708251953, 138.0264892578125, 206.11111450195312, 151.25662231445312, -4.329782485961914, 43.64886474609375, 9.233535766601562, 12.31201171875, 12.575584411621094, 7.750572204589844, 206.8843994140625, 72.8160400390625, -68.91871643066406, -142.7494354248047, 46.536346435546875, -177.24319458007812, -17.340087890625, -16.179656982421875, 192.247802734375, 23.223846435546875, -18.38269805908203, 196.42190551757812, 162.602783203125, 145.5296630859375, 56.23551940917969, 14.137397766113281, 11.319580078125, 124.74176025390625, 35.666748046875, 33.88523483276367, -6.039337158203125, 141.73545837402344, 121.87791442871094, 188.3348846435547, 140.56130981445312, 68.21182250976562, 94.28106689453125, 103.86624145507812, 148.23199462890625, 148.65609741210938, 154.3887481689453, -18.108444213867188, 1.9571151733398438, 31.86760711669922, 1.97259521484375, -85.57650756835938, 169.71896362304688, 192.07940673828125, 180.4013671875, 17.92913818359375, -12.555908203125, -138.73269653320312, 159.34036254882812, -22.50677490234375, 126.58109283447266, 138.57559204101562, 187.58248901367188, 155.29989624023438, -207.87449645996094, 212.486572265625, 172.50619506835938, 0.6791057586669922, 143.46896362304688, -167.22036743164062, 15.625762939453125, 105.28927612304688, -40.877593994140625, 38.69825744628906, 60.953887939453125, 308.75665283203125, 159.9893798828125, 89.88632202148438, 87.39437866210938, -11.228073120117188, -103.4088134765625, 215.47039794921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000334.npy"}
{"epoch": 0.6994764397905759, "step": 335, "batch_size": 128, "mean": 48.23686981201172, "std": 97.96154022216797, "min": -185.2537841796875, "p10": -101.64589996337891, "median": 25.821178436279297, "p90": 174.96318969726562, "max": 226.43423461914062, "pos_frac": 0.6875, "sample": [-24.0946044921875, -162.11517333984375, 56.047935485839844, -46.803009033203125, 159.189208984375, -109.81631469726562, 194.90704345703125, -11.38653564453125, 85.71580505371094, 151.90110778808594, 47.591217041015625, -161.52548217773438, 44.3544921875, -12.954362869262695, -4.405143737792969, 22.946044921875, 1.4115371704101562, -2.473623275756836, -32.8997802734375, -133.4052734375, -16.69891357421875, 188.8570098876953, -5.549407958984375, 5.527862548828125, 15.166839599609375, -0.442474365234375, 159.62295532226562, -2.4211654663085938, -33.80352783203125, 150.56170654296875, -185.2537841796875, 204.24905395507812, -24.434860229492188, 28.9754638671875, 52.13092041015625, 205.99166870117188, 7.458282470703125, 147.2000732421875, 130.59744262695312, 70.45361328125, 170.93777465820312, 161.1668701171875, 159.2669677734375, 136.09349060058594, 206.76678466796875, 176.95343017578125, 12.187057495117188, 139.60739135742188, 184.66024780273438, 178.12374877929688, -7.772789001464844, 99.61882019042969, 17.505962371826172, 130.60968017578125, 141.67562866210938, 127.34619140625, 206.79837036132812, 147.78665161132812, 3.4120559692382812, 39.3707275390625, 100.72637939453125, -152.10543823242188, -104.07841491699219, -30.588348388671875, -126.26071166992188, 164.331298828125, 100.32011413574219, 163.34710693359375, 31.54401397705078, 3.86712646484375, -3.365264892578125, 1.49505615234375, 14.791641235351562, 80.53559875488281, 207.63082885742188, 149.1356201171875, 71.30147552490234, -111.50404357910156, 10.26800537109375, 24.59209632873535, 2.8821563720703125, 157.11508178710938, -7.4031982421875, 119.97697448730469, 177.0847930908203, -15.302467346191406, -14.86279296875, 4.8210906982421875, 154.11376953125, 44.854888916015625, -42.1097412109375, 3.6134033203125, 129.868408203125, 3.9020824432373047, 26.21617889404297, -15.870697021484375, 11.073394775390625, 39.33861541748047, 88.04388427734375, -5.442558288574219, 199.73495483398438, -148.30868530273438, 54.798248291015625, 18.085411071777344, 154.42593383789062, 22.790908813476562, 0.0, 167.4188232421875, 174.1102294921875, 142.95167541503906, -114.23321533203125, 76.24906921386719, 158.9342803955078, -134.431884765625, -129.54922485351562, 226.43423461914062, 6.432819366455078, -2.6704483032226562, 6.802946090698242, -100.6033935546875, 11.61358642578125, 120.64643096923828, 25.426177978515625, 119.01931762695312, -65.25283813476562, -14.944000244140625, 106.9815673828125, 41.07044982910156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000335.npy"}
{"epoch": 0.7015706806282722, "step": 336, "batch_size": 128, "mean": 49.32929229736328, "std": 86.2168197631836, "min": -163.08544921875, "p10": -40.233837890625, "median": 32.02254104614258, "p90": 169.48751831054688, "max": 266.5245361328125, "pos_frac": 0.6953125, "sample": [69.02984619140625, 4.8076171875, 85.23269653320312, -33.275665283203125, 139.2609405517578, 54.134429931640625, -20.49462890625, 137.85882568359375, 137.9920654296875, 167.02163696289062, 9.54364013671875, 32.188568115234375, 131.88165283203125, 78.07463836669922, 3.2899436950683594, 13.655120849609375, -2.19189453125, 191.2853240966797, 9.119117736816406, 169.96795654296875, 185.90309143066406, 112.40901184082031, 124.3700942993164, 149.96279907226562, -64.93890380859375, -16.451995849609375, 9.055908203125, 17.797889709472656, 20.813491821289062, 54.917724609375, 146.19461059570312, 119.49202728271484, -3.8230438232421875, 152.90145874023438, 203.61697387695312, 171.58384704589844, 67.11985778808594, 225.36151123046875, 18.99829864501953, 9.648529052734375, -50.20173645019531, 149.3756561279297, -47.314971923828125, 158.01910400390625, 43.89727783203125, 0.09287261962890625, 0.3688926696777344, 139.39077758789062, -37.41926574707031, 21.006885528564453, 31.85651397705078, 1.447174072265625, 19.548919677734375, 114.18460083007812, 39.05949401855469, -56.432586669921875, -14.15966796875, -13.888107299804688, 127.55562591552734, -4.7247161865234375, 103.00306701660156, 34.678680419921875, -80.84991455078125, 81.3428955078125, -40.6552734375, 108.02085876464844, 50.84002685546875, -26.31121826171875, 125.34527587890625, 146.48951721191406, 1.94927978515625, -9.286651611328125, 158.0000457763672, 32.46223449707031, 6.348331451416016, -40.05322265625, -9.61224365234375, 39.868492126464844, 22.868179321289062, -20.208755493164062, 3.261472702026367, 42.479217529296875, 45.69728088378906, -154.02389526367188, 140.7506866455078, 44.97160339355469, 130.72195434570312, 67.45257568359375, 23.16010284423828, 169.2816162109375, -22.314483642578125, -137.39059448242188, 186.88304138183594, 178.00225830078125, 71.8568115234375, 58.93426513671875, 191.86395263671875, -6.149955749511719, -10.760162353515625, 118.40771484375, -5.547231674194336, 174.86158752441406, -14.2176513671875, -57.9332275390625, -153.84136962890625, 266.5245361328125, 17.706787109375, 161.1056671142578, 206.89614868164062, -163.08544921875, 5.947086334228516, -15.415596008300781, -2.30401611328125, 168.74424743652344, 33.35179138183594, 10.783943176269531, 7.34033203125, -59.7305908203125, -111.2572021484375, 0.0, -20.455963134765625, 38.189605712890625, 89.45263671875, 173.4718017578125, -38.457855224609375, -27.833709716796875, -14.297256469726562, 111.84819030761719], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000336.npy"}
{"epoch": 0.7036649214659686, "step": 337, "batch_size": 128, "mean": 55.332847595214844, "std": 97.5928726196289, "min": -144.4512176513672, "p10": -89.48732299804686, "median": 45.32017517089844, "p90": 178.5402618408203, "max": 216.62188720703125, "pos_frac": 0.703125, "sample": [36.691253662109375, -48.28460693359375, 6.2156982421875, 34.751277923583984, 13.972755432128906, -3.276123046875, -18.8360595703125, 194.793212890625, 167.62213134765625, 209.8763427734375, -1.0906219482421875, 125.30364990234375, 21.334442138671875, 56.13568115234375, 187.47314453125, 177.37176513671875, -11.5799560546875, 73.82858276367188, -17.7083740234375, 14.692276000976562, -116.18661499023438, -128.5228271484375, -112.84103393554688, -39.554107666015625, 23.58264923095703, 4.2149658203125, 35.82977294921875, 172.7560272216797, -3.1675376892089844, 161.433349609375, -7.566925048828125, 191.3563232421875, 96.12945556640625, 134.771484375, 175.88720703125, 144.4410400390625, 53.332672119140625, 8.881095886230469, 112.65625, 108.27880859375, -12.483207702636719, -77.41206359863281, -83.26866149902344, 156.89837646484375, 61.26141357421875, 160.9675750732422, -11.580322265625, -137.1151580810547, -14.24151611328125, 34.500823974609375, 31.711181640625, 146.67156982421875, 0.4012603759765625, 120.12396240234375, 64.0296630859375, -124.82215881347656, 123.06687927246094, 76.46784973144531, 34.642303466796875, 134.414794921875, 1.0709075927734375, 13.310592651367188, 11.830795288085938, 186.83758544921875, 177.73513793945312, -44.16832733154297, 153.24371337890625, 153.29200744628906, -144.4512176513672, 95.7835693359375, 93.30908203125, 188.24575805664062, 42.233917236328125, 24.977203369140625, 137.0052947998047, 5.027618408203125, -6.406951904296875, -27.413368225097656, 177.31912231445312, 151.30807495117188, 14.528282165527344, 73.371337890625, 81.50305938720703, 170.73736572265625, 143.61956787109375, 216.62188720703125, -17.00847625732422, 135.98565673828125, -62.759979248046875, -118.0923080444336, 54.82850646972656, 165.39845275878906, 111.24757385253906, 205.3983154296875, -128.59060668945312, 201.36767578125, -87.77005004882812, -8.206512451171875, 130.9912109375, 36.132415771484375, 79.4608383178711, -15.16973876953125, -93.49429321289062, 180.60293579101562, -143.88104248046875, -108.36554718017578, 189.64129638671875, 91.7344970703125, 14.618104934692383, 3.0389575958251953, -37.98028564453125, 107.54630279541016, 164.21127319335938, 194.01608276367188, -101.67752075195312, 180.41888427734375, -0.3040771484375, 4.189666748046875, -39.298583984375, 147.7188720703125, 150.0780029296875, 128.31072998046875, -134.86471557617188, 48.40643310546875, 15.81475830078125, 170.23092651367188, 163.35459899902344, 95.6201171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000337.npy"}
{"epoch": 0.7057591623036649, "step": 338, "batch_size": 128, "mean": 67.46160888671875, "std": 91.50321197509766, "min": -218.8936309814453, "p10": -30.38696136474609, "median": 44.852760314941406, "p90": 181.64677429199216, "max": 231.204345703125, "pos_frac": 0.7578125, "sample": [190.76522827148438, -41.216819763183594, 7.977294921875, 154.3514862060547, 174.42337036132812, 190.71661376953125, 138.4176025390625, 144.83880615234375, -26.969253540039062, -39.423736572265625, 5.126750946044922, 157.1062774658203, 38.10491943359375, 18.35106658935547, 166.563232421875, 21.9095458984375, 146.62078857421875, 132.754150390625, -5.1194305419921875, -62.03509521484375, 176.46136474609375, 20.127334594726562, -1.6471672058105469, 199.0372314453125, 5.75677490234375, 59.160400390625, 144.46983337402344, 28.75971221923828, -9.974090576171875, 2.0238208770751953, 44.060546875, 10.63970947265625, -4.94683837890625, 29.050079345703125, 154.27687072753906, 137.9058837890625, 132.937255859375, 172.3922119140625, 27.169876098632812, 40.07562255859375, 119.97222900390625, 35.95947265625, 127.505859375, 166.50616455078125, 185.4931640625, 41.385467529296875, -34.65411376953125, 159.865966796875, 52.31591796875, 65.44351196289062, 117.118408203125, -22.86322021484375, 176.41604614257812, -147.82606506347656, 46.71795654296875, -12.696487426757812, -12.341766357421875, 197.93072509765625, -124.09600830078125, 179.17333984375, 147.6551513671875, 231.204345703125, 167.69786071777344, 47.185302734375, 159.05740356445312, 64.05841064453125, -18.295238494873047, -11.9993896484375, -54.681396484375, -5.025169372558594, 6.735355377197266, 141.17750549316406, 200.13052368164062, 22.48919677734375, -28.558181762695312, 3.337339401245117, 64.95576477050781, 172.45941162109375, 179.082763671875, -2.0146656036376953, 29.3336181640625, 0.9620018005371094, 147.3660430908203, -3.9356689453125, 139.66317749023438, 25.75909423828125, -83.67420196533203, 11.151107788085938, 126.94158935546875, 151.97894287109375, 134.57308959960938, 195.3251953125, 25.6895751953125, -8.217132568359375, -140.36048889160156, 173.98037719726562, 23.425567626953125, 204.77032470703125, -142.30679321289062, 127.13475799560547, 45.64497375488281, -25.705413818359375, 6.021751403808594, 149.137939453125, -2.5232696533203125, 179.99832153320312, 101.36076354980469, 115.69117736816406, -40.963584899902344, 197.143310546875, 48.617431640625, 114.01992797851562, 0.3989849090576172, 41.32647705078125, -9.260986328125, 192.31524658203125, 42.62030029296875, 34.72114562988281, 139.07904052734375, 118.60797119140625, -218.8936309814453, 1.349151611328125, 27.2156982421875, 39.75685119628906, 202.546142578125, 198.8195343017578, -37.383148193359375, 148.88539123535156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000338.npy"}
{"epoch": 0.7078534031413612, "step": 339, "batch_size": 128, "mean": 65.71089935302734, "std": 96.0064468383789, "min": -195.5208740234375, "p10": -50.98681640624999, "median": 47.32966613769531, "p90": 175.97797546386718, "max": 266.51849365234375, "pos_frac": 0.765625, "sample": [17.791915893554688, -49.3857421875, 69.6803207397461, 45.561614990234375, 149.85458374023438, -37.36250305175781, 28.382415771484375, -25.229156494140625, 117.86590576171875, -1.96807861328125, 201.24295043945312, 20.0228271484375, 6.57373046875, 9.32781982421875, 153.7274169921875, 4.731842041015625, 49.09771728515625, 41.901123046875, 93.27300262451172, -63.18115234375, 157.0719451904297, 14.024017333984375, 65.35812377929688, 113.15811157226562, 0.0, 266.51849365234375, -103.97744750976562, 147.17428588867188, 0.0, 8.043468475341797, 108.83006286621094, -47.85151672363281, 207.36492919921875, 154.20632934570312, 8.04248046875, -162.1865692138672, 177.7335205078125, 176.41293334960938, 62.230316162109375, 75.14102172851562, 32.478668212890625, 158.3118438720703, -33.77154541015625, 155.84002685546875, 114.12322998046875, 26.873687744140625, 151.63360595703125, -0.00815582275390625, 11.670616149902344, 124.38491821289062, 21.397186279296875, 6.65496826171875, 235.80999755859375, 125.18803405761719, 152.15426635742188, 152.342041015625, 38.237396240234375, 8.981132507324219, 154.9920654296875, 191.16241455078125, 176.08331298828125, -3.3393936157226562, -131.0989990234375, -93.4705810546875, -1.6926898956298828, 189.35604858398438, 2.394317626953125, 132.48739624023438, 134.7696533203125, 33.04345703125, 37.455169677734375, 20.6729736328125, -148.70858764648438, 40.284423828125, 154.39111328125, 28.6806640625, 158.98291015625, 175.93283081054688, -74.9052734375, 123.83770751953125, 3.0295944213867188, 147.51806640625, 172.23094177246094, -11.4979248046875, 164.87765502929688, 237.07147216796875, 86.72369384765625, 0.0, 96.49334716796875, 159.24755859375, 161.84490966796875, 144.156494140625, 160.16552734375, 165.411865234375, 179.197998046875, -10.145782470703125, 142.36175537109375, -7.519989013671875, -70.22894287109375, 175.23727416992188, 80.9931640625, -111.067626953125, 31.2965087890625, -0.534210205078125, 12.103118896484375, 149.42494201660156, 158.154541015625, 7.626434326171875, -54.72265625, 4.225425720214844, 191.03375244140625, 142.57666015625, -195.5208740234375, 165.62005615234375, 25.3896484375, 11.135688781738281, 27.099212646484375, 86.72909545898438, 174.78448486328125, -68.98284912109375, -180.3737335205078, 205.53961181640625, 45.353118896484375, 55.730987548828125, 164.82040405273438, 7.1246490478515625, -23.63109588623047, 160.074951171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000339.npy"}
{"epoch": 0.7099476439790576, "step": 340, "batch_size": 128, "mean": 54.64013671875, "std": 92.54161071777344, "min": -155.73040771484375, "p10": -55.425158691406246, "median": 34.9397087097168, "p90": 169.7015121459961, "max": 237.63336181640625, "pos_frac": 0.7265625, "sample": [215.24403381347656, 98.95745086669922, -19.573577880859375, 156.38429260253906, 1.752105712890625, 166.09519958496094, -67.94219970703125, -19.771804809570312, -56.9088134765625, 162.4164276123047, -65.30642700195312, 143.05506896972656, -76.06427001953125, 7.881935119628906, 0.0, 172.8646240234375, 66.83853149414062, -53.170623779296875, 128.8954620361328, 167.689697265625, 158.3812255859375, 12.64898681640625, -54.789306640625, 181.89431762695312, -134.43490600585938, 50.435455322265625, -26.479156494140625, 151.54815673828125, 10.995872497558594, 164.48899841308594, 206.5003662109375, 4.562244415283203, 157.65142822265625, 169.27902221679688, 28.33563232421875, -14.268234252929688, 0.37940216064453125, -118.30130004882812, 25.444839477539062, 15.938644409179688, 161.76629638671875, 148.00872802734375, -125.57791137695312, 48.174591064453125, 81.46293640136719, 172.9748992919922, -47.616607666015625, 147.44041442871094, 26.629638671875, -13.933731079101562, 118.75732421875, 201.39898681640625, -25.994447708129883, -102.8115234375, 9.666921615600586, -13.95306396484375, 149.77491760253906, 119.23236083984375, 34.104461669921875, -38.240631103515625, 5.1523590087890625, -155.73040771484375, 177.23902893066406, 0.0, 206.03030395507812, 30.939178466796875, 66.31930541992188, 18.65943145751953, 167.12557983398438, 37.4521484375, 18.059814453125, -7.858476638793945, 150.7054901123047, -118.82833862304688, 94.62225341796875, 41.107147216796875, 35.5465087890625, -134.72979736328125, 170.25555419921875, 10.527801513671875, 168.33209228515625, 231.0382080078125, -88.4873046875, 31.082794189453125, -18.41937255859375, 126.77880859375, 69.23260498046875, 44.67102813720703, 159.7398681640625, 22.625503540039062, 38.86529541015625, -16.394250869750977, 63.843048095703125, -16.501007080078125, 152.86553955078125, 123.722900390625, -16.402236938476562, 7.554290771484375, 55.41424560546875, 42.245758056640625, 237.63336181640625, 8.841291427612305, 159.553955078125, -23.754409790039062, 175.53515625, -35.850067138671875, 20.04869842529297, -155.55615234375, 153.12796020507812, 121.712646484375, 147.63458251953125, 34.332908630371094, 14.278858184814453, 120.86854553222656, 170.30108642578125, 169.4640655517578, 17.691802978515625, 125.55066680908203, 12.844085693359375, 46.157928466796875, 28.527389526367188, 75.20260620117188, -32.844451904296875, 19.053165435791016, 62.368446350097656, 18.014663696289062, 147.61172485351562, -9.628726959228516], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000340.npy"}
{"epoch": 0.7120418848167539, "step": 341, "batch_size": 128, "mean": 47.406585693359375, "std": 94.73841094970703, "min": -171.53225708007812, "p10": -66.14277648925781, "median": 26.268966674804688, "p90": 170.69450988769532, "max": 228.724609375, "pos_frac": 0.65625, "sample": [7.387359619140625, -65.61813354492188, 129.135986328125, 166.19244384765625, 192.38601684570312, -46.934295654296875, 24.164215087890625, 20.46624755859375, 54.35798645019531, 1.6405029296875, 154.85760498046875, 40.86106872558594, 158.70785522460938, 11.471240997314453, 203.43673706054688, 190.19882202148438, -120.78854370117188, -1.12530517578125, -8.529876708984375, 43.47941970825195, 156.2784423828125, 15.698150634765625, 142.78419494628906, -50.84800720214844, 198.40606689453125, 59.56919860839844, 12.654991149902344, 44.45896911621094, 9.724716186523438, -8.85052490234375, -2.735931396484375, 35.92333984375, -17.169219970703125, -7.1812744140625, -99.2847900390625, 54.089569091796875, -64.3714599609375, -67.366943359375, -0.5446014404296875, -8.41192626953125, 4.14129638671875, 114.92146301269531, 186.57113647460938, 28.548614501953125, 156.2618865966797, 4.573974609375, 162.00704956054688, 21.111923217773438, -4.774261474609375, 145.97747802734375, 116.69961547851562, 133.2024688720703, 198.29248046875, -24.9842529296875, 170.03799438476562, 141.45498657226562, 104.95831298828125, 137.98684692382812, 16.851470947265625, -8.48980712890625, 113.78387451171875, -98.93017578125, 228.724609375, -10.350791931152344, 126.46499633789062, -135.43751525878906, -88.8101806640625, 160.737060546875, 179.7924041748047, 4.684173583984375, -8.41680908203125, 173.93075561523438, 119.287109375, -21.598167419433594, 148.47665405273438, 157.387939453125, 167.31365966796875, 57.55169677734375, 17.692245483398438, -148.1009979248047, 74.7462158203125, 6.367767333984375, -12.997085571289062, -23.15887451171875, 219.25347900390625, 32.00193786621094, -23.819183349609375, 41.53857421875, 65.20001220703125, 168.73190307617188, -105.76214599609375, 82.74406433105469, 203.14450073242188, 154.9175262451172, -135.38800048828125, -8.33758544921875, -12.774711608886719, -0.4968891143798828, -11.548797607421875, 28.37371826171875, 30.845550537109375, 163.8524169921875, 17.2852783203125, -17.20098876953125, -145.1134033203125, 43.52449035644531, 38.790985107421875, -50.71903991699219, -171.53225708007812, -30.76898193359375, 18.979339599609375, -73.56353759765625, -12.46478271484375, 172.22637939453125, -167.03916931152344, 82.40342712402344, 143.88809204101562, 168.8538818359375, 144.97613525390625, 49.24589538574219, 157.77362060546875, 9.102951049804688, 38.62512969970703, -16.037841796875, 203.882080078125, 12.2464599609375, -7.716522216796875, 12.85750961303711], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000341.npy"}
{"epoch": 0.7141361256544503, "step": 342, "batch_size": 128, "mean": 57.66448974609375, "std": 87.4860610961914, "min": -186.77041625976562, "p10": -26.9428497314453, "median": 38.82736587524414, "p90": 171.8203125, "max": 232.419189453125, "pos_frac": 0.75, "sample": [-6.325469970703125, 23.4322509765625, 68.86099243164062, 22.54327392578125, 155.00579833984375, 221.90713500976562, 20.256092071533203, 113.99603271484375, 8.967933654785156, 41.223876953125, -13.510757446289062, 106.06982421875, 3.4844207763671875, 118.89767456054688, 126.97712707519531, 146.78781127929688, -6.966270446777344, 67.9864501953125, -140.93450927734375, 2.4991607666015625, 40.736053466796875, 79.55270385742188, 14.71551513671875, 34.02924346923828, -52.080535888671875, 14.330070495605469, 20.339920043945312, -44.52191162109375, 79.55923461914062, 144.2820281982422, 40.93083190917969, 125.16854858398438, 140.32582092285156, 67.16387939453125, -8.338623046875, 14.92071533203125, 9.998626708984375, -12.290180206298828, -55.178375244140625, 33.475830078125, 95.03543090820312, 190.13375854492188, 18.65673828125, -124.9345703125, -18.521865844726562, -23.568572998046875, -13.683197021484375, 60.766387939453125, 54.249267578125, 0.938507080078125, 136.3594512939453, -4.7675628662109375, -35.22711181640625, 174.4688720703125, 0.8826751708984375, 209.28265380859375, 168.20761108398438, -1.242431640625, 111.99856567382812, 140.04052734375, 169.50189208984375, 174.76828002929688, 126.87582397460938, 143.75042724609375, 113.85859680175781, -139.64791870117188, -20.301437377929688, 0.0, -86.66876220703125, 46.73443603515625, 190.1295166015625, 3.7586822509765625, 197.2974853515625, 4.363223075866699, 58.247535705566406, 151.80865478515625, 179.71502685546875, 36.918678283691406, 195.0520477294922, 4.156970977783203, -16.568115234375, 160.08428955078125, 41.3978271484375, 127.3583984375, -68.95947265625, 11.175968170166016, -186.77041625976562, -18.2991943359375, 119.20065307617188, 122.569580078125, 171.25689697265625, 126.8109130859375, 30.038658142089844, 29.491424560546875, 50.611724853515625, 58.24049377441406, 141.52548217773438, 21.126007080078125, 0.572296142578125, -130.62261962890625, 20.453689575195312, 152.71124267578125, -13.362152099609375, 232.419189453125, 16.734405517578125, 145.43263244628906, -23.138641357421875, 10.552459716796875, 88.6641845703125, -116.1286849975586, 173.13494873046875, 146.14089965820312, 1.799896240234375, 144.8006591796875, 153.96038818359375, 166.9136962890625, 18.0430908203125, 22.513458251953125, 147.77432250976562, 159.9229736328125, -13.911155700683594, -13.987548828125, -34.816162109375, 208.6630859375, 138.0531005859375, -4.691656112670898, 195.568115234375, 8.9207763671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000342.npy"}
{"epoch": 0.7162303664921466, "step": 343, "batch_size": 128, "mean": 49.57063293457031, "std": 91.92634582519531, "min": -227.90936279296875, "p10": -40.783016967773435, "median": 26.562938690185547, "p90": 167.6034408569336, "max": 225.3353271484375, "pos_frac": 0.7109375, "sample": [-33.65924072265625, 44.585113525390625, 154.54840087890625, 7.1309356689453125, 42.393341064453125, 91.37692260742188, 144.67852783203125, 20.650482177734375, -105.81607055664062, 27.291748046875, 163.466064453125, -43.911529541015625, -7.00384521484375, -40.255096435546875, 120.40087890625, 190.9014892578125, 175.08126831054688, 47.869873046875, -10.399429321289062, -7.115091323852539, 225.3353271484375, 20.721694946289062, 170.87844848632812, 164.8463134765625, 126.0960464477539, 17.932998657226562, 3.3037109375, 150.43157958984375, 29.643951416015625, 145.12948608398438, 84.586181640625, -110.5339126586914, -4.487701416015625, -10.436431884765625, -134.61911010742188, 110.9044189453125, 16.32794189453125, 155.78707885742188, -134.34249877929688, 21.79083251953125, -46.80467224121094, 193.23635864257812, 41.91046142578125, 41.43780517578125, 0.0, -42.01483154296875, 19.1298828125, 108.00733947753906, 35.53319549560547, 220.21240234375, -136.5355224609375, 44.00609588623047, 68.3740463256836, 10.706409454345703, 159.84375, -37.061431884765625, 91.35986328125, 72.06512451171875, 2.2619857788085938, 135.99374389648438, 140.9034423828125, 10.994277954101562, 0.8892250061035156, 2.8392181396484375, 120.96409606933594, 153.93582153320312, -19.448822021484375, 161.17019653320312, 20.65179443359375, 25.766357421875, 111.43952941894531, -19.753143310546875, 131.7884521484375, -25.438690185546875, -0.76556396484375, 37.898643493652344, 16.231597900390625, 223.7833251953125, 138.90174865722656, 167.6331024169922, 40.529136657714844, 136.80911254882812, -93.08245849609375, 13.946319580078125, 149.571533203125, -29.318710327148438, 3.193878173828125, 39.394134521484375, 13.045928955078125, -30.718994140625, 1.8968219757080078, 157.07769775390625, 156.82339477539062, 214.84332275390625, 4.5042724609375, -12.11956787109375, 138.6371307373047, 106.6641845703125, 153.35958862304688, 25.834129333496094, 176.63348388671875, 102.2744140625, -12.039154052734375, 67.30950927734375, -227.90936279296875, 167.59072875976562, -19.0986328125, -6.436283111572266, -24.89532470703125, -25.15679931640625, 9.659732818603516, 143.27713012695312, 171.46548461914062, 2.13287353515625, -147.7279052734375, 14.409042358398438, -7.6613616943359375, -133.6287841796875, 113.22030639648438, 206.02310180664062, 106.42904663085938, -0.5284996032714844, 32.56170654296875, 6.716850280761719, 5.9710693359375, 178.16323852539062, -16.5242919921875, -141.63836669921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000343.npy"}
{"epoch": 0.7183246073298429, "step": 344, "batch_size": 128, "mean": 61.90275192260742, "std": 96.84027862548828, "min": -192.641845703125, "p10": -43.6074607849121, "median": 38.635440826416016, "p90": 187.0983673095703, "max": 235.69921875, "pos_frac": 0.7265625, "sample": [18.167022705078125, -16.995208740234375, 139.4234161376953, 89.68731689453125, -21.218902587890625, 154.09393310546875, 157.18418884277344, 47.602142333984375, 114.68809509277344, -20.048736572265625, 139.76864624023438, 54.11476135253906, 143.5013427734375, -37.82639694213867, 177.31884765625, 198.96133422851562, 17.339996337890625, 133.69869995117188, -171.92913818359375, 198.74459838867188, -73.25473022460938, 0.1219635009765625, -0.4539794921875, -34.24407958984375, 198.29681396484375, 128.55072021484375, -15.15188980102539, 185.48480224609375, 129.9225616455078, 11.97271728515625, 186.68096923828125, -40.4927978515625, -6.697235107421875, 147.788330078125, -192.641845703125, 54.95774841308594, 158.6884765625, 142.8903045654297, 12.550018310546875, -8.520355224609375, -164.16854858398438, 25.863006591796875, 36.110511779785156, 12.732765197753906, -6.90057373046875, 21.384841918945312, -83.64129638671875, 116.55339050292969, 197.42941284179688, 99.24160766601562, -31.21337890625, -133.46731567382812, 0.229461669921875, -50.87500762939453, 120.70074462890625, 116.51075744628906, 193.72938537597656, -15.4486083984375, 202.72921752929688, 183.06130981445312, 161.87745666503906, 161.02511596679688, 224.58059692382812, 212.8909912109375, 170.79441833496094, -18.063095092773438, -17.238922119140625, 219.68310546875, -16.07904052734375, -30.123809814453125, -52.237152099609375, 15.7591552734375, 20.9710693359375, 216.46047973632812, 1.57000732421875, 25.147857666015625, 19.96771240234375, 25.284912109375, 32.326171875, 114.10360717773438, 33.30455780029297, 21.726409912109375, 127.97381591796875, 151.193115234375, -7.317138671875, 82.3865966796875, 1.4686279296875, 161.4820556640625, 24.258773803710938, 137.87338256835938, 203.15643310546875, 117.8191909790039, 17.816818237304688, -124.62234497070312, -21.494422912597656, 172.75534057617188, 79.11618041992188, -1.2509307861328125, 90.86460876464844, -53.868743896484375, -8.269060134887695, 36.0294189453125, 16.600250244140625, 235.69921875, -14.4537353515625, -92.88720703125, 162.9619140625, 41.160369873046875, 20.26799774169922, 0.09625244140625, 3.6629638671875, 134.07664489746094, 8.523178100585938, 180.8536376953125, 115.59588623046875, 61.360015869140625, 160.704345703125, 85.22785949707031, 60.87139892578125, 188.07229614257812, 117.72793579101562, 33.20957946777344, 69.76248168945312, 165.52822875976562, 158.6002197265625, -85.4386215209961, 178.27386474609375, -156.89456176757812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000344.npy"}
{"epoch": 0.7204188481675393, "step": 345, "batch_size": 128, "mean": 47.24263000488281, "std": 83.83718872070312, "min": -207.00823974609375, "p10": -26.13374633789062, "median": 22.80865478515625, "p90": 168.61608123779297, "max": 206.46405029296875, "pos_frac": 0.7421875, "sample": [56.2734375, 83.3110122680664, 106.5687255859375, -135.74082946777344, 189.4459228515625, 103.15345764160156, 8.729248046875, -41.8973388671875, 15.62774658203125, -0.08586692810058594, 30.019378662109375, 11.757423400878906, -0.370361328125, -8.072166442871094, 4.6316986083984375, -7.278564453125, 179.7025146484375, 12.63507080078125, 11.502349853515625, -1.754730224609375, 14.073089599609375, -23.84259033203125, 143.2109375, 168.24851989746094, 3.4832935333251953, 37.62506103515625, 139.22976684570312, 33.248809814453125, -29.58038330078125, 0.7848091125488281, -11.325302124023438, 47.844757080078125, -15.751781463623047, 49.22607421875, 0.680450439453125, 89.74798583984375, 60.0736083984375, 1.447113037109375, 18.31463623046875, 17.150360107421875, 36.298187255859375, -8.110511779785156, -34.51594161987305, 151.32608032226562, 11.8365478515625, 162.3769073486328, 12.19317626953125, 1.5194206237792969, 48.39674377441406, 204.76910400390625, 142.45977783203125, 147.72406005859375, 132.62518310546875, 179.12416076660156, 181.28611755371094, 3.0868377685546875, 21.570068359375, -24.6566162109375, 15.563095092773438, -8.035240173339844, 171.8411865234375, 98.87371826171875, 0.4610748291015625, 31.45391845703125, 23.262298583984375, -80.38043212890625, -207.00823974609375, 15.98604965209961, -132.3929443359375, -18.918548583984375, 176.5561065673828, 184.17758178710938, 159.1333465576172, 23.309860229492188, 31.1339111328125, -31.090927124023438, 206.46405029296875, -118.38020324707031, 49.0670166015625, -3.1094741821289062, 167.53964233398438, 169.47372436523438, 163.509033203125, -182.3807373046875, 83.6575927734375, 40.21809768676758, -36.4342041015625, 202.20248413085938, 35.8721923828125, 137.81011962890625, 180.72166442871094, 153.91342163085938, 24.0859375, 4.01727294921875, 125.083740234375, 15.981353759765625, 35.14788055419922, 155.80422973632812, 165.36239624023438, -24.268310546875, 160.0458984375, -68.3428955078125, 142.10189819335938, -110.16134643554688, 30.239471435546875, 61.1591796875, 3.989774703979492, 22.355010986328125, -11.9097900390625, -10.657470703125, -6.7555694580078125, 134.64657592773438, 20.50634765625, 12.938705444335938, 171.9051513671875, 145.46038818359375, 13.2314453125, -1.6049690246582031, 0.0, 46.803062438964844, 1.6706314086914062, 21.582183837890625, 30.86725616455078, -0.06627655029296875, 156.6773223876953, 6.34197998046875, 92.26681518554688, 45.12455749511719], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000345.npy"}
{"epoch": 0.7225130890052356, "step": 346, "batch_size": 128, "mean": 61.313533782958984, "std": 95.20306396484375, "min": -212.79949951171875, "p10": -40.24158325195312, "median": 36.49128723144531, "p90": 184.3653350830078, "max": 204.1510009765625, "pos_frac": 0.7265625, "sample": [-47.159820556640625, -48.496665954589844, 1.1643333435058594, -37.92073059082031, 142.00604248046875, 71.63421630859375, 185.86691284179688, 188.39035034179688, 141.37344360351562, -8.815383911132812, 144.30223083496094, -23.336410522460938, -49.71502685546875, 22.007461547851562, 156.00588989257812, -152.45889282226562, -115.06272888183594, 2.05120849609375, 22.0372314453125, 192.57284545898438, 28.706893920898438, -16.059616088867188, 22.102432250976562, -56.772674560546875, 45.07284164428711, 146.60617065429688, 11.376113891601562, 200.2037353515625, -15.0572509765625, 2.604583740234375, 40.6121826171875, 146.73583984375, -127.15843200683594, 30.3231201171875, -7.96246337890625, 171.14120483398438, 157.37940979003906, 187.65567016601562, 14.088228225708008, 13.884307861328125, 82.228759765625, 141.28585815429688, -212.79949951171875, 25.1097412109375, -5.5902557373046875, 29.697113037109375, -15.871856689453125, 20.161117553710938, 31.335113525390625, 46.83639907836914, 147.18429565429688, 16.38909912109375, 141.10484313964844, -16.394943237304688, 35.970245361328125, 106.47769165039062, 195.95501708984375, 2.5222015380859375, -6.183021545410156, 158.26626586914062, 182.7957763671875, 201.0506591796875, 28.0048828125, 16.083824157714844, 24.29193115234375, 13.682304382324219, 183.99728393554688, -19.90398406982422, 180.29843139648438, 43.221832275390625, 163.25454711914062, 135.017333984375, 152.296630859375, 201.80279541015625, 176.3118896484375, 165.1700439453125, -111.94166564941406, 185.68096923828125, 57.93170166015625, 32.23187255859375, 142.80340576171875, -127.87344360351562, 173.21377563476562, 4.33172607421875, -6.492919921875, -25.305683135986328, -35.61406707763672, 6.7672271728515625, -24.074310302734375, 192.4661865234375, 185.22412109375, 37.0123291015625, -16.57813262939453, 168.69967651367188, -1.6708526611328125, -18.575279235839844, 136.33323669433594, -204.424560546875, -0.06945037841796875, 178.44741821289062, 0.4107990264892578, 9.531570434570312, 139.62042236328125, 140.97801208496094, 203.34815979003906, 119.93871307373047, 142.3348846435547, 153.80712890625, 3.23614501953125, -0.377532958984375, 0.0, 161.4735870361328, -4.4271087646484375, 73.9151611328125, 146.40087890625, 48.897247314453125, 54.47833251953125, 148.9739990234375, 59.287864685058594, -99.40904235839844, 142.6825408935547, 113.46237182617188, -45.65690612792969, 137.62579345703125, 165.88735961914062, 141.69436645507812, 4.353857040405273, 204.1510009765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000346.npy"}
{"epoch": 0.724607329842932, "step": 347, "batch_size": 128, "mean": 52.3372802734375, "std": 90.791259765625, "min": -222.59649658203125, "p10": -26.72191009521483, "median": 30.917701721191406, "p90": 176.0066131591797, "max": 214.24325561523438, "pos_frac": 0.703125, "sample": [-0.0966644287109375, 80.9951171875, 159.59359741210938, 3.13134765625, 76.11431884765625, 129.4603271484375, 6.666633605957031, 167.85830688476562, 17.057212829589844, 166.87998962402344, -222.59649658203125, 109.977294921875, 165.03639221191406, 26.850616455078125, 182.85671997070312, -156.54949951171875, -42.56817626953125, -0.7776641845703125, 149.57630920410156, 183.6286163330078, 130.40074157714844, -65.21624755859375, 47.13970947265625, 12.392524719238281, 25.576568603515625, 40.57501220703125, 161.02249145507812, 16.55474853515625, 9.329421997070312, -3.57049560546875, -0.72723388671875, 119.8779296875, 195.14744567871094, 70.62782287597656, 164.99822998046875, 56.004356384277344, 172.49691772460938, -87.12191772460938, 80.79000854492188, 72.26787567138672, 2.7510757446289062, 180.996337890625, 193.01800537109375, 27.542999267578125, 41.52044677734375, 117.04571533203125, -16.9251708984375, 12.22772216796875, 28.733352661132812, -128.2108154296875, -4.5884246826171875, -12.296951293945312, 79.41371154785156, -5.4888763427734375, -2.00128173828125, 99.60589599609375, -0.476165771484375, 158.76010131835938, 43.55796813964844, -22.639541625976562, 112.89639282226562, 8.769561767578125, -2.1465301513671875, -20.9918270111084, -7.45220947265625, 117.52616882324219, 59.97261047363281, 55.23725891113281, 197.5289306640625, 108.84539031982422, 30.4500732421875, -148.42404174804688, -3.567169189453125, 21.517486572265625, 214.24325561523438, 0.007354736328125, -141.74166870117188, -15.088005065917969, -38.28495788574219, 4.123687744140625, 8.145790100097656, 136.134033203125, 175.3482666015625, -15.644317626953125, 11.33737564086914, -164.4051513671875, 120.41370391845703, 197.94595336914062, 104.97708129882812, 106.31098937988281, -9.276657104492188, 161.673583984375, -62.169097900390625, 162.90341186523438, 83.41568756103516, -15.38677978515625, 31.385330200195312, -1.5966262817382812, 180.48959350585938, 4.233734130859375, -185.35073852539062, 12.371734619140625, 199.18798828125, 160.44473266601562, 0.0, 20.634490966796875, 22.114776611328125, 32.833343505859375, 124.51995849609375, 173.14151000976562, 32.64703369140625, 10.912994384765625, 164.4259033203125, 142.93260192871094, -11.56939697265625, 44.2274169921875, 39.68212890625, -3.7026519775390625, 147.40512084960938, 45.0028076171875, -2.19769287109375, 178.64816284179688, 192.1462860107422, 17.946685791015625, 26.38225555419922, -36.2474365234375, 177.54275512695312, -10.775062561035156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000347.npy"}
{"epoch": 0.7267015706806282, "step": 348, "batch_size": 128, "mean": 46.10581970214844, "std": 87.89688873291016, "min": -213.30426025390625, "p10": -41.83043136596678, "median": 17.893295288085938, "p90": 167.81937255859376, "max": 215.3971405029297, "pos_frac": 0.671875, "sample": [134.77841186523438, 4.7679443359375, 3.882598876953125, 3.072265625, -7.04803466796875, -15.88299560546875, 154.050048828125, -7.878147125244141, 84.12448120117188, 115.0871353149414, 63.272491455078125, 158.9033660888672, 25.37762451171875, 145.53378295898438, 95.51321411132812, 13.0509033203125, 198.0421142578125, 134.14266967773438, -52.45775604248047, 0.0, 12.88800048828125, 52.539093017578125, 127.30750274658203, -2.8001976013183594, -118.19509887695312, 190.35792541503906, -12.519966125488281, 0.877288818359375, -10.958419799804688, 93.02117919921875, 141.63453674316406, 135.22462463378906, -10.269973754882812, -4.567924499511719, 175.5687255859375, -34.61492919921875, 162.01425170898438, -4.439970016479492, 100.04710388183594, 172.41180419921875, 2.6721630096435547, -19.93310546875, -82.60894775390625, 26.918365478515625, -1.2835693359375, 158.2589569091797, 37.470947265625, 99.09246826171875, 29.464263916015625, 50.673316955566406, 53.16107177734375, 98.25820922851562, -81.28680419921875, 1.9812393188476562, -12.95623779296875, 136.08132934570312, 184.15472412109375, 198.92092895507812, 175.32086181640625, -15.155242919921875, -9.1036376953125, -5.211669921875, -59.85333251953125, 26.145172119140625, -1.2694568634033203, 14.958648681640625, 128.11334228515625, -2.2964725494384766, 7.3709869384765625, 40.758056640625, 131.702880859375, -30.92523193359375, -157.0121307373047, -5.627967834472656, -89.54998779296875, 48.2386474609375, -75.71710205078125, 3.5025787353515625, 110.0626220703125, 7.1964111328125, 207.39105224609375, 9.67431640625, 165.34326171875, 215.3971405029297, 1.0435791015625, 184.26251220703125, -54.4456787109375, -28.15277099609375, -32.80645751953125, 20.82794189453125, -15.069646835327148, 167.63595581054688, 151.04205322265625, -5.822731018066406, 211.96578979492188, -133.54205322265625, 7.974845886230469, 4.425121307373047, -112.82524108886719, 6.5561065673828125, 168.24734497070312, 14.659149169921875, 135.73611450195312, -11.894601821899414, 85.56006622314453, 1.251617431640625, 157.57516479492188, 141.54006958007812, -17.382232666015625, 148.12570190429688, 139.5559844970703, -37.27586364746094, 44.10897445678711, -122.65756225585938, 8.958419799804688, 21.842071533203125, 179.57093811035156, 39.25811767578125, 163.27951049804688, 33.9776611328125, 139.2159423828125, 3.431316375732422, -3.9472732543945312, 13.47412109375, 48.66754150390625, 31.60491943359375, -213.30426025390625, 80.946533203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000348.npy"}
{"epoch": 0.7287958115183246, "step": 349, "batch_size": 128, "mean": 62.071434020996094, "std": 92.74579620361328, "min": -227.57382202148438, "p10": -24.017355346679686, "median": 33.4957160949707, "p90": 182.82364349365233, "max": 251.80755615234375, "pos_frac": 0.75, "sample": [154.4365692138672, 181.63148498535156, 77.60546875, -227.57382202148438, -2.41204833984375, -0.5335884094238281, 251.80755615234375, 85.84921264648438, -109.51173400878906, 58.91156005859375, 46.49859619140625, -50.15733337402344, 158.09756469726562, 192.586669921875, 141.74063110351562, 154.09841918945312, 203.76007080078125, 18.265518188476562, 145.88449096679688, 195.2103271484375, 78.09890747070312, 5.6744384765625, -22.50897216796875, -25.3016357421875, -6.0512542724609375, 122.50054931640625, -7.543212890625, -18.400156021118164, 191.119384765625, 188.2516326904297, 37.86647033691406, 144.5194549560547, 70.8597412109375, -47.79510498046875, 38.2728271484375, 132.4085235595703, 203.55165100097656, -10.24578857421875, 19.765731811523438, 52.635498046875, 7.2151641845703125, 3.4339466094970703, 165.57632446289062, -10.32415771484375, 7.1813507080078125, 73.29107666015625, -8.421417236328125, 221.2763671875, 21.91949462890625, 16.867095947265625, -92.87789916992188, -0.05497169494628906, -27.47747802734375, 180.762939453125, 176.63320922851562, 15.580490112304688, 1.9518508911132812, 15.632766723632812, -124.45291137695312, 90.73078918457031, 0.0, 217.73458862304688, 8.6229248046875, 16.3907470703125, 149.54721069335938, 29.20672607421875, 165.847900390625, 175.0780029296875, 169.06936645507812, -33.562347412109375, 174.26788330078125, -17.8175048828125, -124.02285766601562, 222.07208251953125, 3.1859130859375, 53.7518310546875, -5.018768310546875, 134.8817596435547, 3.8302001953125, 31.510658264160156, 24.3648681640625, 140.75851440429688, 185.6053466796875, 138.41461181640625, 155.64453125, 8.918548583984375, 72.55165100097656, 50.137210845947266, 165.95501708984375, 49.1768798828125, 90.32537841796875, -11.6978759765625, 162.65879821777344, 11.005428314208984, -23.466949462890625, 6.8123321533203125, 162.85052490234375, 0.8648529052734375, 29.417510986328125, -141.43460083007812, 35.48077392578125, 28.12328338623047, -0.1972808837890625, -2.6644287109375, -80.443115234375, 29.24542236328125, 29.381607055664062, 5.200847625732422, 21.848480224609375, 122.10772705078125, 1.33111572265625, 156.27688598632812, 27.952606201171875, 143.2141876220703, 148.15113830566406, 192.30438232421875, 165.24505615234375, -22.4993896484375, 100.83535766601562, 208.86639404296875, 12.164215087890625, 111.41043090820312, 156.775146484375, 139.13348388671875, -163.44976806640625, -4.5395660400390625, 23.016983032226562, 149.1146240234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000349.npy"}
{"epoch": 0.7308900523560209, "step": 350, "batch_size": 128, "mean": 56.56647491455078, "std": 97.48281860351562, "min": -171.67935180664062, "p10": -83.61592712402343, "median": 47.924774169921875, "p90": 178.96033935546876, "max": 230.40591430664062, "pos_frac": 0.765625, "sample": [178.79931640625, 106.67308044433594, -0.10846328735351562, -4.7915496826171875, 10.31789779663086, -2.58197021484375, 119.86676025390625, 195.69927978515625, -0.26485633850097656, 142.3935546875, 31.9271240234375, 17.806991577148438, 175.15029907226562, 42.93067932128906, 28.653411865234375, 177.66140747070312, 162.21621704101562, 176.3851318359375, 12.340812683105469, -52.19428634643555, 158.59022521972656, 15.794883728027344, 5.209197998046875, 134.2726593017578, 111.55883026123047, 144.0882110595703, 43.48139953613281, 137.33505249023438, 209.4346923828125, 200.21511840820312, 52.752899169921875, 137.79129028320312, 14.18429183959961, 17.11979103088379, -48.60205078125, 125.90166473388672, 16.13641357421875, 19.245407104492188, 57.69256591796875, 33.38134765625, 52.754180908203125, 127.94810485839844, 133.37484741210938, -159.88272094726562, 174.43670654296875, -15.8507080078125, 5.4134521484375, 150.03408813476562, -171.67935180664062, 6.0588531494140625, 45.60833740234375, 172.34130859375, 222.3896484375, 92.30613708496094, 30.727447509765625, 137.38360595703125, 191.37875366210938, 158.14547729492188, 65.34634399414062, 208.69747924804688, 109.16387939453125, -89.70928955078125, 174.5109100341797, 10.9044189453125, -19.30859375, -126.64332580566406, 43.239288330078125, 143.50563049316406, -141.49664306640625, 9.03276252746582, 46.33538818359375, 118.20172119140625, 3.644256591796875, -130.57186889648438, -14.471099853515625, 69.13906860351562, -9.154136657714844, 60.01538848876953, 17.70233154296875, 118.21394348144531, 196.81298828125, 166.31893920898438, 121.09160614013672, 103.65309143066406, 32.957000732421875, 162.41830444335938, 17.511474609375, 185.70660400390625, -3.5761566162109375, 9.688201904296875, 192.9781494140625, 73.30867004394531, 31.18359375, 67.86279296875, 77.20791625976562, 8.659942626953125, 32.13397216796875, 153.02731323242188, 230.40591430664062, 89.89083862304688, 173.4178466796875, -162.21826171875, 87.40863037109375, 42.692413330078125, -169.25259399414062, -121.62914276123047, -82.12857055664062, -0.636383056640625, -162.57330322265625, 49.51416015625, -160.61697387695312, -87.08642578125, 10.796173095703125, 194.67868041992188, 96.91725158691406, 51.91693115234375, 11.31634521484375, 11.798690795898438, 181.78134155273438, -14.330718994140625, -14.986541748046875, 56.77920913696289, -23.37225341796875, -9.63433837890625, 179.3360595703125, -135.9947509765625, 49.931976318359375, 113.79124450683594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000350.npy"}
{"epoch": 0.7329842931937173, "step": 351, "batch_size": 128, "mean": 36.540748596191406, "std": 97.41598510742188, "min": -227.71328735351562, "p10": -56.85326080322263, "median": 13.264686584472656, "p90": 174.77344360351563, "max": 248.69146728515625, "pos_frac": 0.625, "sample": [0.0, 159.2271728515625, 12.43896484375, 212.83792114257812, -0.724517822265625, 196.86883544921875, 11.278244018554688, -33.464744567871094, 147.97195434570312, -6.43267822265625, -7.1281585693359375, -141.75753784179688, 17.318878173828125, 98.81375122070312, 173.90740966796875, -133.36402893066406, 19.17496109008789, 122.30382537841797, -34.519554138183594, 19.67938232421875, 36.444793701171875, 152.87657165527344, -40.173858642578125, 220.0260009765625, 39.75914001464844, 166.59323120117188, 9.664098739624023, -18.73870849609375, -11.6378173828125, -0.3145599365234375, 188.94448852539062, 194.9390869140625, -1.278228759765625, 14.974552154541016, -5.3727874755859375, -95.28086853027344, -2.9683361053466797, 152.65359497070312, -114.04898834228516, -91.42385864257812, -165.84326171875, 158.58206176757812, 16.386642456054688, 49.8756103515625, -118.47650146484375, 166.81629943847656, 9.7138671875, 13.66497802734375, -24.791473388671875, 15.251373291015625, -36.549591064453125, -28.1038818359375, 18.475555419921875, -71.38102722167969, 23.78057861328125, -0.902587890625, 166.59487915039062, -48.110076904296875, -12.1943359375, 24.495330810546875, 188.5904541015625, 245.52352905273438, 55.86487579345703, 209.5919189453125, 170.29336547851562, -14.225128173828125, 24.562469482421875, 16.539825439453125, 136.041259765625, -24.1915283203125, -5.9312744140625, 135.55694580078125, -26.53948974609375, 12.864395141601562, 176.794189453125, -138.89218139648438, 26.1422119140625, 11.881805419921875, 0.4935302734375, 14.256103515625, 0.0, 0.05572509765625, 65.26197814941406, 60.781158447265625, 154.61143493652344, -0.7167682647705078, 0.9686431884765625, 17.05506134033203, 138.421630859375, 19.748931884765625, 33.125152587890625, 7.39117431640625, 33.9296875, -14.96661376953125, 0.0, 197.513427734375, 187.06814575195312, 8.2303466796875, 19.523094177246094, -13.772369384765625, 12.816909790039062, -227.71328735351562, 135.15780639648438, 170.49517822265625, 3.8359527587890625, 4.6558380126953125, 248.69146728515625, 70.140380859375, 168.3837432861328, -14.154014587402344, -1.7344989776611328, 0.5206146240234375, -15.142333984375, 70.34675598144531, -4.947139739990234, 161.3780517578125, 22.591049194335938, -49.942535400390625, -50.6270751953125, -168.06674194335938, 126.47442626953125, 127.761962890625, -192.5189208984375, 7.3072967529296875, -42.745819091796875, -92.24452209472656, 230.6231689453125, 59.073211669921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000351.npy"}
