Files

478 lines
1.3 MiB
Plaintext
Raw Permalink Normal View History

{"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-0.45-s_star-0.6-20260430-165125/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-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000002.npy"}
{"epoch": 0.004188481675392671, "step": 3, "batch_size": 128, "mean": 0.0006298869848251343, "std": 0.9153962135314941, "min": -2.280914306640625, "p10": -1.1069564819335938, "median": 0.0028839111328125, "p90": 1.1048004150390625, "max": 3.289520263671875, "pos_frac": 0.5, "sample": [-0.0670166015625, 0.19081497192382812, -0.1837158203125, -0.2694511413574219, 0.2940826416015625, -2.280914306640625, -0.593170166015625, 0.02362060546875, 0.0, -0.829620361328125, 2.00238037109375, -1.1944580078125, -0.584320068359375, 1.089447021484375, -0.28082275390625, 0.208099365234375, -1.56219482421875, 1.9185791015625, 2.0340728759765625, -0.30767822265625, 0.17671966552734375, -0.0079345703125, -0.51318359375, -0.23780059814453125, 0.453338623046875, -0.10460662841796875, 1.140625, -1.05902099609375, 0.79443359375, -1.500640869140625, -0.4191322326660156, 0.08338546752929688, 1.200286865234375, -0.3837890625, 2.46734619140625, 0.18043136596679688, -0.11798095703125, -0.99896240234375, -0.449554443359375, 0.939910888671875, -0.5633392333984375, -0.561859130859375, 0.243408203125, 0.5855255126953125, -0.746337890625, -1.084686279296875, 0.344512939453125, 0.647491455078125, 1.149993896484375, -1.986602783203125, -1.27239990234375, -0.2062530517578125, 0.480316162109375, -0.1231231689453125, 0.099151611328125, -0.158203125, 0.44261932373046875, 1.33233642578125, 0.2963371276855469, 0.2859230041503906, 0.0, 0.229400634765625, 3.289520263671875, -0.916290283203125, -0.6443252563476562, -1.033416748046875, -1.386871337890625, -0.329559326171875, 0.09748458862304688, 0.1485748291015625, -0.67138671875, 0.042083740234375, 2.538665771484375, -1.3668212890625, -0.24415206909179688, 0.14794921875, 0.5763397216796875, -0.035919189453125, -0.675567626953125, 0.75433349609375, 0.0563201904296875, 1.0077972412109375, 0.9709091186523438, 0.24630355834960938, 0.72808837890625, 1.260040283203125, -0.965087890625, 0.2972412109375, 0.420562744140625, -0.01641845703125, 0.1820831298828125, 0.139923095703125, -0.722503662109375, -1.9718017578125, -0.10400772094726562, 0.8574066162109375, 0.12740325927734375, 0.41967010498046875, 0.4613685607910156, 0.1543865203857422, -0.6728057861328125, 0.2643280029296875, -0.178009033203125, -0.314788818359375, 0.7505340576171875, 0.727447509765625, -0.6574249267578125, 0.2945556640625, 0.005767822265625, 0.008453369140625, -1.07769775390625, -1.245086669921875, 1.5996856689453125, -0.989227294921875, -0.0985565185546875, 0.9612045288085938, -1.1589202880859375, -0.931640625, -0.420867919921875, -0.28216552734375, 1.460662841796875, -0.4050445556640625, 0.0230560302734375, -0.3199310302734375, -1.3358993530273438, -1.364898681640625, 0.43817901611328125, 0.47357177734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000003.npy"}
{"epoch": 0.0062827225130890054, "step": 4, "batch_size": 128, "mean": 0.014546826481819153, "std": 0.9029322862625122, "min": -2.64874267578125, "p10": -1.0655792236328125, "median": 0.10058975219726562, "p90": 1.0695465087890625, "max": 2.6322021484375, "pos_frac": 0.5546875, "sample": [-0.507049560546875, -0.061981201171875, 0.2504234313964844, 1.15289306640625, -1.142730712890625, -0.5297622680664062, 0.4997673034667969, 0.75579833984375, -0.813995361328125, 2.2281494140625, -0.257080078125, -0.6185302734375, 0.030517578125, 0.1586761474609375, 0.8337287902832031, 1.00042724609375, -0.084686279296875, -2.5618896484375, 0.11834716796875, 0.1390380859375, 0.34771728515625, 0.79345703125, 0.636505126953125, -0.67364501953125, 0.549591064453125, 1.0746002197265625, 2.6322021484375, 0.032657623291015625, -0.19573974609375, 0.4772491455078125, 0.090667724609375, -0.10977935791015625, -1.064453125, -1.0684051513671875, 0.358795166015625, -0.1036376953125, 0.095428466796875, -1.103424072265625, 0.1443328857421875, 0.44915771484375, -0.3755016326904297, -0.19042205810546875, -0.4076881408691406, -0.21356201171875, -0.3951416015625, 0.4341888427734375, 0.6040191650390625, -0.49658203125, -0.46673583984375, 0.07080078125, 1.289306640625, 0.46903228759765625, 0.3756103515625, -1.39508056640625, 0.309326171875, -0.39775848388671875, -0.5999908447265625, 0.2584228515625, 0.7520751953125, 0.57769775390625, -0.619903564453125, -2.46954345703125, 0.38002777099609375, -0.782958984375, 0.78436279296875, 1.2861328125, 0.6348876953125, -0.4887847900390625, -0.15576934814453125, -0.59014892578125, 0.30326080322265625, -0.334869384765625, -0.183197021484375, -0.42041015625, 0.2434844970703125, -0.67681884765625, -2.5330810546875, -0.7532958984375, -0.3831787109375, 0.49109649658203125, 0.188507080078125, 0.0509033203125, -0.36187744140625, -2.441650390625, 0.9922637939453125, 0.1583251953125, 0.7015914916992188, 0.30523681640625, 1.444793701171875, 0.9210205078125, -0.14495849609375, -0.676910400390625, -1.4237136840820312, 0.421783447265625, 0.007659912109375, -0.178192138671875, 0.32830810546875, 0.10575103759765625, -0.8996124267578125, -0.31915283203125, -1.068206787109375, 0.2881927490234375, -0.20245742797851562, 0.788360595703125, 0.275543212890625, -1.44598388671875, -0.46877288818359375, 0.2734832763671875, -0.7142333984375, 1.0740966796875, 0.7767333984375, 0.807586669921875, -0.32331085205078125, 1.2236328125, 0.727783203125, -1.28204345703125, -0.583251953125, 0.2108154296875, 1.067596435546875, -0.9331207275390625, 0.2591094970703125, 1.1275634765625, 2.380279541015625, 0.1941680908203125, 1.502532958984375, 1.274658203125, 0.2132568359375, -2.64874267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000004.npy"}
{"epoch": 0.008376963350785341, "step": 5, "batch_size": 128, "mean": -0.03181034326553345, "std": 0.932866096496582, "min": -3.455596923828125, "p10": -1.2916107177734375, "median": -0.04880523681640625, "p90": 0.9215377807617186, "max": 2.404998779296875, "pos_frac": 0.4609375, "sample": [-0.0648040771484375, -1.61444091796875, 0.4166259765625, -1.29534912109375, 0.875213623046875, 0.9678192138671875, 0.281585693359375, -0.8910064697265625, 0.2941131591796875, -0.5113525390625, -0.099273681640625, -0.789703369140625, 0.2135467529296875, -1.2352523803710938, 0.32498931884765625, 0.901702880859375, 0.6985435485839844, 1.2564697265625, 2.06878662109375, -1.68157958984375, 0.050323486328125, 0.8741455078125, -0.203094482421875, 0.69097900390625, -0.37567138671875, -0.51458740234375, -0.0448760986328125, 0.550384521484375, -0.32045745849609375, -0.09336090087890625, -1.836181640625, -0.456329345703125, 0.892120361328125, 0.63348388671875, 0.8528594970703125, 0.430450439453125, -0.01483154296875, -0.1868896484375, -0.19757080078125, 1.5208740234375, -0.58038330078125, 0.1414794921875, 0.63140869140625, 1.128570556640625, 0.35235595703125, -0.4418525695800781, -1.318450927734375, 1.104736328125, -0.38968658447265625, -0.052734375, 0.8243408203125, -1.455657958984375, -0.88690185546875, -0.42355918884277344, -0.1422576904296875, -0.577789306640625, 0.819091796875, -0.48077392578125, 0.229522705078125, 0.901123046875, -1.2957763671875, -0.410797119140625, 0.447052001953125, 0.14929962158203125, 0.31424713134765625, -0.2868156433105469, 0.8624420166015625, 0.51409912109375, -1.9779052734375, 0.4254150390625, 0.621246337890625, -0.34598541259765625, 0.5509109497070312, -0.18444061279296875, 0.10528373718261719, -0.8575439453125, -1.097381591796875, 1.7398681640625, -0.253662109375, 1.2049102783203125, 0.199981689453125, 1.73614501953125, 0.439422607421875, -1.290008544921875, -0.037628173828125, -1.499053955078125, -1.4976806640625, -0.315093994140625, 1.899169921875, 0.0977935791015625, -3.455596923828125, 0.047405242919921875, -1.107666015625, 0.639404296875, -0.880767822265625, -0.5277099609375, -2.868865966796875, 0.3321075439453125, -0.44499969482421875, -0.00531005859375, 0.59619140625, -1.184356689453125, -0.07073974609375, -0.36328125, -0.532470703125, -1.35076904296875, -0.0867919921875, -0.06697845458984375, -0.36871337890625, -0.303070068359375, -0.4995880126953125, -0.43779754638671875, 1.4314651489257812, -0.1099090576171875, -0.2155609130859375, 0.32666015625, 0.0324859619140625, 0.69622802734375, -0.058380126953125, -0.035430908203125, 2.132232666015625, -0.7060089111328125, 0.395477294921875, 0.30591583251953125, 0.521240234375, 0.5596771240234375, -0.5209503173828125, 2.404998779296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000005.npy"}
{"epoch": 0.010471204188481676, "step": 6, "batch_size": 128, "mean": 0.0020822137594223022, "std": 0.9400361180305481, "min": -2.689910888671875, "p10": -1.1926971435546874, "median": -0.00469970703125, "p90": 1.319427490234375, "max": 3.051513671875, "pos_frac": 0.4921875, "sample": [-0.5230712890625, 0.41998291015625, -0.5128707885742188, 0.6451416015625, -0.85443115234375, 0.7449493408203125, 0.278045654296875, -0.38776397705078125, -0.11154937744140625, 0.032379150390625, -0.581207275390625, 0.8734283447265625, -1.4100341796875, -0.08306884765625, 0.671600341796875, -0.510986328125, -0.2015380859375, 0.3942604064941406, -0.20740699768066406, 1.69659423828125, -0.102264404296875, 0.5012931823730469, -0.4263458251953125, 1.583251953125, -0.618988037109375, -0.3988494873046875, 0.0733642578125, -0.626861572265625, 0.075103759765625, 0.16764450073242188, -0.0093994140625, -0.120208740234375, 0.5023345947265625, -0.557952880859375, -0.08211517333984375, -0.730987548828125, -0.45587158203125, 0.5738067626953125, -1.4825439453125, 1.364013671875, -0.964630126953125, -1.21942138671875, -1.392425537109375, 1.1646728515625, -0.17002296447753906, -2.081329345703125, 0.11698341369628906, 2.71514892578125, 0.028350830078125, -0.81475830078125, 0.0912017822265625, 0.19647216796875, 2.09747314453125, -1.181243896484375, 1.614990234375, -0.2819099426269531, 1.389556884765625, -0.1011962890625, -0.378387451171875, 0.0152587890625, 0.12870407104492188, -1.560150146484375, 0.32208251953125, 0.647308349609375, -1.87841796875, -0.5048980712890625, -0.04232215881347656, 1.336517333984375, -1.299072265625, 0.02538299560546875, 0.972808837890625, 0.0402679443359375, 1.309234619140625, -1.451416015625, 0.43987274169921875, -0.2001953125, 0.6199264526367188, -0.7546157836914062, 0.0, 0.920440673828125, 0.3669586181640625, 1.312103271484375, -1.38836669921875, 0.10132980346679688, -0.627716064453125, -0.11767578125, 0.478790283203125, 0.264007568359375, 0.026275634765625, 0.5555419921875, 0.232269287109375, 0.017852783203125, 0.2593269348144531, -2.689910888671875, 0.990875244140625, 0.0748748779296875, -0.6160736083984375, 0.36708831787109375, -0.9642333984375, 0.282501220703125, 1.715240478515625, 0.119903564453125, -0.60662841796875, -0.60723876953125, -0.7585067749023438, -0.635345458984375, -0.763031005859375, -1.2220458984375, -0.367584228515625, -0.7496299743652344, -0.23175048828125, 0.060352325439453125, 2.58807373046875, -0.5577850341796875, 0.3523101806640625, 3.051513671875, 1.74871826171875, 1.603271484375, -1.478729248046875, 0.2692718505859375, -0.015871047973632812, -0.84417724609375, -0.278411865234375, -0.730987548828125, -0.2396240234375, 0.46044921875, 0.0270233154296875, -0.08519744873046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000006.npy"}
{"epoch": 0.012565445026178011, "step": 7, "batch_size": 128, "mean": -0.07415376603603363, "std": 0.8794519901275635, "min": -3.567138671875, "p10": -1.1500530242919922, "median": -0.05255126953125, "p90": 0.9584838867187498, "max": 2.429931640625, "pos_frac": 0.4765625, "sample": [0.125152587890625, -0.49749755859375, -0.26849365234375, -0.6856842041015625, 0.316009521484375, -0.365325927734375, -0.2692108154296875, -0.8795166015625, -0.76513671875, -1.2801284790039062, 0.817626953125, -0.356597900390625, -1.63995361328125, 0.20877838134765625, 0.508026123046875, -1.68609619140625, 0.7824859619140625, 1.31561279296875, 0.0, 0.4508857727050781, 0.279693603515625, -0.056732177734375, -0.709930419921875, -0.9547271728515625, 0.25445556640625, -0.986724853515625, -0.8204193115234375, -0.8635330200195312, -0.225372314453125, -0.7764892578125, 0.187255859375, -0.6327972412109375, -0.8153610229492188, 0.4208221435546875, 0.5806884765625, 1.02557373046875, 0.33258056640625, -0.17047119140625, -3.567138671875, 0.13706207275390625, -0.13360595703125, 0.04854583740234375, -0.688079833984375, -0.07781982421875, 0.133056640625, 0.99932861328125, 0.94097900390625, 0.5720634460449219, -0.12392807006835938, 0.20904541015625, 0.59661865234375, 0.039825439453125, 0.035888671875, 0.2555389404296875, 0.5732421875, 0.2665863037109375, -0.5615997314453125, 0.223297119140625, -0.36634063720703125, 0.062103271484375, -1.1702003479003906, 0.89239501953125, 0.5660667419433594, -0.20880126953125, -1.3173675537109375, 1.901214599609375, -0.35089111328125, 0.35457801818847656, 1.16839599609375, -0.27880859375, -1.269683837890625, 1.10772705078125, -0.05819892883300781, -0.048370361328125, 0.39040374755859375, -1.14141845703125, 0.28981781005859375, -0.99627685546875, -1.48724365234375, 0.061309814453125, -1.268310546875, 0.08278656005859375, -0.3574676513671875, 0.28545379638671875, -0.0919189453125, -0.0223388671875, -1.19854736328125, 0.154876708984375, 0.24005126953125, -0.2750396728515625, 1.1710205078125, -1.0004119873046875, 0.2340087890625, -0.4618721008300781, 0.2515411376953125, -0.5701904296875, 0.07355499267578125, -0.0955047607421875, 0.019683837890625, -0.82025146484375, -0.578765869140625, -0.08123779296875, -0.782684326171875, 0.9033203125, -0.0980377197265625, 0.544891357421875, 1.2994384765625, 0.7701416015625, -0.065582275390625, -1.98968505859375, 0.4195098876953125, -0.22870635986328125, 1.65057373046875, 1.8734130859375, -0.6064682006835938, 0.7363433837890625, -1.852447509765625, -0.8820648193359375, 1.74627685546875, 2.429931640625, -0.22802734375, 1.12030029296875, -0.7855377197265625, -0.06953239440917969, -0.05841064453125, -2.42535400390625, 0.8482208251953125, 0.668609619140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000007.npy"}
{"epoch": 0.014659685863874346, "step": 8, "batch_size": 128, "mean": -0.06074722111225128, "std": 1.045549988746643, "min": -3.90191650390625, "p10": -1.2079452514648437, "median": 0.010789871215820312, "p90": 1.1897308349609375, "max": 2.90087890625, "pos_frac": 0.5078125, "sample": [-0.607147216796875, 0.5051651000976562, -0.0475006103515625, -1.04705810546875, -0.29705810546875, -0.02410888671875, -0.6981201171875, -0.22951316833496094, -0.05512046813964844, -2.36492919921875, -0.08659744262695312, -0.997711181640625, 2.01318359375, 1.039276123046875, 0.0626983642578125, 0.5362930297851562, -1.3076019287109375, -0.279388427734375, 0.15137100219726562, 0.18017578125, -1.0128173828125, 1.174285888671875, -0.332977294921875, 0.677978515625, -0.4428253173828125, 0.3748321533203125, -0.777557373046875, 0.18438720703125, -1.28216552734375, -2.41229248046875, -0.8822021484375, 2.90087890625, 0.09417724609375, 0.59735107421875, 0.8223114013671875, -1.935577392578125, -0.09333038330078125, -0.372283935546875, -0.361083984375, -1.2324371337890625, 1.38714599609375, -0.754119873046875, 1.5295867919921875, 0.055450439453125, 0.017974853515625, -0.4726409912109375, 0.12320137023925781, 0.6068649291992188, -2.200103759765625, -1.141143798828125, -0.62774658203125, -0.37066650390625, 0.243408203125, -0.21923065185546875, -1.525238037109375, -0.891357421875, 0.38031005859375, 1.22576904296875, 0.015167236328125, -3.90191650390625, 0.0361328125, 0.3992919921875, -0.5528564453125, -0.08660888671875, 0.08771514892578125, 1.2799072265625, -0.9733428955078125, 1.0831527709960938, -1.690673828125, 1.64984130859375, -0.250152587890625, 1.65777587890625, -0.93585205078125, -0.9440383911132812, 0.199462890625, -0.31695556640625, 0.3062744140625, -0.031772613525390625, 0.14637184143066406, -0.3364410400390625, 0.006412506103515625, 1.508056640625, -0.29443359375, 0.17565155029296875, 0.1868743896484375, 0.3494434356689453, -1.3605804443359375, -0.4137420654296875, 0.12594985961914062, 1.763763427734375, 0.40301513671875, -1.19744873046875, 1.1226348876953125, 0.12845611572265625, -0.32843017578125, 0.57275390625, 2.693267822265625, 0.67333984375, 0.3624916076660156, 0.1319580078125, -0.10694122314453125, 0.26763916015625, -0.455108642578125, 1.1380157470703125, -0.11764907836914062, 0.0818634033203125, -0.02545166015625, 0.3196868896484375, 0.4111328125, -3.33837890625, -0.006992340087890625, 0.22100830078125, 0.4466552734375, 2.125457763671875, 0.02337646484375, 0.7969894409179688, -0.4608154296875, -2.188232421875, 0.2494049072265625, 0.3272857666015625, -0.2800750732421875, -1.132843017578125, 0.9452667236328125, -1.1931076049804688, 0.3647575378417969, -0.7992019653320312, 1.756683349609375, -0.09841156005859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000008.npy"}
{"epoch": 0.016753926701570682, "step": 9, "batch_size": 128, "mean": 0.02586349844932556, "std": 0.9224251508712769, "min": -2.376220703125, "p10": -1.159893798828125, "median": 0.0, "p90": 1.1651290893554687, "max": 3.36248779296875, "pos_frac": 0.4921875, "sample": [-0.18395233154296875, -0.74066162109375, -0.5524673461914062, 0.26711273193359375, 0.177093505859375, -0.109161376953125, 0.8331451416015625, -0.000885009765625, -0.1107177734375, -0.02306365966796875, -0.6104583740234375, 0.3720703125, 0.0479736328125, -1.34539794921875, -1.4006195068359375, -1.12335205078125, 0.689056396484375, 0.7713623046875, 0.36066436767578125, -0.753143310546875, 0.40833282470703125, 1.980255126953125, 0.0194549560546875, 1.2945556640625, -0.8327713012695312, -0.11106109619140625, 0.22979736328125, 0.774658203125, 0.1160736083984375, -2.376220703125, 0.02313232421875, 1.91754150390625, -0.7240753173828125, -0.80364990234375, -0.6834564208984375, 0.07817840576171875, 0.744903564453125, 0.05181884765625, 0.10569000244140625, -0.3494873046875, -0.267730712890625, -0.6726837158203125, 1.99969482421875, -0.108856201171875, 0.560821533203125, -0.035888671875, -0.901153564453125, 0.581573486328125, -1.4459228515625, -0.853912353515625, -0.728912353515625, -0.626953125, -1.15277099609375, -0.3123779296875, 0.906951904296875, 0.6228790283203125, 2.37066650390625, 0.248687744140625, -0.71539306640625, -0.514404296875, -1.416473388671875, -0.1657867431640625, -1.176513671875, -0.3505268096923828, -0.38851165771484375, 1.33404541015625, -0.26395416259765625, 0.938446044921875, -0.534912109375, 0.71527099609375, -0.841339111328125, -0.023807525634765625, -1.35418701171875, 0.6474685668945312, 1.16021728515625, 1.0397491455078125, 0.325531005859375, -0.03458404541015625, 0.22153854370117188, 0.0058441162109375, 1.379608154296875, 1.92498779296875, -1.4036102294921875, 0.47027587890625, -0.0717926025390625, -1.4323272705078125, 0.0, -0.7494354248046875, 0.632171630859375, 0.36572265625, 0.2869873046875, 0.18449783325195312, 0.1393585205078125, 0.15094757080078125, 1.484893798828125, 0.5919189453125, -0.672576904296875, -0.993927001953125, -0.1266937255859375, 0.9718170166015625, 0.5074234008789062, 0.0, 1.082794189453125, 0.58209228515625, -0.6964111328125, 0.45696067810058594, 0.0, 0.4554443359375, -0.2214813232421875, 0.08241653442382812, 1.282958984375, 1.1765899658203125, -1.489898681640625, -2.181671142578125, -1.3031005859375, -0.09212493896484375, -0.2182769775390625, 1.254486083984375, 0.300323486328125, 0.37194061279296875, -0.205078125, 3.36248779296875, -0.4224395751953125, 0.752777099609375, -1.8302841186523438, -0.69866943359375, 1.056640625, -0.408294677734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000009.npy"}
{"epoch": 0.018848167539267015, "step": 10, "batch_size": 128, "mean": -0.0460197776556015, "std": 0.8259887099266052, "min": -1.69305419921875, "p10": -1.06219482421875, "median": -0.07808113098144531, "p90": 0.8450210571289062, "max": 3.3668212890625, "pos_frac": 0.453125, "sample": [-0.22784423828125, -0.6300201416015625, 0.8679962158203125, -0.5340118408203125, 0.5921707153320312, 0.28326416015625, 0.10396575927734375, -0.246551513671875, -0.5764007568359375, -1.2431793212890625, -0.5800628662109375, 0.2878131866455078, 0.2812347412109375, 3.3668212890625, 0.981475830078125, 0.048084259033203125, 0.004787445068359375, -0.59808349609375, 0.835174560546875, 1.78570556640625, -1.1341552734375, -0.15448760986328125, -1.2452239990234375, -1.36224365234375, -0.008941650390625, -0.656219482421875, 1.526763916015625, -0.18628692626953125, -0.6905517578125, -0.466064453125, 0.25464630126953125, 0.1384124755859375, -0.08141326904296875, -0.0867462158203125, -0.943145751953125, -0.275482177734375, 0.69024658203125, 0.33282470703125, 0.468170166015625, 0.7393798828125, -1.321807861328125, 0.16567230224609375, -0.1296710968017578, 0.6710205078125, 0.219329833984375, 0.4356231689453125, -0.4699287414550781, 0.08101654052734375, 0.267791748046875, -1.4105224609375, 0.08415603637695312, -1.09979248046875, 1.031982421875, -1.04608154296875, -0.8574066162109375, 0.3913421630859375, -0.9381866455078125, 1.5033416748046875, -0.27268409729003906, 1.571563720703125, -0.986602783203125, -0.9280204772949219, -0.327117919921875, 0.5528640747070312, 0.20255279541015625, -0.167266845703125, -0.4260711669921875, -1.69305419921875, -0.537384033203125, 0.781890869140625, -0.6975250244140625, -0.0467071533203125, 0.643646240234375, 0.170562744140625, -0.5686569213867188, -0.1926727294921875, -0.00339508056640625, -0.87530517578125, -0.067108154296875, 2.20245361328125, -0.096954345703125, 1.9036865234375, -0.502471923828125, 1.075103759765625, 0.13671875, -1.5169830322265625, -0.2864990234375, -0.796051025390625, -0.296142578125, -1.377532958984375, 0.4971923828125, 0.038288116455078125, 0.2060089111328125, 0.5269775390625, -0.31915283203125, 0.0435638427734375, 0.141021728515625, 0.405364990234375, -0.4756011962890625, 0.7714080810546875, 0.250732421875, -0.4607505798339844, 0.25518798828125, -0.3006439208984375, -1.59332275390625, -0.0980072021484375, -0.107330322265625, -1.485992431640625, -0.6666183471679688, 1.416473388671875, 0.06781005859375, -0.14809036254882812, 0.5826263427734375, 0.0, 0.08412933349609375, -1.4871826171875, -0.07474899291992188, -0.89898681640625, 0.1914215087890625, -0.149688720703125, 1.547454833984375, -0.0982666015625, 0.50732421875, -0.3020172119140625, 0.736541748046875, -0.90667724609375, -0.951934814453125, 0.546417236328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000010.npy"}
{"epoch": 0.020942408376963352, "step": 11, "batch_size": 128, "mean": 0.042532920837402344, "std": 0.9623458981513977, "min": -2.81219482421875, "p10": -1.0127311706542967, "median": 0.04376792907714844, "p90": 1.1346328735351563, "max": 3.2740478515625, "pos_frac": 0.5234375, "sample": [-0.4215087890625, -0.82318115234375, 0.99786376953125, 1.76629638671875, 1.1450042724609375, -0.24498748779296875, -0.22064208984375, -1.305206298828125, 1.262451171875, 0.13787841796875, -0.04164886474609375, 0.1588134765625, 0.894866943359375, -0.1446533203125, 0.087158203125, 0.42661285400390625, -2.81219482421875, 0.1969280242919922, -0.023773193359375, 0.02643585205078125, -0.04368400573730469, 0.392547607421875, 0.4453125, 2.9119873046875, 0.0041656494140625, -0.4060821533203125, 0.577911376953125, -0.3182220458984375, 1.4478759765625, 0.32456207275390625, -0.740875244140625, 0.631134033203125, 0.36064910888671875, 0.33123779296875, -0.0746002197265625, 1.904632568359375, -1.30194091796875, 0.29970550537109375, 1.473907470703125, 0.3595008850097656, 0.4288330078125, -0.040435791015625, -2.592529296875, -0.787933349609375, 0.0918121337890625, -0.04510498046875, -0.167816162109375, 1.776397705078125, 3.2740478515625, 1.46319580078125, -0.5812835693359375, -0.63250732421875, -0.8770980834960938, -1.228302001953125, 0.720550537109375, -0.66455078125, -0.6750946044921875, 0.679962158203125, 0.731292724609375, -0.1533203125, 0.00341796875, 0.1671142578125, -1.27117919921875, -0.786407470703125, 0.28790283203125, -0.0269927978515625, 0.4752655029296875, 0.2812328338623047, 0.1480255126953125, -0.36013031005859375, -0.11614990234375, -0.3333892822265625, -0.68084716796875, -1.3433837890625, 0.25165557861328125, 0.6279754638671875, 0.061100006103515625, -0.6122512817382812, 0.074981689453125, 0.8994140625, 0.16304779052734375, 1.13018798828125, 0.09532928466796875, -0.5674972534179688, 0.167694091796875, 0.61383056640625, -0.12432861328125, 0.3760986328125, -0.40545654296875, 1.091583251953125, 0.423614501953125, -0.5761489868164062, -1.406097412109375, 0.08670425415039062, -0.7691192626953125, -0.68255615234375, 0.4398193359375, -1.776885986328125, -0.3884468078613281, 0.13125038146972656, -0.35649871826171875, -0.304901123046875, 2.31463623046875, -0.160888671875, 1.02130126953125, 0.34161376953125, 1.175750732421875, 0.1174774169921875, -0.201995849609375, -0.0015106201171875, 0.3207244873046875, 0.759765625, -2.4476318359375, -1.0284805297851562, -0.1331634521484375, -2.50189208984375, -1.5404205322265625, -0.0980224609375, 0.428680419921875, 2.081573486328125, -1.0059814453125, 0.5373382568359375, 0.9903717041015625, -0.78118896484375, -0.387542724609375, -0.21372222900390625, -0.42437744140625, 0.8108673095703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000011.npy"}
{"epoch": 0.023036649214659685, "step": 12, "batch_size": 128, "mean": 0.049374401569366455, "std": 0.9006751179695129, "min": -2.590118408203125, "p10": -0.984283447265625, "median": 0.06541633605957031, "p90": 1.25343017578125, "max": 2.4132080078125, "pos_frac": 0.53125, "sample": [-1.043914794921875, 1.376739501953125, 0.4468994140625, 0.59051513671875, -0.034191131591796875, -0.24948883056640625, -1.7182159423828125, 1.6131591796875, 1.65252685546875, -0.23370361328125, -0.5885162353515625, 0.487091064453125, -0.3206787109375, 0.9050445556640625, -0.269989013671875, -0.44908905029296875, 0.620758056640625, 0.3519744873046875, 0.24102783203125, 0.9313201904296875, -0.590057373046875, 1.04656982421875, 2.13568115234375, 0.40821075439453125, -0.5357666015625, -0.8896942138671875, -1.334634780883789, -0.084564208984375, 0.865020751953125, -0.3796234130859375, -0.89019775390625, 0.7600784301757812, -0.6920242309570312, -2.590118408203125, 0.556915283203125, 2.4132080078125, -0.23402023315429688, -0.8347625732421875, 0.390045166015625, -0.7648468017578125, -0.4573822021484375, -1.18017578125, -0.8331298828125, 0.3193359375, 1.3592376708984375, 0.08266067504882812, -0.75482177734375, 0.2679595947265625, -0.470916748046875, 0.018218994140625, 1.0258941650390625, -0.88128662109375, 0.446685791015625, 1.38983154296875, -0.8608322143554688, -0.47769927978515625, 1.2557373046875, 0.41162109375, -0.6424331665039062, 0.8498077392578125, -0.539947509765625, -0.70965576171875, 0.044189453125, -0.9582672119140625, 0.36884307861328125, 0.809600830078125, 0.45349884033203125, 0.6435546875, 0.0481719970703125, 1.2191314697265625, -0.137176513671875, -0.4335784912109375, 0.7035064697265625, 0.873809814453125, 0.50958251953125, -0.416839599609375, 0.30609893798828125, -0.565216064453125, -1.3253936767578125, -0.983367919921875, 0.922393798828125, 0.4286346435546875, -1.465606689453125, 0.84100341796875, 0.04150962829589844, -0.965850830078125, 0.08296585083007812, 0.1583099365234375, 1.25244140625, -0.373687744140625, 1.43890380859375, 0.2809333801269531, -0.7663192749023438, 0.18353271484375, 0.4693603515625, -1.3204345703125, -0.982177734375, -0.37548828125, -0.9982147216796875, 0.264923095703125, -1.13873291015625, -0.4711189270019531, 0.425750732421875, -0.34661865234375, 0.13592529296875, -0.711700439453125, 0.86053466796875, 1.280914306640625, 0.6572723388671875, 0.620849609375, 1.71429443359375, 0.57843017578125, -0.5046615600585938, -0.986419677734375, 1.1093902587890625, 0.154693603515625, -0.057079315185546875, -0.702880859375, -2.09051513671875, -0.8554534912109375, 1.399383544921875, 0.92559814453125, 1.221405029296875, -0.40240478515625, 1.72503662109375, -1.2659149169921875, 0.70751953125, -0.6242523193359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000012.npy"}
{"epoch": 0.025130890052356022, "step": 13, "batch_size": 128, "mean": 0.004402056336402893, "std": 0.8914082050323486, "min": -2.8680267333984375, "p10": -0.9316215515136719, "median": -5.7220458984375e-05, "p90": 1.0496109008789063, "max": 2.91424560546875, "pos_frac": 0.4921875, "sample": [-1.18731689453125, -0.1936206817626953, -0.41774749755859375, -2.56402587890625, 0.21270751953125, 0.64678955078125, 1.16192626953125, -0.90234375, 0.1771697998046875, 0.28205108642578125, -0.541259765625, -0.397491455078125, -0.1513519287109375, 0.003509521484375, 0.1824016571044922, 0.201751708984375, 0.7423095703125, 1.744873046875, -0.0782318115234375, -0.08870697021484375, -0.0209503173828125, -2.28985595703125, 2.91424560546875, -0.00011444091796875, -0.112518310546875, -0.2928962707519531, -0.014200210571289062, 0.434173583984375, 0.10015869140625, -0.61138916015625, 0.067413330078125, -0.11658477783203125, 1.834075927734375, 0.0428619384765625, 0.138824462890625, -0.11919403076171875, 0.24295997619628906, -0.5814666748046875, -0.0028228759765625, 0.66851806640625, 0.46823692321777344, 1.0584869384765625, 0.9351348876953125, 0.74658203125, -0.6866912841796875, 0.435546875, -2.8680267333984375, -0.3332977294921875, 0.176910400390625, 0.33306884765625, 0.4538726806640625, -1.6351165771484375, 0.10954666137695312, 1.1521148681640625, -1.1776123046875, 0.0, -0.2230224609375, 0.024993896484375, 0.0850372314453125, -0.745849609375, -0.05019378662109375, -0.6010589599609375, -1.51824951171875, 0.21848678588867188, 1.46575927734375, 0.04229736328125, 0.164520263671875, 1.16949462890625, 0.1451416015625, -0.02539825439453125, -0.940673828125, 0.189697265625, -0.0373992919921875, 0.3485107421875, -0.54931640625, 0.683349609375, -0.1509552001953125, 1.68536376953125, -0.4868583679199219, 0.6900787353515625, 0.01247406005859375, 0.6297988891601562, -0.7508697509765625, -1.743621826171875, 0.947235107421875, -0.83306884765625, 0.2186737060546875, -0.6208114624023438, 0.69635009765625, -0.064361572265625, 1.938568115234375, -0.11067962646484375, 0.9984588623046875, -0.6223602294921875, 0.07561492919921875, -0.40630340576171875, -0.18352127075195312, -0.7476806640625, -0.6546173095703125, 0.87908935546875, -0.36370849609375, -1.430694580078125, -0.1065826416015625, -0.21320343017578125, -0.57513427734375, -1.802001953125, 1.72705078125, -0.722412109375, 0.4539337158203125, -0.22283935546875, 0.40720176696777344, -0.6622314453125, -1.2052383422851562, 0.13791465759277344, 2.50103759765625, 1.52435302734375, -0.9277420043945312, -0.02490234375, 0.714996337890625, -0.43475341796875, 0.2099609375, -1.06793212890625, 0.220367431640625, 0.14421463012695312, 1.045806884765625, 0.3443794250488281, -0.4843902587890625, -0.14949798583984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000013.npy"}
{"epoch": 0.027225130890052355, "step": 14, "batch_size": 128, "mean": 0.10642816126346588, "std": 0.9013544321060181, "min": -2.055206298828125, "p10": -0.9237709045410156, "median": 0.04822349548339844, "p90": 1.2205978393554686, "max": 3.513916015625, "pos_frac": 0.5234375, "sample": [0.5992584228515625, 0.4235382080078125, -0.14110565185546875, 0.27581787109375, 1.0248565673828125, -1.03961181640625, -0.1971282958984375, -0.0311279296875, -0.03670692443847656, 2.167877197265625, 0.3318634033203125, -0.47188568115234375, -0.8583984375, -0.874725341796875, 0.362884521484375, 0.317718505859375, -0.668731689453125, 0.647674560546875, -0.3356208801269531, -0.8194122314453125, -0.392181396484375, -0.035953521728515625, 0.483642578125, 0.2306060791015625, 0.347686767578125, -1.17242431640625, -0.3896484375, 0.1155242919921875, 0.19476318359375, 0.1949462890625, 1.6359405517578125, -0.1435546875, -0.13201904296875, 1.0727386474609375, 0.41046142578125, 0.20684814453125, -2.055206298828125, -0.18457603454589844, 1.377838134765625, -0.583221435546875, 0.2991447448730469, 0.791839599609375, -0.9198226928710938, 0.619415283203125, 0.0, -0.141693115234375, -0.759063720703125, -0.475311279296875, 0.134124755859375, -1.0283737182617188, 0.7165679931640625, 1.2004241943359375, 0.1510791778564453, 0.01068115234375, -0.2939605712890625, -0.88519287109375, 1.131317138671875, 3.513916015625, 0.920166015625, 1.706298828125, -0.1937255859375, -1.55657958984375, 1.086334228515625, 1.288726806640625, -0.1066131591796875, 0.401397705078125, -1.609222412109375, 0.245849609375, 1.05389404296875, -0.656524658203125, 0.0586395263671875, -0.2290496826171875, -0.4799079895019531, 1.267669677734375, 1.58660888671875, 1.0786514282226562, -0.28812408447265625, 2.02203369140625, -0.893280029296875, -0.2885894775390625, -0.20532608032226562, -0.1600494384765625, 0.32805633544921875, 0.9063568115234375, 0.52728271484375, 0.037807464599609375, 0.06652259826660156, -0.5048675537109375, 0.585479736328125, 0.014270782470703125, -0.8462677001953125, -0.20554351806640625, -1.47332763671875, 0.3222503662109375, -1.700408935546875, 1.368743896484375, 0.0811309814453125, -0.16510009765625, 1.70343017578125, 0.1363983154296875, -0.68572998046875, -0.1175079345703125, 0.34210968017578125, -0.9442138671875, -0.617919921875, 0.7686004638671875, -0.329345703125, -0.638092041015625, 0.1447906494140625, -1.1417083740234375, 0.2048187255859375, 1.0673980712890625, -1.02545166015625, -0.6437530517578125, 0.15550804138183594, -0.9329833984375, -0.3535919189453125, 0.81744384765625, -0.2754364013671875, 0.3510932922363281, -1.212158203125, 0.099761962890625, 1.100860595703125, -0.176300048828125, -0.38858795166015625, 0.0924072265625, 3.17724609375, 1.627716064453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000014.npy"}
{"epoch": 0.02931937172774869, "step": 15, "batch_size": 128, "mean": -0.01959007978439331, "std": 0.8517094850540161, "min": -1.952880859375, "p10": -1.2769622802734375, "median": -0.01082611083984375, "p90": 0.9128158569335937, "max": 3.289581298828125, "pos_frac": 0.4921875, "sample": [0.8413848876953125, -0.717926025390625, -0.15875244140625, -0.0830078125, -1.3786468505859375, 0.6635055541992188, 0.7586212158203125, -0.640655517578125, 0.6519927978515625, -0.19263458251953125, -0.032115936279296875, 0.23828125, -0.03228759765625, -0.4137535095214844, 0.1616363525390625, -0.74127197265625, -0.4170684814453125, -0.19083786010742188, -0.08136749267578125, -0.979888916015625, -0.29339599609375, -0.3337249755859375, -0.741485595703125, -1.558380126953125, 0.53619384765625, 1.1385345458984375, -0.0084686279296875, -1.60467529296875, -0.399993896484375, -0.31386566162109375, 0.838134765625, 0.272430419921875, 0.174713134765625, 0.07498931884765625, -0.1593017578125, 0.09318161010742188, 0.9105224609375, 0.812744140625, 0.125213623046875, 0.17171478271484375, -0.0452423095703125, -1.952880859375, 0.158355712890625, -0.1396331787109375, 0.777374267578125, -1.0134124755859375, 0.2623710632324219, 0.1365203857421875, 0.5145111083984375, 1.27935791015625, -0.24991989135742188, -0.23355865478515625, 0.077972412109375, 1.3480224609375, 1.08941650390625, 0.0792236328125, -1.72064208984375, -0.3658294677734375, 0.001506805419921875, -0.549102783203125, 0.31658172607421875, 0.4230461120605469, -0.56427001953125, 0.0128936767578125, -1.706329345703125, -0.0594482421875, -0.3814544677734375, 0.0204010009765625, 2.435821533203125, -0.939605712890625, -0.14883041381835938, -0.6633224487304688, -0.0325775146484375, -1.5664825439453125, 0.151947021484375, 0.7864227294921875, 0.68792724609375, 0.9437179565429688, -0.505889892578125, 0.9181671142578125, 0.584930419921875, -0.024250030517578125, 1.6669921875, 0.635986328125, -1.84185791015625, 0.404815673828125, -1.30078125, -0.4249267578125, 0.3111763000488281, 0.266448974609375, -0.0809173583984375, -0.2933349609375, 0.00386810302734375, -1.377960205078125, 0.374664306640625, 1.63629150390625, -0.51495361328125, 0.61993408203125, -1.8203125, 1.0693359375, -0.0162811279296875, 3.289581298828125, 0.643585205078125, 0.65631103515625, -0.251068115234375, 0.532196044921875, 0.430419921875, -0.13468170166015625, -1.1834716796875, -0.7779197692871094, 0.6485366821289062, 0.339752197265625, 0.1481781005859375, 0.0664520263671875, -1.377471923828125, -0.044223785400390625, -1.67498779296875, 1.3257827758789062, -0.4616546630859375, 0.07965087890625, -0.01318359375, 0.13594436645507812, -0.960357666015625, -1.266754150390625, -0.0142822265625, 1.3599090576171875, -0.66900634765625, 0.21295166015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000015.npy"}
{"epoch": 0.031413612565445025, "step": 16, "batch_size": 128, "mean": 0.04610002040863037, "std": 0.9196972846984863, "min": -2.28704833984375, "p10": -1.0654815673828124, "median": 0.00063323974609375, "p90": 0.992604064941406, "max": 3.84600830078125, "pos_frac": 0.5, "sample": [-0.16046142578125, -0.09432220458984375, 0.167022705078125, 0.3041839599609375, 0.60888671875, 0.2792816162109375, 0.0012664794921875, 0.59259033203125, 0.0906982421875, -0.97332763671875, -0.008869171142578125, -0.07592010498046875, 0.49770355224609375, -1.12744140625, 3.84600830078125, -0.3620147705078125, 0.22515869140625, 0.52252197265625, -1.64453125, 0.436859130859375, 1.0670623779296875, -0.07784271240234375, 0.739776611328125, 1.544830322265625, 2.00335693359375, 0.017425537109375, 0.636474609375, 3.138336181640625, 0.294952392578125, -1.07977294921875, 2.00067138671875, -0.16278076171875, -1.127899169921875, -0.28643798828125, 1.209747314453125, -0.585601806640625, 1.54083251953125, -0.36163330078125, -0.4929237365722656, 0.050426483154296875, 0.648223876953125, 0.4091796875, -0.3193359375, -1.761962890625, -0.9154052734375, -0.17182350158691406, 0.099945068359375, -1.1696319580078125, 0.28369140625, 0.5090065002441406, 0.5936050415039062, 0.70684814453125, -0.2711181640625, -0.62579345703125, -1.2623291015625, -0.3594970703125, 0.21027374267578125, 0.06256103515625, 0.7696151733398438, 0.2880859375, 0.711181640625, -0.06538009643554688, 0.2340545654296875, -0.0847320556640625, -0.7094268798828125, -0.5167236328125, 0.34088134765625, 0.38677978515625, -0.451171875, -2.28704833984375, 0.37660980224609375, 1.12066650390625, -0.5007476806640625, 0.668731689453125, -0.222808837890625, -1.3404388427734375, -0.58868408203125, 0.9063873291015625, -0.11243438720703125, 0.28765869140625, -0.630126953125, -0.17009353637695312, -0.636962890625, -0.3414764404296875, -0.58795166015625, 0.828948974609375, 0.0, 0.8755111694335938, -0.566558837890625, 0.20892333984375, -1.122833251953125, 0.585662841796875, -0.0247344970703125, -0.15380859375, 0.182373046875, 0.61785888671875, 0.2614479064941406, -0.0648193359375, -0.35650634765625, 0.3119182586669922, 1.436065673828125, 1.5234375, -1.059356689453125, 0.3092193603515625, -0.48028564453125, 0.4806060791015625, -0.366424560546875, -0.192962646484375, -0.8857421875, 0.9451904296875, 0.2703094482421875, -0.0240325927734375, -2.1496734619140625, -0.470428466796875, -2.08966064453125, -0.23297119140625, 1.3513641357421875, 0.26694488525390625, -1.930908203125, -0.287506103515625, -0.9006729125976562, 0.302825927734375, 0.960693359375, -0.34979248046875, -0.3077392578125, 0.51776123046875, -0.896881103515625, 1.87286376953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000016.npy"}
{"epoch": 0.033507853403141365, "step": 17, "batch_size": 128, "mean": 0.08215945959091187, "std": 0.8861317038536072, "min": -2.54754638671875, "p10": -0.8830642700195312, "median": 0.09089851379394531, "p90": 1.0620330810546874, "max": 2.707733154296875, "pos_frac": 0.53125, "sample": [-0.4536590576171875, -0.4615020751953125, -1.6766357421875, -0.30517578125, -0.36651611328125, 0.286376953125, 1.062225341796875, -0.09283447265625, -0.1241607666015625, 0.6024627685546875, 0.2243804931640625, -0.3833160400390625, -0.342987060546875, 2.42620849609375, -0.568206787109375, -0.16017913818359375, 0.130584716796875, 0.20849609375, 0.149658203125, 0.22625732421875, -0.0704345703125, 0.749298095703125, -0.648101806640625, -0.234527587890625, 0.09433364868164062, -0.4259185791015625, -2.406646728515625, -0.139404296875, -1.35162353515625, 0.72113037109375, 0.3560943603515625, 1.05682373046875, 0.816192626953125, 0.28240966796875, -1.28729248046875, 0.05033111572265625, 0.511962890625, -2.54754638671875, -0.08549118041992188, 0.1503753662109375, 0.894683837890625, -0.6987762451171875, 0.18524169921875, -0.5116004943847656, -0.5045623779296875, -0.0682525634765625, 0.45424652099609375, -0.05282783508300781, 0.9853057861328125, 1.0456695556640625, 0.251068115234375, 0.7126617431640625, -1.7272796630859375, 1.90625, -0.6049880981445312, -0.9844970703125, 0.22751998901367188, 0.22637939453125, 0.1346588134765625, 0.1060791015625, 0.393341064453125, -0.033191680908203125, -0.6110305786132812, 0.0301666259765625, 1.5377197265625, -0.8499755859375, -1.7239990234375, 0.21303176879882812, -0.640960693359375, -0.6719131469726562, 0.89178466796875, -1.4268035888671875, -0.8143310546875, -0.628448486328125, 1.7824249267578125, -0.20379638671875, -0.8796539306640625, -0.2762908935546875, 1.50189208984375, 0.2074432373046875, 1.753021240234375, -0.5117034912109375, -1.1800384521484375, -0.551300048828125, -0.07271575927734375, 0.980438232421875, 2.11199951171875, 0.684844970703125, -1.4372406005859375, 0.33563232421875, 0.5450763702392578, 0.5773468017578125, 0.420013427734375, 0.713287353515625, 1.178009033203125, -0.891021728515625, -0.28521728515625, -0.46282958984375, 0.23249053955078125, 0.08746337890625, -0.18022918701171875, 0.9195556640625, 1.165252685546875, 2.707733154296875, 0.52685546875, 1.013885498046875, 0.7401046752929688, 0.3891143798828125, 0.219635009765625, -0.4761962890625, -0.5416984558105469, -0.29857635498046875, 0.40863037109375, -0.544708251953125, 1.06195068359375, -0.35479736328125, 1.26885986328125, 0.89166259765625, 0.0833892822265625, -0.44329833984375, -0.20180130004882812, -0.9062042236328125, 0.68914794921875, 0.1985321044921875, -0.20733642578125, 0.8962554931640625, -0.3377532958984375, 1.853057861328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000017.npy"}
{"epoch": 0.0356020942408377, "step": 18, "batch_size": 128, "mean": -0.11993266642093658, "std": 0.9162041544914246, "min": -3.45184326171875, "p10": -1.12227783203125, "median": -0.060272216796875, "p90": 0.9075103759765623, "max": 2.11895751953125, "pos_frac": 0.4609375, "sample": [-0.999481201171875, -1.70892333984375, 0.4572124481201172, 0.1289520263671875, 0.162506103515625, -0.924163818359375, -0.014484405517578125, -0.19623565673828125, -0.37358856201171875, -0.0968780517578125, 0.607086181640625, 0.3319244384765625, -0.0723876953125, 0.346771240234375, 0.77520751953125, -0.1942138671875, -0.9058837890625, -0.220184326171875, 0.8802337646484375, -0.48325347900390625, 0.5681610107421875, 1.0343780517578125, -0.27984619140625, -0.399505615234375, 1.59124755859375, -0.20367050170898438, -1.644317626953125, 0.656707763671875, -0.92626953125, -0.33596038818359375, 0.084228515625, -1.11920166015625, -0.5390701293945312, -0.4760284423828125, -0.3828887939453125, -0.7162857055664062, 0.717864990234375, 0.3364601135253906, -0.6208877563476562, -0.62493896484375, -0.1255645751953125, -0.019989013671875, -2.49029541015625, 1.5831298828125, 0.02303314208984375, 0.6053009033203125, 1.19586181640625, 0.32830047607421875, 0.782623291015625, -0.79345703125, 0.8859405517578125, 0.446075439453125, 0.06500244140625, 0.1547698974609375, -0.05426025390625, -1.12945556640625, 0.0134429931640625, 0.1212158203125, 0.578094482421875, 0.2634429931640625, 1.197296142578125, -0.35192108154296875, -0.0689697265625, -1.554443359375, -3.45184326171875, -0.959686279296875, -2.54437255859375, 1.01959228515625, -0.41595458984375, -0.22802734375, 0.167755126953125, 0.16343307495117188, -0.6914215087890625, -0.04144287109375, 0.9578399658203125, -0.2344512939453125, -0.881500244140625, -0.3635101318359375, 0.5240936279296875, 0.3867034912109375, -0.424591064453125, 0.10128021240234375, 1.46343994140625, 0.6278533935546875, 0.457611083984375, 0.7060165405273438, -0.670989990234375, 2.11895751953125, 1.05389404296875, -0.22484588623046875, 0.61395263671875, 0.42372894287109375, -0.0187225341796875, 0.0430450439453125, -3.177001953125, -0.66851806640625, -0.24866485595703125, 0.04384613037109375, -0.7124481201171875, 0.39673614501953125, -0.8349533081054688, -0.63287353515625, 0.236968994140625, 0.60260009765625, 0.07515716552734375, -0.85009765625, -0.22625732421875, -0.848968505859375, -0.7408828735351562, -1.3560638427734375, -1.71124267578125, 0.571502685546875, -0.1373291015625, -0.572265625, 1.908294677734375, -0.0662841796875, 0.2381744384765625, 0.297210693359375, -0.15386962890625, 1.30157470703125, 0.42169189453125, -0.3084259033203125, -1.089111328125, -1.40924072265625, 1.2884521484375, -1.1817626953125, 0.6469879150390625, -2.007720947265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000018.npy"}
{"epoch": 0.03769633507853403, "step": 19, "batch_size": 128, "mean": -0.07141060382127762, "std": 0.8510531187057495, "min": -2.72509765625, "p10": -1.0739662170410156, "median": 0.011180877685546875, "p90": 0.8112915039062499, "max": 3.4735107421875, "pos_frac": 0.515625, "sample": [0.4651985168457031, 0.27239990234375, 0.2302398681640625, 0.03125, 1.265869140625, -1.3427734375, 0.497406005859375, 0.2386474609375, -0.6804962158203125, -2.5654296875, 0.07013320922851562, 0.08317756652832031, 0.5423126220703125, -1.0770263671875, 0.18982315063476562, -0.7486839294433594, 0.437744140625, -0.9473876953125, 0.49079132080078125, 1.0097198486328125, 0.506927490234375, -0.872344970703125, -0.290557861328125, 0.819793701171875, 0.3902740478515625, -0.05657958984375, -0.10157585144042969, 0.1743927001953125, -1.56396484375, 0.1674652099609375, 0.807647705078125, 1.27508544921875, -1.45672607421875, -0.6334228515625, -0.2899932861328125, -0.18737030029296875, 1.006011962890625, 0.170806884765625, -0.2693634033203125, 0.30689048767089844, -0.39251708984375, 0.3818798065185547, -1.7626953125, -0.8907012939453125, 0.6590576171875, 1.06805419921875, -0.2845726013183594, 0.075836181640625, -2.321197509765625, -0.23731231689453125, -0.49853515625, -0.41485595703125, 0.2993927001953125, -0.5794677734375, -0.79132080078125, 1.4075927734375, 0.60919189453125, 1.78839111328125, 0.21506500244140625, 0.00470733642578125, -0.3959503173828125, -0.4857330322265625, 0.367401123046875, 0.30681610107421875, -0.399658203125, 0.0038604736328125, 0.18865966796875, 0.6952362060546875, -1.0726547241210938, 1.5685272216796875, -0.86444091796875, -0.6395263671875, -0.3656158447265625, -0.2879638671875, 0.26638031005859375, -1.5827484130859375, -0.56512451171875, 0.5661697387695312, -0.275360107421875, 0.62213134765625, 0.38995361328125, -1.93499755859375, 0.03612327575683594, 0.243255615234375, 0.2332000732421875, -0.2351226806640625, -0.25248146057128906, -1.6746826171875, -0.32054901123046875, -0.129119873046875, -0.07232666015625, -0.1390380859375, 0.779296875, 0.0236053466796875, 0.873687744140625, -0.0860595703125, 0.31329345703125, -0.3785743713378906, 0.6069984436035156, 0.3748922348022461, -0.05611419677734375, -0.2523956298828125, -0.35504150390625, -0.16949462890625, 0.12152099609375, -0.04669189453125, 0.119537353515625, 0.14178466796875, -0.3390655517578125, -1.142486572265625, 0.5155181884765625, -1.0666122436523438, 1.083740234375, -1.04022216796875, -0.836334228515625, -1.4088287353515625, 0.28375244140625, -0.479736328125, 3.4735107421875, 1.1924972534179688, -0.163299560546875, 0.0518798828125, 0.37158203125, 0.0176544189453125, 0.35733795166015625, -2.72509765625, 0.379638671875, -0.17315673828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000019.npy"}
{"epoch": 0.039790575916230364, "step": 20, "batch_size": 128, "mean": 0.07696413993835449, "std": 0.8579336404800415, "min": -2.260986328125, "p10": -0.9065322875976561, "median": 0.0470733642578125, "p90": 1.0321685791015625, "max": 3.53094482421875, "pos_frac": 0.53125, "sample": [-0.0478363037109375, 0.07342529296875, 0.95587158203125, 3.53094482421875, 0.30120849609375, -2.260986328125, -0.105438232421875, 0.58172607421875, 1.0247802734375, -0.7721328735351562, -0.51934814453125, -0.334686279296875, 0.1214599609375, 1.521270751953125, 0.800048828125, 1.1821746826171875, -0.51849365234375, -0.975555419921875, 0.7597579956054688, -0.17901229858398438, -1.06689453125, -0.245819091796875, 0.52191162109375, -0.19002151489257812, 0.0471649169921875, 0.12625885009765625, 0.04791259765625, 0.094940185546875, 0.38135528564453125, 0.2166748046875, 0.01092529296875, -0.09332275390625, 0.23523712158203125, 0.20379638671875, -0.4226226806640625, -0.9667205810546875, 0.2240447998046875, -0.139892578125, -2.16534423828125, -0.978912353515625, -0.811676025390625, 0.42401123046875, -1.50372314453125, 0.4070587158203125, -0.665008544921875, -0.7706298828125, -0.25689697265625, 0.19217681884765625, 0.0612335205078125, 0.04691314697265625, 1.013641357421875, -0.22069549560546875, 0.0, 1.197265625, 2.120269775390625, 0.7271804809570312, -0.0035381317138671875, 1.70367431640625, 0.8763427734375, -0.411041259765625, 0.03876495361328125, 0.42767333984375, -0.47125244140625, 0.061187744140625, 0.6754150390625, -0.581329345703125, 1.0515289306640625, 0.389923095703125, -0.3487091064453125, -1.24444580078125, -0.35882568359375, -0.001903533935546875, 0.9280853271484375, -0.24974441528320312, 0.748992919921875, -0.67572021484375, -0.6074714660644531, 0.290557861328125, -0.35109710693359375, 0.9586029052734375, 0.113861083984375, 0.2142181396484375, -0.12829208374023438, -0.4411125183105469, 0.37945556640625, -0.37659454345703125, -1.439666748046875, 0.4954833984375, -0.2421875, -0.052001953125, -0.1043853759765625, 2.7496337890625, 0.4093017578125, -1.43975830078125, -0.1269683837890625, 0.963134765625, -0.20684814453125, -0.3070068359375, 0.19226837158203125, 0.19584274291992188, -0.85980224609375, 0.94049072265625, -0.03069305419921875, -0.33666229248046875, -1.564056396484375, -0.3041534423828125, 0.0469818115234375, -1.629058837890625, -1.2728652954101562, 1.049407958984375, 1.95867919921875, -0.3450164794921875, 0.8418655395507812, 0.516937255859375, 0.09893798828125, 0.850830078125, 0.3516845703125, 0.5538787841796875, -0.8807373046875, 0.17333602905273438, 1.190704345703125, 1.06097412109375, -0.11785888671875, -0.348175048828125, 0.05185890197753906, -0.7314605712890625, 0.7712249755859375, 1.17913818359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000020.npy"}
{"epoch": 0.041884816753926704, "step": 21, "batch_size": 128, "mean": -0.02755562961101532, "std": 0.8710337281227112, "min": -2.267059326171875, "p10": -1.062060546875, "median": -0.0632781982421875, "p90": 1.138568115234375, "max": 2.064178466796875, "pos_frac": 0.4765625, "sample": [0.464263916015625, 0.0040283203125, -0.2597808837890625, -0.780303955078125, -0.6690673828125, -2.076507568359375, -0.74908447265625, -0.912384033203125, -0.4436187744140625, -0.40929412841796875, 0.5492401123046875, 0.236968994140625, -0.512725830078125, -0.671234130859375, -0.78302001953125, -0.6757125854492188, -0.1368408203125, 0.7847137451171875, -0.98614501953125, -0.14693069458007812, 0.956878662109375, 0.23584747314453125, -2.267059326171875, 0.022705078125, 0.216156005859375, -0.100494384765625, -0.388580322265625, 1.8056640625, -0.6829910278320312, -0.056671142578125, -0.0787200927734375, 1.196197509765625, 0.882415771484375, -0.330902099609375, 0.25054931640625, -0.321380615234375, 0.232421875, -0.614776611328125, -1.6712646484375, 1.699066162109375, -0.38788604736328125, -0.5398406982421875, -0.476318359375, -0.4338493347167969, -0.7668838500976562, -0.672393798828125, 0.537322998046875, 0.006072998046875, 0.418426513671875, 0.420379638671875, -0.629852294921875, 0.65576171875, 0.0, -0.334991455078125, -1.681365966796875, 0.10638427734375, -0.60821533203125, 0.202117919921875, -0.6043529510498047, -0.00669097900390625, -1.75244140625, 0.734375, 0.277313232421875, 1.15960693359375, 0.184326171875, 0.8759918212890625, 0.10186767578125, 0.92376708984375, 0.852996826171875, 0.85552978515625, 0.298248291015625, -0.257965087890625, -1.076202392578125, -1.1214599609375, -0.10677337646484375, -0.2430419921875, -1.45440673828125, 0.733367919921875, -0.122772216796875, -1.055999755859375, -0.6193695068359375, 0.23485183715820312, -1.517303466796875, 1.461669921875, 0.33172607421875, -0.209442138671875, 0.5814018249511719, 0.23139190673828125, 1.4075469970703125, -0.6700477600097656, -0.9725341796875, 0.5207748413085938, 1.13616943359375, 0.07720947265625, 1.8268585205078125, -0.2540855407714844, 0.09578704833984375, 0.7930908203125, 1.538482666015625, 0.6137237548828125, -0.06988525390625, -0.43182373046875, -0.21875, -0.244140625, -1.38104248046875, 2.064178466796875, -0.0909881591796875, -0.5995941162109375, 0.685516357421875, 1.298431396484375, 0.3385009765625, 0.132781982421875, 1.1441650390625, 1.56658935546875, 0.80328369140625, -0.766693115234375, -0.083709716796875, 1.848358154296875, -1.814910888671875, -0.37176513671875, -2.2078857421875, 0.0863037109375, 0.3442649841308594, -1.285400390625, 0.0893707275390625, 0.835357666015625, -0.09018325805664062, 0.46286773681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000021.npy"}
{"epoch": 0.04397905759162304, "step": 22, "batch_size": 128, "mean": 0.14957471191883087, "std": 0.817800760269165, "min": -2.037445068359375, "p10": -0.73912353515625, "median": 0.103729248046875, "p90": 0.9982711791992187, "max": 2.3651885986328125, "pos_frac": 0.59375, "sample": [2.3651885986328125, 0.21574020385742188, -0.247772216796875, -0.6771240234375, 0.88397216796875, 0.2975006103515625, 0.40728759765625, 0.26091766357421875, 0.941680908203125, 1.105865478515625, 0.054962158203125, 0.5820999145507812, -0.735870361328125, -0.4830150604248047, -0.6615142822265625, -2.037445068359375, 0.8067626953125, 0.4727783203125, 0.34466552734375, 0.0305938720703125, -0.4474639892578125, -0.43837738037109375, -0.7427978515625, -0.84228515625, 0.54632568359375, -0.4237213134765625, -0.71295166015625, 0.162353515625, 0.2974395751953125, 0.063690185546875, 0.920379638671875, -0.5299072265625, 0.0, -1.239532470703125, -0.9636001586914062, 0.4600830078125, 0.0599517822265625, 0.102081298828125, 0.7198066711425781, -0.3416290283203125, 0.459747314453125, 0.68634033203125, -0.553070068359375, 1.25677490234375, -0.585357666015625, 0.105377197265625, -1.66748046875, -0.86236572265625, 0.82086181640625, -0.3562469482421875, -0.10631942749023438, 0.07897567749023438, -1.124603271484375, 0.63165283203125, 0.04034423828125, 0.4380645751953125, -0.0843353271484375, 0.696868896484375, 2.2635498046875, 1.73907470703125, -0.579925537109375, -0.7708740234375, 0.257171630859375, -0.339202880859375, 0.016729354858398438, 0.08885955810546875, 2.1090087890625, -0.427398681640625, -0.7089462280273438, -1.66278076171875, 0.271759033203125, -0.258575439453125, -0.9069671630859375, -0.52880859375, -1.25396728515625, -0.85968017578125, 0.2070770263671875, 0.13428497314453125, 0.41064453125, 0.9903564453125, -0.583587646484375, -0.4539642333984375, 0.126739501953125, 0.19554519653320312, 0.490814208984375, 0.697265625, 1.9437255859375, -0.3399658203125, 2.1102294921875, 0.1580810546875, 0.261016845703125, -0.717315673828125, 0.22492599487304688, 0.75677490234375, 0.0211181640625, -0.4865264892578125, -0.30938720703125, 0.19512939453125, 1.46038818359375, -0.62066650390625, 0.793975830078125, 0.3181934356689453, 2.34161376953125, 0.151275634765625, 0.6985321044921875, 0.93560791015625, 0.0242156982421875, 1.0167388916015625, 0.747711181640625, -0.1641998291015625, -0.09418869018554688, 0.850616455078125, -0.19520187377929688, -0.1346588134765625, -0.46637725830078125, -0.737548828125, 1.6571807861328125, 0.886688232421875, 0.695526123046875, 0.525848388671875, -0.2090911865234375, 0.955047607421875, 0.013881683349609375, 0.4184417724609375, -0.1266021728515625, 1.771270751953125, -0.007114410400390625, 0.7041015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000022.npy"}
{"epoch": 0.04607329842931937, "step": 23, "batch_size": 128, "mean": -0.024479106068611145, "std": 0.9381856918334961, "min": -2.6180419921875, "p10": -1.180103302001953, "median": -0.01568603515625, "p90": 0.9903991699218747, "max": 3.548583984375, "pos_frac": 0.484375, "sample": [-0.0466156005859375, 0.6349334716796875, 0.404541015625, -0.8055267333984375, 1.488372802734375, 0.2939453125, -0.8912353515625, 0.8285980224609375, -1.553985595703125, -1.7294921875, 2.3243408203125, 0.965057373046875, -1.2593307495117188, 0.671173095703125, -1.727081298828125, -0.27396392822265625, 0.81597900390625, -0.1924285888671875, 0.004680633544921875, -1.146148681640625, -0.291595458984375, -0.77215576171875, 0.480926513671875, -0.0943603515625, 0.2104034423828125, -0.7445068359375, 0.398651123046875, 1.997650146484375, -0.4692230224609375, -1.093170166015625, 0.0, 0.24549102783203125, 0.5314865112304688, -0.13550567626953125, -0.552581787109375, 0.5257110595703125, -0.0916900634765625, -0.7763671875, 0.252471923828125, -1.31280517578125, 0.265350341796875, -1.54425048828125, -0.328125, -0.7543373107910156, -0.52899169921875, -0.03131103515625, 0.192108154296875, -0.6945075988769531, 3.548583984375, 0.16986846923828125, -0.29901123046875, -0.06309890747070312, -0.6672286987304688, -0.3115386962890625, 0.30694580078125, 0.7791748046875, 0.29245758056640625, -0.5814971923828125, -0.068939208984375, -0.2026214599609375, -1.910400390625, 0.12981796264648438, -1.08013916015625, 0.3441162109375, 0.5166473388671875, -0.37781524658203125, 0.6978759765625, 1.719482421875, 0.055023193359375, -0.282867431640625, 0.5606613159179688, 0.24139022827148438, -1.560943603515625, -6.103515625e-05, 0.784423828125, -1.0211639404296875, 1.430389404296875, 0.5166683197021484, 1.435638427734375, 0.637725830078125, 0.21495819091796875, -0.10338211059570312, 0.42319488525390625, 0.22968292236328125, 0.435699462890625, -2.144073486328125, -1.4398345947265625, -0.776458740234375, 1.049530029296875, 0.477996826171875, 0.15282440185546875, -0.30948638916015625, 1.9541015625, -0.14116287231445312, -0.6306610107421875, -1.378875732421875, 1.49462890625, 1.175506591796875, -0.24259185791015625, -0.899169921875, 1.495147705078125, 0.7411651611328125, 0.47957611083984375, -0.3648529052734375, -0.2739105224609375, 0.29107666015625, -1.114288330078125, 0.43459320068359375, 0.2326507568359375, 1.83966064453125, 0.32439422607421875, -0.3448486328125, -0.14628982543945312, -0.3663787841796875, -0.34383392333984375, -0.89752197265625, -0.952667236328125, 0.257232666015625, 0.1873016357421875, -2.6180419921875, 0.24078369140625, -1.4948959350585938, 0.7455291748046875, 0.241668701171875, 0.877777099609375, -0.87689208984375, -0.5876655578613281, -0.11236572265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000023.npy"}
{"epoch": 0.048167539267015703, "step": 24, "batch_size": 128, "mean": 0.040722861886024475, "std": 0.8713540434837341, "min": -2.349029541015625, "p10": -1.0277008056640624, "median": 0.03304290771484375, "p90": 1.0302978515624999, "max": 2.806304931640625, "pos_frac": 0.53125, "sample": [0.4602813720703125, -0.2024078369140625, 0.14791107177734375, -0.4579315185546875, 0.72686767578125, -0.292724609375, 0.7637939453125, 1.6796875, 0.023529052734375, 1.062469482421875, 0.2185211181640625, 0.09234619140625, 0.040283203125, 0.02660369873046875, 0.359893798828125, -0.662200927734375, -0.23370361328125, 0.470123291015625, 1.1839599609375, 0.3580665588378906, 0.64727783203125, -0.738037109375, -0.475830078125, 0.0, -0.09331512451171875, 0.3419036865234375, -1.1876220703125, -0.65301513671875, 0.02947998046875, -0.980133056640625, -0.53582763671875, -0.274810791015625, 0.53607177734375, 1.51531982421875, 0.0328369140625, -0.459228515625, -1.206207275390625, 0.2276153564453125, -0.05503082275390625, -0.382232666015625, -0.945831298828125, 0.73138427734375, 0.18773269653320312, 0.903564453125, 1.398193359375, -0.48328399658203125, 0.0, -1.7738037109375, -0.4326171875, -0.41164398193359375, 0.817108154296875, -0.01909637451171875, 0.43144989013671875, 0.4528656005859375, -0.07275390625, 0.7905120849609375, -1.069549560546875, 0.1452178955078125, 1.016510009765625, 2.0350341796875, -2.349029541015625, 0.0924530029296875, 1.2789306640625, 0.18503952026367188, -1.46136474609375, 0.9585800170898438, 0.49289703369140625, -0.26889801025390625, -0.676025390625, -1.84686279296875, 0.184478759765625, -0.15802001953125, -0.92816162109375, -0.42431640625, -1.8364105224609375, 0.0332489013671875, 0.4828033447265625, -1.270782470703125, 2.806304931640625, -0.3534393310546875, 0.793304443359375, -1.2852783203125, 0.16473388671875, 0.30523681640625, -0.0694732666015625, 0.184478759765625, -1.009765625, -0.178558349609375, -0.4633941650390625, 0.19968414306640625, 0.626068115234375, 2.0482177734375, 0.825897216796875, -0.9124832153320312, -0.2579345703125, -0.08034324645996094, -0.561920166015625, 0.6536865234375, -0.448699951171875, 0.38995361328125, -1.403564453125, 0.5462646484375, 0.1127166748046875, 1.996124267578125, 0.4624176025390625, -0.1575927734375, 0.4744873046875, -0.39752197265625, -0.363525390625, -0.583282470703125, 1.5635986328125, -0.2629547119140625, 1.1117706298828125, 0.449951171875, 0.44808197021484375, 0.51129150390625, 0.8656005859375, -0.6792755126953125, -0.1131744384765625, -0.0403900146484375, 2.3616943359375, 0.1226806640625, -0.7247810363769531, -1.485870361328125, 0.09130859375, 0.0372314453125, -1.76654052734375, 0.41536712646484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000024.npy"}