{"epoch": 0.7350785340314137, "step": 352, "batch_size": 128, "mean": 64.1005630493164, "std": 92.54032135009766, "min": -207.59722900390625, "p10": -43.12384948730468, "median": 41.561309814453125, "p90": 186.9808624267578, "max": 232.34268188476562, "pos_frac": 0.7578125, "sample": [131.53826904296875, 26.1844482421875, 176.9229736328125, -48.10833740234375, 9.910568237304688, 207.2408447265625, 136.79531860351562, 128.13021850585938, 65.80642700195312, 3.6646957397460938, -9.67901611328125, 157.12094116210938, -136.52365112304688, 16.394012451171875, 158.44032287597656, 5.271598815917969, 33.051300048828125, 191.7750244140625, 156.24920654296875, 127.78518676757812, 213.773193359375, -55.088218688964844, -42.1304931640625, 193.86880493164062, 24.880859375, -207.59722900390625, -14.577972412109375, 32.32600402832031, 18.80120849609375, 133.55599975585938, -1.3623027801513672, 69.10263061523438, 189.370849609375, 21.843704223632812, 49.26177978515625, -128.53067016601562, 22.810592651367188, 91.95745849609375, -19.57763671875, 15.83135986328125, 143.1351318359375, -79.94464111328125, -38.89166259765625, 121.81039428710938, -29.72796630859375, 23.77783203125, 0.0, 170.6632080078125, 8.943511962890625, 186.10385131835938, -1.2200927734375, 198.47291564941406, 140.65023803710938, 33.86083984375, 158.76947021484375, 163.97817993164062, 27.22247314453125, 0.0, 72.70941162109375, 91.718994140625, 122.98777770996094, 0.0448150634765625, 18.233375549316406, 109.700927734375, -61.6119384765625, -45.441680908203125, 32.65990447998047, 114.86337280273438, -104.146728515625, 5.53955078125, 123.80552673339844, 209.56588745117188, 74.0189208984375, 63.14727020263672, 6.672578811645508, -40.9873046875, 189.0272216796875, -5.594242095947266, -105.3411865234375, -100.06918334960938, 232.34268188476562, 191.70286560058594, 184.09275817871094, 23.99018096923828, 31.563873291015625, 158.2299041748047, 0.176422119140625, -45.66998291015625, 0.2880096435546875, 33.451690673828125, 101.62127685546875, -10.29547119140625, 58.20147705078125, 19.783451080322266, 32.79191589355469, 49.402984619140625, 128.56732177734375, 146.38278198242188, -8.686294555664062, 196.14332580566406, 59.06101989746094, 132.43685913085938, 154.44224548339844, 195.2352294921875, 172.4713592529297, -35.804534912109375, 153.33856201171875, 171.13287353515625, 136.70440673828125, 8.986900329589844, 21.704559326171875, 150.1436004638672, 182.4227294921875, 1.3071136474609375, -19.24383544921875, -32.39646911621094, 28.876697540283203, 182.5723876953125, 144.65133666992188, 190.97654724121094, -26.289108276367188, -128.32034301757812, 145.00367736816406, 170.39682006835938, 66.03875732421875, 129.74783325195312, 17.87793731689453, 161.71878051757812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000352.npy"}
{"epoch": 0.7371727748691099, "step": 353, "batch_size": 128, "mean": 64.50082397460938, "std": 91.36217498779297, "min": -180.07647705078125, "p10": -20.09549560546875, "median": 50.79878234863281, "p90": 171.04306182861328, "max": 261.1402893066406, "pos_frac": 0.7578125, "sample": [99.25440979003906, 151.20437622070312, 127.41645812988281, -29.238983154296875, 55.268280029296875, 178.8538818359375, -134.669189453125, 168.2413330078125, 171.42221069335938, 13.509124755859375, 93.75408935546875, 27.094528198242188, 6.735870361328125, 62.487457275390625, -7.6091156005859375, 20.63201904296875, -15.55853271484375, -27.312835693359375, 63.567169189453125, -18.569000244140625, 129.37637329101562, 93.27449035644531, -2.3437881469726562, -180.07647705078125, 150.9547576904297, 36.607421875, 131.6046142578125, 118.241943359375, 243.22589111328125, 8.728374481201172, 189.00390625, 150.94949340820312, 170.59957885742188, -17.1712646484375, 153.8211669921875, -163.74337768554688, 193.50704956054688, 82.874755859375, 75.20947265625, 261.1402893066406, 112.38117980957031, -111.99505615234375, 46.32928466796875, 169.28753662109375, 37.83714294433594, -3.220062255859375, 169.488525390625, 247.30096435546875, 216.896240234375, 144.71826171875, 152.7717742919922, 146.21835327148438, -39.832557678222656, 120.73014831542969, 15.489974975585938, 227.90817260742188, 1.7837791442871094, -21.53235626220703, 140.6917724609375, 167.3192138671875, 195.50112915039062, 97.01760864257812, -5.91357421875, -13.41522216796875, 0.3961029052734375, 100.36019897460938, -141.18386840820312, -17.3203125, 21.72998046875, 28.819915771484375, 28.948028564453125, 9.832595825195312, 149.9669647216797, 161.05368041992188, 153.48468017578125, 136.1305694580078, 14.879249572753906, 154.10623168945312, 11.653762817382812, 6.821968078613281, 157.59579467773438, 43.65447235107422, -8.0860595703125, 24.257568359375, 21.38824462890625, 62.203033447265625, 33.914093017578125, 143.51895141601562, 133.61199951171875, 4.672893524169922, -11.710636138916016, -15.49114990234375, 151.63421630859375, 144.5643310546875, 100.5166244506836, -19.508697509765625, -4.046222686767578, 8.502105712890625, 9.292671203613281, -16.086944580078125, 73.8759765625, 11.73486328125, 28.23333740234375, 3.9605064392089844, 170.98150634765625, 65.66307067871094, 63.84130859375, 55.598968505859375, 165.04725646972656, -117.05728149414062, 26.650856018066406, -0.6379318237304688, 26.553348541259766, 66.04473876953125, 6.660219192504883, 0.0, 217.13858032226562, 170.17666625976562, -31.574798583984375, -21.464691162109375, 3.616119384765625, -167.15908813476562, -13.836822509765625, 20.6429443359375, 171.1866912841797, 104.70745849609375, 151.83636474609375, 173.57818603515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000353.npy"}
{"epoch": 0.7392670157068063, "step": 354, "batch_size": 128, "mean": 41.03722381591797, "std": 102.3831787109375, "min": -184.9288330078125, "p10": -88.13221893310545, "median": 15.444574356079102, "p90": 163.92503814697264, "max": 254.85281372070312, "pos_frac": 0.6171875, "sample": [-10.317413330078125, -52.13804626464844, -130.43128967285156, 154.5587921142578, 48.573455810546875, 219.92575073242188, 116.9366455078125, 19.699542999267578, -0.024700164794921875, 5.2388916015625, 47.19757080078125, 46.75927734375, -116.77839660644531, 16.09112548828125, 162.66000366210938, -19.132125854492188, -13.989471435546875, -0.26671600341796875, 11.25213623046875, -2.836090087890625, 131.86643981933594, 166.3600311279297, 93.45745849609375, -154.5242919921875, 32.43665313720703, 15.921958923339844, -10.784107208251953, 13.198715209960938, 154.58168029785156, 215.79403686523438, -121.7630615234375, -5.5302734375, -0.2391796112060547, 62.103179931640625, 191.57070922851562, 31.6317138671875, 14.96718978881836, -0.1500244140625, 159.17742919921875, 155.71340942382812, 5.7081298828125, -58.64337158203125, 131.12400817871094, -17.6700439453125, 156.84349060058594, 150.6829833984375, -82.1439208984375, 8.828163146972656, -179.08200073242188, 2.2759246826171875, -15.76361083984375, 215.73049926757812, 149.48971557617188, 189.36746215820312, 194.97366333007812, 194.46786499023438, 140.32937622070312, 138.68545532226562, 30.47900390625, 149.74990844726562, 134.93325805664062, -27.463546752929688, -12.684593200683594, -149.85476684570312, -15.427604675292969, 161.1251983642578, 10.018756866455078, 254.85281372070312, -1.7216339111328125, -3.4449539184570312, 162.145263671875, -23.768341064453125, 144.49716186523438, 7.405485153198242, 211.89422607421875, -3.9365386962890625, 125.44403839111328, -138.7059783935547, -13.799530029296875, 0.4039764404296875, -4.0037841796875, -171.10696411132812, 180.57369995117188, 40.25604248046875, -21.465347290039062, 155.71044921875, 45.85841369628906, 155.51536560058594, -42.27325439453125, -72.10969543457031, 17.246826171875, 2.8073806762695312, 162.8814697265625, 11.367477416992188, 112.06959533691406, 24.091629028320312, -8.55780029296875, -64.56988525390625, -101.51235961914062, -82.39787292480469, 135.06716918945312, -3.4388427734375, 12.327957153320312, 146.88380432128906, 138.83685302734375, 147.39706420898438, 5.4234619140625, -23.0069580078125, 44.64738464355469, 200.60235595703125, 160.16046142578125, 99.6744384765625, -12.027275085449219, 167.662841796875, -152.51490783691406, 79.5784912109375, -184.9288330078125, 150.48497009277344, -142.54632568359375, -139.84286499023438, 152.03292846679688, 48.27135467529297, -75.42767333984375, 10.167205810546875, -57.83154296875, -45.66236877441406, 40.6522216796875, 33.62568664550781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000354.npy"}
{"epoch": 0.7413612565445026, "step": 355, "batch_size": 128, "mean": 57.318904876708984, "std": 93.24666595458984, "min": -224.40728759765625, "p10": -32.77637100219726, "median": 39.236541748046875, "p90": 184.46039428710938, "max": 225.37689208984375, "pos_frac": 0.7734375, "sample": [-1.7274246215820312, -224.40728759765625, 185.52838134765625, 38.9688720703125, 21.74169921875, 12.079879760742188, 106.37921142578125, 4.671417236328125, 46.337554931640625, 93.8043212890625, 138.1318817138672, 14.199775695800781, -36.677207946777344, 144.30213928222656, 5.15839958190918, 71.95005798339844, 196.71475219726562, 163.14352416992188, -5.602203369140625, -1.46783447265625, -113.40008544921875, 20.68017578125, -2.016937255859375, 195.37164306640625, 11.245635986328125, 181.1187286376953, -15.465187072753906, -5.834716796875, 39.50421142578125, 12.98455810546875, 27.647369384765625, 16.969757080078125, 44.30902099609375, 199.6434326171875, 190.8150634765625, 52.627288818359375, 201.96762084960938, 28.77413558959961, 12.47235107421875, -28.4609375, 71.67501068115234, 166.26467895507812, 51.594970703125, 15.236709594726562, -59.73173522949219, 84.125, -8.836883544921875, 52.82232666015625, 15.697486877441406, -0.7530517578125, 168.8046875, 11.064132690429688, 173.6214599609375, 12.890899658203125, 1.6485595703125, 45.3173828125, 3.874622344970703, 102.14747619628906, 57.86968994140625, -27.988704681396484, 186.15219116210938, 10.060516357421875, -38.73138427734375, -21.45074462890625, 41.10673522949219, -75.4793701171875, 24.857025146484375, 225.37689208984375, 193.14382934570312, 152.10958862304688, 0.4241485595703125, 191.02838134765625, -166.9915771484375, 167.87435913085938, 0.4087333679199219, 117.89453125, 155.50534057617188, 88.51144409179688, 203.48355102539062, 28.056880950927734, 116.9110107421875, 165.7718505859375, 1.802520751953125, 18.184814453125, 53.495849609375, -57.34063720703125, 54.507232666015625, -2.119598388671875, 34.363525390625, 33.818450927734375, 151.91165161132812, -123.93063354492188, -158.38247680664062, 5.709869384765625, -9.12530517578125, 43.77166748046875, 17.91546630859375, 180.27967834472656, -31.104583740234375, 168.49447631835938, 40.066314697265625, -153.03079223632812, 216.099853515625, 130.2891845703125, 191.49026489257812, 170.90625, 171.6941680908203, 18.887054443359375, -133.0944061279297, -14.322021484375, 13.146697998046875, 35.02824401855469, 12.176406860351562, 160.23446655273438, 107.26699829101562, 157.03912353515625, 161.81280517578125, 141.86294555664062, 136.24114990234375, 136.4366912841797, 78.082275390625, -2.8333740234375, 100.19100952148438, 184.002685546875, 84.92620849609375, -143.35296630859375, 176.45689392089844, 29.312164306640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000355.npy"}
{"epoch": 0.743455497382199, "step": 356, "batch_size": 128, "mean": 49.070831298828125, "std": 91.1596908569336, "min": -188.56576538085938, "p10": -58.2373767852783, "median": 33.37937927246094, "p90": 163.0870819091797, "max": 203.36256408691406, "pos_frac": 0.6953125, "sample": [0.18536376953125, 17.049102783203125, 159.5572509765625, -5.166717529296875, 67.667236328125, -146.7154541015625, 17.393890380859375, -83.1241455078125, 15.55279541015625, 3.8195648193359375, -16.21125030517578, 46.9456787109375, 25.184417724609375, 148.45989990234375, -8.560098648071289, 2.769071578979492, -21.048919677734375, -23.008392333984375, 7.76641845703125, 131.40567016601562, -69.05581665039062, 153.86199951171875, 55.32441711425781, 35.112274169921875, -94.212890625, 188.41619873046875, -129.0303192138672, 151.3912353515625, 1.722503662109375, 186.88946533203125, 40.08380126953125, 167.468505859375, 4.2642822265625, 129.89352416992188, 163.1981658935547, 169.2169647216797, 66.32000732421875, 44.05157470703125, 184.7102508544922, 132.8990936279297, -163.18946838378906, 155.56414794921875, -5.3752593994140625, 9.534309387207031, 115.48690795898438, 34.3604736328125, -8.185348510742188, 191.49078369140625, 169.50823974609375, -7.026599884033203, 137.92431640625, 56.21223449707031, 163.0394744873047, 188.41748046875, -0.18741607666015625, 39.88665771484375, -29.484649658203125, 151.39605712890625, 134.68218994140625, 57.188079833984375, 121.36553955078125, -147.42129516601562, 162.46054077148438, 36.352142333984375, 155.8991241455078, 157.43524169921875, 152.229248046875, -11.248031616210938, 2.2788314819335938, -25.654037475585938, -4.1051788330078125, 25.261474609375, 112.02926635742188, 27.58063507080078, -44.30023193359375, 12.138191223144531, 137.99867248535156, -0.238250732421875, 159.16551208496094, -144.3520050048828, 61.245208740234375, -53.60090255737305, 193.60025024414062, 32.398284912109375, 2.028961181640625, 182.08502197265625, 146.08810424804688, 103.32893371582031, 27.837318420410156, -1.4875679016113281, -94.99597930908203, -4.168846130371094, -15.2552490234375, -188.56576538085938, 104.6419677734375, 16.934295654296875, 50.3062744140625, 0.0, 104.25350952148438, 123.86077880859375, 125.81338500976562, 92.25519561767578, -8.145286560058594, -17.530426025390625, 0.4549713134765625, 16.9847412109375, 24.784175872802734, 144.91461181640625, -1.5002899169921875, -91.10792541503906, 159.31619262695312, 203.36256408691406, 151.2689208984375, 82.993408203125, -109.69564056396484, 169.28054809570312, 157.80178833007812, 10.121719360351562, 35.630615234375, 153.11001586914062, -37.04840087890625, 8.85931396484375, -3.3770904541015625, -148.49169921875, 52.75929260253906, -12.994586944580078, 21.451950073242188, 110.6952896118164], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000356.npy"}
{"epoch": 0.7455497382198953, "step": 357, "batch_size": 128, "mean": 55.659385681152344, "std": 94.46409606933594, "min": -193.8710479736328, "p10": -31.50937805175781, "median": 22.87163734436035, "p90": 185.64807739257813, "max": 262.77484130859375, "pos_frac": 0.703125, "sample": [17.8115234375, 164.35003662109375, -16.715978622436523, 0.5787353515625, 171.2080078125, 42.05810546875, 0.0, -23.3975830078125, 114.63734436035156, 6.5281982421875, 166.77915954589844, -159.062255859375, 6.305267333984375, 262.77484130859375, 215.43914794921875, 221.8941650390625, 6.165279388427734, -5.562957763671875, 25.611404418945312, 30.2864990234375, 16.530349731445312, 17.502685546875, 106.24125671386719, 164.7335205078125, 147.532470703125, -179.02407836914062, 237.802734375, 22.7802734375, 27.55093765258789, 188.99752807617188, 24.44830322265625, 13.25335693359375, -20.61846923828125, 7.088539123535156, 221.03607177734375, 92.91423034667969, -18.546051025390625, -13.666908264160156, 7.03961181640625, 137.02490234375, 131.7520751953125, 216.45806884765625, 115.8421630859375, 127.63943481445312, 151.86341857910156, -32.055633544921875, 5.4901123046875, -76.49710083007812, 177.04315185546875, 8.212890625, -35.873443603515625, 34.334571838378906, 52.32157897949219, 1.678070068359375, 129.5419921875, 136.56150817871094, 196.41522216796875, 184.85784912109375, 37.57769775390625, 1.951934814453125, 156.65252685546875, -0.259979248046875, 42.26947021484375, 29.644561767578125, 46.89593505859375, -9.854705810546875, 160.8645477294922, 163.33529663085938, 132.1390838623047, 35.2540283203125, -16.90325927734375, -31.2752685546875, -93.21279907226562, 6.9810791015625, 1.3712234497070312, 161.517333984375, 16.717575073242188, -62.259307861328125, 21.54266357421875, -37.619415283203125, -57.274200439453125, 0.2797088623046875, 6.7614288330078125, 31.416351318359375, 198.91769409179688, 113.38412475585938, 169.67379760742188, -50.892486572265625, -6.6407623291015625, -0.8150711059570312, -14.394889831542969, 16.5042724609375, -2.8600616455078125, -20.79425048828125, 175.9852294921875, 155.03587341308594, -4.8056640625, -124.24951171875, 0.0, 187.491943359375, -139.90573120117188, 0.0, 148.8741455078125, -5.809906005859375, 179.0430908203125, 22.963001251220703, 10.848594665527344, 74.02708435058594, 175.47833251953125, 158.94815063476562, 188.35800170898438, -193.8710479736328, 170.14352416992188, -5.768890380859375, 2.4423065185546875, 81.4532470703125, -3.837005615234375, 120.63407897949219, 25.854463577270508, 212.83343505859375, 200.15707397460938, 21.068180084228516, -9.423049926757812, -26.024368286132812, 162.49468994140625, -1.17352294921875, 2.9161224365234375, 41.72709655761719], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000357.npy"}
{"epoch": 0.7476439790575916, "step": 358, "batch_size": 128, "mean": 64.22499084472656, "std": 95.37772369384766, "min": -197.78421020507812, "p10": -36.51286392211914, "median": 48.36146545410156, "p90": 183.48709411621093, "max": 223.64215087890625, "pos_frac": 0.7578125, "sample": [0.1390380859375, 2.4680557250976562, 99.09490966796875, 10.671661376953125, 31.085281372070312, 112.2157211303711, -33.69549560546875, 128.03445434570312, 196.148681640625, 216.78668212890625, 8.6876220703125, 4.119091033935547, 17.872268676757812, 169.03746032714844, 190.704833984375, 169.57667541503906, 17.274642944335938, 197.1380615234375, 186.34803771972656, 35.4224853515625, -39.12443542480469, 154.12710571289062, -64.13475036621094, -134.2685546875, 51.881103515625, -9.365676879882812, 131.3592071533203, -5.03704833984375, 191.29193115234375, -10.746246337890625, 16.030899047851562, 132.92437744140625, 74.20782470703125, 145.73243713378906, 65.25836181640625, -37.39617919921875, 213.82199096679688, 6.845062255859375, 31.97119140625, 139.71644592285156, -131.13375854492188, 164.03152465820312, 174.96917724609375, 185.85409545898438, 27.048789978027344, 171.75048828125, 137.86199951171875, 165.90293884277344, 169.83724975585938, -2.2257080078125, -27.6934814453125, -10.81979751586914, -8.1138916015625, 26.990814208984375, 160.18431091308594, -13.153472900390625, 178.41555786132812, 179.581787109375, -16.0889892578125, 223.64215087890625, -23.248565673828125, -197.78421020507812, 168.509765625, -34.554405212402344, 7.674140930175781, 166.38858032226562, 105.931396484375, 33.56455612182617, 85.74090576171875, 12.497528076171875, 153.62246704101562, 13.089118957519531, 22.227752685546875, 27.38818359375, 4.974609375, -90.4317626953125, 161.2732696533203, 89.66365051269531, -15.461174011230469, 196.3682861328125, 112.10639953613281, -140.23233032226562, 132.8866729736328, 51.51544189453125, 169.50344848632812, 37.63922119140625, 166.7255859375, 59.2197265625, 79.07438659667969, 183.75149536132812, 20.950531005859375, 207.33364868164062, 137.78311157226562, 116.06382751464844, 103.10470581054688, 75.578125, -31.4017333984375, 5.770965576171875, 7.28839111328125, 142.43348693847656, 163.041748046875, 48.125213623046875, 151.02182006835938, 183.373779296875, 174.68719482421875, 185.00827026367188, 170.18789672851562, -190.05792236328125, -17.23113250732422, -67.38211059570312, -137.82546997070312, 7.261260986328125, 147.49644470214844, 5.265899658203125, 21.62615203857422, -36.134300231933594, 3.5636672973632812, 13.397241592407227, -2.3533477783203125, 33.2418212890625, 0.0, 139.38491821289062, 55.074676513671875, 48.59771728515625, 156.9560546875, -65.92359924316406, 33.423828125, -98.62461853027344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000358.npy"}
{"epoch": 0.749738219895288, "step": 359, "batch_size": 128, "mean": 57.312503814697266, "std": 93.78173828125, "min": -217.2950439453125, "p10": -33.56585159301758, "median": 35.9036865234375, "p90": 178.50263214111328, "max": 244.03265380859375, "pos_frac": 0.7578125, "sample": [0.41805267333984375, -60.44696044921875, 150.58457946777344, 194.94552612304688, 57.360076904296875, -97.67788696289062, 71.1745376586914, 16.952621459960938, 33.240692138671875, 73.603271484375, 50.44636917114258, 185.4500732421875, 65.60975646972656, -3.37115478515625, -21.162078857421875, -48.041778564453125, 125.19892120361328, 58.52989196777344, 24.12744140625, 12.111907958984375, 29.65691375732422, -11.650161743164062, 158.37664794921875, 183.34844970703125, -94.72366333007812, 80.9610366821289, 5.66650390625, 35.42791748046875, 146.58392333984375, -204.8765869140625, 1.30938720703125, 217.98611450195312, 4.832489013671875, 5.4103851318359375, 2.741353988647461, 144.46725463867188, -8.762861251831055, 155.10140991210938, 175.082275390625, 186.64886474609375, -27.679718017578125, -132.46661376953125, 147.8721923828125, 2.8162002563476562, 127.68049621582031, 53.29266357421875, 2.443695068359375, 165.1781005859375, 146.31265258789062, 139.4043731689453, 170.2987060546875, 1.6904106140136719, 231.47918701171875, -25.153030395507812, 159.29122924804688, 0.088623046875, 86.79135131835938, -173.9267578125, 80.34874725341797, 43.7220458984375, 208.8009033203125, 131.47244262695312, -217.2950439453125, 154.97805786132812, 21.616058349609375, 19.84454345703125, -13.24957275390625, 61.22503662109375, 170.37139892578125, 110.7166748046875, -3.029022216796875, 31.4833984375, -33.173179626464844, 36.37945556640625, 94.23859405517578, 162.87161254882812, -6.42840576171875, 141.90628051757812, 12.027359008789062, 154.76968383789062, -5.622833251953125, -67.98854064941406, 5.24822998046875, 75.98635864257812, 19.544052124023438, 20.17226791381836, 20.656646728515625, 153.2823486328125, 110.14291381835938, 7.2691497802734375, 181.3119354248047, 14.576248168945312, 190.177490234375, 145.88461303710938, 19.62298583984375, -29.964935302734375, -130.88351440429688, 147.9998779296875, -41.204933166503906, -0.9074268341064453, 31.388916015625, 195.38314819335938, 25.980010986328125, -159.82643127441406, 17.263198852539062, -22.53814697265625, 183.92913818359375, -32.132568359375, 62.122833251953125, 118.34429931640625, 151.68606567382812, 110.4513168334961, 4.133331298828125, 128.91183471679688, -7.987548828125, 190.7252197265625, 177.29864501953125, 53.82470703125, 32.017791748046875, -0.6567153930664062, -7.811054229736328, 244.03265380859375, 61.98197937011719, 116.14286804199219, 175.90695190429688, 0.7348480224609375, 142.22055053710938, -34.482086181640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000359.npy"}
{"epoch": 0.7518324607329843, "step": 360, "batch_size": 128, "mean": 53.791656494140625, "std": 84.55347442626953, "min": -224.47125244140625, "p10": -22.39675827026367, "median": 30.71076202392578, "p90": 172.51651153564453, "max": 237.29693603515625, "pos_frac": 0.7265625, "sample": [37.015411376953125, 123.47065734863281, -97.40742492675781, 175.01907348632812, -146.2914276123047, -8.584169387817383, -64.30738830566406, 24.3758544921875, 237.29693603515625, -49.35432434082031, 30.28399658203125, 75.21531677246094, -20.375137329101562, 215.27362060546875, 50.466796875, 190.22113037109375, 14.15142822265625, 163.30987548828125, -32.460052490234375, 48.93635559082031, 191.10226440429688, 31.137527465820312, 17.33477783203125, 90.62727355957031, 139.61358642578125, -21.833587646484375, 3.2757949829101562, 22.0225830078125, 53.148406982421875, 182.23828125, 37.24945068359375, 196.82896423339844, 110.75515747070312, 14.24267578125, 51.881591796875, 6.75048828125, -26.948822021484375, 146.8043212890625, -33.3460693359375, 160.1974334716797, 92.138671875, 21.441009521484375, 161.69454956054688, 28.46685791015625, 111.0267333984375, 21.945545196533203, 20.517059326171875, 154.4227752685547, -10.54864501953125, 41.93096923828125, -12.02813720703125, 155.12600708007812, 93.95480346679688, 80.34869384765625, 210.00924682617188, 11.372116088867188, 186.28245544433594, -41.562652587890625, 161.15872192382812, 14.654182434082031, 107.60305786132812, 43.12371826171875, 168.10055541992188, 134.64768981933594, 18.9561767578125, -16.38629150390625, 184.36727905273438, 19.542022705078125, 50.75050354003906, -122.34127807617188, -11.329513549804688, 88.87713623046875, 1.6071224212646484, 184.38670349121094, -20.026092529296875, 19.846508026123047, 116.18104553222656, 182.87509155273438, 156.75845336914062, 149.92108154296875, -10.13922119140625, 126.369873046875, 6.441379547119141, -11.495498657226562, -224.47125244140625, 108.35075378417969, 40.806732177734375, 6.020599365234375, -21.42352294921875, 19.90838623046875, 163.82119750976562, -35.7784423828125, 200.79495239257812, 106.34783172607422, -0.3838958740234375, -8.12469482421875, 7.00830078125, -139.35971069335938, 8.17633056640625, 0.47422027587890625, 33.50721740722656, 0.0, 52.256256103515625, 38.057586669921875, 171.44398498535156, -13.444774627685547, 66.0726318359375, 47.923583984375, -1.8336334228515625, 6.09510612487793, 70.22000122070312, -5.335235595703125, -1.4794387817382812, 134.21975708007812, 12.288047790527344, 138.7178955078125, -23.71082305908203, -21.50299072265625, 0.0, 12.415809631347656, 138.08404541015625, 23.558731079101562, 37.91667175292969, -14.371162414550781, -15.33099365234375, 19.941864013671875, 171.29364013671875, 165.8333740234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000360.npy"}
{"epoch": 0.7539267015706806, "step": 361, "batch_size": 128, "mean": 71.81221008300781, "std": 102.05006408691406, "min": -195.15469360351562, "p10": -44.91802368164061, "median": 96.23916625976562, "p90": 182.3021057128906, "max": 257.21173095703125, "pos_frac": 0.7109375, "sample": [153.36819458007812, -7.744873046875, -115.84030151367188, 34.64434814453125, 34.77021789550781, -7.4810791015625, 1.4717788696289062, -33.8656005859375, 171.51104736328125, 51.553253173828125, 131.53604125976562, 15.9447021484375, 1.5928192138671875, 228.26031494140625, -8.162302017211914, 171.33607482910156, -1.4938240051269531, 16.954673767089844, -34.629058837890625, 0.8215789794921875, 169.2613525390625, 205.30740356445312, -121.37786865234375, 169.02487182617188, -4.277687072753906, 204.9796142578125, 145.82843017578125, 150.56007385253906, 128.01010131835938, 201.48980712890625, 7.835052490234375, -30.088027954101562, 114.08844757080078, 101.91434478759766, 14.938858032226562, 158.29885864257812, -176.38763427734375, 98.482666015625, 138.0615997314453, 138.70156860351562, 201.40264892578125, 211.46243286132812, 24.752914428710938, 99.27493286132812, 216.46212768554688, 27.3629150390625, -127.53070068359375, 37.61503601074219, 113.8057861328125, -15.774276733398438, -0.8138637542724609, 27.673965454101562, 120.5885009765625, 0.0, 93.99566650390625, 33.1544189453125, 139.6481170654297, 160.73046875, 4.149818420410156, -19.5296630859375, 35.81105041503906, 154.96925354003906, -53.61524200439453, 177.71490478515625, 129.32989501953125, -52.95037841796875, 78.68060302734375, 105.99396514892578, 24.0880126953125, -90.3533935546875, 83.6676025390625, -2.53131103515625, -146.68856811523438, -41.4755859375, 132.46435546875, 0.0, 107.926513671875, -37.71885681152344, -22.482101440429688, -12.818634033203125, 222.9901123046875, 167.29360961914062, 198.6082763671875, 155.17886352539062, -195.15469360351562, 153.62220764160156, -4.923583984375, -53.3568115234375, 257.21173095703125, 151.5697479248047, -140.84814453125, 144.49295043945312, -140.2177734375, 176.37594604492188, 188.51873779296875, 77.53846740722656, 132.75747680664062, 43.02618408203125, 167.63809204101562, 143.69090270996094, 176.23974609375, -7.787607192993164, 225.5626983642578, 169.85525512695312, 175.05133056640625, 166.1015625, 165.97470092773438, 111.09072875976562, -0.249237060546875, 178.50494384765625, 180.3231201171875, 43.85224151611328, -18.838043212890625, 128.66502380371094, 186.91973876953125, 115.81285095214844, -4.94775390625, 170.93267822265625, 179.73928833007812, 127.96142578125, 10.309768676757812, 152.52952575683594, -4.0160675048828125, 132.0027313232422, 50.27069854736328, 23.195030212402344, 164.047607421875, -190.79595947265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000361.npy"}
{"epoch": 0.7560209424083769, "step": 362, "batch_size": 128, "mean": 62.495445251464844, "std": 93.6598129272461, "min": -193.17782592773438, "p10": -37.075765991210936, "median": 32.729278564453125, "p90": 178.45769500732422, "max": 267.1043701171875, "pos_frac": 0.765625, "sample": [20.473037719726562, 172.97491455078125, -12.910160064697266, 71.56744384765625, 169.94451904296875, 116.5081787109375, 152.24880981445312, -2.8023681640625, 103.80844116210938, 13.437774658203125, 103.93670654296875, -193.17782592773438, -158.54660034179688, -59.53118133544922, 244.00753784179688, 4.481195449829102, 177.2294464111328, 9.48297119140625, 27.735565185546875, 139.06373596191406, 104.23602294921875, 155.19586181640625, 267.1043701171875, 47.67987060546875, 11.493896484375, 181.3236083984375, 141.32864379882812, -24.3768310546875, -180.34512329101562, 93.82896423339844, -27.923614501953125, -116.95384979248047, 25.802764892578125, 2.110004425048828, -99.22771453857422, 143.53363037109375, 25.619178771972656, -12.907764434814453, 176.06497192382812, 17.152793884277344, 215.825927734375, 119.583740234375, 201.82559204101562, 168.49253845214844, 191.41836547851562, 165.63693237304688, 145.9677734375, -0.237152099609375, -16.936279296875, 22.82916259765625, -37.8409423828125, 154.89414978027344, 153.214599609375, -36.747833251953125, 2.069143295288086, 3.61236572265625, -36.15789794921875, 204.4854736328125, -6.35980224609375, 159.632568359375, 167.12823486328125, -21.128265380859375, 32.874755859375, 136.07568359375, 84.07218933105469, 129.4525146484375, 28.94078826904297, 22.029159545898438, 128.599365234375, 33.022308349609375, 187.33456420898438, 175.43222045898438, 157.392333984375, 111.9854736328125, -6.491596221923828, 206.939453125, 133.31712341308594, 5.841590881347656, 204.4945068359375, 145.87828063964844, -38.984466552734375, 157.78253173828125, -6.792888641357422, 7.3381195068359375, -54.322265625, -95.00479125976562, -38.29029846191406, 201.80992126464844, 9.816692352294922, 27.14862823486328, 32.459625244140625, 2.836639404296875, -9.286529541015625, 15.025115966796875, 18.42767333984375, 11.625045776367188, 139.7353973388672, 100.90800476074219, 11.937789916992188, 7.5059814453125, -10.56951904296875, 160.6351318359375, 156.24057006835938, 21.31005859375, 3.478515625, 133.79962158203125, -147.65933227539062, 14.540496826171875, 20.00048828125, 19.838119506835938, 198.5774383544922, 10.178960800170898, 134.8638458251953, 119.65187072753906, 32.58380126953125, -10.761377334594727, 43.20648193359375, 190.8922119140625, 42.655548095703125, 60.688682556152344, -20.0228271484375, 36.788734436035156, 151.12632751464844, 167.44137573242188, 65.533935546875, -60.910491943359375, 76.187744140625, 18.374221801757812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000362.npy"}
{"epoch": 0.7581151832460733, "step": 363, "batch_size": 128, "mean": 69.04827880859375, "std": 94.54625701904297, "min": -209.02850341796875, "p10": -18.995333862304676, "median": 48.17637634277344, "p90": 189.3455047607422, "max": 316.56011962890625, "pos_frac": 0.78125, "sample": [163.44842529296875, -3.007293701171875, -14.075225830078125, 183.510498046875, 160.90362548828125, 74.74200439453125, -0.4187297821044922, -138.24429321289062, 53.82275390625, 66.61576080322266, -15.918548583984375, 8.843208312988281, 166.64724731445312, 102.20175170898438, 10.015754699707031, 66.3837890625, 90.96847534179688, 206.88087463378906, 4.182830810546875, -27.052419662475586, 22.5867919921875, 150.09365844726562, 200.43017578125, 101.0201416015625, 21.96612548828125, 190.5106201171875, 205.15505981445312, 238.4466552734375, 14.223159790039062, 46.890106201171875, 164.4290771484375, 28.118553161621094, -13.089324951171875, -7.6812591552734375, 127.70293426513672, 1.7704753875732422, -1.3949089050292969, 85.5281982421875, 167.43992614746094, 3.5340576171875, 9.245689392089844, 24.14795684814453, 112.7718505859375, 140.3772430419922, 53.22694396972656, 133.60223388671875, -15.962997436523438, 124.666015625, 191.8936767578125, 131.24151611328125, 141.25460815429688, 61.892669677734375, 316.56011962890625, 178.49252319335938, -144.23873901367188, 49.462646484375, 3.5602569580078125, 189.11447143554688, 36.95550537109375, 132.2640380859375, 28.557708740234375, 36.29789733886719, 192.58595275878906, 144.87310791015625, 201.35806274414062, 1.5750656127929688, 198.74264526367188, 162.94601440429688, 39.633880615234375, 5.780323028564453, 159.49746704101562, 7.491729736328125, 192.12774658203125, 240.0311279296875, 9.1414794921875, 173.66555786132812, -33.185699462890625, -13.114028930664062, 125.450439453125, 73.2274169921875, 2.6309661865234375, 8.233139038085938, -10.125885009765625, 29.049713134765625, 14.255447387695312, -14.726837158203125, -3.9898681640625, -7.583709716796875, -48.005615234375, 13.43206787109375, 161.54803466796875, -150.3190155029297, 158.693359375, 17.085525512695312, -0.0237274169921875, 188.14224243164062, 137.55458068847656, -29.15924072265625, 178.75454711914062, 87.88543701171875, 23.755935668945312, 127.5108642578125, -152.66770935058594, 187.66021728515625, 187.11764526367188, 121.34414672851562, 28.092628479003906, 59.2305908203125, 178.1359405517578, 61.16459655761719, -209.02850341796875, -26.070785522460938, -33.32914733886719, -1.4427566528320312, 20.548797607421875, 39.858917236328125, 8.912872314453125, 20.726165771484375, 146.7015838623047, 189.88458251953125, 182.61102294921875, 180.28387451171875, 25.863075256347656, 148.4769287109375, -132.86061096191406, 1.033203125, 27.045379638671875, -31.05426025390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000363.npy"}
{"epoch": 0.7602094240837697, "step": 364, "batch_size": 128, "mean": 51.657798767089844, "std": 95.98072814941406, "min": -222.77557373046875, "p10": -45.40536499023436, "median": 34.42448043823242, "p90": 168.6551345825195, "max": 259.00933837890625, "pos_frac": 0.65625, "sample": [64.64939880371094, -3.1879310607910156, 141.18408203125, 61.597171783447266, 127.68815612792969, 13.780746459960938, 139.91571044921875, 140.03475952148438, 134.4835205078125, 66.89152526855469, 48.70588684082031, 112.75405883789062, -12.586273193359375, -10.27734375, -12.4866943359375, 120.86564636230469, 57.45782470703125, 73.8182373046875, 0.431884765625, 138.83218383789062, -166.236328125, -7.09002685546875, 128.2153778076172, -169.91574096679688, 117.9681396484375, -82.49948120117188, 189.58099365234375, 151.60748291015625, -7.214027404785156, 156.22720336914062, 6.84808349609375, 18.210960388183594, 68.2835693359375, 6.343168258666992, -29.95477294921875, -6.4580841064453125, -8.968170166015625, 151.2886199951172, -3.347625732421875, 161.57305908203125, 127.68365478515625, 6.96221923828125, -195.79849243164062, 68.48811340332031, 144.75587463378906, 194.8306884765625, -58.36307144165039, 109.62722778320312, 20.600067138671875, -29.994400024414062, 167.16305541992188, 171.96514892578125, -18.445297241210938, 56.770233154296875, -12.801765441894531, -5.610443115234375, 191.64697265625, -1.6449394226074219, -222.77557373046875, 237.5691680908203, 149.412841796875, -13.91046142578125, -14.18035888671875, -116.36421203613281, 37.1673583984375, 142.33416748046875, -26.327301025390625, 190.75448608398438, -81.7525634765625, 1.0516357421875, 33.81158447265625, 167.23655700683594, 84.419677734375, -16.20184326171875, 100.55015563964844, 8.58868408203125, 201.330810546875, 35.037376403808594, 11.014999389648438, 156.59097290039062, 145.4642333984375, -6.27069091796875, 28.18804931640625, -6.98138427734375, -5.3159027099609375, 115.93507385253906, 24.016952514648438, -24.4139404296875, 59.81402587890625, 239.87518310546875, 143.55303955078125, 15.309234619140625, -37.25787353515625, 1.865997314453125, 163.20263671875, 157.54736328125, -0.06538009643554688, 2.134754180908203, -41.562255859375, 259.00933837890625, 35.62054443359375, 176.4971160888672, -82.82681274414062, 223.00439453125, 132.40887451171875, -54.37261962890625, 93.93338012695312, 35.79791259765625, 1.33245849609375, -20.089324951171875, 114.72314453125, -2.841552734375, -16.542457580566406, 200.6669464111328, -0.22107696533203125, -5.892578125, 2.988525390625, -126.39614868164062, 96.664306640625, -116.94731140136719, 17.424915313720703, 114.68224334716797, 114.17015075683594, 13.787506103515625, -101.14088439941406, 108.16473388671875, 242.90235900878906, 98.44308471679688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000364.npy"}
{"epoch": 0.762303664921466, "step": 365, "batch_size": 128, "mean": 66.91775512695312, "std": 93.88954162597656, "min": -198.4896240234375, "p10": -28.56165618896484, "median": 39.7164306640625, "p90": 191.05033569335936, "max": 225.37478637695312, "pos_frac": 0.765625, "sample": [0.0, 4.963283538818359, 116.97821044921875, 5.44183349609375, 183.99020385742188, 207.81265258789062, 9.189788818359375, 69.80313110351562, 0.0, 42.38214111328125, 25.85693359375, 3.168701171875, 9.614349365234375, 192.8173828125, 16.13146209716797, 12.108291625976562, 19.0347900390625, 96.10376739501953, 15.512130737304688, 179.7644805908203, 45.437255859375, 188.6395263671875, -1.803497314453125, 21.430984497070312, 54.20436096191406, 77.02162170410156, -27.781173706054688, 171.51583862304688, 151.57211303710938, 166.41238403320312, 171.293212890625, -67.73676300048828, -44.712921142578125, 145.05294799804688, -148.0914306640625, -78.12026977539062, 151.31451416015625, 45.79322052001953, 154.2634735107422, 139.54525756835938, -20.57044219970703, 154.021728515625, 195.49786376953125, 135.5797882080078, -30.382781982421875, 199.38406372070312, 4.9129638671875, 149.511962890625, 30.56830596923828, 172.693359375, 156.16909790039062, -21.277587890625, 37.05072021484375, -1.7663726806640625, 16.510055541992188, 8.362884521484375, 123.31401062011719, 180.3362579345703, 173.84512329101562, -34.460357666015625, 27.400634765625, 150.2950439453125, -42.767120361328125, 4.74896240234375, 4.384647369384766, 109.30731201171875, 140.65403747558594, 151.78831481933594, 103.66769409179688, 195.59063720703125, 202.55252075195312, -19.86517333984375, 193.10507202148438, 0.2467193603515625, 17.597412109375, 203.95230102539062, 156.68353271484375, -153.29208374023438, 190.29302978515625, 2.639251708984375, 196.54067993164062, 3.3368301391601562, 225.37478637695312, 13.741470336914062, 14.959245681762695, -27.402542114257812, 162.34765625, 53.31333923339844, 5.779262542724609, -15.994483947753906, 98.71647644042969, 24.679445266723633, 84.9292221069336, 186.78472900390625, -7.935516357421875, 5.3698577880859375, -1.72259521484375, -19.75262451171875, 154.62765502929688, 3.3524169921875, 24.422958374023438, -72.4818115234375, -19.1275634765625, -3.609375, 155.825439453125, 0.0, 44.80561065673828, 33.340721130371094, 156.83860778808594, 196.853515625, 85.42340087890625, 130.61813354492188, 4.323009490966797, -26.09295654296875, -56.88873291015625, -193.0958251953125, 157.81320190429688, 199.794189453125, -43.17236328125, 155.10577392578125, 34.69366455078125, -198.4896240234375, 221.06845092773438, 34.3701171875, 125.13345336914062, 152.74063110351562, 167.60140991210938, 142.406005859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000365.npy"}
{"epoch": 0.7643979057591623, "step": 366, "batch_size": 128, "mean": 63.04789352416992, "std": 92.13922882080078, "min": -215.55819702148438, "p10": -31.24175262451171, "median": 45.767127990722656, "p90": 179.76746826171873, "max": 233.28207397460938, "pos_frac": 0.734375, "sample": [128.84024047851562, 29.740890502929688, 233.28207397460938, 161.3211212158203, 138.24041748046875, 151.03866577148438, 178.39794921875, 151.05322265625, 119.51959228515625, 204.3348388671875, 183.67034912109375, 107.12956237792969, 136.04202270507812, -42.588958740234375, 32.4273567199707, 138.47850036621094, 110.58055114746094, 40.57304382324219, -159.10427856445312, 43.60679626464844, -215.55819702148438, 212.5345458984375, -152.9412841796875, 12.378890991210938, 53.99102783203125, 150.11349487304688, 155.12977600097656, 78.54345703125, 162.69711303710938, 50.188446044921875, -14.42999267578125, 166.99041748046875, 47.324920654296875, -36.751251220703125, 176.864013671875, -200.76284790039062, -3.81494140625, 131.3670654296875, 15.743133544921875, 19.82794189453125, 3.7529640197753906, 2.6551055908203125, -4.94573974609375, -20.83899688720703, -8.80377197265625, 118.53498840332031, 166.43055725097656, -39.915130615234375, 23.170745849609375, -13.40283203125, 111.7470703125, 36.97105407714844, 11.7921142578125, 79.869873046875, 124.36949157714844, -20.898040771484375, 30.451751708984375, 17.6817626953125, 36.0159912109375, 16.429176330566406, -1.4847412109375, 156.61703491210938, 117.76567077636719, 2.4061927795410156, 8.818824768066406, 186.42816162109375, 50.69084167480469, 44.18084716796875, 103.47583770751953, -71.30947875976562, 150.24136352539062, 189.77706909179688, -28.880538940429688, 0.0, -22.75531005859375, 188.95176696777344, 65.87030029296875, 140.43881225585938, 137.77200317382812, -1.7931442260742188, 6.8176116943359375, 44.98040771484375, 206.92269897460938, 97.41712951660156, -4.909912109375, 65.61831665039062, 21.64971923828125, 209.87860107421875, -22.264434814453125, -15.487529754638672, -4.0033111572265625, -46.40386962890625, 92.187255859375, 11.968582153320312, 122.91363525390625, 150.33856201171875, 122.82861328125, -9.51312255859375, 60.06787109375, 174.85826110839844, 161.77268981933594, 16.862319946289062, 6.4540252685546875, 204.81741333007812, 161.74838256835938, 46.55384826660156, 139.76145935058594, -21.730712890625, 213.22171020507812, 2.5006580352783203, 2.35418701171875, 142.5677032470703, 191.63848876953125, 6.72979736328125, -8.712348937988281, -42.603790283203125, 169.04962158203125, -155.9619598388672, 128.39474487304688, 32.7154541015625, 147.86505126953125, 14.07208251953125, -6.7547760009765625, 166.28366088867188, -52.1383056640625, -12.047531127929688, 182.9630126953125, -38.41107177734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000366.npy"}
{"epoch": 0.7664921465968586, "step": 367, "batch_size": 128, "mean": 66.7897720336914, "std": 89.16327667236328, "min": -155.8077392578125, "p10": -24.76010589599609, "median": 44.49993133544922, "p90": 187.75887145996091, "max": 215.75973510742188, "pos_frac": 0.765625, "sample": [150.59051513671875, 28.0205078125, -21.78106689453125, 200.01553344726562, -0.6076717376708984, 20.113876342773438, 70.27164459228516, -150.3411865234375, -8.68072509765625, 131.5133819580078, 156.7264404296875, 19.57746696472168, 4.279685974121094, -2.17071533203125, 78.123046875, 204.84771728515625, 53.24934387207031, 185.59420776367188, -1.599212646484375, 6.22198486328125, 25.0762939453125, 210.80438232421875, 110.45053100585938, -42.177886962890625, 6.481544494628906, 201.15072631835938, -65.09326171875, -12.53363037109375, 28.291915893554688, -20.980575561523438, -23.736953735351562, 141.634765625, 2.821950912475586, -23.50457000732422, 160.57498168945312, -9.78924560546875, 60.913787841796875, 153.16029357910156, 147.04818725585938, -155.8077392578125, 170.92340087890625, 114.15473937988281, 9.049732208251953, 130.578857421875, 179.9881591796875, 3.462554931640625, -104.07882690429688, -7.6607666015625, -8.06964111328125, -10.989791870117188, 54.72283935546875, 6.1659393310546875, 114.96167755126953, 25.865478515625, 159.970947265625, 168.02310180664062, 156.35842895507812, 139.916259765625, 86.01878356933594, -28.229217529296875, 22.075439453125, -27.1474609375, 207.04336547851562, 105.6099853515625, 18.30614471435547, 190.34390258789062, -31.70147705078125, 146.5585479736328, 215.75973510742188, 39.438541412353516, 69.43203735351562, 21.940338134765625, 45.89622497558594, 185.55630493164062, 193.72869873046875, 171.26235961914062, 6.6573486328125, -27.82501220703125, 15.113433837890625, -27.85247802734375, 43.1036376953125, 182.15699768066406, 168.38845825195312, 177.0548095703125, 17.766357421875, 21.937103271484375, 181.65634155273438, 23.91033935546875, -60.564239501953125, 115.60919189453125, 174.77642822265625, 136.4779052734375, 20.67926025390625, -17.2421875, 7.838632583618164, -6.762855529785156, -119.9681167602539, 190.35635375976562, 112.0528564453125, 186.6510009765625, 3.3660812377929688, 195.58734130859375, 168.2880401611328, 3.28411865234375, -13.35601806640625, 107.20193481445312, 149.5069580078125, 191.6884765625, 5.00616455078125, 193.34869384765625, 94.21511840820312, 118.96487426757812, 30.61895751953125, -140.75851440429688, 52.78065490722656, 10.892578125, 4.585906982421875, 152.3045654296875, 78.60565185546875, 161.54298400878906, 13.77684211730957, 102.3741455078125, 19.18280029296875, 6.75592041015625, 210.03799438476562, 126.76554870605469, -19.696624755859375, 46.26282501220703], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000367.npy"}
{"epoch": 0.768586387434555, "step": 368, "batch_size": 128, "mean": 74.9022445678711, "std": 91.01747131347656, "min": -169.5638427734375, "p10": -14.032733154296867, "median": 70.8692626953125, "p90": 180.85609130859376, "max": 305.30572509765625, "pos_frac": 0.8046875, "sample": [10.72906494140625, 198.02801513671875, 132.83851623535156, 168.4525146484375, 20.79931640625, 174.69976806640625, 82.0736312866211, -3.919219970703125, 8.79205322265625, -7.771697998046875, 71.97390747070312, -31.292556762695312, 79.96844482421875, 208.68829345703125, 131.66018676757812, 59.16278076171875, 226.68792724609375, 46.265716552734375, 75.72796630859375, 134.20745849609375, -18.52825927734375, -12.1060791015625, 43.443878173828125, -21.391639709472656, 5.64312744140625, 25.05615234375, -3.6373367309570312, 5.315654754638672, 69.76461791992188, 178.77862548828125, 3.1594467163085938, 51.647705078125, -65.3395767211914, 33.17115783691406, 10.5516357421875, -169.5638427734375, 43.857574462890625, 305.30572509765625, 156.76937866210938, 142.91793823242188, 27.718704223632812, 181.1737060546875, 168.3319549560547, 147.7835693359375, 31.487762451171875, 13.358673095703125, 149.5428466796875, -5.9396209716796875, 39.26659393310547, -169.4184112548828, 6.41497802734375, 160.4481201171875, 7.300331115722656, 108.93205261230469, -76.51250457763672, -7.37310791015625, 198.05667114257812, 151.32992553710938, -2.5235595703125, 136.35585021972656, 30.78643798828125, 65.22891235351562, 121.9012451171875, 166.98924255371094, 189.01416015625, 139.86270141601562, -45.40692138671875, 96.82223510742188, 27.089569091796875, -9.218185424804688, -4.7938232421875, 203.42807006835938, 191.05184936523438, 77.25004577636719, 32.96492004394531, 157.7787322998047, 163.9525146484375, 103.69903564453125, 179.39474487304688, 159.0203857421875, 3.2133636474609375, 133.6767578125, 168.0364227294922, 57.333770751953125, -9.546455383300781, 176.32928466796875, 10.642425537109375, 107.88671112060547, 1.4237823486328125, 140.75198364257812, -1.5780563354492188, 12.066810607910156, 179.08590698242188, 15.848762512207031, 133.79800415039062, 97.3352279663086, 151.24099731445312, 137.74200439453125, 3.3566970825195312, 137.74542236328125, 180.719970703125, -100.90430450439453, 89.87486267089844, 199.984619140625, -77.46337890625, 30.76409912109375, -9.185394287109375, 126.52835083007812, 219.18218994140625, 190.33126831054688, 28.84979248046875, 168.6143341064453, 23.699432373046875, 11.254592895507812, 11.540122985839844, 162.22055053710938, 178.34768676757812, 169.0667724609375, 159.37646484375, 129.34323120117188, 38.19146728515625, -161.03468322753906, -89.07958984375, 5.88031005859375, 85.08492279052734, 182.8811492919922, -78.68997192382812, 170.57949829101562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000368.npy"}
{"epoch": 0.7706806282722513, "step": 369, "batch_size": 128, "mean": 50.717994689941406, "std": 97.50862121582031, "min": -194.42987060546875, "p10": -65.81745147705077, "median": 26.710979461669922, "p90": 175.77272491455076, "max": 244.12210083007812, "pos_frac": 0.734375, "sample": [160.3674774169922, 26.719810485839844, 158.5688018798828, -86.02853393554688, 11.047924041748047, -9.20941162109375, 103.83985900878906, 4.71331787109375, 154.27224731445312, 0.16412353515625, 145.098876953125, 40.51446533203125, -46.22548294067383, 78.149658203125, 198.36297607421875, 194.17727661132812, 131.9272918701172, 149.94223022460938, 24.0869140625, 143.50518798828125, 203.13967895507812, 2.969470977783203, 120.48428344726562, 149.93194580078125, -38.45733642578125, -181.98858642578125, 163.94525146484375, -16.32098388671875, 5.3040771484375, 2.609344482421875, 141.36590576171875, 26.7021484375, 182.87490844726562, 113.3330078125, -8.2567138671875, 154.99169921875, -31.869674682617188, 130.052001953125, 137.28561401367188, 40.42185974121094, 9.650146484375, 1.724029541015625, 168.368408203125, -36.35552978515625, 11.5738525390625, -118.0743408203125, -1.195791244506836, 140.60739135742188, -130.66409301757812, 195.49908447265625, -136.2469024658203, -150.119384765625, 65.208984375, -141.16815185546875, 193.20651245117188, 136.43292236328125, -132.9225616455078, -5.804588317871094, 152.63815307617188, 7.78729248046875, -151.23089599609375, -4.3065185546875, -147.70416259765625, 178.2617645263672, 181.43080139160156, 41.18401336669922, 56.29801940917969, -2.144317626953125, 153.72857666015625, -76.08689880371094, 244.12210083007812, -13.589630126953125, 12.284017562866211, 14.872314453125, 5.22869873046875, 221.65582275390625, 66.57373046875, -5.86431884765625, -55.103919982910156, 144.849365234375, 5.060611724853516, -194.42987060546875, 12.892417907714844, 75.62933349609375, 17.3853759765625, 134.5849609375, 155.59707641601562, 168.0031280517578, -36.524566650390625, 0.0, 17.349838256835938, 153.4444580078125, 49.397979736328125, 128.7691192626953, -7.3905792236328125, 1.517913818359375, 73.4561767578125, -21.237823486328125, 48.95391845703125, 167.82102966308594, 19.858062744140625, 159.26324462890625, 11.39666748046875, -61.416259765625, 69.04437255859375, -16.742727279663086, 158.59243774414062, 18.052993774414062, 174.70599365234375, -91.19015502929688, 42.38246154785156, 89.20763397216797, 14.33599853515625, 150.52963256835938, 180.52908325195312, 3.4908447265625, 33.057830810546875, 187.42333984375, -20.236282348632812, 127.45932006835938, 2.7762203216552734, 187.76849365234375, 13.944671630859375, 24.9447021484375, 27.120040893554688, 13.781387329101562, 43.085693359375, 1.34033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000369.npy"}
{"epoch": 0.7727748691099476, "step": 370, "batch_size": 128, "mean": 66.47425079345703, "std": 100.86552429199219, "min": -210.57220458984375, "p10": -25.590312576293943, "median": 38.179588317871094, "p90": 183.84405517578125, "max": 363.0812683105469, "pos_frac": 0.78125, "sample": [121.31784057617188, -167.89974975585938, 177.17236328125, 35.9227294921875, 38.7850341796875, 1.3208465576171875, -69.87178802490234, 3.090240478515625, 141.8662109375, 14.206588745117188, 40.32663345336914, 151.51145935058594, -11.500919342041016, 19.3453369140625, 143.11795043945312, 8.8084716796875, -0.7816162109375, 266.74310302734375, 0.5011825561523438, 28.6705322265625, 151.68438720703125, 30.105825424194336, 137.54010009765625, 157.7947998046875, -26.578033447265625, -3.8200759887695312, -25.167003631591797, 162.21340942382812, 7.759899139404297, 29.67535400390625, 26.93353271484375, -29.045074462890625, 154.38836669921875, 35.73209762573242, 3.9297332763671875, -0.7575836181640625, 241.70654296875, 104.97401428222656, 4.760944366455078, 189.29612731933594, 147.09133911132812, -174.50257873535156, 184.910888671875, 9.58062744140625, 142.4233856201172, 114.36085510253906, 107.849609375, 37.65705871582031, 42.39837646484375, 0.49956321716308594, 132.21080017089844, 18.574981689453125, 9.259750366210938, 149.25592041015625, 110.40380859375, -20.744903564453125, 151.8535919189453, 4.948486328125, 174.97134399414062, -71.47476959228516, -23.64824676513672, 170.71578979492188, -11.913101196289062, -116.90203857421875, 8.17148208618164, 172.51901245117188, 233.063232421875, 99.19961547851562, 139.84751892089844, -92.70133972167969, 3.85638427734375, -8.966339111328125, 63.59735107421875, 174.70291137695312, 192.84017944335938, -47.409912109375, 197.3868408203125, 187.70907592773438, 134.505859375, 172.08143615722656, 189.49053955078125, 163.1920623779297, 100.76046752929688, 226.9090576171875, 157.23236083984375, 164.72393798828125, -25.05028533935547, 7.801666259765625, -184.98223876953125, 166.26377868652344, 149.42724609375, 11.406957626342773, -100.7713623046875, 23.473052978515625, 363.0812683105469, -10.788745880126953, 16.433197021484375, 49.79216003417969, 164.24986267089844, 166.74444580078125, 179.73931884765625, 189.90350341796875, 19.0330810546875, 183.3868408203125, -11.85107421875, 121.83139038085938, 95.728759765625, 32.95170211791992, 38.702117919921875, 68.1328125, 202.05654907226562, -210.57220458984375, 8.252159118652344, -8.9527587890625, 2.2134170532226562, 159.71826171875, 19.646728515625, 9.062744140625, 176.27908325195312, 14.386383056640625, 12.522159576416016, -152.60299682617188, 40.62158966064453, 169.42282104492188, -13.409019470214844, 11.217422485351562, 141.66986083984375, -5.73968505859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000370.npy"}
{"epoch": 0.774869109947644, "step": 371, "batch_size": 128, "mean": 55.888893127441406, "std": 93.62895965576172, "min": -229.17440795898438, "p10": -37.4199722290039, "median": 36.01155090332031, "p90": 175.74154815673828, "max": 220.73904418945312, "pos_frac": 0.7265625, "sample": [-0.06977081298828125, 122.38955688476562, 58.58599853515625, 158.52276611328125, -12.782392501831055, 0.5459518432617188, 203.09185791015625, -1.8450927734375, 65.14707946777344, 22.125396728515625, 1.4192428588867188, -214.177734375, 6.72137451171875, 14.533660888671875, 30.784866333007812, -9.959716796875, -84.336669921875, 111.13157653808594, 11.406463623046875, -26.14068603515625, -31.283203125, 108.65397644042969, -9.35760498046875, 180.53598022460938, 47.99310302734375, 117.16168212890625, -0.25663185119628906, 42.33404541015625, -27.593368530273438, 37.588653564453125, 144.9813232421875, 180.3489532470703, -96.22981262207031, 11.335403442382812, -229.17440795898438, 94.12081909179688, -41.59033203125, 1.9075927734375, 136.66859436035156, 25.687713623046875, 176.23434448242188, 14.26519775390625, -121.17790222167969, -6.396026611328125, 152.36013793945312, 7.753631591796875, 13.9635009765625, 0.0, -70.52301025390625, 10.794364929199219, 190.38523864746094, 161.8102264404297, 91.708740234375, -75.6607666015625, -166.97164916992188, -5.29058837890625, 32.11463928222656, 190.99163818359375, 162.25497436523438, 11.8016357421875, 12.148643493652344, 105.6868896484375, 46.93684387207031, 207.41635131835938, 31.299560546875, 112.74951171875, 93.81607055664062, 160.70443725585938, 179.06024169921875, -22.447113037109375, 28.86153793334961, 170.16204833984375, 139.72921752929688, -107.87127685546875, 142.83346557617188, 149.02423095703125, 150.03616333007812, 33.11903381347656, 179.62908935546875, 16.193893432617188, -42.035003662109375, -165.5439453125, 17.7186279296875, 214.52308654785156, 126.03509521484375, -20.973052978515625, -126.3439712524414, 148.96560668945312, 156.85382080078125, 59.467689514160156, 0.8350677490234375, -19.76177978515625, 83.38369750976562, 111.79222106933594, 35.968658447265625, -3.2863388061523438, 157.922119140625, 70.37039184570312, 22.919090270996094, 214.91793823242188, 24.893041610717773, -1.5977764129638672, 79.8260498046875, 98.72396850585938, 156.23306274414062, 153.45361328125, -1.1873779296875, 36.054443359375, 33.976905822753906, 150.98928833007812, 165.1396484375, -35.63267517089844, 145.30599975585938, 40.87129211425781, 14.665618896484375, 161.10118103027344, 175.5303497314453, 22.519506454467773, 220.73904418945312, 161.32537841796875, -3.36407470703125, -27.820098876953125, 49.55924987792969, -6.1442108154296875, 74.20632934570312, 155.03509521484375, 126.64241790771484, 188.5709686279297], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000371.npy"}
{"epoch": 0.7769633507853403, "step": 372, "batch_size": 128, "mean": 70.81936645507812, "std": 98.12513732910156, "min": -191.60678100585938, "p10": -20.190407562255857, "median": 58.05554962158203, "p90": 188.65355529785157, "max": 313.305908203125, "pos_frac": 0.7734375, "sample": [204.066650390625, -5.6634521484375, 152.46697998046875, 60.35585403442383, -1.7525768280029297, 150.85693359375, 137.1173095703125, -8.854949951171875, 147.13705444335938, 183.65013122558594, 59.40040588378906, 71.30693054199219, 142.58746337890625, 149.57931518554688, -25.685546875, 143.5435791015625, 15.453845977783203, 140.3353271484375, -13.582197189331055, 19.172027587890625, 82.08181762695312, 36.069175720214844, 55.902099609375, -2.76739501953125, -13.65618896484375, 158.98971557617188, 213.20391845703125, 106.92051696777344, 164.7930450439453, 143.99795532226562, -26.232711791992188, -22.664520263671875, 194.94589233398438, -12.81494140625, 115.25923156738281, 143.98353576660156, 197.5657958984375, -0.3551483154296875, 92.0617904663086, 56.201377868652344, -19.13007354736328, 193.43927001953125, 102.75608825683594, 148.29574584960938, 206.94290161132812, -187.86260986328125, 214.0303955078125, 60.937286376953125, 12.067512512207031, 32.012969970703125, 11.153816223144531, 6.5257415771484375, 144.93341064453125, 0.874114990234375, 23.971038818359375, 32.529327392578125, 121.64788818359375, 178.63653564453125, 139.66969299316406, 29.707489013671875, 73.24807739257812, 34.03057861328125, 145.10736083984375, 19.42779541015625, 4.986572265625, 57.60115051269531, 168.54583740234375, 32.01385498046875, 180.50070190429688, -10.043083190917969, 3.0634765625, 5.64373779296875, 14.951431274414062, 173.23516845703125, 187.8514404296875, 207.719482421875, -155.54248046875, 195.58175659179688, 16.6390380859375, -159.9641876220703, 164.64144897460938, 105.65950012207031, 39.40777587890625, 56.833251953125, -154.77536010742188, 60.21128845214844, 48.352508544921875, 7.4440155029296875, -5.519338607788086, 180.60548400878906, -12.172119140625, -10.017593383789062, 170.40072631835938, 105.78070068359375, 41.5679931640625, -191.60678100585938, -125.93814086914062, 32.8524169921875, 140.22242736816406, -111.19243621826172, 176.91290283203125, -8.025772094726562, -76.375732421875, 2.291229248046875, 188.51687622070312, 154.287109375, -173.86346435546875, 188.97247314453125, 313.305908203125, 200.87091064453125, 58.50994873046875, 226.1905517578125, 180.26205444335938, 3.7247314453125, 57.1650390625, 183.56683349609375, -0.2517738342285156, 146.58740234375, -29.9730224609375, 172.9139404296875, 129.26007080078125, 154.10507202148438, 19.361679077148438, 2.916595458984375, -3.580547332763672, 126.92268371582031, 46.4093017578125, 28.426025390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000372.npy"}
{"epoch": 0.7790575916230367, "step": 373, "batch_size": 128, "mean": 67.40376281738281, "std": 98.70777130126953, "min": -218.0975341796875, "p10": -35.75840110778808, "median": 47.92979431152344, "p90": 195.0307403564453, "max": 220.212646484375, "pos_frac": 0.796875, "sample": [12.989501953125, 36.925010681152344, 180.21109008789062, 30.723403930664062, 0.7639350891113281, -24.406494140625, 17.386322021484375, 210.934814453125, 16.66717529296875, 205.87161254882812, 165.03213500976562, -17.22869873046875, 4.5259857177734375, 173.42752075195312, 129.16824340820312, 78.31329345703125, 2.6013946533203125, 48.35858154296875, -39.22857666015625, 179.66015625, 68.46798706054688, 0.7594070434570312, 135.3095245361328, 78.54815673828125, -46.24346923828125, 47.501007080078125, -6.4824066162109375, -41.865142822265625, 38.93035888671875, 203.92034912109375, 2.7889785766601562, -27.882171630859375, 23.001495361328125, 62.73199462890625, 70.97024536132812, 149.22308349609375, 167.327880859375, -18.52471923828125, 193.175537109375, 81.7266845703125, 109.5216064453125, 159.7352294921875, 194.05471801757812, 99.71184539794922, 17.92535400390625, 190.55877685546875, 14.034347534179688, -95.2342529296875, 49.194366455078125, 195.53469848632812, -34.271183013916016, 174.36776733398438, -116.47742462158203, -44.22802734375, 85.12322998046875, -149.21743774414062, 206.82318115234375, 3.925689697265625, 27.6351318359375, 59.48985290527344, 4.21708869934082, -218.0975341796875, 212.23748779296875, 27.73773193359375, 180.01531982421875, 16.085647583007812, -142.13418579101562, 180.113525390625, 13.411697387695312, 143.23822021484375, 185.469482421875, 178.36264038085938, 12.712081909179688, 101.04100036621094, 23.136703491210938, 82.26423645019531, 57.923828125, 2.3167724609375, 86.97610473632812, 4.27862548828125, -163.30465698242188, -16.921852111816406, 207.35421752929688, 10.026519775390625, 40.627838134765625, 179.10183715820312, 179.02081298828125, 211.27078247070312, 214.6082763671875, 120.27983093261719, 10.1663818359375, 190.16322326660156, -4.1688232421875, 21.412765502929688, 181.12123107910156, 178.3865966796875, 26.17779541015625, 207.827880859375, -131.06027221679688, -156.4635772705078, 99.95549774169922, 169.4502716064453, 220.212646484375, 170.97064208984375, 68.56405639648438, 16.14312744140625, 145.84701538085938, -15.6903076171875, 215.09524536132812, 18.063079833984375, 14.561786651611328, 15.392219543457031, 26.495513916015625, 132.24935913085938, -3.1492042541503906, -11.804611206054688, 159.73995971679688, 5.662261962890625, -68.25041961669922, 108.00454711914062, -3.39129638671875, 194.81475830078125, -13.850833892822266, 4.23968505859375, 184.51211547851562, 206.86785888671875, 151.07440185546875, 14.680267333984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000373.npy"}
{"epoch": 0.7811518324607329, "step": 374, "batch_size": 128, "mean": 45.564208984375, "std": 98.20267486572266, "min": -199.27545166015625, "p10": -71.1726837158203, "median": 23.15599822998047, "p90": 187.04207611083982, "max": 249.71243286132812, "pos_frac": 0.7109375, "sample": [243.98394775390625, 0.0, 9.526092529296875, 232.27542114257812, 5.323486328125, -151.37545776367188, -4.644767761230469, 24.90472412109375, 190.69422912597656, 144.7916717529297, 7.581480026245117, 220.10528564453125, 19.783233642578125, -66.36968994140625, 37.92466735839844, 36.03656005859375, 93.8838119506836, 146.2960968017578, 18.380859375, 12.66473388671875, -143.54373168945312, -41.602294921875, 111.7972183227539, -145.9906005859375, 173.42298889160156, 26.033287048339844, 21.407272338867188, 12.8446044921875, 4.426532745361328, -20.190711975097656, 16.679092407226562, 6.7673187255859375, -37.4764404296875, 61.622406005859375, -8.1446533203125, 8.113395690917969, 26.463287353515625, -3.274993896484375, 197.9478759765625, 152.2551727294922, 54.031158447265625, 10.11083984375, 249.71243286132812, 1.0127639770507812, 64.292236328125, 1.79095458984375, 1.671478271484375, 81.396240234375, -199.27545166015625, -17.998313903808594, 30.077285766601562, -48.710205078125, 185.47686767578125, 114.44998931884766, 25.546356201171875, 61.844482421875, -4.81573486328125, 65.07977294921875, 96.7183837890625, -1.3323478698730469, 15.199882507324219, 200.10321044921875, 205.60928344726562, -19.667144775390625, -1.342376708984375, 91.02608489990234, 51.5152587890625, 93.89471435546875, -25.055419921875, 135.6614227294922, 20.2921142578125, 61.078582763671875, 146.44158935546875, 136.59222412109375, -181.14511108398438, 138.4345703125, -101.5126953125, 9.23040771484375, 15.283477783203125, 161.39569091796875, 147.11273193359375, 7.652923583984375, 72.30355834960938, 163.99746704101562, 0.0, 162.25677490234375, -4.739646911621094, -92.899658203125, -41.223663330078125, -142.29693603515625, 2.8969345092773438, 156.50006103515625, 169.50238037109375, 190.84144592285156, 17.6663818359375, -19.401336669921875, 216.16854858398438, 25.485870361328125, 196.61154174804688, 58.9090576171875, 16.505615234375, -82.37966918945312, -120.79269409179688, 25.2791748046875, 29.008298873901367, 3.184612274169922, 29.1324462890625, 208.43557739257812, -21.48529815673828, -131.52684020996094, -3.26190185546875, 117.05496215820312, 8.604339599609375, -5.5001678466796875, 179.0804443359375, -2.996673583984375, 168.2129364013672, 110.45013427734375, -148.0208740234375, 107.9171142578125, 231.68753051757812, 156.72802734375, 162.77716064453125, 29.612747192382812, 35.51922607421875, -3.484180450439453, 19.497146606445312, -139.8033447265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000374.npy"}
{"epoch": 0.7832460732984293, "step": 375, "batch_size": 128, "mean": 56.25138854980469, "std": 96.69123840332031, "min": -198.61190795898438, "p10": -32.950122070312496, "median": 24.647457122802734, "p90": 183.90309448242186, "max": 272.90252685546875, "pos_frac": 0.7578125, "sample": [2.806976318359375, 85.2619857788086, 43.35186767578125, 6.7375030517578125, 36.736083984375, 2.855072021484375, 23.1754150390625, -130.45065307617188, 6.384254455566406, 40.249420166015625, 197.95562744140625, 0.8675327301025391, 178.29568481445312, 192.19155883789062, 110.64068603515625, -8.39141845703125, 76.30686950683594, 142.9456329345703, 0.4374866485595703, 8.840492248535156, 167.1336669921875, 182.8616943359375, 14.872360229492188, 38.454803466796875, 55.388397216796875, 174.8480987548828, -11.943119049072266, 10.115409851074219, -37.63227081298828, 40.508140563964844, 16.670082092285156, 11.83441162109375, -4.416110992431641, -129.2724609375, 17.58179473876953, 178.22442626953125, 124.29104614257812, 185.31390380859375, 23.589195251464844, -25.245391845703125, 17.711502075195312, 13.762290954589844, -137.20440673828125, 0.414886474609375, -198.61190795898438, 144.0486602783203, 166.84475708007812, 169.6497802734375, 12.983024597167969, -155.231689453125, -153.24374389648438, 3.063629150390625, 104.15475463867188, -6.294824600219727, 24.1160888671875, -5.15093994140625, 150.67446899414062, -11.129295349121094, -35.65789794921875, -138.92361450195312, 196.82138061523438, 176.40823364257812, -18.2044677734375, 230.6168212890625, 46.70365524291992, 23.4586181640625, -3.4822235107421875, 2.5449447631835938, 272.90252685546875, -9.619867324829102, 108.12649536132812, 13.42041015625, 26.594558715820312, 83.60608673095703, 32.867523193359375, 194.015869140625, 48.0264892578125, 6.444061279296875, 50.340179443359375, 44.55140686035156, 155.65679931640625, 93.32369995117188, -7.756134033203125, -35.05360412597656, -3.0471649169921875, 135.119140625, -2.236968994140625, 146.38018798828125, 138.74069213867188, 166.84939575195312, -160.4010009765625, 186.39820861816406, 218.0135498046875, -13.920425415039062, 224.9744873046875, 235.97097778320312, -48.917755126953125, 179.40171813964844, 118.47541809082031, 183.2984619140625, -3.5564651489257812, -128.17755126953125, 181.58743286132812, -2.3948516845703125, 157.095703125, 2.3227481842041016, 141.31134033203125, 191.4171142578125, 74.82693481445312, 6.33367919921875, -3.8011474609375, 190.80429077148438, 18.164306640625, 35.661407470703125, 172.89013671875, -32.04862976074219, 3.4441795349121094, 177.38027954101562, 2.5751800537109375, 17.79583740234375, 5.85272216796875, 25.17882537841797, 63.842071533203125, 147.3460693359375, 12.81219482421875, 155.96653747558594, 10.777618408203125, 121.00611877441406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000375.npy"}
{"epoch": 0.7853403141361257, "step": 376, "batch_size": 128, "mean": 55.088905334472656, "std": 101.54574584960938, "min": -182.64166259765625, "p10": -67.27728271484374, "median": 39.991676330566406, "p90": 183.28978881835937, "max": 286.430908203125, "pos_frac": 0.703125, "sample": [99.48721313476562, 66.55899047851562, 183.60565185546875, 286.430908203125, 136.1424102783203, 237.3822021484375, 100.18756103515625, 46.317562103271484, 55.65803527832031, -134.7742462158203, 181.72406005859375, -25.424118041992188, -155.62905883789062, 32.470184326171875, -24.01123046875, 149.77847290039062, 3.1657333374023438, 45.30029296875, 143.44378662109375, 185.76416015625, 125.36981201171875, -163.68936157226562, -1.0676803588867188, 7.470367431640625, 75.28945922851562, 25.5980224609375, -32.02342224121094, 200.62387084960938, -3.7577667236328125, 168.93695068359375, 166.48495483398438, -4.7845001220703125, 183.869384765625, 41.0123291015625, -18.44232177734375, -19.889984130859375, 183.1544189453125, 39.87413024902344, 178.28424072265625, 12.170867919921875, 0.0, 20.780960083007812, 20.825485229492188, 136.94091796875, -48.021202087402344, 205.25015258789062, -120.10406494140625, 157.15325927734375, 120.8135986328125, 146.70681762695312, -159.71981811523438, 134.384521484375, 45.68792724609375, 10.524658203125, 165.99951171875, 9.56634521484375, 29.24749755859375, 116.30516052246094, -78.554443359375, -11.56744384765625, 142.74476623535156, -1.97442626953125, 231.41268920898438, 218.48101806640625, 20.64306640625, 195.57513427734375, 173.07598876953125, 41.51721954345703, -60.4593505859375, -28.33612060546875, 83.61895751953125, -62.4442138671875, 26.409271240234375, -41.55846405029297, -6.110353469848633, 1.381622314453125, -139.24378967285156, 33.54785919189453, -82.15310668945312, 187.7493896484375, 128.2668914794922, 56.020843505859375, 177.36972045898438, 115.86663818359375, 171.4898681640625, 142.5350799560547, 7.662456512451172, -182.64166259765625, 10.343841552734375, -120.56866455078125, 50.260650634765625, 2.0112571716308594, 157.38064575195312, -33.616397857666016, 9.081748962402344, 146.9574737548828, 104.533447265625, -14.45452880859375, -133.4813232421875, 170.8720703125, 30.657379150390625, 167.5968017578125, -108.05044555664062, 155.0585174560547, 10.557281494140625, 24.66408348083496, 114.2935562133789, 28.945785522460938, 70.31277465820312, 155.85455322265625, -18.563690185546875, -42.051361083984375, -45.23599624633789, 40.109222412109375, 129.69125366210938, -6.130424499511719, 35.43511962890625, -38.255714416503906, 55.739837646484375, -3.4795761108398438, 5.098609924316406, 85.28398132324219, 153.03482055664062, 37.54327392578125, 196.21099853515625, -149.00640869140625, 207.12124633789062, 178.8955078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000376.npy"}
{"epoch": 0.787434554973822, "step": 377, "batch_size": 128, "mean": 51.888980865478516, "std": 92.47039031982422, "min": -169.0107421875, "p10": -37.235357666015624, "median": 29.1298885345459, "p90": 184.18322143554684, "max": 236.6502685546875, "pos_frac": 0.71875, "sample": [194.64105224609375, 100.71615600585938, 158.5745849609375, -14.934417724609375, 98.73599243164062, 9.75372314453125, 5.88067626953125, 108.11700439453125, 156.68797302246094, 17.902130126953125, 17.986129760742188, 60.9825439453125, 146.56484985351562, -69.42214965820312, 118.44180297851562, 155.8458709716797, 201.58087158203125, -33.3448486328125, 0.35448455810546875, -36.973480224609375, 150.6639404296875, 2.4889068603515625, 0.5353851318359375, 29.82235336303711, -13.68220329284668, -13.541358947753906, 74.84176635742188, -10.274261474609375, 182.0252227783203, 36.62689208984375, 126.40444946289062, 57.004119873046875, 99.67692565917969, -125.64422607421875, -9.818527221679688, 168.9886474609375, -3.3909912109375, -4.8739013671875, 10.36920166015625, 44.1715087890625, 15.3094482421875, 132.42367553710938, 12.610176086425781, -69.96484375, 148.43748474121094, 8.75775146484375, 201.726806640625, 68.49163818359375, 13.992111206054688, -131.19668579101562, 11.19464111328125, 217.13839721679688, -5.290252685546875, 0.464935302734375, 190.010009765625, 6.241357803344727, -95.81571960449219, -0.5557098388671875, 158.13186645507812, 156.48931884765625, 171.83383178710938, 25.2349853515625, 81.1363525390625, 126.8695068359375, -20.847015380859375, 159.38360595703125, 33.14695739746094, 0.0, -37.846405029296875, 30.729202270507812, 156.10043334960938, -18.945877075195312, 134.5287322998047, 157.41522216796875, -8.719093322753906, 62.582794189453125, 9.687675476074219, -36.048370361328125, -0.6724948883056641, 155.1810302734375, 21.5411376953125, 224.3240966796875, 8.96429443359375, -26.2813720703125, -16.57513427734375, -135.92037963867188, 24.1405029296875, 4.672050476074219, 180.98263549804688, 46.41455078125, 71.150146484375, 33.4327392578125, 8.579063415527344, 171.46282958984375, -89.89776611328125, 194.83489990234375, 10.309829711914062, 25.06036376953125, 46.80718994140625, 145.66082763671875, 33.163818359375, -137.09393310546875, -11.368980407714844, 59.657623291015625, 206.1568603515625, -92.83540344238281, 212.92001342773438, 31.892364501953125, 65.52951049804688, 158.62799072265625, 31.688461303710938, 11.100112915039062, 28.437423706054688, 3.292724609375, 198.87429809570312, 153.12142944335938, 189.2185516357422, -143.992431640625, -16.652130126953125, 236.6502685546875, 19.62236785888672, 38.506439208984375, -127.36259460449219, 196.31048583984375, -9.0220947265625, -10.427825927734375, -169.0107421875, 115.29043579101562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000377.npy"}
{"epoch": 0.7895287958115184, "step": 378, "batch_size": 128, "mean": 51.65689468383789, "std": 93.63937377929688, "min": -179.50787353515625, "p10": -44.89685592651367, "median": 26.558456420898438, "p90": 185.37154541015624, "max": 229.48443603515625, "pos_frac": 0.703125, "sample": [149.92636108398438, 2.16119384765625, 27.548477172851562, 140.88494873046875, -27.37298583984375, 2.7987918853759766, -30.5565185546875, 9.5965576171875, 198.3924560546875, 64.2841796875, -3.452411651611328, 6.587890625, 30.165069580078125, 147.62625122070312, 149.49127197265625, -59.07362365722656, -179.50787353515625, 189.661376953125, 205.9793701171875, -35.3419189453125, 33.826446533203125, 27.05865478515625, 14.084579467773438, -45.75684356689453, 52.295257568359375, 199.96630859375, 229.48443603515625, 159.80532836914062, 24.96685791015625, 115.2127685546875, 4.640167236328125, 163.838623046875, 117.89036560058594, 20.99163818359375, 11.31304931640625, -155.22592163085938, 3.9228763580322266, 2.960409164428711, 72.40945434570312, 93.97653198242188, 0.0, 172.20086669921875, -5.455596923828125, 165.70266723632812, 202.33993530273438, -48.4031982421875, -13.22481918334961, 214.62628173828125, -16.465335845947266, 201.9371337890625, 122.49298095703125, 129.50953674316406, -16.7047119140625, 2.348876953125, 8.337860107421875, 11.481674194335938, -33.9781494140625, 144.19158935546875, -21.936920166015625, 162.93478393554688, 55.21868896484375, -0.24798965454101562, 146.1627960205078, 159.05905151367188, 38.136016845703125, 150.50860595703125, -44.528289794921875, -80.84222412109375, 185.80255126953125, 114.17428588867188, 128.79855346679688, 26.058258056640625, -0.87432861328125, -20.548995971679688, -138.3079071044922, 185.18682861328125, -86.14424133300781, -118.74896240234375, 32.86224365234375, -2.06396484375, 39.815032958984375, 6.4766082763671875, 80.037109375, 74.99113464355469, 12.3536376953125, 117.76364135742188, 197.451416015625, 9.134124755859375, 20.117828369140625, 112.06405639648438, 175.15869140625, 173.3476104736328, -26.21331787109375, 15.287979125976562, 24.836578369140625, -18.426040649414062, -74.4114990234375, 7.84027099609375, 12.9053955078125, -19.675872802734375, -111.22781372070312, 192.92657470703125, -6.012725830078125, 80.26718139648438, -155.4492645263672, -6.0004425048828125, 28.4947509765625, 140.0430908203125, -16.3485107421875, 37.301422119140625, 66.07635498046875, 51.847137451171875, 0.9848175048828125, -10.741867065429688, 214.33258056640625, 161.77230834960938, 182.58880615234375, 183.41761779785156, 193.50872802734375, 18.216018676757812, -161.78115844726562, 122.35791778564453, 148.78387451171875, 53.35516357421875, 30.270156860351562, -37.132843017578125, 25.97344970703125, -5.65191650390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000378.npy"}
{"epoch": 0.7916230366492146, "step": 379, "batch_size": 128, "mean": 57.36481475830078, "std": 102.1296157836914, "min": -190.74090576171875, "p10": -53.63761291503905, "median": 34.02076721191406, "p90": 185.76500091552734, "max": 223.4564208984375, "pos_frac": 0.7421875, "sample": [-14.551002502441406, 9.85052490234375, 1.4016876220703125, 8.116851806640625, -3.9452438354492188, -28.63763427734375, 4.543251037597656, 163.35903930664062, 50.547027587890625, 1.9949493408203125, 148.18441772460938, 174.05728149414062, 54.023223876953125, 91.06283569335938, 14.023529052734375, 29.746780395507812, 118.95594787597656, -136.216552734375, 205.5184326171875, 38.09527587890625, 186.25106811523438, 91.36962890625, 167.61399841308594, -34.04254150390625, 12.032363891601562, -85.55717468261719, 167.24342346191406, 4.1024169921875, 79.51287841796875, -10.374359130859375, 34.870941162109375, 183.8541717529297, 185.06478881835938, 55.689544677734375, 44.13755798339844, 10.13958740234375, 79.85487365722656, 198.71844482421875, 126.86419677734375, 14.147331237792969, -3.33270263671875, -130.65188598632812, 48.97216796875, 15.305007934570312, 5.93017578125, 213.88531494140625, 11.035362243652344, -60.005096435546875, -16.2657470703125, 7.372589111328125, 153.42996215820312, 3.99800968170166, 33.17059326171875, -176.83517456054688, -10.850128173828125, 8.103118896484375, 180.224609375, 159.314208984375, 138.4456787109375, 142.7962646484375, 125.83312225341797, -115.68985748291016, 140.9716033935547, 100.94053649902344, 160.31741333007812, 221.97344970703125, -1.07275390625, 188.30511474609375, 6.7566070556640625, -47.79345703125, -49.18891143798828, 0.0, -11.6263427734375, 146.1548309326172, 21.07244873046875, 194.73715209960938, 61.867828369140625, 170.09442138671875, 185.46408081054688, 185.5566864013672, 11.30657958984375, 138.99874877929688, 0.0472412109375, -4.247535705566406, 164.00222778320312, 169.89404296875, -20.838050842285156, 137.64366149902344, -20.553054809570312, 165.39614868164062, 125.76643371582031, 2.4744033813476562, 195.57546997070312, 2.770231246948242, 133.72653198242188, -13.15130615234375, 2.314300537109375, 149.51748657226562, -140.22915649414062, 223.4564208984375, 218.64450073242188, -190.74090576171875, 19.064056396484375, 16.800338745117188, 177.82049560546875, 87.34628295898438, 66.71359252929688, -50.90869140625, 186.40652465820312, 187.97512817382812, 196.73046875, 126.23788452148438, 179.43814086914062, 85.70201110839844, -130.72698974609375, -155.50633239746094, 17.147010803222656, 0.5999393463134766, 5.375030517578125, 183.91580200195312, -40.57868957519531, 132.91738891601562, -167.09097290039062, -95.65579223632812, -165.572509765625, 15.009233474731445, -21.308074951171875, 182.73046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000379.npy"}
{"epoch": 0.793717277486911, "step": 380, "batch_size": 128, "mean": 53.50006866455078, "std": 96.56365966796875, "min": -162.33383178710938, "p10": -46.66927795410155, "median": 29.166934967041016, "p90": 188.27547607421872, "max": 270.16534423828125, "pos_frac": 0.7265625, "sample": [180.97329711914062, 52.482208251953125, 76.08857727050781, 20.697463989257812, 156.39154052734375, 17.754547119140625, -157.1595916748047, 3.45501708984375, 1.2183952331542969, -52.205596923828125, 124.20639038085938, 82.26824951171875, 162.21868896484375, -139.44302368164062, 20.18310546875, 16.82604217529297, 127.99240112304688, 44.34161376953125, 158.1588134765625, -3.2091331481933594, 0.0, 47.568115234375, 9.142913818359375, 27.196044921875, 27.414398193359375, 173.42770385742188, 8.284576416015625, 34.39947509765625, -1.8881988525390625, 9.697723388671875, 53.15631103515625, 223.90240478515625, 15.365325927734375, 148.03665161132812, -3.752593994140625, -162.33383178710938, 164.77703857421875, -8.591312408447266, 16.396682739257812, -20.74266815185547, 46.7999267578125, 65.87916564941406, 181.99447631835938, -44.29656982421875, -18.67481231689453, 21.1842041015625, 19.031295776367188, 194.8553466796875, 193.16180419921875, 9.399261474609375, -99.46832275390625, 119.3983154296875, 49.84565734863281, -7.024688720703125, 25.90533447265625, -91.2420654296875, -3.9091758728027344, 16.98040771484375, 194.92733764648438, 165.30972290039062, -88.98223876953125, 177.286376953125, 69.83071899414062, 31.356842041015625, 211.41094970703125, 176.81719970703125, -131.96438598632812, -8.946487426757812, 19.137481689453125, 39.77131271362305, 149.40158081054688, 198.5675048828125, 1.90191650390625, 135.84634399414062, 169.19857788085938, -8.121170043945312, 108.43682861328125, -42.0738525390625, 35.022308349609375, -152.9809112548828, 134.5166473388672, 270.16534423828125, 1.9817886352539062, 154.45582580566406, 179.43609619140625, 28.412208557128906, 161.29556274414062, 186.18133544921875, 180.73147583007812, 60.921234130859375, 37.406097412109375, 232.49581909179688, -71.05078125, 64.1895751953125, -0.06971359252929688, 70.56037902832031, 5.22845458984375, -43.4429931640625, 199.3204345703125, 265.9486999511719, 33.76988220214844, -10.581039428710938, 6.9602508544921875, 172.695068359375, 70.78114318847656, 18.25896453857422, 29.921661376953125, 48.6072998046875, -31.37799072265625, -87.54071044921875, -30.976806640625, -122.85903930664062, 9.703369140625, 5.12890625, 27.352157592773438, 127.01480102539062, -22.140228271484375, 46.3431396484375, 118.57442474365234, 242.03582763671875, 33.820068359375, -17.3751220703125, -43.353851318359375, 198.0321044921875, 1.87506103515625, -86.62872314453125, 219.64053344726562, -14.024772644042969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000380.npy"}
{"epoch": 0.7958115183246073, "step": 381, "batch_size": 128, "mean": 67.25062561035156, "std": 89.68463134765625, "min": -120.1859359741211, "p10": -34.5875305175781, "median": 51.80934524536133, "p90": 184.38140716552735, "max": 205.406982421875, "pos_frac": 0.71875, "sample": [148.53326416015625, 201.522216796875, 187.69073486328125, 82.19113159179688, 201.15003967285156, -63.0078125, 184.9950714111328, 197.79721069335938, 141.78271484375, 5.865447998046875, 197.86456298828125, 19.495697021484375, 17.102081298828125, 159.567138671875, 150.00689697265625, 172.4854736328125, 0.0, 11.301239013671875, 125.27932739257812, 154.78863525390625, 127.4423828125, 66.65411376953125, 11.119354248046875, 204.08926391601562, 50.355804443359375, 77.08949279785156, 146.94583129882812, -94.05751037597656, 0.0, -15.5374755859375, 53.04302978515625, 153.8931884765625, -10.816650390625, 161.2539825439453, -9.018671035766602, 57.22222900390625, 15.021270751953125, 128.44805908203125, 11.65155029296875, 125.13861083984375, 203.7523193359375, 98.62147521972656, -18.286407470703125, 150.0360870361328, 154.01580810546875, 8.3095703125, 181.93017578125, 0.7009353637695312, -83.56169128417969, 165.3194580078125, 164.45533752441406, 177.63494873046875, 46.21832275390625, 0.9875946044921875, 205.406982421875, 184.118408203125, 13.16583251953125, 156.61424255371094, 182.17254638671875, 76.6470718383789, -53.999725341796875, -15.6973876953125, 79.20091247558594, 9.7298583984375, -7.78314208984375, 84.88215637207031, 169.17044067382812, -4.04437255859375, -28.492156982421875, -111.37588500976562, 50.575660705566406, -1.5297565460205078, -0.723480224609375, 168.4398193359375, -23.13629150390625, 41.42755126953125, -1.531158447265625, 121.17840576171875, 120.26724243164062, -106.09187316894531, 177.0576171875, -27.597923278808594, 162.85446166992188, 48.248046875, 195.51324462890625, -2.9139404296875, -26.812255859375, -10.277076721191406, 173.4491424560547, 13.276458740234375, -11.789337158203125, -56.391021728515625, 151.46527099609375, 138.3753204345703, 133.38726806640625, -51.430755615234375, 3.1199607849121094, 9.0203857421875, 53.306488037109375, -120.1859359741211, 121.76942443847656, 8.936836242675781, -47.5953369140625, 112.46517181396484, -112.13363647460938, 128.9988250732422, -20.5985107421875, 7.947181701660156, -12.044479370117188, 153.99732971191406, -8.745208740234375, 194.70704650878906, -87.592529296875, 201.49435424804688, 15.82611083984375, 43.52954864501953, 64.32299041748047, 11.354621887207031, 145.50823974609375, 19.283493041992188, 165.75564575195312, 173.4388427734375, 36.011253356933594, -29.01275634765625, -58.323822021484375, 187.44900512695312, 22.11431884765625, 102.46664428710938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000381.npy"}
{"epoch": 0.7979057591623037, "step": 382, "batch_size": 128, "mean": 77.99198913574219, "std": 99.71882629394531, "min": -152.60995483398438, "p10": -37.80021057128905, "median": 97.65696716308594, "p90": 188.17174987792967, "max": 226.769775390625, "pos_frac": 0.765625, "sample": [-132.7776641845703, 9.882888793945312, 60.108856201171875, 114.53152465820312, 121.5587158203125, 171.3701171875, 187.13250732421875, -14.9498291015625, 20.295560836791992, 143.3956298828125, 149.3895263671875, -147.39614868164062, 126.52667236328125, 14.979209899902344, 191.1588134765625, 117.51959228515625, 36.355926513671875, 57.37968444824219, 165.390869140625, -14.127876281738281, 52.319580078125, -144.3438262939453, 177.5043182373047, 6.84454345703125, 6.7722320556640625, 200.77423095703125, 25.175186157226562, -34.9105224609375, 151.74539184570312, 123.35811614990234, -12.595458984375, 151.7159423828125, 139.58334350585938, 35.271644592285156, 170.15997314453125, 101.4786605834961, -18.88933563232422, 113.34416961669922, -6.234419822692871, 184.2659912109375, 14.145835876464844, 9.713623046875, 182.96371459960938, 25.8345947265625, 176.8994598388672, 180.1119384765625, 196.80296325683594, 48.743804931640625, 29.231597900390625, 12.378402709960938, 186.31295776367188, 155.50059509277344, 6.2394256591796875, 82.8544921875, -2.90069580078125, 130.3006591796875, 226.31854248046875, 211.838623046875, -107.13836669921875, -12.92669677734375, -44.542816162109375, -146.95571899414062, 226.769775390625, 67.04530334472656, 5.243072509765625, 131.08279418945312, 216.3472900390625, 76.29049682617188, 204.79638671875, 29.692779541015625, 208.55850219726562, 26.820907592773438, 81.92768859863281, -1.365692138671875, 186.04055786132812, 124.20718383789062, 150.28506469726562, 21.371688842773438, 150.54913330078125, 179.9698486328125, 32.5235595703125, -31.03228759765625, 59.539093017578125, 157.2694091796875, -141.96762084960938, -4.937896728515625, -93.041748046875, 174.774658203125, -20.720748901367188, 184.41160583496094, 190.59664916992188, -19.82952880859375, 11.928375244140625, 197.22996520996094, 183.085693359375, 96.53742980957031, 138.33554077148438, -116.63980102539062, 7.0263824462890625, -14.988494873046875, 20.106613159179688, 80.5364990234375, 98.77650451660156, 180.15318298339844, 144.21170043945312, -75.32412719726562, 43.401519775390625, 165.6538848876953, -132.43978881835938, 154.1712188720703, 164.5333251953125, -14.170166015625, 156.10903930664062, 134.45175170898438, 114.36038208007812, 150.28871154785156, -152.60995483398438, 146.2950439453125, -15.3594970703125, 137.19996643066406, 181.10842895507812, -22.276416778564453, 223.66363525390625, 160.7863006591797, 194.73373413085938, 173.20086669921875, 186.4810333251953, -53.5924072265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000382.npy"}
{"epoch": 0.8, "step": 383, "batch_size": 128, "mean": 57.45216751098633, "std": 96.02400207519531, "min": -167.920166015625, "p10": -64.91501235961914, "median": 35.147483825683594, "p90": 182.38101196289062, "max": 234.3739013671875, "pos_frac": 0.7265625, "sample": [-8.52783203125, 38.77374267578125, 160.20687866210938, 151.7044677734375, -88.87804412841797, 26.418243408203125, -65.8515625, 21.240936279296875, -1.4942169189453125, 107.95223999023438, 148.4674072265625, -144.50355529785156, 155.6136474609375, 30.873046875, 0.0, -14.045402526855469, 24.04449462890625, -62.24359130859375, 52.351318359375, -84.84249877929688, 21.2845458984375, 190.8399658203125, 167.4412078857422, 144.51031494140625, -134.097412109375, 191.52125549316406, -111.396728515625, -5.864307403564453, -61.84918212890625, 10.939682006835938, 85.59486389160156, 4.9140625, -15.770401000976562, -45.22832489013672, 169.25506591796875, -54.0076904296875, 206.793212890625, 200.4779052734375, 36.89649963378906, 159.76292419433594, 127.4533920288086, 2.5128707885742188, 162.05108642578125, 227.19564819335938, -128.2353973388672, -0.599578857421875, 45.95597839355469, 178.95330810546875, 49.808349609375, 8.658538818359375, 180.872314453125, 1.581085205078125, -7.862642288208008, 160.86611938476562, 103.234375, 23.1334228515625, 5.5959625244140625, -9.230842590332031, -0.034481048583984375, -146.22122192382812, 146.6820526123047, 178.96978759765625, 12.750175476074219, 185.90130615234375, -26.9124755859375, 11.61187744140625, 176.19036865234375, 2.0890960693359375, 6.664794921875, 46.21807861328125, 130.843505859375, 192.0476531982422, 145.40032958984375, 150.0770263671875, 60.588897705078125, 37.379329681396484, 2.933837890625, -3.83795166015625, 39.47093200683594, 156.82981872558594, 3.0912322998046875, 131.21823120117188, 141.1071014404297, 51.31550598144531, 148.65545654296875, 33.398468017578125, 29.29229736328125, 23.229846954345703, 122.89891052246094, -167.920166015625, 189.169677734375, 20.65027618408203, 61.94023132324219, 54.81366729736328, 21.402587890625, 5.614633560180664, -68.73583984375, 207.64788818359375, 153.24415588378906, 149.05776977539062, 153.35604858398438, -83.0609130859375, 234.3739013671875, -116.12019348144531, -109.94122314453125, 25.097023010253906, 101.14176940917969, 19.852630615234375, 0.0, 142.07138061523438, 165.26312255859375, -3.4242286682128906, 59.71112060546875, 145.87774658203125, 115.488037109375, -28.363296508789062, -46.873870849609375, 16.643516540527344, 192.71588134765625, 129.37603759765625, 7.2092437744140625, 157.69326782226562, 213.54525756835938, -64.51363372802734, 20.07525634765625, -2.8754119873046875, 187.7320556640625, 163.87158203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000383.npy"}
{"epoch": 0.8020942408376963, "step": 384, "batch_size": 128, "mean": 39.780609130859375, "std": 101.77757263183594, "min": -184.07763671875, "p10": -106.46401138305664, "median": 22.17156219482422, "p90": 175.9410217285156, "max": 231.75714111328125, "pos_frac": 0.6328125, "sample": [208.61639404296875, -184.07763671875, 125.21438598632812, -23.4193115234375, -24.272003173828125, 163.4366455078125, 41.92681884765625, 231.75714111328125, 28.457107543945312, -22.836563110351562, 165.3489227294922, -19.248973846435547, -8.231475830078125, -88.67214965820312, 36.445220947265625, -131.00900268554688, 167.1575927734375, -3.405132293701172, 194.9830322265625, -22.95843505859375, 4.998659133911133, 190.22085571289062, 144.18051147460938, 150.41018676757812, -25.233047485351562, 172.44580078125, 24.726776123046875, 155.6654052734375, 18.1893310546875, -106.26263427734375, 3.8935546875, -92.33282470703125, -7.4869384765625, 97.00244140625, -25.0345458984375, 141.63380432128906, 25.5155029296875, 37.75311279296875, 163.83665466308594, -14.134605407714844, -2.884490966796875, -138.99874877929688, -1.3516006469726562, 13.305641174316406, 151.27145385742188, 163.76547241210938, 163.14474487304688, -176.1468963623047, 3.1597900390625, 21.234893798828125, 130.32522583007812, 198.85140991210938, -121.61538696289062, -7.56561279296875, -162.75021362304688, 38.44248962402344, 108.1947021484375, 112.58514404296875, -136.4275665283203, -51.319915771484375, 4.22784423828125, -139.47738647460938, 138.510498046875, 156.93014526367188, 183.69528198242188, 48.917572021484375, 78.37677001953125, 56.47149658203125, 111.16302490234375, -7.002864837646484, -10.981670379638672, 126.54129028320312, 174.69189453125, 13.782608032226562, 146.70840454101562, 178.85565185546875, -9.9859619140625, 19.05999755859375, -18.052684783935547, 88.985595703125, 23.108230590820312, -146.09666442871094, 140.5743865966797, 95.7923583984375, 161.26907348632812, 223.59774780273438, 26.524169921875, 0.675079345703125, 96.20979309082031, 40.8822021484375, 191.27374267578125, -12.707405090332031, 161.80813598632812, 62.515472412109375, -6.7142333984375, 2.11920166015625, -61.593505859375, 189.03529357910156, 8.206817626953125, -12.46535873413086, -4.9040374755859375, -49.06982421875, 52.46978759765625, -136.40435791015625, -160.10858154296875, 27.797197341918945, 0.4176025390625, 29.256488800048828, 12.142473220825195, -7.287384033203125, 203.7103271484375, -154.35552978515625, 56.82415771484375, -3.504241943359375, -4.734853744506836, -4.1536865234375, 192.73147583007812, 179.82745361328125, -106.93389129638672, 141.57373046875, 150.84368896484375, -87.94944763183594, 3.30291748046875, 17.894271850585938, 38.758331298828125, 67.492431640625, 12.365478515625, -101.93675231933594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000384.npy"}
{"epoch": 0.8041884816753927, "step": 385, "batch_size": 128, "mean": 54.83083724975586, "std": 96.25867462158203, "min": -187.49530029296875, "p10": -48.970535278320305, "median": 31.141464233398438, "p90": 181.35421752929688, "max": 243.15805053710938, "pos_frac": 0.734375, "sample": [-42.110748291015625, 86.22607421875, 179.18807983398438, 87.07080078125, 37.0340576171875, 27.191162109375, 171.07440185546875, 30.408843994140625, 181.50738525390625, 157.86773681640625, 30.267303466796875, 0.364715576171875, -54.096954345703125, 22.728515625, 24.514968872070312, 116.21949768066406, 33.58050537109375, 137.87478637695312, -79.67057037353516, -2.719970703125, 181.28857421875, -46.77349853515625, 168.27706909179688, -137.80613708496094, 7.970947265625, 176.00755310058594, 146.54901123046875, 11.159965515136719, 143.5491943359375, 24.119781494140625, 7.9148101806640625, 218.30813598632812, 147.26324462890625, 31.87408447265625, 190.10830688476562, 240.34295654296875, 171.91934204101562, 18.717819213867188, 118.65433502197266, 234.79135131835938, 75.29827880859375, -33.91700744628906, 56.723419189453125, 121.80152893066406, 94.31109619140625, -134.1375274658203, 151.6811981201172, 163.27430725097656, -149.3431396484375, -9.995819091796875, 243.15805053710938, 188.51995849609375, 231.82296752929688, -9.94244384765625, 140.46128845214844, 25.560882568359375, 158.61453247070312, 0.6914596557617188, -178.66302490234375, 102.68767547607422, 15.922271728515625, 199.799072265625, -11.8543701171875, 174.815185546875, 6.0997314453125, 2.654388427734375, -29.719482421875, 149.43368530273438, -2.681640625, 61.478424072265625, 106.00039672851562, 3.7858505249023438, -35.73480224609375, 11.272293090820312, 20.371810913085938, -22.604965209960938, 131.56448364257812, 152.46337890625, 151.43576049804688, -109.15966796875, 11.668853759765625, 132.130126953125, 155.4776611328125, 93.10893249511719, 150.68106079101562, 10.106201171875, -24.270034790039062, 27.461898803710938, -187.49530029296875, 2.784759521484375, 64.79217529296875, -5.937541961669922, -1.842864990234375, 29.336219787597656, 190.67568969726562, 52.97052001953125, 0.0, 41.93635559082031, 73.59280395507812, 20.108928680419922, 6.231178283691406, 11.395416259765625, 49.722381591796875, -37.522216796875, 202.52084350585938, -20.712486267089844, -115.86248779296875, 129.01968383789062, -33.39628601074219, 205.45465087890625, 45.6923828125, 0.0, -75.71157836914062, -117.7884292602539, 144.06192016601562, 3.81353759765625, 45.812835693359375, -134.05810546875, -5.447357177734375, -13.258209228515625, -84.26399230957031, 14.556678771972656, 102.26902770996094, 38.74212646484375, 134.885986328125, 72.92855834960938, 183.520751953125, 15.7470703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000385.npy"}
{"epoch": 0.806282722513089, "step": 386, "batch_size": 128, "mean": 69.59428405761719, "std": 100.6490478515625, "min": -251.66976928710938, "p10": -24.950372314453123, "median": 46.280548095703125, "p90": 193.11214294433591, "max": 264.31719970703125, "pos_frac": 0.7734375, "sample": [149.01412963867188, -5.4191741943359375, 4.829551696777344, 143.15798950195312, 70.49822998046875, 2.24041748046875, 72.66748046875, 189.02890014648438, 135.10470581054688, 163.6364288330078, 154.04954528808594, 264.31719970703125, -40.91558837890625, 35.4029541015625, 208.8807373046875, 38.123565673828125, 19.098876953125, 29.353485107421875, 4.0297088623046875, 126.5428466796875, 137.737060546875, 181.48147583007812, 26.918670654296875, 60.18994140625, -15.11767578125, -157.17996215820312, 39.0888671875, -23.8519287109375, 170.38111877441406, 103.30038452148438, 32.99139404296875, -134.04576110839844, 205.55953979492188, 31.5533447265625, -20.418426513671875, 14.3260498046875, 15.770721435546875, -32.25654602050781, 53.400146484375, 181.2130126953125, 245.953125, 192.15350341796875, 75.88601684570312, 40.803741455078125, 61.262054443359375, 150.97439575195312, 205.2283935546875, -91.0994873046875, 76.69526672363281, 219.8056640625, 87.20501708984375, -25.9344482421875, -113.5606689453125, -38.662025451660156, 213.62460327148438, 157.83889770507812, -10.469894409179688, 0.0, -118.35952758789062, -14.367988586425781, -156.7578125, -112.319580078125, 133.61529541015625, 32.258148193359375, 1.8456249237060547, -7.5809326171875, 53.89215087890625, -1.6679954528808594, 169.11524963378906, -195.43902587890625, 200.16708374023438, 1.0885009765625, 160.17398071289062, 216.22097778320312, 214.425537109375, 170.13681030273438, 8.050163269042969, -2.785125732421875, 176.78855895996094, 158.8616485595703, 152.5181427001953, 11.961181640625, -251.66976928710938, 46.01373291015625, -3.09393310546875, 182.90777587890625, 162.05947875976562, 13.53118896484375, 153.1480712890625, 2.555450439453125, 51.467620849609375, 20.251476287841797, 0.55487060546875, 0.0, 214.22494506835938, 148.78892517089844, -10.26318359375, 6.112579345703125, 92.8984375, 166.29965209960938, 180.21112060546875, 180.9291534423828, 46.54736328125, 178.06211853027344, 14.799407958984375, 190.81246948242188, 34.2913818359375, -24.52862548828125, -1.29742431640625, 165.766845703125, 195.34896850585938, 45.5572509765625, 203.79576110839844, 118.99667358398438, 4.518798828125, 25.97259521484375, 33.304046630859375, 14.442352294921875, 97.8623275756836, 158.91659545898438, 34.24148941040039, 189.07705688476562, 14.820976257324219, 10.982597351074219, 175.83978271484375, 0.0, 157.93301391601562, 130.84844970703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000386.npy"}
{"epoch": 0.8083769633507853, "step": 387, "batch_size": 128, "mean": 64.55142211914062, "std": 99.73711395263672, "min": -212.00576782226562, "p10": -41.19287414550781, "median": 45.70142364501953, "p90": 193.46141052246094, "max": 231.97897338867188, "pos_frac": 0.6953125, "sample": [98.11571502685547, 36.44981384277344, 85.25042724609375, 11.681495666503906, 7.48333740234375, 192.37005615234375, 213.17747497558594, -91.62655639648438, -163.08468627929688, -36.579864501953125, 0.0, 205.88418579101562, -10.748870849609375, 219.212890625, 153.43661499023438, 64.5982666015625, 4.954246520996094, -10.725341796875, -29.118881225585938, 134.926025390625, -54.444976806640625, 31.831993103027344, 168.50271606445312, -43.163482666015625, 114.50704193115234, 177.3951416015625, -109.74519348144531, -14.734100341796875, 176.27496337890625, 1.8977985382080078, 168.9359130859375, -83.63067626953125, 175.10757446289062, 33.846435546875, 168.83306884765625, 166.37326049804688, -16.04754638671875, 95.02616882324219, 69.64632415771484, -8.904144287109375, -40.34832763671875, 149.3935089111328, -5.9524993896484375, -32.764373779296875, 196.00790405273438, 46.01484680175781, 213.3038330078125, 128.35177612304688, -38.68567657470703, 1.1355438232421875, -9.516502380371094, 198.80767822265625, -44.3084716796875, -14.537322998046875, 57.30010986328125, 22.6395263671875, 14.039520263671875, -138.74879455566406, 156.3251953125, 175.7574005126953, 139.6269989013672, 37.2772216796875, 68.35601806640625, 29.576416015625, 149.79299926757812, -130.72921752929688, 33.9261474609375, 35.379119873046875, 7.752716064453125, -12.875511169433594, -19.693130493164062, -22.40911865234375, 49.91900634765625, 208.37973022460938, 162.21478271484375, 144.97048950195312, 167.04139709472656, 41.384796142578125, -8.627761840820312, 132.43043518066406, 12.309707641601562, -5.026477813720703, 186.44219970703125, 26.453582763671875, 69.57260131835938, 191.5428466796875, 143.32373046875, 3.235015869140625, 50.22174072265625, 153.94589233398438, -1.7461395263671875, 182.61907958984375, 63.86488342285156, 21.974609375, 217.26513671875, 86.7586669921875, 161.80154418945312, 12.64678955078125, 164.60687255859375, 175.8703155517578, 6.689815521240234, 45.38800048828125, -28.267120361328125, 158.01226806640625, -5.643991470336914, -66.72891235351562, -212.00576782226562, -1.2699623107910156, 154.9657745361328, 231.97897338867188, -13.74725341796875, -19.571823120117188, 88.88793182373047, 119.96658325195312, 209.31524658203125, 154.2273712158203, 174.75425720214844, 47.26617431640625, 221.73760986328125, 13.552253723144531, -85.35409545898438, -163.12969970703125, 228.28268432617188, 209.593994140625, 2.2464675903320312, 171.03955078125, -5.842044830322266, 157.48291015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000387.npy"}