{"epoch": 0.050261780104712044, "step": 25, "batch_size": 128, "mean": 0.10144302248954773, "std": 0.7068220973014832, "min": -2.388427734375, "p10": -0.6485237121582031, "median": 0.13120079040527344, "p90": 0.8563873291015625, "max": 2.61163330078125, "pos_frac": 0.5859375, "sample": [0.2508697509765625, 0.18145751953125, 0.414520263671875, 1.1058807373046875, -0.27642822265625, 1.13165283203125, -1.87884521484375, -0.4587249755859375, 2.61163330078125, -1.0889892578125, 0.0102081298828125, -0.28363037109375, 0.5623321533203125, 0.642913818359375, -0.0146026611328125, -0.701507568359375, -0.06447601318359375, 0.18050384521484375, -1.299346923828125, 0.035919189453125, 0.28246307373046875, -0.13199615478515625, -0.1949462890625, 0.15203857421875, -0.40509033203125, 0.14288330078125, -0.0888214111328125, 0.572845458984375, -0.15834808349609375, -1.027801513671875, 0.12856674194335938, 0.27881622314453125, 0.0, -0.41817474365234375, 1.062896728515625, -0.09258651733398438, 0.1338348388671875, -0.095306396484375, 0.859954833984375, 0.069732666015625, -0.89569091796875, 0.0, 1.28729248046875, 0.396728515625, 0.320404052734375, -0.5797119140625, 0.11968231201171875, 0.920166015625, 0.6009979248046875, -0.24578857421875, 0.26031494140625, -1.40655517578125, 1.171539306640625, 0.1097564697265625, 0.64898681640625, 0.1728363037109375, 0.04840850830078125, -0.159149169921875, 0.513763427734375, 0.68377685546875, 0.6715545654296875, -0.0672607421875, 0.69378662109375, -0.89239501953125, -0.641082763671875, -0.07966423034667969, 0.17987060546875, -0.2716827392578125, -1.112060546875, 1.215545654296875, 0.589569091796875, -0.292266845703125, -0.5053863525390625, 1.7330322265625, -0.48712158203125, 0.447174072265625, -0.5262069702148438, 0.7050819396972656, 0.536041259765625, -0.387542724609375, -0.582489013671875, -0.6658859252929688, 0.9781341552734375, 2.352569580078125, 0.2693634033203125, 0.320098876953125, 0.042079925537109375, 0.5222835540771484, 0.301727294921875, 0.451080322265625, 0.6331787109375, -0.119354248046875, 0.257659912109375, 0.717681884765625, 0.8548583984375, -0.22125244140625, 0.0926971435546875, 1.1431884765625, 0.364501953125, 0.179534912109375, 0.4213523864746094, -0.7696685791015625, 0.45062255859375, -0.024898529052734375, -2.388427734375, 0.343414306640625, 0.05706787109375, -0.60064697265625, -1.32366943359375, -0.599609375, 0.036834716796875, 0.4446258544921875, 0.30767822265625, 0.7658767700195312, -0.14581298828125, 0.540283203125, 0.354827880859375, 0.15313720703125, 0.8465118408203125, 0.24851226806640625, -0.20586013793945312, 0.18037796020507812, 0.7286376953125, -0.61444091796875, -0.5081787109375, -0.391845703125, -0.33782958984375, -0.4848670959472656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000025.npy"}
{"epoch": 0.05235602094240838, "step": 26, "batch_size": 128, "mean": 0.036755383014678955, "std": 0.7451546788215637, "min": -2.203125, "p10": -0.7004104614257812, "median": -0.001983642578125, "p90": 0.9023406982421874, "max": 4.09893798828125, "pos_frac": 0.484375, "sample": [-2.203125, -0.2016162872314453, 4.09893798828125, 0.1366119384765625, -0.160491943359375, -0.39286041259765625, 0.08070945739746094, 0.5077857971191406, 0.398956298828125, 0.12277793884277344, -0.2620277404785156, -0.726776123046875, 0.0, 0.03745269775390625, 0.411651611328125, 0.4207763671875, 0.242706298828125, 0.3793182373046875, 1.10430908203125, 1.186676025390625, 0.109588623046875, -0.4285125732421875, -0.49961090087890625, 1.1806640625, -0.150238037109375, -0.220062255859375, -0.37862396240234375, -0.079864501953125, 0.721466064453125, 0.035308837890625, -0.25334930419921875, 1.154022216796875, 0.21706390380859375, 0.57147216796875, 1.3314208984375, 1.13653564453125, 0.891693115234375, 1.4827880859375, -0.5944366455078125, -0.66851806640625, -0.598480224609375, -0.3601856231689453, 0.0627593994140625, -0.253814697265625, -0.941131591796875, -0.762603759765625, -0.32752227783203125, 0.1888580322265625, -0.4827117919921875, 0.043025970458984375, -1.036895751953125, -1.25299072265625, -0.3072052001953125, 0.23480224609375, 0.982421875, -0.33011627197265625, -0.6893463134765625, -1.0682373046875, -0.440673828125, -1.0936279296875, 0.2114715576171875, -0.726226806640625, -0.142547607421875, 0.39699554443359375, -0.599639892578125, 0.31235504150390625, -0.600189208984375, 0.7756500244140625, 2.70928955078125, 0.41567230224609375, -0.310028076171875, -0.07434463500976562, 0.58465576171875, -0.384185791015625, -0.45114898681640625, -0.00396728515625, -0.2258148193359375, 0.1417388916015625, 0.16766357421875, 0.30457305908203125, 1.02789306640625, 0.92718505859375, 0.4293060302734375, 0.265472412109375, 0.245025634765625, -0.04894828796386719, 0.5276947021484375, -1.173736572265625, -0.5596160888671875, -0.29766845703125, -0.09765625, 0.5977363586425781, 0.20912551879882812, 0.045196533203125, 0.32208251953125, -0.1727294921875, -0.268035888671875, 0.3107948303222656, -0.11512184143066406, 0.4376220703125, -0.18869781494140625, -0.5110511779785156, 0.7359619140625, -0.436279296875, 0.20590972900390625, -0.2171630859375, 0.4456329345703125, 0.0, -1.27947998046875, -0.17298126220703125, 0.50042724609375, -0.31758880615234375, 0.317108154296875, -0.5526962280273438, 0.7440261840820312, 0.474700927734375, 0.380340576171875, -0.68463134765625, -0.65020751953125, 0.0845184326171875, 0.26068115234375, -0.989013671875, -0.2474517822265625, 1.066864013671875, -0.76788330078125, -0.1453094482421875, -0.550201416015625, -0.221343994140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000026.npy"}
{"epoch": 0.05445026178010471, "step": 27, "batch_size": 128, "mean": 0.02639685571193695, "std": 0.9488853812217712, "min": -2.79168701171875, "p10": -0.9555923461914062, "median": 0.10643386840820312, "p90": 0.9759948730468749, "max": 2.79437255859375, "pos_frac": 0.5390625, "sample": [1.369232177734375, 0.786956787109375, -0.2733726501464844, 0.11972808837890625, 0.318389892578125, -2.6611328125, 0.13631439208984375, 0.0931396484375, 0.2362823486328125, 0.2155609130859375, 0.259185791015625, -0.8160400390625, 0.7258148193359375, -0.7772750854492188, -0.53314208984375, 0.39093017578125, -0.14337158203125, 0.318084716796875, 0.16949462890625, 1.442626953125, 0.5918121337890625, 0.33620452880859375, 0.9403076171875, -0.750244140625, 1.7315673828125, 2.79437255859375, 0.7879600524902344, 1.83319091796875, -0.532440185546875, 0.7072525024414062, 0.84344482421875, 0.029876708984375, 0.964385986328125, 0.35247802734375, 0.7913055419921875, -2.79168701171875, -0.2042694091796875, 1.10321044921875, -0.76080322265625, 0.5957794189453125, 2.70220947265625, 0.98748779296875, -0.895782470703125, 0.227691650390625, 0.509002685546875, 0.392333984375, -0.06769943237304688, -0.9579315185546875, 0.029937744140625, 0.7406845092773438, -0.065093994140625, 0.7574615478515625, -0.659454345703125, 0.47733306884765625, 0.8720703125, 1.6600341796875, 0.38681793212890625, -0.2071533203125, -0.12693023681640625, -0.1441650390625, -0.850830078125, -1.1607513427734375, -0.385711669921875, -1.6767578125, 0.3543853759765625, 0.1914520263671875, -0.3912811279296875, -0.14693069458007812, 2.76531982421875, -0.83837890625, -1.268798828125, -0.413421630859375, -0.804962158203125, 0.550048828125, 0.46600341796875, 1.1090774536132812, -1.765869140625, 0.3907470703125, -0.32952880859375, 1.4110107421875, -0.15634918212890625, 0.2225341796875, 0.882293701171875, -0.95458984375, -0.8273162841796875, 0.696014404296875, 0.9710693359375, -0.5374412536621094, 0.4693794250488281, 0.75616455078125, -0.438079833984375, -0.32108306884765625, -2.1619873046875, -0.838134765625, -0.5218353271484375, 0.362945556640625, -2.058502197265625, -0.21196937561035156, -0.30889892578125, -2.1151123046875, 0.43267822265625, -0.029092788696289062, -0.5728073120117188, -1.6868896484375, 0.16915512084960938, 0.55859375, -0.073455810546875, -0.256500244140625, -0.17411041259765625, 0.05237007141113281, 0.5750732421875, 0.4488506317138672, 0.3210620880126953, 0.589569091796875, -0.8885345458984375, -1.1666107177734375, 1.1608734130859375, 0.2932472229003906, -0.428131103515625, -0.760040283203125, -0.64801025390625, 0.24349212646484375, -0.923675537109375, -0.33221435546875, -0.203521728515625, 0.289764404296875, -1.0919952392578125, 0.005767822265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000027.npy"}
{"epoch": 0.05654450261780105, "step": 28, "batch_size": 128, "mean": 0.035564124584198, "std": 0.7673046588897705, "min": -1.8629302978515625, "p10": -0.8963592529296874, "median": 0.004486083984375, "p90": 1.1443222045898436, "max": 2.03936767578125, "pos_frac": 0.5078125, "sample": [0.159332275390625, 0.030517578125, -0.234649658203125, 1.72650146484375, -0.07726287841796875, -0.9504280090332031, -0.092315673828125, -0.007160186767578125, 0.5329399108886719, -0.407958984375, -0.21541786193847656, -0.342559814453125, 0.74993896484375, 0.008819580078125, 0.7869873046875, -0.890045166015625, -0.7835693359375, -0.9533538818359375, -0.73016357421875, 1.1980133056640625, 0.12138557434082031, 0.012615203857421875, -0.910308837890625, 0.663818359375, -0.5610198974609375, 0.12273406982421875, -1.44091796875, 0.2329254150390625, 0.19189453125, -0.686279296875, 0.0085601806640625, 0.4126129150390625, -1.43231201171875, 0.28570556640625, 0.4512290954589844, -1.5260162353515625, 0.7569122314453125, -0.71807861328125, 0.63238525390625, 1.4049072265625, -1.289398193359375, 1.101318359375, -0.5132293701171875, -0.67449951171875, 1.3773345947265625, 1.9759521484375, 0.532928466796875, 1.9642333984375, 0.19677734375, -0.3066558837890625, -0.2022705078125, 0.05890655517578125, 0.883392333984375, -1.03045654296875, -0.003143310546875, -1.0348358154296875, 0.15772247314453125, 0.304931640625, -0.16497802734375, 0.6715545654296875, -0.4290008544921875, -0.11277008056640625, 0.3409881591796875, -0.3177223205566406, 0.275421142578125, -0.383880615234375, -0.7782669067382812, 0.8455352783203125, 0.0439453125, 1.1259613037109375, -0.685791015625, -0.0223388671875, 0.3580322265625, -0.5669326782226562, -0.150543212890625, 0.059478759765625, -1.31512451171875, 0.2763786315917969, -0.25400543212890625, 1.2760009765625, -0.05014801025390625, 0.523681640625, 1.23419189453125, 0.048095703125, 0.457489013671875, 0.2629852294921875, 1.247344970703125, 0.406158447265625, -0.6354141235351562, 0.08815193176269531, -0.708160400390625, -0.423736572265625, 0.100860595703125, -0.3636474609375, -0.42557525634765625, 0.0004119873046875, 0.21282958984375, 0.3375701904296875, 1.371490478515625, -0.81292724609375, -0.08856201171875, -0.19501876831054688, -1.2311553955078125, -0.0662994384765625, -0.04144287109375, -0.890380859375, 1.5902481079101562, -0.35655975341796875, 0.71282958984375, 0.8367881774902344, 1.187164306640625, 0.01446533203125, -0.41748046875, -0.31114768981933594, 0.3070068359375, -0.2282562255859375, -0.2242889404296875, 0.7657546997070312, 2.03936767578125, 0.43958282470703125, -0.2861480712890625, -1.67529296875, -0.02850341796875, 0.14337158203125, 0.75396728515625, -0.32839202880859375, 0.0, -1.8629302978515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000028.npy"}
{"epoch": 0.05863874345549738, "step": 29, "batch_size": 128, "mean": 0.11245732009410858, "std": 0.9597467184066772, "min": -2.2301025390625, "p10": -0.9972198486328124, "median": 0.09664058685302734, "p90": 1.1191070556640625, "max": 3.1773681640625, "pos_frac": 0.546875, "sample": [0.14228057861328125, 1.133270263671875, 1.1048583984375, -0.8634033203125, -0.174560546875, -2.064453125, 0.811248779296875, 2.930419921875, -0.983856201171875, 1.500885009765625, -0.779266357421875, 1.23175048828125, 1.805084228515625, -0.3748779296875, 0.2590827941894531, 0.0, -0.53582763671875, 0.02309417724609375, -0.649017333984375, -0.07608413696289062, 0.824798583984375, 0.441986083984375, -0.5807647705078125, -0.6331787109375, 0.13037109375, 1.113037109375, -0.466461181640625, 0.2221832275390625, -0.28714752197265625, 0.9356155395507812, -0.5059051513671875, -2.026275634765625, 0.435455322265625, 2.9033203125, -0.7562026977539062, 0.93890380859375, 1.49835205078125, 0.010101318359375, 0.6389617919921875, -0.481689453125, 0.239501953125, 0.10461997985839844, 0.65924072265625, 0.7129135131835938, 0.177581787109375, 0.3931922912597656, 0.42791748046875, -0.227325439453125, 0.6545257568359375, -0.3862152099609375, 1.09991455078125, -0.6951904296875, 0.87481689453125, -0.483154296875, -0.0964202880859375, 3.1773681640625, 0.97015380859375, 0.06866455078125, -0.9442138671875, 1.0772705078125, 0.852386474609375, -0.465545654296875, 2.61444091796875, 1.22021484375, -0.6908798217773438, 0.31423187255859375, -1.517578125, 0.1643829345703125, 0.0, 0.5130996704101562, -0.332427978515625, -0.819244384765625, -0.431610107421875, -0.765106201171875, -0.0695037841796875, -0.3126678466796875, -1.05499267578125, 0.10029411315917969, -2.2301025390625, 0.883575439453125, -0.80169677734375, -0.3828125, -1.311981201171875, 1.3137130737304688, 0.945220947265625, -0.7226028442382812, 0.448638916015625, 0.371368408203125, -1.42791748046875, 0.998260498046875, 0.16144561767578125, 0.666748046875, 0.2010955810546875, -0.99041748046875, -1.136199951171875, 0.092987060546875, -0.568939208984375, -0.5059814453125, 0.13587188720703125, 0.3975067138671875, -0.0679931640625, -0.0515899658203125, 1.89031982421875, -1.4001693725585938, 0.6367034912109375, 1.048583984375, -1.3135528564453125, -1.013092041015625, -0.04229736328125, -0.4055023193359375, 0.617523193359375, -0.582733154296875, 0.7789306640625, 0.302703857421875, 0.9587249755859375, 0.03668212890625, -0.0713653564453125, -0.6521453857421875, 1.23760986328125, 0.9586181640625, 0.026952743530273438, 0.2938385009765625, 0.391265869140625, -0.18042755126953125, -1.620819091796875, -1.187652587890625, 0.778472900390625, 0.56842041015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000029.npy"}
{"epoch": 0.060732984293193716, "step": 30, "batch_size": 128, "mean": -0.013647839426994324, "std": 0.8196490406990051, "min": -2.788726806640625, "p10": -0.9040130615234374, "median": 0.0066127777099609375, "p90": 1.0236671447753904, "max": 1.87750244140625, "pos_frac": 0.5078125, "sample": [0.0184326171875, 1.58148193359375, -2.788726806640625, 0.775787353515625, -0.4774971008300781, 0.755340576171875, 0.7537841796875, 0.2265777587890625, -0.612884521484375, -0.5018692016601562, 0.001922607421875, 0.396759033203125, 0.43743896484375, 0.5895233154296875, 1.0894088745117188, -0.173004150390625, -1.413330078125, 0.1300048828125, -0.5190811157226562, -0.1192626953125, -1.739227294921875, 0.672607421875, -0.07763671875, -0.2310028076171875, 0.0432281494140625, -1.506805419921875, -0.02044677734375, 0.05022239685058594, -0.09162139892578125, -0.063079833984375, -0.787109375, -0.3420600891113281, 0.876617431640625, -0.9454345703125, -0.6830711364746094, -1.83575439453125, 0.291961669921875, 0.056430816650390625, 0.39754486083984375, -0.6326904296875, 0.206939697265625, 0.0982513427734375, -1.90667724609375, 1.19775390625, 1.78302001953125, 1.112945556640625, -0.41274261474609375, -0.886260986328125, 0.5779876708984375, -0.9522857666015625, 0.5666351318359375, 1.0997161865234375, 0.53118896484375, 0.2784271240234375, 1.7093505859375, -0.3625335693359375, -1.092071533203125, 0.5846176147460938, -0.22540283203125, -0.85504150390625, -0.8310546875, 1.87750244140625, -0.17181396484375, -0.697509765625, -0.57562255859375, -0.840179443359375, 0.016021728515625, -0.731689453125, -0.15555191040039062, 0.326080322265625, 0.1255207061767578, 0.9246826171875, -0.05068206787109375, -0.361541748046875, 0.57940673828125, 1.844970703125, -0.0400238037109375, -1.1549072265625, -0.08019256591796875, -0.4771270751953125, 1.0144729614257812, 0.663665771484375, -1.50164794921875, 0.392181396484375, 0.5769138336181641, 1.09588623046875, 1.368255615234375, -0.1122894287109375, 1.0451202392578125, -0.4915313720703125, 0.40834808349609375, 1.6048583984375, 0.1234588623046875, -0.17645263671875, 0.464111328125, -0.8463134765625, -0.1865997314453125, 0.062164306640625, -0.501861572265625, 0.05242919921875, -0.26513671875, 0.813568115234375, 0.3523101806640625, -0.5129318237304688, 0.6524658203125, -0.06053924560546875, 0.896820068359375, -0.781982421875, 0.2024688720703125, 0.045013427734375, -0.49346923828125, -0.8722686767578125, 0.678497314453125, 0.356292724609375, -0.2721900939941406, -0.04003143310546875, 0.3665313720703125, 0.1447906494140625, -0.5854072570800781, -1.734466552734375, -0.8421630859375, 0.25357818603515625, 0.3544921875, -1.596405029296875, 0.011302947998046875, -0.247772216796875, 0.2946319580078125, -0.08367919921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000030.npy"}
{"epoch": 0.06282722513089005, "step": 31, "batch_size": 128, "mean": 0.16990603506565094, "std": 0.7966576218605042, "min": -1.8201446533203125, "p10": -0.8127868652343749, "median": 0.0874176025390625, "p90": 1.2530792236328123, "max": 2.72894287109375, "pos_frac": 0.5546875, "sample": [-1.2391204833984375, -0.1527099609375, -0.00274658203125, 1.498626708984375, 0.124786376953125, 1.1877670288085938, 0.7473907470703125, 1.66436767578125, 0.9534912109375, 0.051971435546875, 2.72894287109375, -0.9877777099609375, 0.156982421875, 0.358428955078125, -0.033050537109375, 0.81854248046875, 0.2851295471191406, 0.38362884521484375, 0.3649749755859375, 0.16168212890625, 1.65167236328125, -0.18756103515625, -0.081146240234375, 1.455078125, 1.4583816528320312, 2.45294189453125, -1.372161865234375, 0.2873077392578125, -0.6113967895507812, 0.4066619873046875, 0.30255889892578125, 2.292755126953125, -0.5119552612304688, -0.19569778442382812, 1.563262939453125, 0.1712646484375, -0.0637969970703125, -0.6221923828125, -0.1046905517578125, 0.0, -0.113372802734375, -0.34995269775390625, 0.0, 0.41400146484375, 0.13525390625, -0.924713134765625, 1.1673736572265625, 0.8489494323730469, -0.966461181640625, -1.8201446533203125, 0.1851806640625, -0.66729736328125, 0.557647705078125, 0.9935302734375, 0.13283157348632812, -0.6546783447265625, 0.0107421875, 1.217010498046875, -0.243377685546875, -0.5890274047851562, 0.0, -0.4675750732421875, -1.031280517578125, -0.756683349609375, -0.149078369140625, 0.2431488037109375, -0.5631256103515625, -0.747467041015625, -0.8563232421875, 0.21581268310546875, 0.3465576171875, 0.867645263671875, -0.4641914367675781, -0.029052734375, -0.35479736328125, 0.612640380859375, -0.089080810546875, -0.14731597900390625, 0.759124755859375, 0.550537109375, -0.22758102416992188, -0.1155242919921875, -0.3641815185546875, -0.038059234619140625, 0.5230026245117188, 0.2283477783203125, 0.4697265625, -0.79412841796875, 0.33780670166015625, 0.80963134765625, -0.5335464477539062, 0.2525157928466797, 0.62664794921875, 1.234832763671875, 0.1156768798828125, -0.30300140380859375, -0.553741455078125, -0.7238616943359375, 0.0318450927734375, 0.0, -1.0074462890625, 0.03076171875, 1.3909759521484375, 1.020660400390625, 0.037654876708984375, 0.191741943359375, -0.8994140625, -0.47618675231933594, 1.754669189453125, 0.62786865234375, 1.352447509765625, -0.15077590942382812, 0.24425506591796875, 1.01129150390625, 1.1361083984375, -0.08647918701171875, 0.3418407440185547, -1.260467529296875, 1.295654296875, -0.02471923828125, 0.0792236328125, 0.8764190673828125, 0.20479774475097656, 0.095611572265625, 0.13574790954589844, -0.8851318359375, -0.9577484130859375, 0.05661773681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000031.npy"}
{"epoch": 0.06492146596858639, "step": 32, "batch_size": 128, "mean": 0.10885459184646606, "std": 0.7849098443984985, "min": -2.667694091796875, "p10": -0.7963718414306641, "median": 0.0701446533203125, "p90": 1.0097183227539062, "max": 2.528533935546875, "pos_frac": 0.546875, "sample": [-0.5426025390625, 0.017690658569335938, 0.394317626953125, 1.24603271484375, -0.6292266845703125, -0.8631820678710938, 0.002140045166015625, 0.0654449462890625, 0.3973236083984375, -0.2861328125, -0.303375244140625, -0.967010498046875, 1.11895751953125, 0.9834518432617188, 0.00054931640625, -0.7926445007324219, 0.396240234375, 1.325103759765625, 0.36968994140625, -0.614715576171875, 0.8184051513671875, 0.43035888671875, -0.597900390625, 0.9266510009765625, 1.0438232421875, 0.309295654296875, 0.579437255859375, -0.46138763427734375, 0.700103759765625, 0.4974212646484375, -0.6553955078125, 0.096038818359375, -0.8408203125, 0.4454345703125, -0.159210205078125, 0.564117431640625, -0.8050689697265625, -0.2861328125, 0.4150390625, 0.213348388671875, -0.3550567626953125, -0.06591796875, 0.0675048828125, -0.315338134765625, 1.2186279296875, -0.3554840087890625, 0.5648193359375, 0.5546493530273438, -0.529998779296875, 0.48468017578125, 0.784423828125, -0.70721435546875, -1.163055419921875, 0.9312744140625, -0.3787841796875, 0.037506103515625, 0.54644775390625, 0.68780517578125, 2.528533935546875, -1.18463134765625, -0.5845108032226562, -0.682373046875, -0.4583282470703125, -0.076019287109375, -0.0504150390625, 0.13336181640625, -1.19451904296875, 0.97247314453125, 0.73748779296875, 0.411163330078125, 1.7801513671875, 0.967864990234375, 0.42259979248046875, -2.667694091796875, -0.1748809814453125, -0.0678558349609375, -0.154693603515625, 1.2943115234375, -0.6823348999023438, 1.2388916015625, 0.9698486328125, -1.016357421875, -0.2600250244140625, -0.2461566925048828, 1.354644775390625, 0.9951019287109375, 0.5419464111328125, -0.371856689453125, -0.22787094116210938, -0.69488525390625, 0.3932647705078125, 0.072784423828125, -0.9522247314453125, 0.36029052734375, -0.2999267578125, -1.323699951171875, -0.9023284912109375, 0.7320556640625, 0.45501708984375, 0.742950439453125, -0.2386627197265625, 0.9654541015625, 0.208099365234375, 0.237548828125, 0.95220947265625, -0.323028564453125, -0.4163818359375, -0.02667999267578125, 0.5861663818359375, 0.17407608032226562, 2.219268798828125, 0.1903839111328125, -0.0113983154296875, 0.20208740234375, -0.050567626953125, -2.1947021484375, -0.2445068359375, 0.694061279296875, -0.294952392578125, -0.4212799072265625, -0.7129974365234375, 0.21496963500976562, -0.17547607421875, 1.4593505859375, 0.285186767578125, 1.04901123046875, -0.5875244140625, 0.80401611328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000032.npy"}
{"epoch": 0.06701570680628273, "step": 33, "batch_size": 128, "mean": 0.23624709248542786, "std": 1.0336040258407593, "min": -3.26220703125, "p10": -1.087493896484375, "median": 0.17665863037109375, "p90": 1.4590682983398433, "max": 3.22283935546875, "pos_frac": 0.6015625, "sample": [0.010890960693359375, 1.880157470703125, 0.28631591796875, -0.0880279541015625, -0.00577545166015625, 0.6082611083984375, 1.05780029296875, 0.350128173828125, 1.039794921875, 0.2366485595703125, 0.212677001953125, -0.08777618408203125, 3.008270263671875, -1.05279541015625, 0.0, 0.1282501220703125, 0.6219329833984375, -0.429595947265625, -1.3023300170898438, -1.18994140625, -0.408843994140625, -0.04952239990234375, 0.21616172790527344, -0.487152099609375, 0.4404296875, -0.2612152099609375, 0.16030120849609375, 1.060699462890625, 0.065155029296875, 0.61688232421875, 1.27972412109375, -1.63946533203125, 0.441619873046875, -1.6976318359375, -0.5156269073486328, 2.478729248046875, 0.54266357421875, 0.21807861328125, -0.30401611328125, -0.582305908203125, 0.6509017944335938, -1.025909423828125, 0.3464393615722656, 1.36920166015625, 0.52569580078125, 0.4242401123046875, 2.69610595703125, 0.46657562255859375, 0.0003662109375, -1.2487945556640625, 1.853302001953125, 0.154510498046875, -1.70831298828125, 0.0, -1.66839599609375, -0.05036163330078125, 0.17180252075195312, 1.925994873046875, 1.572021484375, 0.01361083984375, 2.40179443359375, -0.52520751953125, 0.9509162902832031, -0.0706329345703125, 0.1786041259765625, 0.04034423828125, -0.034778594970703125, 0.8092422485351562, -1.06048583984375, 1.9314727783203125, 0.99444580078125, 0.6720657348632812, -1.328277587890625, 1.307647705078125, 0.174713134765625, -0.30413818359375, -0.96954345703125, 0.76568603515625, -0.028289794921875, 0.57330322265625, -0.070892333984375, -0.4488372802734375, 3.22283935546875, -0.10865020751953125, 1.2618408203125, -1.213836669921875, 1.4106597900390625, 0.304443359375, 1.260101318359375, 1.9451141357421875, -0.5297393798828125, 0.21092987060546875, 0.218963623046875, 0.4044837951660156, 0.22817611694335938, -1.245086669921875, -0.779083251953125, 0.743408203125, -3.26220703125, -1.1505126953125, -0.1380157470703125, -0.6217269897460938, 0.5390777587890625, 0.636962890625, 0.9020309448242188, -0.7871170043945312, -0.1674957275390625, 0.9677581787109375, 2.26947021484375, -0.98095703125, -0.72613525390625, 0.9120635986328125, 0.65972900390625, 0.85467529296875, 0.48950958251953125, -0.542724609375, 0.8434600830078125, 0.041370391845703125, 1.333984375, 0.8771820068359375, 0.921600341796875, -0.18115234375, -1.2012939453125, -0.06671142578125, -0.976806640625, 0.16898727416992188, 0.16730880737304688, 1.835052490234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000033.npy"}
{"epoch": 0.06910994764397906, "step": 34, "batch_size": 128, "mean": 0.1527663916349411, "std": 1.0367671251296997, "min": -3.615478515625, "p10": -1.119409942626953, "median": 0.1704559326171875, "p90": 1.2991119384765624, "max": 2.87445068359375, "pos_frac": 0.609375, "sample": [-0.29660797119140625, 1.7885589599609375, 1.44140625, -0.00508880615234375, -0.05318450927734375, -0.874847412109375, -0.490081787109375, 0.3356666564941406, -0.0947265625, 0.94549560546875, -2.5635986328125, -2.351806640625, 2.26617431640625, 0.51800537109375, 0.218658447265625, 1.00982666015625, 2.745361328125, 0.088775634765625, 0.4997215270996094, -0.4769744873046875, 0.08478927612304688, -2.46173095703125, 0.05743408203125, 1.247161865234375, -0.21368408203125, 1.0147247314453125, 2.09619140625, 0.3920631408691406, 0.31294822692871094, 0.26806640625, 1.5615234375, 0.6044692993164062, 0.543121337890625, -1.1529541015625, -1.219940185546875, -1.1050338745117188, 1.08087158203125, 0.0346527099609375, -2.059234619140625, -0.668548583984375, 2.1907958984375, -0.43927001953125, 0.668975830078125, 2.87445068359375, -0.115203857421875, 0.5533447265625, -0.071197509765625, 0.1302947998046875, 0.815093994140625, -0.00933837890625, -0.4338836669921875, 0.4019775390625, -0.2374114990234375, 1.925048828125, -3.615478515625, 0.2141876220703125, -0.04715728759765625, 0.9825439453125, 1.7403564453125, -1.53875732421875, -0.0521240234375, 0.029224395751953125, 0.9994964599609375, -1.273590087890625, 1.105194091796875, 1.14434814453125, 0.0412139892578125, -0.026323318481445312, 0.40721893310546875, 1.0616455078125, 0.3107757568359375, 0.3814697265625, 0.06787109375, 0.2069091796875, 0.74761962890625, -0.16796112060546875, -0.24895095825195312, 0.034603118896484375, 0.183807373046875, -0.4834766387939453, 1.341644287109375, -0.3424072265625, -0.43585205078125, -1.0067596435546875, 0.1960906982421875, 0.74749755859375, 1.381378173828125, 1.223419189453125, 0.29367828369140625, 1.574249267578125, -0.32425689697265625, -2.6417236328125, 0.9046630859375, -0.65142822265625, 0.32281494140625, 0.01934814453125, -1.2671356201171875, -0.90936279296875, -1.17840576171875, 0.331756591796875, 0.04925537109375, 0.82513427734375, 0.8555908203125, 0.240814208984375, 1.2808837890625, -0.3456878662109375, -0.34405517578125, 0.10547637939453125, -0.05498504638671875, -0.147186279296875, 0.1571044921875, -1.206207275390625, 0.4100494384765625, -1.047515869140625, 0.416351318359375, 0.4013214111328125, -0.94903564453125, 0.2897796630859375, -0.713287353515625, 0.274993896484375, 1.1276397705078125, 0.08740234375, -0.1056671142578125, 0.462066650390625, -0.2149658203125, 0.40813446044921875, 0.98614501953125, 1.199371337890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000034.npy"}
{"epoch": 0.0712041884816754, "step": 35, "batch_size": 128, "mean": 0.06308749318122864, "std": 0.9069144129753113, "min": -2.2742919921875, "p10": -1.0167098999023436, "median": 0.04982757568359375, "p90": 1.3629089355468746, "max": 2.55035400390625, "pos_frac": 0.53125, "sample": [-0.21567535400390625, 0.6285400390625, -0.889007568359375, -1.567626953125, 0.1883392333984375, 0.995452880859375, 1.729736328125, 0.603668212890625, -1.52197265625, -1.2250823974609375, 1.4647216796875, 0.41994476318359375, 0.022064208984375, 0.5082855224609375, 0.45053863525390625, 0.31034088134765625, -1.00958251953125, -0.38934326171875, 0.0782928466796875, 0.12141799926757812, 0.45447540283203125, -2.23236083984375, 0.257568359375, 0.373443603515625, -0.30376434326171875, 1.069427490234375, 2.3851318359375, -1.0908432006835938, -0.518157958984375, -0.87933349609375, 0.7871322631835938, -0.6158294677734375, 1.6656494140625, 0.4414520263671875, 1.31927490234375, 2.55035400390625, -0.43341064453125, -0.287811279296875, 0.26134490966796875, 0.0, -0.28118896484375, 0.0687255859375, 0.5370712280273438, 0.408599853515625, 0.4424591064453125, -0.441162109375, 2.04327392578125, -1.0845489501953125, 0.4317626953125, -0.163726806640625, -0.2505340576171875, -0.1874847412109375, -0.649139404296875, -1.273895263671875, -0.015380859375, -0.9515609741210938, 0.574493408203125, -0.3096923828125, 0.15959930419921875, 0.10631942749023438, -0.18463134765625, 1.55731201171875, 1.6593017578125, -0.18506622314453125, -0.21710205078125, -0.131744384765625, 0.242156982421875, 0.48988914489746094, 0.90313720703125, -0.07789230346679688, -0.85345458984375, 0.48101806640625, -0.9168243408203125, 1.8767852783203125, -0.31472015380859375, 0.52679443359375, 0.0973968505859375, 0.2493896484375, 1.79486083984375, -0.5003662109375, 1.034332275390625, 0.09940338134765625, -2.03533935546875, 0.266937255859375, 1.2985992431640625, -0.7108917236328125, -0.17316055297851562, -0.9788818359375, 0.043487548828125, 0.5872116088867188, -2.2742919921875, -0.0886383056640625, 0.26296234130859375, 0.5782394409179688, 0.10940361022949219, -1.33538818359375, -0.1090087890625, 0.2740478515625, 0.324371337890625, 1.49212646484375, -1.534423828125, 1.621917724609375, 0.0, -0.4490203857421875, 0.3951416015625, -0.0220184326171875, 0.0561676025390625, 0.089599609375, 2.30645751953125, -0.23165130615234375, -0.798919677734375, 0.0382080078125, -0.220306396484375, -0.0013275146484375, 0.8693161010742188, 0.171142578125, 0.00066375732421875, -0.3080291748046875, -0.6165313720703125, -0.09520721435546875, 0.42291259765625, -1.432464599609375, -1.0333404541015625, -0.759368896484375, 0.58514404296875, -0.92193603515625, -0.7068328857421875, 0.4173583984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000035.npy"}
{"epoch": 0.07329842931937172, "step": 36, "batch_size": 128, "mean": 0.26891028881073, "std": 1.1109977960586548, "min": -2.95013427734375, "p10": -0.9624649047851562, "median": 0.1544342041015625, "p90": 1.3919853210449218, "max": 3.876312255859375, "pos_frac": 0.6015625, "sample": [0.364990234375, -1.72515869140625, 0.212005615234375, 0.4071044921875, -0.27996826171875, 0.879119873046875, 0.0482177734375, -0.124847412109375, 0.218109130859375, -0.6227607727050781, -1.0157470703125, 0.8705902099609375, 1.0550537109375, 0.591461181640625, -0.479766845703125, 0.0056915283203125, -0.15007781982421875, 1.87890625, 0.078277587890625, 0.40624237060546875, 0.5972900390625, -0.33319091796875, 0.09162139892578125, 0.0, 0.667999267578125, 1.16571044921875, -1.028778076171875, -0.35662841796875, -0.8825531005859375, -0.059917449951171875, 3.75091552734375, 0.0877685546875, -0.5755615234375, 1.184814453125, -0.995758056640625, 0.149993896484375, 0.7328262329101562, 0.130645751953125, 0.0036296844482421875, 0.4178619384765625, 3.869659423828125, 0.0605316162109375, -0.08055877685546875, 0.56390380859375, -0.1072845458984375, -0.304046630859375, 1.155731201171875, 2.3309478759765625, -0.823455810546875, -0.338226318359375, 0.648895263671875, 0.8624114990234375, 0.20062255859375, -0.43102264404296875, -0.9481964111328125, 1.0290985107421875, 0.046356201171875, 0.2530059814453125, 3.876312255859375, 2.74774169921875, 0.0, -1.131378173828125, 1.337188720703125, -0.5598373413085938, 1.0664215087890625, 1.10443115234375, -0.11232757568359375, 0.7701416015625, -0.3696937561035156, -0.8155975341796875, 0.4598388671875, 0.24322509765625, 0.5274658203125, 0.3700408935546875, -1.38665771484375, 0.6927223205566406, -0.385223388671875, 1.40625, -2.25091552734375, 0.4083251953125, 0.68414306640625, -0.127349853515625, 0.388275146484375, 1.3858718872070312, -2.95013427734375, -0.4423866271972656, -0.41473388671875, -0.6220550537109375, -0.38714599609375, 0.738861083984375, 2.1614990234375, 3.65740966796875, 0.199981689453125, 0.41934967041015625, -1.65838623046875, -1.51654052734375, -0.2868213653564453, 1.041473388671875, -1.3212432861328125, 0.3570098876953125, 1.07501220703125, 0.50531005859375, 0.749969482421875, 1.3099212646484375, -1.914337158203125, 1.4531936645507812, 1.11981201171875, -0.435028076171875, 0.15887451171875, -0.592620849609375, -0.1484088897705078, 0.553741455078125, -0.0869140625, 2.57989501953125, -0.2778587341308594, -0.391326904296875, 0.8436279296875, -0.089080810546875, 1.58642578125, 0.5953826904296875, 0.6148681640625, 1.83270263671875, -0.49359130859375, 0.13602256774902344, 0.0029296875, 0.36728668212890625, 0.032135009765625, -1.327484130859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000036.npy"}
{"epoch": 0.07539267015706806, "step": 37, "batch_size": 128, "mean": 0.24163639545440674, "std": 1.005858302116394, "min": -2.89013671875, "p10": -0.8249633789062499, "median": 0.13245010375976562, "p90": 1.3548522949218749, "max": 2.969970703125, "pos_frac": 0.5625, "sample": [-0.5782394409179688, -0.08795166015625, 0.980072021484375, 0.954376220703125, -0.051727294921875, -0.05472564697265625, 0.521484375, 0.19401931762695312, -0.001129150390625, -0.496856689453125, 2.2451171875, 0.4765758514404297, -0.41455078125, 1.1832351684570312, 1.1089019775390625, 1.60516357421875, 1.3037261962890625, 0.23960494995117188, 0.6461181640625, -2.520416259765625, 0.4042816162109375, -0.1434326171875, -0.08291244506835938, -0.5732421875, -0.44439697265625, 1.1824493408203125, -0.26819610595703125, 1.07183837890625, 0.13397216796875, 0.9146881103515625, 0.1676177978515625, -0.017566680908203125, -0.13843536376953125, 0.100616455078125, 0.55755615234375, -0.059967041015625, -1.211578369140625, 0.487518310546875, 1.24462890625, -0.333770751953125, -0.07957839965820312, -0.96905517578125, 0.085723876953125, 2.415496826171875, 1.5590667724609375, 1.92083740234375, -0.08305168151855469, 0.6446609497070312, 0.25638580322265625, -0.654449462890625, 1.3441619873046875, -0.6548614501953125, -0.6212158203125, 2.61151123046875, -0.9832763671875, -0.544647216796875, -0.2847747802734375, 1.2231292724609375, -0.423980712890625, -0.1273193359375, 0.731781005859375, 2.0938720703125, -0.00977325439453125, -0.801544189453125, -2.728363037109375, -0.42071533203125, -2.89013671875, 0.416473388671875, 0.18731689453125, 1.15863037109375, 0.740966796875, 0.18585205078125, 1.14215087890625, -0.076904296875, 0.67486572265625, 1.010772705078125, -0.7140045166015625, 0.828460693359375, -0.1973876953125, -0.078369140625, 0.256561279296875, 2.89959716796875, 0.16583251953125, -1.1197052001953125, 0.3094024658203125, 1.073822021484375, -0.113311767578125, 0.16412353515625, 1.3626708984375, 0.95379638671875, 0.008514404296875, 0.11114501953125, -0.331756591796875, -0.058319091796875, -0.5428237915039062, 0.13092803955078125, 0.930328369140625, 0.2984123229980469, -0.9564132690429688, 0.8240966796875, -0.27496337890625, -0.5974922180175781, -1.188629150390625, -0.325653076171875, 0.7257919311523438, 0.887420654296875, 2.969970703125, 0.06829833984375, 1.7838592529296875, 0.15034866333007812, -0.605560302734375, -0.03265380859375, 0.0902099609375, 0.019561767578125, 1.497283935546875, 0.46636962890625, 2.0916748046875, -1.2146453857421875, 0.47737884521484375, -0.9235076904296875, -0.739959716796875, -2.1616058349609375, -0.297332763671875, 1.128936767578125, 1.35150146484375, 1.2372283935546875, -0.879608154296875, 0.72515869140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000037.npy"}
{"epoch": 0.0774869109947644, "step": 38, "batch_size": 128, "mean": 0.18893998861312866, "std": 1.1249384880065918, "min": -4.023162841796875, "p10": -1.0304054260253905, "median": 0.08387565612792969, "p90": 1.552703857421875, "max": 3.975555419921875, "pos_frac": 0.5390625, "sample": [1.16021728515625, 0.9412002563476562, -0.29125022888183594, 1.1578903198242188, -0.222198486328125, -0.2626914978027344, 0.907867431640625, 2.38604736328125, 0.570953369140625, 1.657379150390625, -0.50469970703125, -0.1915435791015625, 1.2168426513671875, 1.157470703125, -0.586517333984375, 2.883270263671875, 0.132293701171875, -0.5754470825195312, -0.4673309326171875, 1.2410430908203125, -0.464599609375, 0.3292999267578125, 0.3460578918457031, 0.8227615356445312, 1.102691650390625, 0.475311279296875, -1.1435546875, -0.74993896484375, -0.122802734375, -0.42217254638671875, -0.180633544921875, 1.579833984375, -0.2432861328125, -1.27349853515625, 0.61566162109375, -0.2535858154296875, 0.9166717529296875, -0.4022026062011719, 0.6934814453125, -2.268707275390625, -1.0125045776367188, -0.18010711669921875, 0.8116455078125, -0.2656097412109375, 0.763580322265625, 0.0540313720703125, -1.36181640625, 1.16156005859375, 0.78411865234375, -0.1239471435546875, 0.65814208984375, 0.201629638671875, -0.70147705078125, -0.964752197265625, 0.4545936584472656, -0.8210601806640625, -1.101715087890625, -0.1974639892578125, -0.5073089599609375, -0.0357666015625, 0.6365432739257812, 0.019775390625, -0.595062255859375, -1.3907470703125, -0.0121002197265625, 0.08400344848632812, 0.7626953125, -0.5492401123046875, 1.7639007568359375, -0.16497802734375, -1.11309814453125, 0.08966064453125, 1.07708740234375, 0.214324951171875, -0.5184478759765625, -0.101318359375, 3.975555419921875, -0.10550689697265625, 0.1305389404296875, -4.023162841796875, 0.29979705810546875, -0.6925582885742188, 0.295196533203125, -0.31475830078125, 0.08374786376953125, 2.039093017578125, 1.20550537109375, -0.978668212890625, 0.010986328125, 0.43537139892578125, 0.7258224487304688, -0.513916015625, 3.285675048828125, 0.566650390625, 1.6144561767578125, 0.15915679931640625, 0.7490386962890625, 0.9653778076171875, 1.674102783203125, -0.5252685546875, -0.1388092041015625, 0.0, -0.659027099609375, -3.56158447265625, 0.19219970703125, 0.2715301513671875, 1.8555908203125, 0.6998443603515625, 1.158416748046875, 0.01324462890625, -0.462677001953125, -1.689422607421875, -0.4665985107421875, -0.30908203125, 1.8355712890625, -0.646484375, 2.279541015625, 1.1065139770507812, 0.324005126953125, 1.3883056640625, -1.4805908203125, -1.072174072265625, -1.247589111328125, 1.1373138427734375, 1.156890869140625, 1.54107666015625, -0.3708171844482422, 0.7805404663085938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000038.npy"}
{"epoch": 0.07958115183246073, "step": 39, "batch_size": 128, "mean": 0.41931575536727905, "std": 1.0076897144317627, "min": -2.84881591796875, "p10": -0.58245849609375, "median": 0.27050018310546875, "p90": 1.88004150390625, "max": 3.854095458984375, "pos_frac": 0.65625, "sample": [0.498565673828125, 0.0816650390625, -1.200347900390625, -0.21083831787109375, 2.17462158203125, -0.46673583984375, 0.69671630859375, 0.23383331298828125, -1.159576416015625, -0.55126953125, 0.68963623046875, 1.669036865234375, 0.840362548828125, 0.382415771484375, 0.03594970703125, 1.647216796875, 0.0058441162109375, 2.143157958984375, 1.1809234619140625, -0.854888916015625, -0.011566162109375, 0.488037109375, 0.40267181396484375, 0.27154541015625, 2.667633056640625, 1.9907989501953125, 0.38861846923828125, -0.4130096435546875, -0.1495513916015625, -0.64373779296875, 2.33941650390625, 0.30859375, 0.6270751953125, -0.2813262939453125, 2.6036376953125, -1.216522216796875, 0.14001846313476562, 0.38587188720703125, 0.399688720703125, 0.6852951049804688, 0.2264404296875, -0.031970977783203125, 0.851318359375, 0.5760231018066406, -0.47409820556640625, 0.6695556640625, 2.39837646484375, 0.2362651824951172, 1.5051422119140625, 0.95428466796875, -1.3107147216796875, -0.5105361938476562, -1.2222900390625, 1.311309814453125, 0.7728271484375, -0.21916961669921875, -0.591644287109375, 1.2686309814453125, 0.5582275390625, 0.216949462890625, -0.21929168701171875, -0.37066650390625, -0.21282958984375, 0.167755126953125, 1.0202713012695312, 0.881317138671875, -0.18158721923828125, 0.2208271026611328, 0.3033294677734375, -0.7310638427734375, -0.514892578125, 0.7564697265625, -0.31735992431640625, -2.84881591796875, 0.57489013671875, 1.307525634765625, -0.51446533203125, -0.0479736328125, 1.1747894287109375, -0.32696533203125, 1.035430908203125, 0.024776458740234375, -0.287750244140625, -0.1004638671875, 1.351104736328125, 0.2694549560546875, 3.854095458984375, 2.6197509765625, 0.4640655517578125, -0.578521728515625, 0.1964282989501953, -0.618621826171875, -0.2912139892578125, -0.6071968078613281, 0.0, -0.111480712890625, 0.11551284790039062, -1.031036376953125, 1.9202880859375, 1.9930419921875, 1.8420257568359375, 0.3096923828125, 0.262054443359375, 0.5857391357421875, 2.528564453125, 1.5778961181640625, -0.46563720703125, -0.4975128173828125, 1.86279296875, 0.053375244140625, 0.513580322265625, -0.3130035400390625, 0.17673301696777344, 0.6252098083496094, 0.07220458984375, 0.16300201416015625, 1.2752227783203125, 1.130218505859375, -0.4404449462890625, 0.06569862365722656, 0.5480728149414062, 0.36702728271484375, -0.253662109375, 3.19317626953125, 1.236175537109375, 0.2747802734375, 0.33261871337890625, 0.3034820556640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000039.npy"}
{"epoch": 0.08167539267015707, "step": 40, "batch_size": 128, "mean": 0.4909658432006836, "std": 1.1129258871078491, "min": -2.449554443359375, "p10": -0.5984657287597656, "median": 0.3589439392089844, "p90": 1.9077453613281248, "max": 5.05499267578125, "pos_frac": 0.6328125, "sample": [-0.25937652587890625, -2.449554443359375, -0.9347038269042969, 0.303680419921875, -0.43827247619628906, 1.3780975341796875, -1.56243896484375, 1.870361328125, 0.26909637451171875, 1.63836669921875, 0.69805908203125, 0.5837936401367188, 0.70013427734375, 1.42852783203125, 1.199188232421875, -0.21114730834960938, 0.3397674560546875, 2.035552978515625, -0.21457481384277344, 0.6165313720703125, -0.2370147705078125, 1.82366943359375, -0.18871688842773438, 0.003269195556640625, 0.6324958801269531, 2.865142822265625, -0.4860382080078125, 0.6299591064453125, -0.07953643798828125, 0.78912353515625, -0.5292472839355469, 0.462799072265625, -0.8796920776367188, -0.7411041259765625, 0.853515625, -0.284149169921875, 1.1630401611328125, -0.815185546875, 3.40057373046875, 1.315185546875, 5.05499267578125, 1.070648193359375, 0.66827392578125, -0.55401611328125, 0.1741790771484375, -0.287353515625, 0.4206085205078125, 0.2964630126953125, 2.495391845703125, 1.947906494140625, -0.16187286376953125, -0.3594970703125, 0.4319610595703125, 0.66558837890625, 1.317169189453125, 2.02447509765625, -1.0616455078125, 0.50323486328125, 2.62945556640625, -0.234466552734375, 1.0994873046875, -0.10021781921386719, 0.5090179443359375, -0.33935546875, 0.0574951171875, -0.543121337890625, -0.08984375, -0.1277008056640625, 0.388427734375, 0.9322853088378906, 0.5502853393554688, -0.30713653564453125, 0.5939788818359375, 2.2901611328125, 0.59454345703125, -0.8252601623535156, 1.2335205078125, 0.19018936157226562, 0.507568359375, -0.68438720703125, 0.0271148681640625, 1.517547607421875, -0.045928955078125, 0.2754173278808594, -0.489044189453125, 0.5015773773193359, -0.5915145874023438, 1.737579345703125, 0.1032867431640625, 0.3998260498046875, -0.24298095703125, 0.51593017578125, 0.13385009765625, 1.890533447265625, -1.307647705078125, -0.5120391845703125, 1.690765380859375, -0.16402435302734375, -0.084930419921875, 0.37812042236328125, 2.068878173828125, 1.0025177001953125, 0.743743896484375, 0.114654541015625, -0.515350341796875, 3.717864990234375, 0.289398193359375, 0.22916603088378906, -0.5765151977539062, -0.355682373046875, 0.3270263671875, 1.17828369140625, 1.0060882568359375, 0.25514984130859375, 0.781005859375, -1.42291259765625, -0.61468505859375, 0.9390106201171875, 1.20135498046875, 1.48907470703125, -0.20977783203125, -0.038860321044921875, -0.2233123779296875, 2.42510986328125, 0.6242904663085938, 2.9723968505859375, -1.487701416015625, 1.52935791015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000040.npy"}
{"epoch": 0.08376963350785341, "step": 41, "batch_size": 128, "mean": 0.25309881567955017, "std": 1.075480580329895, "min": -3.60650634765625, "p10": -0.9356719970703125, "median": 0.21606063842773438, "p90": 1.634396362304687, "max": 3.52313232421875, "pos_frac": 0.59375, "sample": [1.78997802734375, -0.2548675537109375, 0.220977783203125, -0.3736724853515625, -0.29852294921875, -1.416290283203125, 0.3873291015625, 0.9185028076171875, 0.4778594970703125, 0.5168304443359375, 0.8065261840820312, 2.586395263671875, -0.16126251220703125, -0.3318328857421875, -0.99774169921875, 2.054656982421875, 0.8042144775390625, -0.7261505126953125, 1.44378662109375, 0.8355712890625, 0.6294784545898438, 2.3001861572265625, -0.3175506591796875, 1.3121795654296875, 0.0259246826171875, -0.1439971923828125, 0.27801513671875, 0.339324951171875, 0.099853515625, -0.08850288391113281, 1.718841552734375, -2.310028076171875, 0.013889312744140625, 1.182159423828125, 0.2525634765625, 0.04767417907714844, 2.00286865234375, 1.59820556640625, 0.3562774658203125, 0.2137603759765625, 1.88067626953125, 0.6783676147460938, 1.2340850830078125, 1.191986083984375, 0.654937744140625, 0.270660400390625, 0.9110107421875, -0.0577850341796875, 0.573333740234375, -0.5205078125, -0.1789703369140625, -0.12285614013671875, -0.927276611328125, 0.20953369140625, 3.52313232421875, -0.342041015625, -0.2241363525390625, 0.9749755859375, -0.44134521484375, 1.9027252197265625, -0.3668670654296875, 0.447113037109375, 1.190826416015625, -1.0203857421875, -0.4486083984375, -0.8022918701171875, -0.1368560791015625, 0.26216888427734375, 0.0, -0.259674072265625, -0.4307403564453125, 0.15887451171875, -1.7738800048828125, -0.78546142578125, -3.60650634765625, -2.09881591796875, 0.44936370849609375, 0.22100067138671875, 1.21795654296875, -0.813140869140625, 0.48384857177734375, 0.34514617919921875, 0.39093017578125, 1.203521728515625, 1.062591552734375, 1.215850830078125, 1.1952743530273438, 0.7021484375, 2.8126678466796875, -0.17137908935546875, -2.02435302734375, 0.2070159912109375, 0.9856338500976562, -0.3224029541015625, -0.64306640625, 2.107696533203125, -0.95526123046875, -0.899658203125, 0.1693878173828125, -1.3751373291015625, 1.17645263671875, 0.910858154296875, -0.799285888671875, 0.8601150512695312, -0.8368072509765625, 0.20314788818359375, 0.21836090087890625, -0.1913604736328125, 0.2017822265625, -0.16573333740234375, 1.1971702575683594, 0.52655029296875, 1.090301513671875, -0.41400146484375, -0.628204345703125, -0.199737548828125, -0.2398681640625, -0.572509765625, 0.034423828125, -1.179107666015625, -2.0330810546875, 0.373687744140625, 1.7802734375, 1.094696044921875, 1.7479095458984375, 1.4664306640625, 0.9019622802734375, -1.0062255859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000041.npy"}
{"epoch": 0.08586387434554973, "step": 42, "batch_size": 128, "mean": 0.4242536127567291, "std": 1.1977192163467407, "min": -4.795654296875, "p10": -0.5523996353149413, "median": 0.2562370300292969, "p90": 1.9648986816406249, "max": 3.67913818359375, "pos_frac": 0.65625, "sample": [-0.6137828826904297, -0.0999298095703125, 1.9560546875, 3.17352294921875, -0.24957275390625, 0.97528076171875, -0.175262451171875, 0.797119140625, 0.031219482421875, -0.072540283203125, 1.0565185546875, 3.67913818359375, 0.4639892578125, 0.450469970703125, -0.843994140625, -0.41514015197753906, 1.47344970703125, 1.1700439453125, 0.43842315673828125, 2.861602783203125, -0.0850982666015625, -0.6542892456054688, 0.00384521484375, 0.26904296875, 2.33612060546875, 0.0377044677734375, 1.119384765625, 0.4224815368652344, 1.5242919921875, -0.01739501953125, -0.526092529296875, 1.0404224395751953, 1.08544921875, -0.22894287109375, 3.529388427734375, 0.996917724609375, 1.0947265625, 0.596099853515625, -0.7381134033203125, 0.3193511962890625, -0.5215835571289062, 0.2432384490966797, 1.2677001953125, 0.0582733154296875, 2.361572265625, -0.1640472412109375, 1.1930923461914062, 0.27685546875, -0.5152587890625, -0.13134002685546875, 0.45017242431640625, 0.18979263305664062, 1.083099365234375, -0.2755546569824219, -0.20867156982421875, 0.23839569091796875, -0.50543212890625, 1.523681640625, 1.23358154296875, 0.3525238037109375, 0.02545166015625, -0.17584228515625, 0.0937347412109375, -1.8526992797851562, 0.884002685546875, 0.84698486328125, 0.345733642578125, 0.24343109130859375, -0.4977836608886719, 0.06348419189453125, -0.38616943359375, 0.004100799560546875, -0.257232666015625, 1.191741943359375, -1.226806640625, 0.046108245849609375, -3.114776611328125, -0.45282745361328125, 1.7958984375, -0.24491500854492188, 0.1009674072265625, 1.52099609375, 0.3425140380859375, -0.5154190063476562, 2.523468017578125, -0.1213836669921875, 2.615447998046875, 0.3689117431640625, 0.2832069396972656, 0.720306396484375, 1.8216552734375, -0.6224365234375, 0.980224609375, 0.44109344482421875, 1.743927001953125, 0.8283538818359375, -1.20562744140625, 2.5977783203125, 0.20610809326171875, 1.024139404296875, 0.67059326171875, 0.7699432373046875, 0.0, 0.220794677734375, -1.39727783203125, -0.21342849731445312, -0.129913330078125, 0.3289642333984375, -0.4537086486816406, 1.764923095703125, 2.023681640625, -1.19659423828125, 1.03997802734375, 0.03587532043457031, 0.42156982421875, 1.847686767578125, 1.98553466796875, 0.1859130859375, 2.1162338256835938, 0.017578125, 0.4539794921875, 2.611114501953125, -4.795654296875, -0.2417449951171875, -2.830657958984375, -0.2393035888671875, -0.208648681640625, 0.19918060302734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000042.npy"}
{"epoch": 0.08795811518324607, "step": 43, "batch_size": 128, "mean": 0.4787002205848694, "std": 1.1866506338119507, "min": -2.895172119140625, "p10": -0.7387756347656249, "median": 0.3057746887207031, "p90": 1.9831237792968746, "max": 4.67657470703125, "pos_frac": 0.65625, "sample": [0.47869873046875, -0.1500244140625, 1.749969482421875, -1.64593505859375, 0.3332366943359375, 0.7155303955078125, 3.10333251953125, -0.06941986083984375, -0.0407867431640625, 0.20755767822265625, 0.37267303466796875, -0.0420074462890625, 0.09405517578125, 2.248992919921875, -0.4861602783203125, -0.09429931640625, -0.714111328125, -0.489349365234375, 1.6394195556640625, 0.3162841796875, 1.4891357421875, 0.5293807983398438, 2.070068359375, 1.8361968994140625, -0.594329833984375, -0.8556976318359375, -0.7044677734375, 2.8923187255859375, 0.07196044921875, 2.280059814453125, 0.1949462890625, 0.8334732055664062, 0.44708251953125, -2.895172119140625, 2.249786376953125, 0.5777683258056641, -0.3762054443359375, 1.1441497802734375, 0.09462356567382812, -0.65679931640625, -1.22406005859375, 0.308837890625, 0.23711776733398438, 1.225189208984375, 0.2830696105957031, 0.21251296997070312, 0.1757659912109375, -0.39853668212890625, -1.23687744140625, -1.8421783447265625, -0.404052734375, 0.6754684448242188, -0.010362625122070312, 0.7228546142578125, 2.8560791015625, 0.7236328125, 0.30975341796875, -0.30181884765625, 0.053680419921875, -1.319091796875, 0.865631103515625, -0.157501220703125, 1.5738525390625, -0.1378936767578125, 4.62213134765625, 1.31524658203125, 0.13824462890625, -1.1851806640625, 0.871063232421875, 2.0970306396484375, 0.277862548828125, 1.94586181640625, 1.857513427734375, 0.7221221923828125, -1.4900665283203125, -0.282867431640625, -0.701751708984375, -0.96832275390625, 0.30271148681640625, 1.17236328125, 0.21677398681640625, 1.1900634765625, -0.1669921875, -0.5414886474609375, 1.3074493408203125, -0.014495849609375, 1.2871856689453125, -0.3450775146484375, -0.412078857421875, 0.366790771484375, 1.870635986328125, 0.144287109375, -1.48834228515625, 0.8733291625976562, 1.4097900390625, 0.49570465087890625, 0.7770462036132812, 4.67657470703125, -0.79632568359375, -0.01822662353515625, 0.34832763671875, 1.0839004516601562, 0.39447021484375, -0.4347038269042969, 0.8873291015625, 0.833953857421875, 1.16162109375, -0.096923828125, 1.82061767578125, 3.35272216796875, -0.099822998046875, 0.05389404296875, 0.6090164184570312, -1.6735610961914062, 1.612884521484375, 1.0259857177734375, 2.24761962890625, 0.4876556396484375, 0.635162353515625, 1.5029296875, 0.22607421875, -0.2730712890625, 0.11980056762695312, 2.110076904296875, 0.464080810546875, -0.3478584289550781, 0.19287109375, 0.15500259399414062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000043.npy"}
{"epoch": 0.09005235602094241, "step": 44, "batch_size": 128, "mean": 0.3798980116844177, "std": 1.293203592300415, "min": -3.57598876953125, "p10": -1.0995758056640625, "median": 0.2593841552734375, "p90": 2.1605468749999996, "max": 3.719635009765625, "pos_frac": 0.578125, "sample": [0.282745361328125, -0.111419677734375, 1.7984619140625, 0.6755294799804688, -0.5103759765625, 0.91278076171875, 0.41373443603515625, 0.053470611572265625, 0.84722900390625, 0.000949859619140625, 1.044830322265625, 0.2486572265625, 0.7097625732421875, -0.1878814697265625, 2.51068115234375, 0.94732666015625, -0.340972900390625, -0.013607025146484375, -3.57598876953125, 1.893829345703125, 1.6972808837890625, 1.60693359375, -0.1166534423828125, 1.0012855529785156, 0.94354248046875, 0.721038818359375, 2.6898193359375, 0.28809356689453125, -1.094696044921875, -0.414703369140625, 1.14605712890625, -0.8929595947265625, -0.235076904296875, 0.3173980712890625, 0.08018112182617188, -1.3572998046875, 0.845458984375, -0.5338649749755859, 1.960540771484375, 0.809173583984375, 3.44775390625, -0.23374557495117188, -0.7613372802734375, 2.603240966796875, 1.051788330078125, 3.490081787109375, 0.9022216796875, 2.593048095703125, 1.5768890380859375, -0.0250244140625, -0.3949298858642578, 1.5526123046875, -0.738037109375, -0.7611083984375, -0.116607666015625, 0.28173828125, -1.190521240234375, 2.443603515625, 0.4651298522949219, 1.201324462890625, 0.860565185546875, 0.34124755859375, 0.6110877990722656, -0.41680908203125, -0.286956787109375, 0.0906219482421875, -1.323486328125, -2.258331298828125, -0.01263427734375, 3.52593994140625, -1.1109619140625, -0.0686798095703125, -0.67230224609375, 0.263885498046875, 1.148956298828125, 2.4427947998046875, 1.0677490234375, -2.02923583984375, -0.16130828857421875, -1.512725830078125, 3.110870361328125, 0.1178436279296875, 1.394073486328125, 0.30011749267578125, -0.23223876953125, -0.8840713500976562, 0.550506591796875, -1.51568603515625, 2.6839599609375, 0.1881103515625, -1.484954833984375, 2.297821044921875, 0.0528564453125, -0.181549072265625, -0.37603759765625, 1.48748779296875, -0.9310302734375, 0.2548828125, 0.35675048828125, -0.0234832763671875, 1.25244140625, 0.07487297058105469, -1.0839996337890625, -0.32122802734375, -0.02281951904296875, -0.41375732421875, -1.003448486328125, 2.101715087890625, -0.07122421264648438, -1.15673828125, 3.719635009765625, 1.047149658203125, 0.6462326049804688, 1.61810302734375, -0.0362396240234375, -0.214630126953125, 0.667755126953125, -0.3339996337890625, -3.22314453125, -0.27320098876953125, -1.66986083984375, 0.5070228576660156, 0.7301483154296875, 0.7967071533203125, 1.210906982421875, -0.7451553344726562, 0.9755783081054688, 1.733062744140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000044.npy"}
{"epoch": 0.09214659685863874, "step": 45, "batch_size": 128, "mean": 0.5630996227264404, "std": 1.4184340238571167, "min": -4.948944091796875, "p10": -0.7666900634765624, "median": 0.43250274658203125, "p90": 2.475640869140625, "max": 5.538543701171875, "pos_frac": 0.6640625, "sample": [0.633758544921875, -0.4927825927734375, 1.528839111328125, -0.1983795166015625, -0.18673324584960938, 0.917144775390625, 0.28467369079589844, 1.082550048828125, 0.8869819641113281, 2.3416748046875, 1.7703857421875, 2.49798583984375, 2.009796142578125, -0.10427093505859375, -0.2027740478515625, 0.6877059936523438, -0.07311820983886719, 1.64678955078125, 0.0, -1.296142578125, 2.655120849609375, 0.4893951416015625, -0.3987579345703125, 0.6772613525390625, -1.75286865234375, 3.115875244140625, -0.4801025390625, 0.0401611328125, 0.4028778076171875, -0.04225921630859375, 0.363037109375, -0.17151641845703125, -1.446044921875, -0.5985107421875, -0.10350799560546875, 2.184814453125, 0.571929931640625, -0.0062103271484375, 2.98211669921875, 2.124420166015625, 3.54058837890625, -1.175628662109375, 0.518096923828125, -0.196014404296875, 3.512451171875, -0.007381439208984375, 1.230072021484375, 0.15242767333984375, -0.720703125, 0.22223663330078125, 0.0861053466796875, 0.25469970703125, -0.891693115234375, 0.8421630859375, 0.07682037353515625, 0.5605621337890625, 0.2600860595703125, 0.716400146484375, -0.44956207275390625, -4.948944091796875, -3.8345947265625, 0.9742431640625, -0.314453125, 0.16717529296875, -1.34100341796875, 1.0725860595703125, 1.649169921875, 0.162200927734375, 1.66046142578125, 0.2960491180419922, 0.316619873046875, 3.872711181640625, -0.239593505859375, 0.6281509399414062, 3.1573486328125, 0.640625, 0.82830810546875, 1.7490234375, 1.247039794921875, 1.149383544921875, 1.025177001953125, 3.23748779296875, 0.8680057525634766, 2.642822265625, 0.4163970947265625, -1.358612060546875, 1.1395263671875, -0.1884613037109375, -0.6901321411132812, -0.832977294921875, 0.084197998046875, 2.5184326171875, -0.2115936279296875, -1.6231231689453125, 3.58184814453125, 0.9094772338867188, -1.03302001953125, -0.198089599609375, 0.4986572265625, 0.063995361328125, -0.59185791015625, 0.6574554443359375, 0.05742645263671875, -0.5642547607421875, 0.1771240234375, 1.56622314453125, 0.8206634521484375, -0.4134063720703125, 1.0915298461914062, 0.6968154907226562, 0.5546798706054688, 0.4486083984375, 0.87957763671875, 1.0009765625, 0.357696533203125, 1.481170654296875, 2.0608367919921875, 0.578704833984375, 5.538543701171875, 1.929046630859375, 0.343170166015625, -0.73828125, 1.1555938720703125, 0.946502685546875, 2.466064453125, -0.04254913330078125, -2.6024169921875, -0.3944549560546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000045.npy"}
{"epoch": 0.09424083769633508, "step": 46, "batch_size": 128, "mean": 0.6414192914962769, "std": 1.4651856422424316, "min": -4.146087646484375, "p10": -0.8246063232421875, "median": 0.40668487548828125, "p90": 2.6271087646484372, "max": 5.79779052734375, "pos_frac": 0.625, "sample": [2.8214111328125, 5.79779052734375, 2.6610107421875, -0.34363555908203125, -0.4633331298828125, 0.893890380859375, 2.03729248046875, 3.9619140625, -0.6169891357421875, -0.18628692626953125, -1.7369384765625, 0.2441864013671875, 2.846435546875, -1.03857421875, 0.1082763671875, -1.010955810546875, -0.03179931640625, -0.07007408142089844, -0.2716217041015625, 0.3458251953125, 0.258941650390625, 1.644378662109375, -0.31427001953125, 0.89215087890625, -0.897796630859375, -0.6864471435546875, -2.342132568359375, -0.5249671936035156, 3.70758056640625, -0.5198173522949219, 0.06172943115234375, 0.616119384765625, 1.1610870361328125, -0.0088348388671875, 1.141632080078125, 2.07806396484375, 2.20361328125, 1.192169189453125, -1.6650543212890625, 1.1402587890625, 0.827606201171875, 0.92083740234375, 1.7984619140625, 1.30035400390625, 0.607177734375, -1.92596435546875, 1.212493896484375, 0.61346435546875, 0.9393081665039062, -0.29030609130859375, 2.37005615234375, 2.7106170654296875, 0.7946243286132812, 0.80804443359375, -0.4693756103515625, 1.32763671875, -1.434326171875, 4.1744384765625, -0.24324798583984375, 2.18536376953125, 0.5273971557617188, 0.32900428771972656, -0.841888427734375, -0.5867156982421875, 2.31396484375, -0.6183853149414062, 0.1463756561279297, -1.088714599609375, -0.6666717529296875, 2.46246337890625, -0.5164794921875, 0.13936614990234375, 2.4091796875, -0.101318359375, -1.2764129638671875, -0.058074951171875, -0.2551155090332031, -0.7208786010742188, 0.1883716583251953, 1.35479736328125, 0.5184860229492188, 2.428070068359375, 2.1255035400390625, -2.20513916015625, 1.279815673828125, 0.26107025146484375, -4.146087646484375, 0.1979217529296875, 0.946197509765625, -0.4080810546875, -0.6844024658203125, 0.919464111328125, 1.8885498046875, 0.49561309814453125, -0.26908111572265625, 2.772857666015625, 0.6456756591796875, -0.7691650390625, -0.007415771484375, 0.4361114501953125, -0.2947254180908203, 2.1112060546875, 2.4194793701171875, -0.6039276123046875, 3.56365966796875, 0.8656158447265625, 1.069976806640625, 1.0753631591796875, 0.83306884765625, 0.9910888671875, 3.154571533203125, -0.378173828125, 0.16384124755859375, 0.37725830078125, -0.11927223205566406, 1.77545166015625, 0.317779541015625, 0.199005126953125, 1.60992431640625, 3.05950927734375, 0.0, 1.799957275390625, 0.6420822143554688, -0.81719970703125, -0.07769012451171875, 2.83038330078125, 0.0411376953125, 2.612579345703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000046.npy"}
{"epoch": 0.09633507853403141, "step": 47, "batch_size": 128, "mean": 0.3395553231239319, "std": 1.317376971244812, "min": -3.9481201171875, "p10": -0.9782844543457031, "median": 0.3051605224609375, "p90": 1.9610519409179685, "max": 4.0076904296875, "pos_frac": 0.6328125, "sample": [0.119873046875, 0.08647346496582031, 1.1148757934570312, -0.0421295166015625, 3.66796875, -2.35015869140625, -2.1162109375, 0.2563323974609375, 2.1877593994140625, -0.511749267578125, 0.7025146484375, 0.7452239990234375, -0.662384033203125, -0.3893280029296875, -1.373291015625, -0.8643798828125, 0.50390625, -0.0398101806640625, 1.545196533203125, -0.900360107421875, 2.912567138671875, -0.61456298828125, 0.20245361328125, 0.5001068115234375, -0.378814697265625, 1.710845947265625, 0.543548583984375, 1.00543212890625, -3.9481201171875, -1.341461181640625, 1.12615966796875, 0.0, 2.195770263671875, -0.9360198974609375, -0.832366943359375, 1.34564208984375, 2.40386962890625, 1.1851806640625, 0.0855560302734375, 0.42418861389160156, -0.461151123046875, 0.5568695068359375, 2.025299072265625, 1.402374267578125, 1.0301666259765625, 1.227081298828125, -0.686187744140625, 0.22119140625, 0.61395263671875, 0.07248687744140625, 0.66632080078125, 0.779144287109375, 4.0076904296875, -0.4278717041015625, -2.227935791015625, 2.95477294921875, -2.818359375, 2.984710693359375, 1.405670166015625, 1.0758209228515625, -2.587310791015625, 1.781768798828125, -0.9643478393554688, 0.90325927734375, 1.0434112548828125, -0.7338485717773438, 0.17034912109375, 0.4686279296875, 0.4382743835449219, 1.713592529296875, -0.2557373046875, 0.2943115234375, -0.8563385009765625, 0.8209381103515625, 1.2805938720703125, 1.5589981079101562, 0.0, 1.9335174560546875, 2.73553466796875, 0.83551025390625, 0.4704437255859375, 0.054595947265625, -0.2203826904296875, -1.345916748046875, 1.859954833984375, 0.226715087890625, 0.08658599853515625, -0.344696044921875, 0.307464599609375, 1.26031494140625, 0.48812103271484375, -0.46051025390625, -0.3679389953613281, 2.26019287109375, 0.37517547607421875, -0.17755126953125, 2.483428955078125, -0.0011444091796875, 1.2452392578125, 0.9182662963867188, -0.825927734375, 0.9617156982421875, -0.052276611328125, 0.98883056640625, 0.3028564453125, -2.081695556640625, 0.09402847290039062, -0.681640625, 0.015899658203125, -0.43973541259765625, -0.100189208984375, 0.2557220458984375, 1.45233154296875, 0.46456146240234375, 0.04864501953125, 0.664794921875, -1.01080322265625, 0.9164047241210938, -0.7311859130859375, -0.368682861328125, -1.1808319091796875, 2.2379608154296875, 0.6891250610351562, 0.758575439453125, 1.16741943359375, -3.535888671875, -0.24884033203125, 0.33010101318359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000047.npy"}
{"epoch": 0.09842931937172775, "step": 48, "batch_size": 128, "mean": 0.6258990168571472, "std": 1.6050242185592651, "min": -3.175689697265625, "p10": -0.9870452880859375, "median": 0.4616966247558594, "p90": 2.5901489257812496, "max": 8.5194091796875, "pos_frac": 0.609375, "sample": [-0.0748443603515625, -0.55810546875, 0.6937255859375, 0.8639755249023438, 1.045623779296875, 1.4864501953125, -2.3546142578125, 3.5548095703125, 1.717864990234375, -0.17726898193359375, -0.27753448486328125, 0.07563018798828125, -0.0777130126953125, -2.3846435546875, -0.4614715576171875, 5.9676513671875, -1.00274658203125, -1.8630828857421875, -2.0425567626953125, 0.024078369140625, 2.96209716796875, 0.10784912109375, 0.453948974609375, -0.836883544921875, 1.5295867919921875, 0.406005859375, -1.8145599365234375, 1.4947891235351562, -1.02264404296875, 0.5531463623046875, 0.46944427490234375, 1.974822998046875, 1.6654052734375, 1.0130615234375, 2.1192626953125, 0.9650993347167969, -0.05672454833984375, 4.0479736328125, 1.20452880859375, 3.1452789306640625, -0.5205841064453125, 3.070404052734375, -0.0433502197265625, 0.3311042785644531, 0.8750686645507812, -0.0704803466796875, -3.175689697265625, 0.152587890625, -0.980316162109375, 0.47595977783203125, 2.6640625, 0.6511077880859375, -0.205322265625, -0.6311874389648438, -1.202362060546875, 1.775238037109375, 1.0228652954101562, 0.21143341064453125, -0.15912246704101562, 0.88714599609375, -1.5006103515625, 0.747772216796875, 0.305267333984375, 1.6204376220703125, 0.070526123046875, -0.64947509765625, -0.45654296875, -0.20513916015625, 2.333526611328125, 1.169525146484375, -2.55010986328125, 0.77374267578125, 0.0, -0.0133209228515625, 1.438446044921875, 0.15202903747558594, -0.2451171875, 0.6574287414550781, 0.302886962890625, 2.9957275390625, -0.1974334716796875, 2.736541748046875, 0.60003662109375, 1.93212890625, -0.121063232421875, 3.852386474609375, 2.053802490234375, 0.743316650390625, -0.8139533996582031, -0.580810546875, -0.366455078125, -0.1962127685546875, 1.5601348876953125, -0.720062255859375, 2.5584716796875, 3.033203125, -1.153472900390625, 0.7145843505859375, -0.614105224609375, 0.7333450317382812, -0.37436676025390625, 2.274688720703125, 0.0, 0.55084228515625, 2.087310791015625, 0.227081298828125, 8.5194091796875, 0.114959716796875, -1.6064453125, 0.864654541015625, 0.8912353515625, 0.9149703979492188, 0.66265869140625, 1.9227294921875, -0.5097846984863281, -0.4179229736328125, -0.593109130859375, -0.5366611480712891, 0.5695648193359375, 0.663787841796875, 1.472381591796875, 1.842376708984375, -0.4953765869140625, 2.013153076171875, -0.43914794921875, 1.770751953125, 2.240478515625, 3.118194580078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000048.npy"}
{"epoch": 0.10052356020942409, "step": 49, "batch_size": 128, "mean": 0.5084781646728516, "std": 1.6284102201461792, "min": -6.599761962890625, "p10": -1.4297130584716797, "median": 0.5837860107421875, "p90": 2.4930267333984375, "max": 4.634521484375, "pos_frac": 0.6875, "sample": [1.3150482177734375, 1.26898193359375, 1.0274009704589844, 0.678131103515625, -0.6770477294921875, 0.05755615234375, -1.025115966796875, 3.11517333984375, 1.96173095703125, 0.4142436981201172, -0.32817840576171875, 1.37286376953125, -2.271728515625, -1.15191650390625, 0.7137908935546875, 2.504669189453125, 2.0433502197265625, 1.1036605834960938, 0.09699249267578125, 2.488037109375, -1.461700439453125, -0.017612457275390625, 1.2681884765625, 1.56884765625, 1.48492431640625, -0.800750732421875, -1.55645751953125, 4.634521484375, 0.5428085327148438, 2.1270751953125, 0.8572998046875, 1.262451171875, 3.279876708984375, -0.552398681640625, 3.029296875, 0.9671554565429688, -0.246978759765625, -0.007965087890625, 0.0716400146484375, 0.594940185546875, 0.5294628143310547, 0.6160888671875, -1.160919189453125, 0.7345504760742188, 1.9040679931640625, 1.7604217529296875, 1.759765625, 2.18206787109375, 2.7135009765625, -1.188140869140625, 0.5726318359375, -2.3719100952148438, -0.5477981567382812, 0.8948211669921875, 0.33673858642578125, 2.57568359375, 0.46612548828125, 1.157989501953125, -0.1265106201171875, 0.0077362060546875, 0.5391883850097656, 2.9230194091796875, -0.994720458984375, 1.02044677734375, 0.9122047424316406, 1.0217742919921875, 0.928680419921875, -1.4160041809082031, 2.365020751953125, 1.1448974609375, 2.78955078125, 2.1235198974609375, 0.13385772705078125, -0.4360771179199219, 0.3311748504638672, 2.00091552734375, -4.23822021484375, -1.07806396484375, 1.89910888671875, -0.92950439453125, 0.09937477111816406, 2.207977294921875, 1.639801025390625, -0.6000823974609375, 0.6324539184570312, 1.13543701171875, -4.2132568359375, 0.66802978515625, 2.754669189453125, 1.355743408203125, 1.76483154296875, -0.287445068359375, 0.7513961791992188, 1.357421875, 0.2247467041015625, -1.65850830078125, -0.699554443359375, 1.810943603515625, 0.7187385559082031, -1.69281005859375, -1.96661376953125, -0.524078369140625, -2.7669677734375, 0.53515625, 2.5577392578125, -1.481475830078125, -0.422088623046875, 3.1376953125, 1.3989410400390625, 0.2345428466796875, 4.0952911376953125, 0.16551971435546875, 0.8909683227539062, -0.44243621826171875, -0.058837890625, 0.417724609375, -0.51116943359375, 2.455078125, 1.173248291015625, 0.46319580078125, -6.599761962890625, 0.25023651123046875, 0.003570556640625, 0.033203125, 0.800872802734375, -1.970977783203125, -0.4535369873046875, 0.09030914306640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000049.npy"}
{"epoch": 0.10261780104712041, "step": 50, "batch_size": 128, "mean": 0.36494964361190796, "std": 1.4896831512451172, "min": -3.9095306396484375, "p10": -1.208013916015625, "median": 0.3732719421386719, "p90": 1.9660354614257811, "max": 5.534088134765625, "pos_frac": 0.59375, "sample": [0.4490509033203125, -0.333404541015625, -0.65435791015625, 0.41610145568847656, -0.198455810546875, -0.48480224609375, 0.598236083984375, -0.6581153869628906, 0.107391357421875, 5.534088134765625, 1.03155517578125, -0.12536334991455078, 0.63739013671875, 0.343414306640625, 2.330322265625, 1.8207550048828125, 1.45220947265625, -2.351715087890625, 0.6310272216796875, -0.7256698608398438, -0.1324462890625, -1.244415283203125, -3.9095306396484375, 1.6823272705078125, 0.397430419921875, 3.08990478515625, 1.0784683227539062, -0.10157012939453125, 0.8641357421875, -0.174102783203125, -1.825225830078125, 2.75732421875, 0.9322967529296875, -1.024322509765625, 0.44141387939453125, -1.57366943359375, -0.4676055908203125, -0.11102294921875, 0.591461181640625, 0.021087646484375, -0.5868759155273438, -0.1033172607421875, 1.487335205078125, 0.430908203125, -0.3313751220703125, 0.6341552734375, 0.30047607421875, 0.00067138671875, -0.09026336669921875, -0.21527862548828125, 0.6153564453125, -0.9416961669921875, 0.01329803466796875, 0.03691864013671875, 0.7674484252929688, 1.752532958984375, 2.32989501953125, -1.192413330078125, 1.9459991455078125, 3.12506103515625, 1.0549163818359375, 0.8617401123046875, -3.46234130859375, -0.3751258850097656, -2.276947021484375, -2.356597900390625, -0.635009765625, 0.3034210205078125, 1.4273452758789062, -0.14339447021484375, 0.450439453125, -0.5518417358398438, -0.08951759338378906, 0.646942138671875, -0.09221649169921875, -3.22540283203125, 1.289337158203125, 0.40521240234375, 1.13397216796875, -0.10502815246582031, -0.7141265869140625, 0.557159423828125, -1.446563720703125, -1.8981475830078125, 0.2767219543457031, 2.609832763671875, -2.316741943359375, -0.956817626953125, 1.214599609375, -0.8489990234375, 1.880340576171875, 0.18212890625, 3.674102783203125, 0.963226318359375, 0.8526611328125, 0.74969482421875, 0.7882614135742188, 0.6195068359375, 1.026947021484375, 0.516021728515625, 1.92413330078125, 1.689697265625, -0.1255340576171875, 2.1308746337890625, 0.771514892578125, 0.41230010986328125, 0.9676780700683594, 1.373260498046875, -0.7459945678710938, -0.8211669921875, -0.21917724609375, 1.913787841796875, 5.3857421875, -1.187774658203125, -0.298095703125, -1.86187744140625, 4.68548583984375, 2.252838134765625, -0.1264495849609375, 0.26419830322265625, -0.021240234375, 0.34911346435546875, 1.071258544921875, -0.325347900390625, 0.431976318359375, 1.2583770751953125, 0.47104644775390625, 2.012786865234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000050.npy"}
{"epoch": 0.10471204188481675, "step": 51, "batch_size": 128, "mean": 0.678064227104187, "std": 1.9185408353805542, "min": -7.294189453125, "p10": -1.241033935546875, "median": 0.4565887451171875, "p90": 3.103216552734375, "max": 7.0521240234375, "pos_frac": 0.65625, "sample": [1.911865234375, 0.4431610107421875, 1.7639141082763672, 0.3967018127441406, 1.725341796875, -0.8574447631835938, 0.21453857421875, 4.221954345703125, -0.5816650390625, 0.1722412109375, 2.2422027587890625, 1.048309326171875, 2.2208251953125, 3.27618408203125, -1.1263236999511719, -0.9172515869140625, 1.04095458984375, -1.25799560546875, 0.418609619140625, 3.054229736328125, -7.294189453125, 0.6209716796875, 0.45111083984375, -0.7010650634765625, 2.0497589111328125, 4.7015380859375, -0.5377578735351562, 1.33868408203125, 3.9078826904296875, -0.372955322265625, -1.4190673828125, 2.73809814453125, 7.0521240234375, 1.4661407470703125, 0.9335479736328125, 5.39739990234375, 4.39923095703125, 1.617889404296875, -1.2337646484375, 0.43814849853515625, 0.7504501342773438, 2.3474884033203125, -1.56304931640625, 2.2174072265625, -0.630950927734375, 3.193572998046875, -0.2017364501953125, 0.563262939453125, -0.093017578125, -4.39434814453125, 1.951263427734375, 0.6135711669921875, -3.388336181640625, 0.836212158203125, 0.462066650390625, 1.7603302001953125, -0.02870941162109375, 1.3004150390625, -0.98260498046875, 1.97613525390625, -0.2337646484375, 2.02166748046875, 0.424560546875, 1.4345245361328125, -2.42962646484375, -1.174560546875, 0.05365562438964844, -0.4583282470703125, 0.2169952392578125, 3.500732421875, 0.7865676879882812, 1.047698974609375, 0.0224761962890625, 1.532867431640625, -1.7955322265625, 0.0, 3.290740966796875, -1.1804046630859375, 1.5688629150390625, 3.15350341796875, 0.4088134765625, -0.6352653503417969, 0.2701072692871094, 1.3376922607421875, -2.76995849609375, -1.65032958984375, -0.4428558349609375, -0.37212371826171875, -0.54022216796875, 3.815948486328125, -0.2317962646484375, -0.252838134765625, 0.682769775390625, 0.15428924560546875, 0.2051544189453125, 1.655364990234375, -0.8892822265625, 0.740478515625, 2.2736053466796875, -0.26837158203125, -3.003997802734375, -0.4762725830078125, 0.63934326171875, 3.18145751953125, 2.94476318359375, 3.0816650390625, 1.992095947265625, 1.830810546875, 2.307769775390625, 1.66619873046875, 0.7874755859375, 0.1358795166015625, -0.25665283203125, 1.4197540283203125, 2.204833984375, 1.96417236328125, -2.1680908203125, -2.391693115234375, 0.82110595703125, 1.97540283203125, 0.23223876953125, -0.934356689453125, 0.3017425537109375, -0.5589370727539062, -0.43316650390625, 2.0421295166015625, 0.2554168701171875, 0.3038177490234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000051.npy"}
{"epoch": 0.1068062827225131, "step": 52, "batch_size": 128, "mean": 0.9257479310035706, "std": 1.8488073348999023, "min": -3.7308349609375, "p10": -0.8985610961914061, "median": 0.6886672973632812, "p90": 3.301800537109375, "max": 8.180389404296875, "pos_frac": 0.703125, "sample": [2.0142364501953125, 2.348876953125, -0.223388671875, -0.4639892578125, -0.148773193359375, -0.7570037841796875, 2.492034912109375, 0.8080291748046875, -0.060115814208984375, 5.06427001953125, 0.2123260498046875, 1.474517822265625, -0.8484954833984375, -3.7308349609375, 1.640838623046875, 3.2939453125, 0.87847900390625, 1.25433349609375, -0.3812103271484375, -0.4414215087890625, 1.603271484375, 0.8262138366699219, 1.2812042236328125, 1.0293617248535156, 0.38854217529296875, -1.2530899047851562, 0.45892333984375, 2.110626220703125, 1.443603515625, -1.3386383056640625, -2.62982177734375, -0.49002838134765625, -0.12080574035644531, 1.239166259765625, -0.5380859375, 0.1661376953125, -0.1505584716796875, -3.074981689453125, 0.6031837463378906, -0.218292236328125, 4.38323974609375, 0.5701217651367188, -1.721832275390625, 0.7713775634765625, 1.804229736328125, 0.621185302734375, 0.7008514404296875, 5.08807373046875, 0.2376708984375, 2.24755859375, 1.170135498046875, 1.1509857177734375, 0.3885650634765625, 0.8081817626953125, -0.16637420654296875, 1.5130615234375, -0.19512939453125, 1.632720947265625, 0.29555511474609375, -1.015380859375, 0.88323974609375, -0.118133544921875, 2.626739501953125, -0.30126953125, 1.70269775390625, 0.79278564453125, -1.7972412109375, 0.128662109375, 1.4650726318359375, 0.32745361328125, 0.676483154296875, 1.373443603515625, 2.038421630859375, 3.0118865966796875, -0.83026123046875, -1.30816650390625, 0.7713375091552734, -0.544921875, 1.7275848388671875, 2.285797119140625, 2.610748291015625, 5.81622314453125, 0.5208587646484375, 1.001556396484375, 2.98382568359375, -2.666046142578125, 0.06578445434570312, 4.157806396484375, 5.689971923828125, 1.624908447265625, 2.8830032348632812, 0.0517578125, 0.075897216796875, 3.431610107421875, 0.331298828125, 0.4764747619628906, -0.279052734375, 5.562347412109375, 0.9079437255859375, 0.08650970458984375, -0.12646484375, 3.252960205078125, 1.2577972412109375, -1.60601806640625, 5.00567626953125, 2.0919570922851562, -1.70751953125, 0.5569610595703125, 8.180389404296875, 1.23040771484375, -0.622802734375, 3.32012939453125, 0.058349609375, -0.7046585083007812, 3.3634033203125, 0.6516952514648438, 0.72259521484375, 0.4256477355957031, -0.17932891845703125, 1.259765625, 1.311431884765625, 1.739227294921875, -1.846038818359375, -0.09304046630859375, 0.429931640625, 0.7203369140625, 3.32550048828125, 0.18502044677734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000052.npy"}
{"epoch": 0.10890052356020942, "step": 53, "batch_size": 128, "mean": 0.7723162770271301, "std": 2.032259702682495, "min": -5.37799072265625, "p10": -1.3361846923828122, "median": 0.5145225524902344, "p90": 3.443948364257812, "max": 6.59100341796875, "pos_frac": 0.6328125, "sample": [0.77386474609375, -1.7679443359375, 0.6125335693359375, -1.53155517578125, 1.0450897216796875, -1.50274658203125, 3.1907958984375, -2.549713134765625, -0.27337646484375, 0.02660369873046875, 2.24908447265625, -1.036590576171875, -0.15374755859375, -4.6883544921875, -0.50775146484375, 2.94879150390625, 1.631072998046875, -0.887298583984375, 2.291534423828125, 0.3682861328125, -0.04632568359375, -0.8153533935546875, -0.22711181640625, 1.347900390625, 4.1453857421875, 6.59100341796875, 2.69085693359375, 4.8895263671875, 1.726806640625, 3.42059326171875, 1.7025146484375, 5.3240966796875, -0.3619728088378906, 1.1447296142578125, 1.2365360260009766, -0.041683197021484375, 1.999420166015625, -4.71685791015625, -0.1493816375732422, 1.2004318237304688, 1.2567138671875, -0.25360870361328125, 0.46675872802734375, -2.1233062744140625, -0.05530548095703125, -0.78765869140625, -0.44879150390625, 0.3395538330078125, 3.639404296875, 0.405029296875, 0.86236572265625, 4.33428955078125, 1.5162277221679688, -2.623321533203125, 0.037532806396484375, -0.9623794555664062, 2.48077392578125, 0.6882591247558594, -1.264801025390625, 3.498443603515625, 2.2890777587890625, 0.5712738037109375, -0.11708831787109375, 0.546142578125, 0.2266998291015625, -0.44869232177734375, 2.0101318359375, -0.7654953002929688, 0.6544723510742188, 0.50311279296875, 0.45316314697265625, 2.612548828125, 3.95245361328125, 2.4153594970703125, 4.08074951171875, 2.4439697265625, -0.5431671142578125, 1.24822998046875, 0.70587158203125, 0.5234375, 2.450927734375, -0.1058197021484375, 1.200714111328125, -0.660552978515625, -0.305694580078125, 2.36419677734375, -5.37799072265625, 3.270233154296875, 0.4051532745361328, -1.785980224609375, 1.3352203369140625, 0.1204986572265625, 4.54400634765625, -1.6566848754882812, 1.67620849609375, 6.5797119140625, 0.23680877685546875, -2.490020751953125, -0.7730331420898438, -0.2994384765625, -0.3747901916503906, 1.6766891479492188, -1.0220375061035156, -1.2274169921875, 0.5056076049804688, -0.6066513061523438, 0.929901123046875, 2.313995361328125, 0.7749099731445312, -3.59112548828125, -0.19964599609375, 0.05108642578125, 2.821380615234375, 3.67901611328125, 0.141937255859375, 0.601318359375, 2.862548828125, 0.0, 3.27716064453125, 0.44158935546875, 0.963409423828125, 0.996490478515625, 1.4655838012695312, -0.59326171875, 2.619781494140625, 0.48494720458984375, -0.05419921875, 3.52166748046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000053.npy"}
{"epoch": 0.11099476439790576, "step": 54, "batch_size": 128, "mean": 0.6815527677536011, "std": 2.2250497341156006, "min": -6.910919189453125, "p10": -1.697235107421875, "median": 0.5119590759277344, "p90": 3.4974365234374996, "max": 8.4195556640625, "pos_frac": 0.609375, "sample": [3.473724365234375, 3.9952392578125, 2.000640869140625, 2.1083984375, 0.5823822021484375, 0.6730422973632812, -0.575042724609375, 3.39593505859375, -0.3935260772705078, -0.947784423828125, 0.492340087890625, -0.392303466796875, -0.23311614990234375, -2.00665283203125, 4.9273681640625, -0.377197265625, 6.48883056640625, 0.3341064453125, -0.03741455078125, 1.2466583251953125, 1.15155029296875, 3.308746337890625, 0.291656494140625, 2.22784423828125, -0.9945220947265625, -0.440185546875, 1.73175048828125, 0.011505126953125, 0.6082763671875, 0.6627922058105469, 0.5820159912109375, -0.32574462890625, -2.666748046875, 0.36768341064453125, -0.188812255859375, 2.180816650390625, 5.28271484375, -3.40692138671875, 0.9835205078125, -0.1085205078125, 0.5339431762695312, 0.7146072387695312, 4.421630859375, -0.798675537109375, 4.4083251953125, 0.31182861328125, -3.10675048828125, -1.3335418701171875, -0.9996490478515625, -1.880950927734375, 3.2159881591796875, 8.4195556640625, 0.5315780639648438, 0.230743408203125, 0.11299896240234375, 0.24980926513671875, -1.736328125, 1.4861297607421875, 1.188873291015625, 0.5512542724609375, 2.2604198455810547, 1.179229736328125, -1.133209228515625, -0.70050048828125, 2.2158050537109375, 1.539093017578125, -0.3451995849609375, 0.99853515625, 1.3201751708984375, 2.11669921875, -0.7064895629882812, 0.9621505737304688, 0.022369384765625, 2.4132080078125, -0.455108642578125, 0.669281005859375, 1.716278076171875, -0.662933349609375, 0.643829345703125, -0.06207275390625, -0.0932464599609375, -1.1406631469726562, 0.8915634155273438, 0.53857421875, 0.205322265625, -0.8587646484375, 0.833282470703125, -1.2263031005859375, 1.6250267028808594, -0.1729888916015625, -0.711395263671875, 3.009521484375, -3.66448974609375, 2.70220947265625, 1.41876220703125, -0.18978118896484375, -2.80560302734375, 5.724853515625, 7.1285400390625, 4.11700439453125, 2.779632568359375, 3.92852783203125, 3.23760986328125, 0.8372802734375, 3.9157257080078125, 0.15943527221679688, -3.475738525390625, 0.046783447265625, -6.910919189453125, 2.8173828125, -1.0069580078125, -3.0411529541015625, -0.23248291015625, 0.4749755859375, 3.552764892578125, -0.226837158203125, -1.6397781372070312, -1.68048095703125, -0.513885498046875, -0.6695556640625, 0.897674560546875, 1.6583251953125, 0.8776836395263672, -2.6103515625, -2.186431884765625, -0.3011627197265625, 1.3670921325683594, 1.324188232421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000054.npy"}
{"epoch": 0.1130890052356021, "step": 55, "batch_size": 128, "mean": 0.6309804320335388, "std": 1.9242104291915894, "min": -6.061767578125, "p10": -1.3697448730468749, "median": 0.46500205993652344, "p90": 3.23270263671875, "max": 5.410430908203125, "pos_frac": 0.65625, "sample": [-1.2629547119140625, 2.730865478515625, 0.206085205078125, -1.546844482421875, -0.5501861572265625, -2.73577880859375, 0.6017704010009766, 0.337371826171875, 0.915985107421875, -1.3001708984375, 3.2275390625, 2.4542770385742188, 3.1944427490234375, 1.009521484375, 0.26751708984375, 0.45293235778808594, 0.9192237854003906, -0.5767822265625, 0.4704399108886719, 1.1187629699707031, 0.269500732421875, 1.57879638671875, 2.401336669921875, 2.4654693603515625, 2.1510162353515625, -0.2602977752685547, 0.5957489013671875, 1.815460205078125, 0.312652587890625, -0.71136474609375, -2.169464111328125, -0.47693634033203125, -1.09588623046875, 3.29754638671875, 2.1358642578125, 1.6434707641601562, 0.7235565185546875, 1.34197998046875, 0.292999267578125, 4.4849853515625, -4.553314208984375, 0.0424652099609375, -0.34909820556640625, 0.459564208984375, 0.72442626953125, 1.4891204833984375, 0.65106201171875, -0.576202392578125, 0.48455810546875, 4.57940673828125, 0.48230743408203125, -2.0857696533203125, -1.404693603515625, 0.356109619140625, -0.278289794921875, -2.651458740234375, -1.895111083984375, 2.683258056640625, -0.9628028869628906, 1.056365966796875, -1.354766845703125, 1.340789794921875, 5.244781494140625, -0.4021873474121094, 1.83587646484375, -0.65966796875, -1.0435714721679688, 3.4291229248046875, 0.7006912231445312, -0.8688697814941406, -0.67974853515625, 0.6432876586914062, -0.5393218994140625, 0.3943634033203125, -1.0150299072265625, 0.18060874938964844, 0.32769775390625, 2.748260498046875, 0.015918731689453125, 1.339813232421875, 0.41693878173828125, 3.46063232421875, -3.5704345703125, 0.519744873046875, -0.2728729248046875, 0.05677032470703125, -0.9503593444824219, 3.3688812255859375, 3.83245849609375, 2.2811279296875, -1.01055908203125, 1.4319610595703125, 1.539215087890625, 0.273773193359375, 5.410430908203125, 0.3083324432373047, 2.827301025390625, -1.1200714111328125, 0.6799087524414062, 3.209869384765625, -1.180816650390625, 1.4105148315429688, 0.7612113952636719, 1.78192138671875, 0.7675247192382812, -2.6622467041015625, 3.365447998046875, 1.67047119140625, 2.487377166748047, -1.330963134765625, -6.061767578125, -0.3793792724609375, 2.5546112060546875, 0.0613555908203125, -1.1169586181640625, -0.0154876708984375, 2.620391845703125, -1.4909515380859375, 1.5621681213378906, -0.998687744140625, 1.0968780517578125, -0.10585784912109375, -1.56683349609375, 3.2447509765625, 0.111297607421875, 4.4270782470703125, 5.1578216552734375, 1.2811737060546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000055.npy"}
{"epoch": 0.11518324607329843, "step": 56, "batch_size": 128, "mean": 0.9947658181190491, "std": 2.392529010772705, "min": -6.592376708984375, "p10": -1.1395805358886717, "median": 0.5446186065673828, "p90": 4.077413940429687, "max": 7.6751708984375, "pos_frac": 0.65625, "sample": [-0.234466552734375, -3.3502655029296875, 2.9019775390625, 4.97100830078125, 1.0435638427734375, 1.2385635375976562, 2.0198974609375, -0.17755126953125, 0.509918212890625, 5.37945556640625, 1.35589599609375, 1.79913330078125, 0.31829833984375, -0.7292556762695312, 0.7922821044921875, -0.2445831298828125, -0.3746223449707031, -0.2638397216796875, 4.102386474609375, -1.890594482421875, 2.377044677734375, 0.27264404296875, 2.565826416015625, -0.270355224609375, -1.2237319946289062, 5.8167724609375, 0.389801025390625, 2.20257568359375, 2.2406768798828125, -0.4365692138671875, 2.134674072265625, 3.270294189453125, 0.3466796875, 1.10357666015625, 7.2049560546875, 2.657196044921875, -2.8071441650390625, 0.3925018310546875, 1.6784210205078125, -0.5910110473632812, 6.9652252197265625, 0.32781982421875, -0.0141448974609375, -0.059169769287109375, 0.09792327880859375, 7.6751708984375, 2.316558837890625, -0.06441879272460938, -0.7041015625, -0.484619140625, -1.528839111328125, -1.3514862060546875, 1.6584701538085938, -5.25897216796875, 0.4638824462890625, -4.97747802734375, 0.2884483337402344, -0.3536090850830078, 1.42919921875, 0.11365509033203125, 2.3574676513671875, 4.9893798828125, 3.941925048828125, 2.6482086181640625, -0.82568359375, -0.9124603271484375, 1.339630126953125, 2.8461456298828125, 5.800048828125, 0.5669364929199219, -0.48043060302734375, -0.3989677429199219, -3.26702880859375, 0.5061054229736328, 0.0416107177734375, -0.7548980712890625, -2.6766357421875, 3.48150634765625, -0.410980224609375, 1.301544189453125, -6.592376708984375, 0.93035888671875, 2.58099365234375, -0.1168212890625, 0.9453887939453125, 5.744476318359375, 1.426361083984375, 3.69537353515625, -1.103515625, 1.9006195068359375, -0.14363861083984375, 5.952789306640625, 2.19110107421875, 0.8635482788085938, -0.536865234375, 4.06671142578125, 1.2781600952148438, -0.659271240234375, 0.15961837768554688, 0.5223007202148438, 1.8586578369140625, -0.91259765625, 1.2369918823242188, -1.9771728515625, -3.98541259765625, 5.7811279296875, -0.39636993408203125, 1.18524169921875, -0.1521453857421875, 1.4389801025390625, 1.663726806640625, 1.832122802734375, 3.01190185546875, -0.942779541015625, 0.06224250793457031, 1.312286376953125, 0.25799560546875, 0.1546478271484375, 1.7726898193359375, 1.503143310546875, 2.843780517578125, -0.9835968017578125, 1.957122802734375, 1.6717529296875, 0.3864593505859375, 4.4463653564453125, 3.6077423095703125, 0.46483612060546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000056.npy"}
{"epoch": 0.11727748691099477, "step": 57, "batch_size": 128, "mean": 1.4431445598602295, "std": 2.36293363571167, "min": -2.858642578125, "p10": -1.015465545654297, "median": 1.0082216262817383, "p90": 4.301666259765625, "max": 14.63720703125, "pos_frac": 0.7265625, "sample": [-1.7139739990234375, 2.539093017578125, -2.6722259521484375, 2.70562744140625, -0.1635589599609375, -1.40618896484375, 4.82275390625, -0.06389236450195312, -0.700439453125, -0.4305419921875, 0.0580902099609375, -1.165740966796875, 4.57379150390625, 0.716522216796875, 3.4774703979492188, 0.5733222961425781, 0.9371185302734375, 2.097503662109375, 2.2313232421875, 2.058837890625, 1.0089893341064453, 3.77978515625, 0.569732666015625, -0.01815032958984375, 0.396820068359375, 2.42193603515625, -2.251739501953125, 0.0709228515625, 0.04578399658203125, -1.2347564697265625, 2.180389404296875, 1.0928955078125, 1.38507080078125, 1.081146240234375, 8.1201171875, 14.63720703125, 1.451995849609375, -0.54412841796875, -1.8194847106933594, 2.930877685546875, 4.286651611328125, 1.3284835815429688, -1.292510986328125, 5.5196380615234375, 0.969024658203125, 1.0888900756835938, 0.04547882080078125, -0.119720458984375, 1.56781005859375, 2.43994140625, 1.1097259521484375, 0.26241302490234375, 1.3570556640625, 5.299652099609375, 0.2616729736328125, 2.253173828125, 1.4394912719726562, 1.0275115966796875, 4.336700439453125, -0.11175537109375, 0.42120361328125, -0.97564697265625, 0.7034320831298828, 2.67425537109375, 1.144683837890625, 2.41973876953125, 6.79827880859375, 3.64007568359375, -1.588623046875, 0.27410888671875, 2.760101318359375, 0.368255615234375, -1.0417251586914062, -0.5651817321777344, 4.162445068359375, 3.483001708984375, 0.51312255859375, 3.091156005859375, 2.1235198974609375, 1.56756591796875, 0.541961669921875, -0.003875732421875, 6.324951171875, 4.97149658203125, -0.2232666015625, -1.0550384521484375, -1.4861907958984375, 2.595703125, -0.807281494140625, -0.54815673828125, 2.1949615478515625, 4.94842529296875, -0.1783161163330078, 1.0074539184570312, 0.837188720703125, 0.10645675659179688, 2.704345703125, -0.3574371337890625, 0.105560302734375, -0.362457275390625, -2.858642578125, -1.00421142578125, 6.0311279296875, 2.4121322631835938, 0.32784271240234375, 3.5628662109375, 0.4075813293457031, 2.796539306640625, -0.3691253662109375, 0.4479255676269531, 0.6326904296875, 1.69927978515625, 3.561981201171875, -0.7397918701171875, 1.8886375427246094, 2.49737548828125, 3.633514404296875, 0.061527252197265625, 0.7202663421630859, 0.04472160339355469, 2.13427734375, -0.2489013671875, 4.2274169921875, 1.965911865234375, 2.473175048828125, -0.01726531982421875, 3.345123291015625, 4.948638916015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000057.npy"}
{"epoch": 0.1193717277486911, "step": 58, "batch_size": 128, "mean": 1.0589706897735596, "std": 2.1019153594970703, "min": -10.483856201171875, "p10": -1.2642280578613279, "median": 0.9548835754394531, "p90": 3.4173370361328117, "max": 6.112823486328125, "pos_frac": 0.75, "sample": [1.46240234375, 4.059234619140625, 1.079559326171875, 2.847564697265625, 1.10845947265625, -0.6921195983886719, 2.1034774780273438, 2.79376220703125, 1.5966796875, 4.363861083984375, 1.4995880126953125, 2.3235321044921875, 0.9078292846679688, 4.971405029296875, 1.5343017578125, -1.4746932983398438, 4.4087982177734375, 3.63775634765625, 2.7685775756835938, -1.82684326171875, 6.112823486328125, 3.32513427734375, 0.4356842041015625, -1.5503387451171875, -0.3623046875, 1.5096893310546875, 2.453765869140625, 2.114105224609375, -0.0092010498046875, -3.1202392578125, 0.17514419555664062, 1.4787445068359375, 1.22552490234375, 0.12871551513671875, 2.46075439453125, -0.2679595947265625, 0.578704833984375, 1.5215072631835938, -0.352996826171875, -0.134521484375, -3.31890869140625, 2.3504638671875, 0.220184326171875, 0.15625, 4.87225341796875, -1.2181472778320312, 0.46118927001953125, 3.08538818359375, 0.05947113037109375, -10.483856201171875, 0.4652557373046875, 5.6966552734375, -0.969818115234375, -1.3717498779296875, 0.8292236328125, -0.5536346435546875, -0.09296035766601562, 1.261434555053711, 0.135406494140625, 5.125, 2.0638580322265625, 0.2563018798828125, 1.826446533203125, -0.5420379638671875, 1.82476806640625, 2.112579345703125, 0.069610595703125, 2.4191436767578125, 0.17108154296875, 0.679443359375, 1.5445556640625, -1.5286712646484375, -1.5011520385742188, 2.674652099609375, 1.6320686340332031, 0.12518310546875, 5.47650146484375, 0.6065406799316406, -0.75848388671875, 0.0744171142578125, 1.903076171875, 1.72039794921875, 5.6649169921875, -1.792694091796875, 0.3740081787109375, 0.9099502563476562, -0.5876045227050781, 0.6763992309570312, 0.73101806640625, 2.826141357421875, 0.10790252685546875, 1.9711151123046875, 0.335693359375, 1.92529296875, 1.2984619140625, -1.087066650390625, 3.632476806640625, -1.21435546875, 1.9072265625, 0.9391326904296875, 2.881378173828125, 0.1772441864013672, -0.34688568115234375, -1.729034423828125, 2.3603363037109375, 0.9133453369140625, 0.8341865539550781, 2.5414276123046875, 3.0655136108398438, 2.02667236328125, -1.6099853515625, 2.667724609375, 0.9706344604492188, 3.80548095703125, 2.0398178100585938, 0.6809768676757812, -0.8725128173828125, -1.21368408203125, 2.2112960815429688, 2.645355224609375, 1.7230987548828125, 1.2400341033935547, 0.409912109375, 1.679962158203125, -0.5661392211914062, -1.437347412109375, 2.819305419921875, 0.256866455078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000058.npy"}
{"epoch": 0.12146596858638743, "step": 59, "batch_size": 128, "mean": 0.8495495915412903, "std": 2.654871702194214, "min": -8.00830078125, "p10": -1.6733947753906249, "median": 0.4972991943359375, "p90": 4.1732940673828125, "max": 8.899444580078125, "pos_frac": 0.6171875, "sample": [-1.0438995361328125, 1.316619873046875, 0.05938720703125, -8.00830078125, 1.3206253051757812, 0.217376708984375, -0.870880126953125, 1.76837158203125, 0.3291778564453125, 0.9124755859375, 0.8922958374023438, 1.6196441650390625, 1.582305908203125, 2.3302459716796875, -2.3761138916015625, 0.30877685546875, -0.02457427978515625, -0.36083984375, -1.2692108154296875, 2.1377410888671875, -0.5959014892578125, 0.6895828247070312, 0.2033672332763672, 0.1709442138671875, 2.9084320068359375, 0.5720481872558594, 1.1323089599609375, 4.69378662109375, -3.7264251708984375, 1.71527099609375, -2.696990966796875, 1.0263252258300781, -0.227935791015625, 2.4957275390625, 0.0, 4.2071533203125, 0.06107330322265625, 1.46258544921875, -0.353759765625, -2.513397216796875, 0.246856689453125, 4.46209716796875, 6.916839599609375, 7.932861328125, 0.101226806640625, -4.611053466796875, 1.7926864624023438, 6.7606201171875, 2.85333251953125, 8.2806396484375, 1.7393531799316406, -0.1343994140625, 2.9700927734375, 6.92889404296875, 0.51995849609375, 1.3521957397460938, -0.15362548828125, 1.1571502685546875, -3.09307861328125, -0.2610321044921875, -1.48828125, 2.7235794067382812, 0.332763671875, -1.72607421875, 0.3237457275390625, 4.158782958984375, 6.809967041015625, -0.93218994140625, 2.0642242431640625, 3.0209121704101562, -0.6180419921875, 0.8815460205078125, -0.778411865234375, 6.4788818359375, -1.9239006042480469, -5.93682861328125, 0.7602758407592773, -3.39691162109375, -0.196044921875, 0.1107635498046875, 2.82012939453125, 0.389556884765625, 2.4221954345703125, 2.0397491455078125, 0.9485931396484375, -0.5984039306640625, 0.837066650390625, -0.140228271484375, -1.071258544921875, -0.57891845703125, 3.41046142578125, 1.509857177734375, -0.18468093872070312, -0.55999755859375, -0.7094573974609375, 4.91925048828125, 0.54852294921875, -1.65081787109375, -1.3238525390625, 1.575714111328125, -1.1163787841796875, 0.7451248168945312, 0.474639892578125, 2.0391845703125, -0.31862640380859375, 1.171295166015625, -1.466766357421875, -2.595794677734375, -0.9394454956054688, 1.273681640625, 3.153228759765625, 3.4906005859375, 6.85662841796875, -0.5987548828125, -0.605621337890625, 0.37175750732421875, 8.899444580078125, 1.613861083984375, 2.326324462890625, -0.5260467529296875, -3.7896728515625, 0.8922576904296875, 2.493438720703125, 1.887451171875, -0.22177886962890625, -0.04320526123046875, 0.883575439453125, -0.7073287963867188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000059.npy"}
{"epoch": 0.12356020942408377, "step": 60, "batch_size": 128, "mean": 1.1964894533157349, "std": 2.44051456451416, "min": -5.94842529296875, "p10": -1.3695278167724607, "median": 1.1243324279785156, "p90": 3.8686744689941404, "max": 11.100677490234375, "pos_frac": 0.71875, "sample": [-5.777557373046875, 2.855072021484375, 4.587104797363281, -1.1719512939453125, 2.49810791015625, 3.224853515625, -0.07427406311035156, 2.3888168334960938, 2.688262939453125, 2.53094482421875, 0.6535491943359375, -0.4149513244628906, -0.207977294921875, 2.3788604736328125, 0.104827880859375, 1.9686279296875, 0.4475250244140625, -2.27813720703125, 3.4124908447265625, 3.978759765625, 1.0008316040039062, -1.5888671875, 4.436309814453125, -1.6065216064453125, -1.705841064453125, 0.5590667724609375, 0.6910552978515625, 0.0, -0.7269363403320312, 4.12396240234375, -0.355743408203125, 2.71307373046875, 2.206207275390625, 1.6807098388671875, 0.25457763671875, -1.3219146728515625, 3.4687347412109375, -2.739410400390625, 2.1596450805664062, 0.2779350280761719, -1.1036453247070312, 1.8382339477539062, 1.2086181640625, 4.254371643066406, 0.7950820922851562, 0.1615753173828125, -0.8439655303955078, 1.997314453125, 0.3175201416015625, 0.374114990234375, 0.5165786743164062, -0.50439453125, 1.53466796875, 3.628387451171875, 0.14612579345703125, -1.188934326171875, 3.5598907470703125, -0.83154296875, 1.2386398315429688, 4.1102294921875, 0.5968437194824219, 2.730133056640625, 2.5854949951171875, 0.6869945526123047, -5.94842529296875, 2.2622833251953125, 1.5108642578125, -0.47216796875, -0.19455718994140625, -1.23175048828125, -0.56658935546875, 3.7600555419921875, 3.8214950561523438, 6.2989501953125, 2.822113037109375, 0.52056884765625, -0.0258636474609375, 2.5210723876953125, 1.5526275634765625, 7.116241455078125, 3.677520751953125, 0.761871337890625, -1.4806251525878906, 1.743255615234375, 5.964599609375, 0.8181610107421875, 1.093963623046875, -4.867950439453125, 1.1547012329101562, 2.532623291015625, 0.7552337646484375, 0.29088592529296875, -4.06591796875, 0.3099098205566406, 1.32318115234375, 3.1254425048828125, 0.909423828125, 4.058349609375, 0.47948455810546875, -1.29290771484375, 0.62713623046875, 3.7169189453125, 1.2654876708984375, -1.76727294921875, 5.1653900146484375, 2.8739776611328125, -0.22566986083984375, 2.3046951293945312, 11.100677490234375, 0.0, 6.21807861328125, -2.9517822265625, 1.578643798828125, -0.45086669921875, -1.1687583923339844, 0.70086669921875, 1.586029052734375, 0.3616600036621094, 1.8243255615234375, -4.19232177734375, 1.8374557495117188, 2.2557525634765625, 1.6197586059570312, 3.318695068359375, 2.987579345703125, 1.5357513427734375, 2.5650634765625, 2.297088623046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000060.npy"}
{"epoch": 0.1256544502617801, "step": 61, "batch_size": 128, "mean": 1.1082285642623901, "std": 2.6079630851745605, "min": -5.9212646484375, "p10": -1.755538940429687, "median": 0.7900352478027344, "p90": 4.420492553710937, "max": 9.93798828125, "pos_frac": 0.6953125, "sample": [-2.73199462890625, -2.020111083984375, 0.6978816986083984, 0.112152099609375, 4.05792236328125, 2.731689453125, -0.34014892578125, 0.73052978515625, 0.0293731689453125, 0.9043426513671875, 1.334869384765625, -3.083953857421875, -1.47735595703125, 1.113433837890625, 1.00555419921875, 0.64208984375, -4.25469970703125, -1.2012481689453125, -3.156158447265625, -0.812286376953125, 1.4771804809570312, 0.8047332763671875, 6.309783935546875, 5.22845458984375, 4.1060791015625, 0.4208984375, 4.2298583984375, 0.5496368408203125, 3.401763916015625, 2.1691131591796875, 0.12816619873046875, 0.4134521484375, 0.6838045120239258, -0.506439208984375, 4.250518798828125, 0.6597671508789062, -0.5274505615234375, 3.7933807373046875, 0.990753173828125, 0.46002960205078125, -0.5362701416015625, 0.098480224609375, 2.4733734130859375, -0.05081939697265625, 0.899169921875, 1.0020294189453125, 4.873626708984375, 1.874725341796875, 1.657379150390625, 1.69580078125, 0.729248046875, 2.8077239990234375, 8.862762451171875, 5.31463623046875, 4.408233642578125, 2.52569580078125, 2.244720458984375, 1.3933563232421875, 6.6932830810546875, 0.1924896240234375, 0.40966796875, 2.8603858947753906, 1.94464111328125, -0.7853927612304688, -0.1292572021484375, 4.1903076171875, 4.96923828125, 6.258392333984375, 2.618011474609375, 1.2725372314453125, 2.6151275634765625, 0.2906646728515625, 0.9835586547851562, -1.151123046875, -0.9877700805664062, -0.72247314453125, 0.1533527374267578, 0.9129180908203125, 1.6221466064453125, -2.0362548828125, 5.71795654296875, 0.3834381103515625, 2.509490966796875, 3.9769287109375, 1.367523193359375, 1.9122314453125, 0.1969146728515625, -2.0718231201171875, -0.2724800109863281, 0.992462158203125, -0.2763080596923828, -2.1319122314453125, -2.02490234375, 1.409271240234375, -0.1382274627685547, 1.3233261108398438, 1.3732719421386719, 4.4490966796875, -1.041412353515625, 7.017578125, 0.1549072265625, 9.93798828125, 1.57940673828125, 0.0, 0.3506622314453125, 0.8689193725585938, 3.6105499267578125, 0.6374053955078125, 1.3098678588867188, -5.9212646484375, 4.10479736328125, -3.19854736328125, -0.7236175537109375, -1.1550140380859375, -3.629852294921875, 1.78631591796875, -5.9110107421875, -0.44907379150390625, -1.64215087890625, 0.19500732421875, -0.93011474609375, -0.1951446533203125, -1.1475753784179688, 0.7753372192382812, 1.9128646850585938, 4.85870361328125, 3.743621826171875, -1.5158462524414062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000061.npy"}
{"epoch": 0.12774869109947645, "step": 62, "batch_size": 128, "mean": 1.2413785457611084, "std": 2.7294981479644775, "min": -5.84381103515625, "p10": -1.617840576171875, "median": 0.8416786193847656, "p90": 4.668902587890625, "max": 9.525634765625, "pos_frac": 0.6953125, "sample": [4.43316650390625, 0.43209075927734375, 7.910797119140625, 3.3316116333007812, 4.471160888671875, 0.5092315673828125, 3.749969482421875, 3.21844482421875, -3.06085205078125, -0.3134002685546875, -2.74090576171875, -1.5111770629882812, -1.3474578857421875, 3.99981689453125, 2.43231201171875, 0.8765487670898438, 0.200042724609375, 1.7379684448242188, -0.1781158447265625, -1.5092926025390625, 2.1565093994140625, 0.2025146484375, 2.516204833984375, 0.8082962036132812, -2.7010040283203125, 7.64453125, -4.40252685546875, 0.394073486328125, 0.5138740539550781, 1.3642578125, 0.635467529296875, 0.999267578125, 2.3778228759765625, 1.3848419189453125, 0.15152931213378906, 0.46787071228027344, 1.531005859375, 0.3695526123046875, 1.9333572387695312, -1.236907958984375, 1.7017974853515625, 4.99365234375, 4.0521087646484375, -0.7197265625, -0.5029144287109375, -0.617919921875, 0.464813232421875, 1.50732421875, 3.730621337890625, 9.239227294921875, -5.84381103515625, 4.188873291015625, -1.259429931640625, 4.20343017578125, 2.3671875, -2.8747406005859375, 7.0692138671875, 0.0, 3.4839019775390625, -0.28791046142578125, -3.318084716796875, -1.1693954467773438, 4.7241668701171875, 0.88360595703125, 1.5616378784179688, -0.6354522705078125, 0.8539047241210938, 0.41339111328125, 0.8294525146484375, 4.065521240234375, 3.274566650390625, 0.333038330078125, 0.33582115173339844, -2.7380523681640625, 2.1080703735351562, 4.3226776123046875, 1.046600341796875, -2.33074951171875, 1.2312240600585938, 1.9104557037353516, 0.0812225341796875, -0.9165496826171875, 2.301177978515625, 1.570556640625, -2.20001220703125, 4.158050537109375, 9.525634765625, 0.949951171875, 0.2130126953125, 4.67022705078125, 0.9872913360595703, -1.6434326171875, 7.31109619140625, 1.4022369384765625, 4.6683349609375, 2.9055938720703125, -2.7838287353515625, 6.0816650390625, -1.005462646484375, -1.5916290283203125, 5.4180450439453125, 0.9544830322265625, 0.08025360107421875, 1.995269775390625, 6.5906524658203125, 3.54754638671875, 0.7012939453125, 0.6483001708984375, -0.0617828369140625, 4.681915283203125, 1.57720947265625, 0.03594207763671875, -1.28155517578125, -0.366424560546875, -0.603759765625, -0.9205551147460938, -1.60687255859375, -1.47119140625, 2.5998611450195312, 0.8012046813964844, 0.883087158203125, -0.10498046875, 3.797332763671875, 0.4380645751953125, 0.3524017333984375, 1.852569580078125, -3.09393310546875, -1.5836639404296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000062.npy"}
{"epoch": 0.12984293193717278, "step": 63, "batch_size": 128, "mean": 1.2967699766159058, "std": 2.9713306427001953, "min": -6.876220703125, "p10": -2.4117416381835937, "median": 1.1145133972167969, "p90": 4.7825683593749995, "max": 11.2476806640625, "pos_frac": 0.6640625, "sample": [3.02264404296875, 3.85516357421875, 0.44147300720214844, -4.384765625, -3.61181640625, -0.5284404754638672, 2.5552825927734375, -0.8058929443359375, -0.8241043090820312, 2.554351806640625, -3.469696044921875, -2.309600830078125, -2.4058685302734375, -2.7290420532226562, 3.486480712890625, -0.7324600219726562, 2.1905517578125, 3.61383056640625, 8.6058349609375, 8.790069580078125, -2.43658447265625, -2.425445556640625, 9.79620361328125, -2.247650146484375, 1.05242919921875, 1.530181884765625, 2.232818603515625, -2.1988525390625, 0.1278533935546875, 4.82623291015625, -0.28000450134277344, 2.577545166015625, -0.1212310791015625, 0.4706382751464844, 1.1631431579589844, 11.2476806640625, -3.95916748046875, 0.9728851318359375, -2.710662841796875, 2.84075927734375, -3.619842529296875, 1.1354598999023438, 3.907135009765625, -3.373687744140625, 0.8122634887695312, 3.759490966796875, 7.32806396484375, -6.876220703125, -1.861572265625, 5.007659912109375, 0.5002727508544922, 4.1065673828125, 2.32293701171875, 0.0, -0.2041034698486328, 2.508270263671875, 2.131988525390625, -0.6578445434570312, -0.00089263916015625, 0.5031852722167969, 0.8860092163085938, 0.63836669921875, 1.414306640625, 5.130126953125, -1.23046875, -0.18006134033203125, 2.95745849609375, 1.4273681640625, -1.4553909301757812, 3.56060791015625, 4.885986328125, 1.4386138916015625, 3.85565185546875, 0.59661865234375, 4.897216796875, -3.57159423828125, 3.38665771484375, 1.4039039611816406, 1.684844970703125, -1.000518798828125, 1.61688232421875, 7.780426025390625, 2.509490966796875, -1.85809326171875, 2.345489501953125, 0.4434967041015625, 2.4574203491210938, -0.16233444213867188, 2.5155181884765625, 2.496795654296875, 2.1432647705078125, -0.0289306640625, -1.67803955078125, 4.268157958984375, 0.6733245849609375, 4.5372314453125, 0.05400848388671875, 3.577728271484375, 3.2585906982421875, 4.5269775390625, 1.5841064453125, 2.9334869384765625, 5.007537841796875, -0.4633026123046875, 2.7547531127929688, 0.27117919921875, 0.822601318359375, -0.28765106201171875, 3.50103759765625, 1.09356689453125, -0.52874755859375, -1.003326416015625, 2.845367431640625, 2.679595947265625, 2.1652908325195312, 5.864471435546875, 0.1528778076171875, 0.983673095703125, 0.3179931640625, -0.05332183837890625, -4.79803466796875, 0.2656097412109375, 2.4981536865234375, 1.792938232421875, -1.6805419921875, -0.5124931335449219, 3.610870361328125, 4.76385498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000063.npy"}
{"epoch": 0.1319371727748691, "step": 64, "batch_size": 128, "mean": 1.4545016288757324, "std": 3.169914722442627, "min": -6.345916748046875, "p10": -1.9563262939453125, "median": 1.330129623413086, "p90": 5.952105712890624, "max": 10.87615966796875, "pos_frac": 0.71875, "sample": [3.9245758056640625, -1.4686279296875, -5.358551025390625, 7.39349365234375, -2.90899658203125, 3.797943115234375, 2.5997467041015625, 2.162322998046875, 7.267578125, 0.7657012939453125, 2.885345458984375, 2.26055908203125, 6.21234130859375, 1.03704833984375, 0.9768753051757812, -1.941192626953125, 7.45416259765625, 2.029022216796875, 0.803192138671875, -5.013275146484375, 1.8102569580078125, 1.3010025024414062, -1.3138885498046875, -1.641204833984375, 0.717315673828125, -4.3927459716796875, 2.0072364807128906, 0.7705764770507812, 2.58160400390625, 4.633819580078125, 1.538177490234375, 2.22509765625, 3.32952880859375, 0.40763092041015625, 2.7864990234375, 0.6434860229492188, 0.080322265625, 2.1340179443359375, 2.369476318359375, 0.80987548828125, 3.64239501953125, -0.5354843139648438, -1.5434646606445312, 2.3202667236328125, -1.2948760986328125, 10.199859619140625, 6.7298583984375, -0.9139785766601562, 6.725311279296875, 1.9513092041015625, 1.64434814453125, -2.020050048828125, 0.3128700256347656, -0.8416748046875, 0.7341632843017578, -1.1130828857421875, -1.99163818359375, 2.1639404296875, 4.177696228027344, 2.0474700927734375, 1.45428466796875, 3.416522979736328, -2.3922119140625, 4.894195556640625, 2.82666015625, 0.170135498046875, -0.08638763427734375, -6.345916748046875, 0.07281494140625, 1.97467041015625, 0.26678466796875, 1.364105224609375, 10.87615966796875, 2.2048721313476562, 0.20993423461914062, 2.479644775390625, 1.3424072265625, -2.46258544921875, -0.4840736389160156, 0.3726806640625, -1.6126823425292969, 6.01837158203125, -0.3692131042480469, 0.11604118347167969, 4.587554931640625, 4.77252197265625, 2.4142608642578125, 1.5646896362304688, 0.9061012268066406, -3.02215576171875, 1.974639892578125, 7.303619384765625, 5.9237060546875, 6.76202392578125, 0.39492034912109375, -1.26995849609375, -6.32904052734375, 1.0128936767578125, -1.295501708984375, 1.4107818603515625, 4.197845458984375, 5.37457275390625, 2.595123291015625, 0.0, 0.47310638427734375, 0.2768898010253906, -0.902099609375, 1.2634429931640625, 3.5252685546875, -1.940704345703125, 4.00384521484375, 1.3178520202636719, -0.6922454833984375, 6.877838134765625, 0.9677734375, -0.17847442626953125, 0.01172637939453125, 8.2972412109375, -5.967193603515625, 2.2737503051757812, 0.0, 2.9730224609375, 5.2685546875, 2.7165069580078125, 3.7584915161132812, 2.21124267578125, -5.60357666015625, -0.414459228515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000064.npy"}
{"epoch": 0.13403141361256546, "step": 65, "batch_size": 128, "mean": 1.1527847051620483, "std": 2.7385811805725098, "min": -9.542236328125, "p10": -1.4045211791992187, "median": 1.0684318542480469, "p90": 4.225648498535156, "max": 6.664764404296875, "pos_frac": 0.6875, "sample": [0.056133270263671875, 0.6229248046875, 1.6077880859375, 0.09162521362304688, 6.5904388427734375, 1.121124267578125, 0.7125778198242188, 2.4603271484375, 2.763671875, -6.612762451171875, -3.673126220703125, -4.9407958984375, 1.00909423828125, 3.891632080078125, -1.3748931884765625, 4.36962890625, 4.113037109375, -0.7950439453125, 0.9368553161621094, -1.0872802734375, -1.620391845703125, 0.2717094421386719, 3.696746826171875, 5.844818115234375, 1.9184551239013672, 2.6121368408203125, 4.656494140625, 1.87786865234375, -4.95941162109375, -0.881805419921875, 2.03118896484375, -1.4927978515625, -0.1155853271484375, 2.2251739501953125, 6.2532958984375, -0.708770751953125, 0.0597686767578125, 4.180419921875, -0.16664886474609375, 0.000965118408203125, 3.42919921875, -2.63983154296875, 1.29571533203125, 1.5496597290039062, 3.98089599609375, 0.780670166015625, 5.6395416259765625, 3.1020660400390625, 5.5277099609375, 0.9756393432617188, 1.5544967651367188, 2.265869140625, -0.34869384765625, 1.68402099609375, 2.8065414428710938, -0.0301513671875, 3.8756103515625, 0.8954353332519531, 0.06495475769042969, 3.3584442138671875, -1.0614013671875, 0.19844818115234375, 0.633544921875, -0.8256912231445312, 2.3681793212890625, 0.294525146484375, 2.0303726196289062, -0.8365936279296875, -2.66778564453125, 0.253997802734375, 2.8569412231445312, 3.50189208984375, 0.99755859375, 0.7213821411132812, 3.2567138671875, 0.2115631103515625, -1.078826904296875, 0.3623371124267578, 0.8787994384765625, -0.968658447265625, -6.154571533203125, -1.29833984375, 0.0, 0.0218048095703125, 3.2191314697265625, 1.4993209838867188, -1.3973846435546875, 2.064910888671875, 4.2720794677734375, 2.8594970703125, 1.128143310546875, 2.720306396484375, 6.58953857421875, 3.5231781005859375, 4.975006103515625, -0.088409423828125, 3.99798583984375, 3.1040802001953125, -0.1391124725341797, 6.664764404296875, 1.54852294921875, 3.2457427978515625, -9.542236328125, 2.4169158935546875, 1.5741653442382812, 3.436676025390625, -0.573760986328125, 5.872314453125, 2.3599853515625, 6.348388671875, 1.78973388671875, -0.0308685302734375, 1.5260772705078125, 1.2928924560546875, -5.72259521484375, -1.1423835754394531, 4.20574951171875, -1.88037109375, -0.113037109375, 1.8944091796875, -1.421173095703125, -0.990325927734375, 1.545013427734375, -0.3499755859375, 3.825927734375, -0.008626937866210938, 1.0157394409179688, -0.5700607299804688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000065.npy"}
{"epoch": 0.13612565445026178, "step": 66, "batch_size": 128, "mean": 1.3753262758255005, "std": 3.6771912574768066, "min": -9.758453369140625, "p10": -1.9693660736083984, "median": 1.2421035766601562, "p90": 6.126097106933594, "max": 12.543212890625, "pos_frac": 0.6796875, "sample": [0.687744140625, 1.4151363372802734, 5.084014892578125, -1.89910888671875, -1.5007171630859375, 6.14300537109375, -1.343963623046875, 0.7172813415527344, 1.7264862060546875, 0.914794921875, -0.11467361450195312, -0.4557151794433594, 2.0145339965820312, 1.25360107421875, 2.4708709716796875, 0.381805419921875, -0.4560699462890625, 0.52117919921875, 7.126708984375, 1.637237548828125, 3.074371337890625, -6.088623046875, 4.711723327636719, 3.049407958984375, 3.080535888671875, 2.3099822998046875, 0.1490020751953125, 0.88262939453125, 0.06576919555664062, 6.0690460205078125, 2.339385986328125, -0.0058746337890625, 11.015304565429688, -4.8167724609375, 0.7021865844726562, -0.30773162841796875, -0.83148193359375, 12.543212890625, 2.868988037109375, -1.95361328125, -5.2459564208984375, 2.373666763305664, 6.5836944580078125, 2.064239501953125, 1.2560882568359375, 3.0925979614257812, -0.85174560546875, 10.98455810546875, 7.31451416015625, -0.2877197265625, -6.4307861328125, 1.32208251953125, 1.2306060791015625, 10.10504150390625, 7.91790771484375, 1.0172119140625, 0.3829193115234375, 4.390357971191406, -9.758453369140625, -0.606781005859375, -1.1118621826171875, 0.4946136474609375, 8.755767822265625, 1.2649993896484375, 3.263763427734375, -6.74383544921875, -0.1660308837890625, -0.990386962890625, 3.2039642333984375, -1.0717620849609375, 0.3192138671875, 1.581878662109375, 7.0889892578125, -3.63385009765625, 2.345947265625, 1.4791717529296875, 3.98590087890625, -0.8953094482421875, 2.614349365234375, -6.639129638671875, 3.3883819580078125, 1.4794692993164062, -1.865631103515625, 3.60638427734375, -1.0913162231445312, 2.0111770629882812, -0.49173736572265625, 1.1139678955078125, 4.283203125, 2.62408447265625, 0.5835342407226562, 3.8212890625, 1.58642578125, -8.33795166015625, -0.88446044921875, -1.9122161865234375, -3.7050323486328125, -1.8662261962890625, 2.4398345947265625, 0.657073974609375, 6.7869873046875, 2.42987060546875, 0.78875732421875, 0.0, 2.8391380310058594, -1.4586105346679688, 0.1253662109375, 6.1188507080078125, 1.1533966064453125, 4.606719970703125, 2.499786376953125, 3.20135498046875, 0.13570022583007812, -4.80218505859375, 0.33709716796875, 2.159088134765625, -2.334716796875, 5.560577392578125, 3.472747802734375, 4.247344970703125, -1.499267578125, -0.20904541015625, 1.6132049560546875, -2.006122589111328, 3.458160400390625, 1.0343742370605469, 6.15875244140625, 5.00213623046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000066.npy"}
{"epoch": 0.1382198952879581, "step": 67, "batch_size": 128, "mean": 1.2156729698181152, "std": 3.3596878051757812, "min": -9.924072265625, "p10": -2.5652221679687486, "median": 1.0814552307128906, "p90": 5.237873840332029, "max": 9.206268310546875, "pos_frac": 0.7109375, "sample": [-0.75140380859375, 4.1756591796875, -0.798797607421875, 8.75, 0.4783668518066406, -5.6170654296875, 1.0894241333007812, 3.9903564453125, 5.071380615234375, -0.3697052001953125, 2.0026779174804688, 9.206268310546875, 0.54046630859375, 2.1629638671875, 2.610382080078125, 3.49224853515625, 4.4086761474609375, 6.158782958984375, 0.279754638671875, -2.214630126953125, 1.44903564453125, -1.145538330078125, 3.9126434326171875, 0.0741424560546875, 1.1897125244140625, 6.5583648681640625, -2.1497802734375, -1.287567138671875, 2.2422256469726562, 0.2195587158203125, -4.394287109375, 0.05683135986328125, -0.663970947265625, 2.4427337646484375, 1.32659912109375, -3.65716552734375, 0.0, 0.9285049438476562, 4.7572021484375, 2.0410213470458984, -1.7308502197265625, 4.40069580078125, 3.3898544311523438, 5.8773040771484375, 1.8167037963867188, -1.3363208770751953, 3.0390625, -3.383270263671875, -4.55767822265625, 0.14824295043945312, -3.9808197021484375, 5.6263580322265625, -0.0691986083984375, 7.33660888671875, -0.56890869140625, 2.98773193359375, 6.000244140625, -3.780548095703125, 1.073486328125, 0.8930034637451172, 2.36346435546875, 3.034320831298828, 5.8564453125, -1.3252716064453125, 2.1660079956054688, 0.25127601623535156, 0.7340126037597656, -7.611907958984375, -1.1529541015625, 0.5585155487060547, 6.1144256591796875, 2.7230224609375, 0.12682533264160156, 4.692596435546875, -7.311859130859375, 8.74444580078125, 7.585601806640625, 3.653228759765625, 0.25662994384765625, 1.7327842712402344, 1.3217620849609375, -1.8502120971679688, 2.290985107421875, -0.07672119140625, -5.72021484375, 0.1230010986328125, 5.932037353515625, 0.3258209228515625, 0.7644500732421875, 4.2468719482421875, 4.59295654296875, -9.924072265625, 4.2655029296875, 3.2340011596679688, -6.55804443359375, 3.85577392578125, 1.380523681640625, 2.204437255859375, 2.5172958374023438, 4.519805908203125, 2.90728759765625, 0.8214645385742188, -0.7540664672851562, 0.25617218017578125, 0.96990966796875, 1.2830657958984375, -1.400604248046875, -6.364044189453125, -0.939208984375, 1.5608482360839844, 2.27130126953125, 3.5064697265625, 0.6409530639648438, -1.0933685302734375, -0.46063995361328125, 3.365478515625, 2.948760986328125, 0.3612785339355469, 3.347686767578125, 0.643829345703125, 3.63983154296875, 1.5200996398925781, 4.088653564453125, 0.5312652587890625, 0.5061492919921875, -1.012969970703125, 1.01983642578125, -0.9446601867675781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000067.npy"}
{"epoch": 0.14031413612565444, "step": 68, "batch_size": 128, "mean": 1.705453872680664, "std": 4.03614616394043, "min": -7.71148681640625, "p10": -3.1273330688476557, "median": 1.1689605712890625, "p90": 6.3916580200195305, "max": 16.3983154296875, "pos_frac": 0.6875, "sample": [8.310005187988281, -0.08318710327148438, -1.46588134765625, -2.815093994140625, -0.8885345458984375, -0.026617050170898438, 1.21044921875, 5.269927978515625, -0.7222366333007812, 2.08953857421875, 1.0948867797851562, -0.6996650695800781, 0.94866943359375, 3.793426513671875, 4.7187957763671875, -4.016021728515625, 0.11272430419921875, 0.174560546875, 1.90020751953125, 2.558002471923828, 6.3237762451171875, 1.009918212890625, -2.5479774475097656, 1.178466796875, 16.3983154296875, 0.7255153656005859, 1.549123764038086, 3.263427734375, 0.28832244873046875, -5.0760498046875, 5.03961181640625, 3.16796875, 10.71923828125, 7.5916748046875, 0.34020042419433594, 6.32305908203125, -0.7303619384765625, 1.24755859375, -2.73004150390625, -5.247802734375, -0.447479248046875, -0.1514739990234375, 3.89434814453125, 3.70947265625, 1.0577926635742188, -7.430511474609375, 0.0227508544921875, 2.559844970703125, 5.6849365234375, 7.214263916015625, 2.279296875, -3.5798568725585938, 3.2305908203125, 6.69134521484375, 10.6796875, -0.135223388671875, 4.1065673828125, 5.958160400390625, -0.7801399230957031, 1.4567489624023438, 4.232940673828125, 1.6046295166015625, -0.7496376037597656, 2.374114990234375, 0.28757667541503906, -0.23810958862304688, -7.71148681640625, 4.676849365234375, 1.02642822265625, 11.1148681640625, 0.42672157287597656, -5.0411376953125, 1.4517822265625, 0.270111083984375, -4.94140625, 2.3835296630859375, -0.791595458984375, -3.5506515502929688, 9.818939208984375, -0.3047943115234375, -2.9617462158203125, 1.44708251953125, 0.02936553955078125, 4.68634033203125, 0.44195556640625, 1.583984375, -0.552581787109375, 6.0203704833984375, -7.665283203125, 9.480262756347656, 3.33074951171875, 1.159454345703125, 0.5670547485351562, -1.87408447265625, 0.5921630859375, 5.94464111328125, -4.950164794921875, -3.513702392578125, 5.090850830078125, 2.1509323120117188, 4.5223388671875, 0.95294189453125, 1.9954833984375, 6.550048828125, 8.494171142578125, 0.46307373046875, 5.5747222900390625, 6.285614013671875, 2.7119140625, 4.6996612548828125, 2.41168212890625, -3.6477127075195312, 1.5297088623046875, 1.9650344848632812, -1.10015869140625, 1.3321952819824219, 5.103851318359375, 0.901336669921875, 1.19793701171875, 1.1121139526367188, -0.5131988525390625, -1.8542022705078125, -1.231781005859375, 12.541107177734375, 3.5130615234375, -0.9203338623046875, -0.9999465942382812, 1.0410995483398438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000068.npy"}
{"epoch": 0.1424083769633508, "step": 69, "batch_size": 128, "mean": 2.141019344329834, "std": 3.882119655609131, "min": -9.983306884765625, "p10": -2.1520965576171873, "median": 1.7254562377929688, "p90": 7.460542297363281, "max": 13.61883544921875, "pos_frac": 0.75, "sample": [4.434654235839844, -0.596527099609375, 2.32586669921875, 5.3647308349609375, 0.50897216796875, -9.18646240234375, 3.2140655517578125, 1.5315437316894531, -2.297607421875, -0.28040504455566406, 3.8613204956054688, 1.3352928161621094, 10.922393798828125, 0.5898246765136719, 7.76580810546875, -0.4437103271484375, 7.5479888916015625, 0.6208000183105469, 3.30450439453125, 1.265411376953125, -0.5089569091796875, 0.17643356323242188, -3.1596336364746094, -4.858428955078125, 6.96295166015625, 10.0045166015625, 5.15679931640625, 5.216827392578125, 0.5767593383789062, 2.068389892578125, -9.983306884765625, 4.00970458984375, -0.3378753662109375, 1.454437255859375, 2.5710601806640625, -0.9671630859375, 0.9820384979248047, -0.14327239990234375, 8.098358154296875, 2.7220458984375, 8.69122314453125, -5.6827392578125, 0.04943275451660156, 2.8198699951171875, 1.5337600708007812, 1.4571113586425781, -4.0915069580078125, 7.423065185546875, 3.881420135498047, 0.1693878173828125, 6.99639892578125, 0.135498046875, 7.658721923828125, 5.163116455078125, 5.425872802734375, -6.58587646484375, -5.3486328125, 0.2602996826171875, 1.51239013671875, -0.3980598449707031, 0.7644195556640625, 5.84906005859375, 0.5135307312011719, 3.6403350830078125, 0.756256103515625, 3.5011138916015625, 2.0028839111328125, 4.9493560791015625, 5.977325439453125, 5.334228515625, 10.8861083984375, 3.662017822265625, 2.433135986328125, 0.572662353515625, 3.08795166015625, -0.1772308349609375, 6.77899169921875, 1.1833877563476562, 2.8176116943359375, 2.14410400390625, -1.411264419555664, -0.597869873046875, 2.063610076904297, 1.2492713928222656, -1.587615966796875, 0.877349853515625, 2.58465576171875, 1.7338714599609375, 4.3600311279296875, 7.939117431640625, 6.01995849609375, 7.9658050537109375, 1.717041015625, 4.75543212890625, -2.40557861328125, -2.398040771484375, 0.1531982421875, 0.9492568969726562, 8.382080078125, -0.5436325073242188, 3.9083709716796875, 2.442352294921875, 13.61883544921875, 1.19775390625, 5.615264892578125, 3.970489501953125, -2.291015625, 2.2742767333984375, -1.366119384765625, 2.2089385986328125, -0.2075347900390625, -2.092559814453125, 0.263153076171875, 6.761749267578125, 5.337646484375, -5.3463134765625, -1.7192153930664062, 0.08648681640625, 3.2086257934570312, 8.78302001953125, 1.922149658203125, 3.4227294921875, 0.694366455078125, -0.6545181274414062, -0.06328964233398438, 3.910491943359375, 5.464691162109375, 1.3148193359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000069.npy"}
{"epoch": 0.14450261780104712, "step": 70, "batch_size": 128, "mean": 2.2050421237945557, "std": 3.7933402061462402, "min": -9.73211669921875, "p10": -1.2019340515136716, "median": 1.5714645385742188, "p90": 7.509873962402343, "max": 14.9951171875, "pos_frac": 0.75, "sample": [-0.83221435546875, 3.5026397705078125, 2.1096420288085938, -2.0329208374023438, 0.289031982421875, 0.7584991455078125, 4.3184814453125, 3.3388214111328125, -6.7371826171875, 9.876708984375, -1.4758834838867188, -0.031673431396484375, -0.734344482421875, 4.08544921875, 6.57257080078125, 5.374839782714844, 2.4241180419921875, 3.9635009765625, 0.9544944763183594, 3.2108154296875, 1.0385360717773438, 6.536041259765625, 0.020172119140625, 2.098299026489258, -4.1151123046875, 2.6865081787109375, 0.43088531494140625, 2.4444732666015625, 3.0897598266601562, -0.6463699340820312, 7.05963134765625, 1.2594146728515625, 0.9744873046875, 1.28643798828125, 1.13165283203125, 0.5550537109375, 7.4477691650390625, -3.0869903564453125, -3.86505126953125, 1.239990234375, 1.5559234619140625, 0.22736740112304688, 9.24481201171875, 2.0771331787109375, 1.7786674499511719, -1.411834716796875, 0.5755157470703125, 0.6700897216796875, -1.1119766235351562, -1.54583740234375, 1.55828857421875, 0.8682403564453125, -0.5446319580078125, 0.025356292724609375, -0.7012977600097656, 4.1597900390625, 10.006988525390625, 3.589202880859375, 7.65478515625, 1.081894874572754, 2.4451904296875, 3.673309326171875, -0.543182373046875, 1.2145843505859375, 3.3240966796875, 4.094390869140625, 0.7848892211914062, 2.41485595703125, 1.4494361877441406, 0.061676025390625, -3.6170654296875, 1.6335906982421875, -0.04950904846191406, 2.4721832275390625, 1.7635879516601562, 9.928955078125, -0.692291259765625, 1.5846405029296875, 9.4886474609375, 3.196533203125, 12.913177490234375, 5.2412109375, 2.7140731811523438, 4.2606201171875, 6.1016998291015625, 1.745574951171875, -0.6102142333984375, 2.35223388671875, -0.78057861328125, 9.83673095703125, 5.310997009277344, 6.99530029296875, 0.0, -0.027660369873046875, 2.428955078125, 8.71868896484375, 5.7626800537109375, 3.6619873046875, 0.6776580810546875, 12.841796875, 3.67742919921875, 7.7431640625, 0.72503662109375, -1.9598388671875, 2.2943801879882812, -1.0705718994140625, 8.093597412109375, 0.8516998291015625, 3.689666748046875, -1.8934173583984375, 0.25616455078125, -1.02349853515625, 0.9228057861328125, 3.056427001953125, -6.8248291015625, -9.73211669921875, 2.8444976806640625, 0.12635040283203125, 2.229583740234375, 5.354705810546875, 2.8912811279296875, 2.5600128173828125, 1.53851318359375, -0.31868934631347656, 14.9951171875, 0.0, 0.5715866088867188, -0.40656280517578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000070.npy"}