{"epoch": 0.8104712041884817, "step": 388, "batch_size": 128, "mean": 50.40336608886719, "std": 103.86323547363281, "min": -189.3551025390625, "p10": -67.90598907470702, "median": 20.303525924682617, "p90": 181.9354766845703, "max": 317.58935546875, "pos_frac": 0.6875, "sample": [126.0667724609375, -127.95266723632812, 169.59475708007812, 0.0, 317.58935546875, 172.5514373779297, -5.044403076171875, 203.4937286376953, 55.277618408203125, 174.64297485351562, -167.65252685546875, 157.73703002929688, -135.35247802734375, 4.671592712402344, 6.9840240478515625, 28.24139404296875, 54.98078155517578, 137.8201904296875, 218.95745849609375, -22.341094970703125, 91.59002685546875, 216.83807373046875, -164.70376586914062, -24.5269775390625, -12.683074951171875, -47.98100280761719, 115.1922836303711, 143.29000854492188, 180.87210083007812, 187.74984741210938, 3.58905029296875, 17.31238555908203, 0.0, 15.688507080078125, 19.721908569335938, -108.64117431640625, -22.89520263671875, 13.80816650390625, 130.28173828125, 131.55499267578125, -74.15086364746094, 33.19347381591797, 141.9206085205078, 154.94375610351562, -30.459426879882812, -65.2296142578125, 184.41668701171875, -20.82616424560547, -8.053264617919922, 6.2169342041015625, 118.63731384277344, 0.93133544921875, 194.82635498046875, -37.84382629394531, 13.691299438476562, 146.23483276367188, -3.4915924072265625, 191.73153686523438, 107.251953125, -32.572509765625, -24.43084716796875, 152.35302734375, 155.71658325195312, 3.1974334716796875, 21.440195083618164, -56.83514404296875, 86.81576538085938, -183.22503662109375, 197.68141174316406, -127.83543395996094, 69.35775756835938, 172.31805419921875, 149.74758911132812, 109.53113555908203, 39.064483642578125, 1.9989013671875, -0.74658203125, 174.41796875, 11.776771545410156, -150.46900939941406, -14.95361328125, 201.5599365234375, 169.1966094970703, -17.958282470703125, 19.48147964477539, 11.077285766601562, 28.02593994140625, 154.3875732421875, 134.45779418945312, 20.032503128051758, -146.28488159179688, 35.90785217285156, -3.1580810546875, -189.3551025390625, -9.325298309326172, 199.0469207763672, 175.3939208984375, 158.96920776367188, 20.574548721313477, 176.0081787109375, 4.574729919433594, 135.08377075195312, 178.90155029296875, 211.78662109375, -0.023712158203125, 12.038566589355469, 34.53729248046875, 34.471588134765625, -32.51322937011719, 1.9460830688476562, 184.91363525390625, 14.578033447265625, -38.69398498535156, 12.314788818359375, 12.771514892578125, -118.15117645263672, 153.25656127929688, 13.203346252441406, 160.81488037109375, 115.468017578125, -118.52618408203125, -63.773712158203125, 32.811004638671875, 12.473846435546875, 28.723800659179688, 144.2451934814453, 149.07101440429688, -27.325035095214844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000388.npy"}
{"epoch": 0.812565445026178, "step": 389, "batch_size": 128, "mean": 63.71757125854492, "std": 100.0663070678711, "min": -193.88571166992188, "p10": -48.32160949707031, "median": 54.680999755859375, "p90": 177.53895111083983, "max": 231.40957641601562, "pos_frac": 0.7421875, "sample": [5.2486724853515625, -139.00808715820312, -22.392555236816406, 109.41098022460938, 163.101318359375, 160.056640625, 24.4674072265625, 26.57666015625, -8.740318298339844, 195.41790771484375, 22.9219970703125, 180.4496612548828, 63.7178955078125, 202.33187866210938, 43.08623504638672, 154.1757354736328, 9.203079223632812, 159.7926788330078, 118.15635681152344, 206.35308837890625, -70.89968872070312, 149.08705139160156, 196.50299072265625, 116.14572143554688, 5.888191223144531, 171.67819213867188, -174.54791259765625, 200.98760986328125, 37.44551086425781, 183.5579833984375, 36.570068359375, -180.96517944335938, 141.34799194335938, -36.48878479003906, -118.79794311523438, -4.3792724609375, -50.48454284667969, 149.84414672851562, 195.83428955078125, 231.40957641601562, 158.0753936767578, 0.0, -126.24363708496094, -21.147850036621094, 163.01959228515625, -107.08001708984375, 180.63186645507812, 152.808837890625, 165.33775329589844, 22.68292236328125, -14.1424560546875, 159.18499755859375, 48.429779052734375, 10.015645980834961, 136.53309631347656, 133.52572631835938, 28.553573608398438, 63.664154052734375, -27.693023681640625, 1.3258590698242188, 44.91810607910156, 13.548389434814453, 101.77400207519531, 173.71568298339844, -130.48538208007812, 148.81063842773438, 87.66989135742188, 154.57078552246094, 128.0842742919922, 126.43180847167969, 154.04458618164062, -87.96035766601562, 43.1883544921875, 169.95806884765625, 2.22698974609375, 215.6402587890625, 99.12330627441406, 74.953857421875, -31.77740478515625, -7.43988037109375, 70.453125, 122.45248413085938, 35.880760192871094, 60.932220458984375, 19.043983459472656, 167.8280029296875, -177.86277770996094, 149.29641723632812, 136.37347412109375, 6.087066650390625, 176.2216796875, 23.223388671875, -0.598388671875, 5.473773956298828, -9.633293151855469, 145.53173828125, 9.10723876953125, -193.88571166992188, -31.136810302734375, 176.29150390625, -27.850326538085938, 170.52392578125, -16.27471923828125, -18.498371124267578, 93.13104248046875, 194.40823364257812, 0.0, 85.84046936035156, 165.83106994628906, 208.68580627441406, 4.775421142578125, 9.272872924804688, -3.478515625, 106.0965576171875, 34.09510803222656, 171.0115509033203, 129.1954345703125, 10.912284851074219, -47.39463806152344, 19.425148010253906, 38.46217346191406, 167.2911834716797, -153.5574951171875, 21.191787719726562, 159.80137634277344, -4.721099853515625, 114.14421081542969, 169.906982421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000389.npy"}
{"epoch": 0.8146596858638744, "step": 390, "batch_size": 128, "mean": 62.80315399169922, "std": 103.98272705078125, "min": -195.4913330078125, "p10": -67.06737823486327, "median": 60.608612060546875, "p90": 187.06114501953124, "max": 218.80047607421875, "pos_frac": 0.7109375, "sample": [104.97967529296875, 0.769287109375, 80.37887573242188, 17.285736083984375, 117.041259765625, 153.48294067382812, 0.9965267181396484, 185.0513916015625, 141.0028076171875, 69.34002685546875, -37.227874755859375, 4.2831878662109375, 2.0873489379882812, -77.3291015625, 34.8995361328125, 121.19467163085938, -1.3545150756835938, 197.18328857421875, 171.68533325195312, 153.87844848632812, 150.86465454101562, -33.1385498046875, 176.55958557128906, 17.351486206054688, 70.135009765625, 90.82080078125, -74.1646728515625, 169.0604248046875, 166.099365234375, 85.01651000976562, 175.38604736328125, 154.6943817138672, -125.91184997558594, 66.66822052001953, 177.07452392578125, 36.69061279296875, 100.97074127197266, 24.326217651367188, -3.6578216552734375, 105.04646301269531, -141.880859375, 166.3962860107422, -64.02568054199219, 145.3153076171875, 91.97177124023438, -53.453948974609375, -14.35552978515625, -186.00204467773438, 155.72796630859375, -63.926483154296875, 60.85133361816406, -109.94580078125, -122.85223388671875, 30.71197509765625, 211.63412475585938, 14.014799118041992, 33.674652099609375, 179.4202117919922, 147.79757690429688, -158.95733642578125, -132.82638549804688, 109.38687133789062, 183.23826599121094, 19.606536865234375, 204.39242553710938, -157.29071044921875, 213.40582275390625, 88.12777709960938, -18.450714111328125, 160.32809448242188, 60.36589050292969, 212.1990966796875, 84.9261474609375, 0.0, -27.567596435546875, -9.836944580078125, 171.20758056640625, -31.851089477539062, 156.65863037109375, 190.13380432128906, -44.18229675292969, 187.44784545898438, 216.60565185546875, -195.4913330078125, -16.62127685546875, 179.6385498046875, 218.80047607421875, 172.0215606689453, 6.3507080078125, 8.726272583007812, 6.178291320800781, 199.02340698242188, 35.740875244140625, 76.38101196289062, 127.952392578125, 140.98995971679688, -13.457633972167969, 185.41220092773438, 41.697418212890625, -49.52923583984375, -1.2730541229248047, -95.43655395507812, 200.63134765625, 171.50332641601562, -22.028411865234375, 37.89385986328125, 206.60150146484375, 7.1968994140625, -0.9293022155761719, 0.20770263671875, -37.9766845703125, 3.7559051513671875, -10.218391418457031, 139.47760009765625, 55.82305145263672, 26.734603881835938, 176.49191284179688, 167.88418579101562, 42.00187683105469, 186.72515869140625, 150.41615295410156, 29.774749755859375, -13.681396484375, 187.8250732421875, 186.89541625976562, -138.9368896484375, 138.02207946777344, -8.053871154785156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000390.npy"}
{"epoch": 0.8167539267015707, "step": 391, "batch_size": 128, "mean": 60.35431671142578, "std": 89.56831359863281, "min": -166.443359375, "p10": -24.037284851074208, "median": 37.84182357788086, "p90": 174.5267791748047, "max": 376.31982421875, "pos_frac": 0.7421875, "sample": [-10.951545715332031, -37.31451416015625, 5.960027694702148, 53.749732971191406, 40.730186462402344, 10.758544921875, -3.08758544921875, 5.726524353027344, 8.117286682128906, 125.94509887695312, 4.67938232421875, 10.68585205078125, -16.383155822753906, 99.81175231933594, 197.2490234375, 81.55331420898438, 96.39242553710938, 118.65152740478516, -14.120040893554688, -21.27619171142578, -15.972320556640625, 171.56259155273438, 77.27127075195312, 70.0748291015625, -46.49314880371094, 43.701904296875, 175.10275268554688, 64.4920654296875, 1.8124542236328125, 17.929107666015625, 7.324462890625, -83.2451171875, 173.793212890625, -7.309783935546875, 102.57022094726562, 120.17333984375, 142.0469207763672, 9.940475463867188, -15.463043212890625, 156.16436767578125, 6.988048553466797, 148.27076721191406, 150.7437286376953, 225.2564697265625, 174.69778442382812, -3.925933837890625, 172.20834350585938, 14.71875, -12.195098876953125, -2.88861083984375, 222.0364990234375, 27.771018981933594, -75.00835418701172, 53.888153076171875, -21.449295043945312, 113.45785522460938, 146.72015380859375, 109.24827575683594, 156.432373046875, -33.78790283203125, 41.06005859375, -141.01026916503906, 148.02105712890625, 10.925971984863281, 34.953460693359375, 13.6533203125, 119.40679931640625, 8.552204132080078, 41.86680603027344, 3.2866058349609375, 2.639801025390625, 185.92855834960938, 23.31512451171875, 117.78337860107422, 145.32601928710938, -8.029876708984375, 136.6368408203125, 78.32659912109375, -9.367752075195312, 170.85586547851562, -5.73773193359375, 77.0592041015625, 174.4534912109375, 376.31982421875, 194.66790771484375, 1.785400390625, -10.034126281738281, 34.623809814453125, 163.03225708007812, -0.990631103515625, 12.0343017578125, 3.8869800567626953, -160.8501434326172, 167.135986328125, 42.312957763671875, 57.887176513671875, -38.26878356933594, 136.86607360839844, 190.83444213867188, 188.66259765625, -34.96649169921875, -47.24444580078125, 33.69468688964844, -128.11383056640625, 2.741485595703125, -2.225444793701172, 60.87071228027344, 131.8174285888672, 148.08441162109375, 183.24554443359375, 14.64276123046875, 152.185302734375, 145.20440673828125, -14.0797119140625, 114.46244049072266, 71.1243896484375, 189.66168212890625, 163.273193359375, 224.638427734375, 19.197784423828125, -30.075927734375, 10.217025756835938, 31.90765380859375, 7.630180358886719, -6.916236877441406, 51.79864501953125, -166.443359375, 173.67074584960938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000391.npy"}
{"epoch": 0.818848167539267, "step": 392, "batch_size": 128, "mean": 79.33124542236328, "std": 91.27642059326172, "min": -163.47470092773438, "p10": -9.054138183593748, "median": 54.99873352050781, "p90": 203.18193359375, "max": 249.80160522460938, "pos_frac": 0.78125, "sample": [171.82872009277344, 145.20205688476562, 50.25669860839844, 45.76287841796875, 123.45635986328125, 3.1292648315429688, 92.39566040039062, 148.63259887695312, 80.3004150390625, 137.32843017578125, 15.5750732421875, 87.56674194335938, 138.53831481933594, 33.09527587890625, -8.464080810546875, 2.7707061767578125, -163.47470092773438, 235.019775390625, 129.45211791992188, -43.45054626464844, 203.4312744140625, -4.852268218994141, 5.080535888671875, 49.606231689453125, 197.91485595703125, 224.7637939453125, 0.0, -48.532470703125, 186.63796997070312, 159.02755737304688, 7.7163848876953125, 41.059295654296875, 201.13067626953125, 50.8729248046875, 37.28708267211914, 212.64764404296875, 203.5943603515625, 182.80673217773438, 153.83541870117188, -14.790985107421875, 157.80776977539062, 171.61285400390625, 13.040756225585938, 17.46826171875, -7.921142578125, 22.706344604492188, -8.171623229980469, 1.16851806640625, 125.81658935546875, 176.54510498046875, 44.2479248046875, 185.3994903564453, 209.48004150390625, 137.7195281982422, 203.0750732421875, 147.8458251953125, 6.538848876953125, 131.88250732421875, 223.5430908203125, -6.5042724609375, -15.771308898925781, 130.4100341796875, 17.128448486328125, 201.23779296875, 33.75419998168945, 89.02645111083984, -27.99005126953125, 152.54037475585938, 216.54071044921875, 0.4224853515625, 59.124542236328125, 193.61361694335938, 213.7913818359375, 158.087646484375, 92.2646484375, 206.12075805664062, 40.94673156738281, -155.10888671875, -6.874259948730469, -2.2055091857910156, 10.021682739257812, 171.28912353515625, 3.454345703125, 127.57415771484375, 16.270126342773438, -5.104888916015625, -10.430938720703125, 4.394561767578125, 23.964218139648438, -24.555587768554688, -1.68353271484375, 0.0, 152.7563934326172, 22.05419921875, 167.1109619140625, 214.28961181640625, 62.486572265625, 185.82489013671875, 198.37606811523438, -0.6684722900390625, 189.14627075195312, -0.4777069091796875, 18.8741455078125, 1.7312545776367188, 128.0794677734375, 249.80160522460938, -2.4635543823242188, 209.40814208984375, 21.663497924804688, 101.33206939697266, -132.17120361328125, 193.12338256835938, -5.9417877197265625, 6.6562347412109375, 157.03167724609375, 28.6595458984375, 59.78668212890625, 0.733184814453125, -13.807220458984375, 62.68473815917969, 65.40170288085938, 155.99472045898438, -50.58500671386719, 168.12210083007812, -27.202713012695312, 25.10828399658203, 172.466552734375, 29.299896240234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000392.npy"}
{"epoch": 0.8209424083769633, "step": 393, "batch_size": 128, "mean": 56.42615509033203, "std": 100.69145202636719, "min": -210.75802612304688, "p10": -43.544517517089844, "median": 31.70484161376953, "p90": 174.60760650634765, "max": 341.209716796875, "pos_frac": 0.7421875, "sample": [14.25930404663086, -72.76345825195312, 4.515403747558594, 106.829833984375, 128.19113159179688, -70.489501953125, 162.88626098632812, 120.6844482421875, 162.58233642578125, 1.582305908203125, 11.03350830078125, -3.3453369140625, 82.70169067382812, 1.8702621459960938, 105.56851959228516, 28.878997802734375, -111.50396728515625, 144.31837463378906, 122.74488067626953, 84.0223388671875, 184.2781982421875, 49.12744140625, 13.79119873046875, 41.46173095703125, -39.67138671875, 188.74969482421875, 175.72938537597656, 168.30474853515625, 31.769668579101562, 26.779769897460938, 171.0606689453125, 1.83465576171875, 111.82418823242188, 161.96170043945312, 21.902435302734375, 35.974456787109375, -2.7867431640625, 190.21490478515625, 26.65472412109375, 158.3565216064453, 24.60760498046875, -110.3260498046875, 38.11541748046875, 175.289794921875, -111.3433837890625, 176.7552490234375, -52.163352966308594, -30.855133056640625, -116.26565551757812, 98.60025024414062, 270.6817626953125, 138.65994262695312, 8.415237426757812, 21.280990600585938, 12.372444152832031, 60.88372802734375, 0.7665081024169922, -158.74583435058594, 155.27963256835938, 109.97662353515625, -210.75802612304688, 27.821136474609375, 15.690093994140625, -16.48455810546875, 20.890762329101562, 163.75338745117188, 120.51446533203125, 169.16131591796875, -201.93072509765625, -37.692138671875, -40.044464111328125, 65.37753295898438, 14.068107604980469, 50.4696044921875, 17.568744659423828, 60.518798828125, 213.8140869140625, -15.288421630859375, 45.01409912109375, -8.114463806152344, 10.37099838256836, -15.605155944824219, 158.70248413085938, -175.47459411621094, 18.87592315673828, 117.61420440673828, 163.75698852539062, 135.93441772460938, 154.24595642089844, 165.2393341064453, -28.347305297851562, 193.44631958007812, 177.11209106445312, 174.3157958984375, 175.2884979248047, 29.7320556640625, 31.6400146484375, -42.8349609375, 173.11741638183594, -206.2130126953125, 6.048088073730469, 38.0408935546875, -13.029541015625, -6.256599426269531, 341.209716796875, 166.7784423828125, 43.39080810546875, 0.0, 9.518386840820312, 158.7598876953125, 222.49798583984375, -45.20014953613281, 4.002571105957031, 9.241256713867188, -26.068191528320312, 134.8494873046875, 160.0952606201172, -8.66346549987793, -21.734100341796875, -13.072296142578125, 0.2787742614746094, -29.53131103515625, 169.11643981933594, 169.81021118164062, 3.3100814819335938, 132.013916015625, 137.84320068359375, 160.16064453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000393.npy"}
{"epoch": 0.8230366492146597, "step": 394, "batch_size": 128, "mean": 57.894935607910156, "std": 92.6501235961914, "min": -172.86033630371094, "p10": -40.447900390624994, "median": 36.15449142456055, "p90": 187.03390502929688, "max": 240.25335693359375, "pos_frac": 0.703125, "sample": [181.26333618164062, 113.45693969726562, 182.93545532226562, 151.36285400390625, 47.39166259765625, 168.56991577148438, -14.5694580078125, -3.8757247924804688, 24.029144287109375, 189.14337158203125, 113.201416015625, 12.754013061523438, 35.3651123046875, 126.16162109375, -53.19091796875, -3.4203834533691406, 177.4129180908203, -94.44561004638672, 180.14788818359375, -22.731876373291016, 9.622894287109375, 174.80780029296875, 223.81723022460938, -17.591140747070312, 1.3492660522460938, -27.93182373046875, 21.601654052734375, 157.58206176757812, 128.89208984375, 59.196197509765625, -38.443756103515625, 0.0, -14.875953674316406, -56.674163818359375, 181.951904296875, 12.36370849609375, 16.852325439453125, 138.50799560546875, 5.4609375, 47.241111755371094, 127.96429443359375, 63.339447021484375, 48.44622802734375, -172.86033630371094, 18.21160125732422, 188.35025024414062, 181.53741455078125, 58.58359146118164, 1.733306884765625, 18.12188720703125, -23.460784912109375, -45.124237060546875, 48.278350830078125, -14.794677734375, -23.4842529296875, 192.15335083007812, 36.943870544433594, 96.29644775390625, 11.160049438476562, 59.9737548828125, 85.073486328125, -56.09454345703125, 0.5622329711914062, -83.83243560791016, 177.11050415039062, 3.089008331298828, 56.955039978027344, 116.3648681640625, 19.916358947753906, 153.36758422851562, 166.327392578125, 205.5104217529297, 192.19097900390625, 2.821380615234375, 219.38922119140625, -2.99151611328125, -64.4132080078125, -32.0390625, -8.060958862304688, 240.25335693359375, 9.707592010498047, -135.9991455078125, 43.963592529296875, -9.636322021484375, -45.37640380859375, 20.934478759765625, 40.74896240234375, 29.031143188476562, 8.390823364257812, 112.5296630859375, -0.6988983154296875, -16.53083038330078, 127.62481689453125, 12.723892211914062, -23.46319580078125, -36.99127197265625, 39.26397705078125, 46.86956787109375, 38.887725830078125, 75.3045654296875, -158.70970153808594, 161.51150512695312, 34.574951171875, -34.61170959472656, 198.376953125, 226.677734375, -32.20758056640625, 52.96051025390625, 113.4835205078125, 188.50982666015625, 132.71524047851562, 149.23080444335938, 177.98733520507812, 159.39541625976562, 180.64913940429688, -46.59503173828125, 218.69525146484375, 18.6375732421875, 2.337890625, 191.36334228515625, 4.018524169921875, 186.46975708007812, -98.45095825195312, 146.0411376953125, -7.5824737548828125, -15.460922241210938, -37.746368408203125, 163.43368530273438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000394.npy"}
{"epoch": 0.8251308900523561, "step": 395, "batch_size": 128, "mean": 60.10478973388672, "std": 91.2669906616211, "min": -167.91433715820312, "p10": -30.38642425537109, "median": 28.72216796875, "p90": 185.10230865478516, "max": 225.771484375, "pos_frac": 0.765625, "sample": [49.99639892578125, 5.48291015625, -33.79412841796875, -10.603485107421875, -11.042236328125, -16.790252685546875, 1.47296142578125, 223.50030517578125, 210.3409423828125, 17.668807983398438, 35.851524353027344, 73.61541748046875, 28.200164794921875, 21.265716552734375, 1.1929397583007812, 87.59213256835938, 13.605659484863281, -118.83792114257812, 108.45504760742188, 10.88458251953125, 27.01715087890625, 176.01129150390625, -20.27263641357422, -94.16535949707031, 195.79071044921875, 140.14788818359375, 24.058425903320312, 2.815357208251953, 158.820556640625, 34.39332580566406, 146.23941040039062, 157.42617797851562, 125.10762023925781, 125.40357971191406, 31.563873291015625, 183.20855712890625, 132.53793334960938, 51.7177734375, 0.06560516357421875, -18.93665313720703, 108.7301025390625, 13.009607315063477, -11.72625732421875, 31.12237548828125, 183.1639404296875, 125.07171630859375, -14.337387084960938, -38.10685729980469, 147.50973510742188, 186.8769073486328, 205.96807861328125, -78.49319458007812, 71.44818115234375, 160.09593200683594, 178.48507690429688, 132.69882202148438, 38.855804443359375, 181.17913818359375, 18.5296630859375, 128.3442840576172, -35.6951904296875, -96.00433349609375, 23.66717529296875, 0.577880859375, 173.8604736328125, -29.01812744140625, 18.030906677246094, 121.14083099365234, -16.62969970703125, 10.59112548828125, -23.16162109375, 24.722488403320312, -4.5525054931640625, -25.443756103515625, 29.315521240234375, 14.427749633789062, 158.00721740722656, -0.820587158203125, 10.652618408203125, 0.25201416015625, 25.555084228515625, 19.19245147705078, 225.771484375, 186.28143310546875, 178.3674774169922, -22.42864990234375, 7.976959228515625, -75.44303894042969, 96.9261474609375, 49.17755126953125, 9.654705047607422, 72.21524047851562, -158.9197998046875, 121.06944274902344, -15.320159912109375, 11.526596069335938, 196.35455322265625, 205.78038024902344, 195.25633239746094, 182.52883911132812, -10.060569763183594, 170.18182373046875, 29.244171142578125, -33.57911682128906, 134.264892578125, 119.88423156738281, 173.54103088378906, 11.78216552734375, 166.84524536132812, -155.6490936279297, -3.1045989990234375, 157.59239196777344, 199.72073364257812, -73.901611328125, 12.936676025390625, -167.91433715820312, 34.43817138671875, 6.68998908996582, 196.395751953125, 191.69091796875, 178.9285888671875, 184.5969696044922, 20.872650146484375, 14.110321044921875, 25.0943603515625, 130.41653442382812, 7.6394500732421875, 59.87774658203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000395.npy"}
{"epoch": 0.8272251308900523, "step": 396, "batch_size": 128, "mean": 58.82726287841797, "std": 98.72859954833984, "min": -220.71029663085938, "p10": -43.683537292480466, "median": 44.892486572265625, "p90": 180.778515625, "max": 239.98736572265625, "pos_frac": 0.6953125, "sample": [143.20925903320312, 15.90716552734375, -12.95233154296875, 184.4217529296875, 168.9693603515625, 17.390625, -4.61083984375, 18.158538818359375, -36.939453125, 205.82888793945312, 116.9285888671875, -22.375778198242188, 118.65779113769531, 42.5745849609375, 1.289886474609375, 94.09561157226562, 195.49020385742188, -111.68023681640625, 6.421051025390625, 152.30810546875, -31.425323486328125, -8.853843688964844, -10.79541015625, 237.84088134765625, -3.3764190673828125, 38.85931396484375, 124.9173583984375, 150.132080078125, 78.69056701660156, 139.98370361328125, 106.23492431640625, -132.39035034179688, 75.82940673828125, 180.6378173828125, -139.79629516601562, 183.77886962890625, 9.246009826660156, -0.13580322265625, -6.144084930419922, 132.04441833496094, 28.385971069335938, 152.90203857421875, 80.89840698242188, 176.44793701171875, 200.2164306640625, -28.80859375, 15.207294464111328, 8.846832275390625, -4.73529052734375, 164.02993774414062, 208.11212158203125, 198.06314086914062, 239.98736572265625, 5.02923583984375, 175.30142211914062, 105.38424682617188, 163.317626953125, 2.76959228515625, 16.764495849609375, 2.4390487670898438, 7.545543670654297, -12.681121826171875, -148.55076599121094, -81.2132568359375, 109.30520629882812, -77.25735473632812, 51.982208251953125, -162.93971252441406, 14.38092041015625, -4.440742492675781, 156.083984375, -5.80242919921875, -30.633804321289062, -8.369178771972656, 159.86395263671875, 170.8297119140625, 66.81390380859375, 159.06887817382812, 25.299957275390625, -163.0880889892578, 76.21856689453125, 127.4764633178711, 18.53333282470703, 152.444580078125, 53.3902587890625, 181.1068115234375, 24.28060531616211, 207.3338165283203, -135.2044677734375, 170.1959228515625, -32.88995361328125, 87.14874267578125, -54.6531982421875, 116.50833892822266, 109.2827377319336, 112.28047180175781, 138.2442626953125, -6.8009033203125, 21.02667236328125, 47.21038818359375, 178.8193817138672, 70.47773742675781, -2.564483642578125, -220.71029663085938, -24.223526000976562, 175.6752471923828, -14.452392578125, 200.87744140625, 174.02337646484375, 184.45960998535156, 0.0, 16.7769775390625, 173.23001098632812, 40.0159912109375, 125.76411437988281, 9.378028869628906, 0.0, 49.50633239746094, 33.765472412109375, -19.494247436523438, 76.56185913085938, -149.3902587890625, 85.77810668945312, 167.66714477539062, 164.1278076171875, -41.921783447265625, -47.79429626464844, 155.27488708496094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000396.npy"}
{"epoch": 0.8293193717277487, "step": 397, "batch_size": 128, "mean": 54.94662857055664, "std": 99.0977554321289, "min": -201.63616943359375, "p10": -57.45496826171871, "median": 34.5770263671875, "p90": 182.47417907714842, "max": 218.79632568359375, "pos_frac": 0.703125, "sample": [-8.44384765625, 7.804656982421875, -13.733497619628906, 23.978378295898438, 188.96923828125, 16.289031982421875, -125.74968719482422, 49.51483154296875, -4.7270965576171875, 21.2821044921875, 9.676193237304688, 13.960205078125, 144.1646728515625, -13.07586669921875, 27.161087036132812, 182.04202270507812, 28.255691528320312, 136.87969970703125, 216.02517700195312, 151.1119384765625, -158.95343017578125, 27.808563232421875, -18.26446533203125, 38.9102783203125, 150.80990600585938, 165.36929321289062, 0.0, -1.99359130859375, 203.06619262695312, 40.832794189453125, 0.0, -87.2618408203125, 159.52188110351562, -0.1542205810546875, -4.45263671875, 1.972452163696289, 7.8447265625, 173.6009979248047, -86.43588256835938, 159.57211303710938, 190.30123901367188, 56.05235290527344, 141.15374755859375, 182.04156494140625, -33.269012451171875, -0.13573455810546875, 156.21224975585938, 166.239990234375, 218.79632568359375, -45.39959716796875, 216.10910034179688, -41.3115234375, 203.5814208984375, -201.63616943359375, -126.92987060546875, -40.85693359375, 45.697509765625, -138.6067657470703, 0.995269775390625, -166.59921264648438, -36.785888671875, 47.09843444824219, 183.4825439453125, 153.0626983642578, 108.92402648925781, 28.35033416748047, 146.40310668945312, -10.59722900390625, 47.426971435546875, -1.1618881225585938, 57.8756103515625, 206.74798583984375, 77.57778930664062, 56.28016662597656, 163.45721435546875, 156.150390625, 161.95120239257812, -89.23236083984375, 112.25735473632812, 50.702850341796875, 165.56578063964844, 30.2437744140625, 190.9803466796875, 183.80514526367188, 41.188323974609375, 0.4991779327392578, 159.4233856201172, -6.45068359375, 10.985664367675781, 8.358551025390625, 165.4483642578125, 14.474136352539062, 7.1539459228515625, 11.140365600585938, 165.87322998046875, 158.2367401123047, 10.380561828613281, 19.361061096191406, 186.61639404296875, 73.35073852539062, 20.954345703125, -146.30206298828125, 41.941558837890625, 12.364948272705078, 199.97959899902344, 180.0166015625, 139.3390350341797, 7.717529296875, -16.38031768798828, -19.572219848632812, -25.900436401367188, 177.86590576171875, 144.7518310546875, -83.73077392578125, 50.0284423828125, 42.2088623046875, -142.10455322265625, -15.0137939453125, 156.8780059814453, 12.5787353515625, -41.485565185546875, -46.19390869140625, 147.48797607421875, 155.7340087890625, 88.10394287109375, -108.1239013671875, 136.62002563476562, 141.18414306640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000397.npy"}
{"epoch": 0.831413612565445, "step": 398, "batch_size": 128, "mean": 43.48147964477539, "std": 106.17778015136719, "min": -185.88299560546875, "p10": -91.47850646972653, "median": 20.98400115966797, "p90": 181.1331344604492, "max": 266.8836669921875, "pos_frac": 0.640625, "sample": [14.54901123046875, 20.351654052734375, 162.9490509033203, -9.84344482421875, 266.8836669921875, -127.8546142578125, 65.37515258789062, -21.7606201171875, 120.7198486328125, 184.02117919921875, -156.85382080078125, -61.419921875, -180.11624145507812, 4.70770263671875, 7.716438293457031, -9.729156494140625, 36.266876220703125, -5.712371826171875, 177.12283325195312, 190.24453735351562, 180.11622619628906, -146.6029052734375, -127.191162109375, 148.56488037109375, 39.67694091796875, 132.99853515625, 26.1903076171875, 175.83724975585938, 172.14285278320312, 140.01515197753906, -117.00515747070312, 6.879127502441406, 38.033905029296875, -185.88299560546875, -68.36859130859375, 165.26698303222656, 10.54193115234375, 159.611328125, -3.4832763671875, -129.87255859375, 2.351827621459961, -120.28155517578125, 20.45629119873047, 200.24969482421875, -0.5188560485839844, -80.53851318359375, 8.021316528320312, 146.78038024902344, 20.92822265625, -7.403411865234375, 36.299072265625, -12.555511474609375, 103.91513061523438, -24.092086791992188, 152.07980346679688, 48.4556884765625, -0.24608802795410156, -177.34701538085938, 23.735031127929688, -17.628355026245117, 89.53309631347656, 42.377349853515625, 176.81884765625, -41.05523681640625, 201.07403564453125, 3.267852783203125, 21.79088592529297, -161.10382080078125, 139.5126953125, 187.14698791503906, 21.039779663085938, 215.0267333984375, 29.9854736328125, 226.26486206054688, -13.530502319335938, -16.534957885742188, 10.535314559936523, 175.9752197265625, -33.7545166015625, -12.909820556640625, 191.49951171875, 192.83343505859375, 4.2816925048828125, 31.920915603637695, 124.56075286865234, 5.402587890625, -40.73918151855469, 154.76295471191406, 40.078125, -54.681304931640625, 143.04214477539062, 101.16012573242188, -16.978103637695312, -167.17330932617188, 99.67822265625, -8.740875244140625, 163.4661865234375, -28.852874755859375, -16.32135009765625, 74.98699951171875, -18.02991485595703, -13.853179931640625, 156.4478759765625, 137.0518035888672, 13.577301025390625, 16.14061737060547, 48.194091796875, 183.50592041015625, 176.14065551757812, -42.136932373046875, -74.71814727783203, 199.61724853515625, 153.80047607421875, -66.70145416259766, 178.64898681640625, -5.246856689453125, -60.05625915527344, -184.1712646484375, 15.033157348632812, 179.88421630859375, 147.74786376953125, -9.66156005859375, 6.483823776245117, 234.7999267578125, 50.50433349609375, 35.683197021484375, 136.99777221679688, 96.55133056640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000398.npy"}
{"epoch": 0.8335078534031414, "step": 399, "batch_size": 128, "mean": 52.76074981689453, "std": 93.2054214477539, "min": -188.322265625, "p10": -30.684902954101563, "median": 25.562960624694824, "p90": 183.07017211914064, "max": 227.66983032226562, "pos_frac": 0.6953125, "sample": [-21.646957397460938, 155.68910217285156, 157.2337646484375, 166.95095825195312, 12.5648193359375, -22.4981689453125, -3.49139404296875, 42.819854736328125, 37.824951171875, 191.1290283203125, 49.320518493652344, -34.462188720703125, -22.3251953125, 167.90391540527344, -2.9315872192382812, 47.375213623046875, 115.2137451171875, 3.4809722900390625, 195.6600799560547, 29.56488037109375, -1.608367919921875, -162.5452880859375, -18.70587158203125, 180.52593994140625, 2.6417236328125, -0.97149658203125, 8.613525390625, -149.33612060546875, -188.322265625, 147.8549346923828, -3.150665283203125, 183.57861328125, 26.60617446899414, 190.03961181640625, 17.517433166503906, -10.7232666015625, 164.43856811523438, -7.835365295410156, -2.5947265625, 201.45758056640625, 35.76983642578125, 182.85301208496094, 227.66983032226562, 165.86996459960938, -10.8546142578125, -12.608749389648438, 184.36599731445312, -54.172576904296875, 1.7886066436767578, 161.84857177734375, -30.4166259765625, 30.576133728027344, 144.8933868408203, 121.8680419921875, 200.92291259765625, -10.42041015625, 13.735736846923828, 12.12945556640625, 17.637542724609375, -9.471866607666016, 3.127960205078125, 182.85458374023438, -10.4534912109375, 20.97856903076172, 173.07061767578125, 20.325855255126953, 166.17076110839844, 20.050819396972656, 118.71192932128906, 47.0137939453125, 75.24501037597656, 146.744873046875, -12.015335083007812, -146.6512908935547, -149.80728149414062, -180.9159393310547, -13.826171875, 36.59246826171875, 180.72784423828125, 189.77285766601562, 21.654067993164062, 162.71795654296875, 93.81597900390625, 168.6957244873047, 116.39602661132812, 65.84860229492188, 17.9921875, 0.412689208984375, 53.07745361328125, 125.648193359375, 10.671121597290039, -49.482208251953125, 10.961406707763672, 91.67431640625, 11.01702880859375, 18.11932373046875, -31.310882568359375, 4.460414886474609, 186.049072265625, 8.542236328125, 78.75872802734375, 9.585845947265625, 87.61761474609375, -53.736907958984375, -7.6785888671875, 183.57321166992188, -18.649215698242188, 220.26837158203125, 116.13729858398438, 117.84701538085938, 193.2407684326172, -162.04827880859375, -38.454559326171875, 39.828857421875, 25.96767234802246, 25.158248901367188, 124.42546081542969, 115.1787109375, -3.6058349609375, 156.665771484375, -27.7662353515625, 32.19525146484375, 140.3916015625, 94.60151672363281, -21.035247802734375, -2.401031494140625, 137.3111114501953, 18.477890014648438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000399.npy"}
{"epoch": 0.8356020942408376, "step": 400, "batch_size": 128, "mean": 63.494422912597656, "std": 96.73956298828125, "min": -195.26101684570312, "p10": -36.60224075317382, "median": 38.23085021972656, "p90": 183.91375579833985, "max": 221.415771484375, "pos_frac": 0.71875, "sample": [5.2848052978515625, -101.84799194335938, -2.724609375, 32.32257080078125, 147.8670654296875, 19.285667419433594, 134.70318603515625, 135.11862182617188, 98.67463684082031, -114.59485626220703, 5.0706634521484375, 91.46990966796875, -195.26101684570312, 9.160438537597656, 187.3171844482422, 16.75390625, 13.391067504882812, 175.9763946533203, 141.8125, 174.7560577392578, 213.61419677734375, -19.419540405273438, 20.80162811279297, 126.43484497070312, 176.98684692382812, 154.30029296875, -47.187255859375, 164.64773559570312, -22.52490234375, 186.71417236328125, 45.23051452636719, 9.638103485107422, 19.469947814941406, 142.04806518554688, 45.07891845703125, 183.8170166015625, 181.26483154296875, 111.376953125, 23.95435333251953, -9.674575805664062, -163.0482177734375, 158.67745971679688, 141.1239471435547, 177.1529541015625, 30.8768310546875, 178.49710083007812, -6.626922607421875, 153.392822265625, 5.36224365234375, -7.871856689453125, 41.25018310546875, 153.44732666015625, 189.88916015625, -138.1561279296875, 1.5400390625, -33.935462951660156, 5.797545433044434, 10.47802734375, 190.36416625976562, -2.7066268920898438, -1.3340930938720703, 48.443023681640625, 188.41763305664062, 160.95745849609375, 35.211517333984375, -28.6239013671875, 23.130645751953125, 172.54611206054688, 48.228515625, 187.6494903564453, 0.2481689453125, 4.537895202636719, -26.100982666015625, 28.550186157226562, -42.82472229003906, -82.19961547851562, 177.19012451171875, 189.2357940673828, -1.477203369140625, -47.830230712890625, 116.33389282226562, 60.487274169921875, 148.9638671875, 141.56735229492188, -3.490976333618164, 177.67138671875, 148.75936889648438, -24.003952026367188, 34.94618225097656, 82.88431549072266, 138.81536865234375, -10.464263916015625, -5.4111175537109375, -4.219980239868164, 3.561309814453125, 168.29702758789062, -20.389053344726562, 143.25790405273438, 2.8233489990234375, -129.59384155273438, 89.97705078125, 193.30551147460938, 146.89303588867188, 27.201324462890625, 5.041900634765625, 28.51202392578125, 145.71337890625, -6.980018615722656, -6.8504638671875, 182.8070526123047, 159.95762634277344, 49.43860626220703, 171.78460693359375, -15.595947265625, 173.36843872070312, 186.35850524902344, 167.8519287109375, -25.08441162109375, 221.415771484375, 200.6479949951172, 184.1394805908203, 173.60708618164062, 0.0, -77.11846923828125, -101.3134765625, 1.89849853515625, -162.58755493164062, 41.532066345214844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000400.npy"}
{"epoch": 0.837696335078534, "step": 401, "batch_size": 128, "mean": 73.17318725585938, "std": 93.72574615478516, "min": -244.46240234375, "p10": -23.134516143798827, "median": 64.39787292480469, "p90": 189.44139099121094, "max": 254.439208984375, "pos_frac": 0.7890625, "sample": [3.0403594970703125, 8.709381103515625, 22.826263427734375, -9.186187744140625, 16.3997802734375, 119.96870422363281, 13.97174072265625, 131.88165283203125, 118.36448669433594, -5.190216064453125, 169.80880737304688, 23.139923095703125, 37.51813507080078, 181.8482666015625, 131.96658325195312, 31.799270629882812, 101.296875, 159.83184814453125, -7.134002685546875, 147.5976104736328, -9.399993896484375, 28.231292724609375, -18.68793487548828, 112.36090087890625, -4.904052734375, 181.35482788085938, 73.22689819335938, 1.8194618225097656, 165.887451171875, 115.64431762695312, 194.1822509765625, 221.97039794921875, 156.781005859375, 59.65118408203125, 16.2421875, 15.54193115234375, 165.1062774658203, 151.89852905273438, 9.977348327636719, 155.56985473632812, 254.439208984375, -21.612945556640625, 174.0159912109375, 188.88232421875, 197.18197631835938, 169.22589111328125, 70.90150451660156, 20.376739501953125, 166.75277709960938, 87.5548095703125, -113.07489013671875, 156.22702026367188, 28.67926025390625, 11.897880554199219, -244.46240234375, 131.04632568359375, 11.3677978515625, 206.456787109375, 119.01513671875, -119.16398620605469, -25.292343139648438, 153.97344970703125, 201.91229248046875, 161.46356201171875, 148.9510498046875, 131.9991455078125, -77.45742797851562, -123.61073303222656, 175.20079040527344, -17.5316162109375, -0.8584213256835938, 3.0946731567382812, 217.6932373046875, 27.97637939453125, 36.260498046875, 3.859344482421875, 174.3472442626953, 96.997802734375, 46.096649169921875, 92.1325454711914, 107.302734375, -24.2205810546875, 162.3438720703125, 184.6583251953125, 182.57781982421875, 199.43505859375, 46.634613037109375, -4.1278533935546875, 63.650390625, -24.4197998046875, 82.2896728515625, 10.43585205078125, -11.7254638671875, -27.022552490234375, 85.76632690429688, -25.541824340820312, -28.5050048828125, 191.99908447265625, 220.92005920410156, 22.0399169921875, 30.01934814453125, -2.798126220703125, 188.43905639648438, 174.5071563720703, 190.74588012695312, 18.198287963867188, 4.006134033203125, 4.748565673828125, -22.66905975341797, 65.27964782714844, 65.14535522460938, 1.61383056640625, 5.639190673828125, 15.495922088623047, -5.4291839599609375, 2.2034378051757812, 102.53985595703125, 156.64657592773438, 165.53790283203125, -32.457191467285156, 127.89985656738281, 174.64048767089844, 176.57846069335938, 5.468658447265625, 215.0050506591797, -191.78912353515625, 7.8417205810546875, 194.7900390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000401.npy"}
{"epoch": 0.8397905759162304, "step": 402, "batch_size": 128, "mean": 60.38100051879883, "std": 97.58932495117188, "min": -174.34075927734375, "p10": -38.55734558105468, "median": 39.416664123535156, "p90": 179.71618347167967, "max": 257.5155029296875, "pos_frac": 0.75, "sample": [-4.3948974609375, -3.8993377685546875, -145.02088928222656, 49.053497314453125, 165.9065704345703, -19.373092651367188, -37.9007568359375, -34.7938232421875, 132.86920166015625, -174.34075927734375, -149.6597900390625, 149.37576293945312, 16.72210693359375, -9.977333068847656, 4.697288513183594, 229.29855346679688, 159.9272003173828, 146.041015625, 7.1401824951171875, -29.04180908203125, 159.9286346435547, 3.0240631103515625, 9.091941833496094, 196.60562133789062, -148.7377471923828, 3.9644737243652344, 2.634429931640625, 159.81729125976562, -168.63375854492188, 37.74664306640625, 159.8701171875, -7.11236572265625, 130.94537353515625, 172.85281372070312, 9.39813232421875, 0.0, 227.87701416015625, 178.6650390625, -16.23333740234375, 160.62588500976562, 69.37975311279297, 34.4832763671875, 163.62478637695312, 16.674713134765625, 137.10208129882812, 163.65234375, -136.83554077148438, -10.589813232421875, 188.7877197265625, 174.85633850097656, 27.8052978515625, -88.78125, -40.089385986328125, 157.07415771484375, 144.45504760742188, 41.08668518066406, 65.59117126464844, 132.75718688964844, 3.61322021484375, 147.40493774414062, 12.216766357421875, 214.36962890625, 204.99298095703125, 13.202659606933594, 137.61672973632812, -24.409622192382812, -130.4741973876953, 82.09561157226562, 21.7637939453125, -158.57943725585938, 179.15740966796875, 162.77545166015625, 71.07876586914062, -4.5074462890625, 151.9239044189453, -6.922492980957031, 257.5155029296875, 3.5954551696777344, 67.99578857421875, 185.20950317382812, 175.18994140625, 179.365966796875, 34.76190185546875, 116.14554595947266, 99.26608276367188, 42.16569519042969, 182.7667236328125, 130.50167846679688, -102.47671508789062, 157.66094970703125, 138.75802612304688, 68.41720581054688, 14.133277893066406, -0.638427734375, 34.88275146484375, 109.23046875, 6.675209045410156, 45.8004150390625, 18.6920166015625, 177.12185668945312, 181.82455444335938, 180.53335571289062, 163.44631958007812, -27.363632202148438, 216.05108642578125, -2.6745758056640625, 42.25044250488281, 12.55474853515625, 29.6170654296875, 188.22024536132812, 7.799507141113281, 149.60202026367188, 91.4908447265625, 52.28643798828125, -53.854248046875, 7.59765625, 0.6537628173828125, 14.743804931640625, 87.90580749511719, 3.775482177734375, 103.04241943359375, 1.1759891510009766, 7.664777755737305, -50.32586669921875, -16.6318359375, 23.2642822265625, 129.8135986328125, -1.7451972961425781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000402.npy"}
{"epoch": 0.8418848167539267, "step": 403, "batch_size": 128, "mean": 81.94515991210938, "std": 85.20756530761719, "min": -156.67965698242188, "p10": -14.44401397705078, "median": 70.16524124145508, "p90": 188.52547607421874, "max": 242.045166015625, "pos_frac": 0.828125, "sample": [-16.763442993164062, 153.74502563476562, -8.423065185546875, 178.2998504638672, 159.92840576171875, 53.505126953125, 173.5697021484375, 124.8410873413086, 26.80535888671875, 27.95989990234375, 180.96612548828125, 188.32794189453125, 46.0333251953125, 48.4796142578125, 12.089370727539062, 84.1817626953125, 20.260421752929688, 112.51126098632812, 37.9615478515625, 9.439048767089844, 180.78810119628906, -18.886188507080078, -36.78631591796875, 33.61541748046875, 182.0947265625, 158.4198455810547, 186.46731567382812, 50.0194091796875, 28.87353515625, 206.992431640625, 216.89352416992188, 2.079376220703125, 1.3835296630859375, 137.6136932373047, 0.6412773132324219, 124.42241668701172, 16.96563720703125, 33.9727783203125, 0.0, -14.16485595703125, -0.027688980102539062, 169.88717651367188, 209.39755249023438, 111.3290786743164, 178.29342651367188, 173.96673583984375, 166.39260864257812, 21.47283935546875, 156.38339233398438, 214.78662109375, 16.85198974609375, 70.1380386352539, 8.011627197265625, 2.0019264221191406, 188.98638916015625, 14.34796142578125, 27.822479248046875, -54.466552734375, 118.12371826171875, -15.095382690429688, -24.232879638671875, 70.19244384765625, 2.7427406311035156, 191.06109619140625, 163.5420379638672, 22.00434112548828, 206.57394409179688, 1.3741912841796875, 38.48645782470703, 26.34259033203125, 21.7091064453125, 168.48794555664062, 30.158889770507812, 89.81494140625, -12.489471435546875, 44.643890380859375, 138.4283447265625, 178.65737915039062, 3.8557357788085938, 85.95962524414062, -1.4705390930175781, 57.84991455078125, 194.41995239257812, 14.43536376953125, 179.27105712890625, -20.508544921875, 10.609237670898438, 159.18551635742188, 144.26641845703125, 34.25483703613281, 234.123046875, 110.53561401367188, 242.045166015625, 204.2012939453125, 111.25401306152344, 202.26333618164062, -38.679443359375, 149.28240966796875, 167.57069396972656, 149.07443237304688, 144.58204650878906, 72.68215942382812, 162.8964080810547, 36.023773193359375, -156.67965698242188, 189.49107360839844, -23.2257080078125, 172.091552734375, -1.40081787109375, 156.39389038085938, -53.568206787109375, -4.181205749511719, 18.25927734375, 79.5093002319336, -22.46697998046875, 34.98304748535156, 32.48414611816406, 3.8841552734375, 163.496337890625, -8.429534912109375, -134.14190673828125, 165.98550415039062, 167.3768310546875, 91.84600830078125, 160.8120574951172, 135.17584228515625, 152.25286865234375, 117.82281494140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000403.npy"}
{"epoch": 0.8439790575916231, "step": 404, "batch_size": 128, "mean": 58.747230529785156, "std": 100.42316436767578, "min": -218.99069213867188, "p10": -36.343685913085935, "median": 30.819717407226562, "p90": 190.37378540039063, "max": 249.2103271484375, "pos_frac": 0.6875, "sample": [189.51809692382812, 182.42433166503906, 38.716522216796875, 200.2311248779297, -154.63571166992188, 181.61056518554688, 200.2027587890625, 0.2733612060546875, -43.277137756347656, 0.0, -163.40292358398438, 88.18131256103516, 28.184425354003906, -25.77984619140625, -83.89498901367188, 165.60340881347656, 5.85723876953125, 188.7352294921875, 2.1060562133789062, -56.2406005859375, 179.01486206054688, 108.35342407226562, 20.87896728515625, -7.74859619140625, -21.55670166015625, 199.75421142578125, -218.99069213867188, 171.24356079101562, -23.970359802246094, 5.315452575683594, -3.51898193359375, -38.707275390625, 90.31747436523438, 10.950408935546875, 102.64805603027344, -12.538948059082031, 3.9377174377441406, -12.047698974609375, 249.2103271484375, 136.42959594726562, 167.97430419921875, -4.9115753173828125, -8.719879150390625, -117.59967041015625, 163.01051330566406, 93.31585693359375, 72.07940673828125, 33.028656005859375, 3.997821807861328, 23.50970458984375, 123.08967590332031, 7.1156158447265625, 181.71475219726562, -24.7691650390625, 54.376434326171875, 208.4180908203125, -144.60812377929688, 17.709075927734375, -17.053924560546875, 18.4898681640625, 214.05340576171875, 193.76768493652344, -150.13427734375, 163.72787475585938, 112.00296020507812, 157.83370971679688, 165.99105834960938, -5.23681640625, -4.083984375, 172.4994659423828, 163.66183471679688, -0.34258460998535156, 13.078216552734375, -3.8803272247314453, -9.55242919921875, 144.1409912109375, 7.219451904296875, -19.525726318359375, 200.84341430664062, 46.326446533203125, 116.97285461425781, 15.732650756835938, 20.907196044921875, 64.629150390625, 5.802642822265625, -35.22367858886719, -18.177337646484375, -16.3984375, 196.00216674804688, -89.01171875, -7.2142791748046875, 149.44003295898438, 178.11004638671875, 42.4488525390625, -156.49362182617188, -33.617828369140625, 16.533401489257812, 12.639720916748047, 182.38262939453125, 77.64306640625, -35.330718994140625, 15.616485595703125, 192.37039184570312, 61.04736328125, 184.87796020507812, 120.20809173583984, 42.134727478027344, 62.752532958984375, -134.67507934570312, 197.85733032226562, 126.44235229492188, 215.575927734375, 183.48377990722656, 61.06202697753906, 28.61077880859375, 4.2404022216796875, 74.72129821777344, -11.625659942626953, 84.77587890625, 159.9312286376953, 27.990081787109375, -8.860511779785156, -4.961708068847656, 157.6669921875, 158.0023193359375, 141.7008819580078, 183.66873168945312, 211.3082275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000404.npy"}
{"epoch": 0.8460732984293193, "step": 405, "batch_size": 128, "mean": 64.2371826171875, "std": 92.03898620605469, "min": -169.734619140625, "p10": -31.19099426269531, "median": 44.04092025756836, "p90": 191.4483627319336, "max": 237.0301513671875, "pos_frac": 0.7578125, "sample": [16.8055419921875, 4.417938232421875, 10.179229736328125, 0.0, -42.6785888671875, -8.018524169921875, 10.03009033203125, 158.3582305908203, 72.97430419921875, 174.85171508789062, 3.708648681640625, 19.26617431640625, 46.26593017578125, 91.03724670410156, 155.999755859375, -82.86921691894531, -169.734619140625, 24.250518798828125, 44.52606964111328, 215.33135986328125, 67.34745788574219, 171.47915649414062, -3.9935264587402344, -15.917999267578125, 153.33071899414062, 4.399169921875, -23.319732666015625, 73.16583251953125, 191.8977508544922, 33.473854064941406, 143.8332977294922, 157.24014282226562, 5.194366455078125, -2.551422119140625, 61.51116180419922, 155.53846740722656, 43.55577087402344, -102.49063110351562, -26.719642639160156, 144.98745727539062, -2.814960479736328, 204.1951446533203, 159.09556579589844, 0.9279193878173828, -15.19781494140625, 237.0301513671875, 166.47647094726562, 120.56805419921875, 94.36184692382812, 79.36935424804688, 11.357391357421875, 80.173828125, 191.25576782226562, 139.35061645507812, 12.259994506835938, 0.78619384765625, 148.97421264648438, 172.04803466796875, 153.8391571044922, -25.8509521484375, -115.03482818603516, 126.51510620117188, 102.93038940429688, -15.062576293945312, 14.186370849609375, -6.594474792480469, 67.03863525390625, -7.550048828125, 19.5201416015625, 224.8442840576172, 78.00692749023438, 48.257080078125, 173.55072021484375, 161.597412109375, 211.0372314453125, 215.96771240234375, -34.240447998046875, 140.11160278320312, 8.630195617675781, 46.649871826171875, 220.084716796875, 12.385623931884766, -34.10577392578125, -92.823974609375, 169.42730712890625, 23.40743637084961, 0.01361083984375, 213.86196899414062, 1.7931976318359375, 103.64421081542969, 144.1250457763672, 23.73969268798828, -114.77412414550781, 7.861480712890625, 186.3294677734375, 204.36724853515625, 12.560638427734375, 6.920461654663086, 174.49624633789062, 142.70379638671875, -72.99142456054688, 194.51705932617188, 0.525726318359375, 176.56002807617188, 13.343719482421875, 66.2570571899414, 14.197147369384766, 178.73370361328125, 63.217498779296875, 38.229949951171875, 48.039031982421875, 1.4947738647460938, 188.85494995117188, -3.0750598907470703, -18.518325805664062, -13.94390869140625, -46.81175231933594, 6.0586090087890625, -29.941802978515625, 4.221202850341797, -81.58407592773438, 196.5963134765625, 229.04388427734375, -8.12762451171875, -87.88239288330078, 183.91705322265625, 104.68011474609375, 105.49859619140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000405.npy"}