{"epoch": 0.14659685863874344, "step": 71, "batch_size": 128, "mean": 1.9379534721374512, "std": 4.457602024078369, "min": -12.66436767578125, "p10": -2.9810363769531247, "median": 1.9451522827148438, "p90": 8.13226318359375, "max": 11.99444580078125, "pos_frac": 0.6875, "sample": [-0.3206787109375, 11.99444580078125, -0.8838844299316406, -4.614715576171875, -1.1756591796875, -7.9329833984375, -1.0993194580078125, 10.02886962890625, -8.634185791015625, 3.8182830810546875, 1.1292304992675781, 3.569072723388672, -2.965087890625, 4.7069091796875, 1.99383544921875, 5.775787353515625, 4.7122344970703125, -0.740020751953125, 3.1514739990234375, 8.91094970703125, 2.6825027465820312, -1.20654296875, 2.7848663330078125, 11.768157958984375, 1.9975128173828125, -0.675201416015625, 1.3565597534179688, 7.9085693359375, 2.1473312377929688, 4.658905029296875, -3.01824951171875, 5.368995666503906, 1.672515869140625, 9.334014892578125, -2.160064697265625, 4.55560302734375, 8.182373046875, 0.5770511627197266, 6.94964599609375, -5.681976318359375, -1.26702880859375, -0.9219207763671875, -4.478515625, 2.513439178466797, 1.56927490234375, -0.47113037109375, 1.4813232421875, -12.40576171875, -2.78076171875, -0.6916961669921875, 2.14849853515625, 4.2388916015625, -1.6281700134277344, 4.700778961181641, 1.2902984619140625, 5.46331787109375, 11.1690673828125, 1.7711181640625, -0.35276222229003906, 1.18182373046875, 3.7306365966796875, -1.316802978515625, 8.204925537109375, 7.90087890625, 0.2259082794189453, -0.37554931640625, -0.1382732391357422, -1.764251708984375, 3.48956298828125, -0.7836990356445312, 1.176849365234375, 10.674072265625, 3.6451416015625, -5.5291748046875, -4.057025909423828, 4.652801513671875, 7.0001220703125, -5.70062255859375, -1.9384918212890625, 0.6748008728027344, -2.0350341796875, -1.783966064453125, 0.2296142578125, 2.5105743408203125, 8.916412353515625, 7.3272705078125, 0.083740234375, 3.67547607421875, 5.532958984375, 3.98486328125, 1.4251651763916016, 4.64837646484375, 0.947052001953125, 0.41736412048339844, 0.6054916381835938, 9.163040161132812, -0.555023193359375, 8.141021728515625, 3.386322021484375, 4.03509521484375, 7.8583984375, -3.6698226928710938, 0.2910041809082031, 5.32305908203125, 8.128509521484375, -0.16100311279296875, 2.116424560546875, 1.0974254608154297, -3.93609619140625, 2.0449676513671875, 3.5714263916015625, 1.625650405883789, 3.6514892578125, 2.3316421508789062, 1.220855712890625, 2.1269702911376953, 2.867645263671875, 4.465606689453125, 2.4871788024902344, 2.4909515380859375, 2.45635986328125, 1.8964691162109375, 2.273193359375, 0.23191070556640625, -0.2442188262939453, 11.814483642578125, -12.66436767578125, 2.777099609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000071.npy"}
{"epoch": 0.1486910994764398, "step": 72, "batch_size": 128, "mean": 1.9835209846496582, "std": 4.25370454788208, "min": -10.157028198242188, "p10": -2.989849853515625, "median": 1.8829936981201172, "p90": 7.628860473632811, "max": 14.169769287109375, "pos_frac": 0.6640625, "sample": [2.6107330322265625, 3.0842132568359375, -1.6918792724609375, 1.247161865234375, 11.107666015625, -2.395050048828125, 1.0871429443359375, 4.1703948974609375, 2.4695281982421875, -0.9188079833984375, -1.02581787109375, 0.762451171875, 2.2220306396484375, 2.967132568359375, 4.72027587890625, 9.70208740234375, -3.2781600952148438, 1.7799072265625, -4.062744140625, 0.492340087890625, 3.2909317016601562, 6.431640625, 2.765899658203125, 10.84619140625, 3.55706787109375, 0.3664093017578125, -0.5704421997070312, 2.9926605224609375, -4.404655456542969, 0.45824432373046875, 3.6280364990234375, -10.157028198242188, -1.118682861328125, -5.704109191894531, 1.5164642333984375, 1.9361572265625, 0.706298828125, -1.0505790710449219, 2.001678466796875, 1.1867141723632812, 5.7098388671875, -1.28515625, 8.15234375, 5.914794921875, 10.175369262695312, 3.28631591796875, 2.7590789794921875, -0.330078125, -5.36273193359375, -2.9407958984375, -2.8417816162109375, 5.5255126953125, -2.989776611328125, 0.0, 6.84326171875, 5.96331787109375, 8.556976318359375, -0.22576141357421875, 4.2911376953125, 10.06280517578125, -0.1461029052734375, -2.603912353515625, 3.9877281188964844, 5.804878234863281, 3.18316650390625, 4.221710205078125, 1.2629241943359375, 0.851654052734375, -0.24437332153320312, 2.749267578125, -4.80364990234375, 2.095823287963867, 0.19561767578125, 3.666748046875, 8.753997802734375, 1.2720184326171875, 0.2218799591064453, 5.5970001220703125, -0.2640380859375, 9.31829833984375, 0.6781768798828125, 2.0976219177246094, -1.7736740112304688, -1.70501708984375, -2.506439208984375, -2.990020751953125, 0.978851318359375, 0.3295440673828125, -1.8750686645507812, 7.404510498046875, -5.223876953125, 9.559814453125, -0.2247772216796875, 4.066619873046875, 1.991546630859375, 6.4069061279296875, 5.97821044921875, -1.8726348876953125, 1.9467010498046875, 6.46307373046875, 2.37603759765625, 7.0140228271484375, 0.0, 5.446746826171875, -3.2460670471191406, 0.9934120178222656, 6.6692962646484375, 14.169769287109375, 5.9588623046875, 7.20062255859375, 4.417106628417969, -1.9610748291015625, -0.5147857666015625, -7.247955322265625, 9.92559814453125, -1.7045669555664062, -1.5337448120117188, 0.0, 1.3716583251953125, 3.2414703369140625, 1.9122047424316406, 1.8537826538085938, -7.586029052734375, 3.92047119140625, 2.6025543212890625, -3.14306640625, 3.669189453125, 8.24029541015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000072.npy"}
{"epoch": 0.15078534031413612, "step": 73, "batch_size": 128, "mean": 1.9467735290527344, "std": 4.650934219360352, "min": -17.18292236328125, "p10": -3.531435394287109, "median": 1.734823226928711, "p90": 7.494305419921875, "max": 13.582427978515625, "pos_frac": 0.71875, "sample": [1.7577705383300781, 2.40093994140625, 8.685546875, 4.144561767578125, -0.4810791015625, 2.2131195068359375, 0.69232177734375, -1.0765380859375, -0.5247344970703125, 6.759284973144531, 4.12066650390625, 2.1580810546875, 1.397735595703125, 10.1318359375, 0.1666889190673828, -0.09018707275390625, -7.149658203125, -10.03546142578125, 1.5487823486328125, 0.1565704345703125, 2.275238037109375, 0.1148529052734375, 0.14802932739257812, 2.457916259765625, 0.49399566650390625, 0.1572113037109375, 1.4936904907226562, 1.8616828918457031, 1.613800048828125, 4.4927825927734375, -6.43084716796875, -1.9111785888671875, 4.427703857421875, 9.248016357421875, -2.346588134765625, 1.5679855346679688, -3.1221160888671875, 12.72515869140625, 5.180290222167969, 4.3823089599609375, 5.35906982421875, 1.3921051025390625, 5.5865020751953125, -1.987579345703125, 13.582427978515625, -1.99658203125, -2.337982177734375, 0.34149169921875, 2.844970703125, 11.300140380859375, -6.25567626953125, 1.845703125, 8.7425537109375, -17.18292236328125, 3.357818603515625, 1.56134033203125, 2.04193115234375, -5.482757568359375, 4.608978271484375, -2.7783203125, 4.51458740234375, 1.9842529296875, 3.7585372924804688, -0.636932373046875, -2.082489013671875, 7.62371826171875, 4.692211151123047, -4.5828857421875, 7.4388427734375, 0.38299560546875, -0.7968292236328125, -4.037994384765625, 4.579254150390625, 5.60821533203125, 0.840667724609375, 1.40576171875, -6.2691497802734375, -6.342864990234375, -1.695526123046875, 1.4801483154296875, 6.5272216796875, 4.687713623046875, 4.7669677734375, 0.5987701416015625, -0.8822021484375, 2.962738037109375, -0.5484733581542969, 2.944774627685547, -0.5484466552734375, 4.944549560546875, 1.3493804931640625, 1.43701171875, 7.194580078125, 3.6778564453125, 5.803558349609375, 7.25665283203125, 2.1094436645507812, 2.7642822265625, -6.148223876953125, 5.063690185546875, -0.5549392700195312, 6.635162353515625, 4.388710021972656, 4.627685546875, 4.38177490234375, 4.8524932861328125, 0.1069183349609375, 1.437255859375, 13.12799072265625, 7.9005126953125, 4.38275146484375, -3.2196044921875, 1.8056793212890625, -0.1326446533203125, 10.063156127929688, -3.9228515625, 0.6241703033447266, 7.075714111328125, -3.3636856079101562, -0.5878982543945312, 9.634796142578125, 1.7118759155273438, 8.134735107421875, 2.6974639892578125, 1.1927566528320312, 0.8874740600585938, 5.55389404296875, -4.4300994873046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000073.npy"}
{"epoch": 0.15287958115183245, "step": 74, "batch_size": 128, "mean": 1.8589857816696167, "std": 4.795709133148193, "min": -10.66729736328125, "p10": -4.654179382324219, "median": 1.836212158203125, "p90": 8.01158447265625, "max": 15.21405029296875, "pos_frac": 0.6796875, "sample": [2.09649658203125, 7.947021484375, 2.455474853515625, 1.10137939453125, 8.36566162109375, -0.9530029296875, -4.8563232421875, 1.9910888671875, 6.2106170654296875, 6.575927734375, -1.456573486328125, 8.713455200195312, 0.0, 1.497955322265625, 0.083038330078125, 10.907470703125, 5.215087890625, 10.896209716796875, -1.75946044921875, 0.6546173095703125, 3.44476318359375, 3.7845458984375, 1.26690673828125, 2.6630325317382812, 8.80218505859375, -6.951690673828125, -4.66455078125, -0.49139404296875, -1.95745849609375, -3.6550445556640625, -2.66644287109375, -0.44970703125, -3.82122802734375, -0.7814102172851562, 5.004058837890625, 1.4035186767578125, 0.451385498046875, 6.660350799560547, 0.03993797302246094, 7.8803558349609375, -0.36486244201660156, -2.417327880859375, -9.157684326171875, -6.89129638671875, 1.00262451171875, 3.192474365234375, -0.20477294921875, -2.039947509765625, -0.15570068359375, 4.4583892822265625, -9.715057373046875, -10.19927978515625, -0.57708740234375, 2.702392578125, 1.821990966796875, -1.040414810180664, 1.7353477478027344, 1.97601318359375, -8.56378173828125, 6.752044677734375, 4.647369384765625, 4.0796661376953125, 0.18341064453125, 8.1622314453125, 7.8857421875, 3.344482421875, 8.58905029296875, 4.71221923828125, 1.5319900512695312, 15.21405029296875, 1.488525390625, -0.0241241455078125, -1.0994796752929688, 5.495941162109375, -0.631622314453125, 2.0556640625, 1.850433349609375, 3.23553466796875, 0.0829010009765625, 4.5832366943359375, 2.8715057373046875, 0.8282699584960938, 2.1307373046875, -10.66729736328125, 11.92669677734375, 1.9905853271484375, 0.3262939453125, 1.207855224609375, 1.3126945495605469, -0.8100852966308594, -0.0744171142578125, -4.6497344970703125, 0.19818115234375, 5.5805511474609375, 3.337890625, 4.59814453125, 4.466400146484375, -7.815216064453125, -6.161346435546875, 5.7315673828125, 5.5388031005859375, -5.041412353515625, 2.05279541015625, 8.841827392578125, 2.01214599609375, 6.265720367431641, 2.87103271484375, -4.99273681640625, 5.100860595703125, -2.46636962890625, -0.5006523132324219, 1.8548049926757812, 4.8341064453125, 2.4982757568359375, 12.5565185546875, -0.732757568359375, 1.3844528198242188, 4.9901123046875, 3.4890289306640625, 1.006591796875, 4.068511962890625, 2.5455322265625, 1.3334808349609375, 8.537521362304688, 13.2509765625, 3.8312530517578125, -0.2908935546875, 7.43182373046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000074.npy"}
{"epoch": 0.1549738219895288, "step": 75, "batch_size": 128, "mean": 2.7258448600769043, "std": 4.515727996826172, "min": -10.37298583984375, "p10": -2.6929199218749997, "median": 2.2486677169799805, "p90": 8.104684448242187, "max": 13.95831298828125, "pos_frac": 0.7734375, "sample": [2.5231399536132812, 2.9738540649414062, 1.3269577026367188, 1.335845947265625, -1.353515625, 7.95147705078125, 6.5435028076171875, -2.662322998046875, 4.8143157958984375, 3.1553268432617188, 0.34989166259765625, -4.064208984375, 1.767242431640625, -1.0053253173828125, 8.291412353515625, 0.2785511016845703, 13.95831298828125, 3.5546875, 7.37786865234375, 3.9424591064453125, 7.7430419921875, 6.342342376708984, 5.387847900390625, -3.752593994140625, 0.0393829345703125, 12.22369384765625, -2.55487060546875, 5.2996826171875, 0.62103271484375, 1.9314117431640625, 2.619873046875, 6.7139892578125, 2.96630859375, 2.743194580078125, 0.3305206298828125, -1.8016395568847656, 3.8377532958984375, 3.31317138671875, 3.3184280395507812, 0.81512451171875, 12.0042724609375, 6.271869659423828, 1.15478515625, -10.37298583984375, 1.45245361328125, 1.49072265625, 0.4513816833496094, -0.23382568359375, 7.97491455078125, 0.44622802734375, 4.681243896484375, 4.2779388427734375, 1.9398193359375, 0.39556884765625, 4.67486572265625, 4.9777984619140625, -4.232940673828125, 9.721405029296875, -0.211822509765625, 8.024658203125, 1.21142578125, 2.4821224212646484, 1.0096359252929688, 8.389129638671875, 0.7467994689941406, 1.1092376708984375, 6.96246337890625, -2.769134521484375, 1.6850090026855469, 1.133392333984375, 2.743377685546875, -2.764312744140625, 1.6799983978271484, 11.65191650390625, 11.9227294921875, 0.3239574432373047, -8.4947509765625, -7.4625244140625, -0.427978515625, 4.106819152832031, 12.742431640625, 5.8677520751953125, 6.842315673828125, -5.377754211425781, -0.014827728271484375, 9.446975708007812, 0.7720146179199219, 0.2333984375, 3.2068557739257812, 6.544403076171875, 5.172576904296875, -0.8803253173828125, 4.94647216796875, 2.9091796875, 7.213165283203125, 11.902191162109375, -0.5784530639648438, 3.306884765625, 1.019866943359375, 7.0964202880859375, -6.4100494384765625, -0.011383056640625, 3.56744384765625, 8.437744140625, 7.35888671875, 5.058135986328125, -0.137725830078125, 5.5385894775390625, 1.864471435546875, -3.68304443359375, 6.201171875, -0.1292572021484375, 0.167144775390625, 5.8051605224609375, 4.615549087524414, 6.051025390625, 0.46295166015625, 1.2277984619140625, 6.1938629150390625, -3.00177001953125, 1.267822265625, 2.0152130126953125, 4.919185638427734, 10.8975830078125, -1.95697021484375, -2.347015380859375, -7.3558807373046875, 0.5951385498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000075.npy"}
{"epoch": 0.15706806282722513, "step": 76, "batch_size": 128, "mean": 2.9504592418670654, "std": 4.994762897491455, "min": -11.32183837890625, "p10": -2.061878967285156, "median": 2.1191635131835938, "p90": 9.695321655273435, "max": 21.734024047851562, "pos_frac": 0.7734375, "sample": [6.5687255859375, -2.90380859375, 1.6826629638671875, 2.3511199951171875, 3.4927291870117188, 4.3428192138671875, 2.7867431640625, 0.9891452789306641, 2.578428268432617, 4.237525939941406, 7.72589111328125, 0.17176055908203125, -0.609893798828125, -1.701873779296875, 13.71148681640625, 5.746826171875, -0.8195571899414062, -2.2405853271484375, 0.8471832275390625, -1.391082763671875, 13.087127685546875, 0.47541046142578125, 4.60675048828125, 2.270172119140625, 3.501739501953125, 11.9488525390625, 1.34228515625, 3.1509132385253906, 1.5712432861328125, 0.04180908203125, 2.8720703125, -0.22142791748046875, 6.0376739501953125, 7.6320037841796875, -9.73492431640625, 15.884979248046875, 3.6516265869140625, 5.3853302001953125, 3.18243408203125, 3.52716064453125, 10.18133544921875, 6.0833740234375, 7.440460205078125, 6.89080810546875, -6.12884521484375, 1.1301822662353516, 3.521514892578125, 3.0516510009765625, -0.222503662109375, 9.333221435546875, 0.0, 3.77435302734375, 0.5387115478515625, 0.59161376953125, 7.693046569824219, -3.317626953125, 11.99835205078125, -6.517578125, -1.828521728515625, -3.269378662109375, 5.0444488525390625, 11.679046630859375, -1.6053466796875, 0.16131591796875, 3.142242431640625, 1.1256904602050781, 3.03131103515625, 3.8893356323242188, 5.859306335449219, -2.2541351318359375, -0.2295379638671875, 9.487030029296875, -2.631591796875, 0.7990798950195312, -2.9674224853515625, 0.0, 10.835037231445312, 5.1536102294921875, 4.6319732666015625, 11.812896728515625, 2.1805343627929688, 0.25989341735839844, 1.770111083984375, 2.0577926635742188, 0.7024765014648438, 2.2931442260742188, 1.2743873596191406, 8.15484619140625, 1.21875, 3.2682647705078125, 3.6344146728515625, 1.3331756591796875, 0.9756622314453125, 1.6601715087890625, 0.07606124877929688, -1.5573043823242188, -4.1150054931640625, 0.117889404296875, 2.012939453125, -0.7035751342773438, 5.627197265625, 1.9435272216796875, 7.56597900390625, 13.70367431640625, 9.10736083984375, 16.391815185546875, 11.0439453125, 2.644073486328125, 2.7506103515625, -3.2924346923828125, 0.684967041015625, -1.5244140625, 21.734024047851562, -1.98529052734375, -11.32183837890625, 1.5717315673828125, 3.229248046875, 2.822174072265625, 8.15618896484375, 1.6819992065429688, 1.882080078125, 6.7620849609375, 1.8325424194335938, 1.17938232421875, -1.7182159423828125, 2.5293350219726562, 0.2038421630859375, 0.150634765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000076.npy"}
{"epoch": 0.15916230366492146, "step": 77, "batch_size": 128, "mean": 2.4621851444244385, "std": 5.391096591949463, "min": -12.59893798828125, "p10": -3.7145103454589843, "median": 1.961090087890625, "p90": 9.25700225830078, "max": 19.49615478515625, "pos_frac": 0.6640625, "sample": [-0.9206390380859375, 2.0054168701171875, 1.9167633056640625, 4.5190887451171875, -1.416748046875, 0.8023605346679688, 10.827362060546875, 5.9672393798828125, 0.00054168701171875, 5.4253082275390625, -4.005401611328125, -0.1773834228515625, 9.19647216796875, 3.2716064453125, 11.807907104492188, 1.5450820922851562, 1.0806121826171875, 2.8026046752929688, -12.59893798828125, 6.795135498046875, -1.4086036682128906, -0.24853515625, -0.573333740234375, -1.96673583984375, 6.47613525390625, 9.398239135742188, 4.47607421875, 2.3975830078125, 3.909637451171875, 3.4766845703125, -3.706817626953125, -0.660125732421875, 1.254913330078125, 8.285888671875, -1.86407470703125, 1.1331977844238281, 2.1430892944335938, 2.9400634765625, 8.715087890625, -1.3409500122070312, -0.3426361083984375, 0.775909423828125, 6.509429931640625, 0.111480712890625, 3.8964614868164062, 3.055145263671875, 0.22971343994140625, 19.0770263671875, -1.3582382202148438, 0.080413818359375, 6.00396728515625, -1.05096435546875, 3.774139404296875, 10.185333251953125, 7.9721527099609375, 5.434539794921875, 3.415374755859375, -1.35198974609375, 0.2260284423828125, 0.02878570556640625, -0.2030029296875, -3.7324600219726562, 8.9652099609375, 19.49615478515625, -1.41583251953125, 7.4219970703125, 4.431236267089844, -0.2380218505859375, 10.938385009765625, 3.2453994750976562, 9.70489501953125, 2.708251953125, -8.05987548828125, -2.8848876953125, -3.8338241577148438, -2.015869140625, 6.506317138671875, 0.8487167358398438, 1.606689453125, 2.59088134765625, -2.2718505859375, 11.42156982421875, -3.3319168090820312, -3.062469482421875, 1.6150588989257812, 4.237091064453125, 8.979888916015625, -9.125213623046875, -4.061309814453125, 3.366455078125, 2.4802780151367188, 2.9345321655273438, -2.1102447509765625, 3.53436279296875, 5.1093292236328125, -3.75396728515625, -1.3842926025390625, -6.253814697265625, 8.269561767578125, 5.6353759765625, -4.629241943359375, 6.029998779296875, -1.0070457458496094, 4.14190673828125, 6.4165191650390625, 1.347686767578125, 0.1804351806640625, 10.6177978515625, 13.1072998046875, -0.9342498779296875, 2.1329917907714844, -5.52325439453125, -7.5776824951171875, 16.86700439453125, -0.312042236328125, 3.572998046875, -5.034881591796875, 2.218780517578125, -2.4867591857910156, 0.46661376953125, 6.8760986328125, 1.3449554443359375, 4.25738525390625, 8.732666015625, 5.462982177734375, 15.0302734375, -3.3090438842773438, 0.5068206787109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000077.npy"}
{"epoch": 0.1612565445026178, "step": 78, "batch_size": 128, "mean": 2.950676441192627, "std": 5.390461444854736, "min": -12.357925415039062, "p10": -2.9923072814941403, "median": 2.5267562866210938, "p90": 10.11286926269531, "max": 24.91583251953125, "pos_frac": 0.734375, "sample": [-0.94586181640625, 6.741851806640625, -3.61676025390625, 5.01959228515625, 8.642059326171875, 1.6988906860351562, 2.933258056640625, 3.795318603515625, -4.043785095214844, 5.161598205566406, 0.44183349609375, 2.1085891723632812, 4.539314270019531, 4.764190673828125, -1.0470104217529297, 1.807037353515625, 3.59710693359375, 5.1723175048828125, 10.396774291992188, 0.4503631591796875, 0.0, 10.360992431640625, -4.54107666015625, 2.8095245361328125, 0.5123291015625, 2.265644073486328, 5.72467041015625, 2.9967918395996094, -2.399322509765625, 5.3925933837890625, -2.2267227172851562, 5.65338134765625, 8.871719360351562, 0.281341552734375, 7.151458740234375, -2.5306854248046875, 1.005615234375, 15.02191162109375, 0.05291748046875, -4.8685302734375, 4.566719055175781, 5.553497314453125, -0.8536453247070312, 24.91583251953125, 12.4041748046875, -3.84295654296875, 4.43157958984375, 3.8173980712890625, 0.2513427734375, 3.68634033203125, 4.3109130859375, -12.357925415039062, 2.550872802734375, 3.4164199829101562, 0.50579833984375, -0.5267314910888672, 5.4171600341796875, 4.41522216796875, 2.8295326232910156, -1.5467109680175781, 10.00653076171875, -2.26251220703125, 5.1266021728515625, 6.8431549072265625, 1.6672210693359375, -3.2119064331054688, 1.3306503295898438, 4.135894775390625, 4.0054931640625, -9.04150390625, 15.49517822265625, 3.0117263793945312, 2.92279052734375, 1.1166458129882812, 11.41668701171875, 6.779327392578125, -0.6228504180908203, 2.5026397705078125, -1.4451179504394531, 1.0255584716796875, -4.253143310546875, -1.3010635375976562, 9.100982666015625, 15.566680908203125, 1.0063323974609375, 2.8235397338867188, -0.63494873046875, 3.5810546875, -6.2507171630859375, 6.26171875, 2.8517799377441406, -2.5023956298828125, 8.555953979492188, -1.2248916625976562, -9.081787109375, 1.2426986694335938, 1.516632080078125, -4.08856201171875, 12.517333984375, 0.112884521484375, 0.8415260314941406, 0.7651290893554688, -1.3275985717773438, 8.89483642578125, 6.301666259765625, 3.5224761962890625, 5.901641845703125, -0.9850311279296875, 1.72314453125, 2.0765914916992188, 12.508514404296875, 0.765899658203125, 2.98956298828125, 6.5539093017578125, 4.90484619140625, 2.3600196838378906, 6.203277587890625, 10.713592529296875, 10.814422607421875, -3.9155731201171875, -2.898193359375, -0.15692138671875, 17.313690185546875, 1.4105110168457031, 1.50714111328125, 2.2212753295898438, -0.6427459716796875, 7.62060546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000078.npy"}
{"epoch": 0.16335078534031414, "step": 79, "batch_size": 128, "mean": 3.153383255004883, "std": 6.920065879821777, "min": -39.7431640625, "p10": -3.127577209472656, "median": 2.0979156494140625, "p90": 10.724127197265625, "max": 26.47760009765625, "pos_frac": 0.6875, "sample": [0.22371673583984375, 5.9172210693359375, 9.054183959960938, -2.1820068359375, 6.98114013671875, -2.1490478515625, 1.7898406982421875, -0.810272216796875, 2.7464141845703125, -3.076751708984375, 12.046417236328125, -1.87225341796875, 0.0, 8.57623291015625, -0.22916603088378906, -0.6213150024414062, -0.175994873046875, -3.1253204345703125, -5.859626770019531, -0.46405029296875, 2.3102264404296875, -2.692718505859375, -3.75079345703125, -1.9215583801269531, -3.719970703125, 4.4745941162109375, 3.2881317138671875, 0.0, 8.549545288085938, 0.3555278778076172, -4.912384033203125, 11.647796630859375, 4.642463684082031, 4.990570068359375, 4.167144775390625, 0.932098388671875, 4.115997314453125, 1.683685302734375, 4.105438232421875, -2.414154052734375, 4.620849609375, -7.850761413574219, 9.87066650390625, 0.2748870849609375, 0.4753570556640625, 5.3727874755859375, 6.279758453369141, 15.925811767578125, 1.778167724609375, 2.095947265625, 1.716522216796875, -3.3787841796875, -39.7431640625, -5.123504638671875, 1.7308807373046875, 7.92108154296875, 1.668243408203125, 15.6220703125, 9.495620727539062, 10.693603515625, -0.5099067687988281, 4.9442138671875, 3.292999267578125, 6.593658447265625, 10.79534912109375, 3.478179931640625, 3.9088592529296875, 2.729961395263672, -2.0562896728515625, 3.2750244140625, -0.158599853515625, 3.8395309448242188, 0.6015033721923828, 0.270782470703125, -2.905029296875, -4.81646728515625, 9.427825927734375, 0.648956298828125, 9.922637939453125, 2.099884033203125, -4.2940826416015625, 17.281707763671875, 9.64837646484375, 6.40081787109375, 3.5419921875, 6.00408935546875, 3.7477798461914062, 0.2743721008300781, 1.1336898803710938, 0.7631511688232422, 0.1335773468017578, 5.843360900878906, 0.43032073974609375, 0.963470458984375, 1.6793785095214844, 26.47760009765625, -0.038089752197265625, 4.401336669921875, -3.6552276611328125, 9.996307373046875, 16.61614990234375, 11.721054077148438, 5.22601318359375, 5.1135406494140625, -3.132843017578125, 8.730083465576172, 7.541015625, -0.8061065673828125, 7.975128173828125, -1.0586166381835938, -3.41241455078125, 16.873016357421875, -0.2978515625, -0.5647850036621094, 5.228769302368164, 6.8758087158203125, 8.620834350585938, 4.14361572265625, 1.2629241943359375, 4.5152587890625, -0.29491615295410156, 8.503997802734375, 13.2156982421875, 19.734710693359375, -0.6086959838867188, -1.698211669921875, 1.1535568237304688, 14.27227783203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000079.npy"}
{"epoch": 0.16544502617801046, "step": 80, "batch_size": 128, "mean": 2.574650287628174, "std": 6.054333209991455, "min": -21.148284912109375, "p10": -3.095671081542968, "median": 1.6043701171875, "p90": 10.956109619140623, "max": 23.33538818359375, "pos_frac": 0.6875, "sample": [1.8957481384277344, 4.564094543457031, -2.0474090576171875, 0.8496971130371094, 0.43414306640625, -1.896453857421875, 1.50299072265625, -5.95721435546875, 7.06353759765625, -2.8307647705078125, -0.11159133911132812, -2.5398712158203125, 4.80078125, 2.70465087890625, 5.3753662109375, -2.999755859375, -0.11346435546875, 9.51507568359375, 3.933197021484375, 5.888214111328125, 16.602691650390625, -0.1775360107421875, 2.4149932861328125, 6.064697265625, 1.23870849609375, 11.347564697265625, 6.15869140625, -3.5608673095703125, 3.412668228149414, 11.283935546875, 2.052886962890625, 0.24547767639160156, 2.279541015625, -1.3998756408691406, -10.852996826171875, 1.33917236328125, -21.148284912109375, 5.52880859375, 0.0, 12.591796875, -1.6183624267578125, 8.685699462890625, -2.4218902587890625, -0.754730224609375, 1.48291015625, -8.9232177734375, 0.169952392578125, 5.4918975830078125, -1.319610595703125, -1.5036392211914062, 1.1027603149414062, 1.53179931640625, 1.67694091796875, 5.261474609375, 7.6937255859375, 0.4607353210449219, 0.4024028778076172, 0.6086807250976562, -2.169219970703125, 12.02197265625, 4.076507568359375, 2.0451507568359375, 6.028602600097656, 5.314544677734375, -3.875396728515625, 1.3778533935546875, 5.4083251953125, 0.8546066284179688, 6.720710754394531, 3.1956653594970703, -2.618133544921875, 4.76617431640625, 3.133580207824707, -4.286262512207031, 18.49346923828125, -1.990570068359375, -1.3767547607421875, 0.3067779541015625, 6.12249755859375, -7.283233642578125, 12.20745849609375, 3.636383056640625, 0.0322265625, 10.81561279296875, 5.2264404296875, 2.79620361328125, -5.0059356689453125, 1.07781982421875, -1.6822509765625, 14.704833984375, 6.416511535644531, 2.686553955078125, 12.089584350585938, 3.7156982421875, 3.7872676849365234, 13.152862548828125, 8.17401123046875, 0.8917694091796875, -9.309799194335938, 0.19513320922851562, -9.61798095703125, 12.11285400390625, -3.3194732666015625, 3.79437255859375, 3.4893798828125, 8.99761962890625, 0.97589111328125, -0.7768325805664062, 5.047821044921875, 8.636474609375, -2.170013427734375, -0.482208251953125, 0.02166748046875, -0.7127761840820312, 1.297576904296875, -4.690925598144531, 3.476411819458008, -1.4400787353515625, 0.7877120971679688, -2.86517333984375, 5.716339111328125, 10.378631591796875, 10.60882568359375, 3.995330810546875, 23.33538818359375, -0.3512916564941406, 3.1892929077148438, 14.76458740234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000080.npy"}
{"epoch": 0.16753926701570682, "step": 81, "batch_size": 128, "mean": 2.898014545440674, "std": 6.613247394561768, "min": -17.575439453125, "p10": -3.618984985351562, "median": 2.4909629821777344, "p90": 11.97888793945312, "max": 15.814163208007812, "pos_frac": 0.671875, "sample": [-1.012289047241211, -0.12455558776855469, 9.850120544433594, -9.902420043945312, 4.625, -0.62530517578125, -8.941162109375, 3.3985671997070312, 13.55059814453125, 7.14501953125, 15.814163208007812, 13.99090576171875, -2.3043746948242188, -2.56756591796875, 8.797576904296875, 0.45372581481933594, 4.5981292724609375, 8.45745849609375, 5.655914306640625, 4.29803466796875, 0.7611618041992188, 10.55474853515625, -0.0315093994140625, -2.8071823120117188, 5.30865478515625, 3.902587890625, 2.816680908203125, 0.317901611328125, 9.228179931640625, 6.71356201171875, 12.993896484375, 2.4149169921875, 8.178657531738281, 5.47320556640625, -1.35235595703125, 2.4145736694335938, -5.1141357421875, 15.632781982421875, 13.626815795898438, 2.70068359375, -5.240234375, 0.53985595703125, 11.54388427734375, -0.3807373046875, 0.0423583984375, 1.281951904296875, -16.313140869140625, -2.43060302734375, 7.341552734375, 11.1400146484375, 1.145904541015625, 2.7336578369140625, -6.950416564941406, 8.385162353515625, -4.5722503662109375, -0.49900054931640625, 5.869476318359375, 4.7008819580078125, 2.0166015625, -15.695266723632812, -5.90576171875, 2.4895553588867188, 3.88677978515625, 1.6640777587890625, -0.19338226318359375, 4.651203155517578, 2.1348876953125, 0.03115081787109375, 2.82635498046875, 4.6247100830078125, -0.449188232421875, -1.9453468322753906, -0.4039306640625, 11.005340576171875, 14.126007080078125, 9.707855224609375, -1.36126708984375, 4.5050201416015625, 1.7892818450927734, 6.814361572265625, -12.751220703125, 6.09820556640625, 10.278045654296875, -3.2796478271484375, 2.68084716796875, 1.721588134765625, -0.7167854309082031, -4.05377197265625, -1.037384033203125, 2.7954330444335938, -17.575439453125, -2.408843994140625, -15.145233154296875, 0.4859619140625, 1.8825759887695312, 2.06817626953125, -0.364288330078125, -0.06088829040527344, 13.32403564453125, 13.314697265625, 13.019012451171875, 13.3499755859375, 4.838714599609375, -1.02349853515625, 9.423309326171875, 2.5923290252685547, -2.2001953125, 1.022430419921875, -0.10323905944824219, 6.859222412109375, -3.432647705078125, 5.9300994873046875, 0.21447372436523438, -0.298370361328125, 3.0190277099609375, 10.902053833007812, 8.562271118164062, -1.6674041748046875, 10.633331298828125, 5.02618408203125, 6.59124755859375, 13.2662353515625, 14.222015380859375, 9.792770385742188, 2.49237060546875, 0.9771194458007812, 6.199127197265625, -0.0668792724609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000081.npy"}
{"epoch": 0.16963350785340314, "step": 82, "batch_size": 128, "mean": 2.9830164909362793, "std": 5.966424942016602, "min": -15.506195068359375, "p10": -4.052859497070313, "median": 3.0771865844726562, "p90": 10.936605834960936, "max": 23.8243408203125, "pos_frac": 0.7265625, "sample": [2.0187835693359375, 5.30474853515625, 0.8246231079101562, -7.281219482421875, -1.4323883056640625, 2.35467529296875, 2.5389404296875, 8.555023193359375, 6.2491302490234375, 0.16916275024414062, 1.2461090087890625, 6.67138671875, -4.740936279296875, -3.4286346435546875, -2.824005126953125, 6.8290557861328125, 4.11700439453125, 3.4453887939453125, 13.060684204101562, 5.742919921875, 1.36346435546875, -1.846038818359375, 7.11138916015625, -1.7413482666015625, -0.003093719482421875, 10.861358642578125, -9.945648193359375, 4.133644104003906, 3.37164306640625, 3.838531494140625, 17.907958984375, 2.0723876953125, 11.1121826171875, 5.072509765625, 10.679855346679688, 12.293212890625, -0.215972900390625, 4.006134033203125, 5.11083984375, -15.506195068359375, -3.686004638671875, 0.5025177001953125, 3.00433349609375, 4.506706237792969, -2.35302734375, 9.76080322265625, -3.002044677734375, 6.50360107421875, 16.7147216796875, -1.814453125, -0.9358062744140625, 16.415863037109375, 4.471794128417969, 4.19610595703125, 3.3270263671875, 0.4960365295410156, 13.732940673828125, -4.067108154296875, -0.171356201171875, 2.6203155517578125, 3.4907455444335938, -5.1307220458984375, -13.15765380859375, 3.624187469482422, -4.4140625, 0.5404281616210938, -4.8943939208984375, 1.8502120971679688, 11.26495361328125, 3.5540084838867188, -0.9048290252685547, 7.22259521484375, 2.421539306640625, 3.658294677734375, 0.1275177001953125, 3.24951171875, 0.12689208984375, 1.5467243194580078, 4.773338317871094, 2.7579116821289062, 1.2237205505371094, 23.8243408203125, -4.125823974609375, 8.2689208984375, -0.5728607177734375, 2.2216110229492188, 5.736358642578125, 10.274925231933594, 2.9339599609375, 3.1500396728515625, 3.6715316772460938, -2.1094512939453125, 2.808391571044922, 3.4892578125, 6.772966384887695, 7.02117919921875, -5.9212646484375, 2.0149993896484375, 4.838134765625, 11.8955078125, 5.145416259765625, -0.4142131805419922, 0.23979949951171875, 4.8259124755859375, 7.0462646484375, 3.8120288848876953, -0.0494537353515625, -0.40277099609375, -4.3509521484375, 2.5396270751953125, -2.811431884765625, 3.547332763671875, 6.9546661376953125, 8.065120697021484, -4.0467529296875, 12.40594482421875, 14.6968994140625, 2.6068801879882812, 4.0059051513671875, 5.9412841796875, 3.671051025390625, 4.9942626953125, 11.82171630859375, 6.0876312255859375, 1.0990009307861328, 1.2915210723876953, -11.6043701171875, -3.7360916137695312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000082.npy"}
{"epoch": 0.17172774869109947, "step": 83, "batch_size": 128, "mean": 3.5142087936401367, "std": 7.035341739654541, "min": -15.1707763671875, "p10": -3.419433593749999, "median": 2.6238861083984375, "p90": 12.436788940429686, "max": 30.06756591796875, "pos_frac": 0.7109375, "sample": [6.12335205078125, 0.0, 11.032958984375, 4.499603271484375, 1.6234130859375, -0.5080718994140625, -0.6124267578125, 7.110321044921875, 2.00054931640625, -0.43222808837890625, -1.645721435546875, -15.123291015625, 0.13201904296875, -11.955215454101562, 1.04791259765625, 1.0345458984375, 3.8329315185546875, 11.550933837890625, 8.053794860839844, 2.5235557556152344, 9.037322998046875, 6.8756103515625, -1.8322296142578125, 8.287277221679688, 10.284912109375, -3.914825439453125, -1.720672607421875, -0.3851318359375, -1.8194580078125, 12.977020263671875, 6.7090301513671875, 0.212127685546875, 4.44659423828125, 1.3560867309570312, -7.149993896484375, 15.0135498046875, 3.5646438598632812, 2.702789306640625, 20.685012817382812, 0.6007843017578125, 4.842800140380859, -1.8766403198242188, 3.1839828491210938, 3.716400146484375, -4.400482177734375, 12.117431640625, 5.373016357421875, 5.491035461425781, -0.81768798828125, -2.34466552734375, 4.954620361328125, 6.59075927734375, 17.97723388671875, 7.97509765625, -10.682098388671875, 1.421661376953125, 10.29547119140625, 3.93707275390625, 30.06756591796875, 0.108154296875, -3.1602325439453125, 12.20526123046875, 4.098175048828125, 6.414459228515625, 2.4868392944335938, -3.207122802734375, -0.3692626953125, 0.37564849853515625, -4.2512969970703125, 8.598800659179688, -7.0566864013671875, 5.693817138671875, 0.35025787353515625, -1.6172637939453125, -3.025390625, 3.5879974365234375, 15.95880126953125, -3.062469482421875, 4.3848876953125, 8.29998779296875, -3.16253662109375, 13.392425537109375, 5.32550048828125, 3.64654541015625, -15.1707763671875, -6.544708251953125, 6.795005798339844, 13.0880126953125, -9.5283203125, 10.7685546875, 0.13365936279296875, -1.1321945190429688, 0.95806884765625, -4.750579833984375, 4.771656036376953, 17.447998046875, 16.22283935546875, 1.8555908203125, 2.9521484375, 1.115447998046875, 14.743621826171875, 0.327850341796875, 6.62384033203125, 2.287628173828125, -2.354156494140625, 0.18243408203125, 7.980621337890625, 5.38201904296875, 2.54498291015625, 0.03924560546875, 9.241500854492188, -4.066352844238281, 9.458038330078125, 8.366058349609375, 2.1212158203125, 3.855712890625, 1.56036376953125, 17.0545654296875, 20.63531494140625, -1.9836273193359375, 1.742929458618164, -1.3276824951171875, 10.40863037109375, -0.9294052124023438, 4.5271148681640625, 7.014923095703125, 3.40643310546875, 1.9352340698242188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000083.npy"}
{"epoch": 0.17382198952879582, "step": 84, "batch_size": 128, "mean": 2.83815336227417, "std": 6.847092628479004, "min": -20.78179931640625, "p10": -4.053077697753906, "median": 3.062206268310547, "p90": 11.23279571533203, "max": 18.14508056640625, "pos_frac": 0.6953125, "sample": [-1.0202388763427734, 7.158355712890625, 14.82769775390625, -0.052490234375, -2.7716064453125, -2.8224945068359375, 7.0181884765625, 4.932594299316406, 4.927665710449219, -5.10272216796875, 0.9572830200195312, 7.7730712890625, 1.2673797607421875, -2.5733795166015625, 2.0352783203125, 4.1314849853515625, 7.137153625488281, 1.9571151733398438, -0.24976730346679688, 9.981124877929688, -3.220500946044922, 7.235504150390625, 3.043548583984375, 3.8482666015625, 9.22613525390625, 0.58953857421875, -8.063743591308594, 6.3704833984375, 6.57904052734375, 3.5454254150390625, -1.83294677734375, -19.31201171875, -20.78179931640625, 13.3778076171875, 2.62420654296875, -1.1756820678710938, 6.9146728515625, 1.776214599609375, 1.3753662109375, -0.31809234619140625, 5.558197021484375, 2.20758056640625, 12.412261962890625, -3.8757781982421875, 13.1270751953125, -1.6112213134765625, 1.3019561767578125, -3.7432403564453125, 12.882537841796875, 4.6634521484375, 17.971832275390625, -15.220458984375, 9.24700927734375, 2.34381103515625, 3.5646286010742188, -3.3499298095703125, 3.6565322875976562, -3.9417724609375, 7.6288299560546875, 8.144706726074219, 6.330047607421875, 3.5390167236328125, -0.77203369140625, 4.6610870361328125, 11.8203125, 1.61126708984375, 5.09661865234375, 0.9303512573242188, 1.031463623046875, 3.2482452392578125, -2.0687408447265625, 5.011016845703125, 1.0819091796875, -0.541015625, 3.1800003051757812, 12.152008056640625, 8.064155578613281, 9.441184997558594, -12.689361572265625, 3.928466796875, 3.0808639526367188, 1.5211334228515625, -2.71087646484375, 2.375873565673828, -4.324127197265625, -0.6507339477539062, -0.3905029296875, 8.3743896484375, 15.8021240234375, 0.00994873046875, 10.723876953125, 3.6778106689453125, -6.26513671875, 6.51959228515625, -0.1064453125, -0.269317626953125, 5.0645294189453125, 2.7830276489257812, 3.8348236083984375, 3.3010101318359375, 5.2868194580078125, -1.1203041076660156, 1.937896728515625, 2.1776809692382812, 9.282150268554688, 5.225349426269531, -5.202117919921875, 18.14508056640625, -10.94781494140625, 10.524612426757812, 16.9320068359375, 1.6804885864257812, 6.599702835083008, 3.80609130859375, 16.979400634765625, 9.44171142578125, -5.174072265625, -16.154006958007812, 5.145599365234375, -1.77435302734375, 0.3550376892089844, 10.981002807617188, 14.07855224609375, -2.956512451171875, 3.1296463012695312, 6.3840789794921875, 1.1436767578125, -4.3127899169921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000084.npy"}
{"epoch": 0.17591623036649215, "step": 85, "batch_size": 128, "mean": 1.9722685813903809, "std": 5.8331685066223145, "min": -16.28460693359375, "p10": -5.302519989013672, "median": 2.310246467590332, "p90": 8.756619262695311, "max": 16.668670654296875, "pos_frac": 0.6640625, "sample": [3.390838623046875, -3.6917266845703125, -1.9095115661621094, 16.668670654296875, 6.680889129638672, 5.606689453125, -2.0468292236328125, -2.0876617431640625, 6.025970458984375, 3.40869140625, 1.8220634460449219, -3.5611038208007812, 2.3424110412597656, 8.4228515625, 6.117340087890625, 3.714752197265625, 6.5247802734375, 4.949195861816406, 4.8458709716796875, 3.239959716796875, 0.8897705078125, -0.613616943359375, 8.629562377929688, 13.17523193359375, 2.1058692932128906, -2.7200927734375, 0.20416259765625, -5.156608581542969, -7.8841552734375, 6.140380859375, 6.1126556396484375, 5.1689300537109375, -1.3861083984375, 3.26385498046875, 8.89788818359375, 11.380050659179688, -4.7387237548828125, 4.5350341796875, 0.16397857666015625, -1.324951171875, 11.680984497070312, 6.5677032470703125, -3.22430419921875, 5.3031463623046875, 9.97625732421875, -16.28460693359375, 2.5953750610351562, 2.8619461059570312, 3.6384410858154297, 13.74493408203125, 0.4910888671875, -6.282073974609375, 5.109869003295898, 1.135223388671875, 2.2780818939208984, -4.795951843261719, 2.080718994140625, 1.3114700317382812, 6.148582458496094, 2.2360000610351562, -3.782684326171875, 4.982177734375, 3.8289947509765625, -2.95867919921875, 8.535919189453125, 6.67596435546875, 3.985107421875, 3.194091796875, -4.908988952636719, -9.374176025390625, -1.8228759765625, 6.009860992431641, -7.6259765625, -5.14752197265625, -0.44419097900390625, -7.75128173828125, 11.549957275390625, 8.696075439453125, 0.29482269287109375, 6.0109710693359375, 1.610076904296875, 10.608978271484375, 0.9444732666015625, 6.65789794921875, 15.808837890625, -12.598358154296875, -1.0157470703125, 9.453216552734375, 3.7460289001464844, -4.76068115234375, -9.241416931152344, 1.3738479614257812, 3.607177734375, 2.857114791870117, 9.821487426757812, 2.405282974243164, -0.82012939453125, 4.357421875, -6.94171142578125, 4.76031494140625, -0.7103538513183594, 5.80364990234375, 0.5776329040527344, -1.163330078125, -0.4846954345703125, 1.11798095703125, -7.19677734375, 1.12701416015625, 7.66119384765625, 4.322868347167969, -5.417999267578125, -8.0025634765625, -0.2918701171875, 5.7995147705078125, -4.6946258544921875, 1.006072998046875, 6.1025390625, 0.8427276611328125, 15.560836791992188, -6.910392761230469, -0.876983642578125, 4.238372802734375, -5.253028869628906, 2.13177490234375, 3.529266357421875, -2.369874954223633, 3.6802215576171875, 5.859376907348633], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000085.npy"}
{"epoch": 0.17801047120418848, "step": 86, "batch_size": 128, "mean": 3.5665528774261475, "std": 7.596141815185547, "min": -15.819366455078125, "p10": -4.5686706542968745, "median": 3.454936981201172, "p90": 12.381134033203123, "max": 24.09027099609375, "pos_frac": 0.6640625, "sample": [4.0254364013671875, -9.723052978515625, 4.658477783203125, -7.042083740234375, 4.817535400390625, 2.537841796875, 4.216094970703125, 3.1978225708007812, 6.808990478515625, 4.6300048828125, 23.251312255859375, -2.761505126953125, 1.6249809265136719, 2.32989501953125, 6.290042877197266, -11.616397857666016, -13.65911865234375, -4.3208465576171875, 1.149606704711914, -1.1199378967285156, 3.499359130859375, 5.944478988647461, -2.015970230102539, -1.4607086181640625, 14.30035400390625, -0.28277587890625, -14.27239990234375, -1.48956298828125, 18.478515625, 13.689651489257812, -5.69488525390625, 11.11968994140625, -1.2712249755859375, 9.1593017578125, 21.0552978515625, 11.578155517578125, 10.146270751953125, -4.462249755859375, 2.6949462890625, 13.7496337890625, 3.4187240600585938, -3.570281982421875, 8.301277160644531, -2.4748077392578125, 3.0826339721679688, 1.128326416015625, 8.590370178222656, 12.246826171875, 11.204833984375, 5.479644775390625, -15.819366455078125, 11.48822021484375, -0.138702392578125, 7.4279022216796875, 4.924530029296875, -1.64501953125, 12.69451904296875, -15.41656494140625, 2.0393218994140625, 9.529945373535156, 21.131637573242188, 5.290313720703125, 1.419921875, 12.0775146484375, 7.20318603515625, -1.533203125, 0.0, -3.200775146484375, 2.7859039306640625, 7.832427978515625, 4.6564178466796875, 12.073150634765625, -12.272247314453125, 5.199104309082031, 2.451894760131836, 0.6212615966796875, -0.4242515563964844, -1.835540771484375, 10.387939453125, 10.829269409179688, -2.9182891845703125, 6.969451904296875, 10.8912353515625, -0.1158599853515625, 4.8618316650390625, -8.488983154296875, -11.329254150390625, 9.479537963867188, -4.816986083984375, 0.7299652099609375, -0.7938575744628906, 5.8844146728515625, 1.907470703125, 1.9161491394042969, 2.1739501953125, -5.231201171875, 3.66961669921875, 6.58465576171875, 3.49114990234375, 3.7099170684814453, -0.280426025390625, -4.05584716796875, -3.7445755004882812, 11.90142822265625, 14.774169921875, 9.158233642578125, -1.093912124633789, 24.09027099609375, 2.1534957885742188, 3.60809326171875, -3.068206787109375, 6.280914306640625, 13.066436767578125, 4.262359619140625, 11.871551513671875, -1.1038436889648438, 6.1790924072265625, 14.90130615234375, -0.31128883361816406, 7.081329345703125, 5.0765380859375, 10.75992202758789, 9.779525756835938, 12.883773803710938, -1.194091796875, -0.885040283203125, 1.9964599609375, 2.9089508056640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000086.npy"}
{"epoch": 0.18010471204188483, "step": 87, "batch_size": 128, "mean": 4.09548282623291, "std": 6.694347858428955, "min": -18.510406494140625, "p10": -2.3557922363281247, "median": 3.4492950439453125, "p90": 14.35631103515625, "max": 23.803070068359375, "pos_frac": 0.7265625, "sample": [5.323051452636719, 8.649017333984375, 6.806243896484375, 8.0701904296875, 15.391983032226562, 0.0, -0.25221824645996094, -2.135028839111328, 8.507713317871094, 15.667236328125, -4.019927978515625, 7.240898132324219, -2.9490966796875, 9.661924362182617, 23.803070068359375, -0.9409065246582031, 1.0581550598144531, 5.974761962890625, 14.890045166015625, 4.253257751464844, -1.01397705078125, -7.323883056640625, 1.19940185546875, 5.6274871826171875, 3.427490234375, -2.2746734619140625, -1.4638214111328125, 6.000701904296875, 16.82891845703125, 0.8753185272216797, 11.2418212890625, -10.243682861328125, 6.98126220703125, 16.1419677734375, 5.6246795654296875, 4.159706115722656, 4.571388244628906, 5.318603515625, 4.79620361328125, 1.0989990234375, -1.133687973022461, 0.47406005859375, 8.2896728515625, -2.4580535888671875, -1.2864761352539062, -0.9733085632324219, 9.224563598632812, -4.550079345703125, 10.9505615234375, 3.8179931640625, 7.140403747558594, 0.35007476806640625, 14.669921875, 16.2220458984375, 10.93951416015625, 9.60968017578125, 4.767608642578125, 8.48065185546875, 12.392730712890625, 17.520355224609375, 3.66455078125, 0.7179107666015625, 2.4751930236816406, 10.619766235351562, 3.8891448974609375, 21.0142822265625, -0.251678466796875, 0.0, 3.1571807861328125, 5.6917724609375, 3.2649078369140625, 5.8433074951171875, 12.232452392578125, 3.9334716796875, 14.566558837890625, 3.820556640625, 3.471099853515625, -1.0659103393554688, 9.14532470703125, 0.17840576171875, -7.557861328125, -2.74908447265625, -1.3496627807617188, -0.0331268310546875, -0.5695114135742188, 8.999664306640625, -2.85528564453125, 6.348136901855469, 4.1633148193359375, 8.38555908203125, 4.149261474609375, 1.2653045654296875, 2.8014755249023438, -1.0687255859375, 1.31329345703125, -10.94195556640625, 0.5259628295898438, 1.2248802185058594, -8.273223876953125, 14.266204833984375, -0.100799560546875, 3.2843017578125, -0.06416702270507812, 0.72760009765625, 13.789276123046875, -18.510406494140625, 3.9933624267578125, -5.9444427490234375, 0.16719436645507812, 2.1382980346679688, 14.06390380859375, 16.68145751953125, 0.3345947265625, 7.1160430908203125, -1.69921875, -0.1740264892578125, 8.6800537109375, 2.6418190002441406, 2.752655029296875, 0.32682037353515625, 4.3729248046875, 0.461090087890625, -2.3119659423828125, 2.3489856719970703, 5.8419036865234375, 1.7528228759765625, 15.1070556640625, 1.0092601776123047], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000087.npy"}
{"epoch": 0.18219895287958116, "step": 88, "batch_size": 128, "mean": 3.1990714073181152, "std": 7.199305534362793, "min": -24.52923583984375, "p10": -4.11649169921875, "median": 2.0611772537231445, "p90": 11.88588256835937, "max": 25.981231689453125, "pos_frac": 0.6875, "sample": [8.617111206054688, 14.334625244140625, 11.027099609375, 0.44670867919921875, 16.65667724609375, 2.6687469482421875, -1.9795684814453125, 0.714324951171875, 7.725189208984375, 2.2491378784179688, 2.128885269165039, -4.22662353515625, 2.5418243408203125, 21.924835205078125, -0.8513870239257812, -0.401214599609375, -0.3891448974609375, 7.85516357421875, 1.5991744995117188, -2.95086669921875, 12.76861572265625, -4.554412841796875, 9.786773681640625, -4.26446533203125, -2.52978515625, 6.627716064453125, -6.0667572021484375, 0.46270751953125, 15.07427978515625, -7.4419708251953125, -6.61602783203125, 9.639297485351562, 4.382537841796875, 6.3302001953125, 1.99346923828125, -1.025604248046875, 19.09735107421875, 0.0, 0.3832263946533203, 1.43231201171875, 6.27789306640625, 3.9514617919921875, 10.455642700195312, -0.67388916015625, 13.992660522460938, 13.41009521484375, 2.418365478515625, -7.4059600830078125, 3.3756675720214844, 3.326263427734375, 0.0, 3.590667724609375, 2.695953369140625, 6.59320068359375, 14.348663330078125, 1.0755157470703125, -0.6049385070800781, 0.5112724304199219, 0.055511474609375, 6.519622802734375, 11.137664794921875, -0.5354576110839844, 0.3097381591796875, -5.63226318359375, 5.5105438232421875, 2.647796630859375, -4.162078857421875, 9.84393310546875, -0.173431396484375, 9.3935546875, 3.055957794189453, -4.096954345703125, 10.77215576171875, -0.15850067138671875, -2.143890380859375, -21.783355712890625, 4.517120361328125, 1.4936790466308594, 1.4334716796875, -2.5577774047851562, 7.035552978515625, 7.2621917724609375, 9.472259521484375, -2.124399185180664, 0.355499267578125, -3.0357666015625, -2.4619293212890625, -6.5441741943359375, 6.196006774902344, 7.511474609375, 0.21567535400390625, 4.63818359375, 3.0654449462890625, 1.659423828125, 5.5475006103515625, 15.513397216796875, 0.80810546875, -2.3979339599609375, -2.261322021484375, 16.81170654296875, 2.27813720703125, 3.2648468017578125, -1.4593563079833984, 9.57806396484375, -1.762542724609375, 25.981231689453125, 11.507568359375, -0.4096498489379883, 9.41259765625, -6.72540283203125, 8.376861572265625, 0.7101974487304688, 0.9073486328125, 3.500946044921875, 21.6654052734375, 4.058267593383789, -1.1794986724853516, 0.24809646606445312, 1.60791015625, 1.8624267578125, -24.52923583984375, 6.5430908203125, -2.75823974609375, 7.32928466796875, 1.4828414916992188, 9.1033935546875, 3.564544677734375, 0.09937286376953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000088.npy"}
{"epoch": 0.18429319371727748, "step": 89, "batch_size": 128, "mean": 3.9190523624420166, "std": 6.642943859100342, "min": -17.782958984375, "p10": -2.745429229736328, "median": 3.1548233032226562, "p90": 12.924597167968749, "max": 24.3157958984375, "pos_frac": 0.8046875, "sample": [0.5488433837890625, -9.665069580078125, 15.480499267578125, 3.087371826171875, 5.592010498046875, 16.148468017578125, -1.7497215270996094, 1.4798736572265625, 1.7113189697265625, 8.64520263671875, 2.73687744140625, 3.217437744140625, 2.92974853515625, 12.833648681640625, 2.107513427734375, 10.758392333984375, 0.9266948699951172, -2.953887939453125, 5.288978576660156, 2.70428466796875, 5.301025390625, 13.136810302734375, -2.6851577758789062, 0.7797203063964844, 1.61212158203125, 1.8704986572265625, 3.85125732421875, 0.02679443359375, -0.815185546875, 3.6688690185546875, 3.0922088623046875, 1.4300079345703125, 0.5335693359375, 13.9012451171875, 14.554779052734375, 23.781707763671875, 14.2025146484375, -7.260009765625, 0.763885498046875, 1.9781494140625, -17.782958984375, -1.46575927734375, 5.73773193359375, -8.781890869140625, 1.1392822265625, 4.3885498046875, 7.80181884765625, -10.207489013671875, 0.8612518310546875, -2.8860626220703125, -0.6673316955566406, 8.493927001953125, 4.1353912353515625, 0.141082763671875, 9.2420654296875, 8.27020263671875, 6.9454803466796875, -3.2889060974121094, 7.2983551025390625, 10.44464111328125, 2.464031219482422, 14.995697021484375, 9.383819580078125, 6.8242645263671875, 8.3941650390625, 1.4225845336914062, -6.5558319091796875, 1.83837890625, 0.9562530517578125, -4.8062591552734375, 3.930419921875, 5.815093994140625, 3.9680633544921875, 7.0972747802734375, -6.273468017578125, -0.5984954833984375, 3.5931129455566406, 6.509124755859375, 21.20977783203125, -5.30255126953125, 2.130828857421875, 6.357940673828125, 3.41064453125, 8.129150390625, 1.5944671630859375, 11.060150146484375, 2.9024658203125, 14.680572509765625, 9.623992919921875, 19.44842529296875, 6.004432678222656, -2.45489501953125, 13.65460205078125, 2.143169403076172, 2.117847442626953, 3.4351348876953125, 24.3157958984375, -2.66668701171875, 2.262939453125, 3.93731689453125, 7.0592041015625, 1.5297203063964844, 2.49688720703125, -0.006072998046875, 3.8973770141601562, 5.006431579589844, -2.2962799072265625, -0.913543701171875, 0.9991455078125, 4.82135009765625, 3.67535400390625, 3.4563217163085938, 9.737518310546875, 12.56622314453125, 5.75537109375, 3.4683990478515625, 5.507110595703125, 3.0650634765625, 8.684112548828125, -14.73248291015625, 1.6794414520263672, 11.045654296875, 4.9052886962890625, 2.2629623413085938, 1.6514511108398438, 3.22344970703125, 2.3626251220703125, -1.595794677734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000089.npy"}
{"epoch": 0.18638743455497384, "step": 90, "batch_size": 128, "mean": 3.3884592056274414, "std": 7.407252788543701, "min": -20.38922119140625, "p10": -5.297738647460937, "median": 2.2160110473632812, "p90": 12.081643676757812, "max": 23.180450439453125, "pos_frac": 0.703125, "sample": [9.225784301757812, 5.13226318359375, 1.8961181640625, 18.160430908203125, -10.497200012207031, 1.00677490234375, -5.02008056640625, 1.1888885498046875, -3.76025390625, 12.01434326171875, 1.8933277130126953, 1.9747238159179688, -7.1468048095703125, -0.40763092041015625, -0.5862579345703125, 6.72406005859375, 22.203277587890625, 19.109130859375, 11.8199462890625, 0.8776779174804688, 7.3819122314453125, -0.56121826171875, 2.1089324951171875, 2.579029083251953, -13.9180908203125, -6.2093658447265625, 11.503253936767578, 6.641754150390625, 6.83551025390625, 3.3149871826171875, -1.592681884765625, 7.403495788574219, 9.193084716796875, 2.611297607421875, 7.63006591796875, 1.974884033203125, -4.946197509765625, -2.727386474609375, -0.552459716796875, 5.986320495605469, -5.522016525268555, 1.846221923828125, -0.37985992431640625, -10.172119140625, 9.645416259765625, 11.6395263671875, 4.8022308349609375, 8.722137451171875, 4.7945404052734375, 1.915283203125, 8.461196899414062, -2.7514801025390625, 0.2890625, 1.6757793426513672, 1.39801025390625, 0.3494415283203125, -12.214569091796875, -0.07232666015625, 0.3102264404296875, 6.139122009277344, 0.905059814453125, 0.0, 11.446380615234375, 6.9967803955078125, 10.274971008300781, -0.050140380859375, 8.41888427734375, 3.2404632568359375, -1.34429931640625, 11.7681884765625, 0.671539306640625, 2.077301025390625, 6.079376220703125, 17.05950927734375, 12.238677978515625, 0.338714599609375, 0.5854654312133789, 3.7534713745117188, 3.4352798461914062, 10.98553466796875, 13.5191650390625, -20.38922119140625, 9.71478271484375, -9.69073486328125, 0.4099273681640625, 2.323089599609375, 4.634136199951172, 2.7014694213867188, -0.7552528381347656, -3.5380935668945312, -0.5911712646484375, -5.904052734375, 11.7452392578125, 11.2373046875, -2.660308837890625, 3.322113037109375, -5.282623291015625, 5.614227294921875, 9.00146484375, 10.846664428710938, -2.78289794921875, 11.859687805175781, 7.3966064453125, 0.03505706787109375, -2.130401611328125, 23.180450439453125, -2.7446975708007812, 4.80999755859375, -6.600982666015625, -13.149871826171875, 4.9605865478515625, 1.786163330078125, -5.3330078125, 12.94940185546875, 2.796783447265625, 0.04632759094238281, 13.22735595703125, 2.5069580078125, 15.116256713867188, 1.006591796875, -1.451629638671875, -2.357431411743164, 19.90618896484375, 14.0911865234375, 3.426176071166992, 10.395599365234375, 1.320556640625, 12.975021362304688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000090.npy"}
{"epoch": 0.18848167539267016, "step": 91, "batch_size": 128, "mean": 3.4014909267425537, "std": 7.650609970092773, "min": -13.655181884765625, "p10": -5.756007385253906, "median": 3.093017578125, "p90": 13.28327941894531, "max": 25.447845458984375, "pos_frac": 0.640625, "sample": [6.1256103515625, 1.2068023681640625, -1.3417396545410156, 15.419082641601562, 7.725433349609375, 14.77813720703125, 0.4253883361816406, 3.7066650390625, 5.51507568359375, -2.6371383666992188, 1.116058349609375, 5.87188720703125, 6.884773254394531, 7.501731872558594, 6.402278900146484, 8.420623779296875, 1.9730987548828125, -13.655181884765625, -3.9885025024414062, -0.2129840850830078, 7.950347900390625, 4.2301483154296875, 5.8037109375, 7.68115234375, -1.58343505859375, 5.4964599609375, 0.4339580535888672, -0.5873870849609375, -0.8308124542236328, -8.344818115234375, 12.940841674804688, -0.13130950927734375, -1.73358154296875, -5.21124267578125, -3.5232391357421875, 9.834671020507812, -5.10174560546875, 6.820831298828125, 5.9447021484375, 6.7952728271484375, -8.682830810546875, -0.9657478332519531, 3.3035888671875, -2.701904296875, 2.8824462890625, 18.569732666015625, -6.3857421875, -4.670257568359375, 0.0, -12.019332885742188, -5.849510192871094, -3.9499435424804688, -10.46624755859375, 1.519287109375, 6.731903076171875, 6.96875, 0.4670867919921875, -5.715934753417969, 11.58746337890625, 4.9603729248046875, 9.259307861328125, -8.53485107421875, 4.4426727294921875, -0.13446617126464844, -2.456859588623047, -0.782073974609375, 25.369491577148438, -1.553558349609375, 0.2818603515625, 11.648117065429688, -8.330429077148438, -3.695281982421875, 25.447845458984375, 3.672636032104492, 15.274810791015625, 1.1503467559814453, 4.767547607421875, 0.4694099426269531, 3.467987060546875, 13.13482666015625, -8.452911376953125, -5.5718994140625, 20.462982177734375, 10.090049743652344, 13.815673828125, 0.0, 8.949806213378906, 4.34619140625, 6.884765625, 11.665985107421875, 4.266792297363281, 9.2149658203125, 4.5181427001953125, -7.5616455078125, -0.072021484375, 13.629669189453125, -8.489768981933594, 2.1658477783203125, 5.0092315673828125, 20.30889892578125, 7.51806640625, 8.736312866210938, 7.82611083984375, 4.559844970703125, 4.931671142578125, -10.295906066894531, -1.7685546875, 11.819953918457031, -0.544342041015625, 16.0338134765625, -3.964630126953125, 19.362930297851562, 20.588348388671875, -0.3570728302001953, -1.893280029296875, 10.466552734375, 2.41748046875, -4.6307373046875, 7.1898193359375, 1.251678466796875, 8.77337646484375, -2.8751983642578125, 1.74884033203125, 1.5510787963867188, 4.778228759765625, 1.651397705078125, 8.168136596679688, 0.5620193481445312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000091.npy"}
{"epoch": 0.1905759162303665, "step": 92, "batch_size": 128, "mean": 3.799166679382324, "std": 7.95806360244751, "min": -27.67120361328125, "p10": -5.473458862304687, "median": 3.7222185134887695, "p90": 13.802725219726561, "max": 21.591827392578125, "pos_frac": 0.6953125, "sample": [-5.430328369140625, 0.5996017456054688, -4.46502685546875, 4.1211700439453125, 3.2796707153320312, 6.386482238769531, 4.6011199951171875, 5.951416015625, 13.740798950195312, 7.961761474609375, 1.0301513671875, 2.1566734313964844, 4.333770751953125, 0.956024169921875, 21.190216064453125, 18.4683837890625, 19.92474365234375, -6.23712158203125, 7.9696502685546875, 1.5079421997070312, -7.12884521484375, 7.464691162109375, 14.285537719726562, 12.48468017578125, 3.60394287109375, -3.130859375, 2.325927734375, 13.3475341796875, 1.13421630859375, -1.416961669921875, 4.1800537109375, 1.0916595458984375, 12.007400512695312, 7.65716552734375, 1.0010986328125, -0.621002197265625, -0.282958984375, 4.2186279296875, 14.4727783203125, 7.850830078125, 1.521820068359375, 2.211456298828125, 0.21393966674804688, -27.67120361328125, 12.553543090820312, 2.8361663818359375, 4.568603515625, 2.356353759765625, -0.443695068359375, 1.167083740234375, -5.5740966796875, -8.321075439453125, -9.185165405273438, -0.322174072265625, 15.9825439453125, -4.9943695068359375, -10.032669067382812, 0.0778350830078125, -7.616302490234375, 6.69818115234375, 5.9443511962890625, 6.132835388183594, 5.83990478515625, 5.868194580078125, 6.295013427734375, 4.534263610839844, 9.76388168334961, 0.0, -1.35076904296875, 12.3077392578125, -2.987884521484375, 4.39031982421875, 2.409515380859375, -1.0889339447021484, -3.144287109375, 8.501762390136719, -1.2174186706542969, 0.6548080444335938, 10.487319946289062, -0.9068737030029297, -6.3653717041015625, 13.25103759765625, 6.7338104248046875, -22.083038330078125, -1.9991302490234375, -0.56072998046875, 6.5667724609375, 21.032501220703125, 20.572509765625, 5.566108703613281, 7.5318756103515625, 5.9705352783203125, -0.08726692199707031, -0.4207763671875, -0.1994647979736328, 21.591827392578125, -0.985260009765625, -5.6973419189453125, 19.560516357421875, 4.568794250488281, 10.562286376953125, 10.384323120117188, -11.646484375, 13.947219848632812, 8.67706298828125, 5.10711669921875, 3.08056640625, 3.037139892578125, 3.3379745483398438, 5.579498291015625, 9.3831787109375, 4.334373474121094, 9.753925323486328, -10.143157958984375, 5.370880126953125, 10.7445068359375, -0.013895034790039062, 0.4119415283203125, -1.56427001953125, 8.320098876953125, -4.78350830078125, 9.497161865234375, 3.840494155883789, 1.8314876556396484, 15.65216064453125, 17.7447509765625, 9.559555053710938, -1.3140640258789062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000092.npy"}
{"epoch": 0.19267015706806281, "step": 93, "batch_size": 128, "mean": 3.5520858764648438, "std": 7.500323295593262, "min": -18.280181884765625, "p10": -4.492283630371094, "median": 2.4265480041503906, "p90": 14.714227294921873, "max": 24.856842041015625, "pos_frac": 0.6875, "sample": [0.10592269897460938, 14.56475830078125, 1.309112548828125, 22.99017333984375, -4.5031890869140625, 7.53887939453125, -2.60687255859375, 11.530014038085938, 6.162353515625, 5.207342147827148, 4.4212646484375, -8.65118408203125, 0.451202392578125, 1.790924072265625, 4.0283203125, 11.70062255859375, 2.56939697265625, 17.34674072265625, -3.1437911987304688, 5.131420135498047, 5.707427978515625, 4.8353118896484375, 3.390066146850586, 5.0560302734375, 15.159317016601562, 1.6041412353515625, -0.4202003479003906, 2.4024658203125, 11.30926513671875, 10.83636474609375, 3.395263671875, -7.7285308837890625, 0.9164772033691406, -2.0427703857421875, 2.9365386962890625, 4.97479248046875, 4.659645080566406, 4.6954345703125, -4.48760986328125, 15.06298828125, 3.1245574951171875, 6.788841247558594, 3.4118804931640625, -4.119537353515625, -0.2247467041015625, 4.263969421386719, 0.807952880859375, 15.837615966796875, 1.355987548828125, -0.9213333129882812, 1.5542945861816406, 6.170501708984375, 16.73370361328125, -6.620880126953125, 7.2523193359375, 1.648773193359375, -1.0979232788085938, -0.7049522399902344, 10.542694091796875, 0.42276763916015625, 7.371856689453125, 4.63665771484375, 15.856719970703125, -1.4578094482421875, -1.70025634765625, -8.19677734375, 10.902130126953125, 16.40081787109375, -6.944549560546875, 0.253753662109375, 3.41290283203125, 2.761188507080078, -3.3329086303710938, -1.4482421875, 8.384994506835938, -2.7438812255859375, 13.58905029296875, -0.437591552734375, 19.82598876953125, -18.280181884765625, 19.22259521484375, 7.64178466796875, 14.22015380859375, -6.63970947265625, 6.947052001953125, 1.3570594787597656, 0.0, -7.755859375, 2.3845272064208984, 4.515510559082031, 18.110260009765625, 12.15753173828125, -3.05511474609375, 2.1322479248046875, 6.095706939697266, -15.9783935546875, 7.2646026611328125, -0.9086723327636719, -1.979400634765625, -6.551576614379883, 9.224624633789062, 6.074981689453125, 1.807464599609375, 1.90118408203125, 3.6318359375, 0.0, 1.8669471740722656, -1.8605270385742188, -4.649444580078125, -1.0514335632324219, -0.28600311279296875, 0.04791831970214844, 6.572879791259766, 2.4506301879882812, -12.317901611328125, 22.42181396484375, 24.856842041015625, -2.1170310974121094, -1.4952850341796875, -0.9048995971679688, 4.55633544921875, 2.72625732421875, 2.276611328125, 13.37567138671875, 0.45386314392089844, 2.23455810546875, 0.6543045043945312, 7.748321533203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000093.npy"}
{"epoch": 0.19476439790575917, "step": 94, "batch_size": 128, "mean": 3.9806723594665527, "std": 9.121517181396484, "min": -20.55279541015625, "p10": -7.362551116943359, "median": 3.2945117950439453, "p90": 15.125016784667968, "max": 31.456298828125, "pos_frac": 0.6953125, "sample": [-0.122894287109375, 10.384208679199219, -6.86456298828125, 16.588043212890625, 5.1351318359375, 11.47027587890625, 19.3758544921875, 3.5831661224365234, 5.703155517578125, 8.326034545898438, 0.27130126953125, -6.206298828125, -15.0845947265625, -0.238983154296875, 12.513427734375, 0.0867919921875, 12.90106201171875, 0.6734733581542969, -0.1075439453125, 1.2294158935546875, 13.8603515625, 13.992401123046875, 4.0274658203125, -3.3215789794921875, 18.70330810546875, 14.84442138671875, 1.9973907470703125, 5.079254150390625, -9.91595458984375, -0.360809326171875, 11.184844970703125, 21.6988525390625, -12.194915771484375, 25.30108642578125, 7.317905426025391, 6.9033203125, 6.6746826171875, 1.5375099182128906, -7.110115051269531, 21.593505859375, 4.947364807128906, 31.456298828125, 15.03045654296875, -1.4570541381835938, 11.30126953125, 4.551788330078125, 14.424415588378906, -6.093963623046875, 7.427192687988281, -2.3722381591796875, 1.2791519165039062, 0.36351776123046875, -0.9991931915283203, 7.785736083984375, 3.323932647705078, -6.656524658203125, 7.6489715576171875, -2.199310302734375, 8.92559814453125, -8.99920654296875, -0.43841552734375, -4.68109130859375, 1.5433273315429688, -3.6036758422851562, 2.79437255859375, 13.430511474609375, 15.345657348632812, -0.360321044921875, -6.906639099121094, 12.657867431640625, 0.43701171875, 0.8075408935546875, 8.530387878417969, 12.822906494140625, 17.074432373046875, -6.96087646484375, 4.304718017578125, -1.9627571105957031, 4.143402099609375, 5.14715576171875, 4.19964599609375, 0.7577972412109375, -5.24810791015625, -3.63690185546875, -11.276130676269531, 21.003692626953125, 12.674201965332031, -8.86920166015625, 16.795684814453125, -11.804840087890625, -20.55279541015625, 4.2541656494140625, 2.8309974670410156, 9.152931213378906, 5.419830322265625, 11.747833251953125, 8.600677490234375, 8.151123046875, 16.737579345703125, -8.403717041015625, 7.3216094970703125, -7.951568603515625, 13.38348388671875, 1.9058685302734375, 3.2650909423828125, 6.332763671875, 2.5033111572265625, 10.037528991699219, 1.810272216796875, 1.166015625, 3.34820556640625, 4.738128662109375, 1.0524444580078125, -15.854248046875, 1.4696044921875, 0.09259033203125, 6.640625, 1.0212860107421875, 0.37493133544921875, -8.8306884765625, 9.485641479492188, 0.723663330078125, 27.89361572265625, 12.416152954101562, -10.728363037109375, -0.14178466796875, -0.09906005859375, -1.628631591796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000094.npy"}
{"epoch": 0.1968586387434555, "step": 95, "batch_size": 128, "mean": 4.385439872741699, "std": 9.000085830688477, "min": -22.135101318359375, "p10": -6.739405822753906, "median": 4.227520942687988, "p90": 17.09296722412109, "max": 27.12030029296875, "pos_frac": 0.7109375, "sample": [-3.302358627319336, -0.6084747314453125, 4.11553955078125, 23.73114013671875, 8.79534912109375, 2.356048583984375, 1.3149032592773438, 7.112274169921875, 26.132781982421875, 7.48834228515625, -0.185211181640625, -1.904693603515625, 2.128387451171875, 3.6189117431640625, -9.524063110351562, 4.85943603515625, 3.325286865234375, 6.3953094482421875, 11.684585571289062, 1.8577728271484375, 12.503662109375, 6.528564453125, -10.168212890625, 7.625457763671875, -0.61480712890625, 0.9211883544921875, 25.440704345703125, -4.8394775390625, 15.32818603515625, -22.135101318359375, 22.11065673828125, -0.8360595703125, 9.46501350402832, -6.6736297607421875, 24.81268310546875, 12.59320068359375, 1.501617431640625, 7.1392059326171875, 3.5589141845703125, 9.84637451171875, 4.208854675292969, 5.404541015625, -2.62164306640625, -0.24718856811523438, -0.7306365966796875, 8.490493774414062, 5.344696044921875, 9.094985961914062, 19.3968505859375, 1.94635009765625, -0.747314453125, 9.792106628417969, 5.1577911376953125, -0.2758636474609375, 5.786376953125, 8.275787353515625, 5.245429992675781, 4.515205383300781, 5.07647705078125, 10.54718017578125, 6.62591552734375, -2.4156036376953125, 1.47454833984375, 1.40936279296875, 3.6548004150390625, 7.071014404296875, -3.910064697265625, 10.307464599609375, 5.741172790527344, 4.246187210083008, -6.057701110839844, 18.50628662109375, 9.038818359375, 18.890762329101562, 0.400604248046875, -0.7795944213867188, 0.09759712219238281, 16.487258911132812, 2.6775894165039062, 6.6815643310546875, -9.730865478515625, 9.53164291381836, 4.7312164306640625, 1.3481826782226562, 4.6354217529296875, 23.5048828125, 6.608161926269531, 8.881057739257812, -1.2149200439453125, -14.67462158203125, 5.137931823730469, 7.121124267578125, -2.77630615234375, 1.3418045043945312, -13.05438232421875, 1.036102294921875, 8.18121337890625, 3.6922607421875, -8.931732177734375, 4.6449737548828125, -3.9377593994140625, -2.659374237060547, -6.89288330078125, 2.1630935668945312, 10.941909790039062, -8.316246032714844, 5.410327911376953, -0.9386138916015625, 25.089599609375, -9.561981201171875, 3.6627883911132812, 3.88201904296875, -1.5564117431640625, 6.69525146484375, 2.3354339599609375, 15.34954833984375, 0.6596221923828125, 9.5921630859375, 6.851104736328125, -14.61187744140625, -9.607391357421875, -0.8725223541259766, 6.4270172119140625, 19.8582763671875, 4.452178955078125, 24.26043701171875, -11.78076171875, 27.12030029296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000095.npy"}
{"epoch": 0.19895287958115182, "step": 96, "batch_size": 128, "mean": 4.2129130363464355, "std": 8.396203994750977, "min": -21.13916015625, "p10": -5.137606048583984, "median": 3.713796615600586, "p90": 15.653910064697266, "max": 26.9744873046875, "pos_frac": 0.6953125, "sample": [-3.291778564453125, -10.49725341796875, -6.552490234375, 1.683624267578125, 17.9224853515625, 8.711891174316406, -13.175132751464844, -6.179443359375, 1.057342529296875, 1.454986572265625, 3.6979904174804688, -5.067634582519531, -5.300872802734375, -2.26263427734375, 9.134979248046875, 1.7266368865966797, 4.21307373046875, -5.9120941162109375, 1.4073638916015625, 5.3696136474609375, 9.061492919921875, 2.2431640625, 9.136566162109375, 6.4158477783203125, -3.3724822998046875, 4.3601531982421875, 18.644439697265625, 10.27008056640625, 2.0868759155273438, 3.9053192138671875, 26.2845458984375, 5.76324462890625, 9.28997802734375, 10.678245544433594, 7.1554107666015625, 3.6206283569335938, 3.729602813720703, 3.0049972534179688, 1.638092041015625, 1.173004150390625, 7.4620208740234375, 4.685333251953125, -3.253814697265625, 20.77496337890625, -6.1448974609375, 6.127021789550781, -8.012863159179688, -0.6597213745117188, 4.133270263671875, -0.673980712890625, 4.60198974609375, -3.103424072265625, 9.5712890625, -3.14813232421875, 4.21197509765625, -2.6968994140625, 4.5653228759765625, 13.209259033203125, 4.086700439453125, 14.9852294921875, -1.5547676086425781, 3.8176956176757812, 12.4287109375, 9.266448974609375, 6.4764404296875, 22.22821044921875, 26.9744873046875, 4.102081298828125, -0.9174728393554688, 15.754852294921875, 5.328948974609375, 15.610649108886719, 2.3885574340820312, -9.840866088867188, 23.1083984375, 2.563201904296875, 9.764480590820312, -1.927520751953125, 17.16546630859375, 0.8364715576171875, 8.183807373046875, 0.9072723388671875, 4.800895690917969, -7.7864990234375, 0.4425506591796875, -1.8588523864746094, 0.1948394775390625, 10.916717529296875, 18.918212890625, -0.0088958740234375, -0.7808704376220703, -0.92767333984375, 4.701290130615234, 15.2928466796875, 19.80706787109375, -6.304656982421875, 3.903839111328125, 7.4792938232421875, 9.36197280883789, 13.91473388671875, -15.66448974609375, 20.51263427734375, 2.07696533203125, -2.0264663696289062, -0.46352386474609375, 10.14068603515625, -1.4683837890625, 26.3572998046875, 0.2977294921875, 3.69720458984375, 10.713897705078125, 0.0, 1.6864013671875, 2.879932403564453, 0.8338756561279297, 5.198890686035156, -4.702423095703125, 8.2911376953125, 5.230743408203125, 12.143707275390625, -3.5646018981933594, 7.6163177490234375, -3.3960952758789062, -2.9242401123046875, 2.4076385498046875, 7.0003662109375, -1.1660308837890625, -21.13916015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000096.npy"}
{"epoch": 0.20104712041884817, "step": 97, "batch_size": 128, "mean": 4.713016510009766, "std": 9.844324111938477, "min": -18.79559326171875, "p10": -6.977091217041016, "median": 4.1932525634765625, "p90": 17.31021728515625, "max": 30.1783447265625, "pos_frac": 0.6875, "sample": [5.182279586791992, 6.0838623046875, 6.0328826904296875, -6.7140960693359375, 7.433387756347656, 3.0584869384765625, -1.9755706787109375, 5.511508941650391, 9.61419677734375, 9.036941528320312, -1.1929798126220703, 29.95947265625, 10.100799560546875, 15.31298828125, 0.44376373291015625, 4.203155517578125, 19.005096435546875, -1.45989990234375, -4.73651123046875, 11.930496215820312, 3.62384033203125, 2.7484283447265625, -2.44512939453125, -1.785888671875, 9.9444580078125, -6.432220458984375, 4.872650146484375, -5.3135528564453125, 8.719535827636719, 7.705223083496094, -6.954429626464844, 0.6229400634765625, -4.6242828369140625, -7.02996826171875, 6.889007568359375, 26.085174560546875, 10.376163482666016, -2.91778564453125, -5.788238525390625, -9.103118896484375, 1.81158447265625, 4.183349609375, 2.9390869140625, 24.550048828125, -8.255401611328125, 7.95556640625, 5.26812744140625, 8.945999145507812, 1.959259033203125, 8.419639587402344, 13.006057739257812, -6.02008056640625, 23.89434814453125, 0.217864990234375, 7.909271240234375, 12.158676147460938, -7.919525146484375, 8.624771118164062, 1.077972412109375, -1.711181640625, 10.579574584960938, 6.783233642578125, -1.5579681396484375, 7.695827484130859, 5.4081878662109375, 0.0, 27.476318359375, 8.789703369140625, 9.35638427734375, 30.1783447265625, 0.9164409637451172, 13.106124877929688, 9.513641357421875, 6.153350830078125, -0.5677013397216797, 28.02740478515625, 4.957275390625, -14.3292236328125, -7.749214172363281, 17.73968505859375, 6.820343017578125, 1.4723892211914062, -5.184661865234375, 4.179168701171875, 13.624725341796875, 5.571605682373047, -0.12531280517578125, 26.86383056640625, 5.50042724609375, -3.1994171142578125, 7.2739105224609375, -10.0050048828125, 6.3720703125, 1.4533233642578125, 8.390960693359375, -18.79559326171875, 0.0, 7.65045166015625, 13.779975891113281, 2.420745849609375, 1.4463882446289062, 29.29815673828125, -9.488128662109375, 1.9625930786132812, 3.5083541870117188, 2.5834884643554688, -0.97552490234375, -6.748077392578125, 27.40093994140625, 8.1336669921875, -3.35888671875, 22.943511962890625, -7.716606140136719, -11.2020263671875, 0.520233154296875, 11.789413452148438, 3.03826904296875, 6.59588623046875, -1.0077972412109375, -13.041275024414062, 9.696434020996094, -4.647789001464844, 0.39996337890625, 13.770263671875, 17.12615966796875, -13.71759033203125, 3.439208984375, 9.937042236328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000097.npy"}
{"epoch": 0.2031413612565445, "step": 98, "batch_size": 128, "mean": 4.694764614105225, "std": 10.419342041015625, "min": -28.9783935546875, "p10": -5.862306213378906, "median": 3.7032470703125, "p90": 17.747985839843746, "max": 35.186187744140625, "pos_frac": 0.671875, "sample": [12.416454315185547, 12.656967163085938, 23.84393310546875, 3.038219451904297, 30.427734375, 18.809814453125, 15.40252685546875, -4.820465087890625, 3.858245849609375, -1.2956085205078125, 14.527801513671875, -12.214828491210938, -1.297821044921875, 4.501556396484375, 7.01898193359375, 7.12689208984375, 0.7486724853515625, -11.92242431640625, -5.818023681640625, 2.2388954162597656, 12.587738037109375, -6.955841064453125, -6.380157470703125, 5.5087127685546875, 3.2406768798828125, 4.029380798339844, -0.23632049560546875, -5.2087249755859375, -7.54547119140625, -17.573150634765625, 2.4127235412597656, -28.9783935546875, -5.5003662109375, -7.52984619140625, 3.3942337036132812, -1.6180877685546875, -5.8905792236328125, 12.64727783203125, 2.5446014404296875, 3.472900390625, 2.0638275146484375, 6.80303955078125, 0.4986419677734375, 5.7919158935546875, -1.457611083984375, 17.46563720703125, 0.9405517578125, 20.02239990234375, -5.773712158203125, 13.425548553466797, 33.2127685546875, 13.72003173828125, -1.3190078735351562, 3.66693115234375, 0.236541748046875, -1.5988922119140625, 31.26226806640625, 13.317901611328125, -2.092376708984375, 23.508514404296875, -15.57330322265625, -3.5625381469726562, -2.25091552734375, -5.850189208984375, 23.637908935546875, -1.9139862060546875, 11.223297119140625, 5.4024658203125, 6.567497253417969, -2.8949737548828125, 7.5134429931640625, 16.739959716796875, -13.823028564453125, 2.714935302734375, 20.71221923828125, 13.547317504882812, 5.336944580078125, 12.479248046875, 0.08477783203125, 16.613006591796875, 5.401573181152344, 18.40679931640625, -9.668548583984375, -0.5384902954101562, 16.899612426757812, -3.9747085571289062, -0.00067138671875, 1.8898792266845703, 4.0529327392578125, -1.87451171875, 23.0716552734375, 9.6834716796875, 5.683250427246094, -1.8453598022460938, 7.29046630859375, 8.4547119140625, 1.114776611328125, 6.5495758056640625, 3.1691055297851562, 5.621612548828125, 2.07135009765625, 16.903549194335938, 35.186187744140625, 5.802337646484375, 6.515777587890625, -4.911895751953125, -4.4886932373046875, 5.14715576171875, 5.400764465332031, 4.796630859375, -0.0101470947265625, 1.706451416015625, 3.7125701904296875, 6.732006072998047, -1.30328369140625, 8.925849914550781, 2.6733856201171875, -2.9777069091796875, 6.9135284423828125, 11.122314453125, 12.940536499023438, -4.3307342529296875, 15.898681640625, 6.93890380859375, 19.825592041015625, 3.6939239501953125, -18.72271728515625, 9.312591552734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000098.npy"}
{"epoch": 0.20523560209424083, "step": 99, "batch_size": 128, "mean": 5.21705961227417, "std": 10.812846183776855, "min": -25.46258544921875, "p10": -8.250839233398438, "median": 4.863308906555176, "p90": 18.422927856445312, "max": 27.037689208984375, "pos_frac": 0.71875, "sample": [-19.49102783203125, -0.10024261474609375, -11.601226806640625, 27.037689208984375, 4.427192687988281, 13.251068115234375, -0.9188461303710938, 18.17486572265625, 0.806060791015625, -3.6220703125, 5.863067626953125, -12.426956176757812, 0.0084228515625, 6.934486389160156, -8.0045166015625, 14.992523193359375, 18.707366943359375, 8.252723693847656, 8.465087890625, -1.4162445068359375, 5.136383056640625, 3.0020370483398438, 11.96783447265625, -19.01690673828125, 25.1370849609375, 8.13616943359375, 4.941064834594727, -10.974334716796875, 6.89129638671875, 18.301025390625, 14.663818359375, -3.605194091796875, 21.58892822265625, 2.5675277709960938, 6.563201904296875, -0.390716552734375, 3.872222900390625, 8.815643310546875, 24.933624267578125, 4.9748992919921875, 10.276748657226562, 18.021728515625, 7.2952423095703125, 16.99310302734375, 6.3437347412109375, -4.4339752197265625, 12.944507598876953, 1.7578620910644531, -0.4615631103515625, 2.44537353515625, -13.831954956054688, -20.4010009765625, -19.0826416015625, 10.555305480957031, 10.82452392578125, -4.4189300537109375, 14.446651458740234, -19.557090759277344, 15.82122802734375, 9.576690673828125, 5.700408935546875, 1.6952362060546875, 20.13037109375, 1.3222007751464844, -3.9754791259765625, 16.340484619140625, -23.404281616210938, 8.293212890625, 3.0882797241210938, 18.050277709960938, 1.90093994140625, 17.277740478515625, -1.89874267578125, 0.7286300659179688, -0.7978515625, 7.815338134765625, 5.45002555847168, 19.689208984375, 5.832305908203125, 2.6309051513671875, 9.78460693359375, 1.4919929504394531, -2.697021484375, 3.8794708251953125, 16.948562622070312, 4.299901962280273, 14.168304443359375, -4.460845947265625, 7.3996734619140625, 4.785552978515625, -8.825592041015625, 2.7480010986328125, 2.171661376953125, 8.332809448242188, 20.350250244140625, 0.7005767822265625, 17.41595458984375, -11.685211181640625, 1.6015853881835938, -4.0191650390625, 20.43499755859375, -0.6940460205078125, 0.0, 4.750144958496094, -0.125, 17.0556640625, 2.4275131225585938, -3.3962554931640625, 17.058990478515625, 23.2843017578125, 22.34637451171875, 13.391510009765625, -25.46258544921875, 16.50543212890625, 2.480022430419922, 7.2072296142578125, 11.19805908203125, 22.42315673828125, -2.11865234375, 2.3086090087890625, 3.1724319458007812, -2.81488037109375, 16.372528076171875, -1.0094184875488281, 14.56011962890625, 7.1705780029296875, 23.03778076171875, 3.998138427734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000099.npy"}
{"epoch": 0.20732984293193718, "step": 100, "batch_size": 128, "mean": 4.516352653503418, "std": 9.706889152526855, "min": -16.6219482421875, "p10": -6.965431213378906, "median": 3.311370849609375, "p90": 19.222705078124996, "max": 29.57806396484375, "pos_frac": 0.6796875, "sample": [8.80975341796875, 4.11279296875, 11.958450317382812, 3.0059661865234375, 18.8192138671875, 3.3292236328125, -10.533172607421875, 18.30340576171875, 0.222869873046875, 10.479385375976562, -2.33123779296875, 1.2982177734375, 1.3294830322265625, 4.073944091796875, 14.75225830078125, 26.6668701171875, 0.6967906951904297, 0.2253875732421875, -8.911369323730469, 8.663131713867188, 0.5667724609375, 4.97821044921875, -9.995208740234375, -1.766672134399414, 1.090911865234375, 10.084182739257812, 3.5837249755859375, -8.45135498046875, 0.18343734741210938, 17.710784912109375, -11.79644775390625, -0.23114013671875, -0.13336181640625, 3.119171142578125, 4.305908203125, 4.196830749511719, -1.419830322265625, 20.1641845703125, 3.623260498046875, 14.531265258789062, -2.36346435546875, 12.168121337890625, 6.74151611328125, 2.63470458984375, -1.7318344116210938, 4.1907958984375, 7.737770080566406, 21.658447265625, 21.83538818359375, 13.278350830078125, -1.0811996459960938, -6.9525909423828125, -5.5481414794921875, 5.680349349975586, 3.113710403442383, 11.016845703125, -14.82305908203125, -0.9608154296875, -16.6219482421875, 2.3367538452148438, -0.27579498291015625, 25.34771728515625, -5.092864990234375, 6.0727386474609375, -0.10675048828125, 5.3154144287109375, -4.309814453125, -14.196624755859375, 5.18475341796875, 8.71551513671875, 29.57806396484375, 9.86187744140625, 4.3631439208984375, 5.8611602783203125, -1.7172470092773438, -6.995391845703125, -5.201019287109375, 7.2781829833984375, 0.20211029052734375, 11.641143798828125, -12.486175537109375, 4.600700378417969, 4.076080322265625, 0.691925048828125, 6.582550048828125, 3.29351806640625, -4.601661682128906, -0.9797019958496094, 17.284332275390625, 26.29473876953125, -0.40533447265625, 4.0316009521484375, -4.534576416015625, 2.53851318359375, 6.643310546875, 1.250091552734375, 4.235561370849609, 2.3356781005859375, -0.13603782653808594, 12.432334899902344, 23.107574462890625, 0.22369384765625, 23.8463134765625, 14.480926513671875, 5.68121337890625, -4.746246337890625, 16.474319458007812, 24.967056274414062, 0.8014240264892578, 0.911712646484375, 24.444244384765625, 5.39837646484375, -0.22631072998046875, -4.147239685058594, -10.75299072265625, 15.959671020507812, -0.10925865173339844, 3.58416748046875, 20.309661865234375, -9.831146240234375, 0.7569122314453125, 10.574127197265625, 16.22625732421875, -3.782604217529297, 9.206192016601562, -1.410308837890625, 22.110107421875, -12.264175415039062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000100.npy"}
{"epoch": 0.2094240837696335, "step": 101, "batch_size": 128, "mean": 4.486861228942871, "std": 10.338446617126465, "min": -23.3446044921875, "p10": -6.926290893554687, "median": 4.86199951171875, "p90": 15.892506408691405, "max": 35.24822998046875, "pos_frac": 0.71875, "sample": [6.116539001464844, -23.3446044921875, 6.8923492431640625, 15.765289306640625, 4.7025604248046875, 2.463226318359375, 3.5785865783691406, 4.9354248046875, -4.82818603515625, 0.742828369140625, 17.725067138671875, 9.912368774414062, -4.289306640625, 21.208526611328125, 31.032470703125, 26.928604125976562, 2.660327911376953, 5.185401916503906, -1.3306350708007812, 11.908370971679688, 6.5877685546875, 5.107757568359375, 12.002410888671875, 9.574249267578125, 1.4115428924560547, 19.51910400390625, 6.6148223876953125, 3.0424041748046875, -0.4199066162109375, 27.89288330078125, -3.1906051635742188, 6.8914794921875, 4.5318603515625, 6.42303466796875, 13.0284423828125, 8.739425659179688, 20.691802978515625, 16.189346313476562, 12.397354125976562, 2.29962158203125, -7.060211181640625, 3.82293701171875, 13.21331787109375, 9.70977783203125, 19.227943420410156, 15.750946044921875, 7.97314453125, -4.7756500244140625, -0.15985870361328125, -11.621719360351562, -1.78265380859375, 6.6388397216796875, -17.15380859375, 11.498603820800781, 20.769287109375, 8.178314208984375, 7.7362823486328125, -18.616348266601562, -1.7057075500488281, 2.80218505859375, -5.216705322265625, 8.66473388671875, 2.04718017578125, 12.487342834472656, -0.941009521484375, -4.206243515014648, 6.1546783447265625, 10.861434936523438, -1.408599853515625, 14.514678955078125, -0.015625, 3.750263214111328, 0.387939453125, 5.103370666503906, 0.7955970764160156, 2.1885986328125, 20.920074462890625, 1.1749897003173828, 3.8740310668945312, -21.31976318359375, -21.998321533203125, 9.140838623046875, -9.742172241210938, 13.029754638671875, 3.1761932373046875, 2.1079559326171875, 11.57843017578125, 14.030101776123047, 0.9865684509277344, -2.6270294189453125, 10.23043441772461, 5.262485504150391, 6.43817138671875, 4.78857421875, 6.6202392578125, 4.2902374267578125, -1.52813720703125, 29.943023681640625, 13.93304443359375, 5.0370941162109375, -17.11712646484375, -20.372833251953125, 9.64581298828125, 1.5421142578125, 5.715789794921875, -2.93072509765625, 5.928314208984375, 5.407806396484375, 35.24822998046875, -6.780128479003906, 9.0693359375, -0.48122406005859375, 11.540863037109375, -2.8704833984375, 0.13897705078125, 7.147956848144531, -2.070953369140625, -7.53802490234375, -6.868896484375, 6.297828674316406, -9.345718383789062, -1.51617431640625, 2.529327392578125, 0.26423072814941406, -13.857624053955078, 1.2684459686279297, 7.0319671630859375, 11.031131744384766], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000101.npy"}
{"epoch": 0.21151832460732983, "step": 102, "batch_size": 128, "mean": 6.393280506134033, "std": 9.392809867858887, "min": -17.603271484375, "p10": -5.63642578125, "median": 6.050872802734375, "p90": 17.941480255126955, "max": 32.3719482421875, "pos_frac": 0.7421875, "sample": [-5.0484771728515625, -1.2768707275390625, 11.968772888183594, 5.39453125, 1.88897705078125, 18.012290954589844, 9.330825805664062, 0.032459259033203125, 2.631561279296875, 12.58160400390625, 5.988128662109375, 6.151966094970703, 11.440399169921875, 8.68621826171875, 0.329345703125, 15.371452331542969, 8.833587646484375, -10.439605712890625, -2.4062652587890625, -1.235443115234375, -3.027435302734375, 11.8170166015625, 12.729660034179688, 22.82464599609375, 17.383697509765625, -2.624786376953125, 6.301507949829102, -8.547454833984375, 4.270843505859375, -7.815765380859375, -6.1053924560546875, 7.3952789306640625, 3.7685775756835938, 8.486862182617188, -7.01336669921875, 20.835845947265625, 14.963104248046875, -1.5914230346679688, 1.3495101928710938, 0.9693660736083984, -0.16326904296875, 1.3666305541992188, -2.112689971923828, 16.89141845703125, 7.91351318359375, 5.0059967041015625, 2.529754638671875, -8.37579345703125, -9.32257080078125, -5.8086700439453125, 18.3843994140625, -4.446502685546875, 5.920570373535156, 6.53924560546875, 5.43914794921875, 16.034881591796875, 7.3358917236328125, 32.3719482421875, 9.459014892578125, 16.395172119140625, 1.9466781616210938, -2.8739013671875, 17.152313232421875, 10.351364135742188, 9.508453369140625, -5.5626068115234375, -9.483306884765625, 1.7812347412109375, -3.534496307373047, 6.363334655761719, 10.970359802246094, 17.0489501953125, 21.849868774414062, 15.782394409179688, 5.741601943969727, 18.905410766601562, 18.69265365600586, 6.113616943359375, 7.56707763671875, -14.213623046875, 1.1440849304199219, 8.661468505859375, 0.0, 5.817779541015625, -0.5649566650390625, 18.369476318359375, 0.0, 5.5312652587890625, 13.27838134765625, -0.3644676208496094, 12.765289306640625, 11.944427490234375, -0.7567291259765625, 3.881195068359375, 16.15850830078125, 6.296844482421875, 13.321136474609375, 1.2529296875, 10.16278076171875, 15.393508911132812, 17.42041015625, 5.531211853027344, 2.8834228515625, 23.42919921875, 8.13165283203125, 10.53900146484375, 2.662290573120117, 18.82373046875, 0.9417495727539062, 17.9111328125, -17.603271484375, 10.585845947265625, -2.313446044921875, 3.0956153869628906, 11.71533203125, -12.986083984375, 15.591407775878906, 3.208221435546875, 31.7662353515625, 3.3575286865234375, 17.869873046875, 5.4244384765625, 11.560134887695312, 9.2911376953125, 9.90826416015625, 29.049346923828125, -0.821533203125, -10.997085571289062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000102.npy"}
{"epoch": 0.2136125654450262, "step": 103, "batch_size": 128, "mean": 4.882188320159912, "std": 10.648590087890625, "min": -22.62493896484375, "p10": -6.72821350097656, "median": 2.6732635498046875, "p90": 19.14317626953125, "max": 47.15423583984375, "pos_frac": 0.7109375, "sample": [-20.24847412109375, -8.572998046875, 7.4644775390625, 9.64666748046875, 19.6065673828125, 15.840259552001953, -1.0185165405273438, -3.1025962829589844, 25.38922119140625, 0.98089599609375, -1.364654541015625, 15.676513671875, 19.697341918945312, 2.820098876953125, 0.0, -2.625640869140625, -15.09075927734375, -11.563095092773438, -3.930877685546875, 8.83990478515625, -0.6564216613769531, 1.6876678466796875, 10.145736694335938, -22.62493896484375, 2.3489913940429688, 5.99371337890625, 1.9013137817382812, 2.594696044921875, 5.5050048828125, 1.5291061401367188, 10.785003662109375, 9.311058044433594, 0.3605804443359375, 6.474456787109375, -4.76824951171875, 7.805959701538086, -1.7595367431640625, 16.38542938232422, 8.030990600585938, 8.016239166259766, -1.6760196685791016, 14.702117919921875, 0.365142822265625, 29.951416015625, -2.0519561767578125, 13.625640869140625, 2.904998779296875, 9.76165771484375, 12.090576171875, 47.15423583984375, 17.579925537109375, 13.671875, 8.82110595703125, -9.17266845703125, 1.629974365234375, -0.26885986328125, -10.482353210449219, 9.01580810546875, 9.6741943359375, 11.025955200195312, 19.692535400390625, 8.414764404296875, 2.1075897216796875, 11.616912841796875, 4.776283264160156, 21.47320556640625, 10.83978271484375, 11.418228149414062, 4.103584289550781, 7.00274658203125, 1.0840835571289062, -0.4081268310546875, 0.997711181640625, 2.3255615234375, 1.1743621826171875, 0.9202079772949219, -9.08336067199707, -17.4376220703125, 2.7518310546875, 3.3635406494140625, 22.277679443359375, 13.731216430664062, 0.14825439453125, 13.654144287109375, -1.3091011047363281, 9.866485595703125, 5.2318572998046875, 1.4476089477539062, 27.489501953125, -1.5836181640625, 1.706024169921875, 1.378173828125, 14.591201782226562, 7.62811279296875, -2.5989990234375, 1.9678916931152344, 24.25079345703125, -2.3399658203125, -5.937591552734375, -12.906341552734375, 0.0, 3.3148880004882812, 3.1330032348632812, 16.637115478515625, 0.339111328125, 18.944580078125, -9.8582763671875, 1.0477294921875, 1.3416290283203125, 12.249279022216797, 12.817245483398438, -3.8898391723632812, -2.5884437561035156, 18.74224853515625, 0.568572998046875, 0.2063274383544922, 21.675567626953125, 21.821533203125, 10.6748046875, 20.4041748046875, 0.2816162109375, 6.588630676269531, -2.45355224609375, -15.01873779296875, 1.79150390625, -19.451904296875, -5.373786926269531, -0.68597412109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000103.npy"}
{"epoch": 0.2157068062827225, "step": 104, "batch_size": 128, "mean": 4.424940586090088, "std": 11.586977005004883, "min": -27.716278076171875, "p10": -8.819403839111327, "median": 3.6885757446289062, "p90": 19.01151123046875, "max": 48.62701416015625, "pos_frac": 0.6640625, "sample": [3.61236572265625, -5.7769775390625, 8.475074768066406, 18.7445068359375, -11.52947998046875, 6.485382080078125, 4.955604553222656, -0.17279624938964844, 19.634521484375, -6.176483154296875, 17.39361572265625, -7.693721771240234, 3.978485107421875, -17.424636840820312, 7.278228759765625, 8.07529067993164, 27.7686767578125, 19.833160400390625, -1.26849365234375, -2.4993896484375, 2.9423980712890625, -2.657716751098633, 7.3516998291015625, -20.52374267578125, 25.576751708984375, -4.2735595703125, 3.5736846923828125, -1.815347671508789, 2.747560501098633, 24.1959228515625, 6.7118072509765625, 15.73016357421875, -15.392791748046875, 0.34192657470703125, 2.790740966796875, -9.044044494628906, 0.7429637908935547, -8.723129272460938, 22.0841064453125, -2.288604736328125, 13.021148681640625, 2.2215576171875, 0.6997528076171875, -10.418731689453125, -10.268157958984375, 11.511611938476562, 1.0966644287109375, 7.9797821044921875, -2.6591796875, -1.763031005859375, -4.75799560546875, 9.9698486328125, 9.302490234375, 4.0769500732421875, 26.087615966796875, 6.204700469970703, 18.31256103515625, -5.8729400634765625, -3.0207366943359375, 2.818979263305664, 4.986488342285156, 10.038703918457031, 6.10601806640625, 6.71150016784668, 8.43133544921875, 3.7115325927734375, -1.5766677856445312, 16.98150634765625, 4.352081298828125, -14.264846801757812, 10.261032104492188, 17.945953369140625, 0.5019454956054688, -24.9141845703125, 48.62701416015625, 5.317138671875, 22.79217529296875, 9.581985473632812, 11.705718994140625, 22.19647216796875, 5.5795745849609375, -27.716278076171875, -11.9212646484375, 11.883316040039062, 2.615203857421875, -0.193389892578125, 0.7212944030761719, 6.8492431640625, -7.4720458984375, 18.168060302734375, 23.820755004882812, -0.02337646484375, 4.6225128173828125, 9.1875, 1.1425552368164062, 0.42767906188964844, -8.326263427734375, 0.6339054107666016, 23.11151123046875, 13.077117919921875, 15.281791687011719, 28.3084716796875, -5.266838073730469, 15.9344482421875, -18.350311279296875, 1.0129928588867188, -5.41455078125, 3.665618896484375, 5.834709167480469, -16.195068359375, 0.0, 10.996322631835938, -1.186676025390625, 13.413925170898438, 4.067169189453125, 5.6381378173828125, 13.396697998046875, -6.582756042480469, 8.51861572265625, -0.1307830810546875, -1.30987548828125, 15.542022705078125, 3.0968189239501953, 5.8083343505859375, 18.3865966796875, -3.20263671875, -0.472869873046875, 1.6149444580078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000104.npy"}
{"epoch": 0.21780104712041884, "step": 105, "batch_size": 128, "mean": 8.1506929397583, "std": 11.388684272766113, "min": -24.40240478515625, "p10": -4.790646362304686, "median": 7.4032440185546875, "p90": 22.761151123046876, "max": 34.31396484375, "pos_frac": 0.78125, "sample": [5.662353515625, 4.177978515625, 14.854660034179688, 1.4219017028808594, -7.29632568359375, 18.2667236328125, 8.943962097167969, 15.355056762695312, 10.879974365234375, 2.2174911499023438, 15.6385498046875, 6.414306640625, 22.14129638671875, 19.485122680664062, -11.11309814453125, 22.8243408203125, 6.578468322753906, -1.1555938720703125, 26.942062377929688, 9.444305419921875, -1.1198158264160156, -2.230926513671875, 7.9115753173828125, 6.9014739990234375, 23.502273559570312, 29.3902587890625, -21.898590087890625, 0.23004913330078125, -5.304046630859375, 30.09539794921875, 3.803314208984375, 12.681259155273438, 0.263641357421875, 18.165298461914062, -2.4538421630859375, 21.5548095703125, 23.496536254882812, 3.431640625, 19.4969482421875, -11.4090576171875, 3.37945556640625, 22.0723876953125, 14.5223388671875, 17.902099609375, 2.2177581787109375, 19.043853759765625, 10.08648681640625, -1.179718017578125, 22.675994873046875, 19.376708984375, 3.1138153076171875, 13.306915283203125, -0.54180908203125, 14.021697998046875, -1.9836273193359375, 12.874847412109375, 10.188629150390625, 26.483154296875, 7.7247161865234375, -11.5740966796875, 3.79071044921875, -11.859893798828125, -1.32562255859375, 3.68536376953125, 7.418487548828125, 8.9586181640625, 16.453094482421875, -1.14801025390625, 3.002256393432617, 2.048797607421875, 20.686431884765625, -19.29620361328125, 16.889076232910156, 21.732086181640625, 7.97802734375, -4.57061767578125, 11.9290771484375, 17.763427734375, 27.27679443359375, 1.848419189453125, 6.766357421875, 3.2613067626953125, 7.1194305419921875, 0.5750865936279297, 7.38800048828125, 20.069580078125, -13.54669189453125, 22.73406982421875, 7.869697570800781, -24.40240478515625, 3.2817726135253906, 10.5965576171875, 24.358123779296875, -11.8890380859375, 7.496284484863281, 3.225830078125, 9.42982292175293, -5.8718109130859375, 13.1539306640625, 0.0, 4.438905715942383, 7.326416015625, 0.5897598266601562, -0.312347412109375, 14.543594360351562, -2.9054718017578125, 32.93603515625, 27.471588134765625, 6.40087890625, 5.028968811035156, 7.6358795166015625, 22.8477783203125, 0.4395790100097656, 3.9457168579101562, 18.579071044921875, 11.942031860351562, -4.4784088134765625, 11.61309814453125, 2.7510528564453125, -8.581146240234375, 8.281578063964844, 22.042312622070312, 4.721473693847656, 4.1140289306640625, -0.11456298828125, 34.31396484375, 20.841033935546875, 10.0963134765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000105.npy"}
{"epoch": 0.2198952879581152, "step": 106, "batch_size": 128, "mean": 5.443211555480957, "std": 11.796533584594727, "min": -21.2655029296875, "p10": -9.466888427734375, "median": 4.772998809814453, "p90": 21.485968017578124, "max": 36.147430419921875, "pos_frac": 0.6875, "sample": [-6.248237609863281, -15.332794189453125, 14.99151611328125, -2.578338623046875, -3.909454345703125, 21.651611328125, -7.467315673828125, 9.323135375976562, -12.560760498046875, 15.601264953613281, -13.967315673828125, -0.07547378540039062, 5.5958251953125, 19.460357666015625, 12.660400390625, 31.607391357421875, 27.178802490234375, -9.50213623046875, -5.0007476806640625, -3.3996200561523438, 14.491081237792969, 3.3524322509765625, 5.379180908203125, 5.060096740722656, 11.7725830078125, 0.692596435546875, 3.4437427520751953, 5.6616668701171875, 11.552467346191406, 24.774826049804688, 0.050811767578125, 9.594505310058594, 15.110076904296875, 0.808929443359375, 0.4312267303466797, -5.802909851074219, 9.896453857421875, 3.5428924560546875, 5.508819580078125, 8.13690185546875, 21.829986572265625, 2.243927001953125, 8.642494201660156, -18.946044921875, -1.289581298828125, 15.38067626953125, -7.206573486328125, 0.789581298828125, 12.39825439453125, 6.4174346923828125, 6.204833984375, 8.812324523925781, -5.047616958618164, 17.537841796875, -20.249481201171875, -21.2655029296875, 9.600128173828125, 30.802490234375, 28.593109130859375, 21.41497802734375, 4.098876953125, 8.64453125, -10.561859130859375, -7.0691070556640625, 13.22625732421875, -0.8670501708984375, -6.11053466796875, -4.535400390625, 23.04620361328125, -12.709442138671875, 12.07318115234375, 6.5227203369140625, -0.436065673828125, 7.30548095703125, -13.64007568359375, 1.409646987915039, -3.3676071166992188, 2.071685791015625, 15.83050537109375, 3.8453826904296875, 0.53564453125, 36.147430419921875, 9.633560180664062, 1.884796142578125, -6.050048828125, 2.330078125, 21.19403076171875, -0.329803466796875, 14.853424072265625, -9.4517822265625, 9.784355163574219, 4.306129455566406, 15.986572265625, -0.853668212890625, 1.4616622924804688, 4.414741516113281, 13.405319213867188, 19.375885009765625, 5.968231201171875, 8.786491394042969, 30.34393310546875, -16.59490966796875, 0.9984054565429688, -2.8564987182617188, 6.9004974365234375, 13.296077728271484, 13.139450073242188, -12.894241333007812, 3.398712158203125, -4.751739501953125, -11.09625244140625, -3.4603271484375, 33.07025146484375, 7.486919403076172, -7.16546630859375, 5.862060546875, 11.532440185546875, -5.657234191894531, 19.673309326171875, 1.98553466796875, 22.5272216796875, 4.48590087890625, 9.517074584960938, 25.264617919921875, 19.851104736328125, -0.66522216796875, 10.230270385742188, 1.999053955078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000106.npy"}
{"epoch": 0.22198952879581152, "step": 107, "batch_size": 128, "mean": 7.323891639709473, "std": 12.2823486328125, "min": -54.0699462890625, "p10": -5.277703857421875, "median": 5.696727752685547, "p90": 24.128019714355467, "max": 34.025390625, "pos_frac": 0.7265625, "sample": [0.69622802734375, -3.2635345458984375, 4.1676177978515625, 5.464996337890625, 11.24969482421875, -4.640380859375, 13.341461181640625, 25.1488037109375, 5.662261962890625, 3.13140869140625, -5.30560302734375, 5.3119354248046875, 1.61285400390625, 13.7332763671875, 12.928268432617188, -1.24481201171875, 13.699493408203125, -6.0614013671875, 11.819091796875, 18.8839111328125, -3.222057342529297, 22.001495361328125, 7.266944885253906, 13.230865478515625, -12.199951171875, 28.78204345703125, 8.074981689453125, -13.565048217773438, 17.60009765625, -6.94610595703125, 16.750152587890625, 1.939422607421875, 1.949493408203125, 17.607620239257812, 21.35528564453125, 23.79388427734375, 11.16815185546875, 5.731193542480469, 33.87255859375, -0.120819091796875, 27.696487426757812, 6.844970703125, 6.774749755859375, 11.804153442382812, 3.1270370483398438, 1.1243305206298828, -3.151123046875, 5.43682861328125, 13.790496826171875, 5.749664306640625, 7.751735687255859, -1.1537284851074219, 3.4055099487304688, 3.33935546875, 21.5545654296875, -5.8766632080078125, 3.9581298828125, 1.6203842163085938, 20.218231201171875, 9.905136108398438, 6.7733154296875, 2.51971435546875, -5.2657470703125, 34.025390625, -0.3312225341796875, -6.119720458984375, 4.0567169189453125, 32.680633544921875, 22.92401885986328, 17.855743408203125, -12.19769287109375, 24.907669067382812, -4.1601409912109375, 15.858627319335938, 21.07989501953125, 1.7147216796875, 3.4217529296875, 19.71954345703125, 12.980224609375, 12.524070739746094, 28.207489013671875, 15.23724365234375, -3.06109619140625, -12.011016845703125, 0.278717041015625, 5.2830352783203125, 6.9883575439453125, -0.8082122802734375, 33.841033935546875, -8.701797485351562, 27.17596435546875, 12.528274536132812, -1.4680328369140625, 6.893280029296875, 6.0736846923828125, 1.258108139038086, 16.80035400390625, 1.01995849609375, 0.34893798828125, -4.1961669921875e-05, -54.0699462890625, -9.80303955078125, 10.78900146484375, 26.894744873046875, 17.176292419433594, -3.3125, 26.30194091796875, -0.383758544921875, 7.9187469482421875, 1.4110469818115234, 20.157424926757812, -1.2531661987304688, 9.139678955078125, 3.217559814453125, 7.50286865234375, 18.1357421875, -2.9251861572265625, -1.15447998046875, -5.430938720703125, -2.318359375, 1.3048095703125, 7.7771759033203125, -4.99261474609375, 6.7260284423828125, -0.13533592224121094, 29.440292358398438, 5.150077819824219, 18.0162353515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000107.npy"}
{"epoch": 0.22408376963350785, "step": 108, "batch_size": 128, "mean": 8.182095527648926, "std": 11.017467498779297, "min": -18.674240112304688, "p10": -4.852021789550781, "median": 7.256813049316406, "p90": 25.0998291015625, "max": 34.826934814453125, "pos_frac": 0.7734375, "sample": [25.14605712890625, 16.343460083007812, -1.957275390625, 7.09613037109375, 13.900588989257812, -7.0846405029296875, 19.613082885742188, 9.094642639160156, 3.3199024200439453, 13.046417236328125, 1.599853515625, -4.06591796875, -4.8468170166015625, 12.682220458984375, 4.07623291015625, -7.102287292480469, 27.6270751953125, 5.23077392578125, -2.368499755859375, 9.056396484375, 12.02520751953125, 6.0150146484375, 11.043670654296875, -7.848663330078125, -0.3116912841796875, 9.244796752929688, 27.3818359375, -1.8421630859375, 0.0, -17.68841552734375, 13.603290557861328, 5.249847412109375, 10.10528564453125, 5.028114318847656, 17.057891845703125, 1.90399169921875, 27.9244384765625, 6.281982421875, 16.809814453125, 0.014703750610351562, 3.2117080688476562, 10.277191162109375, 3.42236328125, -4.864166259765625, 5.216033935546875, 26.85406494140625, 11.586151123046875, -15.12060546875, 6.156341552734375, 25.888763427734375, 14.505935668945312, 3.9373397827148438, 9.1922607421875, 34.1708984375, 6.53338623046875, 1.8277740478515625, 0.0, -5.9091796875, 1.8245220184326172, -0.11350059509277344, 2.3299713134765625, 23.54132080078125, 14.068359375, 8.99896240234375, -18.674240112304688, 7.8506622314453125, 20.77069091796875, 12.11556625366211, -0.9889907836914062, 11.4482421875, 10.20098876953125, 6.02496337890625, -3.502899169921875, 2.310821533203125, -4.433662414550781, 25.08001708984375, 20.383392333984375, 1.2259063720703125, 12.50830078125, 17.17596435546875, 12.143035888671875, -6.7957305908203125, 28.57513427734375, 12.6591796875, 3.882415771484375, 1.01019287109375, 22.81939697265625, 10.779273986816406, 9.619918823242188, 29.037841796875, 7.74835205078125, 20.4034423828125, 3.539337158203125, 11.9478759765625, 8.0269775390625, -3.9342498779296875, 32.6583251953125, 16.141860961914062, 29.778106689453125, 19.5599365234375, 15.700874328613281, 4.349853515625, 7.9012298583984375, 0.0, -1.7553253173828125, 10.015266418457031, 6.827934265136719, 34.826934814453125, 4.753376007080078, 23.6488037109375, 5.3668212890625, -8.42388916015625, 3.5413360595703125, 0.18461990356445312, 16.405784606933594, 26.5174560546875, 19.22967529296875, 2.5799102783203125, 2.6584854125976562, 10.637466430664062, -10.677230834960938, -11.531158447265625, -10.330738067626953, 7.4174957275390625, -0.24329376220703125, 9.44964599609375, 2.955322265625, 16.262969970703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000108.npy"}
{"epoch": 0.2261780104712042, "step": 109, "batch_size": 128, "mean": 8.481164932250977, "std": 13.642035484313965, "min": -27.22320556640625, "p10": -6.35946044921875, "median": 6.891632080078125, "p90": 24.39358215332031, "max": 58.3406982421875, "pos_frac": 0.7265625, "sample": [4.50048828125, 12.628707885742188, -10.233718872070312, 22.841140747070312, 4.444091796875, 7.34393310546875, 16.639419555664062, 16.63299560546875, -10.72845458984375, 7.1864013671875, 7.00665283203125, 19.3798828125, -0.024044036865234375, -3.0012054443359375, -6.65972900390625, -4.74493408203125, -1.425689697265625, 31.99420166015625, 40.119140625, 1.7181282043457031, 0.8856239318847656, 0.314971923828125, -2.55804443359375, -11.417327880859375, -7.214687347412109, 11.864471435546875, 1.24969482421875, -0.9674110412597656, -0.0407562255859375, 10.6156005859375, 7.5923004150390625, -2.196929931640625, 16.675445556640625, 35.983001708984375, 25.08380126953125, 9.096153259277344, 13.000473022460938, 21.397979736328125, 1.5993118286132812, 22.238739013671875, 13.212493896484375, 18.3447265625, 5.453895568847656, -0.5664653778076172, 14.38299560546875, 34.416229248046875, -1.259124755859375, 29.29229736328125, 7.191070556640625, 6.51922607421875, 12.914398193359375, 3.3087158203125, 6.776611328125, 6.65264892578125, -21.560272216796875, 5.780731201171875, 7.49676513671875, 7.099231719970703, 12.69140625, 8.62213134765625, 16.161895751953125, 3.3509368896484375, -0.7138671875, 23.429718017578125, 5.4066162109375, 10.844436645507812, -5.72210693359375, 11.360748291015625, 9.059822082519531, 11.752716064453125, 17.00555419921875, 3.995880126953125, -9.016586303710938, 17.812820434570312, 10.666572570800781, 24.166778564453125, 47.681549072265625, -3.088623046875, 1.223846435546875, 30.11053466796875, 5.32965087890625, 5.06219482421875, 19.00914764404297, 13.784187316894531, -0.1142120361328125, 2.0600509643554688, 13.714752197265625, -0.0620880126953125, -10.525970458984375, 41.13037109375, 23.34515380859375, 17.964996337890625, 5.707675933837891, 58.3406982421875, 9.976837158203125, -0.64849853515625, 2.1114730834960938, 37.001953125, -12.274368286132812, 4.135383605957031, -6.23077392578125, 19.184982299804688, 0.0, 1.2481460571289062, 31.60345458984375, 10.333877563476562, 13.870269775390625, 10.250701904296875, -0.78240966796875, 10.241546630859375, -0.8690185546875, 6.013481140136719, 0.8755950927734375, -2.425374984741211, 21.216461181640625, 0.409698486328125, 12.810585021972656, 20.5570068359375, -12.944854736328125, -27.22320556640625, -26.94586944580078, 24.92279052734375, 5.726165771484375, 17.704376220703125, 5.4803466796875, -6.702117919921875, -3.162567138671875, 14.297576904296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000109.npy"}
{"epoch": 0.22827225130890053, "step": 110, "batch_size": 128, "mean": 7.8271331787109375, "std": 13.447150230407715, "min": -39.0040283203125, "p10": -6.074563598632812, "median": 5.776420593261719, "p90": 26.55105514526367, "max": 42.58782958984375, "pos_frac": 0.7265625, "sample": [-0.7999343872070312, -39.0040283203125, -11.013614654541016, 27.1143798828125, -2.7088661193847656, 1.9118976593017578, -3.0020828247070312, 1.444122314453125, -1.333984375, 27.33868408203125, 3.82843017578125, 2.941619873046875, 2.268627166748047, 23.85455322265625, -5.142833709716797, 0.8567276000976562, 12.18389892578125, 4.919403076171875, 16.306610107421875, 4.631011962890625, -8.691474914550781, 18.06707763671875, 1.9518489837646484, 24.054443359375, 7.4199981689453125, 3.9428482055664062, -5.9753570556640625, 12.828620910644531, 2.8873291015625, -4.28118896484375, 27.9378662109375, 13.902740478515625, 26.47979736328125, 2.397390365600586, 34.690399169921875, -3.232666015625, 17.717422485351562, 17.458602905273438, 10.980987548828125, -3.949310302734375, 5.475372314453125, 6.63336181640625, 10.926185607910156, 27.0426025390625, -10.206268310546875, 0.5141487121582031, 16.300857543945312, -8.399627685546875, 11.334732055664062, 0.240509033203125, 22.004547119140625, -20.159347534179688, 12.53753662109375, 0.411224365234375, 29.801055908203125, 12.006195068359375, 3.6362991333007812, -30.41455078125, 15.234100341796875, 18.212646484375, 4.245368957519531, 5.646881103515625, 3.6327056884765625, -1.81268310546875, 0.646209716796875, 28.024871826171875, -6.3060455322265625, 25.80987548828125, 18.59942626953125, -16.623931884765625, -0.46889495849609375, 16.22222900390625, 4.857276916503906, 3.5651168823242188, 11.303024291992188, 6.6038055419921875, 0.44554901123046875, -4.367851257324219, -1.9048309326171875, 12.135711669921875, 0.0, 3.95098876953125, 20.80865478515625, 13.589859008789062, -10.923316955566406, 13.822998046875, 0.5714874267578125, -6.7727203369140625, 7.72967529296875, 26.717323303222656, 16.502334594726562, 36.64654541015625, 14.53533935546875, 25.133544921875, 6.323204040527344, 17.6092529296875, 24.6683349609375, -2.658355712890625, -2.404796600341797, 20.2078857421875, 28.9342041015625, 6.9866180419921875, 1.3674945831298828, 24.478790283203125, -16.985816955566406, 7.806976318359375, 4.874000549316406, -5.684326171875, 24.062591552734375, 18.92657470703125, -3.4651107788085938, 31.38580322265625, -3.4318466186523438, 9.894561767578125, 10.054153442382812, 29.3660888671875, 42.58782958984375, -3.1259765625, 20.035537719726562, 22.030914306640625, -0.29248046875, 19.698516845703125, 5.9059600830078125, 7.43743896484375, -13.24237060546875, 3.91815185546875, -0.7865676879882812, 6.509735107421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000110.npy"}
{"epoch": 0.23036649214659685, "step": 111, "batch_size": 128, "mean": 8.725970268249512, "std": 13.57869815826416, "min": -21.14715576171875, "p10": -8.360445022583006, "median": 7.121608734130859, "p90": 26.856454467773435, "max": 48.08563232421875, "pos_frac": 0.7578125, "sample": [4.569313049316406, -12.2137451171875, 13.074127197265625, -21.14715576171875, 45.693603515625, -8.038219451904297, 11.132858276367188, 26.755828857421875, 0.8941192626953125, -0.5070037841796875, 28.811431884765625, 5.6580810546875, 8.048934936523438, 5.3914794921875, 6.71087646484375, 14.893619537353516, 7.828887939453125, 34.3187255859375, -18.698150634765625, -12.81085205078125, 10.491744995117188, 5.396312713623047, 16.091796875, 3.1878662109375, 15.406524658203125, 2.56109619140625, 5.5088653564453125, 25.928634643554688, 1.1610641479492188, 12.381378173828125, 15.3380126953125, 19.821319580078125, -2.70538330078125, 35.38189697265625, -3.281982421875, 5.3066864013671875, -7.8431396484375, 16.565521240234375, 0.5450592041015625, 10.295623779296875, 1.9364776611328125, 10.959819793701172, -7.260993957519531, 22.510345458984375, 12.238922119140625, 2.6220703125, 16.5665283203125, -1.5917510986328125, 32.130035400390625, 6.887336730957031, 1.1322479248046875, 48.08563232421875, -10.278717041015625, -2.9360809326171875, -0.8245849609375, 17.611083984375, 11.778076171875, 22.146240234375, 3.623931884765625, 30.202484130859375, 6.81927490234375, 11.13494873046875, 5.501930236816406, 11.848175048828125, 2.7433319091796875, -5.594268798828125, 4.0464935302734375, 13.81512451171875, 39.80328369140625, 12.133636474609375, 6.980377197265625, -1.2158966064453125, 7.262840270996094, 0.526702880859375, 4.7275390625, -0.08399772644042969, 1.6885986328125, 27.09124755859375, 9.362289428710938, -16.583740234375, 1.8146858215332031, 10.25439453125, -9.355712890625, 3.021240234375, -2.94122314453125, -5.8936309814453125, 0.27712249755859375, 1.79022216796875, 17.168296813964844, 28.31500244140625, 17.25555419921875, 32.0498046875, 18.374244689941406, -2.1001148223876953, 15.983200073242188, 8.883697509765625, 15.79730224609375, -2.904359817504883, 43.451873779296875, 3.3382492065429688, 21.322853088378906, 9.742813110351562, -14.039581298828125, -9.1123046875, -1.133331298828125, 25.25054931640625, -10.6312255859375, 18.891387939453125, -2.4947357177734375, 20.31035614013672, 14.947113037109375, 7.624725341796875, 16.1634521484375, 37.259552001953125, 20.108612060546875, 16.67578125, 25.4095458984375, -16.6263427734375, 13.27880859375, 13.69096565246582, 7.508880615234375, -11.57562255859375, 13.499359130859375, 6.15087890625, 19.1322021484375, 1.3569488525390625, -11.185302734375, 3.365386962890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000111.npy"}
{"epoch": 0.2324607329842932, "step": 112, "batch_size": 128, "mean": 4.307877063751221, "std": 15.384686470031738, "min": -50.59503173828125, "p10": -13.972511291503904, "median": 3.444408416748047, "p90": 24.991259765625, "max": 46.444183349609375, "pos_frac": 0.6015625, "sample": [16.649932861328125, 3.8918304443359375, -4.299835205078125, 10.88922119140625, 0.7651443481445312, 8.239181518554688, -0.8787307739257812, -31.0443115234375, -3.8879737854003906, -0.4353485107421875, -1.2657623291015625, 6.2779998779296875, -9.508056640625, 19.866485595703125, -0.9508056640625, 3.5815811157226562, 13.4219970703125, 17.829452514648438, 7.099342346191406, 24.584259033203125, -9.277862548828125, 19.681976318359375, 12.232208251953125, 13.49041748046875, -50.59503173828125, -15.41131591796875, -9.577407836914062, 12.188705444335938, 9.43994140625, 8.980636596679688, -10.022247314453125, 5.106575012207031, -26.176300048828125, 5.3507843017578125, 3.9329605102539062, -17.64990234375, 2.9091644287109375, 5.653453826904297, -0.523406982421875, 4.286346435546875, 14.693603515625, -1.94085693359375, 6.11334228515625, 13.944473266601562, 11.857452392578125, -2.2003021240234375, 17.4886474609375, -1.261627197265625, 17.632736206054688, -4.896575927734375, 9.691665649414062, 4.98883056640625, 0.9787826538085938, 15.6051025390625, -17.635772705078125, -3.0881481170654297, -15.835403442382812, -8.944091796875, -35.562225341796875, -8.850433349609375, -1.9068107604980469, -0.9402847290039062, 17.747573852539062, 36.300628662109375, -4.756622314453125, 3.3072357177734375, 32.12591552734375, 16.970474243164062, 0.9837188720703125, 8.069976806640625, 8.008960723876953, -4.714727401733398, 18.482757568359375, 14.16204833984375, 7.9730987548828125, 37.392578125, -3.3006591796875, 0.87457275390625, 28.54498291015625, -18.147994995117188, 0.64935302734375, 8.618301391601562, 0.05084228515625, -0.6988811492919922, 21.03668212890625, 24.29925537109375, 11.459396362304688, -2.06756591796875, -5.616233825683594, 2.7983551025390625, 25.218597412109375, -13.355880737304688, 29.253082275390625, 4.904052734375, -5.353729248046875, 16.062408447265625, -2.6859207153320312, 0.3231792449951172, -38.091278076171875, -6.5605010986328125, -7.751678466796875, 26.11785888671875, -2.5528106689453125, -2.0177001953125, 26.740402221679688, 16.55908203125, -16.24566650390625, 24.893829345703125, 5.1342315673828125, 19.05169677734375, 34.8583984375, -11.33331298828125, 2.9875335693359375, 27.1226806640625, 6.24755859375, 0.8907623291015625, -17.03436279296875, -2.8730621337890625, 2.14422607421875, -0.44073486328125, 6.30010986328125, -5.14501953125, 28.20050048828125, 27.398193359375, 46.444183349609375, -15.85662841796875, -4.1197357177734375, 10.642257690429688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000112.npy"}
{"epoch": 0.23455497382198953, "step": 113, "batch_size": 128, "mean": 10.152936935424805, "std": 14.87888240814209, "min": -29.88922119140625, "p10": -6.881584167480468, "median": 8.411781311035156, "p90": 30.724427795410154, "max": 57.56707763671875, "pos_frac": 0.734375, "sample": [5.492103576660156, 15.988853454589844, 9.6087646484375, 0.3483734130859375, 15.042938232421875, 1.3316497802734375, -5.270355224609375, 16.0120849609375, 27.9259033203125, 7.514801025390625, 6.600616455078125, 8.271728515625, 19.9112548828125, -11.124828338623047, -1.465911865234375, 3.4028396606445312, 3.749898910522461, 2.9932327270507812, 3.88372802734375, 13.09902572631836, 57.56707763671875, -3.98541259765625, -2.1821441650390625, 39.526214599609375, 12.6898193359375, -29.88922119140625, 5.03955078125, 24.1875, 15.887176513671875, -2.2184677124023438, -3.2752685546875, -11.83111572265625, 24.195404052734375, 31.983612060546875, 39.52410888671875, -1.540802001953125, 15.08984375, -6.0800018310546875, 29.447158813476562, -10.352584838867188, 13.808578491210938, 30.650558471679688, -0.6322784423828125, 13.113265991210938, 11.251708984375, 4.515285491943359, -10.41802978515625, 1.28656005859375, 0.6244258880615234, 30.843597412109375, 4.777923583984375, 18.611583709716797, -24.099884033203125, 27.124038696289062, 32.5587158203125, 23.540069580078125, 10.798408508300781, 32.675567626953125, 6.780181884765625, -11.6412353515625, 8.551834106445312, -0.1321258544921875, -5.220607757568359, 24.95751953125, -0.8568191528320312, 7.92218017578125, 6.660442352294922, 6.743904113769531, -12.662933349609375, 17.32269287109375, 4.027923583984375, 31.84954833984375, -1.7374439239501953, 12.121368408203125, -8.342681884765625, 4.0559844970703125, 13.450592041015625, 30.673355102539062, 14.000823974609375, 14.99786376953125, 14.533332824707031, -12.333221435546875, 18.30767822265625, 27.803497314453125, 35.719635009765625, 13.969276428222656, 12.695411682128906, -3.8284454345703125, -4.2810211181640625, 5.656768798828125, 36.813446044921875, 26.80224609375, 19.7894287109375, 13.547027587890625, -6.5504608154296875, 26.257843017578125, 0.04181671142578125, 12.855072021484375, 15.0430908203125, 8.10791015625, 5.786567687988281, -2.76116943359375, 12.995849609375, 3.44293212890625, 24.21722412109375, 37.37548828125, 24.349761962890625, 39.797119140625, -11.41259765625, 1.972259521484375, 22.990814208984375, 18.154632568359375, 11.56341552734375, 42.08233642578125, -5.019966125488281, -7.654205322265625, 2.632415771484375, 3.3442916870117188, 20.247406005859375, 15.850234985351562, -0.23968505859375, 17.038864135742188, -17.691253662109375, 12.75286865234375, 4.2724609375, -2.0047760009765625, 19.1573486328125, -6.2666473388671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000113.npy"}
{"epoch": 0.23664921465968586, "step": 114, "batch_size": 128, "mean": 5.185506820678711, "std": 13.786945343017578, "min": -27.88348388671875, "p10": -12.749347114562987, "median": 4.206756591796875, "p90": 23.09830322265625, "max": 43.8388671875, "pos_frac": 0.6796875, "sample": [3.6862430572509766, 19.615692138671875, 3.7281951904296875, 26.363414764404297, 0.08453369140625, 7.633544921875, 1.9487838745117188, 4.7785186767578125, -1.2169265747070312, -4.8845367431640625, -2.9002227783203125, 23.259033203125, -0.959625244140625, 29.73919677734375, 15.27764892578125, 37.720550537109375, -11.882549285888672, 1.543060302734375, -21.85479736328125, 1.6187896728515625, 1.70330810546875, -15.768508911132812, 5.677757263183594, 15.809463500976562, 0.8943634033203125, 15.62030029296875, 9.946113586425781, 9.99658203125, 11.3388671875, 14.683380126953125, 3.0492401123046875, 15.803573608398438, 5.128013610839844, 7.0869140625, 17.6241455078125, -12.447784423828125, 9.966636657714844, -25.47869873046875, 27.05157470703125, 5.483612060546875, -5.422355651855469, -5.442779541015625, 2.7191009521484375, -27.88348388671875, 5.798583984375, 2.6881866455078125, -0.546142578125, 0.610443115234375, 6.822174072265625, -7.978790283203125, 0.09710693359375, -13.452993392944336, 11.535400390625, -9.113327026367188, -14.100006103515625, -19.366668701171875, 12.262969970703125, 43.8388671875, 22.76397705078125, -8.210281372070312, 25.43536376953125, -24.487335205078125, 20.15191650390625, -7.7191619873046875, -4.549591064453125, 19.6431884765625, 37.33113098144531, -9.902008056640625, 0.8648490905761719, 13.23529052734375, -25.858245849609375, 0.6381072998046875, -14.4595947265625, 9.56121826171875, 2.860504150390625, 33.68817138671875, 1.6066627502441406, -1.053375244140625, -1.0026092529296875, -17.662078857421875, 2.3549118041992188, 9.133682250976562, 7.600006103515625, -16.800445556640625, 1.5280036926269531, -1.3893108367919922, -16.486236572265625, -4.720787048339844, 0.0, 23.92852783203125, 11.424041748046875, -2.060546875, 13.2701416015625, 6.133056640625, -1.3387451171875, 23.0294189453125, -3.4898681640625, 17.4075927734375, 3.34442138671875, 6.11865234375, 10.240577697753906, 32.54182434082031, 11.682014465332031, 10.59112548828125, 38.1162109375, 17.04491424560547, 8.291351318359375, 4.62469482421875, -6.77557373046875, 0.0, 33.073577880859375, -1.391510009765625, -6.44024658203125, 13.39794921875, 7.93914794921875, 20.729515075683594, 6.517341613769531, 3.788818359375, 5.883934020996094, 6.694099426269531, 3.448577880859375, 6.980682373046875, 5.822998046875, -0.42620849609375, 13.60137939453125, 7.895721435546875, 3.2461700439453125, 15.229461669921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000114.npy"}
{"epoch": 0.2387434554973822, "step": 115, "batch_size": 128, "mean": 5.805802345275879, "std": 15.266173362731934, "min": -34.196441650390625, "p10": -12.18950729370117, "median": 4.928667068481445, "p90": 28.677209472656248, "max": 57.839599609375, "pos_frac": 0.6796875, "sample": [32.43780517578125, -14.08160400390625, 13.435150146484375, -10.441062927246094, -0.8543701171875, -13.318115234375, -9.057586669921875, 2.0447998046875, 2.8568191528320312, 19.666797637939453, -5.041265487670898, 9.508049011230469, 8.149154663085938, 7.785797119140625, 3.429534912109375, -9.969734191894531, 9.886283874511719, -3.730133056640625, 15.083602905273438, 2.8211669921875, -12.514274597167969, 2.4874267578125, 6.917938232421875, 6.124603271484375, 4.785861968994141, -4.35162353515625, 2.13214111328125, 14.164924621582031, 13.08880615234375, 15.394287109375, -3.3207550048828125, 13.508743286132812, 12.453643798828125, 4.4556884765625, 3.4175186157226562, 8.156524658203125, -7.9169921875, -20.536331176757812, -1.0102691650390625, 32.2613525390625, 5.915275573730469, 29.55596923828125, 39.3265380859375, 8.775405883789062, -21.22674560546875, -18.781326293945312, 23.72618865966797, -10.512283325195312, 27.843612670898438, 42.40350341796875, 6.570884704589844, 57.839599609375, -8.76470947265625, -1.954620361328125, -12.432647705078125, 14.750274658203125, 29.708831787109375, 22.619171142578125, 13.739669799804688, -24.154266357421875, -15.077178955078125, -6.432861328125, 28.30059814453125, 2.439849853515625, -4.2475738525390625, 11.9803466796875, 5.1375885009765625, 1.8311767578125, 5.672630310058594, 30.406005859375, 0.9469566345214844, 10.26495361328125, 14.811824798583984, 9.108985900878906, -2.6692123413085938, 6.74180793762207, 44.5467529296875, 2.39453125, -18.202117919921875, -8.212944030761719, 2.5697021484375, 1.658111572265625, -12.43865966796875, 34.669769287109375, 20.44829559326172, 5.07147216796875, -28.42523193359375, 9.792930603027344, 2.456165313720703, 31.428619384765625, -4.158227920532227, 3.26690673828125, -6.6417236328125, 34.85528564453125, 14.427886962890625, -3.5408935546875, 2.7761993408203125, 7.58380126953125, -8.433395385742188, 7.8645172119140625, 9.130767822265625, 4.276466369628906, 9.028457641601562, 7.89849853515625, 7.8603515625, 2.45355224609375, 41.5701904296875, 5.1988983154296875, -9.272064208984375, -12.085304260253906, 11.175323486328125, 4.628453254699707, 16.00428009033203, 6.215789794921875, 12.669845581054688, -11.043731689453125, -5.710235595703125, 13.022747039794922, -1.1710433959960938, 21.888748168945312, -34.196441650390625, -2.816741943359375, 9.305221557617188, 0.4578857421875, 12.860763549804688, 3.577178955078125, -1.8191986083984375, 11.8017578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000115.npy"}
{"epoch": 0.24083769633507854, "step": 116, "batch_size": 128, "mean": 8.193466186523438, "std": 14.74716854095459, "min": -40.65972900390625, "p10": -7.602320861816406, "median": 7.215747833251953, "p90": 25.99936828613281, "max": 58.2635498046875, "pos_frac": 0.7265625, "sample": [3.590301513671875, -4.134449005126953, 5.997528076171875, 3.6698455810546875, 6.5029449462890625, 10.376708984375, 10.073883056640625, 25.74237060546875, 42.430450439453125, 8.116912841796875, 19.313690185546875, 9.353515625, 20.70832061767578, 3.6823196411132812, 15.94073486328125, 11.372650146484375, -6.864173889160156, 12.6973876953125, 4.362644195556641, -10.08795166015625, -21.1922607421875, -6.176576614379883, 22.996353149414062, 23.425079345703125, 0.38915443420410156, -8.56646728515625, 11.3145751953125, -0.5591964721679688, 6.3489990234375, 0.782073974609375, 2.1082210540771484, 25.76580810546875, -10.15869140625, 18.000030517578125, -5.1953582763671875, 2.95513916015625, 21.653457641601562, 15.65283203125, -7.7411651611328125, -6.4374847412109375, -2.5336151123046875, 20.495880126953125, 15.946563720703125, -9.8115234375, 8.472610473632812, 4.5098419189453125, -2.8387451171875, 6.45684814453125, 21.469512939453125, 34.24560546875, -40.65972900390625, 3.38031005859375, 14.2864990234375, -16.828704833984375, -0.2597503662109375, 29.3583984375, 12.157684326171875, 24.162445068359375, -3.575836181640625, 1.0037193298339844, 9.752838134765625, 11.15380859375, 5.3082275390625, 3.560882568359375, 9.024261474609375, 0.0, 10.043991088867188, 35.2855224609375, -2.4833297729492188, -9.868377685546875, 18.316665649414062, 33.2689208984375, 0.933135986328125, -5.9596099853515625, -7.542816162109375, 1.3660354614257812, 1.9854888916015625, 1.3712921142578125, 6.5552825927734375, 23.38775634765625, 4.1060638427734375, 38.91766357421875, 17.404296875, 0.0, 12.693307876586914, 6.871498107910156, 20.671966552734375, 4.531494140625, 35.476287841796875, -2.842864990234375, -1.8224334716796875, 58.2635498046875, 13.73638916015625, -35.77813720703125, 7.60430908203125, 32.229736328125, 18.67420196533203, 4.557537078857422, 9.66384506225586, 9.066497802734375, 11.142776489257812, -1.275726318359375, 7.652801513671875, 27.071334838867188, 20.21942138671875, -21.52265167236328, 8.36370849609375, 11.839080810546875, 2.2881851196289062, -12.446205139160156, 11.109420776367188, 7.55999755859375, 44.31756591796875, -2.3035202026367188, 26.544342041015625, 16.110519409179688, 6.3973846435546875, 16.83380126953125, 21.873611450195312, 13.153106689453125, -6.823760986328125, 7.806373596191406, -0.0955657958984375, 28.201385498046875, -12.68951416015625, 5.579204559326172, -3.73028564453125, 18.45355224609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000116.npy"}