{"epoch": 0.8481675392670157, "step": 406, "batch_size": 128, "mean": 55.69919204711914, "std": 101.89807891845703, "min": -205.89028930664062, "p10": -54.020271301269524, "median": 28.305984497070312, "p90": 181.10791931152343, "max": 230.63580322265625, "pos_frac": 0.71875, "sample": [157.4629669189453, -36.15375518798828, -39.23040771484375, 16.40283203125, -145.26287841796875, 211.22772216796875, 119.48001098632812, 165.58724975585938, 129.9145965576172, 176.3439178466797, -11.76416015625, 0.7062606811523438, -22.8614501953125, 192.0994873046875, 60.607177734375, -32.10426330566406, 210.48379516601562, -5.142463684082031, 130.28392028808594, 26.334182739257812, -141.41033935546875, -52.60960388183594, 18.714935302734375, 142.95684814453125, 165.83090209960938, 11.010086059570312, 137.76617431640625, 213.80621337890625, 30.6234130859375, 149.64447021484375, 59.239349365234375, 175.46669006347656, -174.37521362304688, -1.56280517578125, 17.816131591796875, -147.2432861328125, -16.218017578125, -6.288536071777344, -23.079795837402344, 172.23089599609375, -95.74148559570312, -50.30732727050781, -143.449951171875, -45.7447509765625, 30.2060546875, 180.9786376953125, 19.140396118164062, 158.20635986328125, 151.2874755859375, 177.63229370117188, 0.598175048828125, 22.50958251953125, 170.71327209472656, 127.24361419677734, -205.89028930664062, -9.130645751953125, 149.58157348632812, 181.40957641601562, 123.23816680908203, 21.7193603515625, 2.23089599609375, 169.366943359375, -136.63497924804688, 194.70977783203125, 24.389846801757812, 29.49725341796875, 10.09222412109375, 0.44781494140625, -31.81512451171875, 172.10736083984375, 230.63580322265625, 167.21957397460938, 11.322174072265625, 36.464111328125, 187.24395751953125, 35.571014404296875, 4.5428924560546875, 155.68038940429688, 15.41104507446289, 204.33123779296875, 174.33059692382812, -20.250282287597656, -161.11566162109375, 9.09756088256836, 170.81114196777344, 156.6138916015625, 125.5753173828125, -11.302703857421875, 27.114715576171875, 135.70407104492188, 69.7157974243164, -13.0263671875, 131.66249084472656, 149.48483276367188, -100.26958465576172, 223.97564697265625, 136.03292846679688, 4.257957458496094, 142.02488708496094, 62.3687744140625, 82.03207397460938, 159.24606323242188, 36.56280517578125, 161.51168823242188, -26.942596435546875, 59.73561096191406, -128.15988159179688, 201.401123046875, 3.8614501953125, 37.60516357421875, 16.829559326171875, 110.72463989257812, -40.485321044921875, 150.8717041015625, 219.2364501953125, -2.5845947265625, 2.2943038940429688, 15.294715881347656, -5.770355224609375, -21.772987365722656, -59.6519775390625, 21.69525146484375, 10.23394775390625, 50.70751953125, 201.86288452148438, -57.31182861328125, 15.29254150390625, 18.62311553955078], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000406.npy"}
{"epoch": 0.8502617801047121, "step": 407, "batch_size": 128, "mean": 61.099708557128906, "std": 96.4862060546875, "min": -167.18478393554688, "p10": -39.438534545898435, "median": 31.607295989990234, "p90": 185.63241882324218, "max": 240.67230224609375, "pos_frac": 0.703125, "sample": [-11.197303771972656, 28.76556396484375, 112.45892333984375, -14.666244506835938, 170.16433715820312, 12.900806427001953, 176.72601318359375, 91.47305297851562, -36.55303955078125, 20.29143524169922, -2.3909072875976562, -65.62631225585938, -146.78561401367188, -4.1241607666015625, -65.56021118164062, 147.97640991210938, -2.4625701904296875, -0.6317787170410156, 19.706298828125, -29.257476806640625, 154.45755004882812, -1.4264068603515625, -7.292205810546875, 151.9197998046875, 184.99581909179688, -42.81434631347656, 38.817535400390625, 112.36611938476562, -13.574714660644531, -13.52752685546875, -167.18478393554688, 24.252944946289062, -40.6929931640625, 38.808990478515625, 127.72889709472656, 190.13336181640625, 176.01605224609375, 189.14926147460938, -48.13392639160156, -2.300750732421875, 68.74240112304688, 25.2850341796875, 32.31231689453125, 4.821037292480469, -38.80975341796875, 212.56570434570312, 107.41345977783203, 120.3526611328125, 203.608154296875, 4.2864990234375, 140.2107391357422, 161.41574096679688, 9.091156005859375, 208.36837768554688, 12.1378173828125, 165.40475463867188, -92.60564422607422, -1.2538299560546875, 24.837860107421875, 131.71841430664062, 153.6451873779297, 26.55963134765625, 12.241180419921875, 11.959659576416016, 174.7725830078125, 87.0047607421875, -106.4998779296875, -20.248504638671875, 134.72567749023438, 185.08404541015625, 17.950653076171875, -31.991195678710938, 154.8892822265625, 177.21267700195312, 202.748291015625, 16.058879852294922, 4.44281005859375, 30.90227508544922, 0.0, 212.18032836914062, 174.86532592773438, 13.546104431152344, 166.67877197265625, -38.708343505859375, 190.01498413085938, -103.1829833984375, 202.50955200195312, 129.72470092773438, 17.611114501953125, 148.7667236328125, 178.86123657226562, 3.189117431640625, -16.5560302734375, 146.6804656982422, 35.20880126953125, 36.64752960205078, -38.900909423828125, 35.50750732421875, 40.210662841796875, 138.38424682617188, 180.33218383789062, -22.3826904296875, 211.12442016601562, -148.3887939453125, 76.14535522460938, -145.06951904296875, 100.98391723632812, 79.29388427734375, 181.18093872070312, 10.704391479492188, 164.43353271484375, 186.91195678710938, 54.52337646484375, -145.3645782470703, 196.11746215820312, 11.683624267578125, 39.59516906738281, -30.82109260559082, 10.697868347167969, 177.2259521484375, 161.07952880859375, 168.87054443359375, 22.113433837890625, 9.021646499633789, -5.2810821533203125, 240.67230224609375, 177.85336303710938, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000407.npy"}
{"epoch": 0.8523560209424084, "step": 408, "batch_size": 128, "mean": 66.41944885253906, "std": 93.96672058105469, "min": -210.58621215820312, "p10": -30.7677001953125, "median": 43.58264923095703, "p90": 182.26480560302736, "max": 215.08685302734375, "pos_frac": 0.7578125, "sample": [-3.8267593383789062, 9.122528076171875, -38.29015350341797, 80.26788330078125, 24.180519104003906, 173.906005859375, 25.009674072265625, 201.5118408203125, 189.810302734375, 173.08242797851562, -31.051498413085938, 22.270889282226562, 80.62493896484375, 134.87469482421875, -58.96588134765625, 171.33810424804688, 12.311004638671875, -16.97882080078125, -0.5873870849609375, 124.16864013671875, 1.38037109375, 19.045093536376953, 14.531829833984375, -10.715438842773438, 196.219482421875, 164.41656494140625, 198.16326904296875, 187.89532470703125, -8.271697998046875, 1.89141845703125, -21.909133911132812, 167.35910034179688, 150.59649658203125, 152.64889526367188, -39.13665771484375, 134.07516479492188, 208.81410217285156, 137.90684509277344, 170.65689086914062, 42.7498779296875, -12.388214111328125, -206.18206787109375, -210.58621215820312, 12.88409423828125, 194.21286010742188, 161.44761657714844, 4.482696533203125, 43.01145935058594, -28.86083984375, 173.0719757080078, 75.00747680664062, 134.957763671875, -7.92559814453125, -6.870697021484375, 105.05796813964844, 44.153839111328125, 151.96527099609375, 167.29507446289062, -29.125030517578125, 97.25819396972656, 174.61074829101562, 165.0442352294922, 214.632568359375, 12.311027526855469, -6.8243408203125, 177.45220947265625, -42.39837646484375, -30.646072387695312, 155.828125, -140.0238037109375, 185.43792724609375, 34.00738525390625, 147.63955688476562, 13.714111328125, 99.76307678222656, 177.76812744140625, -21.117767333984375, 129.3206024169922, 2.08245849609375, 27.046310424804688, 174.74627685546875, 97.0965576171875, 7.148521423339844, 41.17523193359375, -27.9959716796875, 2.975677490234375, 157.91461181640625, 167.06463623046875, 144.50946044921875, -61.21257019042969, 39.16729736328125, 15.347854614257812, 44.249542236328125, 172.30038452148438, 172.02450561523438, -2.9680328369140625, -175.05987548828125, 69.57598876953125, 182.41851806640625, 41.7291259765625, -55.31895446777344, 182.1989288330078, 18.13427734375, 62.31318664550781, 134.50564575195312, -7.393890380859375, 17.664993286132812, 10.483428955078125, 7.04656982421875, 49.070587158203125, 60.700401306152344, 19.962127685546875, 22.83966827392578, 0.0, 12.349197387695312, 215.08685302734375, 130.1784210205078, 1.7988605499267578, 137.70217895507812, -44.924835205078125, -125.24674987792969, 29.29876708984375, 180.2032470703125, 79.564453125, 201.48309326171875, 85.0576171875, 142.5396728515625, 192.5712890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000408.npy"}
{"epoch": 0.8544502617801047, "step": 409, "batch_size": 128, "mean": 67.12085723876953, "std": 102.78360748291016, "min": -178.14804077148438, "p10": -46.05863037109375, "median": 35.95757293701172, "p90": 191.27402343749998, "max": 260.33984375, "pos_frac": 0.7421875, "sample": [172.24285888671875, 151.76893615722656, 167.3311767578125, 88.73562622070312, 198.15380859375, -43.7967529296875, 138.748046875, 195.54443359375, 1.46173095703125, 42.697601318359375, 19.40319061279297, 3.376678466796875, 25.075942993164062, 5.42901611328125, 13.8883056640625, 217.07693481445312, 23.084083557128906, 22.0460205078125, 196.63165283203125, -164.421875, -12.950016021728516, -86.38595581054688, -1.1723556518554688, 183.67503356933594, 162.974853515625, -128.2781982421875, 16.89190673828125, -124.99227905273438, 186.1416473388672, -174.32489013671875, 189.12899780273438, 174.63046264648438, -48.179466247558594, 10.056671142578125, -6.411376953125, 204.4945068359375, 27.239013671875, 41.439208984375, 207.98382568359375, 187.0589599609375, 153.03704833984375, 180.89443969726562, -99.07620239257812, 31.896530151367188, 2.4388885498046875, 29.713836669921875, 117.27261352539062, 174.33541870117188, 10.205474853515625, 216.45330810546875, 172.29010009765625, 159.06390380859375, 133.793701171875, 162.3155517578125, 251.0567626953125, -45.423309326171875, -97.3035888671875, 92.13003540039062, 23.69720458984375, -10.992019653320312, 229.39825439453125, 138.4005126953125, 22.416549682617188, 116.488525390625, 168.33843994140625, 37.035797119140625, 22.448699951171875, -42.821380615234375, -4.8156585693359375, -41.95396423339844, 170.4781494140625, -29.228164672851562, -157.5833740234375, 188.2240447998047, 3.429920196533203, 126.153564453125, 201.60202026367188, 260.33984375, 78.69571685791016, 25.90045166015625, 5.458209991455078, -61.32207107543945, 125.80526733398438, 0.0, 112.37374877929688, 240.6029052734375, 34.87934875488281, 189.44384765625, 169.41360473632812, -162.72103881835938, -18.47863006591797, 168.51300048828125, 17.373291015625, -2.26409912109375, 56.19337463378906, -0.18710899353027344, 144.08734130859375, 171.01815795898438, 157.8802947998047, 180.13580322265625, 31.86743927001953, -13.013954162597656, 120.9844970703125, 84.1949462890625, 40.922454833984375, -3.9251327514648438, 22.660568237304688, -47.541046142578125, -2.5865936279296875, -11.798545837402344, 57.34260559082031, 3.7212629318237305, -178.14804077148438, 32.07118225097656, 164.19049072265625, 138.83615112304688, 8.036657333374023, 139.652587890625, 34.64381408691406, -23.77398681640625, 166.90521240234375, 11.765304565429688, 8.111892700195312, 145.2635040283203, -5.583038330078125, 146.17169189453125, 215.52206420898438, 122.52493286132812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000409.npy"}
{"epoch": 0.856544502617801, "step": 410, "batch_size": 128, "mean": 56.45503234863281, "std": 101.3469467163086, "min": -193.987060546875, "p10": -50.58290634155273, "median": 40.42856216430664, "p90": 190.05061035156248, "max": 253.67697143554688, "pos_frac": 0.6953125, "sample": [104.97894287109375, -73.14518737792969, 17.862632751464844, -41.08843994140625, -18.036022186279297, 164.5997314453125, 11.484878540039062, 25.188323974609375, -30.700897216796875, 112.36000061035156, 235.25210571289062, -24.432586669921875, 42.007408142089844, -82.8602294921875, 38.96271514892578, -167.78564453125, -17.1888427734375, 70.61636352539062, 179.05242919921875, 131.05831909179688, 196.40948486328125, -193.987060546875, 126.19131469726562, 13.491653442382812, 171.53871154785156, -9.1153564453125, 145.50396728515625, 86.85159301757812, 125.89323425292969, 152.6128692626953, 17.906219482421875, 130.7647705078125, 38.6116943359375, 169.4803466796875, 0.017459869384765625, 150.70474243164062, 23.07470703125, 56.386444091796875, 32.03826904296875, 187.32537841796875, 178.91546630859375, 57.255828857421875, -8.615249633789062, 176.18731689453125, -149.66192626953125, 150.42300415039062, -165.99832153320312, -14.284332275390625, -22.973403930664062, -55.014404296875, 113.193359375, 135.49331665039062, 96.41270446777344, -61.882476806640625, -9.665069580078125, 41.8944091796875, -5.494117736816406, 167.69432067871094, 50.084442138671875, 48.885589599609375, -29.253318786621094, 207.34100341796875, 44.15692138671875, 200.08319091796875, -3.58038330078125, 10.172836303710938, -29.606048583984375, -2.052001953125, 145.66195678710938, 93.62165832519531, 147.9537353515625, 167.3052978515625, 25.258052825927734, -10.86968994140625, -153.57015991210938, 0.0, 36.67121124267578, 25.6221923828125, 214.244140625, 187.1865234375, 212.25772094726562, 17.356170654296875, 16.622543334960938, -86.21083068847656, -37.31713104248047, 172.90591430664062, 18.330543518066406, 16.92205810546875, -29.723678588867188, 29.079795837402344, -154.2344512939453, 103.616455078125, 0.3486328125, 155.11952209472656, -26.007080078125, 216.30410766601562, -154.04922485351562, 185.230224609375, 4.775642395019531, 37.72773742675781, 78.43934631347656, -48.683692932128906, 136.83456420898438, -165.96743774414062, 59.836151123046875, 55.69819641113281, 59.90191650390625, 158.63980102539062, 239.74200439453125, 146.3256072998047, 198.11563110351562, -20.698028564453125, 146.60287475585938, 118.81182861328125, -0.188262939453125, 54.362060546875, 223.26873779296875, -25.75341796875, 47.91758728027344, 1.3669586181640625, 4.415163040161133, 197.85147094726562, 222.55462646484375, 18.81707763671875, 104.67903137207031, -35.211578369140625, 253.67697143554688, -3.2497386932373047], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000410.npy"}
{"epoch": 0.8586387434554974, "step": 411, "batch_size": 128, "mean": 68.26472473144531, "std": 95.99127197265625, "min": -215.80072021484375, "p10": -19.463177490234372, "median": 41.36842346191406, "p90": 186.71532287597657, "max": 336.23883056640625, "pos_frac": 0.7421875, "sample": [143.41384887695312, 180.53353881835938, 34.07366943359375, 155.4334716796875, 26.69049072265625, -1.0928192138671875, -1.9228363037109375, 7.689643859863281, 38.0660400390625, 150.30917358398438, 165.99530029296875, 152.5794677734375, -2.261974334716797, -18.5614013671875, -150.01998901367188, -153.03091430664062, 216.65850830078125, -13.99786376953125, 145.8438720703125, 156.36265563964844, 44.492706298828125, 64.0135498046875, 160.74256896972656, 98.30072021484375, 151.05442810058594, 20.232269287109375, 50.913482666015625, 336.23883056640625, 149.628173828125, -17.30657958984375, 192.40460205078125, 80.91351318359375, 23.100059509277344, 19.506423950195312, 146.19174194335938, 43.05218505859375, 188.40292358398438, 171.82794189453125, 20.5379638671875, 26.79293441772461, 8.372222900390625, 34.68290710449219, -5.4801177978515625, 72.40736389160156, 0.7112884521484375, -68.30579376220703, -1.73907470703125, -4.357364654541016, -17.27984619140625, 138.34580993652344, 154.96609497070312, 26.734405517578125, -14.190032958984375, 4.144252777099609, 15.611639022827148, 179.89105224609375, -31.52203369140625, 42.510894775390625, 27.4208984375, 117.49231719970703, -0.4713611602783203, 40.2259521484375, -10.10137939453125, -90.47689819335938, 158.28933715820312, 166.81271362304688, 136.1025390625, 181.1954345703125, 27.972335815429688, 203.57138061523438, 4.653865814208984, 149.8854522705078, 23.832977294921875, 189.75863647460938, -1.7942733764648438, 204.38131713867188, 26.204864501953125, 2.31500244140625, 171.22006225585938, 156.48776245117188, 25.73863983154297, -215.80072021484375, 191.46951293945312, 16.479923248291016, 10.986129760742188, -148.43069458007812, 37.35536193847656, -1.8127632141113281, 165.3633270263672, 215.23800659179688, 183.11306762695312, -123.3564453125, 30.52093505859375, 33.539886474609375, 88.18502807617188, 132.5804901123047, 131.69351196289062, 221.25381469726562, 153.9361572265625, -49.041168212890625, 185.9920654296875, -21.56732177734375, -79.92471313476562, -13.89630126953125, 9.637351989746094, 42.7723388671875, 153.20196533203125, 50.28662109375, 35.524505615234375, 141.29644775390625, -21.930328369140625, 65.62261962890625, 169.1731414794922, -7.962257385253906, 42.9222412109375, -8.789688110351562, 104.53225708007812, 226.2587890625, 158.16494750976562, 163.87814331054688, 163.46636962890625, 213.4106903076172, 138.68212890625, 194.06173706054688, 19.320465087890625, -7.47998046875, -92.59846496582031, -9.46820068359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000411.npy"}
{"epoch": 0.8607329842931937, "step": 412, "batch_size": 128, "mean": 56.34794616699219, "std": 102.26834106445312, "min": -201.91436767578125, "p10": -52.87414550781249, "median": 36.52009963989258, "p90": 183.2864990234375, "max": 263.1546325683594, "pos_frac": 0.671875, "sample": [-143.255615234375, 180.46624755859375, 6.284290313720703, 199.24151611328125, 68.6712646484375, 132.04196166992188, 40.564971923828125, 81.994384765625, -37.13005065917969, -16.883026123046875, 21.80230712890625, 37.615264892578125, -2.3197479248046875, -13.882715225219727, 23.5780029296875, -201.91436767578125, 0.5438385009765625, -27.433624267578125, -110.36204528808594, 30.8880615234375, 4.0044708251953125, 1.8408851623535156, 21.35198974609375, 87.67694091796875, 178.4115753173828, 174.17181396484375, 187.89959716796875, 37.286888122558594, -9.315399169921875, 11.302619934082031, 136.05332946777344, 139.62872314453125, -23.3956298828125, 176.57183837890625, 200.43325805664062, 175.34066772460938, 182.19024658203125, 17.003379821777344, 172.54373168945312, 74.95751953125, 149.01280212402344, -26.2657470703125, 188.56329345703125, -143.67364501953125, 99.7169189453125, -25.93963623046875, -4.2430572509765625, -32.516998291015625, -165.674072265625, 53.067047119140625, 116.43145751953125, 194.12820434570312, 217.8909912109375, -18.514129638671875, 183.96261596679688, 68.9718017578125, -9.06414794921875, 111.91732788085938, -87.85871887207031, -58.694732666015625, -112.37545776367188, -8.171133041381836, 225.90582275390625, 62.903076171875, -4.3467559814453125, 182.99673461914062, 144.46353149414062, 263.1546325683594, 42.156585693359375, 63.6156005859375, -116.00726318359375, 64.83000183105469, -50.379608154296875, -12.000335693359375, 21.32489013671875, 35.75331115722656, 212.7500457763672, 177.43057250976562, 34.749603271484375, -183.00653076171875, 0.81768798828125, 128.00437927246094, 176.02308654785156, -3.0394287109375, 186.42196655273438, 2.0998687744140625, 176.04302978515625, 172.494384765625, 197.2452392578125, -116.60206604003906, 17.782882690429688, 182.92422485351562, 164.28314208984375, -134.67050170898438, -6.561256408691406, 39.075469970703125, -15.464447021484375, 158.14535522460938, -12.33416748046875, 143.94442749023438, -7.264678955078125, -0.5465335845947266, 219.36495971679688, 125.528076171875, 160.96185302734375, 7.543478012084961, -36.757080078125, 129.76480102539062, 11.464553833007812, 138.43435668945312, 171.07144165039062, 28.326248168945312, 0.0, 51.68791198730469, -14.6806640625, 173.27398681640625, 8.386520385742188, 131.3770751953125, -120.70086669921875, 4.414356231689453, 33.47669982910156, 160.25558471679688, -49.037261962890625, 57.87249755859375, -13.025711059570312, -20.659423828125, 151.31393432617188, 180.62332153320312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000412.npy"}
{"epoch": 0.86282722513089, "step": 413, "batch_size": 128, "mean": 52.09455108642578, "std": 92.30155944824219, "min": -192.53225708007812, "p10": -43.99647521972656, "median": 20.941635131835938, "p90": 178.73829193115233, "max": 257.7968444824219, "pos_frac": 0.734375, "sample": [24.074644088745117, -22.029640197753906, -0.9535770416259766, 9.493568420410156, 8.017257690429688, 162.0577850341797, 62.9541015625, 19.23846435546875, -192.53225708007812, -42.342559814453125, 194.23870849609375, 38.43988037109375, 195.58026123046875, -54.610321044921875, 34.56671142578125, 19.593666076660156, 9.29107666015625, 190.54144287109375, 67.91555786132812, -141.06639099121094, 50.5107421875, 13.556285858154297, 52.76676940917969, 20.06573486328125, 28.763015747070312, -18.124473571777344, 8.708534240722656, 204.08404541015625, -1.6390323638916016, 120.77368927001953, 172.49581909179688, -35.661773681640625, -56.179290771484375, 166.22198486328125, -45.519073486328125, 168.22206115722656, 179.67127990722656, 160.51515197753906, 162.80673217773438, 21.88617706298828, 127.07048034667969, -4.783477783203125, -103.73757934570312, 4.41387939453125, 65.16204833984375, -31.1533203125, -106.51322174072266, 54.58045959472656, 197.70993041992188, 84.936279296875, -31.24884033203125, 6.157501220703125, -19.392288208007812, 114.052978515625, 24.13775634765625, 139.88949584960938, 178.33843994140625, -26.139755249023438, 43.881195068359375, 130.31883239746094, 184.07534790039062, 26.4346923828125, 15.855697631835938, 10.587997436523438, 38.13739013671875, 216.88555908203125, -123.54315185546875, 44.92437744140625, 167.46458435058594, -27.701171875, 163.20233154296875, 30.06871795654297, -60.79475402832031, 257.7968444824219, 12.499614715576172, -5.3291778564453125, -2.9400482177734375, 21.614089965820312, -20.405349731445312, 118.53216552734375, 6.9954986572265625, 164.68353271484375, -51.76678466796875, 162.6065673828125, 152.96041870117188, -25.4775390625, 18.7857666015625, 167.14657592773438, 2.6956214904785156, -43.34393310546875, 172.69935607910156, 37.266136169433594, -90.29788208007812, 153.5623779296875, 17.07672119140625, -21.254920959472656, 15.304214477539062, 16.951171875, 164.875244140625, 11.435775756835938, 10.065982818603516, -3.0247344970703125, 5.8196258544921875, 148.20578002929688, 118.48370361328125, 75.209716796875, 0.7151603698730469, 1.924774169921875, 139.98825073242188, 136.02362060546875, 20.269180297851562, 19.03802490234375, 141.27276611328125, 170.570556640625, -11.295013427734375, -123.40350341796875, 51.4561767578125, 182.081787109375, -148.23904418945312, 222.60101318359375, 1.062744140625, 199.20437622070312, 5.30413818359375, -0.2164897918701172, 195.30172729492188, 172.38238525390625, 17.379180908203125, 15.577392578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000413.npy"}
{"epoch": 0.8649214659685864, "step": 414, "batch_size": 128, "mean": 61.079620361328125, "std": 105.21012878417969, "min": -221.37762451171875, "p10": -47.468807983398435, "median": 37.70429992675781, "p90": 193.34490509033202, "max": 332.71917724609375, "pos_frac": 0.6953125, "sample": [135.8246612548828, -152.4447479248047, 82.36323547363281, 167.57345581054688, 1.2940521240234375, 156.69778442382812, 82.58502197265625, 96.7347412109375, 3.115966796875, -125.731201171875, -9.272598266601562, 9.060104370117188, 191.07601928710938, 19.1395263671875, 9.772445678710938, -1.8298282623291016, -43.265594482421875, -126.873779296875, 166.97845458984375, -8.750221252441406, -221.37762451171875, -102.55023193359375, 229.21881103515625, 7.5252685546875, 147.3154296875, -167.2888641357422, -31.456817626953125, -2.556652069091797, 159.08071899414062, 157.2161865234375, -7.957820892333984, -31.863067626953125, 154.01983642578125, 172.5670623779297, -58.5255126953125, 193.1848602294922, 159.3448486328125, -8.114334106445312, 215.30661010742188, -10.210113525390625, 24.64324951171875, 173.59390258789062, 69.13360595703125, -7.936279296875, 143.7504119873047, 71.65093994140625, 5.048095703125, -3.902008056640625, 40.64501953125, 216.96578979492188, -6.566131591796875, -10.419731140136719, -3.145244598388672, 219.83035278320312, 166.3660430908203, -12.16302490234375, 164.07398986816406, 193.6926727294922, 191.22671508789062, -98.45599365234375, 147.273193359375, 2.3490962982177734, 1.6296005249023438, 165.25128173828125, 7.2535400390625, 25.100494384765625, 117.87358856201172, 137.64064025878906, 57.34686279296875, 0.668914794921875, 166.6737060546875, -8.605384826660156, 38.3768310546875, 209.47894287109375, 160.13092041015625, 189.47772216796875, 168.37728881835938, 1.17083740234375, 39.427764892578125, 138.374755859375, -1.5334625244140625, -15.39019775390625, -149.55133056640625, 217.99295043945312, 332.71917724609375, 177.04759216308594, 167.322509765625, 124.23418426513672, 6.842811584472656, 158.66940307617188, -210.17144775390625, 9.595947265625, 37.031768798828125, 204.92584228515625, 15.266143798828125, 193.19586181640625, -48.64935302734375, -2.539154052734375, -14.602157592773438, 13.897422790527344, -143.4976806640625, 65.56240844726562, 238.8612060546875, 41.47209167480469, 7.585014343261719, 14.466903686523438, 166.67906188964844, 130.11386108398438, 123.44612121582031, 15.313385009765625, -46.962860107421875, 89.12964630126953, 214.92413330078125, 57.610931396484375, -3.731996536254883, 12.031951904296875, 135.54879760742188, 198.1656494140625, 25.33795166015625, 38.78102111816406, 24.792800903320312, -2.20623779296875, -58.370849609375, -3.695068359375, 91.978515625, 62.80108642578125, 205.64126586914062, -10.120620727539062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000414.npy"}
{"epoch": 0.8670157068062827, "step": 415, "batch_size": 128, "mean": 73.05023956298828, "std": 100.5458984375, "min": -188.7303466796875, "p10": -31.552714538574218, "median": 54.186553955078125, "p90": 193.60432128906248, "max": 266.42529296875, "pos_frac": 0.75, "sample": [17.625499725341797, 151.0735626220703, 13.369575500488281, 55.811279296875, -39.238037109375, -5.9309844970703125, 161.56390380859375, -0.8890380859375, 154.36886596679688, 11.432647705078125, 87.40656280517578, -1.5915336608886719, 195.2083740234375, 120.22537231445312, -27.24370574951172, -11.998733520507812, 188.09732055664062, 5.42950439453125, 158.979248046875, 61.08203887939453, 164.70913696289062, 22.381988525390625, 201.67724609375, 3.151275634765625, 168.6826934814453, 64.95654296875, 51.71476745605469, 174.95550537109375, 175.15756225585938, 145.26617431640625, -147.50271606445312, 207.1170654296875, 199.89627075195312, 25.05340576171875, -160.12698364257812, 23.83843994140625, 54.5152587890625, 188.01852416992188, -10.904830932617188, 128.68109130859375, 131.3138427734375, 15.264999389648438, 0.0135345458984375, 18.589080810546875, 45.979515075683594, -23.347274780273438, 171.91802978515625, 35.470458984375, 189.51708984375, 154.83934020996094, -13.340278625488281, 27.88397216796875, -176.5179443359375, 44.331695556640625, 166.0738067626953, 187.44094848632812, -68.74649047851562, 192.0528564453125, 47.02874755859375, -28.9423828125, 92.09713745117188, 171.0680694580078, -144.0823974609375, 142.7027587890625, 266.42529296875, -23.54888916015625, 28.470077514648438, 99.64420318603516, -56.07958984375, -0.7240047454833984, -31.378860473632812, 206.4078369140625, 157.58413696289062, 20.205368041992188, 16.31829833984375, -10.406951904296875, 174.1761474609375, 212.28387451171875, 134.96900939941406, -46.3970947265625, 76.40530395507812, 13.6549072265625, 133.39724731445312, -66.19158935546875, 192.9168701171875, -28.68994140625, -111.30994415283203, 150.23202514648438, 117.4375, -25.821868896484375, 2.2241973876953125, 2.5958251953125, 40.760711669921875, 168.01516723632812, 128.34832763671875, -31.9583740234375, 53.85784912109375, 151.12042236328125, 175.820068359375, -188.7303466796875, 212.71685791015625, 106.81417083740234, 181.4375, 15.531448364257812, 29.823455810546875, 202.615966796875, 2.4495849609375, -20.544815063476562, 178.02468872070312, 159.03321838378906, 161.89181518554688, 53.78966522216797, 164.7341766357422, 147.74705505371094, -117.80396270751953, 232.032470703125, 24.373336791992188, 208.28932189941406, 204.26788330078125, -6.958271026611328, 4.01959228515625, 22.773681640625, 171.66387939453125, -8.027481079101562, 178.9068603515625, 160.70913696289062, -11.103790283203125, 234.56134033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000415.npy"}
{"epoch": 0.8691099476439791, "step": 416, "batch_size": 128, "mean": 71.72634887695312, "std": 98.8337631225586, "min": -205.44891357421875, "p10": -30.914633178710936, "median": 57.05024719238281, "p90": 189.0691879272461, "max": 245.77520751953125, "pos_frac": 0.765625, "sample": [180.09744262695312, 217.25448608398438, 194.86273193359375, 25.885406494140625, 4.5941925048828125, 4.2897491455078125, -32.345672607421875, -5.78070068359375, -16.465103149414062, 134.31263732910156, -38.6173095703125, 245.77520751953125, 75.00616455078125, 41.52447509765625, 57.52435302734375, 38.978248596191406, 1.943878173828125, 197.6788330078125, 225.18017578125, 127.92340087890625, 145.91636657714844, 166.86395263671875, 54.558074951171875, 186.0532684326172, 3.974822998046875, 42.061492919921875, -30.30133056640625, 151.80853271484375, -128.42593383789062, 22.224929809570312, 147.44046020507812, 173.31637573242188, 159.255859375, -7.1923828125, 10.24221420288086, 189.38717651367188, -156.0810089111328, 180.37179565429688, 64.9886474609375, 20.47515869140625, 34.048431396484375, 45.44207000732422, -20.138824462890625, -205.44891357421875, -28.233867645263672, 197.114501953125, 27.913955688476562, 147.57899475097656, 151.26119995117188, 198.23114013671875, 40.68865966796875, 188.9329071044922, 14.135955810546875, 35.68012237548828, -35.438507080078125, 167.74679565429688, 187.21737670898438, 151.99114990234375, -5.7667999267578125, -12.342620849609375, 146.78500366210938, 60.0079345703125, -152.58502197265625, -25.18310546875, 217.15902709960938, 158.88540649414062, 8.841468811035156, 149.85528564453125, -17.252899169921875, 152.96017456054688, 0.8197860717773438, -33.98045349121094, 73.95843505859375, 173.5684814453125, 219.127685546875, 156.7000274658203, 99.75674438476562, 169.49395751953125, -90.92181396484375, 1.1202239990234375, 116.194091796875, 29.298583984375, 143.00286865234375, -20.305843353271484, -121.7750244140625, 124.60185241699219, 70.0430908203125, 101.12908935546875, -3.739410400390625, 23.905624389648438, 47.055015563964844, 146.26771545410156, 76.88336944580078, 147.79525756835938, -182.228759765625, 146.96180725097656, 56.576141357421875, 1.6334686279296875, 27.30364990234375, 178.03819274902344, -101.0411376953125, 161.979248046875, 44.672607421875, 12.0623779296875, 62.305267333984375, -12.9547119140625, 205.3748779296875, 143.79177856445312, 228.00921630859375, -12.191482543945312, 4.389610290527344, 216.17095947265625, 49.663482666015625, 27.4229736328125, -10.20758056640625, 169.20001220703125, 173.12350463867188, 172.88232421875, 104.89295959472656, -8.34384536743164, 27.612457275390625, -119.3668212890625, 183.46389770507812, -3.507129669189453, 169.00003051757812, 17.157272338867188, 162.52243041992188, 177.95680236816406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000416.npy"}
{"epoch": 0.8712041884816754, "step": 417, "batch_size": 128, "mean": 57.535972595214844, "std": 102.45470428466797, "min": -183.89859008789062, "p10": -77.4532485961914, "median": 45.126983642578125, "p90": 189.10315551757813, "max": 234.47311401367188, "pos_frac": 0.7109375, "sample": [-1.7854461669921875, 35.50721740722656, 39.20616149902344, 119.25926208496094, -30.516387939453125, 73.31912231445312, 33.251800537109375, -61.2520751953125, 199.3465118408203, 196.01507568359375, 26.081680297851562, 43.7034912109375, 4.184471130371094, 154.08456420898438, -167.5169677734375, 7.2464599609375, 9.315216064453125, -6.8502349853515625, -14.58099365234375, 168.21878051757812, 179.21221923828125, -28.518768310546875, 195.7203369140625, -109.21713256835938, -143.7505340576172, -160.094970703125, 121.58245849609375, 6.045280456542969, -43.17277526855469, 213.09835815429688, 74.7137451171875, -8.445098876953125, 14.15185546875, 96.89153289794922, 127.96266174316406, 187.5584716796875, 24.67413330078125, -133.23297119140625, 156.01878356933594, 148.17068481445312, 175.15728759765625, 52.16204833984375, 133.70252990722656, -6.28717041015625, 155.155517578125, 141.52487182617188, 16.715499877929688, 76.604736328125, 140.07687377929688, 69.99531555175781, 23.778261184692383, -0.3612937927246094, -77.28132629394531, -82.7862548828125, -53.890899658203125, 46.79109191894531, -2.1160888671875, 167.52963256835938, 77.81210327148438, -11.5086669921875, 107.592041015625, 9.94793701171875, 3.390239715576172, 14.686691284179688, 192.24862670898438, 0.7138671875, 176.32260131835938, -27.95880126953125, -183.89859008789062, 3.0594482421875, -28.919647216796875, 192.8699951171875, 100.5084228515625, 141.70745849609375, 55.32464599609375, -123.05474853515625, 115.98390197753906, 173.67462158203125, 0.0, -2.4017333984375, 118.83565521240234, 0.0, 44.165496826171875, 165.01132202148438, 129.66781616210938, 155.3774871826172, 107.97702026367188, 3.00445556640625, 124.1199951171875, -21.152313232421875, -102.88052368164062, 187.099365234375, 176.1448974609375, 168.67105102539062, 188.11456298828125, 71.714111328125, -180.60374450683594, 119.12052917480469, 231.701171875, 4.4131011962890625, 185.8990478515625, -152.23196411132812, -77.85440063476562, 23.817668914794922, 173.55441284179688, 192.1744842529297, 198.48074340820312, -53.40008544921875, -135.70303344726562, 8.3773193359375, 5.25244140625, -15.139602661132812, 6.774604797363281, 177.007080078125, 197.42205810546875, 188.28683471679688, -15.46759033203125, 50.663055419921875, 191.00790405273438, 7.401607513427734, 48.058197021484375, 46.088470458984375, 38.670143127441406, -4.44989013671875, 81.79244995117188, 146.02847290039062, 234.47311401367188, 216.94268798828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000417.npy"}
{"epoch": 0.8732984293193717, "step": 418, "batch_size": 128, "mean": 49.79821014404297, "std": 94.77705383300781, "min": -212.47042846679688, "p10": -56.16210327148437, "median": 30.429550170898438, "p90": 172.40072326660155, "max": 263.6909484863281, "pos_frac": 0.734375, "sample": [-30.1458740234375, 162.97824096679688, 34.0914306640625, 48.033966064453125, -104.515625, -159.0313262939453, 0.1961669921875, 201.10931396484375, -46.186859130859375, -212.47042846679688, 37.3614501953125, 82.28681182861328, 29.905059814453125, -77.5501708984375, 55.817596435546875, 90.42572021484375, 149.680419921875, -113.86013793945312, 124.61367797851562, 18.982330322265625, 181.91494750976562, 133.12353515625, -15.169189453125, 0.9767341613769531, -1.114013671875, 10.79351806640625, -18.401931762695312, 146.43130493164062, 36.233123779296875, 200.64556884765625, 112.20528411865234, 36.7381591796875, 174.14320373535156, 137.57823181152344, 17.406204223632812, 85.60354614257812, -25.14337158203125, 153.51589965820312, -48.051666259765625, 3.9231491088867188, -0.6018447875976562, 160.35488891601562, 5.61065673828125, 171.165771484375, -1.7703704833984375, 17.373443603515625, 1.9426097869873047, 65.70941162109375, 29.019439697265625, -22.53857421875, 2.4188079833984375, 178.8221435546875, 130.60604858398438, 26.126220703125, 16.9163818359375, -11.109710693359375, 8.9288330078125, 22.596923828125, 66.33287048339844, 131.29156494140625, -77.9365234375, 148.61056518554688, 181.16839599609375, 23.640731811523438, 35.1058349609375, 9.152336120605469, 157.65097045898438, -114.02909088134766, -153.98321533203125, 9.788764953613281, -41.82073974609375, 6.933464050292969, 152.8924102783203, 40.24224853515625, 60.182159423828125, 193.76589965820312, 130.173828125, -10.937957763671875, 87.488525390625, -54.890380859375, 42.114288330078125, 161.39794921875, -8.457122802734375, 61.834625244140625, 131.92236328125, 170.8819580078125, 162.0999298095703, -61.13434600830078, 3.2431640625, -59.12945556640625, 14.91839599609375, -33.664093017578125, -21.070159912109375, 149.29244995117188, 163.67469787597656, 17.18346405029297, 182.95208740234375, 121.69239807128906, 22.28015899658203, 170.98008728027344, -54.099700927734375, 1.3812103271484375, 8.962684631347656, 7.6699981689453125, -135.42889404296875, 41.35638427734375, 45.48211669921875, 154.96142578125, -189.15533447265625, 182.34674072265625, 53.4173698425293, 15.099700927734375, 51.76179504394531, -0.48413658142089844, 171.17025756835938, 30.95404052734375, 251.810546875, 178.9105224609375, 99.19027709960938, 236.67623901367188, 76.06210327148438, -15.168952941894531, -10.15289306640625, 17.567840576171875, -84.38777160644531, 171.65394592285156, 263.6909484863281, 12.441757202148438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000418.npy"}
{"epoch": 0.875392670157068, "step": 419, "batch_size": 128, "mean": 68.2968521118164, "std": 110.50336456298828, "min": -249.20079040527344, "p10": -64.89147644042967, "median": 53.06028747558594, "p90": 191.0975570678711, "max": 301.16217041015625, "pos_frac": 0.7578125, "sample": [76.6923828125, 12.395919799804688, 9.37042236328125, 185.04962158203125, 30.10272979736328, 26.605560302734375, 5.919406890869141, 17.153305053710938, 53.2158203125, -126.98110961914062, -119.96623992919922, 86.213134765625, -146.5205078125, 182.97274780273438, 2.9944839477539062, 26.75823974609375, 220.4951171875, 123.10301208496094, 160.61404418945312, 181.25051879882812, 190.39820861816406, 174.1187744140625, 192.7293701171875, 6.3791656494140625, 200.591796875, -5.686737060546875, 29.142501831054688, -0.73480224609375, -16.85009765625, -166.847900390625, 178.38836669921875, 165.3037109375, 152.67913818359375, -4.7919464111328125, 68.66563415527344, -30.039749145507812, 87.6734619140625, -0.4694175720214844, 167.38092041015625, -166.63267517089844, 186.5946044921875, 3.0694427490234375, 179.4500732421875, 171.94381713867188, 301.16217041015625, 10.088645935058594, 225.0134735107422, -184.16201782226562, 55.03302001953125, 218.86361694335938, 124.3358154296875, 87.952880859375, -19.837135314941406, 48.379539489746094, 35.03948974609375, 182.75686645507812, 223.484130859375, 179.17933654785156, 1.955810546875, -12.68841552734375, -134.3570556640625, 46.80633544921875, 0.01071929931640625, 2.4829978942871094, 244.10357666015625, -180.54107666015625, 170.37368774414062, 194.3669891357422, 12.198776245117188, 37.04945373535156, -28.953956604003906, 163.38284301757812, 156.1734619140625, 21.090667724609375, 121.02720642089844, 0.0, -105.14469909667969, -59.77008056640625, 16.804824829101562, 0.0, 24.487274169921875, 152.040283203125, 155.4717559814453, 188.860107421875, -11.470794677734375, 168.2748260498047, 26.280364990234375, 184.2078857421875, 98.42691040039062, 21.079483032226562, 187.05792236328125, 28.4840087890625, 189.35089111328125, 17.08648681640625, -5.7092132568359375, 15.09765625, 142.6595916748047, 196.77420043945312, 165.40956115722656, 187.204833984375, 211.27520751953125, 14.4686279296875, 236.3416748046875, 16.17236328125, 183.57644653320312, -42.30047607421875, 162.69937133789062, 166.5139617919922, 63.0443115234375, 159.16397094726562, -115.98675537109375, -76.84140014648438, 199.57327270507812, 164.9215545654297, -249.20079040527344, 70.2869873046875, -159.58041381835938, -16.724212646484375, 167.20907592773438, -24.1634521484375, 128.38319396972656, 170.67343139648438, -10.916900634765625, 21.561676025390625, 6.8453826904296875, 68.74393463134766, 118.7220458984375, 52.904754638671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000419.npy"}
{"epoch": 0.8774869109947644, "step": 420, "batch_size": 128, "mean": 65.5490951538086, "std": 99.06062316894531, "min": -189.24957275390625, "p10": -41.11048774719238, "median": 42.19407653808594, "p90": 187.01409301757812, "max": 237.492431640625, "pos_frac": 0.7734375, "sample": [-114.39266967773438, 15.113189697265625, 32.46337890625, -184.87628173828125, -25.93859100341797, -32.66999053955078, 14.440155029296875, 174.81729125976562, -52.504608154296875, 94.340087890625, 178.88833618164062, 6.946533203125, -162.57113647460938, 162.2880859375, 140.03077697753906, 11.447090148925781, 165.4818115234375, -145.35763549804688, 153.0008544921875, 106.81366729736328, 32.722137451171875, 55.349571228027344, -5.799507141113281, 45.6759033203125, 6.9979095458984375, 186.95562744140625, 13.500732421875, 18.812355041503906, 157.43792724609375, 35.585784912109375, 10.117767333984375, 227.4920654296875, 42.08290100097656, 172.08978271484375, 16.5946044921875, -3.40228271484375, 152.868408203125, 145.03466796875, 203.08450317382812, 14.326171875, 218.765869140625, 134.9259490966797, 18.96472930908203, 165.3519287109375, 21.3438720703125, 30.1917724609375, 167.42935180664062, -7.6580810546875, -6.1346282958984375, 111.79576110839844, -7.66778564453125, 136.05389404296875, 160.3271484375, 176.27932739257812, 147.95718383789062, 152.69720458984375, 0.0, 42.30525207519531, -39.976898193359375, -152.8245086669922, 148.99688720703125, 148.0718231201172, 2.208953857421875, 9.710281372070312, 13.674957275390625, 33.8707275390625, -5.230010986328125, 207.28274536132812, 223.469482421875, 153.54486083984375, 158.8985595703125, -128.38638305664062, 28.2010498046875, 74.53775024414062, -40.56341552734375, 194.05838012695312, 139.30674743652344, 23.3897705078125, 178.91952514648438, 31.00537109375, -14.388839721679688, 12.426048278808594, 237.492431640625, -49.123382568359375, 166.70616149902344, 93.25308227539062, 13.35614013671875, -42.38698959350586, 181.978759765625, 193.7440643310547, 93.32162475585938, 23.282928466796875, 187.1505126953125, 193.134521484375, 9.537139892578125, 12.394012451171875, 124.5850830078125, 61.51287841796875, 214.76422119140625, -7.77655029296875, 204.26870727539062, 137.22366333007812, -139.0941925048828, 0.1398162841796875, 27.272186279296875, 174.67398071289062, 92.27976989746094, 4.669189453125, -68.42892456054688, -104.905029296875, 93.15936279296875, 54.52410888671875, 134.02281188964844, 175.19554138183594, 174.1826171875, 23.00390625, 116.29721069335938, -189.24957275390625, 21.572540283203125, 229.98550415039062, 178.60879516601562, 88.633544921875, -38.228424072265625, -6.938039779663086, 4.831718444824219, 119.81027221679688, -0.20937347412109375, 47.636077880859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000420.npy"}
{"epoch": 0.8795811518324608, "step": 421, "batch_size": 128, "mean": 74.37167358398438, "std": 96.58332824707031, "min": -172.30108642578125, "p10": -31.138548278808585, "median": 73.727294921875, "p90": 188.05456085205077, "max": 224.62185668945312, "pos_frac": 0.734375, "sample": [205.3167724609375, 127.85243225097656, 139.62490844726562, 135.03924560546875, 153.54248046875, 129.9550018310547, 158.52056884765625, 11.64166259765625, -169.65084838867188, -6.890083312988281, 161.55612182617188, -40.20343017578125, 10.765899658203125, 129.26278686523438, 25.918212890625, 144.93502807617188, 36.0501708984375, -157.9293212890625, 33.056365966796875, 141.0479278564453, -85.67304992675781, -51.187599182128906, 116.31669616699219, 10.584510803222656, 158.26748657226562, 219.79019165039062, 180.69259643554688, 44.55169677734375, 10.935791015625, 119.3839340209961, 0.0, -97.8055419921875, 76.94644165039062, -5.7869720458984375, 9.961044311523438, 109.37576293945312, 186.4027099609375, 185.35012817382812, 23.239944458007812, 171.301025390625, 198.68875122070312, 166.72073364257812, 45.958740234375, 49.98405075073242, -7.129108428955078, 22.96442413330078, 137.26654052734375, -24.313812255859375, 215.52896118164062, 145.2659912109375, 147.46029663085938, -18.28832244873047, 135.32131958007812, 111.8804931640625, -19.11199951171875, 198.5140380859375, -136.47720336914062, 174.52027893066406, 9.567657470703125, 64.46726989746094, 40.785400390625, -58.804237365722656, 221.27877807617188, 204.26824951171875, 182.2844696044922, 95.05867004394531, 188.49546813964844, 177.665771484375, 172.6286163330078, 164.3654327392578, 173.6779327392578, 209.02560424804688, 117.8841552734375, 182.24307250976562, 137.79342651367188, 53.439697265625, -17.65850830078125, 26.514495849609375, 180.54354858398438, 13.782463073730469, 181.84454345703125, 13.369918823242188, 42.19427490234375, 185.781982421875, -5.3702392578125, 126.22024536132812, 35.62867736816406, -23.580612182617188, 163.00186157226562, -2.9688339233398438, 145.8555908203125, 146.38990783691406, 214.27960205078125, 127.1641845703125, -28.33499526977539, 70.50814819335938, 134.94598388671875, -134.9072265625, -172.30108642578125, 86.77239227294922, -12.19512939453125, -44.3507080078125, 31.05865478515625, -9.55474853515625, 17.49591064453125, 187.8656005859375, 7.222625732421875, -7.57830810546875, 117.05802917480469, -82.19033813476562, -28.536514282226562, 184.70648193359375, 37.73211669921875, 158.68399047851562, -23.491943359375, 198.71124267578125, 131.8519287109375, 0.0, -37.2099609375, -23.164154052734375, 224.62185668945312, 80.83905029296875, 28.426528930664062, 0.0, 31.4617919921875, -6.4736328125, 209.82095336914062, 4.148223876953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000421.npy"}
{"epoch": 0.881675392670157, "step": 422, "batch_size": 128, "mean": 49.56951141357422, "std": 104.28485870361328, "min": -202.38241577148438, "p10": -91.18880767822264, "median": 37.79603576660156, "p90": 186.58282318115235, "max": 226.21737670898438, "pos_frac": 0.703125, "sample": [17.16705322265625, -20.13794708251953, 177.10935974121094, -21.561233520507812, 189.71328735351562, -19.222389221191406, 173.71194458007812, -166.24569702148438, 25.047393798828125, 1.67340087890625, 145.5487060546875, 6.0330810546875, 47.446746826171875, 174.15673828125, 151.07217407226562, -200.06173706054688, 12.67437744140625, 14.19744873046875, -14.0509033203125, 137.74481201171875, 135.41665649414062, 38.999053955078125, 120.9429931640625, 22.315597534179688, -58.52398681640625, 150.6702880859375, -155.4356689453125, 140.77294921875, 143.8667449951172, 194.69622802734375, 44.6590576171875, 4.690185546875, -12.539321899414062, 72.20712280273438, 8.727672576904297, 151.85028076171875, 7.146629333496094, -129.89385986328125, 197.91737365722656, 10.7476806640625, 4.962005615234375, 180.3680877685547, -55.523712158203125, 36.593017578125, 42.3370361328125, -180.53912353515625, -12.074722290039062, 226.21737670898438, 46.07672119140625, 170.54550170898438, 41.017333984375, 201.525390625, 140.83078002929688, 10.27950668334961, 135.09518432617188, -47.541656494140625, 60.818115234375, 11.204462051391602, 14.8504638671875, 63.3922119140625, 148.15451049804688, -102.14653778076172, -31.383895874023438, -161.43087768554688, 205.36663818359375, 6.892822265625, 126.86153411865234, 12.540679931640625, 2.6642608642578125, 208.91802978515625, 93.51290893554688, -202.38241577148438, 179.4354705810547, -46.12567138671875, 66.18673706054688, -31.179424285888672, 161.9366912841797, -86.49263763427734, 188.85897827148438, 6.487270355224609, 212.2091064453125, -1.5617904663085938, 152.11126708984375, 43.38099670410156, 194.0455322265625, 60.018096923828125, -125.46420288085938, 4.855377197265625, 188.01327514648438, -0.04309844970703125, -15.478462219238281, -19.092041015625, 41.438079833984375, -25.81573486328125, 160.21185302734375, 170.04763793945312, -70.79168701171875, -4.206390380859375, 54.1375732421875, 140.07608032226562, 104.83293151855469, 15.137405395507812, 213.4013671875, 133.831298828125, -113.03469848632812, 156.0911865234375, 68.95443725585938, -164.9608154296875, -22.1038818359375, 105.88055419921875, 0.0, -137.091064453125, 185.9697723388672, 157.9429168701172, -135.61614990234375, 34.435302734375, 104.49758911132812, 149.68914794921875, 39.056358337402344, -1.9542312622070312, -35.34315490722656, 116.30841827392578, 4.6014251708984375, 184.1060333251953, 29.712051391601562, 31.504547119140625, -0.3175849914550781, 202.91549682617188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000422.npy"}
{"epoch": 0.8837696335078534, "step": 423, "batch_size": 128, "mean": 70.97218322753906, "std": 104.08800506591797, "min": -173.39947509765625, "p10": -41.693487548828124, "median": 48.88043212890625, "p90": 197.9728973388672, "max": 224.32586669921875, "pos_frac": 0.765625, "sample": [1.579681396484375, 196.6685028076172, -17.870346069335938, 9.87115478515625, 176.71600341796875, 157.30746459960938, 8.468814849853516, -13.9923095703125, 2.5639305114746094, 161.72445678710938, 113.7872314453125, 168.47378540039062, 33.73822021484375, 151.75759887695312, -152.4072265625, -11.809173583984375, 173.98382568359375, -133.01148986816406, 200.51708984375, 17.196762084960938, 6.2033538818359375, -32.909881591796875, 23.959243774414062, 159.37399291992188, -159.38092041015625, 109.34939575195312, 134.70106506347656, 13.104476928710938, 170.0786590576172, 199.44259643554688, 4.62188720703125, 24.41717529296875, 79.888671875, 28.379638671875, 125.89370727539062, 149.89151000976562, 190.34353637695312, 5.2633819580078125, -22.06549072265625, 145.25289916992188, 28.561264038085938, 187.71914672851562, -11.919906616210938, 182.49365234375, -1.9526214599609375, 26.666748046875, 9.593719482421875, 215.7939453125, -43.49676513671875, -39.09979248046875, 197.84835815429688, -40.920654296875, -63.97930908203125, 65.12408447265625, 188.38616943359375, -38.552825927734375, -126.63473510742188, -158.97909545898438, 192.65011596679688, 180.34609985351562, 36.563751220703125, 0.1078033447265625, 166.08108520507812, 11.90386962890625, -12.798622131347656, 137.18203735351562, 1.1603717803955078, 185.38491821289062, 10.102424621582031, 29.76031494140625, 157.8414306640625, 171.06100463867188, 138.75408935546875, 152.3746337890625, 23.944778442382812, 65.269287109375, -60.950286865234375, 183.89752197265625, 218.12484741210938, 73.75967407226562, 17.40753173828125, 192.7965087890625, 17.6297607421875, 209.23428344726562, 203.3380126953125, 61.94367980957031, 186.42523193359375, 25.563766479492188, 0.7527313232421875, 206.38583374023438, 156.39576721191406, 168.05723571777344, 208.73712158203125, -11.377166748046875, 217.00924682617188, 147.14117431640625, 3.2682857513427734, -58.906402587890625, 198.26348876953125, -6.68597412109375, 2.014678955078125, 150.97412109375, 40.12646484375, 202.39822387695312, 224.32586669921875, -39.5443115234375, -5.0826263427734375, 145.02532958984375, 202.8771514892578, -16.960906982421875, -155.1314697265625, 41.017784118652344, -106.88082885742188, 42.477294921875, 17.383365631103516, 186.43145751953125, 55.2835693359375, 158.53512573242188, 188.089111328125, 83.4344482421875, 163.38189697265625, 158.08132934570312, 152.52508544921875, -165.43865966796875, 33.368011474609375, 117.86807250976562, -2.366586685180664, -173.39947509765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000423.npy"}
{"epoch": 0.8858638743455497, "step": 424, "batch_size": 128, "mean": 69.75753784179688, "std": 90.74818420410156, "min": -182.62191772460938, "p10": -35.123121643066405, "median": 58.92888641357422, "p90": 185.82355194091795, "max": 233.98858642578125, "pos_frac": 0.7578125, "sample": [80.34053039550781, 54.48040771484375, 0.046356201171875, -9.143798828125, 83.36634826660156, 12.80438232421875, -6.304107666015625, -5.3756103515625, 34.97654724121094, -5.127105712890625, 4.40521240234375, -136.7210693359375, 233.98858642578125, 55.65892028808594, -6.202362060546875, 117.7629623413086, 18.4708251953125, 207.7766876220703, 175.3414306640625, 160.60372924804688, -7.2649993896484375, 5.2171630859375, 206.88516235351562, 160.1058349609375, 112.35081481933594, 59.237030029296875, -0.8929405212402344, -54.184967041015625, 133.5386962890625, 136.4697265625, 115.1777572631836, 209.89019775390625, 165.59564208984375, -71.92034912109375, 142.269775390625, 157.1057891845703, 36.459228515625, 160.52786254882812, 187.70205688476562, -70.6717529296875, 27.1126708984375, 174.1536865234375, 171.10617065429688, -182.62191772460938, 0.5895919799804688, 203.33642578125, 191.26260375976562, 68.91238403320312, -165.7004852294922, 107.4969482421875, -28.062816619873047, 2.637939453125, -2.76837158203125, 179.13743591308594, 64.13568115234375, 154.79962158203125, 20.505096435546875, 170.31072998046875, 72.66795349121094, 77.12161254882812, 36.87605285644531, -6.810089111328125, 190.8660888671875, 166.155029296875, 206.06390380859375, 150.12454223632812, 27.4464111328125, 35.6173095703125, -60.67333984375, -34.247039794921875, -38.187896728515625, 76.7626953125, 150.90379333496094, 207.135498046875, 219.11749267578125, 169.80593872070312, -3.6278610229492188, 44.34656524658203, 156.74327087402344, 120.9952392578125, 49.936553955078125, 158.12680053710938, -2.9550018310546875, 3.9561004638671875, 185.0184783935547, 96.88462829589844, -23.0552978515625, 0.0, -37.16731262207031, 33.65666198730469, 13.117218017578125, 150.49105834960938, 164.69131469726562, -9.276641845703125, 125.82164764404297, 170.0917510986328, 101.52606201171875, 164.23141479492188, 165.41610717773438, 147.48065185546875, 27.017799377441406, -79.71481323242188, 141.0848388671875, 154.92105102539062, 135.93374633789062, 1.528177261352539, -8.975078582763672, 12.0418701171875, -20.208770751953125, 6.839458465576172, 140.85211181640625, -65.7656021118164, 217.50494384765625, 161.15646362304688, -38.05828857421875, 7.903474807739258, 16.79524803161621, 70.2344970703125, 30.07830810546875, 70.43206787109375, 26.748870849609375, 58.62074279785156, -100.97897338867188, 33.83518981933594, 79.0245361328125, 15.980781555175781, 1.4152946472167969, 198.43081665039062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000424.npy"}
{"epoch": 0.8879581151832461, "step": 425, "batch_size": 128, "mean": 61.92911911010742, "std": 96.95599365234375, "min": -223.967529296875, "p10": -30.21939544677734, "median": 27.0521240234375, "p90": 189.31590728759764, "max": 241.54205322265625, "pos_frac": 0.7109375, "sample": [162.57162475585938, 167.91494750976562, 54.75677490234375, 85.28352355957031, -12.357421875, 15.3800048828125, 45.4915771484375, 160.82913208007812, 38.95182800292969, -15.542984008789062, 174.24960327148438, 22.634674072265625, 185.44705200195312, 12.315460205078125, 147.819580078125, -0.747802734375, 136.8607177734375, 157.806884765625, 1.500478744506836, 25.0941162109375, 92.20440673828125, 158.56527709960938, 12.131561279296875, -3.166950225830078, 169.7152099609375, 113.61439514160156, 4.337872505187988, 21.4820556640625, -66.35813903808594, -4.453971862792969, 45.6685791015625, 6.53643798828125, -20.66827392578125, 15.043121337890625, 203.5811767578125, -5.220546722412109, -29.298736572265625, 205.29803466796875, 7.60479736328125, 15.032798767089844, -22.25146484375, 207.7041015625, 155.59188842773438, 33.055694580078125, 7.5411376953125, 5.931877136230469, -40.52449035644531, 0.901824951171875, -166.38919067382812, 239.64117431640625, 81.80816650390625, 156.49038696289062, -105.49468994140625, 48.5948486328125, 185.06646728515625, -6.9161224365234375, 159.910888671875, 238.07415771484375, 39.1080322265625, 27.212127685546875, -39.35809326171875, 26.892120361328125, 8.02291488647461, 43.15275573730469, 194.90081787109375, -48.16375732421875, 142.4621124267578, -4.170352935791016, 178.09080505371094, 156.83056640625, 155.44912719726562, 11.12005615234375, 215.297119140625, 7.130151748657227, -16.104034423828125, -5.894401550292969, -12.67794418334961, 31.60498046875, -19.04723358154297, 116.35010528564453, -0.2374725341796875, 183.2552490234375, 29.675384521484375, 165.98089599609375, 190.84695434570312, -126.16820526123047, 15.858467102050781, -32.36759948730469, 23.342300415039062, -22.3079833984375, 18.428466796875, -24.978633880615234, 117.46080780029297, -121.63291931152344, -42.3779296875, -56.122222900390625, 125.72100830078125, 1.5535354614257812, 153.95504760742188, -149.20428466796875, 46.92181396484375, 241.54205322265625, 14.58734130859375, 170.29493713378906, 122.34571838378906, 124.96327209472656, 180.4197998046875, 5.5718994140625, 6.6781005859375, 169.80963134765625, 0.0, -11.950042724609375, 168.60894775390625, 207.22247314453125, 231.45498657226562, -4.620912551879883, 25.477622985839844, 175.4287872314453, -223.967529296875, 48.48779296875, 209.18634033203125, 188.6597442626953, 142.11117553710938, 173.30239868164062, 198.41983032226562, -20.09412384033203, -8.271347045898438, -1.1975631713867188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000425.npy"}
{"epoch": 0.8900523560209425, "step": 426, "batch_size": 128, "mean": 64.90850830078125, "std": 89.16984558105469, "min": -161.35943603515625, "p10": -23.824724578857417, "median": 33.33757019042969, "p90": 187.40202331542966, "max": 223.41336059570312, "pos_frac": 0.78125, "sample": [159.106201171875, 16.854141235351562, 2.2070446014404297, -37.61663818359375, 205.60470581054688, 25.896896362304688, 132.84164428710938, 1.8787689208984375, -146.64151000976562, 47.10552978515625, -112.67041015625, -46.8885498046875, -49.19203186035156, 158.88250732421875, 167.28955078125, 178.48196411132812, 167.62603759765625, 26.6893310546875, 153.04379272460938, 6.92578125, -42.719879150390625, 49.67059326171875, 199.71011352539062, 21.783615112304688, 4.683349609375, 24.433929443359375, 185.26980590820312, 198.8367919921875, 156.67724609375, 192.377197265625, 137.9158477783203, 4.093036651611328, 206.524658203125, 116.60095977783203, 169.52310180664062, 3.191741943359375, 0.066253662109375, 26.105255126953125, 182.42474365234375, 192.85499572753906, 43.50701904296875, 198.16847229003906, 8.786972045898438, -4.295440673828125, 174.361083984375, 177.08505249023438, -2.671142578125, 134.0354766845703, 18.4287109375, 13.3177490234375, -121.09162902832031, 31.2562255859375, 92.51641845703125, 5.247283935546875, 29.407588958740234, 176.84262084960938, -41.866363525390625, 37.57677459716797, 174.33160400390625, 145.57359313964844, 147.64999389648438, 163.1790008544922, -32.509613037109375, -3.27362060546875, -1.9115142822265625, 85.51861572265625, 59.62385559082031, 150.10382080078125, 0.382568359375, -9.283203125, 69.2557373046875, 25.215927124023438, -16.34270477294922, 17.19818115234375, -6.410619735717773, 214.05487060546875, -27.87779998779297, -6.4160308837890625, 36.74888610839844, 156.14642333984375, -52.44316864013672, 167.0413818359375, 68.6541748046875, 132.2973175048828, 108.67425537109375, 35.418914794921875, 23.8009033203125, 24.70447540283203, 17.571666717529297, 156.6715087890625, 92.57464599609375, 20.560760498046875, 1.430196762084961, -22.087692260742188, 196.35098266601562, 37.832305908203125, 5.3651123046875, 193.59951782226562, 2.1728515625, -14.719793319702148, 114.5042953491211, 4.45648193359375, 0.0, 10.881088256835938, 16.502197265625, 207.91824340820312, 78.01495361328125, 3.453704833984375, 4.203125, 96.77732849121094, 154.64688110351562, 28.76708984375, -132.45849609375, -3.685791015625, -8.558052062988281, 150.2183837890625, -21.851715087890625, 178.79244995117188, 163.2906494140625, 120.82568359375, 138.59751892089844, 4.789009094238281, -161.35943603515625, 122.24371337890625, 223.41336059570312, 39.942474365234375, -18.11328125, 197.5882110595703], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000426.npy"}
{"epoch": 0.8921465968586387, "step": 427, "batch_size": 128, "mean": 83.62887573242188, "std": 94.12740325927734, "min": -174.37261962890625, "p10": -23.77573547363281, "median": 70.50328063964844, "p90": 195.47029876708984, "max": 331.8133544921875, "pos_frac": 0.7890625, "sample": [18.943275451660156, 205.24920654296875, -8.048202514648438, -28.214599609375, -105.10488891601562, 13.413009643554688, 162.27610778808594, 115.3128662109375, 160.74612426757812, 56.32893371582031, 127.54693603515625, 35.73872375488281, 202.75961303710938, 18.93170928955078, 171.76025390625, 187.89480590820312, 170.4463348388672, 6.6757354736328125, 160.913818359375, 206.42819213867188, 202.48806762695312, 51.92948913574219, -19.192581176757812, 136.94644165039062, 168.3646240234375, 161.63882446289062, 146.058349609375, 93.1263427734375, 36.34409713745117, 49.644187927246094, 229.4393310546875, 192.51266479492188, 120.50897216796875, 43.526947021484375, 0.25506591796875, 3.853717803955078, 137.3396453857422, 7.269493103027344, -2.488037109375, -33.750457763671875, 189.23118591308594, 20.767913818359375, -82.64875793457031, 75.12945556640625, -142.73843383789062, 157.81289672851562, -2.7155914306640625, 194.7900848388672, -25.492599487304688, 164.62692260742188, 194.23684692382812, 32.3411865234375, 162.15492248535156, 79.3748779296875, 208.98419189453125, 191.4720458984375, 222.73672485351562, -10.308914184570312, 13.226303100585938, -11.826087951660156, 158.29441833496094, 43.771881103515625, 213.68704223632812, 21.36688232421875, 118.89523315429688, 58.58154296875, -19.245330810546875, 165.19097900390625, 165.04408264160156, -104.63240051269531, -174.37261962890625, 187.85894775390625, 124.14249420166016, -23.240509033203125, 133.91357421875, 21.044164657592773, -25.02459716796875, 162.88778686523438, 138.6775360107422, 35.48509216308594, 234.48379516601562, -42.061737060546875, 164.0066375732422, -32.5213623046875, 16.442337036132812, 70.2039794921875, 164.8079376220703, 70.80258178710938, 100.23258972167969, 111.26214599609375, 152.66714477539062, 189.32705688476562, -30.333984375, 11.4110107421875, 178.15350341796875, 160.12661743164062, 197.05746459960938, 52.67340087890625, -0.7375411987304688, 140.34046936035156, -14.905975341796875, 14.007644653320312, 43.15582275390625, -27.795700073242188, 8.347549438476562, 331.8133544921875, -11.105209350585938, 154.16159057617188, 2.7005615234375, 180.94549560546875, 198.55694580078125, 36.81135559082031, 152.99794006347656, -5.695709228515625, 49.22613525390625, -11.171554565429688, 176.2833251953125, 65.8623046875, 1.0783233642578125, 3.83184814453125, 182.06427001953125, 27.774002075195312, 0.052093505859375, -11.930370330810547, 34.439453125, 181.24325561523438, 213.47747802734375, 180.63182067871094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000427.npy"}
{"epoch": 0.8942408376963351, "step": 428, "batch_size": 128, "mean": 79.72618103027344, "std": 97.15668487548828, "min": -182.7711181640625, "p10": -11.580528259277342, "median": 64.38313293457031, "p90": 186.4956787109375, "max": 282.3045959472656, "pos_frac": 0.796875, "sample": [23.859832763671875, -11.12677001953125, -138.3975372314453, 179.41473388671875, 204.87979125976562, -19.050384521484375, 56.3616943359375, 61.57611083984375, -117.62222290039062, 8.991592407226562, 282.3045959472656, 23.7557373046875, 84.08392333984375, 148.35806274414062, -9.0001220703125, 179.3058624267578, 24.30029296875, -8.25537109375, 17.306650161743164, -7.837066650390625, 209.05718994140625, -8.269561767578125, -174.03225708007812, 212.35220336914062, 185.95101928710938, 205.4935302734375, 169.81773376464844, 160.5377197265625, 148.67295837402344, 249.6422119140625, 42.72001647949219, 193.9652099609375, 2.594900131225586, 102.1998291015625, -17.57391357421875, 14.908954620361328, -1.709625244140625, 152.55667114257812, 38.366455078125, 161.969970703125, 56.49053955078125, -5.567878723144531, -156.13265991210938, 162.00238037109375, 174.2884521484375, 214.72427368164062, -120.9396743774414, 31.677757263183594, 4.880340576171875, 14.096817016601562, 168.1338653564453, 76.09326171875, 149.30270385742188, 120.70378112792969, 27.870635986328125, 152.00125122070312, 168.40756225585938, 125.78729248046875, -40.655029296875, 235.66436767578125, -124.99659729003906, 215.59280395507812, 165.60011291503906, 47.35675048828125, 174.4757080078125, 6.0648193359375, 150.90692138671875, 184.30931091308594, 86.21537780761719, 50.03936767578125, 19.01647186279297, 33.94325256347656, 178.14564514160156, 167.90896606445312, 279.58154296875, 15.41046142578125, 48.954925537109375, -2.699188232421875, 94.92166137695312, 21.613128662109375, 48.68647003173828, 15.949462890625, 170.57554626464844, 67.19015502929688, -10.039199829101562, 4.581939697265625, -19.15300750732422, 27.036865234375, 125.46249389648438, 144.24440002441406, 272.02386474609375, 184.2493133544922, 185.93710327148438, 5.537109375, 90.5406494140625, -182.7711181640625, -18.216644287109375, 186.73553466796875, 115.25360107421875, 0.0, 26.584136962890625, 178.6414337158203, 98.82731628417969, 14.717544555664062, 175.54629516601562, 0.45920562744140625, -12.639297485351562, 146.0357666015625, 142.40972900390625, 57.22955322265625, 24.045623779296875, 154.67401123046875, 25.512863159179688, 85.24608612060547, 177.62380981445312, 79.8145751953125, 19.626922607421875, 131.858642578125, 140.9039306640625, -1.942718505859375, 37.462554931640625, 30.264404296875, 186.39288330078125, -0.5820980072021484, -2.934417724609375, 170.08209228515625, 153.50140380859375, 148.14315795898438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000428.npy"}
{"epoch": 0.8963350785340314, "step": 429, "batch_size": 128, "mean": 62.12968444824219, "std": 111.98297119140625, "min": -211.1816864013672, "p10": -88.7017852783203, "median": 38.9014892578125, "p90": 201.02444458007812, "max": 232.093505859375, "pos_frac": 0.6953125, "sample": [173.0671844482422, -207.16439819335938, -10.924407958984375, 141.74337768554688, 178.97862243652344, 175.4627227783203, 26.328163146972656, 28.94549560546875, -59.94340515136719, 142.36996459960938, 5.9146575927734375, -165.86573791503906, 4.02972412109375, 169.97171020507812, 205.20526123046875, 171.37217712402344, 1.3667354583740234, 166.67169189453125, -25.3741455078125, 25.39044189453125, 138.1715545654297, 148.52447509765625, 6.168487548828125, 176.78750610351562, -146.36178588867188, 195.833984375, 134.81747436523438, 211.75364685058594, 16.54425048828125, 89.9306640625, -145.23361206054688, 155.2342987060547, 151.10699462890625, 175.73873901367188, 32.03477478027344, 32.360076904296875, -136.31658935546875, 7.506866455078125, 178.30303955078125, 232.093505859375, 185.80908203125, 166.68341064453125, -108.16691589355469, 116.60333251953125, 114.3658447265625, -6.9854278564453125, 216.6629638671875, 119.040283203125, 8.476402282714844, -45.2393798828125, 5.695085525512695, 206.45681762695312, 209.7724609375, 184.40185546875, 13.315994262695312, 53.903106689453125, 222.86764526367188, -114.23316955566406, -97.82354736328125, 170.67953491210938, -5.3619384765625, 45.442901611328125, 183.68612670898438, 198.69827270507812, -33.3507080078125, 18.089805603027344, 53.761962890625, 82.5936279296875, 28.116737365722656, 78.49313354492188, -21.7464599609375, -93.46871948242188, -44.469207763671875, -41.41093444824219, 163.3990478515625, -211.1816864013672, 193.5146484375, -14.417015075683594, 17.23187255859375, 185.40748596191406, -3.799795150756836, 0.0, 179.45556640625, 219.885009765625, 188.01596069335938, 58.865814208984375, 181.2346649169922, -4.683403015136719, -4.0729217529296875, 6.262351989746094, 209.46380615234375, -30.01336669921875, 142.4378204345703, -29.20037841796875, 10.893325805664062, 57.237518310546875, 21.29513168334961, 3.3574447631835938, 2.253387451171875, 131.33352661132812, -149.20208740234375, -13.50689697265625, -28.397003173828125, 20.254013061523438, -40.14044189453125, 76.31939697265625, 147.39840698242188, 202.508056640625, -176.738525390625, 152.31356811523438, 204.75625610351562, 166.17913818359375, -50.35331726074219, 208.08106994628906, 197.9583740234375, -23.44781494140625, 9.543310165405273, 4.382843017578125, 214.69012451171875, 200.38861083984375, -86.6588134765625, 153.3568115234375, -19.13983154296875, -29.433364868164062, 102.13546752929688, -1.49078369140625, 123.73529052734375, -160.9725341796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000429.npy"}
{"epoch": 0.8984293193717278, "step": 430, "batch_size": 128, "mean": 53.741641998291016, "std": 106.35418701171875, "min": -190.0936279296875, "p10": -102.39556884765624, "median": 31.439102172851562, "p90": 191.96429138183592, "max": 237.62506103515625, "pos_frac": 0.7109375, "sample": [-152.75213623046875, -19.6536865234375, -156.9247589111328, 5.96282958984375, 138.70187377929688, 143.3544921875, -165.7521209716797, 181.386474609375, -8.84576416015625, 56.80363464355469, 155.79443359375, 14.892436981201172, 31.616607666015625, 15.478683471679688, 97.27705383300781, -190.0936279296875, 198.52935791015625, 23.0968017578125, 1.8397293090820312, 165.57330322265625, 141.89617919921875, -115.34237670898438, -31.339035034179688, 144.45022583007812, 10.76904296875, 30.513153076171875, -44.65753173828125, 0.0, 191.45285034179688, 82.379638671875, 110.9380111694336, -23.92816162109375, 167.28472900390625, 58.99085998535156, 20.328529357910156, 0.0, 91.239501953125, -2.3773651123046875, 4.06439208984375, 151.52029418945312, -112.6097412109375, 209.92047119140625, -98.01806640625, 181.47640991210938, 201.46377563476562, 135.0665283203125, 37.557769775390625, 23.694000244140625, -25.55743408203125, 237.62506103515625, 155.27032470703125, -14.03955078125, 50.33271789550781, 17.84698486328125, 175.804443359375, 151.49920654296875, 43.00236511230469, -151.59591674804688, 190.66973876953125, -170.93533325195312, 169.4053497314453, 181.78028869628906, 193.15765380859375, 177.7886962890625, 1.4099884033203125, 64.72808837890625, 180.56900024414062, 156.1105194091797, 11.311016082763672, 199.5994873046875, 23.7208251953125, 34.53178405761719, 221.6494140625, 158.47238159179688, 1.44622802734375, -91.09673309326172, -7.963836669921875, 3.648956298828125, 2.0956954956054688, 70.57659912109375, 175.6278076171875, 40.17987060546875, -32.649993896484375, 13.032329559326172, -18.99738311767578, 118.66964721679688, 65.9603271484375, 222.93392944335938, 0.0, 5.26959228515625, 64.88655853271484, -29.080078125, 116.53146362304688, 11.536697387695312, 214.7235107421875, 143.203369140625, 31.2615966796875, 193.1593017578125, 7.644170761108398, -135.1736602783203, 211.9367218017578, 200.27853393554688, 53.197265625, -48.406402587890625, -143.77597045898438, 11.523666381835938, 159.88890075683594, -1.4403457641601562, 193.476806640625, 186.92025756835938, -126.5140380859375, 4.6576080322265625, 60.912445068359375, -12.329727172851562, 27.679779052734375, 184.5177001953125, 9.52923583984375, -1.272735595703125, -4.2104644775390625, -151.64193725585938, -8.205780029296875, -169.28549194335938, 76.09368896484375, 147.48760986328125, 147.0635223388672, 22.3660888671875, -43.693878173828125, 167.57424926757812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000430.npy"}
{"epoch": 0.900523560209424, "step": 431, "batch_size": 128, "mean": 49.277442932128906, "std": 98.9319076538086, "min": -171.4075927734375, "p10": -61.45898132324218, "median": 26.353065490722656, "p90": 191.84110717773436, "max": 256.385986328125, "pos_frac": 0.6875, "sample": [-9.569366455078125, -171.4075927734375, 104.7293701171875, 167.7391815185547, 155.73330688476562, 74.02978515625, 22.546096801757812, 72.84317016601562, 2.92889404296875, -127.87875366210938, -2.96282958984375, 28.4259033203125, 4.65252685546875, 50.780609130859375, -44.75634765625, 60.81842041015625, -43.156585693359375, 137.5482177734375, 129.60391235351562, 8.104318618774414, -3.1068115234375, 149.91204833984375, 113.95271301269531, -123.71296691894531, 123.22381591796875, 160.32550048828125, 101.64572143554688, -58.854461669921875, 169.5414276123047, 215.682373046875, 32.58930969238281, 24.280227661132812, 147.82708740234375, -143.4759521484375, -55.88636779785156, 161.41702270507812, 71.08158111572266, -22.806243896484375, -6.133018493652344, 120.95223999023438, -43.44498825073242, -161.10235595703125, 6.975889205932617, 2.21221923828125, 151.86817932128906, 54.23858642578125, 196.96267700195312, -15.926109313964844, -53.51238250732422, -166.8984375, 3.1132640838623047, 36.00067138671875, 1.6050090789794922, 160.03094482421875, -0.9205322265625, 169.9212646484375, 48.4852294921875, 212.035888671875, 173.83023071289062, -21.933197021484375, 1.7364501953125, 17.47982406616211, 187.01878356933594, 33.48046875, 21.4600830078125, 118.713134765625, 69.30810546875, 190.72015380859375, -24.50243377685547, -21.777587890625, 194.4566650390625, -11.395408630371094, -3.80914306640625, 199.58880615234375, -2.4742431640625, 237.05419921875, 8.88671875, 230.01275634765625, 46.33467102050781, 30.792205810546875, 204.67547607421875, 256.385986328125, 47.88037109375, -112.62091064453125, -1.3319110870361328, -14.64886474609375, 161.74856567382812, 29.044769287109375, -38.60272216796875, 18.96723175048828, 221.88275146484375, 158.86843872070312, 189.2320098876953, -116.56904602050781, 3.8568572998046875, 235.44757080078125, -67.53619384765625, 213.49957275390625, -0.5442638397216797, 9.92974853515625, 38.64997863769531, 38.14942932128906, 13.141769409179688, 3.21044921875, 29.478668212890625, 6.0836944580078125, 160.49732971191406, 61.41343688964844, 138.58383178710938, -13.810745239257812, 16.75841522216797, -12.490158081054688, 1.5055694580078125, 154.20632934570312, 23.693145751953125, 87.32331848144531, -7.0530853271484375, -2.014129638671875, -92.21903991699219, 36.16551208496094, -86.64108276367188, -73.685791015625, 15.023918151855469, 3.32171630859375, -125.07257080078125, 221.6993865966797, 160.16122436523438, 36.033294677734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000431.npy"}
{"epoch": 0.9026178010471204, "step": 432, "batch_size": 128, "mean": 69.41046142578125, "std": 97.20526123046875, "min": -194.589599609375, "p10": -36.91207275390625, "median": 42.934722900390625, "p90": 201.17841033935545, "max": 238.58370971679688, "pos_frac": 0.6953125, "sample": [-11.134552001953125, 178.5548095703125, 4.925743103027344, 179.70172119140625, 223.97415161132812, 213.0836181640625, 174.126953125, 61.75689697265625, -39.557861328125, -165.111328125, 222.80859375, -1.8456268310546875, 180.42669677734375, 40.91986083984375, -17.694656372070312, 13.849327087402344, 36.220184326171875, 200.71629333496094, -33.47163391113281, 6.2901458740234375, 144.93072509765625, -194.589599609375, 77.42195129394531, -1.0217399597167969, -62.882476806640625, 7.014825820922852, -25.4827880859375, 181.02964782714844, 70.94017028808594, -9.092864990234375, 25.9490966796875, 183.2969207763672, 206.77609252929688, 187.83131408691406, -37.6610107421875, -6.2878265380859375, 163.58248901367188, -36.89654541015625, 194.81243896484375, 21.386627197265625, 141.08811950683594, 159.34609985351562, 0.115020751953125, 2.7783050537109375, -6.484626770019531, -9.290985107421875, 84.97296142578125, 12.506233215332031, -100.63671875, -2.7335205078125, 141.92742919921875, 87.21803283691406, 17.916427612304688, 202.25668334960938, -21.673095703125, 11.8966064453125, -72.19927215576172, 168.28204345703125, -141.12001037597656, -8.302940368652344, -9.240631103515625, 28.478240966796875, 172.47509765625, -12.313690185546875, 167.72450256347656, 5.314262390136719, 204.489990234375, 156.66839599609375, 155.95025634765625, 157.18621826171875, -69.1702880859375, 137.0415802001953, 5.498565673828125, -19.47626495361328, 16.061851501464844, 208.77203369140625, 153.24969482421875, 213.3135986328125, 26.949859619140625, 177.37310791015625, 44.9495849609375, 4.627281188964844, 238.58370971679688, -14.132919311523438, 166.61019897460938, 141.19171142578125, 184.3440399169922, 136.7564697265625, -29.3619384765625, 142.877197265625, 12.367233276367188, -14.695262908935547, 0.0, 203.78106689453125, 179.60928344726562, -22.9671630859375, 10.827117919921875, -43.395263671875, -8.6917724609375, 72.62728881835938, 66.80307006835938, -5.202301025390625, -36.94830322265625, 168.49542236328125, 133.77276611328125, 138.6503143310547, 213.57247924804688, 165.48101806640625, 207.77191162109375, 174.0577850341797, -1.32940673828125, 78.5889892578125, 44.99761962890625, 213.81036376953125, 1.4187240600585938, 48.17913818359375, 75.67938232421875, 149.24452209472656, 4.893928527832031, 171.4437255859375, 117.24966430664062, 159.1524658203125, -52.54376220703125, 4.224067687988281, 33.20082092285156, 142.5843963623047, -4.992160797119141, -37.431488037109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000432.npy"}
{"epoch": 0.9047120418848168, "step": 433, "batch_size": 128, "mean": 55.49558639526367, "std": 98.5360336303711, "min": -215.63900756835938, "p10": -41.4122589111328, "median": 46.12807083129883, "p90": 169.01038665771483, "max": 340.15582275390625, "pos_frac": 0.75, "sample": [139.079345703125, -166.56100463867188, -192.71121215820312, 211.30020141601562, 85.55496215820312, 33.31537628173828, 136.6885986328125, -15.602294921875, -8.646780014038086, 38.1143798828125, 23.4434814453125, 21.444366455078125, 156.76895141601562, 93.50335693359375, 53.9776611328125, 27.1832275390625, -9.38751220703125, 46.80542755126953, 66.72137451171875, 17.655044555664062, 39.64080810546875, 14.06378173828125, 147.42196655273438, -80.45822143554688, -0.046916961669921875, 22.478221893310547, 202.08901977539062, 5.3704681396484375, 204.8360595703125, 172.36041259765625, 158.31460571289062, -11.029998779296875, 151.64056396484375, 56.27264404296875, 100.85479736328125, 203.1759033203125, 159.92501831054688, 144.3138885498047, -127.5997314453125, 42.470458984375, -8.302497863769531, 20.3487548828125, -1.6505126953125, 33.89958190917969, 94.01435852050781, 86.66262817382812, 114.54537963867188, -0.27712249755859375, 16.5570068359375, 155.9967041015625, 154.6522674560547, 103.4623031616211, 48.42083740234375, 8.64691162109375, 54.818603515625, -131.07632446289062, 52.1978759765625, 150.53807067871094, 13.634563446044922, 10.791275024414062, -20.484344482421875, 117.0428466796875, -112.46118927001953, 137.22874450683594, 179.44577026367188, 76.5130615234375, 148.03048706054688, -215.63900756835938, 168.6938934326172, 157.17076110839844, -48.118804931640625, 18.182205200195312, 169.74887084960938, -8.185810089111328, 6.12176513671875, 35.235748291015625, 133.646728515625, 147.18484497070312, 157.0581512451172, -190.16629028320312, -125.50140380859375, 173.8836212158203, 103.87901306152344, -15.715240478515625, 27.54559326171875, -2.16094970703125, 170.39739990234375, -9.563922882080078, 75.42544555664062, 45.450714111328125, 155.11245727539062, -4.6635589599609375, -196.72543334960938, 37.12398147583008, 22.567665100097656, 158.4068145751953, 115.54031372070312, 191.1519317626953, 138.80734252929688, -125.03659057617188, 48.559051513671875, -16.137344360351562, 4.282981872558594, 340.15582275390625, 40.6236572265625, 135.91400146484375, 168.65484619140625, 149.63169860839844, 177.42420959472656, 1.6884231567382812, 50.713165283203125, 14.2027587890625, 61.219207763671875, -35.272125244140625, 18.982803344726562, 114.1934814453125, 87.26546478271484, -7.360660552978516, 152.6683807373047, 213.14300537109375, 34.43585205078125, 137.6373291015625, -172.84909057617188, 25.021068572998047, -24.024864196777344, -38.53802490234375, 81.29116821289062, 5.113857269287109], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000433.npy"}
{"epoch": 0.9068062827225131, "step": 434, "batch_size": 128, "mean": 67.24725341796875, "std": 98.94953918457031, "min": -193.63848876953125, "p10": -26.23843078613281, "median": 41.722503662109375, "p90": 186.9083740234375, "max": 253.949462890625, "pos_frac": 0.703125, "sample": [94.88627624511719, -6.0996856689453125, -12.72509765625, 5.291748046875, 181.6568145751953, 25.0201416015625, 158.56265258789062, 149.5183563232422, -28.0771484375, 250.76824951171875, -9.000465393066406, 168.90672302246094, -124.732666015625, 3.00335693359375, -10.541160583496094, 9.6224365234375, -21.752227783203125, 142.18313598632812, 90.79450225830078, -0.5362548828125, 14.33984375, 235.59112548828125, -4.94012451171875, 253.949462890625, 182.439453125, -88.16510009765625, 0.0, -16.250144958496094, 0.9387283325195312, 157.11383056640625, 214.1466064453125, 171.67596435546875, -21.96411895751953, 161.64547729492188, 157.61941528320312, 162.23492431640625, -11.6376953125, 72.3040771484375, 26.65289306640625, 216.61444091796875, 51.388763427734375, 137.498046875, 196.26608276367188, -50.67730712890625, 35.48063659667969, -164.9564666748047, 128.349853515625, 130.3759765625, 29.4639892578125, 127.89413452148438, 185.85247802734375, 183.9640655517578, 128.15579223632812, 245.6141357421875, 29.24237060546875, 142.34005737304688, 166.03765869140625, 12.890899658203125, -32.61962890625, 189.37213134765625, 40.64484405517578, 134.46875, 216.6929931640625, -19.602096557617188, 11.658050537109375, -3.683258056640625, 79.65719604492188, 175.32113647460938, 23.05157470703125, 43.85385513305664, 207.99606323242188, 3.046905517578125, -21.821327209472656, 194.09857177734375, -63.9952392578125, 1.9784355163574219, 32.193084716796875, 208.10250854492188, 4.110107421875, 149.517578125, 175.233642578125, -25.450408935546875, 138.6812744140625, 4.3184814453125, 64.41854858398438, 173.80941772460938, 178.25753784179688, -191.2576904296875, 151.06402587890625, -23.753753662109375, 63.22398376464844, -24.404296875, -129.7635498046875, 0.0, 166.25018310546875, 42.80016326904297, 27.51092529296875, 181.3795166015625, 182.46929931640625, 14.16268539428711, -21.686431884765625, -5.849822998046875, -29.984426498413086, 196.15362548828125, -9.093154907226562, 9.655853271484375, -4.721622467041016, -29.666168212890625, 166.84683227539062, 106.38687133789062, 44.977996826171875, 3.089324951171875, 0.0, 93.85430908203125, 140.6062774658203, -6.979278564453125, 23.755691528320312, 167.78854370117188, 96.69473266601562, 68.73355102539062, -193.63848876953125, 14.995849609375, 25.52703857421875, -8.105789184570312, -129.06964111328125, 182.109130859375, 135.69921875, 160.33538818359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000434.npy"}
{"epoch": 0.9089005235602095, "step": 435, "batch_size": 128, "mean": 60.2728271484375, "std": 94.2633056640625, "min": -193.16183471679688, "p10": -32.338008117675784, "median": 39.32904815673828, "p90": 183.4046875, "max": 247.26556396484375, "pos_frac": 0.7578125, "sample": [152.16500854492188, 4.930267333984375, 4.982570648193359, 69.52783203125, 34.08869934082031, 196.1360626220703, 168.2661895751953, -137.2349853515625, 142.46853637695312, -6.682159423828125, -13.237049102783203, -23.397811889648438, -193.16183471679688, 46.402671813964844, 5.938512802124023, -25.25762939453125, 77.06732177734375, 73.07736206054688, 175.60870361328125, 118.43753051757812, 13.5081787109375, 26.052011489868164, 184.95779418945312, 0.0, 3.4487781524658203, -20.830841064453125, 68.14456176757812, 25.846580505371094, 143.97503662109375, 11.58416748046875, 154.75411987304688, 168.89927673339844, 139.32525634765625, -93.27571868896484, -9.246826171875, 215.76177978515625, 117.50437927246094, 184.21051025390625, 44.8948974609375, 9.49554443359375, 183.58853149414062, 186.17721557617188, 183.32589721679688, 27.169784545898438, 162.00164794921875, 15.311614990234375, 158.44277954101562, -32.655975341796875, 182.83790588378906, 37.93817138671875, 155.76873779296875, 101.54083251953125, 210.44448852539062, 166.32073974609375, 21.882081985473633, 4.593711853027344, 46.62225341796875, 19.86151123046875, 121.26127624511719, -46.5343017578125, 11.439239501953125, 214.11907958984375, 10.037704467773438, -159.2578887939453, 18.390670776367188, 14.179962158203125, 38.8759765625, 183.97894287109375, -141.25933837890625, 20.868377685546875, 33.18742370605469, -3.41107177734375, 50.070587158203125, 203.35739135742188, 194.725341796875, -18.638824462890625, 157.80186462402344, 175.14161682128906, -43.08946990966797, 135.7916717529297, 2.936309814453125, 160.38461303710938, 67.31402587890625, 157.41883850097656, -150.0076904296875, 14.446533203125, 81.229248046875, 4.596080780029297, -0.5042877197265625, -36.070404052734375, -8.99432373046875, 153.77655029296875, 128.63717651367188, 247.26556396484375, 89.6241455078125, 176.05812072753906, 112.47492980957031, 136.229248046875, -21.953956604003906, 62.518035888671875, -7.2147979736328125, 39.78211975097656, 131.1261749267578, 3.7583484649658203, 18.62823486328125, 12.99334716796875, 41.854888916015625, 7.8481597900390625, 150.636474609375, 7.052986145019531, -7.942047119140625, -41.284912109375, -157.26312255859375, -28.571617126464844, -0.96368408203125, 224.13760375976562, 10.9637451171875, 16.5869140625, 123.43441772460938, -158.48397827148438, 50.03515625, 144.84140014648438, 171.817626953125, -1.6105499267578125, -32.20173645019531, 165.11203002929688, 48.7442626953125, 142.38267517089844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000435.npy"}
{"epoch": 0.9109947643979057, "step": 436, "batch_size": 128, "mean": 43.15338134765625, "std": 93.61460876464844, "min": -193.70614624023438, "p10": -48.95629425048828, "median": 17.843799591064453, "p90": 179.82846527099608, "max": 231.3037109375, "pos_frac": 0.6796875, "sample": [90.3441162109375, -35.67431640625, 217.18450927734375, 147.82882690429688, 2.903532028198242, 22.506515502929688, 174.85044860839844, 5.2379150390625, -37.57086181640625, 98.212890625, -130.13467407226562, 62.61822509765625, -8.049270629882812, 32.0386962890625, -6.486785888671875, 206.41510009765625, 29.26776123046875, -20.098548889160156, -93.84074401855469, -85.8763427734375, 11.551193237304688, 130.5045623779297, 155.94766235351562, 146.1129150390625, 188.88223266601562, -30.486106872558594, 199.2169189453125, -0.0958251953125, 156.42254638671875, 21.72784423828125, 48.140846252441406, 12.313796997070312, -5.495716094970703, 24.363922119140625, 178.5869598388672, 127.17950439453125, -16.009620666503906, 45.53931427001953, -0.485504150390625, -52.22749328613281, 142.23434448242188, 10.6868896484375, 18.41510772705078, 128.69985961914062, 182.72531127929688, 31.43438720703125, 6.2142791748046875, 7.5438232421875, -25.2919921875, 174.67681884765625, 24.8128662109375, -23.860687255859375, 44.263275146484375, -14.63909912109375, 167.69961547851562, -34.516990661621094, 199.68829345703125, -193.70614624023438, 5.78607177734375, 183.70089721679688, 160.54379272460938, 33.35649108886719, 12.30267333984375, 85.27906799316406, 54.12353515625, 152.88758850097656, -8.746795654296875, -144.60568237304688, 114.9205322265625, 150.66151428222656, 0.6711578369140625, 230.26861572265625, 77.3170166015625, -132.96856689453125, 23.511703491210938, 75.77859497070312, 6.122777938842773, 141.2555694580078, 11.410736083984375, -2.53955078125, 78.47470092773438, -16.939208984375, 17.272491455078125, -6.781455993652344, 66.16302490234375, 64.22311401367188, -5.5397186279296875, 5.001617431640625, 59.47130584716797, -94.20986938476562, 186.7254638671875, 1.2186279296875, -6.245733261108398, -47.554351806640625, 13.81903076171875, 21.5552978515625, 231.3037109375, -17.818771362304688, 3.8975982666015625, -30.021530151367188, -17.760986328125, -183.07476806640625, 16.645004272460938, 114.15350341796875, 139.80096435546875, 12.60870361328125, -156.46652221679688, 158.24790954589844, 227.35211181640625, -14.069656372070312, 14.405708312988281, 54.97239685058594, -22.63848876953125, 141.60171508789062, 9.877227783203125, 134.21559143066406, 12.23895263671875, 221.0035400390625, -69.36019897460938, 8.14398193359375, -58.49365234375, 46.2891845703125, -169.36468505859375, 111.51165771484375, -28.237548828125, 189.674560546875, 38.36067199707031, -23.50384521484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000436.npy"}
{"epoch": 0.9130890052356021, "step": 437, "batch_size": 128, "mean": 71.97354125976562, "std": 101.97712707519531, "min": -177.65283203125, "p10": -22.97371215820312, "median": 64.9552001953125, "p90": 191.8220458984375, "max": 259.442626953125, "pos_frac": 0.75, "sample": [28.972900390625, 172.02853393554688, 42.960693359375, 158.8583984375, -177.4542236328125, 230.6357421875, 2.88299560546875, 8.202499389648438, 80.546142578125, 15.46820068359375, 185.784912109375, 21.969223022460938, 149.16966247558594, 72.91796875, 11.716537475585938, 168.2154541015625, -12.10003662109375, 5.57952880859375, 47.6219482421875, 163.63973999023438, 42.103363037109375, 199.97457885742188, 259.442626953125, 179.5019073486328, 14.711265563964844, 29.1490478515625, -3.7210235595703125, 130.40090942382812, 162.82550048828125, 30.412200927734375, -17.05902099609375, -3.657440185546875, 7.483795166015625, 106.81883239746094, -1.260894775390625, 50.21082305908203, -157.39999389648438, -30.154129028320312, 10.557968139648438, 33.14208984375, 165.58482360839844, 63.419921875, 180.29763793945312, 173.50222778320312, -44.644927978515625, 5.533172607421875, 31.660400390625, 13.804931640625, 199.7083740234375, 162.9375, 148.233642578125, 215.1070098876953, 140.9892578125, 180.06021118164062, 8.414852142333984, 81.51840209960938, -26.874725341796875, -15.765132904052734, 203.80029296875, 191.67559814453125, 171.14364624023438, -4.57562255859375, 192.16375732421875, 209.08843994140625, 226.976318359375, 121.34945678710938, 0.6992454528808594, 155.215576171875, -12.663192749023438, -21.301849365234375, 42.36334991455078, 11.44708251953125, -40.71337890625, 186.97772216796875, -174.78695678710938, 168.14462280273438, 7.167263984680176, 105.40946960449219, -172.22183227539062, 0.0, 173.07455444335938, -142.67929077148438, 8.896926879882812, -7.633356094360352, 189.99197387695312, 145.66970825195312, -92.78079223632812, 119.13931274414062, -156.67697143554688, 171.3280029296875, 203.45831298828125, 171.77517700195312, -2.3801612854003906, 140.03756713867188, 66.490478515625, -7.1964111328125, 165.779296875, 26.99071502685547, 88.87130737304688, 111.49609375, 194.89512634277344, 200.97607421875, 123.09539794921875, 190.28948974609375, 31.256622314453125, -4.610710144042969, 143.312255859375, 127.58480072021484, 177.19937133789062, -148.22593688964844, 5.0889129638671875, -0.861419677734375, 157.4022216796875, -177.65283203125, -7.303558349609375, 163.57794189453125, 66.57125854492188, 165.33612060546875, 144.68695068359375, -0.2415924072265625, 122.72158813476562, -13.372711181640625, 24.591400146484375, 74.12286376953125, 0.0, 14.04339599609375, 150.03463745117188, 242.49789428710938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000437.npy"}
{"epoch": 0.9151832460732985, "step": 438, "batch_size": 128, "mean": 29.754268646240234, "std": 100.73963165283203, "min": -218.0040283203125, "p10": -113.89723510742188, "median": 11.0263032913208, "p90": 172.54274139404296, "max": 246.0245361328125, "pos_frac": 0.6328125, "sample": [9.451347351074219, -183.50186157226562, 133.444091796875, -15.300323486328125, -91.07254028320312, 180.6312255859375, 53.51898193359375, 37.421173095703125, 0.0, 11.867158889770508, 189.8143310546875, -152.45663452148438, 17.150108337402344, 171.44735717773438, 41.85845947265625, 10.170341491699219, 16.60821533203125, 171.5716094970703, 154.2476806640625, -20.234375, -48.772796630859375, 0.49261474609375, 1.939178466796875, -2.0352096557617188, -125.70098876953125, -30.168540954589844, -22.57672119140625, 132.3966064453125, 9.91717529296875, 44.8079833984375, 11.640390396118164, 206.4373779296875, 149.43106079101562, -10.333236694335938, 30.44891357421875, 122.0653076171875, -146.10536193847656, -16.860763549804688, 48.174346923828125, -182.39620971679688, -218.0040283203125, -150.40237426757812, 53.87823486328125, 144.770751953125, 10.299850463867188, -12.779541015625, 2.2309417724609375, 25.502288818359375, -168.1566162109375, -3.1227684020996094, -4.2066650390625, -153.30284118652344, 1.7818527221679688, 11.526275634765625, 9.38446044921875, 34.388519287109375, 126.71624755859375, -61.58251953125, 46.827606201171875, 22.109413146972656, 11.841567993164062, -6.145284652709961, -0.9906139373779297, -163.33203125, 131.72555541992188, -15.64975357055664, 11.4468994140625, 3.8164215087890625, 186.92813110351562, -134.83670043945312, 65.13992309570312, -50.470672607421875, 181.10806274414062, -24.13086700439453, 25.58575439453125, 80.702392578125, 147.7522430419922, -132.8984832763672, 166.5614013671875, 164.30886840820312, -25.490814208984375, 160.84783935546875, -77.9585189819336, -4.99420166015625, 211.37457275390625, -2.715789794921875, 154.30934143066406, 63.675140380859375, -112.766357421875, 7.699249267578125, 174.8087158203125, 17.559669494628906, 2.4803218841552734, -30.75946044921875, 188.6318359375, 153.04638671875, 11.924369812011719, 181.6875, 238.93746948242188, -26.4434814453125, -2.005767822265625, -39.69549560546875, 40.04449462890625, 124.92225646972656, 146.58758544921875, -34.3182373046875, 4.6568450927734375, 59.79766082763672, -19.393280029296875, 158.82083129882812, 187.23553466796875, 10.605707168579102, -93.43624877929688, 22.166622161865234, 0.0, 1.2038421630859375, 145.1321563720703, 8.944587707519531, -72.20977020263672, 72.08241271972656, 187.20050048828125, -116.53594970703125, 6.387355804443359, 162.69683837890625, 246.0245361328125, -2.3988418579101562, 13.225677490234375, 29.191543579101562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000438.npy"}
{"epoch": 0.9172774869109948, "step": 439, "batch_size": 128, "mean": 56.88280487060547, "std": 101.0040283203125, "min": -220.18896484375, "p10": -82.87948455810546, "median": 26.696243286132812, "p90": 187.72586364746093, "max": 228.69003295898438, "pos_frac": 0.71875, "sample": [-125.79644775390625, -122.5933837890625, -159.50238037109375, 210.32608032226562, -56.03338623046875, 180.51571655273438, 17.056396484375, 17.08039093017578, 91.15702819824219, 176.6836395263672, -3.9013214111328125, 20.744598388671875, -1.7795085906982422, 4.881134033203125, 144.77264404296875, -155.73233032226562, 10.0989990234375, 127.4229736328125, -1.3981170654296875, 63.991668701171875, 202.88677978515625, 165.1765899658203, 10.304290771484375, 32.57647705078125, 3.99163818359375, 162.93545532226562, 72.9024658203125, 159.30593872070312, 187.03146362304688, 19.801542282104492, -137.83441162109375, 104.41067504882812, 14.331314086914062, 209.88198852539062, -19.415771484375, 136.59262084960938, 211.89346313476562, 57.871490478515625, 228.69003295898438, 25.560562133789062, -36.764984130859375, 18.019939422607422, 163.63095092773438, 142.22459411621094, 5.509613037109375, 143.83253479003906, 156.74685668945312, 10.83575439453125, 2.0422515869140625, 7.784515380859375, 7.970703125, 18.432846069335938, -94.78289794921875, 13.41943359375, 29.861053466796875, 192.32110595703125, 152.7476043701172, -26.14842987060547, 38.7413330078125, 0.0, 27.831924438476562, 0.0, -101.5504150390625, 14.493919372558594, -91.00497436523438, 180.27972412109375, 24.9112548828125, -84.269775390625, 10.708553314208984, -35.855621337890625, 2.619964599609375, 189.34613037109375, 20.528411865234375, -24.676467895507812, -7.58428955078125, 54.717750549316406, -2.3229713439941406, 127.1877212524414, -42.206390380859375, 103.68902587890625, -30.751815795898438, 125.01293182373047, 171.80551147460938, 24.690689086914062, 143.86349487304688, 18.163116455078125, -114.89369201660156, -4.5648651123046875, -107.17711639404297, -1.66497802734375, 174.86036682128906, -148.356201171875, 0.0, 194.19171142578125, 207.56292724609375, 121.46468353271484, 202.5845947265625, 163.50814819335938, 42.0838623046875, -52.34898376464844, 48.83978271484375, -82.28364562988281, 157.0605010986328, -11.254486083984375, 56.874664306640625, 16.443359375, 135.23825073242188, 146.02308654785156, 145.86505126953125, 41.09796142578125, 177.8954620361328, 75.7349853515625, 116.21003723144531, 182.89154052734375, 161.81454467773438, 0.3612232208251953, -220.18896484375, 202.46119689941406, 123.12026977539062, 196.3873291015625, 177.78060913085938, 145.70785522460938, 20.056076049804688, -19.8662109375, 199.8750762939453, 0.0, 185.59156799316406, 171.0706787109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000439.npy"}
{"epoch": 0.9193717277486911, "step": 440, "batch_size": 128, "mean": 46.95436096191406, "std": 100.80087280273438, "min": -195.65338134765625, "p10": -76.59870300292968, "median": 26.495941162109375, "p90": 177.16534576416012, "max": 251.50506591796875, "pos_frac": 0.671875, "sample": [-32.292388916015625, 0.0, 33.652099609375, -154.6490478515625, -15.636177062988281, 88.10113525390625, -162.9237060546875, 122.17062377929688, -15.477317810058594, 140.0636444091797, -23.83258056640625, 20.831756591796875, -115.1351318359375, 1.525238037109375, 34.649505615234375, -141.25558471679688, 119.98605346679688, 171.00509643554688, 141.01171875, 155.66839599609375, 48.44830322265625, 195.2608642578125, -2.9529800415039062, 29.034027099609375, -74.03750610351562, -7.196693420410156, -195.65338134765625, 128.6529083251953, 77.40576171875, 88.51522827148438, 198.25848388671875, 226.77099609375, 8.709522247314453, -11.480712890625, -18.04290771484375, 68.25909423828125, 83.33599853515625, -167.81546020507812, 0.0, 48.65087890625, -176.07382202148438, -3.732666015625, 7.332275390625, 131.86441040039062, 185.34654235839844, -29.95648956298828, -76.98495483398438, 1.2867889404296875, 17.417076110839844, 159.8826446533203, 209.14279174804688, -14.420478820800781, 146.67860412597656, 0.4738311767578125, 39.05938720703125, 194.36151123046875, -2.893695831298828, -138.73974609375, -48.075225830078125, -34.14324951171875, -52.60235595703125, -1.483245849609375, 8.112091064453125, -41.87701416015625, 251.50506591796875, 152.26220703125, 214.84170532226562, 145.79705810546875, 163.97723388671875, 9.789031982421875, 0.40610504150390625, 30.38427734375, 11.657272338867188, 161.4134979248047, 170.63424682617188, 2.2230987548828125, -2.24542236328125, -125.37077331542969, 17.273223876953125, 17.867950439453125, 134.02247619628906, 141.23126220703125, 168.90155029296875, 98.22738647460938, 203.37440490722656, -107.84793090820312, 33.96380615234375, 199.1173095703125, 39.237518310546875, -93.6862564086914, 206.22418212890625, -139.5821075439453, 161.62881469726562, -18.903823852539062, 151.71966552734375, 110.73446655273438, 8.366958618164062, 30.116500854492188, 164.21363830566406, 46.06158447265625, 173.65911865234375, -32.627784729003906, 20.64727783203125, 21.173805236816406, 50.737548828125, 70.43450927734375, 6.34454345703125, 195.68438720703125, 148.46731567382812, 69.53265380859375, 10.77508544921875, 57.154876708984375, 23.957855224609375, 173.0699920654297, -44.25445556640625, 165.67453002929688, 169.43072509765625, 21.644561767578125, 32.082489013671875, 119.65237426757812, -25.141754150390625, 189.4142608642578, -8.59149169921875, -76.43316650390625, -6.488800048828125, 15.427566528320312, -18.886268615722656, 156.5187225341797], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000440.npy"}
{"epoch": 0.9214659685863874, "step": 441, "batch_size": 128, "mean": 68.48980712890625, "std": 92.60041809082031, "min": -188.04917907714844, "p10": -17.016912078857416, "median": 43.69507598876953, "p90": 187.6084777832031, "max": 232.837158203125, "pos_frac": 0.765625, "sample": [-12.961471557617188, 31.16900634765625, -141.20022583007812, 71.68597412109375, 0.30029869079589844, -14.847381591796875, 199.3436279296875, 0.28076171875, 3.488311767578125, -9.226806640625, -47.92303466796875, 232.837158203125, 24.77777862548828, 142.83892822265625, 31.26171875, 154.62765502929688, 173.49087524414062, 163.55731201171875, 141.2217559814453, 69.46400451660156, 51.70330810546875, 148.2946014404297, 9.294309616088867, 18.49066162109375, 185.3157958984375, 92.02651977539062, 164.7431640625, 174.9402618408203, 40.38629150390625, 3.4488525390625, 9.78863525390625, 161.132080078125, 102.90147399902344, -15.597663879394531, 156.130615234375, 113.99578857421875, 12.19049072265625, 177.28677368164062, 146.6238250732422, -150.2742919921875, 175.80441284179688, 163.80880737304688, -22.958824157714844, -0.5994873046875, 220.62725830078125, -1.8122901916503906, 47.489532470703125, 40.18681335449219, 165.13485717773438, 201.6754150390625, 165.03416442871094, 6.46136474609375, 51.152130126953125, 19.703475952148438, 18.72905731201172, 16.53326416015625, -1.959320068359375, 164.23919677734375, 43.22218322753906, -12.048583984375, -4.05609130859375, 31.909393310546875, 29.29302978515625, 157.7242431640625, 8.37396240234375, 0.0, -0.3188018798828125, 106.06675720214844, 71.48884582519531, 218.81515502929688, -70.88327026367188, -72.29460144042969, -20.3284912109375, 11.033004760742188, 0.0, 157.0962677001953, 44.16796875, 38.568756103515625, 160.27978515625, 49.571533203125, 12.814773559570312, 166.10894775390625, 24.557540893554688, -69.0587158203125, 49.311248779296875, 92.06982421875, 156.73043823242188, 154.8401336669922, 0.04949951171875, 23.275390625, 5.143829345703125, 8.60113525390625, 39.48565673828125, 167.49993896484375, 160.91531372070312, -11.09228515625, 183.4515838623047, -113.03661346435547, 187.3955078125, 54.09918212890625, 188.10540771484375, -188.04917907714844, 58.91082763671875, 4.99102783203125, -106.35873413085938, -0.7170867919921875, 212.115478515625, -127.73280334472656, 174.14393615722656, -7.481201171875, 32.902442932128906, 193.39596557617188, 190.11538696289062, 190.9153594970703, 204.19534301757812, 132.55718994140625, 116.20845031738281, -3.64404296875, -31.88189697265625, 9.138763427734375, 227.96783447265625, -8.663078308105469, 164.57745361328125, 148.2264404296875, 193.030517578125, 128.02113342285156, 148.76165771484375, 35.84173583984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000441.npy"}