{"epoch": 0.24293193717277486, "step": 117, "batch_size": 128, "mean": 8.396854400634766, "std": 13.773359298706055, "min": -24.24664306640625, "p10": -6.946688079833985, "median": 7.275758743286133, "p90": 29.364291381835937, "max": 43.285980224609375, "pos_frac": 0.7109375, "sample": [3.2375335693359375, 7.403446197509766, 12.873939514160156, 16.25750732421875, 5.4633026123046875, 14.319198608398438, 2.3687286376953125, -1.4372062683105469, 8.859161376953125, -3.9350204467773438, 14.091079711914062, -7.1360015869140625, -1.160430908203125, 13.556533813476562, 18.915069580078125, 3.345916748046875, 9.137237548828125, 9.4111328125, 25.13299560546875, 3.7230224609375, 19.13763427734375, -9.316505432128906, 7.116249084472656, 1.9985828399658203, 13.570993423461914, 31.51531982421875, 0.5686492919921875, -19.066795349121094, 10.99261474609375, 43.285980224609375, 15.877695083618164, 29.312591552734375, 32.491607666015625, 14.15129280090332, 9.606729507446289, -9.872726440429688, 24.340484619140625, 0.3919830322265625, 1.0680809020996094, -9.653076171875, 33.209228515625, 7.187641143798828, 4.807373046875, 30.23162841796875, 0.0, 23.34014892578125, -4.188629150390625, 6.6140289306640625, 38.56671142578125, 24.015167236328125, 9.541213989257812, 34.189910888671875, -23.2696533203125, 19.628631591796875, 24.11572265625, 21.92803955078125, -6.92681884765625, 3.294187545776367, 3.7374343872070312, 7.3638763427734375, 16.60858154296875, 14.1097412109375, 13.351654052734375, 11.861194610595703, 4.524932861328125, 13.116012573242188, 12.8656005859375, -0.391937255859375, -6.993049621582031, -6.7499847412109375, 8.444282531738281, -21.871551513671875, -0.20558929443359375, 9.412826538085938, 27.464263916015625, -12.876739501953125, 10.9664306640625, 11.929290771484375, 21.32281494140625, 1.5267562866210938, -1.7350997924804688, -1.8765869140625, 8.21234130859375, 34.012603759765625, 5.779998779296875, -6.72259521484375, -5.1381072998046875, -14.5079345703125, -1.347991943359375, 3.974485397338867, -24.24664306640625, 17.22576904296875, 29.48492431640625, 29.52960205078125, 8.318832397460938, 4.853485107421875, -10.411651611328125, -1.61700439453125, -1.4604454040527344, 29.909698486328125, 4.1441650390625, -2.992706298828125, 2.4272212982177734, -5.403839111328125, -4.481597900390625, 8.121871948242188, 27.080902099609375, 6.875053405761719, 26.35009765625, -4.684658050537109, 15.502601623535156, 15.919281005859375, 19.9901123046875, -0.0692901611328125, -0.41721343994140625, -3.21466064453125, 4.966203689575195, 0.2397918701171875, 5.4544219970703125, -3.2559242248535156, -19.005935668945312, 37.80609130859375, 11.89892578125, 13.77978515625, 1.886016845703125, 17.207958221435547, 38.376312255859375, 10.31268310546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000117.npy"}
{"epoch": 0.2450261780104712, "step": 118, "batch_size": 128, "mean": 8.977682113647461, "std": 15.309951782226562, "min": -25.482421875, "p10": -10.039096069335937, "median": 7.5762939453125, "p90": 26.612521362304687, "max": 51.279388427734375, "pos_frac": 0.7265625, "sample": [6.428947448730469, 9.501678466796875, 17.58441162109375, 4.555908203125, 29.918777465820312, 24.593902587890625, -23.55841064453125, -13.95034408569336, 19.814407348632812, 27.120880126953125, 15.95074462890625, 6.8399658203125, 6.768184661865234, 6.51751708984375, -15.2593994140625, 33.34222412109375, 17.732818603515625, 20.047698974609375, 22.93466567993164, 5.495643615722656, 10.801910400390625, 19.1500244140625, 32.61515808105469, 0.704925537109375, -24.098785400390625, -25.482421875, 17.381439208984375, 2.94012451171875, -20.172653198242188, 22.443603515625, 17.97418212890625, 24.31542205810547, 44.3892822265625, 13.498626708984375, 3.7281723022460938, 6.458038330078125, 2.144775390625, 6.46588134765625, 17.674484252929688, 20.799713134765625, -3.64013671875, 23.7554931640625, 32.22462463378906, -2.5989913940429688, 2.3773651123046875, -4.49932861328125, 18.576797485351562, 17.457763671875, 9.47027587890625, -0.3424186706542969, 0.4647369384765625, 2.893402099609375, 26.3946533203125, 19.73089599609375, 27.8311767578125, 39.8018798828125, -7.5929412841796875, 12.64361572265625, -0.14634132385253906, 21.023757934570312, -6.420982360839844, 0.0, -9.6639404296875, 10.563583374023438, -12.43994140625, -6.680925369262695, -20.24456787109375, 5.714752197265625, 15.664665222167969, -10.914459228515625, -0.10271453857421875, 31.95458984375, -0.4195556640625, 17.56353759765625, 6.7941741943359375, 25.567626953125, 24.78887939453125, 16.596298217773438, -9.416061401367188, 25.939788818359375, 4.7794189453125, 22.739593505859375, 9.499725341796875, -5.7378387451171875, 12.901771545410156, 25.266510009765625, -20.26275634765625, 1.299102783203125, -11.131500244140625, -8.381149291992188, 51.279388427734375, -5.383209228515625, 22.736373901367188, 7.4959259033203125, -1.7734375, 9.026092529296875, 1.8816871643066406, -13.403457641601562, 12.35748291015625, -7.3095855712890625, 5.188945770263672, 36.24702453613281, 8.7794189453125, 13.3443603515625, -0.1078948974609375, 10.655364990234375, 4.853515625, 3.3578643798828125, 10.44383430480957, 32.45310974121094, -6.712169647216797, -8.3555908203125, 17.764122009277344, 13.138626098632812, 5.429931640625, 3.110626220703125, -21.977325439453125, 11.0413818359375, 7.6566619873046875, 22.177734375, 1.9521522521972656, 15.04327392578125, 4.108009338378906, 22.791748046875, 0.0, 4.9748077392578125, 16.2799072265625, 50.84051513671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000118.npy"}
{"epoch": 0.24712041884816754, "step": 119, "batch_size": 128, "mean": 7.648219585418701, "std": 18.33664321899414, "min": -40.08428955078125, "p10": -14.516098022460938, "median": 5.276100158691406, "p90": 30.548767089843746, "max": 67.14141845703125, "pos_frac": 0.671875, "sample": [-40.08428955078125, -18.606460571289062, 24.179656982421875, 2.6006698608398438, 2.1949844360351562, 3.2475509643554688, -9.353019714355469, -6.94390869140625, 8.13330078125, -3.3242034912109375, 22.0814208984375, -11.107215881347656, -12.3580322265625, 32.31292724609375, -18.3458251953125, 10.037384033203125, -14.662109375, 24.74554443359375, 28.16741180419922, -3.023223876953125, 8.030120849609375, 18.80767822265625, -1.1641578674316406, -9.117050170898438, 6.6631317138671875, 15.919044494628906, 21.087196350097656, 6.232421875, 32.94346618652344, 5.2290802001953125, 4.494424819946289, 2.357574462890625, -3.513336181640625, 1.425994873046875, 11.876861572265625, 11.8179931640625, 13.003814697265625, 4.253143310546875, 8.28485107421875, 2.145122528076172, 23.824249267578125, -9.444541931152344, -1.513153076171875, 0.395263671875, 31.708740234375, 0.9581146240234375, 3.66534423828125, -35.44903564453125, -14.453521728515625, -21.78021240234375, 51.013214111328125, 11.852554321289062, 4.0691986083984375, 29.680450439453125, -6.3191070556640625, -1.14874267578125, 13.528167724609375, 2.5354080200195312, 15.963706970214844, 23.31945037841797, 3.370330810546875, -0.12127685546875, 23.672897338867188, 14.86602783203125, 19.35955810546875, -8.69744873046875, 40.516143798828125, 11.704341888427734, 32.583892822265625, 18.82659912109375, 0.9039878845214844, 7.56292724609375, 13.35345458984375, 0.0, 39.747100830078125, -0.45306396484375, -0.8271026611328125, -24.07421875, 21.030029296875, -24.204208374023438, 21.957256317138672, 12.332687377929688, 5.930694580078125, -15.60089111328125, 13.50823974609375, 0.4810028076171875, 30.0516357421875, -7.410297393798828, 16.516799926757812, 6.5261077880859375, 22.35894775390625, -16.63824462890625, -3.054473876953125, 58.8699951171875, -4.9870147705078125, 67.14141845703125, 7.533935546875, -1.030914306640625, 8.48828125, -26.48956298828125, 24.12835693359375, 32.83709716796875, 4.55609130859375, 18.505020141601562, -6.6678619384765625, 20.573974609375, -6.904380798339844, 5.3231201171875, 4.490974426269531, -11.40765380859375, 3.0572052001953125, -22.83148193359375, 50.432098388671875, 14.913703918457031, 56.7694091796875, 7.1571502685546875, -2.9372711181640625, 16.5196533203125, -0.19525146484375, 19.056732177734375, 7.744171142578125, 4.2483673095703125, 27.898681640625, -7.526811599731445, 39.1646728515625, 4.1594085693359375, 5.54351806640625, -18.31768798828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000119.npy"}
{"epoch": 0.24921465968586387, "step": 120, "batch_size": 128, "mean": 9.147024154663086, "std": 17.108680725097656, "min": -30.65838623046875, "p10": -13.176852416992187, "median": 10.137710571289062, "p90": 32.19825439453125, "max": 57.695159912109375, "pos_frac": 0.671875, "sample": [-12.80462646484375, 24.485687255859375, -5.8602294921875, 12.902084350585938, -2.7284011840820312, 11.301513671875, 23.434112548828125, -11.935745239257812, 7.4416961669921875, 15.595123291015625, -2.28424072265625, -5.0749969482421875, 21.131668090820312, 4.3060455322265625, -15.758056640625, -1.19573974609375, 32.581787109375, -2.963205337524414, 10.351776123046875, 17.6657772064209, -5.238433837890625, 32.563323974609375, 32.095916748046875, 26.69354248046875, 26.218109130859375, 11.396743774414062, 6.0687408447265625, -1.66400146484375, -5.3896026611328125, 31.24017333984375, 25.14141845703125, 0.0599365234375, 6.748992919921875, 4.677001953125, -26.051971435546875, 7.971771240234375, 0.38817596435546875, 14.892814636230469, 21.64385986328125, 20.296737670898438, 0.5515556335449219, -1.2038726806640625, 13.52178955078125, -25.29595947265625, 0.0, -2.4308509826660156, -21.371337890625, -1.8229522705078125, -6.325904846191406, 4.7808685302734375, 11.14984130859375, 12.307567596435547, 52.130615234375, 20.568756103515625, 14.6112060546875, 9.92364501953125, -14.045379638671875, 14.564788818359375, 7.989898681640625, 2.2753829956054688, -4.1092681884765625, 0.0, 12.671295166015625, 20.7808837890625, 57.695159912109375, 15.76666259765625, 16.462921142578125, 37.52227783203125, -16.2919921875, 20.696861267089844, 20.5128173828125, 8.485183715820312, 4.150718688964844, 18.13360595703125, 13.539138793945312, -6.079986572265625, 11.224609375, 33.26741027832031, 19.2388916015625, 19.407958984375, 12.79742431640625, -16.085784912109375, -18.652389526367188, -7.899261474609375, -11.250640869140625, 38.417205810546875, 9.293357849121094, 35.76287841796875, 43.886688232421875, 15.920890808105469, -0.7531089782714844, 27.2259521484375, -29.3603515625, 32.437042236328125, 38.1029052734375, 6.348663330078125, -0.2136993408203125, 17.275802612304688, 11.885055541992188, 17.854782104492188, -22.091575622558594, 2.615398406982422, 0.7781906127929688, 18.710433959960938, 26.879104614257812, -0.8361701965332031, 12.070236206054688, -1.5410003662109375, 39.9371337890625, -30.65838623046875, 23.049835205078125, 21.919708251953125, 20.206016540527344, -17.035400390625, -23.795562744140625, 15.96258544921875, 14.108283996582031, 19.60272216796875, 6.566070556640625, -0.7321319580078125, -3.9146194458007812, 15.775749206542969, -3.3406829833984375, 51.18682861328125, 2.4487380981445312, 3.33160400390625, -3.8985748291015625, 15.221023559570312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000120.npy"}
{"epoch": 0.2513089005235602, "step": 121, "batch_size": 128, "mean": 5.406813621520996, "std": 15.347919464111328, "min": -35.970001220703125, "p10": -12.352769470214843, "median": 4.288200378417969, "p90": 24.039091491699207, "max": 47.33880615234375, "pos_frac": 0.65625, "sample": [-1.3105850219726562, -0.349273681640625, 4.351072311401367, 6.966468811035156, 0.7086029052734375, 3.7672882080078125, -3.616455078125, 6.934619903564453, 8.3463134765625, 28.758544921875, 3.2777366638183594, -0.96551513671875, 34.705535888671875, 4.778240203857422, 17.45049285888672, 0.699005126953125, -2.8715667724609375, 3.8831329345703125, 0.2418804168701172, 22.701492309570312, 18.060272216796875, -17.219497680664062, 18.666229248046875, 16.27532958984375, -23.99603271484375, 0.0, 8.603446960449219, 2.89105224609375, 2.56201171875, 21.284637451171875, 9.986703872680664, 8.700050354003906, -33.9813232421875, 27.16015625, 5.6106414794921875, -35.970001220703125, -3.1497650146484375, 5.473197937011719, 22.50604248046875, 12.437698364257812, -3.938446044921875, -10.335662841796875, 7.649772644042969, 1.792485237121582, 6.13690185546875, 1.5668182373046875, 16.595199584960938, -12.181304931640625, -9.782440185546875, 0.0, 7.860012054443359, -4.441851615905762, -0.8944778442382812, -1.59576416015625, 14.613113403320312, -7.377845764160156, 0.34228515625, 5.357086181640625, 18.688568115234375, -12.752853393554688, -2.37017822265625, -18.432876586914062, -2.715320587158203, 45.9058837890625, 13.008956909179688, 9.402587890625, 43.73762512207031, -5.6776123046875, -3.907073974609375, -2.063079833984375, -4.746822357177734, 11.708984375, -3.38995361328125, 14.296867370605469, -9.539505004882812, -0.68310546875, 4.380535125732422, 7.2495269775390625, 0.5256500244140625, 15.84149169921875, 13.439773559570312, -1.3132038116455078, -6.15692138671875, 28.768356323242188, 46.346099853515625, 5.871826171875, -25.624908447265625, -1.09027099609375, -20.193603515625, 47.33880615234375, 15.943801879882812, 19.326766967773438, 18.45611572265625, 4.237457275390625, 33.38653564453125, 2.2476348876953125, -23.10302734375, 5.2320709228515625, 13.07110595703125, 0.092864990234375, 2.604856491088867, 7.4300689697265625, 14.1500244140625, 7.4827880859375, 4.173248291015625, 14.90423583984375, 0.10595703125, 4.3389434814453125, -3.4127578735351562, -3.38104248046875, 4.191925048828125, 35.322357177734375, 10.709884643554688, 5.00103759765625, 9.232666015625, 9.058082580566406, 29.477203369140625, -23.75115966796875, 28.23876953125, 13.982421875, -0.938934326171875, -15.63128662109375, 40.64326477050781, 13.137359619140625, 0.7254047393798828, -17.860382080078125, -24.3453369140625, 16.013229370117188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000121.npy"}
{"epoch": 0.2534031413612565, "step": 122, "batch_size": 128, "mean": 7.318086624145508, "std": 18.359853744506836, "min": -38.3751220703125, "p10": -16.577923583984376, "median": 6.8195648193359375, "p90": 31.022917175292967, "max": 60.22015380859375, "pos_frac": 0.65625, "sample": [22.347625732421875, -16.831451416015625, 26.204376220703125, -14.3287353515625, 14.1019287109375, 22.442794799804688, 6.962371826171875, 3.534759521484375, -34.22662353515625, 29.422454833984375, 25.850143432617188, 0.0, 9.229736328125, -26.728790283203125, 0.22368621826171875, -38.3751220703125, 10.301345825195312, 15.861114501953125, -12.902206420898438, 14.279365539550781, -3.26123046875, 21.79840087890625, -22.84356689453125, -32.201904296875, 6.6767578125, 36.21685791015625, -11.7738037109375, 9.983844757080078, 45.728363037109375, 2.9695281982421875, 8.534271240234375, -2.865753173828125, 7.7618560791015625, 8.039337158203125, 17.951751708984375, -6.212898254394531, 4.3457489013671875, -21.938629150390625, 1.48797607421875, 10.535484313964844, -28.5897216796875, 13.3741455078125, 7.18170166015625, 45.40679931640625, -2.653594970703125, -24.312744140625, -11.360076904296875, 8.74183464050293, 40.117706298828125, -9.105194091796875, -6.89990234375, 10.030960083007812, 11.62677001953125, -19.953948974609375, 12.5235595703125, 15.024871826171875, -1.3414459228515625, 22.72500228881836, -28.7608642578125, 10.083984375, 13.69482421875, 22.35595703125, 18.31158447265625, 0.1429290771484375, -7.407783508300781, 33.403778076171875, 15.584541320800781, 25.8699951171875, 21.36029052734375, 31.513519287109375, 46.4664306640625, 12.99305534362793, 0.9601287841796875, 36.74359130859375, 60.22015380859375, 7.798545837402344, 4.7680511474609375, -1.290252685546875, 30.823867797851562, 24.16290283203125, -5.882698059082031, -16.469268798828125, 2.297698974609375, -9.179229736328125, -5.8724365234375, -4.15350341796875, -1.4734382629394531, 12.016685485839844, 6.2725982666015625, -7.1597442626953125, 17.08544921875, 0.43536376953125, 2.3768081665039062, -19.55560302734375, -3.49200439453125, -19.38665771484375, -1.4619598388671875, 27.6593017578125, 24.24029541015625, -1.58465576171875, 4.8846893310546875, -6.5086669921875, 22.632949829101562, 18.85394287109375, 2.353179931640625, 24.3895263671875, 21.319915771484375, 29.01287078857422, 5.311851501464844, 1.41351318359375, -0.8793487548828125, -0.14575958251953125, -7.4215850830078125, 16.24591827392578, 48.873046875, 0.77978515625, 32.758819580078125, -3.40618896484375, 2.3785400390625, 25.501113891601562, 11.9561767578125, 22.7137451171875, -3.9961681365966797, 34.09210205078125, -5.153999328613281, 1.985687255859375, 31.48736572265625, 7.93194580078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000122.npy"}
{"epoch": 0.2554973821989529, "step": 123, "batch_size": 128, "mean": 10.546405792236328, "std": 18.723281860351562, "min": -49.053375244140625, "p10": -12.573419189453125, "median": 9.449958801269531, "p90": 34.120846557617185, "max": 67.47772216796875, "pos_frac": 0.7421875, "sample": [24.897918701171875, -13.027496337890625, 30.328201293945312, -12.04486083984375, 7.015899658203125, 6.023479461669922, 10.341690063476562, 3.446380615234375, 13.00408935546875, 11.236114501953125, 0.09164810180664062, 3.829376220703125, -2.7179718017578125, 2.798431396484375, 5.9558868408203125, -11.645050048828125, -12.378814697265625, 16.7176513671875, 44.2957763671875, 8.790313720703125, -24.31317138671875, -23.400146484375, 9.189071655273438, 17.005945205688477, 49.33056640625, 3.334136962890625, 20.7364501953125, 29.4737548828125, 28.8828125, 1.566131591796875, 28.430908203125, 23.1822509765625, 20.03302001953125, 24.35650634765625, 8.029296875, 23.79345703125, 24.02410888671875, -24.631317138671875, -0.67218017578125, 29.805999755859375, 17.31292724609375, 4.491790771484375, 0.42572021484375, 34.943359375, 10.43670654296875, 57.94036865234375, 1.3280353546142578, 3.325439453125, 18.9189453125, 30.4825439453125, 3.5343017578125, 18.400253295898438, 21.284759521484375, 67.47772216796875, 45.75469970703125, 12.211692810058594, -6.103916168212891, -0.2114410400390625, 3.30487060546875, 47.049896240234375, -2.431976318359375, 6.016225814819336, -49.053375244140625, 34.1290283203125, 9.123710632324219, -16.839111328125, 23.561538696289062, 6.44091796875, 36.498565673828125, 7.488325119018555, 11.22894287109375, 0.605621337890625, 7.52906608581543, -18.267311096191406, 11.743415832519531, 26.731658935546875, 23.393035888671875, -3.641845703125, -19.617279052734375, 41.816070556640625, -2.4732666015625, 10.27276611328125, 20.957508087158203, -4.48931884765625, 8.449462890625, 20.137420654296875, -18.91845703125, 17.79399871826172, 18.51531982421875, -7.337554931640625, 0.0, -3.314453125, 37.5810546875, 10.931365966796875, 9.43377685546875, 0.0, 13.916763305664062, 10.559494018554688, 0.9724884033203125, 19.089202880859375, -5.242061614990234, -13.266998291015625, 6.542816162109375, 27.415679931640625, 0.60284423828125, 12.626914978027344, 9.9481201171875, 34.117340087890625, 15.891006469726562, -16.776901245117188, -3.0863037109375, -3.48077392578125, 31.069671630859375, 9.466140747070312, -32.62049865722656, 47.958282470703125, 12.9898681640625, -7.717193603515625, 14.08306884765625, -2.4481277465820312, 0.1632843017578125, 12.27801513671875, 6.487579345703125, 34.02801513671875, 45.195556640625, -19.8104248046875, 16.8140869140625, 20.75323486328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000123.npy"}
{"epoch": 0.25759162303664923, "step": 124, "batch_size": 128, "mean": 9.483943939208984, "std": 17.332170486450195, "min": -52.377960205078125, "p10": -9.356423950195312, "median": 7.222286224365234, "p90": 33.62046508789062, "max": 57.41514587402344, "pos_frac": 0.765625, "sample": [0.50396728515625, 28.794532775878906, -52.377960205078125, 18.319580078125, -3.1669178009033203, 2.902252197265625, 26.161605834960938, 7.9594573974609375, 33.748382568359375, 46.25238037109375, 5.6508941650390625, -27.82135009765625, 6.400299072265625, -9.162322998046875, 22.962493896484375, 4.63330078125, 6.4685211181640625, 18.909149169921875, -6.409612655639648, 13.569366455078125, 6.1591644287109375, 5.071998596191406, -0.0100250244140625, 10.66021728515625, 20.66796875, 3.412750244140625, -1.554168701171875, -4.612964630126953, 3.5796127319335938, 8.763641357421875, 27.370819091796875, -13.8248291015625, 4.237220764160156, 23.896575927734375, 44.450347900390625, 38.49760437011719, 19.055152893066406, 12.264083862304688, 5.142822265625, 23.97540283203125, 22.241043090820312, 17.535614013671875, 39.099639892578125, 9.8028564453125, 3.462188720703125, -1.9983367919921875, -0.3281402587890625, 13.89984130859375, 45.6993408203125, -9.809326171875, 9.08197021484375, -17.729534149169922, 7.293296813964844, 0.0, 2.2892932891845703, -2.9752120971679688, 26.00372314453125, 6.83111572265625, 3.776580810546875, -11.495536804199219, 33.565643310546875, -37.503082275390625, -27.02020263671875, 38.9903564453125, 13.51007080078125, 24.1136474609375, 7.151275634765625, 29.930816650390625, 25.28965950012207, 0.75128173828125, 9.68792724609375, 5.141265869140625, -20.4591064453125, -26.789398193359375, 23.06043243408203, 1.4645881652832031, 4.490730285644531, 12.520195007324219, 34.22064208984375, 21.516860961914062, 10.452301025390625, 30.656158447265625, 8.03717041015625, -16.1134033203125, -5.2020416259765625, 6.246429443359375, 2.3260650634765625, 14.88427734375, 35.000762939453125, 15.887531280517578, 2.93450927734375, 14.076347351074219, -4.1643218994140625, 12.102554321289062, 12.431716918945312, 8.178091049194336, -1.280303955078125, 57.41514587402344, 12.258039474487305, 20.683570861816406, -3.067962646484375, 0.59039306640625, 14.554962158203125, 3.2068023681640625, 0.527618408203125, 54.08062744140625, -11.775299072265625, 0.085693359375, 6.441562652587891, 14.9354248046875, -2.0562591552734375, -0.2820281982421875, 5.866424560546875, 11.211044311523438, 7.4413604736328125, 20.70003890991211, 9.338211059570312, 35.628509521484375, 0.25920677185058594, -1.701263427734375, 5.0698394775390625, 15.34490966796875, 23.58648681640625, 43.3265380859375, 4.09796142578125, 10.2559814453125, 4.612281799316406, -12.960311889648438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000124.npy"}
{"epoch": 0.25968586387434556, "step": 125, "batch_size": 128, "mean": 9.2630033493042, "std": 18.938236236572266, "min": -34.09315490722656, "p10": -12.47395725250244, "median": 6.531959533691406, "p90": 34.610949707031246, "max": 63.96490478515625, "pos_frac": 0.6328125, "sample": [5.6175079345703125, -9.985355377197266, -32.98487091064453, 38.45904541015625, 1.7731952667236328, 13.54638671875, 16.337478637695312, -11.24102783203125, 3.528008460998535, -13.698944091796875, 3.0966796875, -1.4376220703125, 32.79046630859375, -1.891845703125, 13.75140380859375, 26.022247314453125, -27.226318359375, 14.404632568359375, 13.91033935546875, 14.768959045410156, -8.43609619140625, -3.655956268310547, 18.6141357421875, -5.436973571777344, 14.42315673828125, 13.097099304199219, 10.657318115234375, 15.7008056640625, -4.6957244873046875, 44.200927734375, 4.578605651855469, -6.441253662109375, 11.944808959960938, -1.6364364624023438, 6.82049560546875, 32.61106872558594, 21.2576904296875, 35.98736572265625, 14.958786010742188, -3.2349395751953125, 0.134185791015625, -22.787353515625, -12.542022705078125, -6.994293212890625, -4.4087982177734375, 43.7613525390625, 2.622406005859375, -7.616241455078125, -34.09315490722656, 15.09246826171875, 15.133544921875, -17.088653564453125, 24.272369384765625, 63.96490478515625, 6.6805267333984375, 14.381446838378906, 8.275260925292969, 19.690704345703125, 19.14581298828125, 27.5400390625, -2.809661865234375, 36.61773681640625, 13.845184326171875, -0.9594535827636719, -0.11083984375, 16.275848388671875, 4.303253173828125, 1.5329170227050781, 6.58636474609375, 5.8532562255859375, 56.01361083984375, 10.83553695678711, 29.510971069335938, 41.14959716796875, -12.460226058959961, -19.983444213867188, 6.3561553955078125, 22.6973876953125, 34.02105712890625, 24.45611572265625, 31.116180419921875, 3.6313934326171875, -0.574188232421875, 44.9876708984375, -1.9857177734375, 13.2921142578125, -13.389389038085938, -7.54962158203125, 21.128955841064453, 31.506256103515625, 28.574752807617188, 3.48736572265625, -3.940185546875, 12.624267578125, -9.561676025390625, -1.3166351318359375, -9.403045654296875, 28.146759033203125, 28.132537841796875, -1.591949462890625, -6.6834259033203125, 0.0, 3.2209033966064453, -5.1775665283203125, 40.159149169921875, -8.828826904296875, -0.0023956298828125, 4.5760040283203125, 6.4775543212890625, 45.2149658203125, -7.51654052734375, -12.505996704101562, 21.339263916015625, 30.741897583007812, 0.6486282348632812, 26.542938232421875, 49.52972412109375, 10.485931396484375, -15.44879150390625, 17.683349609375, 26.559051513671875, -32.2374267578125, -7.204704284667969, 47.60882568359375, 14.528450012207031, 19.037460327148438, -3.1436538696289062, -17.007354736328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000125.npy"}
{"epoch": 0.2617801047120419, "step": 126, "batch_size": 128, "mean": 10.66677474975586, "std": 19.36330795288086, "min": -44.74896240234375, "p10": -11.654122924804687, "median": 9.438804626464844, "p90": 34.6331069946289, "max": 53.1827392578125, "pos_frac": 0.703125, "sample": [21.928009033203125, 51.246002197265625, 13.578857421875, 17.441513061523438, 31.502685546875, 42.608001708984375, 12.185569763183594, 39.8067626953125, 4.1004638671875, -1.0654449462890625, -44.74896240234375, 24.714500427246094, 2.6702423095703125, 18.225906372070312, 27.33884048461914, 5.4422760009765625, 26.51470947265625, -5.8136749267578125, -21.96728515625, -5.8387451171875, -14.597381591796875, 30.672164916992188, -10.69952392578125, -7.185842514038086, 2.5245361328125, -9.718032836914062, -4.56866455078125, -2.9364242553710938, 4.442901611328125, -7.226226806640625, 8.0963134765625, -2.1625518798828125, 16.902931213378906, 12.461959838867188, 32.395233154296875, -12.016021728515625, -8.474815368652344, 1.272308349609375, -1.5977935791015625, 4.253843307495117, 9.100234985351562, 23.810611724853516, 33.753509521484375, -1.3511962890625, 14.13909912109375, 23.35443115234375, 7.861774444580078, 18.520599365234375, -15.325851440429688, -3.486083984375, 33.873931884765625, -0.12807846069335938, 11.881668090820312, 49.37860107421875, 3.304595947265625, 36.59446716308594, 15.915824890136719, 6.870941162109375, 4.0883331298828125, 20.98226547241211, 11.348304748535156, 32.66650390625, 21.115478515625, 25.928192138671875, -15.005523681640625, 17.049713134765625, 5.094730377197266, 24.034820556640625, 21.68499755859375, 47.43328857421875, 39.6290283203125, -3.571990966796875, 27.599273681640625, 28.622695922851562, -2.0798568725585938, 8.097259521484375, 10.4464111328125, -7.8852996826171875, 0.3890838623046875, -25.1134033203125, 28.535919189453125, 11.70526123046875, -2.8820133209228516, 44.51507568359375, 20.752655029296875, 3.121410369873047, -30.4715576171875, 6.153656005859375, 20.8780517578125, -35.35882568359375, 12.31500244140625, -10.386054992675781, 25.978195190429688, 31.744598388671875, -43.58148193359375, -10.684890747070312, 4.146276473999023, 9.777374267578125, 6.453521728515625, -16.755767822265625, 23.756736755371094, 27.062744140625, 34.17543029785156, 43.62371826171875, 7.894134521484375, 20.591094970703125, -7.7647705078125, 47.711029052734375, 21.544281005859375, 39.79948425292969, -12.2210693359375, 19.941986083984375, 17.96569061279297, 35.701019287109375, 20.513347625732422, 7.144989013671875, 4.171966552734375, -1.022613525390625, 10.30499267578125, -1.3736572265625, 13.846435546875, -11.4990234375, 2.5814208984375, 4.5469970703125, 29.830291748046875, 53.1827392578125, 4.702354431152344, -19.6534423828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000126.npy"}
{"epoch": 0.2638743455497382, "step": 127, "batch_size": 128, "mean": 12.303199768066406, "std": 19.349016189575195, "min": -42.4425048828125, "p10": -10.493290710449218, "median": 8.87132453918457, "p90": 38.175595092773435, "max": 55.9068603515625, "pos_frac": 0.734375, "sample": [-0.9582023620605469, -1.3227386474609375, -9.82244873046875, 25.669891357421875, 10.048614501953125, 2.02471923828125, 18.5926513671875, 11.860950469970703, -11.319770812988281, -42.4425048828125, 30.08099365234375, 2.771331787109375, 19.305206298828125, 17.794401168823242, 9.863067626953125, -31.874755859375, -8.215972900390625, 12.752334594726562, -17.612754821777344, 7.46209716796875, 7.52996826171875, 7.20306396484375, 21.6761474609375, 3.817718505859375, 20.5325927734375, 48.0784912109375, 22.51287841796875, 35.680328369140625, 12.557174682617188, 25.190170288085938, 37.900177001953125, 15.026073455810547, 29.34228515625, 4.242170333862305, 11.07745361328125, 46.03204345703125, -10.130729675292969, -19.703460693359375, -1.3254165649414062, 8.670539855957031, 46.805938720703125, 7.9356689453125, 7.2337493896484375, 5.66717529296875, -15.0870361328125, -25.150283813476562, 35.865020751953125, 36.42729187011719, 11.907112121582031, -21.265380859375, 18.726242065429688, 34.736602783203125, 20.660858154296875, -7.87139892578125, 29.426971435546875, 41.90777587890625, 24.760833740234375, 15.061294555664062, -3.74627685546875, 3.369190216064453, 2.01531982421875, 5.143341064453125, 37.192657470703125, 23.25067138671875, 8.007743835449219, 34.2264404296875, -5.503631591796875, 5.3509368896484375, 29.607025146484375, 38.8182373046875, 26.006881713867188, -2.8706283569335938, -11.489990234375, -2.3180694580078125, 9.9951171875, -11.261489868164062, -3.2105712890625, -15.146591186523438, 7.9618377685546875, -3.209331512451172, 2.894561767578125, 22.139480590820312, 55.9068603515625, -2.042858123779297, 41.072265625, 18.130661010742188, 4.452056884765625, 55.6417236328125, 29.522308349609375, 21.55609130859375, 46.2257080078125, -8.909408569335938, 6.72686767578125, 7.4874267578125, 3.9869422912597656, -2.3140697479248047, -20.531442642211914, 9.571929931640625, 9.07210922241211, 31.5906982421875, -4.6190643310546875, 8.195526123046875, -9.95611572265625, 30.619094848632812, 23.49346160888672, 4.854949951171875, 39.938690185546875, 48.74969482421875, 1.994140625, -4.5898284912109375, 6.8405914306640625, 2.686248779296875, 24.0458984375, 34.67755126953125, -18.12909698486328, 4.3893585205078125, 36.257354736328125, 41.17706298828125, -2.210773468017578, 11.435379028320312, 29.0894775390625, 22.5126953125, 3.47760009765625, 32.378387451171875, 2.5755615234375, -10.1640625, 23.45916748046875, 48.94671630859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000127.npy"}
{"epoch": 0.26596858638743454, "step": 128, "batch_size": 128, "mean": 13.558063507080078, "std": 20.886171340942383, "min": -43.57171630859375, "p10": -10.929864501953125, "median": 12.845868110656738, "p90": 37.81118316650391, "max": 74.89227294921875, "pos_frac": 0.734375, "sample": [27.295120239257812, 3.37493896484375, 19.872024536132812, 25.17779541015625, -32.47119140625, -33.120208740234375, -18.5150146484375, 6.70843505859375, 25.586517333984375, 28.605575561523438, 24.098655700683594, 2.286865234375, 35.40428924560547, 0.662109375, 37.84407043457031, 0.7606277465820312, 3.507427215576172, 24.055984497070312, 27.889434814453125, 24.232818603515625, -2.8910675048828125, -1.5695114135742188, 36.04607391357422, 67.58563232421875, 4.5831298828125, 7.268531799316406, -3.52960205078125, 5.215179443359375, 29.323760986328125, 20.897613525390625, 9.752212524414062, 12.101282119750977, -19.157562255859375, 34.39836120605469, -7.68189811706543, 32.05351257324219, -16.738143920898438, 6.4686279296875, 0.784210205078125, -21.12957763671875, 25.300750732421875, 3.823974609375, -2.664703369140625, 18.2769775390625, 35.332305908203125, -1.514739990234375, -0.11992263793945312, 24.36883544921875, 19.18975830078125, 37.797088623046875, 35.8741455078125, 27.0714111328125, -3.777130126953125, 35.423095703125, -2.5452728271484375, -6.4234466552734375, 3.95135498046875, 25.01300048828125, 19.20263671875, 41.37677001953125, 27.275821685791016, 3.1956634521484375, 14.06744384765625, 0.05400848388671875, 0.8275527954101562, 7.4585418701171875, 40.7874755859375, 29.101318359375, 26.903472900390625, 13.5904541015625, 15.009857177734375, 28.585525512695312, 49.832275390625, 15.8505859375, 47.87091064453125, 35.3134765625, 9.339111328125, -5.773284912109375, -2.95391845703125, 0.20505142211914062, 40.79362487792969, 31.4827880859375, -14.841278076171875, 19.366180419921875, 29.166290283203125, -43.57171630859375, 21.562774658203125, -21.476577758789062, 18.582359313964844, 74.89227294921875, 7.25323486328125, -11.152374267578125, 36.07530212402344, 16.853546142578125, -8.59197998046875, 0.9710044860839844, -16.126022338867188, -7.0288848876953125, 7.23052978515625, 11.717460632324219, 36.723541259765625, -5.6428070068359375, 5.14129638671875, 46.957130432128906, 22.345420837402344, 34.576568603515625, -7.256481170654297, -6.13201904296875, 55.3731689453125, 0.7508201599121094, 6.910736083984375, 0.0, -22.392791748046875, -7.5719451904296875, 26.901123046875, -11.650146484375, -10.834503173828125, 15.283477783203125, 23.899459838867188, 42.041168212890625, 4.46905517578125, 18.3931884765625, 10.950546264648438, 39.775726318359375, 62.703460693359375, 25.192413330078125, 19.21435546875, -2.3837661743164062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000128.npy"}
{"epoch": 0.2680628272251309, "step": 129, "batch_size": 128, "mean": 12.47940444946289, "std": 21.58934211730957, "min": -48.25146484375, "p10": -12.749890136718749, "median": 11.716987609863281, "p90": 41.63941345214844, "max": 79.21585083007812, "pos_frac": 0.7578125, "sample": [-48.25146484375, 29.1300048828125, -21.6988525390625, -13.566650390625, 11.950447082519531, 45.09889602661133, 27.52398681640625, 37.293663024902344, 30.19927978515625, 3.46484375, 32.54888916015625, 2.8173961639404297, 12.913055419921875, -1.513580322265625, -6.832790374755859, -10.931854248046875, 27.517120361328125, 35.06634521484375, 11.724380493164062, -9.964988708496094, 12.013500213623047, 12.056938171386719, 1.0099716186523438, 20.18865966796875, 42.033355712890625, -6.9622344970703125, 16.096267700195312, -2.575366973876953, 79.21585083007812, 62.6131591796875, 4.855194091796875, 23.476654052734375, 53.41746520996094, 14.67041015625, 1.5236587524414062, -44.09033203125, 34.48992919921875, 6.08465576171875, 62.23345947265625, 10.804561614990234, -4.893856048583984, 13.46600341796875, 40.20306396484375, 44.849578857421875, 12.756332397460938, 43.37652587890625, -23.9049072265625, 10.52436637878418, 0.7389907836914062, 4.003410339355469, 26.445953369140625, 18.047760009765625, 9.681961059570312, 26.965576171875, -12.5159912109375, 8.263236999511719, 5.40972900390625, 43.67620849609375, 2.657470703125, 28.078216552734375, 1.8433990478515625, 2.3264389038085938, 41.4705810546875, 22.44507598876953, 61.14923095703125, -7.77850341796875, 16.43939208984375, 16.417495727539062, 26.914337158203125, 4.5060577392578125, 4.72320556640625, 6.9658050537109375, 2.7733612060546875, -5.022285461425781, 2.46173095703125, -13.295654296875, 26.773338317871094, 19.70842742919922, -3.4072608947753906, 59.710205078125, 27.59564208984375, 9.404365539550781, 19.909912109375, 3.9513778686523438, 2.640178680419922, 2.329315185546875, 29.76483154296875, 3.9976806640625, 18.935272216796875, 38.92314147949219, -9.91387939453125, -27.1475830078125, -2.2817344665527344, -0.30682373046875, -7.739021301269531, 12.88525390625, -11.417999267578125, -32.615020751953125, 15.61895751953125, 11.7095947265625, 45.680084228515625, -15.948165893554688, 13.747161865234375, -4.019819259643555, -18.9508056640625, -21.720916748046875, 12.028244018554688, 3.030303955078125, 15.352340698242188, 23.190582275390625, 26.464645385742188, 6.2452239990234375, 16.9542236328125, 9.897003173828125, 9.133697509765625, 46.8165283203125, 10.325393676757812, 30.8126220703125, -19.97711181640625, 18.609359741210938, 11.889518737792969, -0.5995693206787109, 1.336181640625, 30.547073364257812, 13.855316162109375, -15.102058410644531, 26.9842529296875, 31.94119644165039], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000129.npy"}
{"epoch": 0.27015706806282724, "step": 130, "batch_size": 128, "mean": 11.577238082885742, "std": 19.5683536529541, "min": -38.839599609375, "p10": -11.15162963867187, "median": 9.14584732055664, "p90": 37.92990722656249, "max": 56.71501159667969, "pos_frac": 0.71875, "sample": [6.709075927734375, -3.2859954833984375, 30.40411376953125, 5.141876220703125, 5.387115478515625, 40.69927978515625, 9.044036865234375, 28.01556396484375, 41.355628967285156, -9.396942138671875, -2.353961944580078, 9.490966796875, 21.712142944335938, 40.41363525390625, -2.04376220703125, 5.94403076171875, -3.736083984375, -16.890350341796875, 9.376869201660156, 20.47833251953125, 16.674598693847656, 40.27935791015625, -1.674041748046875, 9.110946655273438, -2.686901092529297, 56.71501159667969, 1.634552001953125, 19.760650634765625, 47.23052978515625, 49.04875183105469, -31.56939697265625, -0.234588623046875, 0.0, -9.782989501953125, 25.985748291015625, 13.638214111328125, 23.023712158203125, -37.6307373046875, 9.103553771972656, 21.465301513671875, 25.473796844482422, 14.584442138671875, 48.62664794921875, -8.787574768066406, -38.839599609375, -35.912811279296875, -4.494842529296875, 34.48133850097656, 4.77056884765625, -1.126495361328125, 37.3382568359375, 1.4017410278320312, 12.156326293945312, 33.07660675048828, -17.405517578125, -0.587005615234375, 12.881834030151367, 19.086517333984375, 35.77996826171875, -2.55291748046875, 9.180747985839844, 1.0895843505859375, 21.78759765625, 17.618408203125, 5.9647064208984375, 43.74494934082031, 29.94244384765625, -16.693572998046875, -14.345123291015625, -1.288299560546875, 40.604034423828125, 7.6246490478515625, 30.486572265625, -23.063720703125, 5.560829162597656, 18.92498779296875, -15.003303527832031, 28.01239013671875, 8.806503295898438, 16.21675682067871, 28.803268432617188, 17.43084716796875, 30.465972900390625, 18.2833251953125, 32.354949951171875, 1.08074951171875, 32.736236572265625, 8.3798828125, 27.803985595703125, -3.4472427368164062, 10.563720703125, 6.4476318359375, 3.721588134765625, 2.895282745361328, 22.666000366210938, -28.5389404296875, -5.293670654296875, 4.7440185546875, 3.010162353515625, 46.20526123046875, 42.8592529296875, 34.994384765625, 20.899368286132812, 1.5791702270507812, -24.83843994140625, 27.453948974609375, 4.080497741699219, 5.299041748046875, 19.45281982421875, -4.9572296142578125, 34.64723205566406, 7.210277557373047, 27.779891967773438, -6.728584289550781, 19.09759521484375, 19.43396759033203, 39.3104248046875, 20.0555419921875, -0.4067535400390625, 17.497955322265625, 14.551414489746094, -6.753204345703125, -0.5761566162109375, 3.796539306640625, 28.302001953125, -25.2996826171875, 4.902774810791016, 26.24908447265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000130.npy"}
{"epoch": 0.27225130890052357, "step": 131, "batch_size": 128, "mean": 9.663520812988281, "std": 22.029722213745117, "min": -46.61297607421875, "p10": -17.412063598632812, "median": 8.610496520996094, "p90": 39.515966796875, "max": 71.7974853515625, "pos_frac": 0.640625, "sample": [30.484283447265625, 10.062276840209961, 36.991241455078125, -21.958343505859375, 12.465118408203125, 2.190093994140625, 2.134136199951172, 23.277984619140625, -16.612014770507812, 40.92816162109375, 31.774879455566406, 30.877655029296875, 43.837371826171875, 17.907958984375, -16.7362060546875, 1.4378509521484375, 2.633544921875, -16.26434326171875, 9.211593627929688, 49.50146484375, 34.977569580078125, 9.64410400390625, -5.05810546875, -4.991044998168945, 29.817962646484375, -4.8817596435546875, -10.958343505859375, 42.034912109375, 10.185676574707031, 12.73974609375, -2.0215988159179688, 58.16679382324219, -6.4267425537109375, 1.069366455078125, 20.38653564453125, -3.254301071166992, -23.537826538085938, 43.12042236328125, -1.897918701171875, 30.85650634765625, 8.025482177734375, 3.67254638671875, -2.0678367614746094, -16.993988037109375, 26.06085205078125, 6.7594757080078125, 57.44854736328125, -18.3875732421875, 19.286178588867188, 25.98992919921875, -2.3892974853515625, -3.938934326171875, -15.746917724609375, 14.84014892578125, 13.708038330078125, 38.381256103515625, -23.0218505859375, -5.7686767578125, 25.401042938232422, 34.266143798828125, -8.153030395507812, -16.9302978515625, -20.29266357421875, 30.78917694091797, 6.77398681640625, 5.869972229003906, 16.62812042236328, 29.25408935546875, 71.7974853515625, 7.580902099609375, -46.61297607421875, 3.3579025268554688, 10.104766845703125, -9.047649383544922, -3.4069900512695312, 18.985198974609375, -26.28302001953125, 62.867889404296875, -6.506595611572266, 35.75541687011719, 10.425689697265625, -22.958648681640625, 1.9887123107910156, -13.153045654296875, 12.714630126953125, 19.756072998046875, -13.61669921875, 39.166107177734375, 44.241302490234375, -24.201828002929688, 0.11931610107421875, -3.3244857788085938, 43.18182373046875, -21.007587432861328, 27.047698974609375, 6.473686218261719, 1.8618965148925781, 9.245758056640625, -6.7748260498046875, 17.219810485839844, 10.019767761230469, -21.47894287109375, 28.36572265625, -2.5958633422851562, 10.2166748046875, 33.92933654785156, -5.331512451171875, -31.888458251953125, 50.89990234375, 8.321853637695312, 34.240478515625, 40.332305908203125, 12.459991455078125, 20.0887451171875, -14.89263916015625, -4.3920135498046875, 20.4075927734375, -9.506317138671875, 27.400123596191406, 8.899139404296875, 19.08197021484375, 14.567276000976562, 23.65576171875, 17.54693603515625, -6.59796142578125, -21.085948944091797, -8.346588134765625, 2.0349998474121094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000131.npy"}
{"epoch": 0.2743455497382199, "step": 132, "batch_size": 128, "mean": 12.927886962890625, "std": 24.70322608947754, "min": -46.0321044921875, "p10": -17.112712097167968, "median": 10.2257080078125, "p90": 45.04612426757812, "max": 94.747802734375, "pos_frac": 0.6953125, "sample": [-10.70269775390625, 26.062255859375, 37.423858642578125, 13.53350830078125, -1.0093460083007812, -9.476654052734375, 26.56578826904297, 11.677017211914062, -4.454345703125, 37.67181396484375, 4.400402069091797, -11.049453735351562, -21.376495361328125, 35.31683349609375, 21.351226806640625, 6.8749237060546875, 4.4117431640625, 45.520538330078125, 43.67669677734375, 58.21728515625, 35.090728759765625, 31.134689331054688, 73.15377807617188, 47.722686767578125, 15.035240173339844, 1.790679931640625, -1.0634307861328125, -32.98927307128906, 14.469879150390625, 20.369735717773438, -2.0219879150390625, 46.534698486328125, 1.982666015625, 20.182373046875, -23.43731689453125, -16.868911743164062, -17.701522827148438, 8.823379516601562, 4.438323974609375, 2.428924560546875, 42.73272705078125, 16.195877075195312, -9.567276000976562, -13.754898071289062, -22.48639678955078, 0.084075927734375, 5.913055419921875, -17.68157958984375, 13.842864990234375, -11.45343017578125, 15.7587890625, -11.187873840332031, 18.59392738342285, 30.570205688476562, 4.14739990234375, -0.10650634765625, 4.24072265625, 17.327255249023438, 94.747802734375, 0.2305755615234375, 25.169219970703125, 26.799468994140625, 21.10693359375, -26.56085205078125, 36.26789855957031, -0.66888427734375, -2.153820037841797, 9.154464721679688, -18.5350341796875, 0.63836669921875, 15.496673583984375, 3.983264923095703, 12.925643920898438, 37.73332214355469, -5.8608856201171875, -1.4975738525390625, 0.0, 35.969818115234375, 14.55419921875, -18.971923828125, 9.080093383789062, 37.31683349609375, 48.8851318359375, -33.206573486328125, -46.0321044921875, 56.664520263671875, 13.17718505859375, 17.657196044921875, 3.836669921875, 11.110107421875, 22.752044677734375, 13.097702026367188, 33.58154296875, 24.69580078125, 53.25288391113281, 89.33267211914062, -15.08514404296875, 27.462539672851562, -0.532470703125, 14.95306396484375, 24.16534423828125, 41.4046630859375, 3.553741455078125, 0.6481342315673828, 10.404510498046875, 52.5699462890625, 6.99395751953125, 7.212921142578125, 0.0, -16.362762451171875, 33.054290771484375, 39.886322021484375, -10.164680480957031, 1.5425872802734375, -8.696807861328125, 10.046905517578125, -1.9569091796875, 54.955657958984375, -27.577987670898438, 44.842803955078125, 33.27012634277344, 27.74853515625, -23.561355590820312, 15.405426025390625, -15.554534912109375, 17.953201293945312, 0.4442138671875, 63.133819580078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000132.npy"}
{"epoch": 0.2764397905759162, "step": 133, "batch_size": 128, "mean": 13.333931922912598, "std": 22.227149963378906, "min": -65.4732666015625, "p10": -12.526146697998044, "median": 11.290933609008789, "p90": 41.48431396484375, "max": 70.5887451171875, "pos_frac": 0.765625, "sample": [20.566497802734375, 1.4418563842773438, -34.767303466796875, 0.0, 1.105010986328125, 32.613433837890625, -0.7735252380371094, 13.67348861694336, 3.991851806640625, 28.305862426757812, 16.364334106445312, 17.57244873046875, 5.609283447265625, 25.137474060058594, 39.088623046875, 50.34173583984375, -10.457275390625, -1.7991943359375, -1.4351654052734375, -14.663009643554688, 33.85209655761719, 49.41583251953125, 70.5887451171875, 5.4490966796875, 25.955810546875, 43.58740234375, 21.764362335205078, 53.17913818359375, 0.0, -0.4597625732421875, 18.058319091796875, 21.847381591796875, -4.369050979614258, 1.893096923828125, 26.611953735351562, 32.92326354980469, 21.650711059570312, 25.300277709960938, -28.127899169921875, 32.61773681640625, 39.195343017578125, 6.178489685058594, -5.9346923828125, -6.066375732421875, 8.982589721679688, 0.475799560546875, -16.0709228515625, 8.321754455566406, 5.430206298828125, 7.81072998046875, 16.874237060546875, 43.302001953125, 2.658203125, 29.142791748046875, 48.48541259765625, 9.085586547851562, 6.318389892578125, 41.594970703125, -4.2723541259765625, 30.122589111328125, -20.961334228515625, 10.521610260009766, 34.458953857421875, 2.1551513671875, 17.527084350585938, 7.1446533203125, 24.720001220703125, 41.4368896484375, 29.617828369140625, 28.625808715820312, -37.7225341796875, 0.9808349609375, 12.060256958007812, 9.659713745117188, 50.531463623046875, 8.235506057739258, -5.787483215332031, -20.654693603515625, 4.4486083984375, 44.15655517578125, -7.127712249755859, 1.469879150390625, 9.230056762695312, -8.28656005859375, 22.2454833984375, -11.670295715332031, 7.802276611328125, 26.694480895996094, 25.1048583984375, 40.975341796875, -4.5526123046875, 47.725311279296875, 28.53778076171875, 13.7637939453125, 14.497314453125, 3.575439453125, 6.987152099609375, 39.1640625, 24.944732666015625, 44.31300354003906, 8.518218994140625, 29.753219604492188, 21.220802307128906, 25.895870208740234, -27.550567626953125, 20.91619110107422, 22.40155029296875, 29.166412353515625, -25.894676208496094, 47.829010009765625, 21.38641357421875, 8.17898941040039, 5.0438232421875, -14.52313232421875, -50.688995361328125, 16.850631713867188, -65.4732666015625, 5.01561164855957, 2.6936492919921875, 36.76812744140625, 5.700225830078125, 25.6768798828125, 36.257694244384766, -4.6168365478515625, -24.96087646484375, 1.9301090240478516, 28.18109130859375, 39.228878021240234], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000133.npy"}
{"epoch": 0.27853403141361255, "step": 134, "batch_size": 128, "mean": 9.307854652404785, "std": 20.924606323242188, "min": -41.151092529296875, "p10": -14.834086608886718, "median": 7.4997711181640625, "p90": 32.76624603271484, "max": 90.82232666015625, "pos_frac": 0.671875, "sample": [38.38177490234375, 26.412933349609375, -16.625457763671875, -7.492340087890625, -25.396072387695312, -14.640274047851562, 8.26556396484375, -16.209335327148438, -5.3087158203125, 25.0028076171875, -41.13671875, -7.028961181640625, 11.3289794921875, 19.193115234375, 27.52886962890625, -0.558685302734375, -5.488779067993164, 8.674015045166016, 22.891937255859375, 7.063762664794922, 10.06439208984375, 61.2896728515625, 0.0, 20.302734375, 21.72698974609375, 21.880584716796875, 30.64471435546875, -3.1652679443359375, 7.22021484375, 44.820709228515625, -29.376935958862305, -6.1316680908203125, 17.521331787109375, 2.808368682861328, 28.614025115966797, 3.339141845703125, 26.380691528320312, 15.82147216796875, 14.7425537109375, 5.39886474609375, -10.727020263671875, 24.956073760986328, 14.301620483398438, 26.96026611328125, 90.82232666015625, 33.424774169921875, 3.455291748046875, -12.312255859375, 67.02105712890625, 28.110626220703125, -21.6806640625, 32.628578186035156, 10.7337646484375, -11.890018463134766, -0.32106781005859375, 1.6340408325195312, 30.935028076171875, 22.668289184570312, 39.24198913574219, 8.389190673828125, -11.728172302246094, -2.7966461181640625, 9.6644287109375, 7.779327392578125, -26.942626953125, 40.44908142089844, -2.9004364013671875, 15.31988525390625, 25.220245361328125, 12.341102600097656, 2.5331764221191406, 16.75408935546875, -17.62240219116211, -4.630271911621094, 0.94342041015625, 41.3629150390625, 11.059219360351562, 24.260528564453125, 7.021205902099609, -33.87226867675781, 18.3668212890625, 32.27214050292969, 27.1024169921875, -11.18927001953125, -1.06982421875, -2.404693603515625, 9.930152893066406, 13.974489212036133, 43.670806884765625, 1.6856422424316406, -0.399017333984375, -5.5637664794921875, 9.278491973876953, 0.019927978515625, -41.14825439453125, 26.776451110839844, 22.006988525390625, 13.8145751953125, 6.9569549560546875, -8.995819091796875, 5.591678619384766, 11.377044677734375, -13.401641845703125, -41.151092529296875, -15.28631591796875, 20.6983642578125, -16.0933837890625, 1.6414794921875, 35.430755615234375, 2.565521240234375, 7.0459442138671875, 31.81277847290039, 0.4350929260253906, 21.660888671875, 38.46484375, 25.848209381103516, 32.94450378417969, 4.13787841796875, 4.471893310546875, -8.320785522460938, -2.874755859375, 1.60980224609375, -3.4447402954101562, -5.806396484375, 32.689849853515625, 18.345945358276367, 4.598197937011719, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000134.npy"}
{"epoch": 0.2806282722513089, "step": 135, "batch_size": 128, "mean": 13.12112045288086, "std": 20.596376419067383, "min": -50.49644470214844, "p10": -9.542979431152343, "median": 10.41619873046875, "p90": 43.43393554687499, "max": 59.997802734375, "pos_frac": 0.78125, "sample": [0.4863090515136719, -26.10784912109375, 11.240856170654297, 21.65289306640625, -9.125686645507812, 41.818328857421875, 0.8860034942626953, 13.54010009765625, 30.756072998046875, 3.3074417114257812, 53.177520751953125, 18.462982177734375, -6.044036865234375, -2.4718170166015625, 11.5360107421875, 6.56109619140625, 9.123855590820312, 25.56414794921875, 25.381103515625, 3.4319686889648438, 2.230377197265625, 7.399284362792969, -0.9503250122070312, 2.8964309692382812, -30.37774658203125, 0.6592845916748047, 33.250640869140625, 25.930931091308594, 18.30541229248047, 45.90081787109375, 34.0208740234375, 0.07696914672851562, 28.54656982421875, 5.031303405761719, -18.236602783203125, 36.10302734375, 4.301727294921875, 16.746612548828125, 27.114837646484375, 9.237625122070312, 4.6418609619140625, 12.999969482421875, 13.928741455078125, 36.930938720703125, 56.393768310546875, 19.059364318847656, 41.431121826171875, 11.005828857421875, 1.24847412109375, -1.95269775390625, 18.72571563720703, 20.120635986328125, 1.7011489868164062, -10.51666259765625, -18.432296752929688, -8.177658081054688, 42.90765380859375, 6.7967529296875, 38.576446533203125, 17.345306396484375, 22.898757934570312, 17.2896728515625, 33.22550964355469, -6.596771240234375, -2.19287109375, 10.963470458984375, -4.217376708984375, 19.92169189453125, 18.656951904296875, 2.816375732421875, 16.872596740722656, 6.471080780029297, 51.787750244140625, 16.701034545898438, 34.499755859375, 30.89153289794922, 15.868728637695312, 7.16448974609375, -5.9754638671875, 49.176177978515625, 20.021942138671875, -15.808258056640625, -1.24151611328125, -0.42352294921875, 7.642425537109375, 49.250396728515625, 0.69140625, 50.91815185546875, 36.605133056640625, 49.35064697265625, 20.8094482421875, 38.9771728515625, 3.25897216796875, -40.00457763671875, 44.66192626953125, 4.601776123046875, 6.695945739746094, 6.375387191772461, 2.00360107421875, 24.86578369140625, 7.971961975097656, 12.475830078125, 29.948745727539062, 59.997802734375, 2.738544464111328, 0.4747161865234375, 4.163753509521484, 15.114471435546875, 46.694488525390625, 56.598876953125, -15.69903564453125, -14.175048828125, 6.17095947265625, 9.917526245117188, 0.0, 21.265594482421875, -19.650634765625, -50.49644470214844, -0.8335418701171875, 10.914871215820312, -5.58245849609375, 20.402755737304688, 5.234100341796875, 46.36419677734375, 28.529876708984375, -12.848396301269531, -18.8670654296875, 1.101776123046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000135.npy"}
{"epoch": 0.28272251308900526, "step": 136, "batch_size": 128, "mean": 12.299373626708984, "std": 22.5756778717041, "min": -54.64068603515625, "p10": -8.70827407836914, "median": 10.205007553100586, "p90": 42.021603393554685, "max": 83.87991333007812, "pos_frac": 0.765625, "sample": [43.04510498046875, 16.842514038085938, 1.928436279296875, 15.795425415039062, 34.98058319091797, 36.36029052734375, 45.716522216796875, 10.83905029296875, 31.220008850097656, 59.155792236328125, 6.375434875488281, 3.15692138671875, 42.781341552734375, 6.207221984863281, 25.81948471069336, 38.24626159667969, 13.338104248046875, 39.15473937988281, 10.111907958984375, 0.304901123046875, -42.62725830078125, 35.91017532348633, 8.04534912109375, 24.331436157226562, -18.208587646484375, 4.737377166748047, -7.5908203125, -42.2071533203125, 6.393218994140625, 13.311477661132812, 4.583282470703125, -0.7414703369140625, -6.157684326171875, 7.400238037109375, 23.66130828857422, -27.596755981445312, -8.334228515625, 27.710060119628906, 13.046272277832031, -3.6712493896484375, 26.10693359375, -19.929367065429688, 58.150146484375, -54.64068603515625, 11.1168212890625, -7.2777557373046875, 3.69000244140625, 4.878692626953125, 15.409393310546875, -4.125740051269531, 13.55902099609375, -34.67425537109375, 42.77583312988281, 15.754638671875, 29.50347900390625, 70.29852294921875, -8.212677001953125, 6.76263427734375, 10.23175048828125, 31.457992553710938, 43.16119384765625, 19.24365234375, 36.841766357421875, -11.562458038330078, 10.262893676757812, 0.1456146240234375, -0.630523681640625, 10.32379150390625, 32.931549072265625, -1.879150390625, 3.6067962646484375, 14.8668212890625, 5.441398620605469, 10.22808837890625, 4.2095947265625, 51.12835693359375, -0.6186561584472656, 15.16986083984375, 10.181926727294922, -8.06787109375, -0.143524169921875, -25.3021240234375, 22.9619140625, 54.612701416015625, 34.478515625, 23.97686767578125, 1.9866600036621094, 17.67254638671875, -6.566545486450195, 42.6031494140625, 16.69573211669922, 10.0093994140625, -9.581047058105469, 14.066970825195312, 4.891510009765625, -0.22100830078125, 55.6060791015625, 14.210594177246094, 27.920089721679688, 22.575836181640625, 0.09459877014160156, 5.697235107421875, 32.63165283203125, 0.082183837890625, 32.19110107421875, -12.871856689453125, -5.6056060791015625, 2.3256683349609375, 32.538055419921875, 41.772369384765625, 35.1812744140625, 9.415695190429688, -20.658416748046875, -7.001678466796875, 6.07745361328125, 9.8509521484375, 83.87991333007812, 5.234718322753906, 1.53515625, 9.639358520507812, 12.009735107421875, 14.2022705078125, 12.116119384765625, 3.099151611328125, 15.77340316772461, -49.54193115234375, 29.129364013671875, 3.8723602294921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000136.npy"}
{"epoch": 0.2848167539267016, "step": 137, "batch_size": 128, "mean": 11.112654685974121, "std": 21.50867462158203, "min": -45.14288330078125, "p10": -15.867064666748046, "median": 8.247573852539062, "p90": 37.476278686523436, "max": 73.496337890625, "pos_frac": 0.7578125, "sample": [-29.6246337890625, 3.0947265625, 37.31689453125, -5.905113220214844, 11.747573852539062, 18.8924560546875, 39.17897033691406, 11.529571533203125, 3.4881134033203125, 6.6109619140625, 19.395057678222656, 15.71728515625, 22.95062255859375, 55.31695556640625, 17.13970947265625, 2.27838134765625, 0.4790496826171875, 30.094970703125, 1.061767578125, -14.382415771484375, 28.832923889160156, 20.95330810546875, 8.5086669921875, 11.65582275390625, -43.6014404296875, 20.725814819335938, 6.122406005859375, -15.717529296875, 28.938194274902344, 7.871978759765625, 48.555450439453125, 15.231231689453125, -2.5022125244140625, 38.50347900390625, -17.490371704101562, -1.8504638671875, 3.640167236328125, 7.6043243408203125, 1.2745361328125, -8.961761474609375, 6.8424072265625, -3.731273651123047, -19.57251739501953, 4.17388916015625, -44.01251220703125, 5.607269287109375, 7.6971435546875, 16.842056274414062, -45.14288330078125, 15.300254821777344, 6.1714630126953125, 43.539398193359375, 6.999359130859375, 11.037841796875, 13.117279052734375, 33.22637939453125, -17.223388671875, 39.5364990234375, -17.643646240234375, 2.7401580810546875, 5.16558837890625, -1.10809326171875, 29.1700439453125, -0.809356689453125, 12.135101318359375, 55.930938720703125, -27.764892578125, 24.25151824951172, 8.12921142578125, -3.04205322265625, 3.207855224609375, 24.711761474609375, 30.916290283203125, 11.113887786865234, 33.92661666870117, 2.190460205078125, 1.2392997741699219, 73.496337890625, 27.6937255859375, 13.188888549804688, 7.451263427734375, 55.822265625, -1.9796371459960938, 34.97627258300781, 19.262725830078125, 16.8858642578125, 55.6185302734375, 19.779647827148438, -5.114158630371094, 32.620391845703125, 1.1986770629882812, -16.215980529785156, -0.4869232177734375, 8.24395751953125, 1.8284492492675781, 63.280120849609375, 25.280517578125, 2.9794769287109375, 19.31194305419922, 15.178482055664062, -35.20074462890625, -30.935882568359375, 8.251190185546875, -30.613540649414062, -8.740852355957031, 52.61564636230469, -14.30660629272461, 29.322113037109375, 37.848175048828125, 32.483177185058594, 23.02130126953125, 33.275848388671875, -0.977447509765625, 20.683197021484375, 13.823455810546875, 22.55685806274414, -2.50372314453125, 6.18145751953125, 0.705810546875, 7.3080902099609375, 7.289787292480469, 19.2943115234375, 7.789398193359375, 13.613800048828125, 15.5631103515625, 21.9219970703125, 28.6431884765625, -6.338951110839844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000137.npy"}
{"epoch": 0.2869109947643979, "step": 138, "batch_size": 128, "mean": 10.971098899841309, "std": 24.069299697875977, "min": -45.6419677734375, "p10": -17.207801818847656, "median": 8.725624084472656, "p90": 41.1763729095459, "max": 82.21319580078125, "pos_frac": 0.671875, "sample": [66.9840087890625, 40.90093994140625, 45.16802978515625, 17.35936737060547, -11.532356262207031, 4.8651123046875, -5.646331787109375, 65.30374145507812, 7.28485107421875, 22.913665771484375, 12.067916870117188, 10.33551025390625, 3.0601348876953125, 11.962814331054688, 11.259212493896484, -30.507522583007812, -17.851715087890625, 30.047332763671875, -9.97369384765625, -28.71826171875, 23.12921142578125, 2.93731689453125, -12.31298828125, -8.16522216796875, -45.6419677734375, 25.078353881835938, -4.288810729980469, 35.00733947753906, 1.9249496459960938, 82.21319580078125, -5.411956787109375, 5.487762451171875, 3.096923828125, 41.81904983520508, 1.1361198425292969, 24.420394897460938, -24.835693359375, 9.115447998046875, 39.61651611328125, -1.2309455871582031, 11.308349609375, -2.035898208618164, 31.021560668945312, 36.028289794921875, 15.78314208984375, 31.728073120117188, 6.7913970947265625, 3.968090057373047, 14.980682373046875, -14.41845703125, 11.198074340820312, 0.081146240234375, 36.491790771484375, -16.931838989257812, 43.107208251953125, 11.55462646484375, 31.51495361328125, -0.7037200927734375, 20.42926025390625, -20.52972412109375, -21.030059814453125, 0.6092071533203125, -1.042938232421875, -1.5439224243164062, 21.619110107421875, -29.766738891601562, 66.70941162109375, 30.591407775878906, 27.892242431640625, -0.584716796875, -25.142120361328125, 17.20257568359375, -6.173736572265625, 53.49932861328125, 14.16412353515625, -7.966644287109375, -4.0470733642578125, 20.1531982421875, 12.738616943359375, 0.960296630859375, 14.008316040039062, 31.3759765625, 27.98504638671875, 33.2843017578125, -40.223175048828125, -11.64874267578125, 12.877349853515625, -0.2121124267578125, -13.08673095703125, -15.191680908203125, 27.833297729492188, 54.97613525390625, 44.27915954589844, 7.6192626953125, -2.476869583129883, 27.1776123046875, 27.297027587890625, 25.420074462890625, 2.96624755859375, -3.795654296875, 28.319091796875, -16.031402587890625, 72.45608520507812, -5.018497467041016, 8.335800170898438, 10.359161376953125, -8.347328186035156, -15.65277099609375, 5.3142242431640625, 38.429176330566406, -18.25372314453125, 3.0220947265625, 2.2477035522460938, 18.02264404296875, 26.162506103515625, 2.0732574462890625, 54.4678955078125, -34.098480224609375, 32.01434326171875, -36.32513427734375, 4.057613372802734, 22.712234497070312, 6.7988128662109375, 20.560577392578125, 48.569091796875, 27.592971801757812, 9.745635986328125, -12.284072875976562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000138.npy"}
{"epoch": 0.28900523560209423, "step": 139, "batch_size": 128, "mean": 8.220113754272461, "std": 23.849695205688477, "min": -46.725128173828125, "p10": -21.458368682861327, "median": 6.309952735900879, "p90": 40.35035400390625, "max": 67.62655639648438, "pos_frac": 0.640625, "sample": [-36.91638946533203, -3.0043182373046875, -30.560089111328125, 6.914276123046875, -29.73406982421875, 14.295394897460938, 30.71411895751953, 19.392578125, -1.9274444580078125, -10.54888916015625, 47.581268310546875, 4.8180389404296875, 21.727996826171875, 10.38262939453125, -27.8336181640625, 45.338165283203125, 51.6048583984375, 10.674057006835938, -12.745025634765625, 8.368545532226562, 5.307609558105469, 24.8970947265625, 30.988296508789062, -14.535354614257812, 12.113372802734375, -11.176849365234375, 6.343660354614258, 34.65702819824219, -5.719673156738281, -46.725128173828125, 0.7387962341308594, -6.8096466064453125, 35.19621276855469, 39.141021728515625, 25.989913940429688, 0.11876392364501953, 12.478744506835938, -10.203689575195312, 3.464691162109375, 29.09881591796875, -3.7643051147460938, -20.70648956298828, 1.57965087890625, 21.380905151367188, 11.217887878417969, -5.989715576171875, -40.419036865234375, -0.498321533203125, -13.134960174560547, 37.679412841796875, -9.11004638671875, 0.578765869140625, -23.212753295898438, -20.11041259765625, 7.82733154296875, 42.58099365234375, 1.40972900390625, 1.000579833984375, -24.95318603515625, -18.0308837890625, 40.04376220703125, -12.188516616821289, 43.599945068359375, 23.730209350585938, 29.136444091796875, 15.84881591796875, 0.29465675354003906, 0.881591796875, 41.06573486328125, 21.87769317626953, -15.234352111816406, 19.777191162109375, 13.29168701171875, 42.19622802734375, -3.780792236328125, -10.442657470703125, 4.964202880859375, 67.62655639648438, -11.71612548828125, 39.78118896484375, 25.454238891601562, 22.200531005859375, 15.262191772460938, 50.86445617675781, 4.823089599609375, -15.26776123046875, 52.1534423828125, -36.0098876953125, 20.836212158203125, -4.473358154296875, 28.63031005859375, 46.404388427734375, 19.1873779296875, 15.52655029296875, -18.216156005859375, 16.060020446777344, -2.48028564453125, -37.21435546875, -5.589141845703125, 2.0511093139648438, 24.930587768554688, 6.2762451171875, -9.459075927734375, 3.930339813232422, 12.891494750976562, -7.86676025390625, 25.193138122558594, 66.02264404296875, 9.518020629882812, 59.2568359375, 39.85028076171875, 18.03009033203125, 4.07855224609375, -33.350860595703125, 8.112846374511719, -32.054779052734375, 0.0, 7.0239410400390625, 13.945274353027344, 32.205379486083984, -8.244277954101562, -33.714599609375, 19.87590789794922, -5.270500183105469, 20.751678466796875, -13.72760009765625, 38.761444091796875, 5.020896911621094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000139.npy"}
{"epoch": 0.29109947643979056, "step": 140, "batch_size": 128, "mean": 14.819308280944824, "std": 22.09014320373535, "min": -41.1376953125, "p10": -9.351691436767577, "median": 10.433700561523438, "p90": 46.64678955078125, "max": 79.1082763671875, "pos_frac": 0.6875, "sample": [-5.47198486328125, 24.1466064453125, 31.97509765625, -7.278778076171875, -9.628349304199219, 47.8355712890625, -5.369956970214844, 39.56224060058594, -1.3979644775390625, -0.2867279052734375, -15.438735961914062, 9.2518310546875, 30.19708251953125, 18.31859588623047, 6.472526550292969, 7.86151123046875, 3.4227294921875, 21.01202392578125, 46.0709228515625, 11.19219970703125, 8.10064697265625, -17.334945678710938, 28.17108917236328, -23.36846923828125, 21.121673583984375, 38.86578369140625, -2.559051513671875, 6.89752197265625, 42.59828186035156, 42.25634002685547, 29.52532958984375, 32.13804626464844, 63.9141845703125, 38.1326904296875, -9.233123779296875, 57.10955810546875, -5.372772216796875, 3.4901123046875, 47.50526428222656, 47.43695068359375, -2.1950912475585938, -26.9168701171875, 11.703659057617188, 0.0483856201171875, -2.23541259765625, 35.0933837890625, 48.754058837890625, 10.490081787109375, -6.78790283203125, 79.1082763671875, 9.67755126953125, 23.801467895507812, 30.230850219726562, 10.606376647949219, 3.160125732421875, 21.643280029296875, 20.64385986328125, 5.330488204956055, -8.3206787109375, 22.56804656982422, 26.055267333984375, 0.7819976806640625, 2.94781494140625, 21.738677978515625, 6.1640777587890625, -0.5628662109375, -3.9395484924316406, -12.592529296875, 0.0, -41.1376953125, -1.18994140625, -29.454833984375, 65.35737609863281, -0.12465667724609375, 5.075477600097656, 14.08673095703125, 23.028533935546875, 31.50323486328125, 3.05694580078125, 47.57838439941406, 10.053459167480469, 4.9373779296875, -0.3621540069580078, 42.132415771484375, 50.91802978515625, 34.700462341308594, 30.03155517578125, -2.6880626678466797, 33.626800537109375, 37.713348388671875, 5.961416244506836, 8.249549865722656, 19.241012573242188, -2.6411895751953125, 27.792152404785156, 30.568832397460938, 0.0, -6.797334671020508, 19.4014892578125, -1.2960128784179688, -8.066650390625, 47.16632080078125, -15.49896240234375, 46.42413330078125, 8.839462280273438, -18.086456298828125, 42.43939208984375, 21.377120971679688, 20.712890625, 22.587074279785156, 44.719970703125, 20.9906005859375, 50.046142578125, 19.8221435546875, 27.032806396484375, -6.487548828125, -1.049591064453125, -13.9246826171875, 8.406341552734375, -21.6072998046875, 10.3773193359375, 53.9879150390625, 11.6854248046875, -2.7173080444335938, 8.634613037109375, -16.66632080078125, 25.080047607421875, 22.485382080078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000140.npy"}
{"epoch": 0.2931937172774869, "step": 141, "batch_size": 128, "mean": 14.077037811279297, "std": 24.861181259155273, "min": -74.4921875, "p10": -11.049827575683594, "median": 12.408161163330078, "p90": 46.056359863281244, "max": 88.188232421875, "pos_frac": 0.7265625, "sample": [39.783843994140625, 16.085205078125, -24.44003677368164, -8.318405151367188, 6.91680908203125, 14.53900146484375, -7.45440673828125, 22.304031372070312, -50.8857421875, 88.188232421875, 5.893798828125, 9.382843017578125, 30.109397888183594, 32.25628662109375, 15.047073364257812, 41.29229736328125, 24.159805297851562, 19.106521606445312, 15.708072662353516, 21.878021240234375, 57.587646484375, -5.7315673828125, -0.13702392578125, 44.2001953125, -0.2151355743408203, -14.279624938964844, 38.12083435058594, 12.538276672363281, 1.4361114501953125, 48.088470458984375, 32.07666015625, -16.5594482421875, -9.722808837890625, 67.3701171875, 2.8184814453125, 17.512847900390625, -25.32159423828125, -7.1471405029296875, 10.970550537109375, -17.775299072265625, 41.490966796875, 35.40155029296875, 12.771341323852539, 20.462738037109375, -6.599189758300781, 8.37744140625, 1.783721923828125, -7.319404602050781, 58.400146484375, 8.407760620117188, 11.021697998046875, 22.69561767578125, 3.232542037963867, -11.146652221679688, 29.525115966796875, 21.603851318359375, 1.6697006225585938, 1.4516067504882812, 24.688720703125, 1.338531494140625, 33.402435302734375, 33.37223815917969, -15.33697509765625, -42.57936096191406, -3.948089599609375, 67.41415405273438, -2.036680221557617, 25.155235290527344, -5.8856048583984375, 36.86338806152344, 5.4212799072265625, 13.325325012207031, 7.094326019287109, 23.15904998779297, 29.732940673828125, 35.874671936035156, 32.4969482421875, -6.719512939453125, 0.0, -13.777212142944336, 41.69132995605469, -2.76165771484375, -0.973876953125, 60.1390380859375, 5.027824401855469, 52.08197021484375, -74.4921875, 17.21875, 19.538970947265625, -0.868621826171875, 21.723403930664062, 19.29711151123047, 34.21067810058594, -43.93870544433594, -5.4426727294921875, 17.628677368164062, 1.1494522094726562, 54.79078674316406, -0.03485107421875, 31.358657836914062, -6.902984619140625, 10.65643310546875, 19.33209228515625, 39.4771728515625, 15.42474365234375, 4.425355911254883, 23.17596435546875, 45.185455322265625, 42.38511657714844, -11.008331298828125, 7.845020294189453, 11.70159912109375, 54.5848388671875, -35.16450500488281, 0.9041061401367188, 9.234527587890625, 10.1810302734375, 12.278045654296875, 19.60931396484375, 20.253402709960938, -7.9327239990234375, 48.277313232421875, 49.26275634765625, 6.768341064453125, 11.95438003540039, 54.84600830078125, 14.8375244140625, 3.2551345825195312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000141.npy"}
{"epoch": 0.29528795811518327, "step": 142, "batch_size": 128, "mean": 13.686077117919922, "std": 24.706485748291016, "min": -64.32904052734375, "p10": -14.481500244140618, "median": 9.651115417480469, "p90": 45.95123138427734, "max": 77.12741088867188, "pos_frac": 0.6875, "sample": [-4.15423583984375, -2.8809280395507812, 8.903854370117188, 28.292449951171875, 2.0515480041503906, -20.01416015625, 10.39837646484375, -0.275543212890625, 71.18045043945312, 38.034332275390625, 43.56793212890625, -3.7566375732421875, -7.125274658203125, 37.138214111328125, 17.2716064453125, 28.06219482421875, 0.30968475341796875, -3.057159423828125, 52.0841064453125, 24.51318359375, 0.22967529296875, 23.987159729003906, 0.0, 20.35259246826172, 21.06890869140625, 45.751495361328125, 11.217254638671875, 4.5314483642578125, -2.5226516723632812, 0.337738037109375, -1.743011474609375, 26.376861572265625, -2.5024642944335938, -22.746673583984375, 2.1394500732421875, 2.2691917419433594, 24.947265625, 14.496009826660156, -23.557952880859375, 26.259628295898438, 21.8675537109375, 47.17987060546875, 43.15850830078125, 17.831512451171875, 17.63043212890625, -9.44378662109375, -0.9826431274414062, 44.169708251953125, 18.97113037109375, 67.23934936523438, 14.3697509765625, -6.307781219482422, 4.1330718994140625, 13.269386291503906, -12.38653564453125, 2.787046432495117, -0.3046722412109375, 32.62847137451172, -19.77960205078125, -19.3697509765625, 22.08282470703125, 31.61077880859375, -9.077194213867188, -8.01702880859375, 46.41728210449219, 36.938133239746094, 14.95068359375, 0.5244331359863281, 44.766807556152344, -3.3104248046875, 7.231536865234375, 20.586822509765625, 2.2734832763671875, 5.46046257019043, -23.0501708984375, 30.515457153320312, -38.151580810546875, 27.653396606445312, 41.423736572265625, -24.162689208984375, -5.0098876953125, 67.2623291015625, 77.12741088867188, -2.09759521484375, -2.0284194946289062, 11.891319274902344, 10.405105590820312, 49.0711669921875, 8.364921569824219, 5.6376953125, -2.5374927520751953, -3.3776779174804688, 31.63470458984375, 40.730987548828125, 56.8565673828125, 7.603271484375, 34.365264892578125, 24.863113403320312, 20.455780029296875, 52.12115478515625, -64.32904052734375, 53.5277099609375, -31.925689697265625, 0.8470020294189453, -0.46721649169921875, -26.002777099609375, 34.783470153808594, 16.490867614746094, 31.397247314453125, 2.2241668701171875, 36.716461181640625, -32.93609619140625, 2.3299560546875, 5.752777099609375, 18.40843963623047, -8.013916015625, 28.57257080078125, -1.6858634948730469, 4.700714111328125, 1.54736328125, -22.434341430664062, 61.77777099609375, 0.16149520874023438, -2.2051544189453125, 30.5355224609375, 28.01404571533203, 65.02845764160156, 40.89861297607422], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000142.npy"}
{"epoch": 0.2973821989528796, "step": 143, "batch_size": 128, "mean": 15.79372501373291, "std": 26.572114944458008, "min": -77.0760498046875, "p10": -9.860083007812499, "median": 13.205524444580078, "p90": 46.60120544433594, "max": 97.7799072265625, "pos_frac": 0.765625, "sample": [23.141963958740234, 46.7091064453125, 55.30804443359375, 29.8956298828125, 38.087738037109375, 17.60260009765625, 29.050765991210938, 9.851051330566406, -13.822616577148438, -7.060762405395508, 27.06890869140625, 0.10725021362304688, 21.71234130859375, 12.325469970703125, 2.3725662231445312, -10.78643798828125, 46.554962158203125, 5.81463623046875, -17.033462524414062, 97.7799072265625, -18.3502197265625, -4.599517822265625, 43.62391662597656, 45.3638916015625, 20.183494567871094, 42.858917236328125, 35.138763427734375, 58.75628662109375, 86.51385498046875, 2.2923583984375, -3.28778076171875, 42.20171356201172, 28.40093994140625, 31.533130645751953, 24.468544006347656, -3.19903564453125, -77.0760498046875, -2.4530868530273438, 28.098068237304688, 8.7640380859375, 8.672737121582031, -29.195404052734375, 6.5098114013671875, 40.220458984375, 3.9538650512695312, 5.1021728515625, 10.108699798583984, 21.31292724609375, 15.29439926147461, -63.23931884765625, 6.868408203125, 27.349365234375, 2.8265380859375, 38.30506896972656, 18.3603515625, 21.1976318359375, 0.010009765625, 21.019208908081055, -15.36480712890625, 0.7365646362304688, -9.46307373046875, 0.36991310119628906, 65.92864990234375, 17.882293701171875, 4.9546966552734375, -2.1812973022460938, 56.95172119140625, 31.407562255859375, 35.37225341796875, 3.903717041015625, 29.305572509765625, 33.70283508300781, -5.440982818603516, 3.781557083129883, 32.30663299560547, -8.4305419921875, -36.354736328125, 0.37946319580078125, 12.961257934570312, 13.449790954589844, 4.2111053466796875, -1.57586669921875, 7.191184997558594, 16.57354736328125, 9.8076171875, -2.867279052734375, -16.1793212890625, 1.92645263671875, -1.4820785522460938, 14.896697998046875, 34.03081130981445, 13.889556884765625, -3.135772705078125, 21.455459594726562, 19.985015869140625, 26.680511474609375, 0.5855712890625, 16.619110107421875, -3.712453842163086, 3.6555633544921875, 80.87054443359375, 30.538726806640625, -5.38763427734375, 11.252494812011719, -29.552398681640625, 73.11865234375, 30.116546630859375, 3.480764389038086, 4.369731903076172, -2.1409149169921875, 26.46728515625, 56.173248291015625, 28.055946350097656, 66.38021850585938, 26.67559814453125, 60.94036865234375, 41.591278076171875, 42.65350341796875, -38.02947998046875, 24.770702362060547, -16.127899169921875, 50.1463623046875, -7.4227294921875, 9.37921142578125, 43.939910888671875, 6.40478515625, 14.37249755859375, 1.2562141418457031], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000143.npy"}
{"epoch": 0.2994764397905759, "step": 144, "batch_size": 128, "mean": 12.768148422241211, "std": 24.1916561126709, "min": -64.03504943847656, "p10": -12.653567504882812, "median": 6.9107208251953125, "p90": 44.22145538330078, "max": 87.95111083984375, "pos_frac": 0.734375, "sample": [-0.5674953460693359, 17.881736755371094, 41.0457763671875, 4.656257629394531, 4.7047119140625, 11.10430908203125, -21.087181091308594, -9.800628662109375, 1.3566913604736328, 87.95111083984375, 2.1566009521484375, -9.27093505859375, 6.925651550292969, 54.84564208984375, 4.6815185546875, 33.421173095703125, 0.7276458740234375, 44.467132568359375, 9.090568542480469, 0.144195556640625, 1.204376220703125, 18.16217041015625, -12.808059692382812, 3.2043609619140625, 2.6707763671875, -4.899993896484375, -21.179771423339844, -2.485851287841797, 26.319915771484375, 4.832284927368164, -36.7303466796875, 5.481227874755859, -1.342193603515625, 24.1873779296875, 49.7056884765625, 37.190185546875, 29.45037841796875, 70.6956787109375, 7.224479675292969, 55.44708251953125, -0.9563808441162109, -10.549209594726562, 7.133934020996094, 21.52325439453125, -0.6114082336425781, -17.695632934570312, 7.830108642578125, 33.329833984375, -16.244171142578125, 22.150909423828125, 12.769371032714844, 53.302642822265625, 36.04618835449219, -17.4349365234375, -64.03504943847656, -25.8564453125, 13.767974853515625, -11.902137756347656, -11.20654296875, -0.3214263916015625, 32.77349853515625, 1.8223876953125, -7.755767822265625, 6.415740966796875, 27.39971160888672, 12.76611328125, 67.1292724609375, 60.527587890625, 38.89508056640625, 36.5576171875, 1.5722122192382812, 18.25487518310547, 46.63079833984375, -27.681915283203125, 21.044036865234375, 24.478302001953125, 2.91412353515625, 8.162696838378906, 44.11616516113281, 38.402191162109375, 41.9520263671875, 36.81856155395508, 3.5153350830078125, -7.6530914306640625, 14.046615600585938, 24.714946746826172, -2.045166015625, 4.114837646484375, 1.5952033996582031, 30.41778564453125, 67.69879150390625, 32.07568359375, 12.420936584472656, 0.31890869140625, 10.788986206054688, 41.728759765625, 2.9392471313476562, -3.306304931640625, 6.2255706787109375, 35.43524169921875, 6.904815673828125, 0.08981132507324219, 0.0, -0.717010498046875, -12.587356567382812, -11.621002197265625, 3.89007568359375, -31.825836181640625, 24.4100341796875, 25.26490020751953, 18.644439697265625, 0.259063720703125, 40.70806884765625, -6.496406555175781, 23.005645751953125, 1.76153564453125, 17.852691650390625, 5.472930908203125, 16.92132568359375, -24.57611083984375, 12.517501831054688, 34.665191650390625, 6.9166259765625, 45.62762451171875, -16.535308837890625, 0.1217498779296875, 68.9105224609375, 4.7048492431640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000144.npy"}
{"epoch": 0.30157068062827225, "step": 145, "batch_size": 128, "mean": 16.63412094116211, "std": 25.11893653869629, "min": -52.896087646484375, "p10": -12.820843505859376, "median": 14.652206420898438, "p90": 49.434985351562496, "max": 81.2923583984375, "pos_frac": 0.7109375, "sample": [-12.77783203125, 36.91755676269531, -39.93853759765625, 33.90338134765625, 13.11018180847168, -27.552093505859375, 37.95037841796875, 17.8231201171875, 76.350830078125, 24.973556518554688, 22.01422119140625, 10.839569091796875, 23.307876586914062, 32.510345458984375, 29.58722686767578, 61.84552001953125, 3.2337646484375, 30.18231201171875, 7.6458587646484375, 6.254486083984375, 0.8687667846679688, -7.327178955078125, 50.01416015625, -5.652683258056641, -7.1587982177734375, 65.2451171875, -26.073883056640625, 49.186767578125, 0.6948394775390625, 17.0966796875, 3.1544189453125, 30.106307983398438, 8.241874694824219, 34.32122802734375, 54.113189697265625, 20.91168212890625, 70.97970581054688, -52.896087646484375, 8.453643798828125, -0.721649169921875, 16.571945190429688, 18.8248291015625, -13.102733612060547, 61.928985595703125, 65.80416870117188, 14.4888916015625, 36.147369384765625, 32.059234619140625, 11.932723999023438, 18.6934814453125, 17.638408660888672, -6.246742248535156, -6.293251037597656, 31.684341430664062, 45.76588439941406, 40.941680908203125, 0.493377685546875, -4.767547607421875, -14.656051635742188, 3.381254196166992, 14.815521240234375, 5.7228546142578125, -8.217864990234375, 41.59503173828125, 61.69854736328125, -5.244781494140625, 54.50494384765625, -9.726882934570312, 53.108123779296875, 29.68180274963379, 47.08587646484375, -10.427444458007812, 5.996757507324219, 32.74053955078125, 42.6900634765625, 11.341461181640625, 1.5916481018066406, 9.633499145507812, 26.130020141601562, -21.576629638671875, -17.845008850097656, 6.305633544921875, 9.053474426269531, 29.730804443359375, 81.2923583984375, 18.98261260986328, 4.920265197753906, 27.620025634765625, -24.532028198242188, -8.660240173339844, -2.6417388916015625, 0.0, 7.892112731933594, 19.834228515625, 9.615005493164062, 3.4629058837890625, 38.79833984375, 68.95843505859375, 39.262237548828125, 24.314117431640625, -2.3666534423828125, -9.591384887695312, 26.1068115234375, 22.882553100585938, 13.112945556640625, -12.92120361328125, 24.87908935546875, 28.02508544921875, -2.7169342041015625, -0.9489212036132812, -15.211700439453125, 48.951263427734375, 25.917396545410156, 44.115966796875, -3.1090774536132812, -1.378204345703125, 3.4053802490234375, 28.045324325561523, 41.546722412109375, 49.17303466796875, 29.268524169921875, -14.779052734375, -1.8689632415771484, -3.1598968505859375, 26.57586669921875, -16.733642578125, -8.2646484375, 17.67486572265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000145.npy"}
{"epoch": 0.3036649214659686, "step": 146, "batch_size": 128, "mean": 12.816333770751953, "std": 23.92302131652832, "min": -58.219329833984375, "p10": -14.926327514648436, "median": 7.490669250488281, "p90": 43.92696075439453, "max": 74.7652587890625, "pos_frac": 0.734375, "sample": [4.778168678283691, 2.7231979370117188, 18.582027435302734, -30.2828369140625, 12.401763916015625, 37.922210693359375, -23.328933715820312, 8.930953979492188, 32.22705841064453, -21.881195068359375, 74.7652587890625, 69.77291870117188, 9.6517333984375, -8.44342041015625, 17.09979248046875, 2.4593963623046875, 41.003570556640625, 21.52117919921875, 2.7042694091796875, 2.751495361328125, 28.263381958007812, 1.3758544921875, 11.497119903564453, 6.2577056884765625, -18.729522705078125, -8.848724365234375, 3.334543228149414, 20.137435913085938, 17.34667205810547, 19.19744873046875, 11.312355041503906, 0.7871627807617188, 38.20355224609375, 2.090728759765625, 3.26458740234375, 17.224349975585938, 15.430618286132812, -21.28668975830078, 10.780433654785156, 3.6551666259765625, 7.6193389892578125, 45.59010314941406, 0.8925132751464844, 34.1580810546875, 0.0, 44.80250549316406, -4.717041015625, -1.5840892791748047, 27.478012084960938, 22.3692626953125, 5.090301513671875, -15.0802001953125, 6.989992141723633, -2.45391845703125, 49.745849609375, 59.297821044921875, -58.219329833984375, 28.523643493652344, 5.410026550292969, 33.52803039550781, 33.24005126953125, 4.6611328125, 3.6300888061523438, 45.43341064453125, 33.339599609375, -12.621780395507812, -7.8172607421875, -11.682464599609375, 6.471466064453125, -17.311874389648438, 49.181884765625, -17.908401489257812, 2.79266357421875, 36.08270263671875, -21.5484619140625, 59.98907470703125, 7.7994384765625, 7.9753570556640625, -9.816192626953125, 67.09503173828125, 3.735931396484375, 30.618865966796875, 2.082307815551758, -10.051025390625, 32.66302490234375, -0.26116180419921875, -6.7321624755859375, 3.4085540771484375, 0.0, 34.754051208496094, -5.8612823486328125, -5.584503173828125, 19.83544921875, -22.73486328125, 1.7318115234375, 57.46070861816406, 19.447479248046875, 37.36952209472656, 6.549690246582031, 48.33067321777344, -20.57666015625, -14.860382080078125, 0.794769287109375, 9.98529052734375, 36.96124267578125, -7.206451416015625, 34.251708984375, 7.36199951171875, 47.005706787109375, 42.058441162109375, 43.551727294921875, -56.5938720703125, 26.06930160522461, 39.697959899902344, 16.18487548828125, 32.78814697265625, -12.33028793334961, 41.06732177734375, 1.373138427734375, 33.14677429199219, 0.42266845703125, 5.3315277099609375, 22.206817626953125, 41.289642333984375, 27.637420654296875, -11.944442749023438, 29.53515625, -6.56103515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000146.npy"}
{"epoch": 0.3057591623036649, "step": 147, "batch_size": 128, "mean": 11.028444290161133, "std": 26.96625328063965, "min": -70.8204345703125, "p10": -20.071949768066407, "median": 7.481380462646484, "p90": 44.306671142578125, "max": 87.2232666015625, "pos_frac": 0.6328125, "sample": [-0.0689697265625, -1.993988037109375, 6.029178619384766, 33.5574951171875, -7.604881286621094, -2.7118988037109375, 81.31625366210938, -20.075485229492188, 0.0, 66.78271484375, -17.894683837890625, 22.308792114257812, 14.839111328125, 10.950714111328125, 24.082918167114258, 4.338401794433594, -20.0704345703125, 39.806243896484375, 53.6068115234375, 3.4214134216308594, -3.689788818359375, 19.842559814453125, -0.6637802124023438, 36.00878143310547, -1.307891845703125, 47.94148254394531, 0.9270057678222656, 41.43962097167969, 10.257110595703125, -0.73431396484375, 62.884307861328125, 34.727203369140625, 16.270050048828125, 21.839141845703125, 15.0771484375, -27.008758544921875, -39.28553771972656, -6.283935546875, -16.8408203125, 17.321807861328125, 11.38232421875, 28.95235824584961, 30.072811126708984, 5.1255950927734375, -23.643108367919922, 27.4947509765625, -27.36474609375, -1.9979724884033203, -0.8805122375488281, -1.5875396728515625, -5.62811279296875, 49.195159912109375, 10.097808837890625, 11.166885375976562, 26.22565460205078, 30.977462768554688, 77.4114990234375, 10.217094421386719, 4.5732574462890625, -4.0612945556640625, 19.1356201171875, 17.847412109375, 54.60255432128906, 5.416019439697266, 44.923187255859375, 1.395233154296875, -0.11053466796875, 55.87054443359375, 37.38764572143555, 2.9859771728515625, -41.42669677734375, 11.197296142578125, 2.129119873046875, 1.026092529296875, 29.67999267578125, 5.4700775146484375, 11.003618240356445, 4.2724151611328125, -0.09383773803710938, -0.26184844970703125, 26.572372436523438, -9.389823913574219, -1.262115478515625, 33.36231231689453, 14.321006774902344, 37.384429931640625, 7.72747802734375, 33.28857421875, -11.455169677734375, 28.87261199951172, 13.600677490234375, 40.7781982421875, -26.98309326171875, -70.8204345703125, -35.08784484863281, -2.0327224731445312, 10.663330078125, 24.491119384765625, -15.465896606445312, -40.66627502441406, -8.527481079101562, 3.7375946044921875, 44.042449951171875, 14.636962890625, -17.2113037109375, -7.3680572509765625, 42.13807678222656, -2.987060546875, -37.7635498046875, -15.431396484375, -62.321144104003906, 6.047271728515625, 7.8172454833984375, 35.644386291503906, 48.04115295410156, 35.15272521972656, 6.925010681152344, 50.650390625, -2.126251220703125, -5.44012451171875, 7.235282897949219, 40.94410705566406, 43.284088134765625, 19.3177490234375, 87.2232666015625, 9.827423095703125, -4.217525482177734, -25.049591064453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000147.npy"}
{"epoch": 0.3078534031413613, "step": 148, "batch_size": 128, "mean": 11.602202415466309, "std": 24.507627487182617, "min": -54.72877502441406, "p10": -20.404626464843748, "median": 10.540946960449219, "p90": 44.797750854492186, "max": 87.13259887695312, "pos_frac": 0.6484375, "sample": [10.3641357421875, 11.754058837890625, 16.46605682373047, -14.2401123046875, -1.6265010833740234, 66.41943359375, -0.5243930816650391, 18.7049560546875, 71.366943359375, 20.336761474609375, 5.158788681030273, 20.513961791992188, -4.8253631591796875, 52.04193115234375, 19.744873046875, 35.31547546386719, 9.977294921875, -21.56170654296875, 18.478607177734375, 20.41244888305664, -21.0009765625, 39.78765869140625, -44.808563232421875, 0.8749961853027344, 7.046844482421875, -21.84814453125, 14.048812866210938, -5.959251403808594, 16.595367431640625, -8.217519760131836, 16.354347229003906, 16.559097290039062, -7.506416320800781, -54.72877502441406, -1.60296630859375, -12.82025146484375, 57.886749267578125, 6.0876007080078125, -20.1490478515625, 19.87310791015625, 54.50103759765625, 1.4572086334228516, -21.7716064453125, -22.736083984375, 1.55877685546875, 5.29301643371582, 1.6880035400390625, -14.0911865234375, 16.36199951171875, 22.63201904296875, 5.344158172607422, 7.2974700927734375, 6.308799743652344, 23.2266845703125, -21.040145874023438, 55.500701904296875, 9.024826049804688, 36.402801513671875, 10.727546691894531, 43.26873779296875, 22.164215087890625, -0.2462615966796875, -0.1944580078125, -1.336761474609375, 0.0, 7.962207794189453, 47.46929931640625, 16.95294189453125, 34.35230255126953, 36.55828857421875, -3.152080535888672, -23.277305603027344, -7.2200775146484375, -28.865768432617188, 31.353561401367188, 10.717758178710938, -20.057838439941406, 30.84747314453125, -0.8029556274414062, 27.466995239257812, -2.434713363647461, 17.266738891601562, -21.982879638671875, 18.9913330078125, 52.58198547363281, 10.294349670410156, 1.3917579650878906, 16.28558349609375, -8.877899169921875, -15.22406005859375, 23.22112274169922, 23.94293975830078, -0.5133037567138672, 22.136672973632812, 19.126785278320312, 44.50177001953125, -19.950027465820312, 11.756729125976562, -10.655952453613281, 59.07353210449219, 2.7406463623046875, 22.329086303710938, 87.13259887695312, 34.118682861328125, 1.9844818115234375, -5.546480178833008, 23.591705322265625, -25.2476806640625, 30.929656982421875, 25.99498748779297, 45.488372802734375, 42.772300720214844, 16.809967041015625, 26.95770263671875, -32.832420349121094, -4.095161437988281, -18.33642578125, 16.141876220703125, 58.766326904296875, 0.0, -8.59075927734375, 12.23101806640625, 32.41785430908203, -0.9891510009765625, 13.861419677734375, 42.292449951171875, 54.36167907714844, -3.531890869140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000148.npy"}
{"epoch": 0.3099476439790576, "step": 149, "batch_size": 128, "mean": 14.148704528808594, "std": 23.64820671081543, "min": -39.07905578613281, "p10": -13.399522399902342, "median": 11.542678833007812, "p90": 40.87962341308593, "max": 92.21142578125, "pos_frac": 0.765625, "sample": [43.705902099609375, -14.27801513671875, 22.93109130859375, 32.21613311767578, 38.920257568359375, 41.955322265625, 12.167861938476562, 4.836524963378906, 7.16485595703125, 5.130157470703125, 30.087158203125, 2.956817626953125, 40.418609619140625, 20.22856903076172, 19.8289794921875, -3.50811767578125, 43.64080810546875, 8.653778076171875, -1.925018310546875, -9.915054321289062, 13.610359191894531, 4.5761260986328125, 17.98809814453125, 3.8165206909179688, 15.366077423095703, 9.467315673828125, -36.07176208496094, 25.537368774414062, 45.79351806640625, 11.403663635253906, 11.435333251953125, 17.673736572265625, 22.709869384765625, 15.10809326171875, 1.7176227569580078, 0.2648773193359375, 38.75575256347656, 1.7431926727294922, -13.023025512695312, 1.3658599853515625, 31.666732788085938, 70.74774169921875, -3.7425003051757812, 1.298553466796875, 73.14915466308594, 38.88238525390625, 10.064788818359375, -26.792510986328125, 36.152740478515625, 49.791473388671875, 19.492809295654297, -2.725830078125, 12.148193359375, 16.396484375, 7.5076751708984375, 74.0506591796875, -28.1346435546875, 14.11370849609375, 13.891555786132812, 7.8341064453125, -10.342529296875, 13.188972473144531, -20.177978515625, 30.842010498046875, -23.751312255859375, 2.361358642578125, 8.80902099609375, -2.2646484375, 20.942855834960938, 19.839746475219727, 37.365081787109375, 22.980545043945312, 21.396516799926758, 13.233734130859375, 33.462738037109375, 13.219573974609375, 36.951904296875, 8.087760925292969, 52.9683837890625, 0.00806427001953125, 39.30633544921875, 0.5209236145019531, 29.93140411376953, 25.521240234375, 7.928497314453125, -9.8660888671875, 52.8309326171875, 27.614471435546875, -35.720367431640625, -2.1375885009765625, 11.313949584960938, 26.100738525390625, -0.732269287109375, -20.932373046875, -2.01385498046875, -4.501373291015625, 59.4622802734375, 4.60247802734375, 33.81575012207031, 7.3214263916015625, -1.3997344970703125, 4.6681671142578125, 30.41138458251953, 0.4417877197265625, 26.140579223632812, 10.40655517578125, -24.6123046875, -21.550201416015625, 0.4433135986328125, 92.21142578125, -3.037567138671875, 6.556182861328125, 10.005859375, 24.99884033203125, 4.645782470703125, -10.52984619140625, 37.361778259277344, 32.51860046386719, -39.07905578613281, -33.52728271484375, 20.401046752929688, 36.58355712890625, -22.850372314453125, 69.44320678710938, 38.609474182128906, 13.256942749023438, 11.6500244140625, -2.8707275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000149.npy"}
{"epoch": 0.31204188481675393, "step": 150, "batch_size": 128, "mean": 13.607524871826172, "std": 23.249595642089844, "min": -45.9635009765625, "p10": -10.607862854003907, "median": 10.082033157348633, "p90": 40.85015411376953, "max": 74.19580078125, "pos_frac": 0.7421875, "sample": [21.48614501953125, 23.756088256835938, -3.142375946044922, 40.6241455078125, 13.736587524414062, -18.875946044921875, -7.9407501220703125, 24.279586791992188, -30.19036865234375, 40.18975830078125, 22.809707641601562, 34.181610107421875, 39.6431884765625, 15.37298583984375, -12.447509765625, 4.3697509765625, 6.622581481933594, 37.227386474609375, 7.12890625, 31.654998779296875, 22.655303955078125, -7.5153045654296875, -1.114349365234375, 3.4629364013671875, -5.424137115478516, 10.20550537109375, 9.071685791015625, 0.7238922119140625, 60.63465881347656, 24.779258728027344, 64.61627197265625, 1.5816802978515625, 12.141357421875, 7.560083389282227, 16.181503295898438, 24.551193237304688, 19.6219482421875, -43.42803955078125, 36.636470794677734, 12.1446533203125, 16.910186767578125, 41.20497131347656, 12.64080810546875, 36.666839599609375, -5.2279052734375, -6.683052062988281, 1.5369338989257812, 5.17327880859375, 10.96307373046875, 57.022216796875, 7.386438369750977, 71.00469970703125, 9.428619384765625, 30.724761962890625, 39.74029541015625, 48.79144287109375, 36.591156005859375, 5.949695587158203, 0.13037109375, 1.7635917663574219, 1.1942138671875, -0.9267597198486328, -20.521987915039062, 42.02058410644531, -4.25054931640625, -2.120208740234375, -0.5091094970703125, 1.2815399169921875, 39.813575744628906, 44.83563232421875, 19.23675537109375, 40.5938720703125, -4.33935546875, 36.89982604980469, 13.971948623657227, -10.523025512695312, -36.8607177734375, 8.270538330078125, 2.631124496459961, 22.09832000732422, 9.958560943603516, -35.8043212890625, -5.8089141845703125, -25.477951049804688, -4.84918212890625, 15.53448486328125, -19.42706298828125, 74.19580078125, 34.291900634765625, 8.497161865234375, 11.36614990234375, -10.805816650390625, 5.860931396484375, 28.708053588867188, -45.9635009765625, 7.8986663818359375, 35.58586120605469, -1.9056396484375, -17.002532958984375, 1.0868587493896484, 21.655349731445312, 37.92760467529297, 16.155792236328125, 45.800933837890625, 7.640192031860352, 1.06304931640625, 37.77178955078125, 15.28680419921875, 40.698089599609375, 29.656951904296875, -1.16351318359375, 9.446380615234375, 2.3989334106445312, 52.412384033203125, 14.197372436523438, 1.1447601318359375, 25.113426208496094, -10.245162963867188, 50.517333984375, -5.75390625, 0.0, 22.6544189453125, 70.979736328125, 0.854522705078125, 14.87234878540039, 8.587844848632812, 21.52422332763672, -29.461631774902344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000150.npy"}
{"epoch": 0.31413612565445026, "step": 151, "batch_size": 128, "mean": 19.177261352539062, "std": 25.917072296142578, "min": -47.4111328125, "p10": -13.076171112060546, "median": 20.939821243286133, "p90": 50.34688262939453, "max": 83.92178344726562, "pos_frac": 0.75, "sample": [24.286788940429688, 22.330413818359375, 0.7191581726074219, -20.595802307128906, 0.3558349609375, 69.22393798828125, 48.784271240234375, -19.830795288085938, 3.8168487548828125, 35.5469970703125, -2.162109375, -6.857810974121094, 28.865676879882812, 34.92327880859375, -7.2101898193359375, -12.537666320800781, -12.892074584960938, 27.982192993164062, 36.288665771484375, 26.164833068847656, -47.4111328125, 2.7472991943359375, -5.40765380859375, 7.190345764160156, 2.0159912109375, 28.8311767578125, -3.420806884765625, -4.0232086181640625, 33.5435791015625, 45.063934326171875, 50.544281005859375, -0.25115966796875, -3.458465576171875, 15.196968078613281, 28.103275299072266, 20.457992553710938, 12.955230712890625, 13.711082458496094, -4.686576843261719, 73.59970092773438, 19.990081787109375, 29.292434692382812, 12.52886962890625, 15.37451171875, 42.09912109375, -32.32586669921875, 29.874237060546875, 9.6151123046875, 41.44581604003906, 42.54937744140625, 40.217506408691406, -8.462646484375, 1.4117794036865234, 0.6652908325195312, 28.244003295898438, 35.68492126464844, 14.5382080078125, 32.631500244140625, -28.02691650390625, 18.262557983398438, -4.69586181640625, 64.803466796875, 36.49822998046875, -2.6016464233398438, 14.886428833007812, 29.036209106445312, 39.24725341796875, -21.17218017578125, 23.224029541015625, 54.97210693359375, -32.550628662109375, -14.5859375, 38.919921875, 38.22233581542969, 40.3388671875, 18.65673828125, 49.839141845703125, 35.81687545776367, 61.3309326171875, -26.3961181640625, -22.4952392578125, 34.687713623046875, 65.63888549804688, 30.671661376953125, 67.32363891601562, 6.757591247558594, 42.384033203125, -0.5156097412109375, 0.0, 22.597366333007812, 30.473419189453125, 27.75634765625, 30.10040283203125, -9.838119506835938, 19.011032104492188, 30.451263427734375, 49.275634765625, 63.67669677734375, -13.505729675292969, 83.92178344726562, -35.591156005859375, 21.797958374023438, 5.332305908203125, 26.154136657714844, 42.557891845703125, 43.21490478515625, 6.210609436035156, -6.7945556640625, 28.984954833984375, 19.2501220703125, 29.787506103515625, 51.07861328125, 11.021743774414062, -6.022178649902344, 70.4599609375, 18.584930419921875, 46.8873291015625, 50.26228332519531, 62.711761474609375, 23.119232177734375, 25.956787109375, 4.539581298828125, 1.748504638671875, 16.59124755859375, 21.421649932861328, 1.4161376953125, 12.092010498046875, -28.36200714111328], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000151.npy"}
{"epoch": 0.3162303664921466, "step": 152, "batch_size": 128, "mean": 14.61042594909668, "std": 25.749433517456055, "min": -52.19847106933594, "p10": -15.033905029296868, "median": 12.837196350097656, "p90": 48.762509155273435, "max": 91.95956420898438, "pos_frac": 0.7109375, "sample": [33.75848388671875, 44.727081298828125, 2.741546630859375, 9.038200378417969, 5.813446044921875, 29.36944580078125, 11.0303955078125, 31.197799682617188, 49.3778076171875, 16.787094116210938, 4.8759002685546875, 34.602294921875, 63.01493835449219, 4.41571044921875, 26.05633544921875, -20.974578857421875, -1.842071533203125, 2.419586181640625, 1.7991943359375, -20.79913330078125, -52.19847106933594, 27.618743896484375, -41.478729248046875, 23.272293090820312, -7.840579986572266, 21.84197998046875, 27.241065979003906, -0.6461257934570312, 45.870635986328125, 0.806854248046875, -0.9289474487304688, 43.37809753417969, 46.181610107421875, 27.012969970703125, 33.60023498535156, 30.801345825195312, 44.83551025390625, 46.325592041015625, 55.57080078125, 26.186981201171875, -33.142601013183594, 91.95956420898438, -25.550018310546875, -3.12005615234375, 25.934776306152344, -1.640359878540039, -10.290206909179688, 19.38165283203125, -9.137323379516602, 0.5082244873046875, 15.610946655273438, -7.77911376953125, 11.847320556640625, -4.6160125732421875, -6.0774688720703125, 22.777359008789062, 3.56646728515625, -19.2557373046875, -8.913711547851562, 1.904296875, 55.595947265625, 13.827072143554688, -22.93310546875, 18.764495849609375, 62.43157958984375, -20.5260009765625, 27.362548828125, -3.3003158569335938, 1.046661376953125, 31.87969970703125, 5.601158142089844, -7.17413330078125, -3.3316917419433594, 9.85491943359375, 0.9646072387695312, 62.703399658203125, 33.076263427734375, 18.300689697265625, 56.89399719238281, 27.18570899963379, 51.3115234375, -32.530303955078125, 1.6481304168701172, -1.971954345703125, -2.273468017578125, 5.713775634765625, 22.1356201171875, 26.4541015625, 61.40283203125, 8.18647575378418, 0.4391021728515625, -36.967498779296875, -9.105316162109375, 61.04010009765625, 34.85417175292969, 15.806983947753906, -24.56475830078125, 6.17388916015625, 23.214569091796875, -13.22454833984375, 17.36669921875, 42.8099365234375, 23.0225830078125, 11.16033935546875, -11.7769775390625, 60.8282470703125, 3.981718063354492, 19.1153564453125, -10.706268310546875, -5.822654724121094, 38.570438385009766, 19.001174926757812, -38.108642578125, 21.096923828125, -2.2876853942871094, 7.5728912353515625, 21.44023895263672, 4.202644348144531, 36.43962097167969, 53.520050048828125, 41.45904541015625, 48.498809814453125, 44.399681091308594, 35.73896789550781, 3.344562530517578, 15.392059326171875, -12.221817016601562, 27.300193786621094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000152.npy"}
{"epoch": 0.3183246073298429, "step": 153, "batch_size": 128, "mean": 22.240020751953125, "std": 24.175687789916992, "min": -44.45939636230469, "p10": -2.6093925476074213, "median": 20.424072265625, "p90": 56.495050048828126, "max": 84.4010009765625, "pos_frac": 0.84375, "sample": [13.799530029296875, -0.3276176452636719, 68.99136352539062, 9.132781982421875, 52.706573486328125, 43.223655700683594, 41.402923583984375, 2.7161712646484375, 84.4010009765625, 7.782562255859375, 44.43743896484375, 3.5867996215820312, -4.65966796875, 22.924072265625, 32.605018615722656, 4.430885314941406, -19.5281982421875, 4.2136383056640625, 19.931076049804688, 62.29408264160156, 51.4627685546875, 66.29042053222656, 31.739948272705078, 2.4107818603515625, 15.506072998046875, -2.935760498046875, 36.11970520019531, -0.3395500183105469, 1.2182197570800781, 1.9334716796875, -13.317169189453125, 1.4021453857421875, 42.3172607421875, 4.1508941650390625, 16.45953369140625, 14.515823364257812, 49.19720458984375, 38.326194763183594, 23.09267234802246, 0.0, 56.464599609375, 29.705780029296875, 47.97381591796875, 25.4156494140625, 43.5875244140625, 29.26019287109375, 38.465545654296875, 1.072174072265625, 15.37701416015625, 23.068862915039062, 22.226806640625, 47.29119873046875, -41.679595947265625, 4.9133758544921875, -3.49951171875, 29.620849609375, 41.345703125, 33.502655029296875, -0.5478363037109375, 54.57989501953125, 17.37164306640625, 4.02935791015625, 14.834579467773438, 67.54034423828125, -13.3834228515625, 43.537750244140625, 7.6009521484375, 19.5869140625, 17.654983520507812, 21.44580078125, -14.068206787109375, 28.94140625, 7.914337158203125, 30.993576049804688, -0.82794189453125, -1.1513824462890625, 10.333160400390625, 28.803131103515625, 17.14642333984375, 27.57098388671875, 31.450210571289062, -6.4486236572265625, 39.025482177734375, 63.94757080078125, 23.818313598632812, 24.068771362304688, 1.5544815063476562, 13.236587524414062, 35.9674072265625, 0.10443115234375, 5.157806396484375, 20.521392822265625, 60.83355712890625, 39.96942138671875, 49.53388977050781, 8.892215728759766, 6.28141975402832, 11.371429443359375, 11.77984619140625, 20.326751708984375, 47.418121337890625, 43.70033264160156, 36.99342346191406, 6.031806945800781, 6.0348663330078125, -2.4695205688476562, 16.169921875, 73.38473510742188, 26.7440185546875, 65.49176025390625, 58.029449462890625, 20.884300231933594, 6.064727783203125, 57.46246337890625, -13.134725570678711, 27.451324462890625, 9.90737533569336, 56.56610107421875, 35.97303771972656, 9.295791625976562, -17.367431640625, 67.4019775390625, 45.866241455078125, 9.803909301757812, -44.45939636230469, -31.5438232421875, 30.789031982421875, 29.206573486328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000153.npy"}
{"epoch": 0.3204188481675393, "step": 154, "batch_size": 128, "mean": 15.860121726989746, "std": 28.102550506591797, "min": -70.45907592773438, "p10": -17.90050048828125, "median": 16.77863311767578, "p90": 53.54966125488281, "max": 81.77178955078125, "pos_frac": 0.71875, "sample": [27.40106201171875, 42.4139404296875, -17.8238525390625, 56.6275634765625, 31.434799194335938, 7.418403625488281, -5.488273620605469, 53.175750732421875, 6.0042724609375, 18.250030517578125, -44.066619873046875, 34.0919189453125, 8.265289306640625, 12.2032470703125, 21.950057983398438, -70.45907592773438, 28.702072143554688, -23.927093505859375, 24.655128479003906, 3.633544921875, 11.912002563476562, 3.8780670166015625, 7.745819091796875, -18.079345703125, 4.2622222900390625, -0.572967529296875, 31.86053466796875, 7.4009246826171875, 3.97894287109375, 66.54241943359375, 13.031646728515625, -6.431558609008789, 54.422119140625, 52.91644287109375, 4.901153564453125, 47.411346435546875, 31.64349365234375, 62.501953125, 29.821807861328125, 56.876220703125, 4.389739990234375, 57.85150146484375, 54.664886474609375, 26.604278564453125, -49.11070251464844, 18.27203369140625, 73.1495361328125, -1.4117202758789062, -4.662078857421875, -23.404052734375, -3.2561798095703125, -0.8341217041015625, 17.4151611328125, -21.5242919921875, 25.61602783203125, 36.338409423828125, 24.344512939453125, -6.1550140380859375, 3.4430084228515625, -19.768646240234375, -2.5851974487304688, 20.669631958007812, 41.7015380859375, 30.757110595703125, 28.416549682617188, -48.99066162109375, -17.055166244506836, -0.595062255859375, -1.5499496459960938, -15.53057861328125, 1.1846599578857422, -39.61494445800781, 46.28572082519531, 61.777557373046875, -12.526908874511719, -9.27142333984375, 8.446235656738281, -5.62255859375, 25.73760986328125, 8.811477661132812, -24.499603271484375, 9.263931274414062, -2.6276626586914062, 29.15557861328125, 30.048614501953125, 12.971466064453125, 43.117340087890625, 24.19403076171875, -2.001708984375, 29.692176818847656, 15.823272705078125, 21.53619384765625, 52.752471923828125, 44.486244201660156, 32.67535400390625, -9.78466796875, 20.147659301757812, 11.17681884765625, 81.77178955078125, 11.267791748046875, 21.92901611328125, 44.070465087890625, -13.4659423828125, -57.678497314453125, 37.234771728515625, 16.142105102539062, 29.645217895507812, 28.155853271484375, 18.706634521484375, 12.7701416015625, 59.0474853515625, 35.03143310546875, 28.596145629882812, 50.06340026855469, -1.15289306640625, 3.025653839111328, 72.21197509765625, 17.655517578125, 15.002372741699219, 22.05596923828125, -35.86749267578125, -12.34381103515625, 50.563201904296875, 24.210067749023438, 3.7297897338867188, 22.55853271484375, 34.89278793334961, 59.24523162841797], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000154.npy"}
{"epoch": 0.3225130890052356, "step": 155, "batch_size": 128, "mean": 11.726255416870117, "std": 26.172279357910156, "min": -53.666015625, "p10": -19.156424713134765, "median": 8.664226531982422, "p90": 44.888937377929686, "max": 78.70574951171875, "pos_frac": 0.6640625, "sample": [62.009674072265625, 22.264755249023438, 13.2728271484375, -6.4295501708984375, 29.753971099853516, -0.34762001037597656, 9.010894775390625, -5.38970947265625, 7.039398193359375, 44.554412841796875, 39.95798110961914, 3.641569137573242, 28.77313232421875, -22.214447021484375, 22.002960205078125, 28.584991455078125, -16.762115478515625, 65.70291137695312, 78.70574951171875, -5.82049560546875, 4.66156005859375, 28.283172607421875, 41.746429443359375, 78.69964599609375, 4.5101165771484375, 38.07098388671875, -38.195648193359375, 0.0, -24.729782104492188, -9.10400390625, -17.835289001464844, 53.14593505859375, 27.965545654296875, 41.131996154785156, 7.142791748046875, -14.719924926757812, -19.41876983642578, 34.57563018798828, -6.69207763671875, -2.7182464599609375, 7.56634521484375, 31.60101318359375, 17.4815673828125, 0.6133956909179688, 33.271331787109375, 50.176513671875, 27.62176513671875, 13.702728271484375, 8.632682800292969, 31.996030807495117, -8.657989501953125, -45.1229248046875, 7.626689910888672, 7.447052001953125, 2.3675498962402344, -31.932647705078125, -1.56414794921875, 32.78692626953125, 66.09442138671875, 46.37774658203125, 3.3772964477539062, 3.5393753051757812, 21.306793212890625, -9.253364562988281, -38.874000549316406, -7.027000427246094, 41.759765625, 51.741912841796875, 15.566417694091797, 17.03173828125, -3.4542694091796875, -26.110984802246094, 36.320220947265625, 37.905517578125, -12.898223876953125, 29.354400634765625, 8.695770263671875, -6.518806457519531, 32.82318115234375, -24.50799560546875, 30.492828369140625, 48.98744201660156, -19.043991088867188, -14.594314575195312, 6.8108367919921875, 7.361412048339844, 30.15399169921875, -39.36572265625, 1.006134033203125, 36.53663635253906, 15.914825439453125, 31.716354370117188, 21.2022705078125, -0.754180908203125, -38.96844482421875, 12.205490112304688, 7.1174163818359375, 57.48466491699219, 9.970458984375, -13.234016418457031, -14.431396484375, -11.579345703125, -3.8781661987304688, 8.133663177490234, 18.233673095703125, 36.9107666015625, 3.850189208984375, -5.5249176025390625, -53.666015625, 3.79937744140625, -37.47344970703125, 10.743087768554688, 34.062374114990234, -9.1397705078125, 45.66949462890625, 19.87274169921875, 31.95005989074707, 22.49114990234375, 14.904083251953125, 56.44366455078125, -18.9405517578125, -0.6602325439453125, 1.59857177734375, 14.320587158203125, 16.22113037109375, 20.721771240234375, 14.079086303710938, -2.4462890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000155.npy"}
{"epoch": 0.32460732984293195, "step": 156, "batch_size": 128, "mean": 19.497596740722656, "std": 29.958253860473633, "min": -70.3712158203125, "p10": -21.12957305908203, "median": 23.158668518066406, "p90": 56.078060913085935, "max": 87.5987548828125, "pos_frac": 0.765625, "sample": [30.840545654296875, 38.35865020751953, 25.609375, 19.18560791015625, -1.580047607421875, 0.5915756225585938, 40.898834228515625, 41.97685241699219, 24.972381591796875, 24.55448341369629, 39.31092834472656, 24.865570068359375, 44.58905029296875, 31.910919189453125, 3.2009201049804688, -22.215377807617188, -5.100990295410156, 33.1148681640625, 6.72418212890625, 37.94427490234375, 3.03338623046875, 40.72442626953125, -14.291961669921875, -22.516727447509766, 27.4593505859375, 9.846481323242188, -3.2444305419921875, 16.872421264648438, 0.47983551025390625, 40.39385986328125, 65.13082885742188, -3.857452392578125, 67.0601806640625, -5.2972564697265625, 8.88490104675293, -20.787033081054688, -33.437255859375, 6.438594818115234, 41.575592041015625, 3.2736053466796875, 27.754234313964844, 54.690765380859375, 29.03857421875, 6.7659912109375, 2.6960906982421875, 2.5391845703125, 40.32538604736328, -31.415863037109375, 42.17471694946289, 32.996124267578125, 2.7385826110839844, 31.849531173706055, 17.54656982421875, 24.79248046875, 76.924072265625, -23.599365234375, -32.10453796386719, 79.1319580078125, -1.66497802734375, 6.9837799072265625, -6.78399658203125, 77.5784912109375, 29.123245239257812, 44.75697326660156, 26.962093353271484, 53.2041015625, 47.31219482421875, 6.7696380615234375, 55.59088134765625, 7.896583557128906, 14.880950927734375, 1.1316909790039062, -2.975574493408203, 58.8265380859375, 28.8067626953125, 32.142974853515625, 87.5987548828125, 18.425994873046875, 0.5461254119873047, -3.251241683959961, 74.58905029296875, 34.39350891113281, 65.48196411132812, 1.053436279296875, 8.293861389160156, 24.505661010742188, 2.369495391845703, 29.034835815429688, 29.038124084472656, -25.390945434570312, 36.583251953125, 46.641387939453125, 35.943206787109375, 40.28912353515625, 25.1109619140625, 10.50091552734375, -21.9288330078125, 58.38032531738281, 21.811676025390625, -24.871978759765625, 57.214813232421875, 12.9365234375, 12.91949462890625, -19.16485595703125, -70.3712158203125, 49.682281494140625, -0.8043994903564453, 74.71563720703125, -4.1255035400390625, -35.6572265625, 4.407379150390625, 26.63055419921875, 80.8499755859375, 10.461288452148438, -69.910400390625, -33.367759704589844, -1.754119873046875, 42.044952392578125, 48.4395751953125, 52.193328857421875, 1.5552177429199219, 26.709671020507812, 37.70599365234375, 49.16717529296875, -16.774322509765625, 52.95166015625, -6.915252685546875, 5.9226531982421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000156.npy"}
{"epoch": 0.3267015706806283, "step": 157, "batch_size": 128, "mean": 13.500454902648926, "std": 26.457229614257812, "min": -53.316925048828125, "p10": -18.775393676757812, "median": 9.490623474121094, "p90": 46.89030456542968, "max": 90.3238525390625, "pos_frac": 0.71875, "sample": [33.641510009765625, 1.005645751953125, 22.360626220703125, 57.84967041015625, -2.863433837890625, -18.143890380859375, 12.75152587890625, 18.139678955078125, 47.779510498046875, -7.33306884765625, 24.8680419921875, 0.0274200439453125, -39.874176025390625, 27.302505493164062, 0.5073356628417969, 31.4515380859375, 8.882553100585938, 0.8233566284179688, 2.518756866455078, -10.805404663085938, -16.125732421875, 57.61895751953125, 25.145919799804688, -8.695404052734375, 43.64963912963867, -3.714946746826172, -31.97650146484375, -2.0262298583984375, 5.522125244140625, 15.3211669921875, 7.10577392578125, -5.07501220703125, 7.0552978515625, -45.3436279296875, 32.48577117919922, 41.814422607421875, 19.34686279296875, 27.795867919921875, -38.21282958984375, -53.316925048828125, 70.00506591796875, 15.468246459960938, 29.849609375, 21.383819580078125, 74.26022338867188, 32.86041259765625, 9.270614624023438, -5.652915954589844, 17.830841064453125, 0.0, 15.175582885742188, 44.22093200683594, 7.1316070556640625, -4.950042724609375, 59.78790283203125, 12.17889404296875, 34.419525146484375, -6.126129150390625, -6.412109375, 1.1127490997314453, 46.50921630859375, 24.002532958984375, 47.79869842529297, 7.436431884765625, 26.60619354248047, -2.9925289154052734, 46.22705078125, -4.712249755859375, -42.692138671875, 31.5772705078125, 3.7938232421875, -13.555999755859375, -39.793670654296875, 11.22930908203125, -10.11126708984375, 16.33868408203125, -30.4102783203125, 41.09393310546875, 6.996437072753906, 49.070526123046875, 19.80902099609375, 37.4906005859375, 0.3516998291015625, -5.6985321044921875, 3.10015869140625, 9.71063232421875, 6.9513702392578125, 13.627204895019531, 24.105152130126953, 0.0, -25.14361572265625, 54.825897216796875, 21.144119262695312, 16.381637573242188, 5.7602691650390625, -0.67047119140625, 25.8117618560791, 48.81031036376953, 87.52099609375, 45.912567138671875, 8.4981689453125, -20.2489013671875, 26.51116943359375, 7.854398727416992, 4.2158203125, 0.5575180053710938, 30.664749145507812, 30.14385223388672, 22.672515869140625, -6.7659912109375, -22.083770751953125, 40.608642578125, 3.95269775390625, 33.11016845703125, 29.01116180419922, 30.168914794921875, -4.118743896484375, 0.27969837188720703, -24.56085205078125, 42.787696838378906, -25.78912353515625, 7.1681976318359375, 5.770423889160156, 23.893033981323242, 6.62158203125, 17.678199768066406, 90.3238525390625, 53.8411865234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000157.npy"}
{"epoch": 0.3287958115183246, "step": 158, "batch_size": 128, "mean": 14.883365631103516, "std": 25.770984649658203, "min": -41.295806884765625, "p10": -21.094876098632813, "median": 14.631803512573242, "p90": 46.92423095703125, "max": 75.0322265625, "pos_frac": 0.71875, "sample": [4.518766403198242, -2.0108489990234375, 22.10809326171875, 48.433258056640625, -32.084136962890625, 4.161834716796875, -0.5313873291015625, 14.8804931640625, -13.313766479492188, 50.705596923828125, -14.34814453125, 1.1778564453125, -36.282135009765625, -32.87451171875, 22.279022216796875, 20.18646240234375, 30.476581573486328, 21.6085205078125, 5.21929931640625, 26.481246948242188, 15.964805603027344, 53.5318603515625, 36.04241943359375, 3.2533798217773438, 14.472846984863281, 31.75732421875, 14.513973236083984, -38.4876708984375, 39.046234130859375, 26.718597412109375, -11.929962158203125, 28.400070190429688, -15.07080078125, 47.660675048828125, 17.70318603515625, 6.552703857421875, 34.508514404296875, -34.181636810302734, 19.71044921875, -35.3648681640625, 34.2874755859375, 7.12164306640625, 10.767789840698242, -21.215728759765625, 12.1412353515625, -1.1443023681640625, -6.10296630859375, 15.561203002929688, 31.329086303710938, 20.374797821044922, 55.837493896484375, -4.803855895996094, 1.7768707275390625, 12.720748901367188, 31.254615783691406, 41.25868225097656, -32.26433563232422, -17.7315673828125, 57.57466125488281, 12.474151611328125, 56.057861328125, 7.0609130859375, -3.4290771484375, 10.086395263671875, 34.19970703125, 13.291351318359375, -17.98297119140625, 0.0, 27.952613830566406, 14.28936767578125, 64.2620849609375, -5.139862060546875, -1.77520751953125, 44.883056640625, 23.385162353515625, 46.608612060546875, 4.8753509521484375, -21.070220947265625, -9.1827392578125, -2.324350357055664, 43.96476745605469, 7.9761962890625, -41.295806884765625, 13.674789428710938, 27.669906616210938, 34.49969482421875, 32.233245849609375, 0.3038368225097656, 39.0858154296875, 34.91339111328125, 40.820892333984375, 28.171356201171875, 11.286842346191406, 15.083282470703125, 58.019775390625, 38.49859619140625, 69.47036743164062, -5.470859527587891, -21.73919677734375, -21.15240478515625, 10.512313842773438, 75.0322265625, -12.468170166015625, 33.56285095214844, 43.67779541015625, 38.047943115234375, 46.567169189453125, 26.815887451171875, 33.380035400390625, -0.6872291564941406, 14.7496337890625, 27.50885009765625, 32.371002197265625, 7.6710357666015625, 27.211715698242188, -13.019989013671875, 48.11981201171875, 3.833721160888672, -27.98290252685547, -7.7362060546875, 29.1864013671875, 37.86907958984375, -39.1898193359375, 9.560150146484375, 73.85467529296875, 17.245040893554688, 18.81279754638672, 1.6905097961425781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000158.npy"}
{"epoch": 0.3308900523560209, "step": 159, "batch_size": 128, "mean": 11.95935344696045, "std": 30.79499626159668, "min": -82.91680908203125, "p10": -26.631573486328122, "median": 10.285621643066406, "p90": 51.5971206665039, "max": 88.6226806640625, "pos_frac": 0.6171875, "sample": [-24.187652587890625, 49.946807861328125, 21.19219970703125, 31.9677734375, -2.458648681640625, -14.717193603515625, -35.178070068359375, 48.83489990234375, -3.05322265625, 55.248291015625, -46.89448547363281, 14.566131591796875, -82.91680908203125, -10.776847839355469, 12.405014038085938, 30.842445373535156, 12.5494384765625, -6.9420623779296875, -34.9312744140625, 39.569671630859375, 25.301910400390625, 2.966461181640625, -11.897918701171875, -22.781234741210938, 24.538116455078125, 50.964752197265625, 25.207260131835938, 53.07264709472656, 28.198760986328125, -5.5757904052734375, 20.587690353393555, 1.034759521484375, 40.492523193359375, 13.380409240722656, 11.898170471191406, -16.57733154296875, -2.726104736328125, -20.9569091796875, -3.144306182861328, -30.229278564453125, 41.262367248535156, 14.611465454101562, 30.164794921875, 36.889122009277344, -9.0899658203125, 17.26153564453125, -25.6856689453125, 3.555208206176758, -29.050445556640625, -17.93463897705078, 0.6000595092773438, -9.253524780273438, 44.06170654296875, 9.227462768554688, 19.25537109375, 43.22901916503906, 2.9289703369140625, 43.443359375, -4.862823486328125, -7.20562744140625, 88.6226806640625, 7.827239990234375, -24.71459197998047, 13.2420654296875, 66.57293701171875, 12.714447021484375, 46.5880126953125, 3.75347900390625, -9.455718994140625, 11.113075256347656, 3.0905303955078125, 29.39532470703125, -3.7807960510253906, 40.28289794921875, 32.927978515625, -28.83868408203125, 2.53997802734375, 58.456024169921875, -15.448867797851562, 18.36138916015625, -16.1279296875, -37.96014404296875, 25.81342315673828, 67.13143920898438, 70.669189453125, -11.584617614746094, -13.496162414550781, -31.214569091796875, -10.515182495117188, 48.77996826171875, 25.536109924316406, 17.66009521484375, -14.377044677734375, 20.27416229248047, 10.724197387695312, 30.246109008789062, -28.9376220703125, 54.498779296875, -10.017570495605469, 37.12002944946289, 25.263137817382812, -24.296510696411133, -29.411697387695312, -3.6679229736328125, 83.26434326171875, 72.30670166015625, 22.076385498046875, -45.190757751464844, 36.825523376464844, 3.599925994873047, -5.348724365234375, -7.1741943359375, 57.37535095214844, -41.4208984375, 70.50784301757812, -5.76153564453125, 44.54412841796875, 33.151336669921875, 45.81024169921875, 40.547271728515625, 9.8470458984375, 7.260471343994141, -3.240447998046875, 66.84857177734375, 9.285469055175781, 7.637298583984375, -9.3006591796875, 39.75870895385742], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000159.npy"}
{"epoch": 0.33298429319371725, "step": 160, "batch_size": 128, "mean": 17.872238159179688, "std": 26.215808868408203, "min": -41.64453125, "p10": -12.14179992675781, "median": 15.922149658203125, "p90": 52.215673828125, "max": 92.684814453125, "pos_frac": 0.7578125, "sample": [-4.028717041015625, 38.754547119140625, 0.0, 71.70263671875, 22.06865692138672, 25.55572509765625, 30.443359375, 19.730987548828125, 67.16983032226562, -1.3470458984375, 47.07054138183594, 62.2879638671875, 27.153179168701172, 58.6077880859375, 18.15045928955078, -1.260833740234375, 66.38702392578125, 23.8145751953125, 8.141212463378906, 14.736114501953125, 3.01953125, 13.275848388671875, 21.72698974609375, 29.27410888671875, 2.4051055908203125, -8.877853393554688, -5.9517364501953125, 73.76631164550781, 1.8868446350097656, -39.79510498046875, 14.51287841796875, 92.6256103515625, -30.391693115234375, 40.30895233154297, 17.408042907714844, 4.449270248413086, 5.315986633300781, 3.7897605895996094, 19.818862915039062, 15.814605712890625, 30.593040466308594, 43.79144287109375, 8.54461669921875, 0.0, 32.13970947265625, -12.98529052734375, 29.685585021972656, -24.990310668945312, -11.780303955078125, 0.62713623046875, 12.491607666015625, 34.302978515625, 92.684814453125, 62.47721862792969, -10.452926635742188, 11.898651123046875, 28.387451171875, -14.03314208984375, 9.789497375488281, -11.301483154296875, -4.80523681640625, 47.72901916503906, 7.67645263671875, -23.06683349609375, -9.874832153320312, 49.183990478515625, 43.97943115234375, 50.02323913574219, 4.589639663696289, 9.1986083984375, 23.914756774902344, 1.695648193359375, -0.10448265075683594, 11.86932373046875, 18.946914672851562, 32.14912414550781, 61.43798828125, 44.14580535888672, 30.794281005859375, 16.029693603515625, -22.024215698242188, -5.1822052001953125, 37.80413055419922, -27.590354919433594, -15.097686767578125, 55.85050964355469, 29.85511016845703, 3.28765869140625, 37.8565673828125, -10.313690185546875, 7.190338134765625, 1.8560867309570312, 6.767791748046875, -26.420074462890625, 26.896148681640625, -41.64453125, -1.76507568359375, 25.177642822265625, 52.721832275390625, 3.013946533203125, 12.214698791503906, 17.70494842529297, -19.13657569885254, 25.68988037109375, 19.548858642578125, 10.862895965576172, 14.33676528930664, -7.4051666259765625, 45.332611083984375, 21.757278442382812, 46.76519775390625, 0.3037567138671875, 17.013031005859375, 37.699188232421875, -6.052001953125, 23.99555206298828, 44.834381103515625, 10.281890869140625, 25.108383178710938, 50.14013671875, 54.0382080078125, 27.711471557617188, -28.8323974609375, 32.08740234375, 17.070999145507812, 51.998748779296875, 2.822021484375, 4.6113128662109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000160.npy"}
{"epoch": 0.33507853403141363, "step": 161, "batch_size": 128, "mean": 18.486648559570312, "std": 28.437725067138672, "min": -58.33946228027344, "p10": -14.201095581054688, "median": 16.451248168945312, "p90": 60.40927734375, "max": 70.96347045898438, "pos_frac": 0.7265625, "sample": [-3.4241485595703125, 2.760284423828125, -0.59771728515625, -56.62458801269531, 1.0124034881591797, 22.223114013671875, 1.8598480224609375, 12.319732666015625, 5.57330322265625, 6.9222412109375, 3.5934104919433594, 69.57522583007812, 27.909393310546875, -26.412094116210938, 5.2417144775390625, 2.3299407958984375, 68.5089111328125, 18.644683837890625, 33.86252975463867, -3.36737060546875, 37.487335205078125, 16.090576171875, 49.99066162109375, 70.464111328125, -17.563121795654297, 3.183990478515625, 6.828815460205078, 65.859130859375, 8.13690185546875, 28.508682250976562, 4.47607421875, 25.797414779663086, -5.1244049072265625, -14.353515625, 29.58014678955078, 25.477523803710938, 36.560028076171875, 47.1328125, -1.1457595825195312, 8.393157958984375, -5.18353271484375, -35.46661376953125, -27.1048583984375, -6.7321624755859375, 11.560287475585938, 41.580902099609375, -14.135772705078125, -3.06048583984375, 0.0, 35.295654296875, 17.0906982421875, 32.554718017578125, 6.381126403808594, -0.7110309600830078, 9.011962890625, 1.61090087890625, 46.004127502441406, 48.56340789794922, 36.27899932861328, 26.73865509033203, 50.474456787109375, 58.85406494140625, -18.1700439453125, 16.811920166015625, -11.4033203125, 58.411041259765625, 50.60540771484375, 63.458587646484375, -18.770660400390625, 12.989959716796875, 69.57177734375, 23.898876190185547, 32.89410400390625, 57.847198486328125, -35.95526123046875, 61.679534912109375, 29.242713928222656, -2.77392578125, 1.03778076171875, 31.886817932128906, -12.999122619628906, 0.21788787841796875, 62.08355712890625, 60.544158935546875, 21.889251708984375, 63.792938232421875, 1.5853729248046875, 42.188934326171875, 70.96347045898438, -1.0297489166259766, 1.665496826171875, 22.88848876953125, -14.880889892578125, 37.416961669921875, -11.4508056640625, 28.5810546875, 60.668670654296875, -7.214454650878906, 35.74249267578125, -7.495758056640625, 25.581649780273438, 52.77519226074219, 21.01080322265625, -16.4664306640625, 13.838539123535156, 36.962493896484375, -13.236282348632812, 20.835296630859375, 34.65833282470703, 42.06597900390625, 51.19171142578125, 58.2906494140625, -8.867874145507812, 24.689605712890625, 23.327245712280273, -45.53436279296875, 60.351470947265625, 9.135848999023438, 11.257362365722656, 0.0, -1.87255859375, 61.6104736328125, 52.536773681640625, -58.33946228027344, 19.632843017578125, 15.441940307617188, 2.5548858642578125, 47.14141845703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000161.npy"}