{"epoch": 0.9235602094240838, "step": 442, "batch_size": 128, "mean": 51.37265396118164, "std": 107.68428039550781, "min": -221.6046142578125, "p10": -88.60974807739255, "median": 26.850372314453125, "p90": 186.64112243652343, "max": 264.77508544921875, "pos_frac": 0.71875, "sample": [152.77169799804688, 17.47955322265625, 21.318374633789062, -31.25992202758789, 226.73583984375, 179.20310974121094, 6.307861328125, -12.066879272460938, 183.30865478515625, 3.241973876953125, 109.196533203125, 26.417068481445312, -108.80657196044922, -11.409835815429688, 166.43264770507812, -161.6796875, 27.283676147460938, 22.778472900390625, 24.318603515625, -150.39083862304688, 12.5927734375, 39.857540130615234, -1.871185302734375, -4.482175827026367, 139.54708862304688, -5.493682861328125, 147.66773986816406, -23.061859130859375, 206.6620330810547, 190.27273559570312, -161.97640991210938, 3.973114013671875, -42.725341796875, 162.7113037109375, 65.78675842285156, 162.97891235351562, -30.185287475585938, 24.70391845703125, -159.59259033203125, 8.635696411132812, 19.391708374023438, -80.18553924560547, 153.41116333007812, 14.776931762695312, -22.7744140625, -183.66787719726562, 12.952774047851562, 38.979339599609375, 10.075889587402344, -52.13067626953125, 14.9227294921875, 5.080841064453125, 100.18150329589844, 225.5880126953125, 7.554344177246094, 114.8553466796875, 148.0182342529297, 0.0, 93.32919311523438, -170.17007446289062, -77.00607299804688, 12.216392517089844, -167.19886779785156, 171.0143280029297, -167.18731689453125, 191.62437438964844, 87.265380859375, -2.6077728271484375, 32.08331298828125, 145.6871795654297, 0.030181884765625, -221.6046142578125, 195.37197875976562, -8.084564208984375, 117.36965942382812, 62.278411865234375, 219.40264892578125, -23.29681396484375, 137.7881317138672, 66.7333984375, 29.54998779296875, 79.4620361328125, -10.45572280883789, 228.33035278320312, 158.17288208007812, 190.9541778564453, -108.2662353515625, -154.8623046875, -53.423675537109375, -40.786956787109375, 13.70697021484375, 100.9443359375, 13.594696044921875, 65.28411865234375, 15.34869384765625, 153.04046630859375, 114.58375549316406, 156.0108184814453, 24.939834594726562, 171.70523071289062, 198.51760864257812, 195.99212646484375, 171.30328369140625, 185.084716796875, 115.21466064453125, 264.77508544921875, 9.8043212890625, -57.52385711669922, -27.763595581054688, 221.90695190429688, 164.10250854492188, 167.39508056640625, 119.66337585449219, 151.0879364013672, 18.824142456054688, 78.223388671875, 18.15013885498047, 78.908935546875, 157.09982299804688, -55.384979248046875, 125.16868591308594, 158.73410034179688, 175.54849243164062, 23.50201416015625, -150.3400421142578, 60.522491455078125, 153.1749267578125, 56.923858642578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000442.npy"}
{"epoch": 0.9256544502617801, "step": 443, "batch_size": 128, "mean": 54.193939208984375, "std": 98.53472137451172, "min": -197.2818603515625, "p10": -42.677111053466795, "median": 25.004051208496094, "p90": 184.28759002685547, "max": 242.23040771484375, "pos_frac": 0.6953125, "sample": [120.37632751464844, 8.76861572265625, 102.79385375976562, 21.938430786132812, -46.38482666015625, 56.85284423828125, 197.91273498535156, 14.901031494140625, 198.3192138671875, -15.104667663574219, 14.643707275390625, 74.901611328125, -12.390678405761719, 86.75762939453125, -10.303436279296875, -150.57582092285156, -16.2548828125, 113.0924072265625, 182.6336669921875, -5.431095123291016, 42.75251770019531, 87.901611328125, 57.19548034667969, 46.500518798828125, -5.4964599609375, 23.516738891601562, 20.887298583984375, 169.52845764160156, 29.274490356445312, 5.092079162597656, -31.8504638671875, 64.2564697265625, -176.52911376953125, 4.219547271728516, -10.99993896484375, 34.248443603515625, 176.58392333984375, 17.39739990234375, 175.22900390625, 11.190269470214844, 183.76416015625, -17.804290771484375, -8.762176513671875, -0.5034637451171875, 51.81422424316406, 55.04508972167969, 242.23040771484375, 5.5511932373046875, 143.53134155273438, -36.996734619140625, 69.0062255859375, -182.27978515625, -104.32948303222656, -83.060791015625, 71.71096801757812, 157.2918243408203, 51.58622741699219, 235.99905395507812, 99.1552734375, 206.06765747070312, 132.302978515625, 26.491363525390625, -10.047615051269531, 163.6337890625, 189.37417602539062, -8.105873107910156, -3.7382049560546875, 0.07314300537109375, 172.50254821777344, 157.90216064453125, -8.778121948242188, -4.65740966796875, 201.068359375, 15.827880859375, 184.98341369628906, -4.1995849609375, -4.524116516113281, 160.1055145263672, -99.7178955078125, 13.520538330078125, -13.195198059082031, 183.9893798828125, 180.88656616210938, 124.82332611083984, 111.4888916015625, -44.65772247314453, 15.81390380859375, -123.255126953125, 150.4236602783203, 213.9423828125, 220.21759033203125, 172.552734375, 104.3828125, 22.43414306640625, -111.72190856933594, 141.04278564453125, 10.672348022460938, 207.12252807617188, 138.9495086669922, 164.2451171875, 46.425079345703125, -41.828277587890625, -168.43087768554688, 180.5986785888672, -8.0133056640625, 2.5769195556640625, 122.19439697265625, 145.542236328125, 57.998260498046875, 7.9028472900390625, 165.45045471191406, 23.165313720703125, -40.671875, -6.712974548339844, 3.5230560302734375, -12.100028991699219, 67.39422607421875, 185.24964904785156, 6.094821929931641, -197.2818603515625, 182.4639892578125, 167.05313110351562, 189.82167053222656, 3.7509689331054688, -3.32366943359375, -124.84707641601562, 9.931625366210938, 11.360198974609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000443.npy"}
{"epoch": 0.9277486910994764, "step": 444, "batch_size": 128, "mean": 57.92145538330078, "std": 102.31398010253906, "min": -212.05255126953125, "p10": -42.0871841430664, "median": 30.383487701416016, "p90": 190.55965881347655, "max": 334.859130859375, "pos_frac": 0.6875, "sample": [3.4668807983398438, -3.88555908203125, 187.703857421875, 23.29852294921875, 78.87380981445312, 66.56340026855469, -18.158233642578125, 2.70538330078125, 20.55860137939453, -13.88787841796875, -9.859954833984375, 35.720062255859375, -33.819091796875, 38.073974609375, 161.00054931640625, 29.9482421875, 147.68946838378906, 66.54403686523438, -27.546234130859375, 30.34838104248047, 109.19761657714844, 24.119049072265625, 108.32373046875, -117.5982666015625, 20.161956787109375, -36.315704345703125, 165.14239501953125, 20.110206604003906, 6.4878692626953125, 51.65740966796875, 16.561874389648438, 167.625244140625, 109.69044494628906, -21.915359497070312, 181.72706604003906, -28.941131591796875, -40.33204650878906, 44.47576904296875, 175.35687255859375, 164.2315673828125, 184.10272216796875, 224.94085693359375, -15.10986328125, -4.010461807250977, 41.37225341796875, 174.17953491210938, -1.21026611328125, -153.05108642578125, -138.63848876953125, -44.453582763671875, 232.38214111328125, 150.71145629882812, -80.06488037109375, 218.17041015625, -5.509273529052734, 72.5478515625, 240.64813232421875, 189.2198486328125, -0.167327880859375, -13.18255615234375, 2.0860652923583984, 334.859130859375, 149.80032348632812, -39.75514221191406, 23.520614624023438, -7.423187255859375, 197.62673950195312, 133.41404724121094, -25.62493896484375, 225.7242431640625, 162.56614685058594, 44.09303283691406, -79.013671875, 195.8558349609375, -14.11798095703125, 70.7371826171875, 206.8079833984375, 173.874267578125, 179.02227783203125, 108.45028686523438, 182.87631225585938, -2.7360801696777344, 0.2810020446777344, 208.9476318359375, -7.417156219482422, 172.12710571289062, -9.178062438964844, 158.39019775390625, 163.8931884765625, -91.48937225341797, 3.911102294921875, 203.89227294921875, 10.928115844726562, 157.1049346923828, 43.935707092285156, 30.418594360351562, 59.548858642578125, 30.149002075195312, 10.901809692382812, -73.92828369140625, -25.16064453125, 7.431083679199219, 193.68588256835938, 5.596103668212891, -131.44435119628906, 48.90777587890625, -102.90538787841797, 166.03167724609375, -41.07301330566406, 147.89486694335938, 6.632286071777344, 65.84681701660156, 9.254302978515625, 136.5867462158203, 109.73272705078125, 32.04747009277344, 193.7547607421875, 149.47694396972656, 160.39859008789062, 146.19210815429688, -3.1380233764648438, 17.864032745361328, -124.23123168945312, -0.09833335876464844, -185.00772094726562, 25.212921142578125, 143.46868896484375, -212.05255126953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000444.npy"}
{"epoch": 0.9298429319371728, "step": 445, "batch_size": 128, "mean": 47.76304626464844, "std": 98.40782165527344, "min": -179.00294494628906, "p10": -58.017144775390626, "median": 20.131103515625, "p90": 184.28409118652343, "max": 231.38137817382812, "pos_frac": 0.671875, "sample": [8.298675537109375, -178.72117614746094, 202.80517578125, 92.62652587890625, 45.79974365234375, 204.24139404296875, -5.378578186035156, 109.898681640625, 180.74581909179688, -136.11318969726562, 11.007041931152344, 66.59947204589844, 167.7323455810547, -67.54031372070312, 4.295989990234375, 4.176483154296875, 45.005706787109375, 14.240631103515625, 217.12939453125, 15.416854858398438, 186.95364379882812, -10.1195068359375, 6.693138122558594, -20.754493713378906, 26.142974853515625, 147.32080078125, 62.650848388671875, -2.5697669982910156, 179.61500549316406, 24.654296875, 214.62213134765625, 78.42178344726562, 35.98638916015625, -1.7389602661132812, -26.805938720703125, -13.014633178710938, 7.9273834228515625, 51.93913269042969, 157.71749877929688, 76.0604248046875, 8.4246826171875, 185.0657958984375, 209.4200439453125, -0.922088623046875, 65.81001281738281, 3.200746536254883, 0.134918212890625, 30.36285400390625, 18.09158706665039, -141.15997314453125, -37.61469268798828, 68.43545532226562, 183.89892578125, -72.67532348632812, 151.53976440429688, 0.21216201782226562, 53.420013427734375, 188.63714599609375, 20.18426513671875, 187.9060516357422, 136.47486877441406, 96.615966796875, 231.38137817382812, -6.980133056640625, 176.79425048828125, 150.25332641601562, 131.7283172607422, -179.00294494628906, 54.117950439453125, 206.82260131835938, 25.571044921875, 28.774658203125, -26.178070068359375, 147.2241668701172, 152.3941650390625, 5.51654052734375, -19.142364501953125, -157.95608520507812, 173.48794555664062, 154.4196319580078, -0.6910400390625, -31.69580078125, -48.73484802246094, -143.9879150390625, 4.084239959716797, 21.205482482910156, 16.603469848632812, -91.21514892578125, 136.4727783203125, -0.20813751220703125, 96.792236328125, 161.69313049316406, 199.93167114257812, 162.67849731445312, 44.03485107421875, 166.59307861328125, -3.2913875579833984, -155.83346557617188, 184.2030029296875, -2.68865966796875, 1.7445449829101562, 136.52337646484375, 184.47329711914062, 23.572772979736328, -5.1522216796875, 20.07794189453125, 9.372573852539062, 19.269668579101562, 158.23046875, -2.057647705078125, -45.903289794921875, -13.760238647460938, 0.0, -16.066864013671875, 15.187309265136719, -57.602264404296875, 3.9550018310546875, -4.27398681640625, -8.678085327148438, 179.7728271484375, -58.985198974609375, -53.31396484375, -176.49710083007812, -1.7696037292480469, 55.04266357421875, 180.32003784179688, -101.2248764038086, 166.80857849121094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000445.npy"}
{"epoch": 0.9319371727748691, "step": 446, "batch_size": 128, "mean": 66.57752990722656, "std": 101.696044921875, "min": -218.819580078125, "p10": -38.764317321777334, "median": 41.758296966552734, "p90": 194.10093994140624, "max": 269.0357971191406, "pos_frac": 0.7421875, "sample": [122.25146484375, 235.65087890625, -3.7703323364257812, 6.619140625, 0.6136398315429688, 34.26660919189453, 151.44264221191406, 99.53356170654297, 170.0716094970703, -16.918685913085938, 4.9789581298828125, -48.31349182128906, 137.95291137695312, 37.75200653076172, 6.687286376953125, 25.969322204589844, 221.24880981445312, 22.382217407226562, -18.008468627929688, -6.52777099609375, 8.970779418945312, 141.181396484375, 194.15716552734375, 169.85995483398438, -8.337844848632812, -43.09211730957031, 269.0357971191406, 7.356071472167969, 12.203971862792969, -110.7086181640625, 11.779998779296875, 13.294715881347656, -27.9742431640625, 38.954917907714844, -61.62017822265625, 26.66132354736328, 178.00128173828125, 4.479377746582031, 17.14764404296875, 21.928688049316406, 194.07684326171875, 1.9237442016601562, 108.2848892211914, 17.611312866210938, 47.17279052734375, 160.43182373046875, 191.068115234375, 172.91357421875, 179.38003540039062, -218.819580078125, 142.39364624023438, 138.38739013671875, 217.5047607421875, 4.389331817626953, 148.6418914794922, 169.37696838378906, 6.51788330078125, 48.55133056640625, -210.70892333984375, 125.56011962890625, 182.05624389648438, -1.0584678649902344, -43.88911437988281, 0.0, 44.561676025390625, 177.48843383789062, 50.684173583984375, 121.63323974609375, 172.44482421875, 159.35548400878906, -12.016632080078125, 171.8064727783203, 135.3688507080078, 173.55474853515625, 175.844482421875, 9.866790771484375, 177.33035278320312, 149.2307891845703, -61.783203125, 0.0, 80.13623046875, -99.59490966796875, 215.19546508789062, -10.55950927734375, 27.346817016601562, -132.3905029296875, 102.75686645507812, 220.46185302734375, 148.8974609375, 82.27877807617188, -34.25532531738281, 251.86386108398438, 209.82815551757812, 176.43350219726562, 131.52676391601562, 50.201416015625, 128.33966064453125, 154.23886108398438, -29.01901626586914, 89.32525634765625, 21.792720794677734, -13.01611328125, 207.63592529296875, 209.162353515625, -208.69674682617188, -152.9818572998047, -13.018447875976562, 9.9310302734375, -11.677978515625, 4.36639404296875, 38.218692779541016, 0.0, -36.9095458984375, -8.820159912109375, 153.13912963867188, -46.03143310546875, 201.4432373046875, 120.864990234375, 222.64797973632812, 15.643844604492188, 4.275390625, 57.02354431152344, 25.302978515625, -13.7852783203125, 173.42108154296875, 66.00527954101562, 151.72747802734375, 106.94842529296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000446.npy"}
{"epoch": 0.9340314136125655, "step": 447, "batch_size": 128, "mean": 72.11605834960938, "std": 92.1263427734375, "min": -201.9998779296875, "p10": -19.833419418334955, "median": 52.262451171875, "p90": 187.02913818359374, "max": 234.36322021484375, "pos_frac": 0.7734375, "sample": [-50.560333251953125, 12.010021209716797, 155.40762329101562, 158.0111846923828, 25.251083374023438, 9.51852798461914, 202.10028076171875, 14.473312377929688, -33.50933837890625, 167.37521362304688, 101.93614196777344, 161.72889709472656, 50.107269287109375, 187.61822509765625, 185.76107788085938, 101.97100830078125, 0.0, -13.60455322265625, 195.83718872070312, 11.537906646728516, -13.4039306640625, 32.22357177734375, 166.02880859375, 26.40484619140625, 158.034912109375, -15.53680419921875, 0.9322509765625, 161.5608367919922, -147.71420288085938, 148.90142822265625, -137.03648376464844, 26.073486328125, 149.54612731933594, 130.8099365234375, -16.78530502319336, 69.42820739746094, 178.70152282714844, 164.96571350097656, 50.97535705566406, 147.6508026123047, 0.22662734985351562, 172.892578125, 188.7750701904297, 186.77667236328125, 70.13993835449219, 1.622833251953125, 168.5565185546875, 158.24298095703125, 183.6185302734375, 0.0, 192.32705688476562, 22.999114990234375, 63.4571533203125, 144.28262329101562, -36.01441955566406, -79.94012451171875, -5.036823272705078, 2.2933349609375, -136.43905639648438, 175.6737823486328, 12.6473388671875, 130.52081298828125, -2.416107177734375, 92.53851318359375, -17.617691040039062, 215.70452880859375, 158.62069702148438, 4.61541748046875, -44.202789306640625, 17.48926544189453, 155.43673706054688, -23.356658935546875, -27.191375732421875, 9.65838623046875, 114.77069091796875, 1.3600540161132812, 115.51597595214844, -56.64044189453125, 70.11199951171875, 188.33474731445312, 234.36322021484375, -3.49664306640625, -4.231405258178711, 51.114471435546875, -201.9998779296875, 0.0, 6.354286193847656, 149.68966674804688, 9.607101440429688, 184.13217163085938, 53.410430908203125, 40.064697265625, 30.167388916015625, 149.38351440429688, 219.79547119140625, 151.93499755859375, 167.31365966796875, 231.3194580078125, 0.0, 176.98451232910156, 95.44775390625, 150.9971923828125, 30.28021240234375, 21.819793701171875, 40.03106689453125, 80.96826171875, 30.189300537109375, -145.51177978515625, 189.4478302001953, 97.7117919921875, -7.071533203125, 151.96099853515625, 34.09564208984375, 190.25433349609375, 46.738677978515625, 45.785919189453125, 16.37518310546875, 144.29896545410156, -8.012191772460938, -18.32345962524414, 63.016571044921875, 180.8836669921875, 161.7164764404297, 104.31999206542969, 11.474884033203125, 197.89605712890625, 42.397308349609375, 120.6721420288086], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000447.npy"}
{"epoch": 0.9361256544502617, "step": 448, "batch_size": 128, "mean": 62.62259292602539, "std": 97.75274658203125, "min": -177.26812744140625, "p10": -29.126977539062494, "median": 38.36216735839844, "p90": 183.4240478515625, "max": 267.3643798828125, "pos_frac": 0.7421875, "sample": [20.059890747070312, 9.961776733398438, -136.62460327148438, 10.312286376953125, 81.83697509765625, 191.3192138671875, 8.888885498046875, -177.26812744140625, 125.85723876953125, 160.57150268554688, -24.652130126953125, -165.77682495117188, 154.72235107421875, 40.659698486328125, 1.53857421875, 99.89764404296875, 155.96078491210938, 185.47235107421875, 60.8028564453125, -7.268957138061523, 150.52548217773438, 189.765625, 267.3643798828125, 6.522312164306641, 105.86076354980469, 26.506057739257812, 28.700592041015625, -22.131832122802734, 16.37127685546875, 8.122589111328125, -14.027824401855469, 39.0830078125, 203.5540771484375, -120.509765625, 159.39974975585938, -43.87469482421875, 166.3624267578125, 182.54620361328125, 83.85379028320312, 75.13629150390625, 132.47360229492188, 21.198829650878906, 195.15179443359375, -2.277862548828125, 91.44116973876953, 6.109912872314453, 84.12889099121094, 137.57864379882812, -3.955047607421875, 220.35504150390625, -11.548271179199219, -5.889862060546875, -9.921630859375, 97.51878356933594, 2.3747024536132812, -14.414398193359375, 35.47389221191406, 146.76568603515625, -46.09771728515625, 197.2535400390625, 151.59063720703125, 65.989013671875, 13.34515380859375, -0.3310966491699219, 177.578857421875, 219.90054321289062, 40.625728607177734, 14.056533813476562, 97.95834350585938, 211.326416015625, 171.80526733398438, 135.82301330566406, 131.7930908203125, 37.04595947265625, 12.213951110839844, 137.2471923828125, 187.64852905273438, 175.9111328125, 163.28570556640625, 120.64877319335938, -12.053207397460938, 191.67111206054688, -0.5055580139160156, 166.3429718017578, 12.009902954101562, -161.32461547851562, -151.22360229492188, -165.5184326171875, 170.04129028320312, 12.93499755859375, -12.344512939453125, -2.4726104736328125, 171.86282348632812, 173.11497497558594, 21.166839599609375, -148.94000244140625, 76.578125, -1.4882068634033203, 13.57574462890625, 162.02389526367188, 163.51165771484375, 16.61629295349121, 181.15565490722656, 8.21432876586914, 144.27078247070312, -27.484283447265625, -35.15858459472656, 115.15496826171875, 37.641326904296875, -26.840103149414062, 175.83326721191406, -32.959930419921875, -17.56524658203125, 4.89288330078125, -103.45772552490234, 22.41119384765625, 7.828643798828125, 19.730743408203125, 155.43771362304688, 1.5354042053222656, 139.17742919921875, -11.364044189453125, 9.459102630615234, 174.22567749023438, 235.568359375, 154.38771057128906, 51.0103759765625, 92.42205810546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000448.npy"}
{"epoch": 0.9382198952879581, "step": 449, "batch_size": 128, "mean": 70.43926239013672, "std": 100.18292236328125, "min": -198.84555053710938, "p10": -30.793037414550774, "median": 62.047698974609375, "p90": 192.20977783203125, "max": 227.118408203125, "pos_frac": 0.75, "sample": [173.82376098632812, 122.85037231445312, -36.32261657714844, 131.42173767089844, 60.755279541015625, -16.0706787109375, 14.087860107421875, 205.8511962890625, -5.45489501953125, 14.667930603027344, 12.08740234375, 30.239349365234375, 128.5482177734375, 140.97726440429688, -8.259811401367188, 150.52728271484375, 209.0406494140625, 190.37210083007812, 19.0511474609375, -2.503976821899414, -167.12109375, 150.9655303955078, 203.830078125, 160.59487915039062, 60.0399169921875, 111.66455841064453, -112.9932861328125, 72.23858642578125, 6.071831703186035, 194.127685546875, 210.44403076171875, 188.02999877929688, 167.53460693359375, 199.55145263671875, 26.53955078125, 177.11029052734375, 14.979217529296875, 110.79623413085938, 166.82579040527344, 143.7456512451172, 185.3815460205078, 26.79150390625, 145.72732543945312, 16.751136779785156, 227.118408203125, 218.86782836914062, 0.594085693359375, -5.246526718139648, 201.3349609375, 170.93368530273438, -96.80530548095703, 138.3232421875, 12.170547485351562, 17.665817260742188, 6.637556076049805, 161.08592224121094, 21.436431884765625, 150.0182342529297, 5.54827880859375, 104.46539306640625, 24.762161254882812, 51.27398681640625, 174.25790405273438, 18.37908935546875, 119.42776489257812, -16.691055297851562, 67.12420654296875, 63.340118408203125, -9.446319580078125, 174.12274169921875, -62.84710693359375, 85.43183898925781, 181.90182495117188, 163.47344970703125, -198.84555053710938, -2.05487060546875, 1.4014968872070312, 160.99456787109375, 32.53448486328125, 172.62322998046875, 159.38433837890625, 107.24470520019531, 221.71990966796875, 87.38412475585938, 138.42828369140625, 127.17234802246094, 172.67691040039062, 20.7147216796875, 142.19766235351562, -27.333194732666016, -3.2296829223632812, 33.182769775390625, -5.60870361328125, 200.4268798828125, 190.52255249023438, 191.3878173828125, -144.05276489257812, -5.999330520629883, 160.12283325195312, 65.04534912109375, -38.060546875, 153.67181396484375, -3.8193435668945312, -22.671249389648438, 104.37094116210938, -28.4232177734375, -143.50820922851562, -12.382652282714844, 162.658203125, 19.752960205078125, 5.281883239746094, 153.65371704101562, -45.16542053222656, -2.705944061279297, 39.365631103515625, -3.5266265869140625, 162.9373779296875, 135.50177001953125, 3.823394775390625, 200.65145874023438, 37.14630126953125, -173.48143005371094, 201.99090576171875, 16.044288635253906, 55.05877685546875, -173.49398803710938, -143.05877685546875, -7.403106689453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000449.npy"}
{"epoch": 0.9403141361256544, "step": 450, "batch_size": 128, "mean": 50.99037170410156, "std": 96.783935546875, "min": -183.59231567382812, "p10": -38.34573974609374, "median": 28.123727798461914, "p90": 186.87276611328124, "max": 223.13485717773438, "pos_frac": 0.6953125, "sample": [0.0, -148.35113525390625, 173.20713806152344, -48.72772979736328, 67.40447998046875, -15.33380126953125, 163.33209228515625, -13.0218505859375, 186.53350830078125, 47.01776123046875, 129.786865234375, 200.54550170898438, -1.5675277709960938, 88.764892578125, 168.21914672851562, 11.008613586425781, -35.13311767578125, 137.67938232421875, 16.384262084960938, 42.17321014404297, 0.0226287841796875, 123.72819519042969, 201.15505981445312, 193.47422790527344, 140.70510864257812, 24.8031005859375, -17.06926727294922, 16.792022705078125, 100.35714721679688, 15.974609375, 1.35394287109375, -2.5798110961914062, -15.1754150390625, 2.6116180419921875, 60.75810241699219, 13.94110107421875, 218.49313354492188, -157.0727081298828, 122.86988830566406, -104.21823120117188, -2.601062774658203, 195.0694580078125, 17.179744720458984, 95.19613647460938, 15.117965698242188, 86.30313873291016, -25.705663681030273, 29.329559326171875, 199.14279174804688, 81.42706298828125, -2.2938232421875, 174.935302734375, 17.644264221191406, 10.277992248535156, 78.5545654296875, -4.7843780517578125, 37.533905029296875, 199.8152313232422, 10.455291748046875, -166.6947021484375, 48.590057373046875, -183.59231567382812, 164.3975830078125, 51.663726806640625, -45.84185791015625, 196.357421875, 161.4039764404297, 2.9873714447021484, 153.98904418945312, 128.61441040039062, -2.3342132568359375, 136.29544067382812, -2.257568359375, 180.59832763671875, 5.473846435546875, 26.917896270751953, 197.00115966796875, 159.001953125, -101.46647644042969, 15.344039916992188, 73.80184936523438, 45.15496826171875, -13.065048217773438, 1.198150634765625, 193.36444091796875, 32.665771484375, -1.2593307495117188, 8.88092041015625, 135.90249633789062, -21.74060821533203, 10.219146728515625, 53.064517974853516, 187.66436767578125, 223.13485717773438, 24.295135498046875, 171.78872680664062, -30.322357177734375, -145.34326171875, -30.5146484375, -142.787841796875, 14.280216217041016, 98.4320068359375, 81.38980102539062, 166.34751892089844, 207.01095581054688, 52.978057861328125, -180.28822326660156, 117.47366333007812, 11.233642578125, 169.9074249267578, 180.21299743652344, 7.026885986328125, 35.62274932861328, -19.00677490234375, 181.12881469726562, -136.86297607421875, -5.3297271728515625, -17.2750244140625, 41.6478271484375, 137.86843872070312, 0.0, -13.497398376464844, 120.97726440429688, 40.87633514404297, -13.3553466796875, 165.70870971679688, -23.921844482421875, -119.8132553100586], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000450.npy"}
{"epoch": 0.9424083769633508, "step": 451, "batch_size": 128, "mean": 58.45191192626953, "std": 96.5649185180664, "min": -199.20962524414062, "p10": -35.34173316955566, "median": 35.004905700683594, "p90": 176.8959487915039, "max": 250.29052734375, "pos_frac": 0.75, "sample": [53.01043701171875, 30.99774169921875, -23.235198974609375, 7.838222503662109, 209.8760223388672, 165.00926208496094, 48.97918701171875, 2.565166473388672, -199.20962524414062, -30.70477294921875, 139.91966247558594, 7.785505294799805, 203.1915283203125, 22.332733154296875, 25.24041748046875, 132.502197265625, 83.69015502929688, 9.595062255859375, 193.62203979492188, -84.74838256835938, 169.11932373046875, -9.394256591796875, 9.406173706054688, 40.82843017578125, 217.9501953125, -0.51617431640625, 247.13577270507812, 12.63360595703125, -19.287307739257812, 34.7738037109375, -88.50846862792969, -1.1131210327148438, -182.86912536621094, 193.7652587890625, 118.97817993164062, 47.72955322265625, 24.335594177246094, 122.01629638671875, 33.599334716796875, 151.98974609375, 17.06502914428711, 166.74502563476562, 155.7850341796875, 157.58160400390625, 158.12100219726562, 176.6903533935547, 17.958816528320312, -4.7341766357421875, 205.65802001953125, -0.7803955078125, 6.0807952880859375, 212.50927734375, 22.51708984375, 218.79959106445312, 163.5753173828125, 165.30001831054688, -40.30892562866211, 55.5938720703125, -190.5762939453125, 160.09373474121094, 154.8441162109375, 41.590789794921875, 28.844589233398438, 173.30294799804688, 2.98394775390625, 78.10040283203125, 2.8463821411132812, 0.4912872314453125, 98.1629638671875, 33.97186279296875, 1.5107421875, 9.603179931640625, -31.620147705078125, 171.82791137695312, 11.124755859375, -42.409698486328125, 48.72149658203125, 11.190185546875, 170.96163940429688, 115.84201049804688, -1.867095947265625, -3.8071365356445312, -29.25079345703125, 41.26134490966797, 185.2515869140625, -30.146987915039062, 35.23600769042969, 109.355224609375, 14.27947998046875, 15.777137756347656, 145.94882202148438, 135.54013061523438, -22.4769287109375, 174.93959045410156, 41.58906555175781, -178.34249877929688, 38.04277801513672, -52.208953857421875, -33.21293640136719, 177.37567138671875, 46.07421875, 167.19485473632812, 70.29629516601562, 164.35321044921875, -9.644256591796875, -114.45286560058594, -119.40423583984375, -105.26318359375, 135.0858154296875, 5.182525634765625, 170.18118286132812, 21.3673095703125, 185.53271484375, 174.1885986328125, -3.7571640014648438, 12.747386932373047, 91.35211181640625, 4.85247802734375, 96.39231872558594, 250.29052734375, 52.406307220458984, -3.0724334716796875, -26.400894165039062, 2.7438125610351562, 147.76483154296875, 161.26727294921875, -41.65960693359375, 156.545166015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000451.npy"}
{"epoch": 0.9445026178010472, "step": 452, "batch_size": 128, "mean": 56.89568328857422, "std": 95.7474365234375, "min": -171.06365966796875, "p10": -32.42921142578125, "median": 32.08221435546875, "p90": 188.84803009033203, "max": 243.028564453125, "pos_frac": 0.75, "sample": [-69.37432861328125, 159.68280029296875, 39.67230224609375, 211.5029296875, -159.69281005859375, 12.266227722167969, 11.977088928222656, 199.6343994140625, -12.478195190429688, -26.544830322265625, 190.09521484375, -0.995330810546875, 5.1351165771484375, 13.281227111816406, -2.693775177001953, 8.87237548828125, 69.52294921875, -8.57806396484375, -8.650146484375, 17.678955078125, -0.253936767578125, -160.4145965576172, 130.23257446289062, 170.92214965820312, 158.07827758789062, 54.462860107421875, 47.529083251953125, 98.50929260253906, 0.08612060546875, -118.42245483398438, -123.6953125, 210.50051879882812, -29.051616668701172, 211.632080078125, 165.56121826171875, -31.487701416015625, 190.85885620117188, 180.51840209960938, -16.92675018310547, 153.76052856445312, 106.02791595458984, 14.91265869140625, 7.283596038818359, 49.18505859375, 151.88827514648438, 14.485902786254883, 66.4477767944336, 142.99424743652344, 116.99925231933594, 17.574684143066406, 185.547119140625, 41.1451416015625, 145.6226806640625, -10.942413330078125, 8.713005065917969, -52.17071533203125, 13.380355834960938, 18.064361572265625, 13.808258056640625, 22.030120849609375, 11.833656311035156, 42.968048095703125, -110.2015380859375, 120.85025024414062, -72.51754760742188, 159.9976043701172, 175.21640014648438, 184.36660766601562, 5.03759765625, 14.011322021484375, 8.35345458984375, 190.18362426757812, 23.65772247314453, 13.468994140625, 180.3060302734375, 243.028564453125, 164.69483947753906, -16.740936279296875, 40.89276123046875, 42.99085998535156, -157.91476440429688, 113.62631225585938, 194.61949157714844, -4.11407470703125, 188.3135223388672, 167.39627075195312, -6.402381896972656, 156.2460479736328, 210.54864501953125, 28.1793212890625, 39.16157531738281, 51.0521240234375, 22.176631927490234, -0.49481201171875, 11.3447265625, -41.76763916015625, 162.4447784423828, 46.08354187011719, 61.912506103515625, -18.51104736328125, 225.87969970703125, 56.32275390625, 24.726165771484375, 158.54425048828125, 65.35578918457031, 2.396228790283203, -29.14666748046875, 21.360916137695312, 182.88327026367188, 197.07505798339844, 227.92294311523438, 186.8925323486328, 12.302261352539062, 38.292724609375, 175.4508056640625, 0.0, 106.54400634765625, -2.1139907836914062, -153.37387084960938, -34.626068115234375, 35.985107421875, 51.329559326171875, -171.06365966796875, 51.8311767578125, 4.226165771484375, 178.65997314453125, 6.79443359375, 0.18558311462402344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000452.npy"}
{"epoch": 0.9465968586387434, "step": 453, "batch_size": 128, "mean": 67.36186218261719, "std": 91.94129943847656, "min": -188.21929931640625, "p10": -27.187750244140616, "median": 51.66327667236328, "p90": 187.9800842285156, "max": 252.29742431640625, "pos_frac": 0.7890625, "sample": [25.633689880371094, 157.70205688476562, 14.33447265625, 130.73321533203125, 161.02081298828125, 5.251068115234375, 20.497879028320312, -2.5318603515625, 162.52650451660156, 108.75892639160156, 61.23682403564453, 76.76966857910156, 61.29804992675781, 55.14129638671875, -18.29437255859375, -51.090484619140625, 111.47354125976562, -188.21929931640625, 29.032012939453125, 188.75100708007812, -23.217132568359375, 181.58978271484375, 178.81964111328125, 155.94334411621094, 51.66412353515625, 15.732421875, -32.99560546875, 173.87335205078125, 47.97276306152344, 175.55262756347656, 24.305221557617188, 148.61607360839844, 41.74168395996094, 147.50526428222656, 35.446502685546875, 27.00360107421875, 38.92486572265625, -4.782806396484375, 195.21385192871094, 195.18804931640625, 8.048553466796875, 184.01849365234375, 34.2216796875, 104.08303833007812, 157.38369750976562, -166.27911376953125, 15.152191162109375, 12.988296508789062, 12.5157470703125, -52.40252685546875, 5.350250244140625, 171.6717529296875, 2.7788543701171875, 21.061355590820312, -177.6851806640625, 137.49154663085938, -4.937347412109375, -47.170013427734375, 51.66242980957031, 0.409637451171875, 144.60440063476562, -5.0620574951171875, 0.0, 138.25765991210938, 91.52101135253906, 25.905990600585938, 130.5220947265625, 13.213104248046875, 170.78213500976562, 24.808013916015625, 92.63467407226562, 85.92803955078125, 53.2166748046875, -166.030517578125, 24.07415771484375, 187.64968872070312, 207.21319580078125, 176.71536254882812, -3.076568603515625, 166.11814880371094, -10.357589721679688, 200.74972534179688, 47.362579345703125, -8.6588134765625, 6.845733642578125, 18.81475830078125, 196.28521728515625, -18.538330078125, -32.143890380859375, 198.60577392578125, 46.189300537109375, 137.0181884765625, -53.478271484375, -35.73480224609375, 176.50418090820312, 22.251739501953125, -166.95703125, 170.55322265625, -10.36383056640625, 252.29742431640625, 99.59908294677734, 46.489593505859375, 80.30065155029297, 117.98158264160156, 64.37966918945312, 66.354248046875, 1.5194778442382812, 4.0499420166015625, 195.47616577148438, 108.7545166015625, 171.6514434814453, 140.84625244140625, 27.24853515625, 166.8448486328125, 56.6475830078125, 54.74125671386719, 192.6510009765625, 192.98846435546875, -11.08544921875, 22.710891723632812, 176.38229370117188, 7.562164306640625, 140.25360107421875, -25.063690185546875, 190.92681884765625, -48.0615234375, 105.3436508178711, 224.09918212890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000453.npy"}
{"epoch": 0.9486910994764398, "step": 454, "batch_size": 128, "mean": 61.74987030029297, "std": 96.13810729980469, "min": -246.76202392578125, "p10": -22.063314819335933, "median": 27.513458251953125, "p90": 192.60648803710936, "max": 240.856201171875, "pos_frac": 0.7578125, "sample": [42.96588134765625, 181.2220458984375, -160.72421264648438, -53.88494873046875, -5.363739013671875, 10.15863037109375, 192.27545166015625, 159.46389770507812, 9.297607421875, 38.02330017089844, 19.726608276367188, 22.5079288482666, -8.106584548950195, 140.13394165039062, 13.88646125793457, 122.25138092041016, -157.94412231445312, 7.127265930175781, 108.30459594726562, -2.818479537963867, 158.37197875976562, 68.8494644165039, 26.401611328125, 5.081390380859375, 225.7799072265625, 240.856201171875, -8.28839111328125, 201.04425048828125, -246.76202392578125, 7.714202880859375, 19.07372283935547, 163.19618225097656, 124.3250503540039, 142.66249084472656, 180.5145721435547, 15.460006713867188, 213.08755493164062, -30.435379028320312, 18.2169189453125, 18.196792602539062, 185.95870971679688, 18.288223266601562, 203.73736572265625, 151.80450439453125, 30.922080993652344, 159.93319702148438, 130.79916381835938, -11.70623779296875, 194.07302856445312, -12.682861328125, 16.5126953125, 100.39581298828125, 34.729949951171875, -1.5532684326171875, 168.05654907226562, -17.18304443359375, 140.548095703125, 5.5442657470703125, 176.7852783203125, 193.37890625, 136.00942993164062, 164.89317321777344, 198.429443359375, 117.49781799316406, 207.1563720703125, 189.47882080078125, 177.71728515625, 25.3428955078125, 15.193740844726562, -1.652984619140625, 200.95330810546875, 173.79827880859375, -0.98541259765625, -71.32327270507812, 11.982879638671875, 35.74894714355469, -20.23535919189453, 62.6754150390625, 30.18115234375, -157.812255859375, -17.252334594726562, 3.1834945678710938, 189.800537109375, -40.31182861328125, -25.092681884765625, 32.85894775390625, 156.65341186523438, 67.1107177734375, -20.75225830078125, 169.84732055664062, 73.43162536621094, 14.870773315429688, 18.51593017578125, 194.58062744140625, 223.41000366210938, 2.05596923828125, 24.75542449951172, 7.8634033203125, 187.8227996826172, 3.074939727783203, -26.930679321289062, 14.62567138671875, 181.3154296875, 44.23755645751953, 23.582061767578125, 26.878662109375, 204.92343139648438, 79.46810150146484, -111.09002685546875, 164.46893310546875, 3.0892562866210938, -2.819000244140625, 77.89361572265625, 1.623138427734375, 28.14825439453125, 148.82664489746094, 147.69354248046875, -4.878692626953125, -83.2900390625, 34.98133087158203, 15.624053955078125, -11.868728637695312, 33.61920166015625, 185.46881103515625, -20.7650146484375, -3.594329833984375, 13.1761474609375, -112.09129333496094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000454.npy"}
{"epoch": 0.9507853403141361, "step": 455, "batch_size": 128, "mean": 49.86511993408203, "std": 95.5196533203125, "min": -205.3111572265625, "p10": -41.43996429443359, "median": 26.81792449951172, "p90": 174.74907684326172, "max": 240.50900268554688, "pos_frac": 0.7109375, "sample": [158.38592529296875, 104.31117248535156, 27.043563842773438, -10.852828979492188, 144.79339599609375, 8.949188232421875, 133.9068603515625, 176.89732360839844, 27.1065673828125, -20.56622314453125, -13.410003662109375, 85.90151977539062, -2.916717529296875, 157.62225341796875, 70.445068359375, 171.6156463623047, 184.96041870117188, 154.4563446044922, 231.25851440429688, -56.434356689453125, 9.030487060546875, 40.916595458984375, 5.3508453369140625, 162.2742919921875, 240.50900268554688, 177.2969207763672, 168.89532470703125, 16.45318603515625, 33.56454086303711, 10.711212158203125, -20.400924682617188, -182.00845336914062, -0.4456901550292969, 26.59228515625, 34.95208740234375, 4.107421875, -13.212799072265625, -23.690887451171875, -106.4242935180664, 24.217864990234375, 119.10659790039062, -9.65966796875, 145.94619750976562, 221.30703735351562, 22.1907958984375, 36.5703125, 164.19935607910156, -3.95123291015625, 172.7547607421875, 25.94134521484375, 10.172821044921875, 10.540191650390625, 28.265304565429688, -7.914520263671875, -5.9933929443359375, -0.78509521484375, 15.68310546875, -176.98464965820312, -124.86739349365234, 1.3044986724853516, 24.537643432617188, -30.614120483398438, 5.303834915161133, -34.72418212890625, -31.221588134765625, -102.43254089355469, 127.13008117675781, 43.43495178222656, 156.7546844482422, 133.85842895507812, -170.46078491210938, 145.54803466796875, 181.77740478515625, 173.6368408203125, 47.02069091796875, -16.190643310546875, -18.70050048828125, 178.1917724609375, 173.82839965820312, 87.64424133300781, 87.51754760742188, -21.070266723632812, 9.332111358642578, -102.63219451904297, 165.11952209472656, -1.9029083251953125, -175.6172637939453, 94.72686004638672, -2.42132568359375, 24.548595428466797, -6.542816162109375, 23.858154296875, 2.9161376953125, 36.5570068359375, 17.668533325195312, 139.96759033203125, 221.186767578125, 221.26043701171875, 127.39091491699219, -40.1871337890625, 9.54901123046875, 6.070049285888672, 139.21673583984375, 51.264801025390625, 3.78985595703125, -3.3132705688476562, 18.11022186279297, 182.03167724609375, 22.12066650390625, 55.979591369628906, -44.36323547363281, 129.03662109375, 136.20779418945312, 32.33259582519531, 169.68032836914062, -100.21875, -205.3111572265625, 47.02398681640625, -71.19296264648438, 172.82623291015625, 155.2684326171875, 60.085296630859375, 203.69705200195312, 54.81272888183594, 20.70819091796875, 40.5260009765625, 181.4788055419922, 33.330535888671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000455.npy"}
{"epoch": 0.9528795811518325, "step": 456, "batch_size": 128, "mean": 74.89949035644531, "std": 107.79398345947266, "min": -196.98904418945312, "p10": -52.12667236328125, "median": 51.36519241333008, "p90": 202.0148727416992, "max": 343.85992431640625, "pos_frac": 0.7421875, "sample": [43.829986572265625, -8.786937713623047, -53.1949462890625, 168.15142822265625, 19.85723114013672, -9.388755798339844, 163.9552459716797, -111.63790893554688, 192.16937255859375, 125.4361572265625, 181.11572265625, -196.98904418945312, 21.506362915039062, 2.1785888671875, -34.2935791015625, 207.71463012695312, 144.80616760253906, -53.115753173828125, 7.838531494140625, 13.4520263671875, 186.43426513671875, 151.96435546875, -11.887939453125, 174.0667724609375, 1.2943143844604492, -17.747051239013672, 127.1470947265625, 13.522994995117188, 170.87966918945312, -74.91165161132812, 156.43319702148438, -150.40924072265625, 152.45846557617188, -34.17686462402344, 154.03753662109375, 187.8770751953125, 130.7464141845703, 159.14321899414062, 172.3925018310547, 141.49090576171875, 227.9197998046875, -22.76806640625, 192.98492431640625, 25.00140380859375, 200.53857421875, 169.4691925048828, 207.77459716796875, -5.414459228515625, 10.94049072265625, 167.45452880859375, 38.882537841796875, 137.8997802734375, -10.862518310546875, 32.803062438964844, -52.4017333984375, 16.1639404296875, 2.464202880859375, 19.33438491821289, 27.78668212890625, 217.064697265625, 18.78790283203125, -6.598115921020508, 17.743064880371094, -127.85851287841797, -29.800552368164062, 191.2689208984375, 158.4241943359375, -152.9293975830078, 27.83575439453125, 21.948974609375, 14.51174545288086, 146.7159423828125, -5.426540374755859, 144.60809326171875, 58.90039825439453, 151.05535888671875, 219.96792602539062, 152.070068359375, -101.66738891601562, 69.51394653320312, 157.17848205566406, 210.1983642578125, -171.84841918945312, 194.4818115234375, 13.245285034179688, -4.140655517578125, 21.265090942382812, 211.75848388671875, 195.13973999023438, 204.10948181152344, 139.5887451171875, 211.136962890625, -4.113700866699219, 189.13717651367188, 0.0, 3.997335433959961, 133.01612854003906, 184.96670532226562, 148.65036010742188, 132.3750457763672, 13.9532470703125, 12.88360595703125, 117.279052734375, -1.6639862060546875, 134.39559936523438, -143.950439453125, 172.93951416015625, 5.21380615234375, 14.24163818359375, -19.727294921875, 0.740570068359375, 139.1107940673828, 113.77264404296875, 137.38656616210938, 201.67054748535156, 1.2445030212402344, 279.4859619140625, 76.58822631835938, 343.85992431640625, 231.60354614257812, 185.00942993164062, 31.756378173828125, -10.21624755859375, -0.92852783203125, 178.00624084472656, -52.0087890625, -135.94064331054688, 202.81829833984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000456.npy"}
{"epoch": 0.9549738219895288, "step": 457, "batch_size": 128, "mean": 65.94767761230469, "std": 99.80509948730469, "min": -209.48031616210938, "p10": -37.57227706909179, "median": 40.240264892578125, "p90": 190.92884521484373, "max": 233.46556091308594, "pos_frac": 0.7109375, "sample": [197.44732666015625, 16.893402099609375, -29.04657745361328, -8.116668701171875, 154.29217529296875, 34.740234375, 172.6168670654297, 182.16159057617188, 217.80999755859375, 28.756240844726562, 173.783447265625, -0.232757568359375, 207.54534912109375, 131.96002197265625, 8.19049072265625, 214.77664184570312, -12.140228271484375, 34.25848388671875, -139.32696533203125, 188.92379760742188, 171.69821166992188, 164.3181915283203, 164.28414916992188, 233.46556091308594, 158.39459228515625, 40.635498046875, 185.99606323242188, 14.090408325195312, 185.5321044921875, 20.05615234375, 39.84503173828125, -20.678192138671875, -3.27642822265625, -25.789993286132812, 158.78985595703125, 182.39007568359375, -36.42755889892578, 49.74530029296875, -30.954132080078125, 124.85107421875, -56.420928955078125, 62.03314971923828, -32.57472229003906, -31.41143798828125, 5.069141387939453, -209.48031616210938, 203.76748657226562, 27.10791015625, 193.58306884765625, -7.334136962890625, -46.9036865234375, 75.35223388671875, 135.39370727539062, -55.071533203125, 173.1680450439453, -5.690900802612305, -171.77056884765625, 149.18589782714844, 49.78887939453125, 24.52984619140625, -18.2674560546875, -21.30461883544922, 30.4234619140625, -28.398521423339844, 25.55645751953125, 60.09423828125, 144.5212860107422, 16.7205810546875, 230.2533721923828, -12.92718505859375, 30.447677612304688, 169.60659790039062, 156.24588012695312, 41.26153564453125, 18.739166259765625, -160.7255096435547, -49.188079833984375, 36.832305908203125, 61.665924072265625, 110.35545349121094, 23.908447265625, 183.15261840820312, 55.38154602050781, 167.9427490234375, 146.45590209960938, -30.347442626953125, 140.3999481201172, -3.7736358642578125, 174.22576904296875, 219.20367431640625, 7.346540451049805, -18.37213134765625, 179.692138671875, -140.16729736328125, -50.39959716796875, 182.43402099609375, -29.44940185546875, 130.83660888671875, -20.643218994140625, 189.79132080078125, 159.1706085205078, 33.68133544921875, -41.13069152832031, 51.347442626953125, -40.2432861328125, 212.886474609375, 29.49322509765625, 133.52273559570312, 15.130569458007812, 125.25257873535156, 195.89602661132812, 52.719154357910156, 4.43963623046875, 198.0012664794922, 184.43496704101562, 150.79415893554688, 5.481452941894531, 35.27911376953125, -153.35003662109375, 98.47785949707031, 201.18927001953125, 1.3325424194335938, 0.0, 188.622802734375, -2.0256195068359375, 158.4364471435547, 51.784393310546875, 6.567047119140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000457.npy"}
{"epoch": 0.9570680628272251, "step": 458, "batch_size": 128, "mean": 59.10076904296875, "std": 98.0514144897461, "min": -192.43731689453125, "p10": -31.635995483398435, "median": 25.4671630859375, "p90": 195.05085449218748, "max": 247.8152618408203, "pos_frac": 0.75, "sample": [13.538551330566406, -50.10272216796875, 12.20587158203125, -167.55899047851562, 209.48431396484375, 18.384963989257812, -47.536712646484375, 187.49069213867188, -12.296859741210938, -33.3016357421875, 18.85601806640625, -130.5675048828125, -41.9202880859375, 0.7008895874023438, 2.7553672790527344, 109.64314270019531, 34.896942138671875, 51.442535400390625, -22.23663330078125, 56.056610107421875, 196.57437133789062, -1.5786361694335938, 110.12442016601562, -1.9738082885742188, 25.752288818359375, 4.564231872558594, 181.462890625, 194.39791870117188, -9.4864501953125, 162.4404296875, 17.791976928710938, -1.3832473754882812, 147.92788696289062, 222.92434692382812, 36.556671142578125, 21.85638427734375, 17.64288330078125, 106.614501953125, 190.10870361328125, -30.922149658203125, 198.76898193359375, -3.376810073852539, -168.4942626953125, 25.182037353515625, 7.503021240234375, 5.9472503662109375, 12.923095703125, 11.716915130615234, 6.124824523925781, 67.2665786743164, 4.690101623535156, -87.83563232421875, -18.358856201171875, 179.71270751953125, 156.13858032226562, -24.39581298828125, 1.99371337890625, -0.20200347900390625, 142.67544555664062, 210.372314453125, 37.7518310546875, 3.018899917602539, 238.1328125, 53.518463134765625, 142.36341857910156, 7.8355712890625, 12.313720703125, -143.12557983398438, -28.934967041015625, 212.5394287109375, 176.20272827148438, 41.43450927734375, -38.968231201171875, 29.392333984375, -83.237548828125, 165.3998260498047, 136.01768493652344, -18.0240478515625, 7.312126159667969, 153.9375, 170.44146728515625, -28.148162841796875, -10.359588623046875, 87.6307373046875, 205.61990356445312, 29.138320922851562, 247.8152618408203, 21.1885986328125, 42.446075439453125, 164.95449829101562, 35.69610595703125, 209.21780395507812, 160.94654846191406, 143.968505859375, 220.26138305664062, -4.842796325683594, 152.39144897460938, -6.75616455078125, 6.217142105102539, 141.47207641601562, 44.0435791015625, 12.71148681640625, 8.05999755859375, 179.73553466796875, 134.21902465820312, 109.75572967529297, 9.582542419433594, 0.2362060546875, 171.30921936035156, 188.2851104736328, 29.647125244140625, 209.92164611816406, 162.52276611328125, 149.1441650390625, 190.40191650390625, 19.3145751953125, 145.85736083984375, -0.7883148193359375, -183.717041015625, -192.43731689453125, 146.21237182617188, 196.69305419921875, 20.916034698486328, 3.35052490234375, 148.5857391357422, 32.905670166015625, 6.040473937988281, -1.4733428955078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000458.npy"}
{"epoch": 0.9591623036649215, "step": 459, "batch_size": 128, "mean": 43.55192565917969, "std": 103.16400146484375, "min": -200.436279296875, "p10": -66.09205932617186, "median": 18.05237865447998, "p90": 188.28225860595703, "max": 241.0264892578125, "pos_frac": 0.65625, "sample": [5.6851959228515625, 2.7207794189453125, 18.54054832458496, 168.77902221679688, 0.4195556640625, 168.64785766601562, -155.30337524414062, 45.88671875, 14.393524169921875, 31.589675903320312, 65.74642944335938, 127.87905883789062, 65.96734619140625, -17.10400390625, 19.63001251220703, 6.710689544677734, 189.06590270996094, 112.120361328125, -18.612821578979492, 17.564208984375, 70.36712646484375, 19.40594482421875, -54.504180908203125, 31.246414184570312, 26.762054443359375, -17.376541137695312, 158.26504516601562, 1.4121932983398438, 169.0716552734375, -14.891159057617188, -16.759498596191406, 0.0, 3.8269500732421875, -39.117156982421875, 177.35003662109375, -136.48822021484375, 97.63697052001953, -25.984161376953125, -14.36676025390625, 91.22677612304688, 184.3995361328125, -14.428375244140625, 18.87841796875, 66.30111694335938, -163.03012084960938, 49.909019470214844, -28.757659912109375, 27.247802734375, -16.443111419677734, 196.46142578125, 4.9097442626953125, 9.722127914428711, -3.3799571990966797, 227.51571655273438, 59.95552062988281, -4.699180603027344, 29.023452758789062, -163.51821899414062, 181.7929229736328, 241.0264892578125, 143.97756958007812, -144.6372833251953, 191.313720703125, 161.48704528808594, 165.99615478515625, 211.7613525390625, 50.065673828125, 155.5576171875, -176.57086181640625, -169.14688110351562, 15.5299072265625, 43.32391357421875, 170.96226501464844, 169.9853515625, -16.820785522460938, -190.3677978515625, 204.5633544921875, 130.66342163085938, 8.72613525390625, -64.52740478515625, -200.436279296875, -4.310791015625, -4.97393798828125, -163.91354370117188, 187.9464111328125, 50.891998291015625, -17.0440673828125, 186.00421142578125, 0.0, 7.094245910644531, 156.7812042236328, -22.24267578125, 24.548004150390625, 225.6871337890625, -118.96929931640625, 218.578369140625, -5.8040924072265625, 141.67945861816406, 17.034805297851562, 26.871604919433594, -70.66091918945312, 151.12152099609375, 17.269775390625, -9.929275512695312, -15.263763427734375, 92.97406768798828, 196.36062622070312, 12.743194580078125, -0.122650146484375, -4.16400146484375, -46.520050048828125, -69.742919921875, 1.045928955078125, -41.943359375, 173.2470245361328, 204.74273681640625, -0.011688232421875, 16.739486694335938, 183.65081787109375, 32.57769775390625, -5.839389801025391, 199.35128784179688, 215.7608642578125, 13.65496826171875, 41.123992919921875, 7.2535400390625, 66.65353393554688, 145.01133728027344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000459.npy"}
{"epoch": 0.9612565445026178, "step": 460, "batch_size": 128, "mean": 43.92329406738281, "std": 106.98551177978516, "min": -205.3743896484375, "p10": -104.93582077026366, "median": 23.445289611816406, "p90": 177.65089721679686, "max": 242.07919311523438, "pos_frac": 0.6953125, "sample": [9.688491821289062, 86.101318359375, -133.02359008789062, 165.92047119140625, 64.36654663085938, 173.60305786132812, -178.85385131835938, 138.41079711914062, -44.627685546875, 199.30084228515625, -149.4505615234375, 18.389495849609375, 15.71783447265625, 12.4169921875, 23.96478271484375, -8.6666259765625, -123.80349731445312, 16.44720458984375, 176.87442016601562, 199.250732421875, -60.696990966796875, 28.06182861328125, 69.57257080078125, 136.84591674804688, 93.46626281738281, 10.644775390625, 169.41561889648438, -8.190200805664062, 209.86871337890625, -8.15374755859375, -17.7073974609375, 169.8793487548828, 95.1207275390625, 240.00054931640625, 90.39041137695312, -84.97343444824219, 140.81454467773438, -157.06564331054688, 239.83984375, -20.46795654296875, -185.33164978027344, 172.74429321289062, -89.35540771484375, 0.8879756927490234, 99.757568359375, 169.51629638671875, -36.58892822265625, 143.2941131591797, 5.4452056884765625, -98.77056884765625, 89.27645874023438, 0.5350151062011719, 0.244384765625, -0.7932052612304688, -110.73735809326172, 23.055862426757812, -14.543968200683594, -0.9371337890625, -2.4312591552734375, -19.725006103515625, 197.39328002929688, 66.91268920898438, 159.2823486328125, 54.334434509277344, 19.376251220703125, -125.84515380859375, 0.0021991729736328125, 53.6275634765625, 170.16049194335938, -0.40003204345703125, -0.6871337890625, 124.051513671875, 19.48492431640625, 20.2781982421875, 70.734619140625, 139.26914978027344, 171.55404663085938, 163.21331787109375, 23.834716796875, 31.06191635131836, 202.93505859375, 31.241195678710938, -24.39599609375, 236.670166015625, 49.7542724609375, 242.07919311523438, 194.08547973632812, 176.35696411132812, -148.22381591796875, 183.64288330078125, 117.04345703125, 133.239501953125, -160.26974487304688, 7.848846435546875, -47.21095275878906, 7.60174560546875, 164.8479766845703, -205.3743896484375, 153.19345092773438, -70.99557495117188, 115.80586242675781, -152.4757080078125, 156.99642944335938, 9.272506713867188, 16.408103942871094, -83.32228088378906, -6.12359619140625, -69.77767944335938, 172.086181640625, 41.454681396484375, 21.811309814453125, 179.46267700195312, 34.025634765625, 169.70703125, -126.9365234375, 5.587371826171875, 6.266899108886719, 154.91146850585938, 13.10028076171875, 86.611572265625, 180.79666137695312, 7.6709747314453125, 38.45849609375, -102.44944763183594, -94.36997985839844, 4.200777053833008, 24.165374755859375, 72.89212036132812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000460.npy"}
{"epoch": 0.9633507853403142, "step": 461, "batch_size": 128, "mean": 57.33232879638672, "std": 106.91297912597656, "min": -206.25619506835938, "p10": -72.50583496093746, "median": 39.51051330566406, "p90": 191.8295684814453, "max": 238.32952880859375, "pos_frac": 0.7109375, "sample": [18.238235473632812, -168.52041625976562, 195.0738525390625, 52.478431701660156, 42.86763000488281, 168.27032470703125, 167.39756774902344, 24.0057373046875, -136.0031280517578, 200.89926147460938, 170.6632080078125, 0.0, 14.0413818359375, 238.32952880859375, 150.8604736328125, 154.88214111328125, 157.26370239257812, 80.65646362304688, -3.49969482421875, 11.352874755859375, 91.33570098876953, 191.38323974609375, 52.471046447753906, -97.03134155273438, 75.52871704101562, -7.456939697265625, 13.4417724609375, 208.08737182617188, -38.59281921386719, 163.8480224609375, 0.2949333190917969, 211.03640747070312, 57.27349853515625, 161.11911010742188, -40.3994140625, 164.59475708007812, 206.89739990234375, 19.926193237304688, -179.5201416015625, -50.373992919921875, 192.87100219726562, -164.9443359375, 1.8997955322265625, 17.1849365234375, 146.3193359375, 3.8832473754882812, 136.48101806640625, 154.30374145507812, 192.90232849121094, -6.595184326171875, 154.64724731445312, -123.82894897460938, 22.659515380859375, -1.726043701171875, 12.128498077392578, -10.430679321289062, 165.02728271484375, -194.06463623046875, -39.98974609375, 130.57766723632812, 144.79180908203125, 2.1942672729492188, -14.584426879882812, 163.43826293945312, 200.97695922851562, 24.622589111328125, -206.25619506835938, 0.041046142578125, 184.2171173095703, 7.9826202392578125, -14.282695770263672, 174.83963012695312, -0.753173828125, 183.43951416015625, -14.460479736328125, 209.60455322265625, 85.1385498046875, 170.99822998046875, -61.994903564453125, -38.47308349609375, 158.16880798339844, 5.0804443359375, 146.58236694335938, 85.07516479492188, 168.01388549804688, 73.28075408935547, -25.025115966796875, -120.2896728515625, -8.66157341003418, 119.521728515625, 112.40045166015625, 136.74693298339844, 12.714599609375, 150.23744201660156, 39.256744384765625, 206.61790466308594, 0.92095947265625, 36.92730712890625, 165.207763671875, -140.78604125976562, 112.0403060913086, 1.5346565246582031, 197.0495147705078, 21.328079223632812, -47.9398193359375, -37.06040954589844, 18.654144287109375, 163.82415771484375, 165.39599609375, -11.706344604492188, -102.62954711914062, 132.26519775390625, 1.7458152770996094, 143.60009765625, -9.907424926757812, 2.4670181274414062, 45.59626770019531, 159.8207550048828, -170.11569213867188, 13.948005676269531, 39.7642822265625, -5.127784729003906, -156.95208740234375, 197.8765869140625, -20.248788833618164, 80.45730590820312, 161.78514099121094, 185.17404174804688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000461.npy"}
{"epoch": 0.9654450261780104, "step": 462, "batch_size": 128, "mean": 75.97261047363281, "std": 106.86404418945312, "min": -194.04248046875, "p10": -32.78356246948242, "median": 79.0080337524414, "p90": 201.67142639160156, "max": 252.04705810546875, "pos_frac": 0.765625, "sample": [155.83619689941406, 64.31155395507812, 6.2104339599609375, 83.81005859375, 190.27197265625, 44.479583740234375, 172.23342895507812, 123.7640380859375, 15.775703430175781, 201.56393432617188, -158.1319580078125, 159.07318115234375, 202.75259399414062, 22.539031982421875, 202.761962890625, 3.009002685546875, 134.2461395263672, 166.37567138671875, 127.1734619140625, 12.376495361328125, 34.6563720703125, 154.56060791015625, 148.67721557617188, 48.8863525390625, 27.362049102783203, -10.08978271484375, 183.42726135253906, -1.6200332641601562, 172.1378173828125, -136.5993194580078, -159.67816162109375, 160.27923583984375, -2.822113037109375, 3.123443603515625, -2.07940673828125, 167.10894775390625, 201.9222412109375, -135.2288818359375, -16.799957275390625, 13.603729248046875, 167.3201904296875, 166.72921752929688, 189.637451171875, 6.9887237548828125, 240.5836181640625, 134.7454833984375, 170.12168884277344, 189.27041625976562, 135.0299072265625, 3.090789794921875, 204.83189392089844, 7.692665100097656, 163.9859619140625, -158.23165893554688, 14.769302368164062, -16.529251098632812, 193.261962890625, 30.767303466796875, 9.577003479003906, 16.517822265625, 165.07501220703125, 220.68170166015625, 25.738994598388672, 174.19309997558594, -21.683746337890625, 105.89080047607422, -141.99356079101562, 98.265625, 249.25390625, -31.294876098632812, 182.92288208007812, 179.57888793945312, 170.84498596191406, 144.045654296875, 197.839599609375, 88.26849365234375, -15.6370849609375, 24.341552734375, 11.235677719116211, -15.42340087890625, 15.696353912353516, -29.94573974609375, -0.458526611328125, 2.5794525146484375, -171.1590576171875, 74.20600891113281, 155.06515502929688, 237.2843017578125, -85.44960021972656, 133.60299682617188, -120.73029327392578, 211.86318969726562, 58.430992126464844, 45.19324493408203, 139.8289031982422, 252.04705810546875, -36.257164001464844, 124.39328002929688, 148.67019653320312, 175.47537231445312, 133.41233825683594, -10.37823486328125, 246.84896850585938, 146.05389404296875, -15.98052978515625, 20.37664794921875, 156.00341796875, 18.622314453125, 27.297271728515625, 12.28729248046875, 171.5902099609375, 7.08184814453125, 69.53460693359375, 192.165283203125, -93.43771362304688, 219.79342651367188, -30.270126342773438, 139.95968627929688, 141.72409057617188, -19.626022338867188, 205.57241821289062, 20.360015869140625, -3.981170654296875, -91.33740234375, 140.79965209960938, 137.40875244140625, 182.7551727294922, -194.04248046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000462.npy"}
{"epoch": 0.9675392670157068, "step": 463, "batch_size": 128, "mean": 50.54184341430664, "std": 94.24430084228516, "min": -193.40573120117188, "p10": -33.89945983886719, "median": 20.995532989501953, "p90": 184.3289855957031, "max": 230.5369873046875, "pos_frac": 0.71875, "sample": [77.5804443359375, 70.42652893066406, 65.48123168945312, 2.3242874145507812, 22.799102783203125, 1.9303874969482422, 202.281005859375, 2.168701171875, -179.9290771484375, 88.07763671875, 173.8577880859375, 0.7521820068359375, -162.6205291748047, -41.22492218017578, -5.034576416015625, 195.57330322265625, 123.33563232421875, 117.76480102539062, 14.65838623046875, -3.796659469604492, 19.051551818847656, -193.40573120117188, -24.29180908203125, 6.2178955078125, 20.666749954223633, -0.655975341796875, -103.49856567382812, 36.21258544921875, 190.96954345703125, 189.7918701171875, -166.6676788330078, 31.35230255126953, 12.34124755859375, -31.09674072265625, -34.48809814453125, 201.4573974609375, 199.280029296875, 169.79420471191406, 230.5369873046875, -47.54066467285156, 54.46856689453125, -25.240631103515625, 118.41073608398438, 219.485595703125, 81.69831848144531, 6.538818359375, 206.848876953125, -3.4888992309570312, 27.31280517578125, 6.0574798583984375, 16.114112854003906, 31.89057159423828, 182.48406982421875, 58.72015380859375, 9.019859313964844, 17.6708984375, 190.44131469726562, -28.711181640625, 188.6337890625, 148.2769317626953, 95.5484619140625, 178.115478515625, 9.511749267578125, -7.3729400634765625, -27.430511474609375, 19.966934204101562, 176.25933837890625, 3.4397735595703125, 71.6612548828125, 59.73358154296875, 172.62257385253906, -35.106109619140625, -7.44525146484375, 68.11407470703125, 46.641815185546875, 6.09161376953125, 135.531494140625, -32.08345031738281, -26.56182098388672, 14.86431884765625, 167.66839599609375, 11.098175048828125, 182.46868896484375, 142.70181274414062, 73.76705932617188, 6.672025680541992, 21.00469970703125, 155.6904296875, 53.99620056152344, -23.204614639282227, -15.96337890625, 125.42889404296875, -33.647186279296875, 0.0, -9.34619140625, 5.269344329833984, 8.053386688232422, 199.00765991210938, 32.827301025390625, -12.227470397949219, 100.680908203125, 104.42585754394531, -15.436630249023438, 3.2913856506347656, 20.986366271972656, -124.06887817382812, 19.23565673828125, 19.50330352783203, 136.64559936523438, -0.5586872100830078, 13.965972900390625, -123.20343017578125, -75.3819580078125, 119.07539367675781, -2.576141357421875, 167.36476135253906, 170.68121337890625, 41.5357780456543, 175.47122192382812, 129.922119140625, 220.50393676757812, 38.75341796875, 46.351715087890625, 172.29888916015625, 138.44943237304688, 172.4422607421875, -171.54864501953125, -21.887603759765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000463.npy"}
{"epoch": 0.9696335078534032, "step": 464, "batch_size": 128, "mean": 56.64286804199219, "std": 97.58517456054688, "min": -214.65533447265625, "p10": -37.090686035156246, "median": 33.7830753326416, "p90": 182.5225402832031, "max": 215.6905517578125, "pos_frac": 0.7265625, "sample": [0.0, 138.9051513671875, 89.65809631347656, 161.16390991210938, -9.987991333007812, 152.14120483398438, 179.73446655273438, -32.62745666503906, 182.99551391601562, -0.06699562072753906, 54.66172790527344, 96.69415283203125, 156.95484924316406, 46.7530517578125, 153.2404022216797, 26.439552307128906, 173.513671875, 7.6092529296875, 165.3408203125, 123.00175476074219, 48.8802490234375, 167.16049194335938, 10.283319473266602, 41.80133056640625, 52.6534423828125, 170.10720825195312, -7.292320251464844, -167.03411865234375, 215.6905517578125, 15.616226196289062, -4.24615478515625, -15.34405517578125, 22.479843139648438, 203.8385772705078, 191.73248291015625, 182.94342041015625, 124.33261108398438, 8.124443054199219, -7.524383544921875, 122.31221008300781, 1.47100830078125, 25.07403564453125, 185.6805419921875, 98.08085632324219, 149.22544860839844, 4.33966064453125, -14.433837890625, 15.82720947265625, 0.2352447509765625, 46.58160400390625, 14.192977905273438, 24.83612060546875, -3.445709228515625, 150.18087768554688, 34.38986587524414, 133.35586547851562, 162.17919921875, 48.011566162109375, -176.30889892578125, 140.76141357421875, -3.020263671875, 165.329833984375, -8.291473388671875, 166.9775390625, 195.65274047851562, -9.46453857421875, 118.53438568115234, 28.845932006835938, 31.512939453125, 140.16375732421875, -130.00418090820312, -214.65533447265625, -164.41622924804688, 3.4251365661621094, -21.786041259765625, 12.89105224609375, -16.65887451171875, -49.2315673828125, 106.631103515625, 47.23102569580078, 175.40533447265625, 210.08990478515625, 167.85736083984375, 206.40097045898438, -48.087677001953125, 182.3421630859375, 1.7339859008789062, -39.1348876953125, 12.8385009765625, 3.663909912109375, 40.50201416015625, -173.23715209960938, 174.03964233398438, -13.482269287109375, -71.22637939453125, 7.274131774902344, -1.389434814453125, 152.77249145507812, -36.214599609375, 10.242889404296875, 167.8399658203125, 189.25648498535156, 49.06395721435547, 184.1046142578125, 189.687255859375, -50.159332275390625, 167.8931884765625, -2.435577392578125, 154.0576629638672, -130.2723388671875, 14.760894775390625, 9.463279724121094, -15.864601135253906, 62.949676513671875, 192.680908203125, 45.61430740356445, 144.33213806152344, 159.47698974609375, 151.73135375976562, -194.55389404296875, 19.021728515625, 32.85986328125, -35.01799774169922, 33.17628479003906, 18.686573028564453, -24.794082641601562, 41.014434814453125, 30.78021240234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000464.npy"}
{"epoch": 0.9717277486910995, "step": 465, "batch_size": 128, "mean": 72.89688873291016, "std": 102.73009490966797, "min": -196.59268188476562, "p10": -33.192492675781246, "median": 44.2574348449707, "p90": 199.69063415527344, "max": 238.31390380859375, "pos_frac": 0.734375, "sample": [-2.2453155517578125, -7.985260009765625, 13.431106567382812, 31.439376831054688, 180.2205810546875, -40.37494659423828, -31.687896728515625, 200.41778564453125, 6.17205810546875, 23.783447265625, 0.9929962158203125, 6.47869873046875, -5.2425689697265625, 10.857025146484375, -138.75054931640625, -37.766357421875, 199.45452880859375, 171.44757080078125, 215.2774658203125, 190.3343505859375, 170.97265625, -48.34381103515625, 43.38624572753906, 174.44622802734375, 37.0118408203125, -1.3019218444824219, 4.04412841796875, 179.2866668701172, 200.24154663085938, 159.62013244628906, 2.5464324951171875, 158.49620056152344, 32.35772705078125, 144.33636474609375, -5.1028594970703125, -13.078948974609375, -35.344573974609375, 194.05157470703125, 156.24310302734375, 230.21881103515625, -3.51153564453125, 184.27899169921875, -0.7779541015625, -32.270172119140625, -106.21124267578125, 32.822998046875, -8.67877197265625, 44.308570861816406, -22.9508056640625, 154.47132873535156, 147.2945098876953, 179.90640258789062, 107.09295654296875, 149.9998779296875, 42.26814270019531, -131.928466796875, 224.568359375, 44.206298828125, -10.548919677734375, 0.88214111328125, 17.19024658203125, 127.3822021484375, 47.189971923828125, 130.8470001220703, 187.318603515625, 46.30757141113281, 206.4310302734375, 227.5382080078125, -18.0810546875, 19.47270965576172, 124.9400634765625, 170.94076538085938, 162.71653747558594, 156.6054229736328, 125.23793029785156, -78.8448486328125, -7.85430908203125, 46.4952392578125, 25.853256225585938, 120.34037017822266, 68.1488037109375, 10.596649169921875, 3.227081298828125, 12.054107666015625, 6.916770935058594, 150.728759765625, 200.77627563476562, -131.55152893066406, 42.60839080810547, -17.44207763671875, 192.92877197265625, 215.96368408203125, 238.31390380859375, 21.300575256347656, 153.679931640625, 100.10113525390625, 206.68710327148438, 183.9278564453125, 209.06793212890625, 3.7540054321289062, 185.72067260742188, 180.5245819091797, 206.00152587890625, 184.05157470703125, 10.8070068359375, 159.97085571289062, -19.803680419921875, 183.9053955078125, -163.74465942382812, 37.7987060546875, -99.74972534179688, 1.357696533203125, 187.18865966796875, 188.2940673828125, 168.77587890625, 198.63172912597656, -156.1279296875, 165.18716430664062, 101.675537109375, 11.047977447509766, -19.138671875, 124.78292846679688, -17.3865966796875, -18.155357360839844, 176.00961303710938, -196.59268188476562, -10.232421875, 84.62249755859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000465.npy"}
{"epoch": 0.9738219895287958, "step": 466, "batch_size": 128, "mean": 77.7585678100586, "std": 93.32164001464844, "min": -193.48968505859375, "p10": -16.1899917602539, "median": 70.8380126953125, "p90": 184.28876495361328, "max": 236.80026245117188, "pos_frac": 0.8046875, "sample": [7.527290344238281, 138.3449249267578, 158.0446319580078, 175.26577758789062, 144.2718048095703, 183.74203491210938, 24.681121826171875, 0.0, 116.6739501953125, 137.4955291748047, 172.81626892089844, -27.328731536865234, 120.93765258789062, 122.5169677734375, 86.25909423828125, 230.10809326171875, 119.82059478759766, -0.2934684753417969, 159.08786010742188, 77.83155822753906, 143.19207763671875, 26.23845672607422, 103.63090515136719, -153.0924072265625, 25.38580322265625, 49.705474853515625, 158.96792602539062, 27.541778564453125, 149.92645263671875, 140.5486602783203, -3.0984535217285156, 173.98953247070312, 5.004600524902344, 144.90145874023438, 164.5634765625, 1.8047142028808594, 182.05075073242188, -7.6454010009765625, 170.0315399169922, 71.61785888671875, 235.63819885253906, 12.13671875, 227.92849731445312, 56.000152587890625, 181.79815673828125, 192.191162109375, 20.558265686035156, -19.409942626953125, -140.59669494628906, 7.061279296875, 14.488128662109375, 154.20315551757812, 6.062774658203125, 161.2063446044922, 189.87850952148438, 175.79669189453125, -12.6075439453125, -193.48968505859375, 25.9293212890625, 183.09182739257812, 89.74909973144531, 161.7392578125, 136.5134735107422, 70.05816650390625, 18.598724365234375, 150.6158447265625, 181.66806030273438, 146.8282928466797, 209.2923583984375, 159.29397583007812, 0.0, 164.82949829101562, 230.59442138671875, -31.817047119140625, -14.810012817382812, 24.40691375732422, 140.52093505859375, -29.765060424804688, 60.25970458984375, 236.80026245117188, 7.448234558105469, 9.668731689453125, 218.71466064453125, 114.12295532226562, 30.066970825195312, -182.739013671875, 22.993408203125, 32.802642822265625, 24.599853515625, 137.69781494140625, 170.17514038085938, 20.544052124023438, 20.97515869140625, -21.911727905273438, 219.13568115234375, 3.4359302520751953, 31.333938598632812, 166.251220703125, -89.43922424316406, -118.89031982421875, 76.77114868164062, 28.953125, -4.9606475830078125, 185.56446838378906, 24.511627197265625, 0.0, 173.58596801757812, 19.7095947265625, -0.06310081481933594, 93.87957763671875, 206.78070068359375, 124.71897888183594, -3.9668655395507812, 160.71771240234375, 113.58726501464844, -9.823318481445312, 25.77203369140625, 30.841293334960938, 21.390995025634766, 2.546875, 26.213836669921875, 11.979606628417969, 207.7467041015625, 168.40200805664062, -23.4583740234375, -32.82906723022461, 171.44003295898438, 29.787353515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000466.npy"}
{"epoch": 0.9759162303664921, "step": 467, "batch_size": 128, "mean": 57.694068908691406, "std": 100.4228744506836, "min": -176.30426025390625, "p10": -67.37840042114257, "median": 26.322181701660156, "p90": 189.60839233398437, "max": 224.99246215820312, "pos_frac": 0.6640625, "sample": [180.2042236328125, 170.98001098632812, 84.79515075683594, 127.70025634765625, 39.484291076660156, -12.900054931640625, 108.64447021484375, 91.66155242919922, 111.17529296875, -3.87896728515625, -150.92727661132812, -176.30426025390625, -10.8907470703125, -10.6859130859375, 19.58984375, 160.41366577148438, -114.77859497070312, 214.09149169921875, 4.6651611328125, 104.93487548828125, -21.511688232421875, 148.434814453125, -18.261810302734375, 132.8240203857422, 180.90216064453125, 82.24078369140625, 26.223114013671875, 151.92527770996094, -1.1331100463867188, 14.873291015625, 24.6705322265625, 168.52586364746094, 75.47540283203125, 166.34002685546875, 193.92684936523438, 99.95779418945312, 26.421249389648438, 195.68707275390625, 99.91239929199219, -40.38787841796875, 200.2386474609375, 1.694549560546875, 0.20726585388183594, 193.24896240234375, 9.794342041015625, 161.52874755859375, -15.684459686279297, -41.906280517578125, 189.03375244140625, -145.22018432617188, 12.528152465820312, -79.65493774414062, -68.55684661865234, 170.61019897460938, -55.081512451171875, 152.9561767578125, -66.87335205078125, -46.089508056640625, 169.4866943359375, 216.0341796875, 19.1473388671875, 25.634592056274414, -1.2031745910644531, 9.834136962890625, 149.68215942382812, 17.79266357421875, 0.8308792114257812, 99.41683197021484, -2.550811767578125, -80.08953857421875, 5.420379638671875, -158.20887756347656, 180.41275024414062, 0.0, -1.0114898681640625, -4.9971160888671875, 94.4991455078125, 201.8234100341797, 5.56988525390625, -1.5249099731445312, 0.0, 49.18363952636719, 174.3634033203125, -10.656646728515625, 130.23345947265625, 112.05168151855469, 153.9786376953125, -38.21888732910156, 219.0311279296875, -125.09844970703125, -102.99214172363281, 206.51593017578125, 181.75579833984375, 159.14781188964844, 188.05892944335938, -2.2755603790283203, -21.206069946289062, 22.34686279296875, -9.071197509765625, 133.6739044189453, -35.56512451171875, 179.37469482421875, 172.28195190429688, 41.73614501953125, -14.738052368164062, 27.97882080078125, -17.094253540039062, -77.39605712890625, 14.776022911071777, 93.27923583984375, 190.94921875, 169.18125915527344, -128.59230041503906, 112.81834411621094, 8.243328094482422, 15.13824462890625, -8.768203735351562, 137.084716796875, 224.99246215820312, 17.285198211669922, 176.09127807617188, 28.726409912109375, 38.594573974609375, 177.9379425048828, 199.17218017578125, 203.27377319335938, -121.22500610351562, -25.31219482421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000467.npy"}
{"epoch": 0.9780104712041885, "step": 468, "batch_size": 128, "mean": 49.85354232788086, "std": 92.85494232177734, "min": -152.54354858398438, "p10": -51.63597412109374, "median": 24.487869262695312, "p90": 177.77929077148437, "max": 258.86395263671875, "pos_frac": 0.6953125, "sample": [36.734771728515625, 205.1217041015625, 156.5936737060547, -99.12086486816406, 59.44496154785156, 22.542236328125, -14.072341918945312, 177.01858520507812, 26.550260543823242, 33.77454376220703, 32.159423828125, -45.12030029296875, 21.4066162109375, 200.328125, -84.5784912109375, -15.318862915039062, -0.884857177734375, 182.44241333007812, -123.66590881347656, -23.395233154296875, 73.24179077148438, 47.33301544189453, -131.42337036132812, 135.6820068359375, -19.987152099609375, 229.3602294921875, 55.00360107421875, 136.140869140625, -14.919525146484375, -25.323345184326172, 10.317138671875, 25.3948974609375, 195.19277954101562, -32.146034240722656, 35.010467529296875, -8.275341033935547, -20.307388305664062, 174.17709350585938, -120.51416015625, 22.61114501953125, 7.840263366699219, 29.569976806640625, -59.543792724609375, 167.9908447265625, -2.7383651733398438, 3.3135986328125, 75.0394287109375, -125.416748046875, 38.121002197265625, 5.47210693359375, -12.440750122070312, 4.401193618774414, 87.90533447265625, -6.103515625e-05, 16.8162841796875, 24.5535888671875, 11.652069091796875, 177.16647338867188, 150.841552734375, 165.45252990722656, 179.20919799804688, -21.618438720703125, -152.54354858398438, 12.547073364257812, 258.86395263671875, 8.567230224609375, 15.099861145019531, 197.61566162109375, 21.32073974609375, 108.84991455078125, 55.801605224609375, 140.63766479492188, -28.384307861328125, 77.9478759765625, 159.97445678710938, 169.6102294921875, -41.570526123046875, 18.234375, 9.463600158691406, 139.10020446777344, -64.67166137695312, 210.581787109375, 26.706012725830078, 185.25546264648438, 42.9814453125, 141.31597900390625, 148.82659912109375, -21.608081817626953, 150.1417236328125, 125.83356475830078, -7.637245178222656, 5.9938507080078125, 2.840576171875, 207.34698486328125, 2.71075439453125, 34.79310607910156, 136.25631713867188, 3.082855224609375, 66.30206298828125, 4.13525390625, 163.9827423095703, 174.14614868164062, 127.38423156738281, 6.7349853515625, -46.67364501953125, -30.937103271484375, -46.90093994140625, 69.46917724609375, 56.74151611328125, 104.40103149414062, -8.138107299804688, -5.08709716796875, 152.74267578125, -121.02180480957031, -58.774871826171875, 15.65673828125, 161.6472625732422, 5.517051696777344, -4.2312164306640625, -130.64013671875, 161.15257263183594, 184.5804443359375, 24.422149658203125, 183.88824462890625, -48.576446533203125, 175.91265869140625, -59.7196044921875, 163.13479614257812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000468.npy"}
{"epoch": 0.9801047120418848, "step": 469, "batch_size": 128, "mean": 66.12458801269531, "std": 104.24964141845703, "min": -232.54708862304688, "p10": -32.75726623535156, "median": 37.86549377441406, "p90": 189.34980316162108, "max": 243.29248046875, "pos_frac": 0.78125, "sample": [21.349334716796875, 183.60147094726562, 0.0, 15.031089782714844, 31.63812255859375, 168.6805877685547, -105.97161865234375, 21.73577880859375, 13.651519775390625, 92.12294006347656, 141.7474365234375, 33.873779296875, 72.5916748046875, -148.4378662109375, 19.420562744140625, 5.01324462890625, 210.0977783203125, 168.6899871826172, 21.9354248046875, 147.91555786132812, -3.0972442626953125, 80.0050048828125, 6.455974578857422, -2.9307708740234375, -174.65911865234375, 180.6898193359375, 22.2301025390625, 172.9212646484375, 188.15789794921875, 25.483917236328125, 154.01544189453125, 59.9378662109375, 231.3206787109375, 186.86636352539062, 2.3468170166015625, 121.97978210449219, 50.997802734375, 82.79730224609375, 182.00328063964844, 3.20672607421875, -2.1237106323242188, -25.599853515625, 31.966278076171875, 0.04913330078125, 154.0287628173828, -0.391693115234375, 27.745010375976562, 16.4976806640625, 49.690399169921875, 37.44854736328125, 151.2318878173828, 188.5712890625, 176.00244140625, 195.29052734375, -167.07241821289062, 148.80606079101562, 171.3775634765625, 212.33633422851562, 179.07810974121094, -148.24163818359375, 5.468555450439453, -10.5782470703125, 203.59808349609375, -179.4678955078125, -113.40861511230469, 70.6004638671875, 26.887420654296875, 1.95257568359375, -33.755462646484375, 187.70169067382812, 19.156448364257812, 152.56614685058594, 80.10653686523438, 174.97496032714844, -232.54708862304688, 13.532541275024414, 37.9656982421875, 163.80511474609375, -7.2271881103515625, 243.29248046875, 178.7219696044922, 157.84494018554688, 37.765289306640625, 1.10205078125, 181.86080932617188, -16.322296142578125, -220.11602783203125, 5.740360260009766, 86.67646789550781, 23.15277862548828, 72.52099609375, 203.84713745117188, -5.957000732421875, 182.1064453125, -38.92724609375, -30.166046142578125, 0.4815406799316406, 1.0164260864257812, -14.592700958251953, 191.1663360595703, -71.14991760253906, -2.2625961303710938, -32.3294677734375, 226.98178100585938, 4.608869552612305, 7.5047607421875, 187.92242431640625, 132.53634643554688, 188.31695556640625, 200.27386474609375, 168.55413818359375, 8.190963745117188, 182.28904724121094, 50.2896728515625, 2.8433609008789062, 61.409423828125, 24.77691650390625, 167.63479614257812, 145.05889892578125, 29.78498077392578, 195.11813354492188, 194.19708251953125, 157.33383178710938, -15.663482666015625, 150.33779907226562, 38.14947509765625, 217.01641845703125, -40.429840087890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000469.npy"}
{"epoch": 0.9821989528795811, "step": 470, "batch_size": 128, "mean": 73.5052490234375, "std": 104.4378433227539, "min": -193.22747802734375, "p10": -65.44570388793944, "median": 48.14399719238281, "p90": 201.1071350097656, "max": 248.1992645263672, "pos_frac": 0.796875, "sample": [-17.46173095703125, 16.218170166015625, 188.84793090820312, 45.19329833984375, 2.0363616943359375, 150.44998168945312, 7.1269073486328125, 166.59744262695312, 181.1973876953125, 27.39129638671875, -32.782958984375, 4.6226654052734375, 1.2947463989257812, 181.28326416015625, 137.2486572265625, 158.690185546875, 187.15838623046875, 1.537017822265625, -13.7391357421875, 50.4154052734375, 138.04898071289062, -142.1505126953125, -116.37911987304688, -145.55618286132812, 223.7974853515625, 157.96676635742188, 139.9512939453125, 15.67828369140625, 174.87982177734375, 188.35385131835938, 236.21270751953125, 146.61724853515625, -64.8249740600586, -1.4454345703125, 30.48895263671875, -66.89407348632812, 214.94403076171875, 158.78646850585938, 27.30157470703125, -124.02922058105469, 12.10125732421875, 221.2991943359375, 145.42510986328125, 18.56414794921875, 4.527801513671875, 169.017578125, 204.78945922851562, 15.02044677734375, 28.413944244384766, 141.6646728515625, 12.633270263671875, 176.47418212890625, 27.7584228515625, 32.939186096191406, -17.4683837890625, 168.2642822265625, 162.3334503173828, -0.8919448852539062, 161.57284545898438, 18.78338623046875, 70.06107330322266, 4.221050262451172, 170.4961395263672, 86.76329040527344, 20.918197631835938, 25.038925170898438, 9.182273864746094, -193.22747802734375, 20.040252685546875, 183.62985229492188, 43.19728088378906, 102.61397552490234, 153.02597045898438, -30.31060791015625, -135.1741943359375, -81.6816177368164, -94.0087890625, 146.1236572265625, 187.9877471923828, 164.52359008789062, 36.473541259765625, 42.99407958984375, 153.55624389648438, 219.14013671875, 136.8866729736328, -21.63823699951172, 7.632843017578125, 205.22625732421875, 200.02011108398438, 203.64352416992188, -26.239242553710938, 167.64791870117188, 148.8504638671875, 162.30117797851562, -1.1952667236328125, 107.40475463867188, 58.291526794433594, 143.03639221191406, 44.838226318359375, 25.30206298828125, 185.77587890625, 194.73419189453125, 138.58944702148438, 248.1992645263672, 17.501220703125, 192.1114501953125, 0.1823272705078125, 6.345794677734375, -12.068145751953125, -165.63308715820312, 15.776214599609375, -86.39390563964844, 211.37127685546875, -107.75204467773438, 138.81402587890625, 221.35528564453125, 6.3079833984375, 45.872589111328125, 73.2335205078125, 168.98370361328125, 206.8367919921875, -130.05154418945312, 52.3775634765625, 27.989288330078125, 212.8563232421875, 180.13137817382812, 184.57672119140625, -25.240386962890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000470.npy"}
{"epoch": 0.9842931937172775, "step": 471, "batch_size": 128, "mean": 55.88236999511719, "std": 98.52395629882812, "min": -173.18357849121094, "p10": -31.12346534729004, "median": 31.344158172607422, "p90": 188.40685272216797, "max": 259.795654296875, "pos_frac": 0.6796875, "sample": [-100.43428039550781, 171.0204620361328, 31.595855712890625, 103.72116088867188, 162.21780395507812, 222.42657470703125, 170.56443786621094, -134.7987060546875, -12.788904190063477, 166.5955352783203, -56.45170593261719, -21.84515380859375, -3.024627685546875, 138.82281494140625, -2.8362655639648438, 72.2686767578125, 20.80364990234375, 68.11868286132812, -23.85009002685547, -110.2296142578125, 30.573486328125, 28.29156494140625, 19.38323974609375, 51.84423828125, -27.559600830078125, -12.312828063964844, 205.311279296875, 0.468292236328125, -27.22021484375, 36.21339416503906, 216.04232788085938, 43.515228271484375, 60.96929931640625, 13.065826416015625, -4.451995849609375, 101.66758728027344, 173.15220642089844, 132.52645874023438, 31.09246063232422, 14.519561767578125, 28.155258178710938, -6.18743896484375, 174.98135375976562, 194.0821533203125, -31.04912567138672, 136.02272033691406, 28.079605102539062, 216.38194274902344, 160.43972778320312, -7.029510498046875, 5.3307952880859375, 8.168701171875, 179.14190673828125, 88.70758056640625, -147.7078857421875, 153.3880615234375, 139.07005310058594, 3.692392349243164, -2.1798248291015625, 30.86713409423828, 226.15573120117188, 259.795654296875, 13.959405899047852, -1.65374755859375, -31.296924591064453, -3.8604793548583984, -6.401641845703125, 6.4506072998046875, -173.18357849121094, 16.69671630859375, 151.0181884765625, 111.08653259277344, 161.64527893066406, 39.96820068359375, -146.84556579589844, 34.45782470703125, 36.42333984375, 139.18331909179688, 75.58534240722656, -30.1019287109375, 126.52944946289062, -18.517364501953125, -2.20379638671875, -21.497283935546875, 54.5380859375, 149.2061767578125, 10.332103729248047, -7.259712219238281, 170.82107543945312, -3.232025146484375, 163.66439819335938, 167.12295532226562, -12.028308868408203, 57.48472595214844, 6.9608001708984375, 188.09742736816406, 186.480224609375, -73.76020812988281, -7.7252197265625, -28.850738525390625, 32.7621955871582, 19.94196319580078, 4.4019775390625, 105.48699951171875, 158.9203643798828, 188.35523986816406, 50.49053955078125, 243.60189819335938, 124.96171569824219, 8.0545654296875, 157.66970825195312, 188.52728271484375, 198.9324951171875, -147.2088623046875, 197.16973876953125, 57.71705627441406, 111.34854125976562, 0.0, 46.045074462890625, -162.15927124023438, 197.22613525390625, 164.99935913085938, 194.3704833984375, -19.45623779296875, 19.73150634765625, -14.06195068359375, -106.64443969726562, -154.82955932617188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000471.npy"}
{"epoch": 0.9863874345549738, "step": 472, "batch_size": 128, "mean": 70.65408325195312, "std": 100.1778564453125, "min": -176.90567016601562, "p10": -36.253720092773435, "median": 47.60946273803711, "p90": 190.65385894775392, "max": 357.9688720703125, "pos_frac": 0.7421875, "sample": [357.9688720703125, -5.513671875, 5.465780258178711, -37.75303649902344, 5.296228408813477, -1.61181640625, -14.59835433959961, 157.94058227539062, 0.7943115234375, 28.117416381835938, 129.73805236816406, 104.86460876464844, 190.02984619140625, 32.51477813720703, -50.23206329345703, 162.27554321289062, -22.32585906982422, 9.201347351074219, -36.74395751953125, 79.67988586425781, 6.69439697265625, 33.589874267578125, 141.981201171875, 48.473785400390625, -12.710884094238281, 201.95123291015625, 149.96502685546875, -147.58721923828125, 3.3078346252441406, 163.39370727539062, 44.37384033203125, -107.07626342773438, 20.324859619140625, 220.23419189453125, 48.02995300292969, -29.308547973632812, 178.9890594482422, 178.8933868408203, -36.1014404296875, 5.2133026123046875, 26.111419677734375, 148.72596740722656, 132.06988525390625, 137.96939086914062, -29.415084838867188, 20.588348388671875, 13.383209228515625, -3.7395477294921875, 12.4364013671875, 17.322357177734375, 150.18386840820312, 58.95878601074219, 188.06881713867188, -93.18194580078125, 164.80291748046875, 161.6876220703125, 179.25430297851562, 166.75607299804688, 139.13217163085938, 30.306503295898438, -19.220016479492188, 100.81951904296875, 2.502216339111328, -1.79803466796875, 158.60137939453125, 190.7303466796875, 158.7677001953125, 70.9154052734375, 136.4672088623047, -0.790313720703125, 46.42201614379883, 47.18897247314453, 23.153045654296875, 179.4883575439453, 161.6248016357422, 219.33853149414062, 186.59848022460938, -9.63760757446289, 180.2097930908203, 25.727394104003906, -21.42901611328125, 247.07049560546875, 175.21897888183594, 222.84469604492188, 29.6966552734375, 218.57843017578125, 214.28060913085938, 184.57159423828125, 159.35079956054688, -176.90567016601562, -22.32855224609375, -125.44525146484375, 142.8687744140625, 97.27069091796875, 3.57574462890625, 145.7193603515625, 103.79959106445312, 9.107681274414062, -36.609039306640625, 121.93585205078125, 179.88681030273438, -13.101226806640625, 191.10430908203125, 143.75363159179688, 31.22906494140625, 95.54208374023438, -2.772705078125, 13.501007080078125, -3.343883514404297, 0.84039306640625, 203.56619262695312, 190.62107849121094, 4.520441055297852, 19.662811279296875, 48.233245849609375, -155.27818298339844, -107.55763244628906, 251.91748046875, -11.07073974609375, 54.88671875, 198.72958374023438, 124.74005126953125, -131.62905883789062, 107.2589111328125, 142.99407958984375, 150.71963500976562, -49.10224151611328, -15.572437286376953], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000472.npy"}
{"epoch": 0.9884816753926702, "step": 473, "batch_size": 128, "mean": 69.88441467285156, "std": 100.83721923828125, "min": -195.1175079345703, "p10": -29.132363891601546, "median": 53.528221130371094, "p90": 197.21399536132813, "max": 232.8345947265625, "pos_frac": 0.765625, "sample": [12.106964111328125, 198.46243286132812, 52.39698791503906, -3.3899173736572266, 157.83279418945312, 5.21435546875, 34.70013427734375, 17.89739990234375, 69.5389404296875, -17.625030517578125, 119.10360717773438, 29.07781982421875, -66.21633911132812, 34.855133056640625, -130.13140869140625, 57.1213264465332, 27.6641845703125, 7.256866455078125, 149.24832153320312, 170.8984375, 133.92987060546875, 157.71812438964844, 208.3135986328125, 232.8345947265625, 76.63433837890625, -133.77032470703125, -17.001487731933594, 139.40663146972656, 141.44715881347656, 20.319564819335938, 197.12176513671875, 186.36749267578125, 175.84561157226562, 38.1318359375, -16.82347869873047, -39.66485595703125, 4.739349365234375, -123.94750213623047, 220.70248413085938, -10.693336486816406, 140.72283935546875, 162.18377685546875, 39.8428955078125, -19.00735855102539, -13.850479125976562, -22.76177978515625, -42.4593505859375, 202.2005615234375, 54.659454345703125, 25.14031982421875, 18.832138061523438, -1.4657135009765625, -0.2416229248046875, 203.15545654296875, -18.06781005859375, 193.85650634765625, -24.618438720703125, 167.2185516357422, 213.9896240234375, -183.01504516601562, 62.865020751953125, 199.17291259765625, 202.15206909179688, 4.9742431640625, 16.350631713867188, -113.28271484375, 228.455078125, 185.96603393554688, 93.42721557617188, 204.83502197265625, 88.2757568359375, 152.37051391601562, 130.7786865234375, 20.639297485351562, 3.995370864868164, 163.5513916015625, 157.60159301757812, 18.37488555908203, 16.575225830078125, 195.487060546875, -3.3418426513671875, -46.397499084472656, 183.23388671875, 152.2467041015625, 161.44790649414062, 193.19451904296875, -195.1175079345703, -135.29676818847656, 180.3653564453125, -20.038558959960938, 105.23138427734375, 150.66244506835938, 3.124828338623047, 50.18858337402344, 0.35675048828125, -126.33808898925781, 20.082019805908203, 16.468017578125, 55.48548889160156, 2.7804718017578125, 197.42919921875, 157.60281372070312, 150.3856658935547, 0.0, 165.772216796875, 128.260009765625, -175.22525024414062, 153.79595947265625, 32.009239196777344, 151.04791259765625, 159.99588012695312, 18.000320434570312, 1.2420310974121094, 188.17880249023438, 18.249542236328125, -4.164031982421875, 3.811126708984375, -16.137191772460938, 29.9169921875, 45.21018981933594, 168.67034912109375, 81.55343627929688, 173.81170654296875, 67.63643646240234, 221.6617431640625, 144.1317901611328, 191.3634490966797, 96.15194702148438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000473.npy"}
{"epoch": 0.9905759162303664, "step": 474, "batch_size": 128, "mean": 48.95247268676758, "std": 97.7584457397461, "min": -198.89984130859375, "p10": -72.55186767578125, "median": 29.47928810119629, "p90": 176.6666290283203, "max": 319.0506591796875, "pos_frac": 0.734375, "sample": [9.020065307617188, 1.1026611328125, 319.0506591796875, -2.850067138671875, -16.417829513549805, -27.24981689453125, 169.2007598876953, 205.1209716796875, 65.95591735839844, 169.6591339111328, 7.0863494873046875, 25.93115234375, -6.11962890625, -157.60723876953125, 7.701313018798828, -153.18907165527344, 4.9767608642578125, -7.483036041259766, 108.8883285522461, 38.774635314941406, 126.64827728271484, 190.75588989257812, 5.19366455078125, -74.81695556640625, 172.52731323242188, -41.11600875854492, -101.1748046875, 28.163330078125, -18.63970184326172, 68.6064453125, 152.9873046875, 53.06781005859375, -3.5377044677734375, 150.39053344726562, 51.972198486328125, 149.90399169921875, 240.14215087890625, 22.976085662841797, 0.46065521240234375, 193.86343383789062, 151.00164794921875, 94.3668212890625, 194.90899658203125, 186.37669372558594, 68.27651977539062, -6.070301055908203, 0.0, 26.616104125976562, 27.07525634765625, 10.286598205566406, -2.086639404296875, -10.75506591796875, -130.0430450439453, 16.234329223632812, 41.68607711791992, 132.278076171875, 10.188087463378906, -98.72246551513672, 177.91006469726562, 153.93228149414062, 5.9344024658203125, -6.86444091796875, -7.05621337890625, 169.50247192382812, 155.68853759765625, 24.118560791015625, 166.39053344726562, 67.48947143554688, -38.8104248046875, 12.234603881835938, 85.89566040039062, 29.321186065673828, -151.17279052734375, 10.10477066040039, 38.288116455078125, -29.00299072265625, 199.53390502929688, 139.7654266357422, 122.66030883789062, -6.19464111328125, -71.58111572265625, 22.598358154296875, -0.1609039306640625, 5.742513656616211, -0.5351753234863281, -198.89984130859375, -125.36128997802734, 157.40370178222656, -171.52052307128906, 157.7830047607422, 86.27520751953125, -193.65347290039062, 3.81683349609375, 2.04693603515625, 153.04611206054688, 8.839683532714844, 44.15924072265625, -16.75946044921875, 34.758636474609375, 145.391357421875, 82.15704345703125, 176.13372802734375, 181.87408447265625, 87.82415771484375, 220.48788452148438, 111.52171325683594, 162.92486572265625, 12.747833251953125, 55.702728271484375, 24.868362426757812, 35.1861572265625, 32.78099060058594, 48.692474365234375, 61.626800537109375, 38.609619140625, 125.01937866210938, 164.44859313964844, 186.96701049804688, -85.42933654785156, 33.162506103515625, 22.28411865234375, 182.17320251464844, 35.18072509765625, -150.69863891601562, 28.5927734375, 20.849761962890625, 29.63739013671875, 135.98812866210938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000474.npy"}
{"epoch": 0.9926701570680628, "step": 475, "batch_size": 128, "mean": 55.152618408203125, "std": 97.38682556152344, "min": -207.91604614257812, "p10": -54.8746856689453, "median": 33.51170349121094, "p90": 174.78130493164062, "max": 261.9085388183594, "pos_frac": 0.7109375, "sample": [204.76211547851562, 171.533935546875, -126.45905303955078, 124.94007110595703, 154.04891967773438, 41.947784423828125, -207.91604614257812, 221.83511352539062, -120.20865631103516, -50.237457275390625, 1.4054107666015625, -147.2252197265625, -46.196258544921875, -43.238929748535156, 8.171600341796875, 152.09408569335938, 153.0039825439453, 164.71620178222656, 174.47134399414062, -8.867607116699219, 202.0333251953125, -15.544586181640625, 16.678329467773438, 107.8659896850586, 184.21568298339844, 167.9364471435547, 156.58154296875, -10.390228271484375, 236.04205322265625, -1.878671646118164, 118.80490112304688, 124.74818420410156, -12.302459716796875, 170.7162628173828, 34.143585205078125, 20.69086456298828, -117.9596176147461, 29.788284301757812, 0.4891490936279297, 172.66940307617188, 148.50537109375, 38.45734405517578, -146.762939453125, 91.92897033691406, 95.89595031738281, -17.90533447265625, 150.69845581054688, -9.26531982421875, 164.60781860351562, -27.155746459960938, 15.661479949951172, -103.52820587158203, 176.11846923828125, 167.81167602539062, -131.60302734375, 166.35675048828125, 194.45236206054688, -2.027956008911133, 20.37786865234375, 116.2379150390625, 167.67713928222656, 32.87982177734375, 8.215995788574219, -21.77886962890625, 27.61676025390625, 128.99180603027344, 20.245285034179688, 46.34857177734375, 8.445220947265625, 27.7880859375, 14.83258056640625, 180.40679931640625, 128.3062286376953, -16.21112060546875, 148.5096435546875, -41.224517822265625, 0.863861083984375, 261.9085388183594, 38.883514404296875, -3.873504638671875, 18.565383911132812, 208.78070068359375, 170.31109619140625, -28.855560302734375, 162.5673828125, -77.603515625, 54.5312614440918, 12.60784912109375, 56.28515625, 60.38872528076172, -1.004302978515625, 40.91691207885742, 123.90536499023438, 145.00946044921875, -2.628753662109375, 167.0196533203125, 153.57315063476562, 175.50454711914062, 23.765640258789062, 94.45101928710938, 11.864128112792969, -124.80145263671875, 34.2108154296875, 0.2202911376953125, -7.33563232421875, -32.644195556640625, -65.69488525390625, 11.65655517578125, 163.00778198242188, 67.35592651367188, 91.04176330566406, 56.380584716796875, -3.3230133056640625, 12.340484619140625, 166.5171661376953, -29.41119384765625, 50.43199157714844, -78.26602172851562, -90.875244140625, 21.22930908203125, 203.50119018554688, 35.98675537109375, 20.316043853759766, 210.03164672851562, 118.09661865234375, 13.17669677734375, -41.83162307739258, 12.6575927734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000475.npy"}
{"epoch": 0.9947643979057592, "step": 476, "batch_size": 128, "mean": 65.86012268066406, "std": 99.12320709228516, "min": -192.44903564453125, "p10": -39.833348083496084, "median": 33.18645477294922, "p90": 190.03265686035155, "max": 253.32879638671875, "pos_frac": 0.7109375, "sample": [-52.587677001953125, -135.95297241210938, -3.045135498046875, -37.981964111328125, 18.957672119140625, 161.22543334960938, -1.4955902099609375, -9.089454650878906, 18.05487060546875, 112.96162414550781, 147.0830535888672, 120.99163818359375, 182.2877960205078, 210.3472900390625, 146.01412963867188, 220.8560791015625, -32.0863037109375, 193.2869415283203, 156.6700439453125, -4.313941955566406, -8.627086639404297, 37.7196044921875, 19.4014892578125, 25.173065185546875, 145.41159057617188, -17.45708465576172, 177.05300903320312, -1.4437179565429688, -190.43875122070312, 199.74795532226562, 14.87200927734375, 22.567596435546875, -44.15324401855469, 200.35882568359375, 178.30029296875, 16.091156005859375, 189.55130004882812, -69.93416595458984, 30.30657958984375, -19.641693115234375, -3.656494140625, -192.44903564453125, 141.2435302734375, 21.731674194335938, 173.45504760742188, 30.398590087890625, 146.77133178710938, 199.7210693359375, 25.33989715576172, -69.31314086914062, 127.02125549316406, 181.053955078125, 11.4716796875, 10.172309875488281, -2.191143035888672, 211.0938720703125, 13.580322265625, -19.17193603515625, 183.5920867919922, 28.580352783203125, 165.8992919921875, 109.53279876708984, -102.335205078125, 184.5291748046875, 20.35333251953125, 201.52044677734375, 192.22833251953125, 10.98486328125, 150.2100830078125, 63.644683837890625, 30.344482421875, -52.83026123046875, 235.0536346435547, 39.514404296875, 116.91744232177734, -1.7089309692382812, 154.06161499023438, -147.92056274414062, 169.18487548828125, 45.0867919921875, 160.63282775878906, 151.97402954101562, 154.90162658691406, 2.83392333984375, 9.348060607910156, 2.9451751708984375, -3.4652175903320312, 148.75808715820312, -181.26475524902344, 138.2884521484375, -12.680557250976562, 131.74169921875, 191.15582275390625, 3.6861343383789062, 154.7117462158203, -22.049148559570312, -16.082366943359375, 57.016151428222656, 2.45245361328125, 157.85882568359375, -22.9949951171875, 33.82830810546875, -60.6771240234375, 253.32879638671875, 19.427391052246094, 21.232803344726562, -12.143157958984375, 32.54460144042969, 118.17906951904297, 109.69354248046875, 83.45904541015625, 165.98898315429688, 177.8193359375, 96.41200256347656, 5.63824462890625, -20.382781982421875, 129.460693359375, -26.389678955078125, -4.0054779052734375, 136.62210083007812, 175.76968383789062, 145.42047119140625, -89.66815185546875, 186.0146026611328, -31.037109375, 193.1275634765625, 156.75286865234375, 104.15216064453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000476.npy"}
{"epoch": 0.9968586387434555, "step": 477, "batch_size": 128, "mean": 68.07359313964844, "std": 105.90400695800781, "min": -203.2001953125, "p10": -49.65098571777343, "median": 49.226043701171875, "p90": 193.16015319824217, "max": 247.04180908203125, "pos_frac": 0.6875, "sample": [-148.95840454101562, 11.05377197265625, 170.62644958496094, -133.0496826171875, 191.30462646484375, 37.13740539550781, -146.4930419921875, 158.779052734375, -27.879501342773438, 119.59536743164062, -8.454416275024414, 191.72998046875, -1.9236602783203125, 187.68682861328125, 60.36273193359375, -14.211212158203125, 223.3507080078125, 103.48330688476562, -203.2001953125, -26.19012451171875, -5.973876953125, -32.899200439453125, 27.517959594726562, 186.63473510742188, 207.70079040527344, 199.65536499023438, 159.3347625732422, -11.890876770019531, 15.983062744140625, 27.67095947265625, 159.93247985839844, 187.55661010742188, 28.61785888671875, 228.78753662109375, 153.15951538085938, 96.9808349609375, 240.5946044921875, 40.679481506347656, -10.77008056640625, 0.0, 17.673416137695312, 185.8904571533203, 4.0992279052734375, -30.453353881835938, 89.46024322509766, -15.50726318359375, 9.373291015625, 156.3671417236328, 171.86346435546875, -101.60308837890625, -7.240386962890625, 17.7049560546875, -28.438186645507812, 181.53469848632812, 31.392822265625, 15.703155517578125, 163.41351318359375, 30.094100952148438, 165.049560546875, 44.81121826171875, 151.9383544921875, 177.80953979492188, 246.452880859375, -16.26971435546875, 131.22406005859375, 135.5029296875, 203.69638061523438, -126.9989013671875, 93.15093994140625, -35.8004150390625, 140.15087890625, 178.22799682617188, 129.05003356933594, 162.06362915039062, 93.82464599609375, 198.94732666015625, -44.66234588623047, -0.9618682861328125, 1.0843276977539062, 167.12936401367188, -27.027313232421875, 173.865966796875, 141.9896240234375, -18.884544372558594, -53.591583251953125, -168.70428466796875, 172.58560180664062, 115.38252258300781, 55.70554733276367, 143.29486083984375, -2.7106504440307617, 14.50042724609375, -13.288997650146484, 211.47500610351562, -4.966339111328125, 164.3807373046875, -115.35415649414062, 150.27664184570312, 33.589080810546875, 40.49102020263672, -62.090118408203125, 247.04180908203125, -93.21324157714844, -176.89724731445312, 17.99957275390625, -22.03912353515625, -47.962158203125, -2.8165283203125, 210.48056030273438, 109.79460144042969, 44.19061279296875, 234.41998291015625, 188.6546630859375, 196.49722290039062, -0.6097259521484375, 173.60885620117188, 183.81216430664062, -77.26454162597656, 181.79983520507812, 13.579925537109375, 53.640869140625, 149.8411865234375, 154.8647918701172, 107.1640853881836, 4.8213348388671875, 115.67251586914062, 57.9422607421875, 6.7042999267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.43-s_star-0.3-20260430-192039/margin_logs/step_0000477.npy"}