{"epoch": 0.33717277486910996, "step": 162, "batch_size": 128, "mean": 16.583608627319336, "std": 27.278804779052734, "min": -67.25584411621094, "p10": -17.015994262695312, "median": 16.264564514160156, "p90": 53.663018798828126, "max": 79.0301513671875, "pos_frac": 0.765625, "sample": [-5.2263336181640625, 7.743064880371094, 59.38623046875, 36.10472106933594, 2.4633560180664062, 15.581939697265625, -16.891326904296875, 26.39862060546875, 7.5921173095703125, 21.804649353027344, 42.09245300292969, 69.55337524414062, 9.085071563720703, 20.429519653320312, -4.7470703125, -10.439788818359375, 20.78887939453125, 27.215789794921875, 1.7056388854980469, 4.301654815673828, -29.21722412109375, 30.044403076171875, -19.234573364257812, 25.96197509765625, 16.53179931640625, 22.5970458984375, 57.187744140625, 9.32586669921875, -33.5174560546875, 6.5561676025390625, -1.3806915283203125, 19.854248046875, 30.967681884765625, -20.84870147705078, 62.700164794921875, 6.2932586669921875, 16.6883544921875, -55.837310791015625, -11.490671157836914, 30.875930786132812, 1.7665977478027344, 47.21221923828125, 3.0290470123291016, 53.63037109375, 56.48614501953125, -67.25584411621094, 52.78521728515625, 26.897003173828125, 24.957275390625, 53.85430908203125, -3.8712081909179688, 35.309600830078125, 37.99969482421875, 1.04937744140625, 29.455551147460938, -2.581958770751953, -10.47802734375, 36.19941711425781, 64.886962890625, 25.36309814453125, 64.5667724609375, 79.0301513671875, 0.4144287109375, 0.3377532958984375, 31.92791748046875, 47.436248779296875, 49.168495178222656, 58.53801727294922, -17.306884765625, 42.75419616699219, 0.0073947906494140625, 8.46535873413086, -22.974212646484375, 17.548736572265625, 78.7100830078125, -16.362655639648438, -23.639938354492188, 2.302276611328125, 16.023345947265625, 47.983154296875, -6.410402297973633, 7.5386962890625, 36.41645431518555, 26.246307373046875, 53.61834716796875, 47.543426513671875, 6.371955871582031, 0.842041015625, 24.31402587890625, 16.505783081054688, 9.391006469726562, 9.895233154296875, 9.56715202331543, -49.719146728515625, 1.558837890625, 25.87261962890625, 44.43755340576172, 28.472991943359375, 17.4423828125, 53.73919677734375, 34.864593505859375, -32.814727783203125, 41.876556396484375, 5.845703125, 0.9306564331054688, 54.37847900390625, -11.649185180664062, 4.0412445068359375, -3.2159767150878906, 42.883758544921875, -26.813766479492188, 5.25311279296875, -8.99298095703125, 34.817779541015625, 22.884536743164062, 33.042633056640625, 11.958282470703125, -4.182271957397461, 24.984893798828125, 16.630523681640625, 8.16033935546875, -19.71966552734375, -0.9878768920898438, -0.163726806640625, 9.271324157714844, 51.198822021484375, 22.737442016601562, 11.210906982421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000162.npy"}
{"epoch": 0.3392670157068063, "step": 163, "batch_size": 128, "mean": 18.67947769165039, "std": 23.593486785888672, "min": -42.934173583984375, "p10": -11.283311462402343, "median": 19.959896087646484, "p90": 50.07977294921874, "max": 72.47079467773438, "pos_frac": 0.765625, "sample": [20.413360595703125, 0.2090301513671875, 35.055320739746094, -11.848739624023438, 23.112030029296875, -12.641937255859375, 14.412887573242188, 13.67742919921875, 27.53759765625, 0.5871429443359375, -4.8145904541015625, 55.31251525878906, 35.785919189453125, 12.32568359375, 60.242584228515625, 8.169403076171875, 10.732772827148438, -8.24835205078125, 8.476974487304688, -6.8675537109375, -12.66217041015625, 32.35342788696289, 21.5406494140625, -3.2891769409179688, 14.2833251953125, 43.344085693359375, -4.810029983520508, 8.954452514648438, 42.59710693359375, -4.35186767578125, -5.9568939208984375, 49.67607116699219, -1.0104522705078125, 39.4886474609375, 27.593841552734375, 20.54168701171875, 11.020263671875, 39.8865966796875, -10.5592041015625, -5.121742248535156, 32.360137939453125, 26.970199584960938, -25.727920532226562, 7.1681365966796875, 12.758636474609375, 23.175552368164062, -33.11029052734375, 54.250885009765625, 40.307533264160156, -26.52923583984375, 31.065082550048828, 5.7958984375, 33.028961181640625, 28.858230590820312, -0.90594482421875, 48.37187194824219, 17.17095947265625, 52.84326171875, 47.56321716308594, 34.098876953125, 0.779449462890625, -3.50396728515625, -9.395751953125, 26.179420471191406, -8.144927978515625, -11.040985107421875, 9.941703796386719, 40.31488037109375, 3.16888427734375, 29.199752807617188, -42.934173583984375, 53.03791809082031, 20.19878387451172, 19.24468994140625, 11.788818359375, -15.661865234375, 63.59161376953125, -12.6729736328125, 51.727996826171875, 55.52667236328125, 27.124176025390625, 37.97364807128906, 19.72100830078125, 17.479175567626953, 30.196006774902344, 32.39643096923828, 29.211441040039062, 0.4156951904296875, 51.64759826660156, 41.12471008300781, 3.0139293670654297, 6.63580322265625, 20.490440368652344, 72.47079467773438, 29.49700164794922, -25.59307861328125, -21.809951782226562, 22.09807586669922, 67.16302490234375, 25.954330444335938, 46.30494689941406, 41.843170166015625, -14.653533935546875, 40.6109619140625, -3.451934814453125, 15.12286376953125, 51.02174377441406, 16.749557495117188, 24.143692016601562, 26.289649963378906, 49.43878173828125, 29.111251831054688, 18.58782958984375, 0.420867919921875, 16.286903381347656, 13.483123779296875, 27.167938232421875, 49.42103576660156, 27.21856689453125, 57.66057586669922, -3.567136764526367, 2.44342041015625, -26.972991943359375, 47.74153137207031, 38.748565673828125, 27.62891387939453, 8.399946212768555, 0.5546798706054688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000163.npy"}
{"epoch": 0.3413612565445026, "step": 164, "batch_size": 128, "mean": 16.419584274291992, "std": 25.772798538208008, "min": -42.369384765625, "p10": -9.58774070739746, "median": 13.803144454956055, "p90": 51.07965087890624, "max": 88.7906494140625, "pos_frac": 0.71875, "sample": [-9.32427978515625, 14.37314224243164, -2.7288475036621094, 46.839691162109375, -0.0684814453125, -0.8779296875, -5.142848968505859, -6.716827392578125, 19.901290893554688, 28.59722900390625, 30.13116455078125, -10.7845458984375, 34.35321044921875, 41.10009765625, 19.812744140625, -40.189453125, -1.80548095703125, 44.190643310546875, 13.34273910522461, 61.1221923828125, 0.218597412109375, -36.09222412109375, 41.89385986328125, 8.972747802734375, 10.393730163574219, -21.690757751464844, 22.117279052734375, 4.8766937255859375, -2.27593994140625, 5.712074279785156, 29.99395751953125, -0.1441802978515625, 4.277496337890625, 49.70527648925781, -2.686614990234375, 41.206939697265625, -38.663482666015625, 2.1847076416015625, 1.2207260131835938, -33.57014465332031, -4.117218017578125, 59.848052978515625, 34.789703369140625, 24.50439453125, 53.162017822265625, 5.3916168212890625, 46.2847900390625, 1.5829010009765625, -0.259002685546875, 2.624542236328125, -24.19727325439453, 12.110397338867188, 20.069000244140625, 24.668672561645508, -10.571640014648438, 53.55268859863281, -1.6526336669921875, 35.30262756347656, 22.098037719726562, 28.458892822265625, 1.0309600830078125, 8.256820678710938, 52.78369140625, 50.36444091796875, -8.524124145507812, 54.119720458984375, 18.099716186523438, -0.590606689453125, 38.951904296875, 11.20672607421875, 70.18310546875, 28.61333465576172, 58.07012939453125, 0.086669921875, -7.261463165283203, 0.611297607421875, 14.900993347167969, 6.590991973876953, -37.94769287109375, 16.864967346191406, 27.98956298828125, 42.92790222167969, -10.015705108642578, 0.8634567260742188, -0.3934326171875, 28.379425048828125, 39.45245361328125, 55.25555419921875, 35.905609130859375, 20.76580810546875, -3.7517642974853516, 39.082679748535156, 1.3517074584960938, 43.35955810546875, 8.126953125, 40.213043212890625, 41.58654022216797, 52.74847412109375, -9.404327392578125, -0.18781280517578125, -30.7996826171875, 21.487472534179688, 1.3923263549804688, 23.220314025878906, 2.6844863891601562, -5.05133056640625, 46.87689208984375, -5.254852294921875, 35.44367218017578, 7.61328125, 26.663955688476562, 37.545135498046875, 5.376152038574219, 14.2635498046875, -42.369384765625, -18.870849609375, 7.921669006347656, -8.493133544921875, 88.7906494140625, 15.158843994140625, 26.146770477294922, 38.700714111328125, 12.5003662109375, 31.152679443359375, 37.566558837890625, 52.75665283203125, 84.58383178710938, 16.60596466064453], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000164.npy"}
{"epoch": 0.34345549738219894, "step": 165, "batch_size": 128, "mean": 13.46900749206543, "std": 24.55954360961914, "min": -51.67523193359375, "p10": -13.029089355468749, "median": 8.233810424804688, "p90": 46.69356231689453, "max": 75.24282836914062, "pos_frac": 0.6640625, "sample": [3.8676300048828125, 45.285400390625, -3.51904296875, 34.21783447265625, -17.7637939453125, 2.9814682006835938, 24.890029907226562, 6.024993896484375, 40.2230224609375, 40.94671630859375, 60.0787353515625, 3.5621490478515625, -49.490814208984375, 32.37590026855469, 46.63720703125, 13.9266357421875, -4.48248291015625, 57.038330078125, -6.3774261474609375, 39.9188232421875, 25.442855834960938, -45.395477294921875, 46.82505798339844, 32.975624084472656, 5.2283477783203125, -21.039798736572266, -20.4879150390625, -9.27923583984375, 13.599029541015625, 31.116241455078125, 5.510467529296875, -7.9036865234375, -3.70123291015625, 0.62115478515625, 1.2004623413085938, 12.761138916015625, 22.890426635742188, -5.15362548828125, -7.145368576049805, -3.6704483032226562, 15.24066162109375, 16.672927856445312, 43.806427001953125, -16.0699462890625, 12.67669677734375, 48.11346435546875, -0.8779850006103516, -1.7462844848632812, -15.56671142578125, 8.614730834960938, 75.24282836914062, 1.4365692138671875, 2.5911712646484375, -2.909515380859375, 50.969146728515625, 18.910018920898438, -8.266815185546875, 4.606163024902344, -4.103694915771484, 12.25750732421875, -0.912384033203125, 53.39752197265625, -5.284996032714844, 42.1658935546875, 23.705169677734375, -14.28057861328125, 20.740005493164062, -0.393096923828125, -1.97039794921875, -5.9487762451171875, -0.88916015625, -1.5370712280273438, 11.77825927734375, 17.87813377380371, 23.403717041015625, 4.2915802001953125, 33.52496337890625, -22.03045654296875, 27.841339111328125, 15.281005859375, -19.528076171875, 23.61908721923828, 59.456207275390625, -12.49273681640625, 7.106773376464844, 30.016082763671875, 0.495880126953125, 55.20159912109375, 40.092041015625, 18.319580078125, 8.781524658203125, 0.108184814453125, 32.614990234375, 35.97489547729492, 43.60980224609375, -11.390327453613281, -6.3472442626953125, 19.277252197265625, 36.27850341796875, 5.320167541503906, 72.83688354492188, -2.2392425537109375, 30.899581909179688, 33.21269989013672, 54.38749694824219, -26.148475646972656, -10.137950897216797, 15.517578125, 44.29658508300781, 0.43310546875, 0.979522705078125, 4.8289794921875, 51.30072021484375, 28.98077392578125, 39.76622009277344, 38.5655517578125, 4.812774658203125, 9.531608581542969, 7.8528900146484375, 39.28223419189453, -1.3654403686523438, 34.558868408203125, -51.67523193359375, -25.3367919921875, -3.0205001831054688, 50.909400939941406, -8.18511962890625, -4.4192352294921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000165.npy"}
{"epoch": 0.34554973821989526, "step": 166, "batch_size": 128, "mean": 12.678672790527344, "std": 29.776508331298828, "min": -65.10064697265625, "p10": -20.96471252441406, "median": 9.313007354736328, "p90": 51.45399398803711, "max": 93.70660400390625, "pos_frac": 0.65625, "sample": [13.83148193359375, -6.053436279296875, -23.317962646484375, 12.392715454101562, -0.675201416015625, 8.333724975585938, 56.257568359375, 43.70135498046875, -2.5929088592529297, 2.8534603118896484, 6.829803466796875, -5.469429016113281, 93.70660400390625, -10.456245422363281, 2.44598388671875, -6.794891357421875, 18.941726684570312, 29.682708740234375, 7.1294708251953125, 1.8397903442382812, -0.9066009521484375, -28.869842529296875, 11.733840942382812, 49.10215759277344, -51.7066650390625, -20.098297119140625, 1.466970443725586, 26.309921264648438, -45.43260192871094, -10.24578857421875, -5.619415283203125, 16.47705078125, 24.621795654296875, 62.68377685546875, 51.294891357421875, 68.2950439453125, 3.0151634216308594, -2.3260345458984375, -14.183784484863281, -61.203643798828125, 5.819526672363281, -22.612533569335938, 5.193138122558594, -1.8022079467773438, -8.437179565429688, 64.47433471679688, 20.70504379272461, 18.087905883789062, -16.594390869140625, 29.3341064453125, -0.853179931640625, 74.24578857421875, -36.265380859375, 56.01416015625, 80.02264404296875, -43.82652282714844, -24.946044921875, 82.65939331054688, 22.067718505859375, 34.447265625, -18.923545837402344, -8.940948486328125, -20.37860107421875, 31.143692016601562, 40.9169921875, -22.332305908203125, 46.416656494140625, 30.8154296875, 5.76434326171875, -14.61865234375, 19.2022705078125, 21.3125, 59.096923828125, -8.323335647583008, -6.6428375244140625, 18.116455078125, 13.225120544433594, -0.7929344177246094, 9.359382629394531, 4.106292724609375, -4.106719970703125, 31.0318603515625, 14.608108520507812, 10.123977661132812, 10.1328125, 65.01712036132812, 2.27593994140625, 32.502830505371094, 14.064849853515625, -27.700271606445312, 5.556243896484375, -4.863739013671875, 10.534172058105469, 40.1461181640625, 42.669464111328125, 23.996444702148438, 12.44927978515625, 40.514625549316406, 22.795799255371094, 6.15057373046875, 19.054000854492188, 0.13674354553222656, -3.7096481323242188, 39.90001678466797, 1.8555469512939453, 27.17022705078125, 45.264434814453125, 39.6968994140625, -9.022216796875, 51.46228790283203, -31.458709716796875, -65.10064697265625, 8.22552490234375, 43.208831787109375, 49.079071044921875, -16.871315002441406, 9.883590698242188, -0.9322509765625, 18.73516845703125, -7.350616455078125, 51.450439453125, -1.3738555908203125, 91.83599853515625, 12.667213439941406, 1.810394287109375, 26.631942749023438, 9.266632080078125, 10.204132080078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000166.npy"}
{"epoch": 0.34764397905759165, "step": 167, "batch_size": 128, "mean": 14.476500511169434, "std": 26.091190338134766, "min": -49.99378967285156, "p10": -13.025990295410155, "median": 11.632217407226562, "p90": 50.85534057617186, "max": 82.88311767578125, "pos_frac": 0.6875, "sample": [37.96015167236328, 4.694725036621094, 10.04669189453125, 35.2052001953125, 7.71807861328125, 38.3992919921875, 39.05206298828125, 59.885955810546875, 27.836395263671875, 12.878738403320312, 44.34864807128906, -44.413360595703125, 53.54949951171875, 1.0970611572265625, 13.17987060546875, 64.042724609375, 13.4312744140625, 56.929046630859375, -0.60552978515625, 12.145751953125, 2.74810791015625, 28.49786376953125, -44.64337158203125, -48.915130615234375, -1.7291336059570312, -0.11438751220703125, -16.7266845703125, 44.140655517578125, 20.044509887695312, 43.0950927734375, 24.301544189453125, 49.86125183105469, 11.277610778808594, 53.924774169921875, 33.64387512207031, 8.7177734375, 14.491142272949219, -23.6829833984375, 19.314605712890625, -7.410346984863281, 57.09434509277344, -2.265533447265625, 16.8824462890625, 4.064422607421875, -1.1340484619140625, -49.99378967285156, 0.0, 2.8037109375, 2.5833740234375, 26.437423706054688, 14.605270385742188, 60.8800048828125, -19.209877014160156, 53.17488098144531, -13.213241577148438, 11.331695556640625, 37.237396240234375, 11.9327392578125, 18.744338989257812, 0.72747802734375, -30.446319580078125, 6.3924560546875, 63.43505859375, 0.3442535400390625, 19.554718017578125, 11.069625854492188, 17.069412231445312, -12.94573974609375, 33.382354736328125, 3.563234329223633, -1.08306884765625, 0.7717323303222656, 31.889076232910156, -27.484176635742188, 39.10986328125, 26.037857055664062, 41.65531921386719, -0.96307373046875, 21.23876953125, -0.506622314453125, -2.7370071411132812, 44.45210266113281, 29.681884765625, -21.815826416015625, 43.766143798828125, -6.020164489746094, 17.32550048828125, -6.282482147216797, 30.110595703125, 28.284774780273438, 29.64715576171875, 5.3655242919921875, 39.125038146972656, 17.74609375, 25.2275390625, 68.0478515625, -4.302886962890625, -12.0546875, -10.395057678222656, -1.443695068359375, -5.5305328369140625, 4.594310760498047, -6.689544677734375, 58.95098876953125, -10.517608642578125, 6.9332275390625, 82.88311767578125, 15.49810791015625, 37.17620849609375, -11.669906616210938, 5.1302490234375, 32.29638671875, -24.530487060546875, -4.245567321777344, -2.4165802001953125, 4.323936462402344, 26.674530029296875, -4.989898681640625, -3.26336669921875, 41.302459716796875, 33.95268249511719, 16.50476837158203, 8.048477172851562, -2.5649185180664062, -44.01312255859375, 41.7996826171875, 7.586090087890625, 63.05320739746094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000167.npy"}
{"epoch": 0.34973821989528797, "step": 168, "batch_size": 128, "mean": 18.861879348754883, "std": 26.86654281616211, "min": -61.686279296875, "p10": -9.802281188964841, "median": 15.948217391967773, "p90": 53.18707275390625, "max": 84.05816650390625, "pos_frac": 0.734375, "sample": [22.123046875, 15.497772216796875, 65.36798095703125, 50.5042724609375, 30.98052978515625, 13.243852615356445, 34.54142761230469, 16.044475555419922, 44.37939453125, -6.37603759765625, 52.998291015625, -11.376800537109375, -1.960418701171875, 18.584930419921875, -1.232757568359375, 3.56390380859375, 21.531326293945312, 27.2864990234375, -6.514820098876953, 41.87654113769531, 12.808372497558594, 50.66766357421875, 16.850753784179688, 23.152191162109375, 52.16644287109375, -2.075939178466797, 14.989112854003906, 42.50387954711914, -7.916923522949219, 0.00089263916015625, 0.0, 14.513763427734375, 66.07275390625, 13.336044311523438, 47.718780517578125, 33.408782958984375, 15.847412109375, 0.331268310546875, 25.67987060546875, 1.42926025390625, 54.706298828125, 12.22088623046875, -1.619873046875, 36.67529296875, -9.127487182617188, 73.93865966796875, 15.229339599609375, 28.48004150390625, 67.31988525390625, -11.4476318359375, 45.90679931640625, 8.326828002929688, 26.01192855834961, 11.5848388671875, -35.28057861328125, 28.66717529296875, 41.629676818847656, 39.5185546875, 39.910377502441406, 4.1140594482421875, -20.062652587890625, -16.47093963623047, 1.89642333984375, -0.83544921875, 37.642852783203125, -0.9593372344970703, 39.263427734375, 35.345703125, 84.05816650390625, 23.052764892578125, 59.908935546875, 7.000667572021484, -61.686279296875, 34.7945556640625, 54.502197265625, 0.6311836242675781, 6.88189697265625, 36.318328857421875, 4.4685821533203125, -1.701751708984375, -4.161041259765625, 6.882965087890625, 26.49652099609375, 19.51171875, 57.362640380859375, 42.42010498046875, 48.596893310546875, 33.627532958984375, 64.6734619140625, 36.61250305175781, -15.520523071289062, -5.7525482177734375, 50.292877197265625, -14.169036865234375, 5.753959655761719, 42.9755859375, -7.970741271972656, 20.14556884765625, 17.762252807617188, -45.20951843261719, 15.851959228515625, 8.271575927734375, 44.78260040283203, 31.70819091796875, 24.437637329101562, 59.64434814453125, -2.9809341430664062, -3.3237762451171875, 3.873241424560547, -1.4700851440429688, 6.3970947265625, -1.58148193359375, -18.97076416015625, -32.88847351074219, 44.654205322265625, 14.6051025390625, 33.0057373046875, 53.6275634765625, 47.023590087890625, 38.15928649902344, -2.03857421875, 22.0679931640625, 11.362030029296875, -7.01226806640625, -32.60870361328125, 10.237960815429688, 66.45947265625, -52.667327880859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000168.npy"}
{"epoch": 0.3518324607329843, "step": 169, "batch_size": 128, "mean": 21.89925765991211, "std": 27.337188720703125, "min": -43.88087463378906, "p10": -6.0947242736816385, "median": 15.94253921508789, "p90": 61.619363403320314, "max": 85.4998779296875, "pos_frac": 0.796875, "sample": [-2.1386489868164062, 35.00769805908203, 74.89395141601562, 29.25433349609375, -7.4520263671875, 10.65652084350586, 2.6067123413085938, 7.6896514892578125, 2.8275146484375, 22.5987548828125, 21.300704956054688, 44.61297607421875, 47.10541915893555, 42.9346923828125, -22.91156005859375, -1.1867828369140625, 44.261932373046875, 6.4383544921875, 4.270818710327148, 2.5206298828125, -1.151824951171875, 8.085433959960938, 40.94104766845703, 12.803207397460938, 42.624420166015625, 48.293609619140625, 23.813018798828125, -8.473876953125, 17.736053466796875, 41.353973388671875, 51.695220947265625, 52.628204345703125, 43.5289306640625, -7.6240997314453125, 4.496238708496094, 7.4455108642578125, -11.897066116333008, -43.88087463378906, 22.407546997070312, -3.855224609375, -5.513023376464844, 55.38226318359375, -9.511947631835938, 12.965751647949219, 0.0, -34.37517547607422, 8.17315673828125, 18.199737548828125, 13.189697265625, 16.34613037109375, -3.192413330078125, 8.536865234375, 83.06103515625, 54.395721435546875, 27.434444427490234, 39.88572692871094, -36.622230529785156, 42.253456115722656, 0.4022655487060547, -1.8250274658203125, 63.283203125, 37.323211669921875, 9.303054809570312, -4.230781555175781, 34.33551025390625, 62.18896484375, 52.08677673339844, 71.52194213867188, 14.897796630859375, 49.0791015625, 0.8194580078125, 71.945068359375, 5.4415283203125, 0.67388916015625, 34.760345458984375, -28.835556030273438, 6.524833679199219, 62.059234619140625, 22.527236938476562, -7.9267578125, 44.98872375488281, 6.902153015136719, 6.268749237060547, 37.000274658203125, 16.19947052001953, 41.53913879394531, 53.27618408203125, -3.786346435546875, 31.614044189453125, 28.3514404296875, 1.4088668823242188, 35.54437255859375, 5.052711486816406, 85.4998779296875, 28.907867431640625, 10.613208770751953, 9.979766845703125, 4.5465087890625, 70.56773376464844, 37.34454345703125, 5.987981796264648, 40.02410125732422, 75.98239135742188, 14.6014404296875, -24.059844970703125, -23.1796875, 6.4942626953125, -0.8452377319335938, 79.63946533203125, 23.111587524414062, 40.59716796875, 0.12185096740722656, 57.026702880859375, 44.76715087890625, 42.821746826171875, 0.0, 13.347381591796875, 3.6100921630859375, 15.68560791015625, 34.86756896972656, 18.9971923828125, 0.2998790740966797, 10.482147216796875, 16.23623275756836, 71.1407470703125, 73.75689697265625, 61.43084716796875, -4.885749816894531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000169.npy"}
{"epoch": 0.3539267015706806, "step": 170, "batch_size": 128, "mean": 18.284976959228516, "std": 24.920488357543945, "min": -46.4642333984375, "p10": -8.65802307128906, "median": 15.525829315185547, "p90": 49.67543792724609, "max": 86.97344970703125, "pos_frac": 0.765625, "sample": [3.96002197265625, 10.607330322265625, 36.01789855957031, -10.048538208007812, 40.736328125, 8.29779052734375, -0.7431259155273438, 86.97344970703125, 21.31475067138672, 9.782958984375, 27.840179443359375, 73.13156127929688, 29.534591674804688, 19.93639373779297, -34.7061767578125, 78.32626342773438, -20.96551513671875, 27.860107421875, -0.002349853515625, 11.961381912231445, -23.59967041015625, 5.54754638671875, 25.670013427734375, 35.95635986328125, 5.634452819824219, -16.233871459960938, -14.020233154296875, 3.138774871826172, 50.660400390625, -4.1068115234375, 26.365325927734375, 14.333640098571777, 25.26361083984375, 4.691366195678711, 14.596778869628906, 1.614166259765625, 0.431243896484375, 5.0868377685546875, 43.39056396484375, 48.51104736328125, 1.52716064453125, 25.757003784179688, 4.7232818603515625, 55.73077392578125, 38.68731689453125, 30.171432495117188, 29.273895263671875, -2.055419921875, -16.013671875, 52.234039306640625, 51.13191223144531, 43.932533264160156, 38.85821533203125, -36.33424758911133, 16.399810791015625, 0.426971435546875, 55.201141357421875, -10.633224487304688, 49.25331115722656, 31.66643524169922, 12.12786865234375, 59.6279296875, 46.68284606933594, -5.3867645263671875, 32.791500091552734, 29.821067810058594, -3.017425537109375, 5.3693084716796875, 47.88580322265625, 2.46234130859375, 20.906883239746094, 34.92156982421875, 7.286886215209961, 16.80230712890625, 38.83385467529297, 26.540008544921875, 0.0, 44.07666015625, 14.822502136230469, 36.45957946777344, -23.52386474609375, 40.029632568359375, 8.849609375, 36.297882080078125, 46.072662353515625, 25.035736083984375, -33.133331298828125, 8.74810791015625, 35.338623046875, -46.4642333984375, 21.032989501953125, 18.082275390625, -4.64837646484375, 27.7469482421875, -0.9261474609375, -2.9153594970703125, 45.570098876953125, -0.6689910888671875, 24.48724365234375, 85.90716552734375, 0.4987640380859375, -8.062088012695312, 24.045143127441406, -2.661376953125, -5.81378173828125, 6.949432373046875, 19.586883544921875, 10.434616088867188, 2.3071250915527344, 13.053070068359375, -2.725982666015625, 2.03985595703125, 68.0548095703125, 18.491729736328125, -11.473968505859375, 13.164321899414062, 7.3113250732421875, 42.8814697265625, 16.229156494140625, 2.0988636016845703, -1.3587398529052734, 30.572235107421875, 56.46966552734375, 47.17340850830078, 5.639442443847656, 20.085548400878906, -0.309661865234375, 53.176727294921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000170.npy"}
{"epoch": 0.35602094240837695, "step": 171, "batch_size": 128, "mean": 18.689701080322266, "std": 25.587400436401367, "min": -28.072784423828125, "p10": -14.48537139892578, "median": 16.422260284423828, "p90": 57.1923828125, "max": 80.47233581542969, "pos_frac": 0.7421875, "sample": [75.3323974609375, 12.266845703125, 11.850448608398438, 11.17718505859375, 49.438079833984375, 51.14674377441406, 49.720787048339844, -21.723785400390625, -5.853912353515625, 19.811355590820312, -16.723068237304688, 1.2528533935546875, 62.47563171386719, -12.421920776367188, 1.0908660888671875, -0.4336395263671875, -12.217910766601562, 21.364822387695312, 25.823486328125, 61.211212158203125, 42.29109191894531, -3.544483184814453, 16.762939453125, 66.15374755859375, 39.23036193847656, 20.2806396484375, -21.761627197265625, 10.840774536132812, -20.811309814453125, 16.8280029296875, 8.429718017578125, 8.382308959960938, -10.013580322265625, 31.88595199584961, 43.7913818359375, 63.167388916015625, 43.206298828125, 4.836700439453125, 12.322982788085938, -2.6909103393554688, 14.0196533203125, -11.311065673828125, -5.554630279541016, -6.2340087890625, -11.520339965820312, 59.28326416015625, 49.997222900390625, 9.77128791809082, -28.072784423828125, 3.6575775146484375, 56.85052490234375, 23.96368408203125, 8.64190673828125, 49.8660888671875, 27.984130859375, 22.060165405273438, 31.51177978515625, 59.9869384765625, 57.99005126953125, -15.659454345703125, 34.535888671875, 19.908592224121094, -9.050277709960938, 45.16449737548828, 28.07870101928711, 48.29438781738281, 21.161865234375, 8.568029403686523, 43.96142578125, 34.457496643066406, 16.453590393066406, 63.500457763671875, -1.7042694091796875, 4.443511962890625, -23.537620544433594, -14.33984375, 58.69282531738281, 4.898578643798828, -22.14642333984375, 6.022369384765625, -0.4738140106201172, 2.629364013671875, -9.219024658203125, -22.366500854492188, 15.597900390625, -15.999069213867188, 38.55034255981445, 26.798126220703125, -2.56170654296875, 24.557159423828125, 38.821495056152344, 74.38836669921875, -26.952091217041016, 17.276756286621094, 25.813568115234375, 45.99137878417969, 42.527984619140625, 2.4058589935302734, 52.260101318359375, 19.99505615234375, 18.21624755859375, 28.46258544921875, -14.824935913085938, 7.66876220703125, 12.31109619140625, 13.83453369140625, 16.948226928710938, 48.642974853515625, 13.071617126464844, 2.1123428344726562, 16.39093017578125, -16.280105590820312, 45.024658203125, 38.44024658203125, 33.73973083496094, 15.142242431640625, 20.9365234375, 4.489982604980469, 19.369232177734375, 13.061721801757812, 80.47233581542969, 39.723907470703125, -7.419189453125, 1.044931411743164, -2.6259918212890625, 59.73480224609375, 21.35986328125, -5.581390380859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000171.npy"}
{"epoch": 0.3581151832460733, "step": 172, "batch_size": 128, "mean": 12.475286483764648, "std": 26.213340759277344, "min": -69.32684326171875, "p10": -12.977438163757324, "median": 11.263862609863281, "p90": 48.07240600585937, "max": 67.210693359375, "pos_frac": 0.6953125, "sample": [11.151840209960938, -8.044586181640625, 4.660797119140625, 11.93646240234375, 22.709915161132812, 10.545524597167969, -3.74200439453125, -8.237716674804688, 6.220314025878906, -0.177215576171875, 1.8938217163085938, 0.4584827423095703, -46.251007080078125, 4.169177055358887, 34.71015930175781, 10.084341049194336, 21.195022583007812, 57.44830322265625, 24.65325927734375, 8.315963745117188, -6.3709716796875, 9.72674560546875, 23.5595703125, 57.56231689453125, 3.381805419921875, 32.5823974609375, 28.182693481445312, -7.1187744140625, -5.298095703125, 40.975830078125, 2.30279541015625, 24.196380615234375, -51.82330322265625, 3.96221923828125, 16.91259765625, 35.361900329589844, -13.413238525390625, 57.684295654296875, -22.1282958984375, -0.9468307495117188, -6.2596540451049805, 14.227340698242188, -4.54193115234375, -1.474813461303711, 62.26298522949219, -69.32684326171875, -5.9132537841796875, 31.08306884765625, 50.95123291015625, 39.848419189453125, -4.195098876953125, -27.50933837890625, 18.082275390625, 0.726959228515625, 22.668853759765625, 13.635162353515625, 43.3365478515625, 54.11558532714844, -11.26983642578125, -4.338264465332031, -3.297210693359375, 14.46380615234375, 20.647918701171875, -2.208038330078125, 43.97210693359375, 33.980194091796875, -8.26165771484375, 3.815643310546875, 14.797470092773438, 5.8623199462890625, 62.305908203125, -44.553436279296875, 4.934089660644531, 36.112579345703125, 56.0838623046875, 47.913543701171875, 4.2965850830078125, 2.6840782165527344, 15.6097412109375, -46.5389404296875, 26.2906494140625, 7.749671936035156, 20.322769165039062, -43.71026611328125, 15.505699157714844, 32.958160400390625, 6.897819519042969, 52.20233154296875, 11.375885009765625, -33.40118408203125, 56.30560302734375, -4.605340957641602, 48.443084716796875, -27.7733154296875, -21.33747100830078, 15.557876586914062, 18.544830322265625, 16.613067626953125, -12.790666580200195, 12.749977111816406, -4.5072021484375, 67.210693359375, 34.416351318359375, 34.41854476928711, 16.404312133789062, 21.26226806640625, -5.6439971923828125, 19.57763671875, 25.470882415771484, 6.99371337890625, 14.1168212890625, 41.122581481933594, 23.756404876708984, 46.39337158203125, 44.49095153808594, 28.289108276367188, -4.971282958984375, 8.8565673828125, 6.872396469116211, 43.41278076171875, -55.738250732421875, 52.657508850097656, -2.5149383544921875, 8.748916625976562, 0.0, 34.190673828125, 26.53460693359375, -4.60675048828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000172.npy"}
{"epoch": 0.36020942408376966, "step": 173, "batch_size": 128, "mean": 15.613748550415039, "std": 25.27781867980957, "min": -56.2747802734375, "p10": -12.782439422607416, "median": 13.757827758789062, "p90": 47.92542343139648, "max": 92.80178833007812, "pos_frac": 0.7109375, "sample": [3.658203125, -56.2747802734375, -35.13396453857422, 30.471664428710938, 24.09320068359375, -3.403472900390625, 41.8939208984375, -33.791168212890625, 42.4312744140625, 74.74664306640625, -4.949146270751953, 22.178375244140625, -8.438175201416016, 34.71368408203125, 32.267364501953125, 46.3333740234375, 3.26202392578125, 4.0290985107421875, 17.498443603515625, 60.805328369140625, 50.987152099609375, 8.067459106445312, 5.456024169921875, 31.58123779296875, 12.089347839355469, 11.4112548828125, -6.4529876708984375, 7.959381103515625, 21.595733642578125, 37.81944274902344, 10.5723876953125, 30.296646118164062, -2.079631805419922, 5.601528167724609, 6.00262451171875, 31.370346069335938, -30.76861572265625, 22.366363525390625, 16.470138549804688, 39.288787841796875, -7.4888763427734375, 20.575927734375, 28.470947265625, 29.726882934570312, 38.32975769042969, 31.6439208984375, -4.203277587890625, -3.8729095458984375, 4.081703186035156, 9.84075927734375, -3.5901412963867188, -30.138992309570312, -19.391098022460938, 92.80178833007812, 13.02099609375, -25.8397216796875, 3.3969192504882812, -17.702484130859375, 26.419601440429688, 75.86416625976562, 46.27606201171875, 49.16650390625, 2.1377220153808594, 39.83488464355469, 40.8751220703125, 47.9508056640625, -35.88433837890625, 18.34307861328125, 14.361328125, 17.495437622070312, 19.414783477783203, 2.5996551513671875, 47.91558837890625, 35.82170867919922, 3.2747802734375, 36.78155517578125, 9.034408569335938, 53.788482666015625, -3.2328529357910156, 7.949851989746094, 18.36187744140625, -24.614044189453125, 2.444364547729492, 6.479827880859375, 42.25910949707031, 5.5812530517578125, 33.523216247558594, 45.51580810546875, -1.6533203125, 4.92425537109375, 13.4984130859375, -2.406158447265625, 39.87458801269531, 21.427078247070312, 51.45220947265625, 22.65447998046875, 34.8570556640625, 28.009368896484375, 36.954437255859375, 43.46893310546875, 51.91114807128906, 7.6824951171875, -4.506616592407227, -8.004623413085938, -3.966764450073242, 47.94837188720703, -18.66912841796875, 45.71919250488281, 21.231185913085938, 16.17181396484375, -4.8023223876953125, -0.624176025390625, -11.133636474609375, 14.017242431640625, 1.4718570709228516, -5.232666015625, -2.819091796875, -16.62964630126953, 54.061004638671875, -5.503059387207031, 54.520111083984375, -6.876434326171875, 0.0, 21.80522918701172, 20.259552001953125, -25.264129638671875, -6.4443511962890625, 17.747528076171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000173.npy"}
{"epoch": 0.362303664921466, "step": 174, "batch_size": 128, "mean": 19.734516143798828, "std": 25.332143783569336, "min": -39.157958984375, "p10": -9.08310852050781, "median": 15.544021606445312, "p90": 54.97734985351562, "max": 97.76568603515625, "pos_frac": 0.78125, "sample": [41.18516540527344, 1.4172229766845703, 3.6072463989257812, 0.7614364624023438, 67.11181640625, 43.677825927734375, 57.361663818359375, -24.55975341796875, 8.2314453125, 0.9566612243652344, 50.087432861328125, 38.31726837158203, 1.3858642578125, 39.606773376464844, 13.18756103515625, 24.04656982421875, 49.93377685546875, 82.77703857421875, 5.945098876953125, 10.113021850585938, -3.2319488525390625, 54.50469970703125, 1.130279541015625, 63.176361083984375, 64.9893798828125, -6.47479248046875, 32.77893829345703, 8.271453857421875, 64.42398834228516, 37.472381591796875, -3.586578369140625, 38.26727294921875, 97.76568603515625, -16.72686767578125, -12.821220397949219, -7.0405731201171875, 44.680908203125, 56.0802001953125, 66.33944702148438, 2.4215087890625, 14.746696472167969, 23.797027587890625, 27.20647430419922, 20.877120971679688, -2.2889881134033203, 59.3193359375, 40.895263671875, 29.542236328125, -0.395477294921875, 78.68157958984375, -8.70745849609375, -17.147705078125, 41.123924255371094, 28.49822998046875, -3.1639862060546875, 2.75311279296875, 5.954250335693359, -1.5446319580078125, 21.281768798828125, 18.19305419921875, 10.241973876953125, 3.6336212158203125, 59.379302978515625, 42.43009948730469, 15.172515869140625, -11.573684692382812, -3.4488372802734375, 33.5963134765625, 10.734764099121094, 29.77532958984375, 16.251327514648438, 38.422332763671875, 46.60852813720703, 42.050506591796875, 34.416473388671875, 21.828506469726562, 7.682952880859375, 10.041168212890625, 49.0186767578125, -9.959625244140625, -24.433273315429688, 29.495521545410156, 1.3818511962890625, 27.3302001953125, 17.004638671875, 2.2266483306884766, 49.20127868652344, 7.621786117553711, -13.67535400390625, 13.15256118774414, 8.055831909179688, 14.5631103515625, 43.08552551269531, 27.553253173828125, 34.6246337890625, 31.125335693359375, 1.719482421875, 65.15643310546875, -2.291095733642578, 13.815208435058594, 40.46136474609375, -13.603973388671875, 33.933319091796875, 19.347305297851562, -5.871490478515625, 10.349205017089844, -4.877494812011719, 1.3768062591552734, -1.943603515625, -26.680328369140625, 30.940872192382812, 0.977874755859375, 2.347991943359375, 26.272254943847656, 21.449615478515625, -13.4798583984375, 53.64227294921875, -11.4468994140625, 6.1080322265625, -39.157958984375, 9.158172607421875, 15.91552734375, 29.63006591796875, 18.75634765625, 35.504364013671875, 1.1236801147460938, 20.247161865234375, -4.67205810546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000174.npy"}
{"epoch": 0.3643979057591623, "step": 175, "batch_size": 128, "mean": 15.807592391967773, "std": 27.68558692932129, "min": -76.60928344726562, "p10": -15.749522399902338, "median": 14.414527893066406, "p90": 48.833232116699214, "max": 98.807373046875, "pos_frac": 0.765625, "sample": [40.293701171875, 25.4815673828125, 29.276611328125, -2.4079666137695312, 1.5354499816894531, -32.855552673339844, 20.2705078125, 6.6156158447265625, 71.21600341796875, -39.691650390625, 10.80755615234375, 11.253990173339844, 19.683319091796875, 20.059219360351562, 37.2126579284668, 98.807373046875, 3.1657447814941406, 34.53692626953125, 16.820297241210938, 57.09550476074219, -27.9962158203125, 43.51849365234375, 44.69776916503906, 36.436279296875, 31.764816284179688, -2.52044677734375, -3.1239471435546875, -25.33270263671875, 21.930389404296875, -21.802780151367188, -3.7790603637695312, -3.7064743041992188, 18.68707275390625, 47.578857421875, -3.3436126708984375, 46.83741760253906, 9.582168579101562, 2.6384735107421875, 56.907135009765625, 37.8275146484375, 2.962371826171875, 10.393035888671875, 12.92083740234375, 17.33648681640625, 7.653167724609375, 54.735809326171875, 40.916900634765625, 41.18865966796875, -11.68310546875, 63.5655517578125, 16.63128662109375, 10.811866760253906, 3.42608642578125, 10.09857177734375, -11.612701416015625, -19.3212890625, 13.543731689453125, -1.7917404174804688, -0.5919876098632812, 53.129638671875, 17.929973602294922, 48.71380615234375, 7.52862548828125, 30.05336570739746, 45.216522216796875, 6.0166168212890625, 1.5915069580078125, 4.376596450805664, 0.447265625, 7.1705322265625, 44.46197509765625, -2.0088729858398438, 5.5785980224609375, 47.66436767578125, 35.601722717285156, -27.260757446289062, 0.642608642578125, 49.11189270019531, 0.0, -76.60928344726562, 16.2333984375, 32.06781005859375, 59.678558349609375, 7.2193603515625, 18.111083984375, -3.3236083984375, 20.114959716796875, -14.218765258789062, -28.12664794921875, 66.00146484375, 36.583587646484375, 4.217670440673828, 32.912689208984375, 39.749542236328125, 39.203094482421875, 28.862762451171875, -7.669769287109375, 9.33305549621582, -21.053878784179688, 5.485809326171875, 35.91239929199219, 37.781578063964844, 8.697647094726562, 19.456634521484375, 5.156402587890625, -57.6258544921875, 17.599212646484375, 15.285324096679688, -7.42803955078125, 19.18548583984375, 26.691791534423828, 16.543014526367188, 23.327865600585938, 23.99749755859375, 60.48553466796875, 4.38427734375, 19.804025650024414, 65.58184814453125, 35.59992980957031, 9.383819580078125, 12.099639892578125, 6.0950164794921875, 18.3804931640625, -58.07843017578125, -45.384002685546875, 0.0, 3.843700408935547, 68.69461059570312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000175.npy"}
{"epoch": 0.36649214659685864, "step": 176, "batch_size": 128, "mean": 11.684707641601562, "std": 28.631805419921875, "min": -85.88272094726562, "p10": -17.934457397460932, "median": 9.258586883544922, "p90": 46.025569915771484, "max": 81.05615234375, "pos_frac": 0.71875, "sample": [-15.82965087890625, 10.484954833984375, -1.90673828125, -16.58502197265625, 18.393051147460938, 34.69813537597656, -35.02732849121094, -37.107666015625, 8.450908660888672, 14.06793212890625, -64.25775146484375, 2.376312255859375, 17.155139923095703, 46.226348876953125, 9.7904052734375, 33.10939025878906, 33.319488525390625, 2.3401947021484375, 2.155975341796875, -6.91552734375, -33.326385498046875, 4.7457275390625, 9.32563591003418, -54.60797119140625, 12.438018798828125, 59.642120361328125, 23.959571838378906, 75.44497680664062, 26.505203247070312, -2.0815582275390625, 68.17758178710938, -26.317642211914062, -1.9413604736328125, 14.154144287109375, 5.985042572021484, 4.282989501953125, -10.049102783203125, 20.67822265625, 12.895782470703125, 5.611595153808594, -35.10479736328125, 9.191537857055664, 38.402740478515625, -2.473846435546875, 6.174430847167969, 45.93952178955078, 4.856708526611328, 1.4392242431640625, 20.805145263671875, 60.0020751953125, -23.148292541503906, 39.96771240234375, 33.624053955078125, -14.605880737304688, 12.9853515625, 45.36248779296875, -3.8448486328125, 32.85942840576172, 27.4171142578125, 3.6085281372070312, 3.5145263671875, 0.6439743041992188, -4.289276123046875, -12.865264892578125, 36.55206298828125, 45.195404052734375, 29.920501708984375, -10.318603515625, 51.3165283203125, 3.374359130859375, -16.615081787109375, 56.7237548828125, 14.173202514648438, 7.9417724609375, 81.05615234375, 42.976104736328125, 0.8223056793212891, -14.217620849609375, 14.196746826171875, 13.681167602539062, 24.165992736816406, 30.156478881835938, 39.13881301879883, 33.252479553222656, 29.333797454833984, 43.27660369873047, -3.311248779296875, 7.0288848876953125, 2.611114501953125, 64.03167724609375, 22.934677124023438, 3.7294769287109375, -25.334182739257812, 22.745147705078125, -10.25628662109375, 1.3263530731201172, 43.12762451171875, -59.751678466796875, -21.01300048828125, 35.2835693359375, -14.963348388671875, 40.34303283691406, 2.8404312133789062, -14.183563232421875, 6.0739593505859375, -50.580322265625, 27.58251953125, 12.513336181640625, -11.558258056640625, 7.291423797607422, 51.71802520751953, -5.694755554199219, 62.861907958984375, 42.041900634765625, 64.46527099609375, 19.902244567871094, 1.5966796875, 20.514659881591797, -10.500762939453125, -85.88272094726562, 48.6429443359375, 1.51715087890625, 0.5687274932861328, 23.583717346191406, 26.28271484375, 10.613861083984375, 18.123146057128906, -0.2498321533203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000176.npy"}
{"epoch": 0.36858638743455496, "step": 177, "batch_size": 128, "mean": 11.52630615234375, "std": 30.550771713256836, "min": -85.71490478515625, "p10": -23.232177734374996, "median": 8.574177742004395, "p90": 51.51683654785156, "max": 84.03704833984375, "pos_frac": 0.6875, "sample": [37.46588134765625, -2.0201683044433594, 2.380645751953125, 62.80696105957031, -12.440483093261719, 53.03985595703125, 8.858699798583984, -20.8065185546875, 28.13861083984375, 5.821052551269531, 47.823699951171875, -33.05006790161133, -68.16546630859375, -11.696268081665039, 2.5769195556640625, -17.871826171875, 31.146991729736328, 5.02838134765625, 36.476898193359375, 6.895599365234375, 32.60748291015625, 21.055984497070312, 37.8662109375, 43.18463134765625, -19.151763916015625, 24.630569458007812, -25.467315673828125, 61.684112548828125, 43.424163818359375, 60.54327392578125, 5.2782745361328125, 22.04839324951172, -50.655731201171875, 69.8477783203125, -1.326446533203125, 24.887832641601562, 51.50531005859375, 2.9724273681640625, -9.380630493164062, 2.88134765625, 31.014190673828125, -22.274261474609375, -3.1266231536865234, 2.0368194580078125, -3.481231689453125, -20.820152282714844, 3.963695526123047, 11.392593383789062, 36.426307678222656, 35.62040710449219, -9.77777099609375, 1.773162841796875, 68.38897705078125, 19.873779296875, -2.7498931884765625, 51.577239990234375, 18.166717529296875, -4.657989501953125, 54.19258117675781, 3.218505859375, -30.968093872070312, 58.41607666015625, 0.00140380859375, -21.869873046875, 50.7763671875, -0.98486328125, -30.282745361328125, -5.8813323974609375, 5.910030364990234, 51.543731689453125, -9.719970703125, 76.3685302734375, -5.42852783203125, 7.930944442749023, -3.7439956665039062, 36.88798522949219, 36.055755615234375, -64.35714721679688, 21.25836181640625, 16.73126220703125, 20.362213134765625, -39.880950927734375, -34.415489196777344, 22.82940673828125, -49.0108642578125, 7.9188232421875, 26.152915954589844, 0.48956298828125, -13.000152587890625, 8.544647216796875, 16.045654296875, 47.242919921875, 27.623775482177734, 29.9698486328125, -6.562255859375, -14.074249267578125, 14.068084716796875, 9.20556640625, 44.029052734375, 4.992744445800781, -18.402923583984375, -3.583892822265625, 9.928131103515625, 34.2435302734375, 8.603708267211914, -59.108184814453125, 5.9394989013671875, 2.0345611572265625, -32.87847900390625, -85.71490478515625, 19.74346160888672, 33.36079406738281, 53.83699035644531, 26.94610595703125, 12.934310913085938, 48.146484375, 31.791015625, 23.626113891601562, 84.03704833984375, 9.20703125, 40.0311279296875, 1.2147445678710938, 0.201080322265625, -3.0022125244140625, 15.627685546875, 0.916595458984375, 27.811676025390625, 45.09862518310547], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000177.npy"}
{"epoch": 0.3706806282722513, "step": 178, "batch_size": 128, "mean": 18.767261505126953, "std": 29.050281524658203, "min": -62.60552978515625, "p10": -13.0168701171875, "median": 14.842044830322266, "p90": 58.95133056640625, "max": 108.43560791015625, "pos_frac": 0.75, "sample": [45.502777099609375, 13.4913330078125, 11.704048156738281, -5.906280517578125, -5.388313293457031, 6.684965133666992, 34.958984375, 40.03868865966797, 25.227214813232422, 27.425172805786133, 108.43560791015625, 44.58131408691406, 26.77397918701172, 17.137313842773438, -13.171234130859375, 65.57888793945312, 0.0, 47.22816467285156, 42.803619384765625, 14.018455505371094, 11.694183349609375, 68.01141357421875, 5.560722351074219, 56.95660400390625, 42.063934326171875, -12.950714111328125, -7.1748809814453125, 78.99566650390625, -12.806976318359375, 36.3248291015625, 3.505949020385742, -2.062225341796875, 0.32952880859375, 36.769622802734375, -50.83372497558594, 8.55029296875, 24.438995361328125, 51.82652282714844, -25.06988525390625, 41.059791564941406, 59.64697265625, 23.50408935546875, 31.74761962890625, 35.4078369140625, 55.7593994140625, 51.566131591796875, 4.509857177734375, -34.55186462402344, -2.233612060546875, 73.82833862304688, 42.512451171875, 72.26712036132812, -33.92022705078125, -62.60552978515625, -3.087921142578125, 10.749320983886719, 13.6800537109375, 34.546409606933594, -22.718368530273438, 20.6319580078125, 0.10774421691894531, 79.49130249023438, 6.153755187988281, 33.6409912109375, 48.16412353515625, 14.127124786376953, 23.91949462890625, 26.087677001953125, 19.019775390625, 6.117897033691406, 70.41329956054688, 16.900161743164062, -9.659473419189453, -1.226318359375, 10.866003036499023, 63.2247314453125, -27.70098876953125, 10.658905029296875, -7.608894348144531, 3.13623046875, 32.539306640625, 19.02825927734375, 33.65362548828125, -11.78515625, -33.80914306640625, -5.197441101074219, 58.6531982421875, -5.49945068359375, 13.2772216796875, 21.386138916015625, 8.341289520263672, 5.12028694152832, 0.2928466796875, 15.689483642578125, 15.58258056640625, 69.67282104492188, 8.98382568359375, 5.187477111816406, 0.914215087890625, 50.66021728515625, 8.158439636230469, -9.658275604248047, -18.15765380859375, 72.25946044921875, 18.800743103027344, -13.282516479492188, 10.489952087402344, 45.277679443359375, 48.29270935058594, -10.280364990234375, -4.305915832519531, 73.81304931640625, -24.27796173095703, 7.438568115234375, 32.72833251953125, 19.1651611328125, -15.78668212890625, 14.607307434082031, 29.437225341796875, 3.8951416015625, 15.0767822265625, 27.9249267578125, 24.510772705078125, 1.61175537109375, -5.26544189453125, 23.531402587890625, 36.99355697631836, 27.132049560546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000178.npy"}
{"epoch": 0.37277486910994767, "step": 179, "batch_size": 128, "mean": 13.885309219360352, "std": 27.909015655517578, "min": -51.729248046875, "p10": -23.444366455078125, "median": 10.881141662597656, "p90": 52.29017333984375, "max": 82.34817504882812, "pos_frac": 0.703125, "sample": [29.486968994140625, 0.39257049560546875, 52.9752197265625, 7.787384033203125, 9.920379638671875, 50.16749572753906, 8.048622131347656, 8.713947296142578, -37.652252197265625, 0.8623504638671875, 12.200958251953125, -1.258087158203125, 21.281402587890625, 4.2559814453125, 51.498199462890625, -23.323287963867188, 33.198974609375, 22.5985107421875, 6.808135986328125, -3.7043075561523438, 20.70947265625, 31.30780792236328, 4.757232666015625, 26.263931274414062, 16.74224853515625, -22.165021896362305, -31.381103515625, -5.854953765869141, -4.08624267578125, -37.13008117675781, 36.28864288330078, 7.35986328125, -33.670127868652344, 13.23876953125, -17.64324951171875, 33.35394287109375, 33.713104248046875, 53.195648193359375, 16.79925537109375, -4.983085632324219, -21.2021484375, -31.37872314453125, -1.7029495239257812, 20.433013916015625, -17.05059814453125, 30.449615478515625, 4.846900939941406, -7.736503601074219, 47.56103515625, -38.6239013671875, -18.823776245117188, -23.726882934570312, 21.765968322753906, 18.376541137695312, 8.72509765625, 62.539031982421875, 53.97248077392578, 46.534950256347656, 10.74365234375, -2.459625244140625, -51.729248046875, -10.217391967773438, 72.29608154296875, 9.239181518554688, -8.841690063476562, 30.88751220703125, 50.418792724609375, -10.557144165039062, -1.6416015625, 10.845062255859375, 9.073261260986328, 33.9503173828125, 17.6973876953125, 27.00714111328125, 0.1554718017578125, 6.848012924194336, 59.092559814453125, 9.486602783203125, 2.5814552307128906, 30.514244079589844, 3.972858428955078, 5.5833892822265625, 73.97943115234375, -12.350288391113281, 11.305892944335938, -34.12835693359375, 63.22834014892578, 51.99658203125, 0.6009025573730469, -28.616043090820312, 36.71936798095703, 9.239830017089844, 61.305419921875, 27.552490234375, 26.075469970703125, 14.9044189453125, 29.478424072265625, 30.589157104492188, 13.02490234375, 27.145370483398438, 64.96273803710938, 29.594772338867188, 51.175628662109375, 11.414894104003906, -32.493408203125, 38.083892822265625, -40.39784240722656, 5.9540863037109375, 21.068687438964844, 48.45024871826172, 13.67657470703125, 29.04437255859375, -23.956398010253906, 82.34817504882812, -14.2391357421875, -9.086669921875, 37.63184356689453, 38.49732971191406, 3.86175537109375, -8.802474975585938, 62.61541748046875, 36.88923645019531, -0.2537841796875, 10.917221069335938, -10.22454833984375, -15.72955322265625, 53.6934814453125, 39.591102600097656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000179.npy"}
{"epoch": 0.374869109947644, "step": 180, "batch_size": 128, "mean": 17.848716735839844, "std": 26.816219329833984, "min": -47.36090087890625, "p10": -13.120556640624997, "median": 15.007080078125, "p90": 54.679911804199214, "max": 79.39315795898438, "pos_frac": 0.734375, "sample": [-0.7316474914550781, -4.299709320068359, -27.40936279296875, -7.563873291015625, 3.0927963256835938, 21.3363037109375, 50.430511474609375, 6.53680419921875, -8.847747802734375, 49.18135070800781, -1.531829833984375, 35.7730712890625, 73.00088500976562, -43.77771759033203, 7.169464111328125, 47.518646240234375, -3.007080078125, 20.55157470703125, 7.484798431396484, -47.36090087890625, 37.408538818359375, 75.0426025390625, 26.24514389038086, 30.2158203125, 21.430633544921875, 38.727935791015625, 16.157928466796875, 47.08074951171875, 78.77409362792969, -7.627870559692383, -38.67903137207031, 5.2551727294921875, 51.7265625, 79.39315795898438, 61.34228515625, 9.731239318847656, 11.003448486328125, 8.926406860351562, 6.0814666748046875, 16.415283203125, 51.133544921875, -8.936477661132812, 34.387245178222656, 25.684734344482422, 70.48233032226562, 17.552032470703125, -18.545974731445312, -2.4760360717773438, 5.4776611328125, 0.9906463623046875, -20.4273681640625, -4.721343994140625, 3.4171295166015625, 24.637939453125, 59.3287353515625, 0.58709716796875, 16.82724380493164, 38.40262222290039, 9.59161376953125, 9.958747863769531, 66.054931640625, 31.396438598632812, -3.891693115234375, 75.35919189453125, 40.425201416015625, -15.539093017578125, 24.062034606933594, 27.3485107421875, -24.569183349609375, 18.490951538085938, 28.880859375, 7.896568298339844, 18.938011169433594, 55.236602783203125, 54.44132995605469, 53.249420166015625, -0.7208175659179688, -0.4096488952636719, 49.550750732421875, 32.32557678222656, 72.65188598632812, 9.378631591796875, -5.20953369140625, 38.55438232421875, 37.148948669433594, 31.492507934570312, 17.99531364440918, -17.18425750732422, 2.19329833984375, 30.515716552734375, 0.678131103515625, 14.145751953125, 1.4806804656982422, 27.150634765625, 41.39556884765625, 10.119583129882812, 18.144866943359375, 26.034393310546875, -0.4349517822265625, -5.62176513671875, 3.5744781494140625, 20.677215576171875, 52.539337158203125, 10.599700927734375, 5.747285842895508, -14.867698669433594, 4.0319366455078125, -15.258453369140625, 58.67193603515625, 4.732902526855469, -12.5447998046875, 31.430850982666016, -3.825183868408203, -11.59417724609375, 37.573341369628906, 27.979736328125, 25.77996826171875, -5.8585205078125, -15.236251831054688, 39.73991394042969, 1.2094669342041016, 19.99908447265625, 1.8314132690429688, 4.100147247314453, -8.298187255859375, 65.79025268554688, -14.4639892578125, 15.868408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000180.npy"}
{"epoch": 0.3769633507853403, "step": 181, "batch_size": 128, "mean": 16.60419464111328, "std": 28.53624725341797, "min": -55.52288818359375, "p10": -20.875672912597658, "median": 15.194225311279297, "p90": 50.2266860961914, "max": 93.38812255859375, "pos_frac": 0.734375, "sample": [0.9772663116455078, 6.307487487792969, 0.7665939331054688, 16.576690673828125, 0.0, 27.305526733398438, 48.984344482421875, 30.7005615234375, 8.98455810546875, 18.535606384277344, 58.32708740234375, -9.696949005126953, 37.32368469238281, -5.0291595458984375, 69.051513671875, 19.393280029296875, 0.0, 51.18357849121094, 31.591724395751953, -16.865982055664062, 38.72833251953125, 4.089691162109375, 49.81658935546875, 8.607208251953125, -55.3111572265625, 66.48713684082031, -12.4547119140625, 19.131450653076172, 5.494842529296875, 21.899993896484375, 44.50065994262695, 3.0079421997070312, 15.976696014404297, 46.53045654296875, 23.77020263671875, 31.099578857421875, 42.65240478515625, 15.952117919921875, 17.6773681640625, 27.74072265625, 15.029830932617188, 32.151123046875, -20.883331298828125, -5.557605743408203, 9.17340087890625, 11.343391418457031, 6.3648529052734375, 11.93377685546875, 73.51830291748047, -25.02032470703125, 25.097076416015625, -2.6046371459960938, 40.34742736816406, 48.11865234375, -13.537124633789062, -49.516845703125, -2.203502655029297, 13.47833251953125, -0.50323486328125, 9.302459716796875, 28.1077880859375, 1.90802001953125, -20.290771484375, 58.839508056640625, 25.627731323242188, 25.602008819580078, 14.360809326171875, 48.72737121582031, 2.23431396484375, 26.7550048828125, 13.92181396484375, 31.472999572753906, 15.358619689941406, 49.07383728027344, 81.72576904296875, 93.38812255859375, -40.7353515625, 10.103317260742188, -26.64923095703125, 47.841033935546875, -20.872390747070312, 11.917724609375, -5.132137298583984, -39.264312744140625, -7.4972991943359375, 45.53861999511719, 16.77508544921875, -6.776863098144531, 12.211761474609375, 8.80767822265625, -16.582260131835938, 22.55828857421875, 59.0667724609375, -21.158462524414062, 9.1761474609375, 24.993980407714844, 7.2736968994140625, 25.189834594726562, 17.360427856445312, 65.09710693359375, 59.7904052734375, -23.163177490234375, 20.185684204101562, 11.87677001953125, -25.79205322265625, -8.489830017089844, 7.57611083984375, 66.7960205078125, -11.15557861328125, -55.52288818359375, 45.59681701660156, 22.54278564453125, 49.037479400634766, -11.28271484375, 26.643539428710938, 65.63040161132812, -32.051544189453125, 13.309982299804688, 13.6705322265625, -5.543426513671875, 9.168869018554688, 30.533035278320312, 47.15576171875, -26.226760864257812, 47.48529815673828, 32.27374267578125, 31.43243408203125, 41.95611572265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000181.npy"}
{"epoch": 0.37905759162303665, "step": 182, "batch_size": 128, "mean": 20.58496856689453, "std": 24.46780776977539, "min": -51.683074951171875, "p10": -7.579636955261229, "median": 20.648574829101562, "p90": 53.089620971679686, "max": 74.65859985351562, "pos_frac": 0.8125, "sample": [30.71795654296875, 43.41316223144531, 7.2020263671875, -0.33513832092285156, 16.494155883789062, 10.697052001953125, 49.869422912597656, 32.023193359375, 18.89495849609375, 5.5959625244140625, 49.0406494140625, 24.88916015625, 21.32977294921875, 43.31927490234375, 30.314727783203125, -0.774261474609375, 27.982025146484375, 43.54754638671875, 49.85260009765625, 32.426597595214844, 33.185630798339844, 13.75851058959961, 34.57495880126953, 59.4644775390625, -9.65350341796875, 53.23760986328125, -5.8538818359375, 34.36114501953125, 70.6007080078125, 28.071807861328125, 41.512969970703125, 17.269912719726562, 34.238067626953125, 30.67864990234375, 13.85477066040039, -3.419057846069336, -11.448745727539062, 6.62701416015625, 4.890777587890625, 9.05224609375, -36.232269287109375, 16.163116455078125, -51.683074951171875, 28.399330139160156, 47.11212158203125, -7.1279449462890625, 61.117584228515625, -46.075439453125, 46.378326416015625, 74.65859985351562, 24.972747802734375, 46.211578369140625, 47.196533203125, 53.07891845703125, 44.100189208984375, -2.896575927734375, 9.505805969238281, 5.978233337402344, 7.8890380859375, 32.087554931640625, 24.855255126953125, 52.10404968261719, 19.967376708984375, 19.015396118164062, 27.365264892578125, 9.386520385742188, 29.936119079589844, 30.955596923828125, 73.00167846679688, 15.596290588378906, -13.398094177246094, 7.550031661987305, 27.7615966796875, 11.674392700195312, 0.89263916015625, 14.7037353515625, 27.371124267578125, 31.73222541809082, 5.429756164550781, 6.6314697265625, 8.659027099609375, -5.608642578125, 3.764251708984375, 36.40275573730469, -3.288970947265625, 0.254852294921875, 29.794189453125, 3.43731689453125, 13.63006591796875, 1.276641845703125, 53.45098876953125, 46.7646484375, 63.310028076171875, 55.66424560546875, -29.073226928710938, -8.633584976196289, 60.940093994140625, 35.53205108642578, 33.6522216796875, 24.94012451171875, -19.662918090820312, -3.2132568359375, 9.72125244140625, 11.840251922607422, -1.72772216796875, 30.683979034423828, 22.055206298828125, 25.113479614257812, 53.79718017578125, 9.22503662109375, -10.406036376953125, 40.36967468261719, 53.14996337890625, 1.9994525909423828, 0.9117584228515625, 5.7923583984375, 9.226188659667969, 48.357574462890625, 16.538719177246094, -22.686004638671875, -13.96807861328125, 48.489227294921875, 31.364501953125, 1.06640625, 53.114593505859375, -6.5673370361328125, 22.63916015625, -22.089317321777344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000182.npy"}
{"epoch": 0.381151832460733, "step": 183, "batch_size": 128, "mean": 16.932781219482422, "std": 26.974395751953125, "min": -48.5095100402832, "p10": -16.747874450683593, "median": 13.454946517944336, "p90": 49.67654571533203, "max": 81.63357543945312, "pos_frac": 0.7265625, "sample": [-20.26776123046875, 31.162841796875, -15.246551513671875, 40.79389953613281, -25.429794311523438, 56.52049255371094, -19.99432373046875, 11.499404907226562, 45.90834045410156, 21.582923889160156, -9.470947265625, 17.954185485839844, 27.117935180664062, 3.11431884765625, 12.51318359375, -41.53607177734375, -0.9575176239013672, 39.99641418457031, 12.61651611328125, 0.8824234008789062, 16.55670166015625, 13.713592529296875, 40.03031921386719, -16.281509399414062, 10.877128601074219, -26.014556884765625, 46.513671875, 13.089126586914062, 8.52264404296875, 9.079582214355469, -4.8253173828125, -29.034820556640625, 11.918609619140625, 9.328338623046875, 7.810646057128906, 40.49568176269531, 9.875320434570312, 53.03858947753906, 23.54632568359375, -17.8360595703125, 58.61029052734375, -5.782440185546875, 25.75334358215332, -13.70806884765625, 47.778411865234375, 23.86687469482422, 42.69456481933594, -12.774444580078125, 61.01116943359375, 31.374752044677734, -0.77874755859375, 40.601959228515625, 43.79579162597656, -33.968017578125, 0.0, 0.6309127807617188, -3.371063232421875, 10.219940185546875, -16.108657836914062, 43.42169189453125, 11.404975891113281, 58.91473388671875, 0.04017829895019531, 68.1119384765625, 47.74273681640625, -24.89373779296875, 10.799320220947266, 62.6964111328125, 36.686553955078125, -13.613571166992188, 31.531723022460938, 69.41329956054688, 81.63357543945312, 39.48552703857422, -1.6883087158203125, 21.10364532470703, -34.28474426269531, 22.686370849609375, -3.225799560546875, 29.13079833984375, 29.135650634765625, 19.9669189453125, 10.849090576171875, 13.196300506591797, -4.202323913574219, 9.8927001953125, 21.606292724609375, 32.11652374267578, 10.226760864257812, 10.82952880859375, -0.7933120727539062, 19.322509765625, 3.2758865356445312, 3.6169986724853516, 1.751617431640625, 45.543609619140625, 50.09062194824219, 49.49908447265625, 32.635955810546875, 0.45700836181640625, -48.5095100402832, 5.4292449951171875, -11.089370727539062, 17.11431884765625, -1.1948356628417969, 38.05232238769531, 0.7901725769042969, 48.89662170410156, -2.8983001708984375, 65.22320556640625, 21.973419189453125, 68.98712158203125, 35.5638427734375, 41.929954528808594, 21.114105224609375, -35.88262939453125, 16.30548095703125, 31.226470947265625, -32.256011962890625, 78.6448974609375, -5.913726806640625, 38.11292266845703, 28.441818237304688, 29.8546142578125, -7.212188720703125, 21.42688751220703, 42.60743713378906, 35.53253173828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000183.npy"}
{"epoch": 0.3832460732984293, "step": 184, "batch_size": 128, "mean": 24.853824615478516, "std": 29.838788986206055, "min": -52.8336181640625, "p10": -14.320629119873043, "median": 23.61715316772461, "p90": 62.04123229980468, "max": 88.58303833007812, "pos_frac": 0.796875, "sample": [35.742218017578125, 88.58303833007812, 57.30816650390625, 12.667221069335938, 13.022506713867188, -9.900299072265625, 16.449493408203125, -6.302162170410156, 53.886749267578125, 39.745635986328125, 20.6727294921875, 12.198486328125, -29.046051025390625, 14.840118408203125, -8.211105346679688, 1.2884979248046875, 74.85870361328125, 43.05908203125, 24.081199645996094, 38.04032897949219, 45.856781005859375, 59.99072265625, -52.8336181640625, 58.996185302734375, -5.24420166015625, 63.826202392578125, 40.260894775390625, -7.44677734375, -19.508926391601562, 72.24234008789062, 11.726043701171875, 68.8648681640625, 25.548065185546875, -13.480392456054688, 26.3447265625, 59.46820068359375, 3.423187255859375, 52.46186828613281, 39.38673400878906, 69.76275634765625, 23.153106689453125, 75.87139892578125, -42.46275329589844, 33.65218734741211, 27.52935791015625, 81.06777954101562, -37.842071533203125, 56.8896484375, -18.160614013671875, 36.91455078125, 17.084148406982422, -9.356048583984375, 25.063568115234375, 22.75933837890625, -18.230438232421875, 31.566375732421875, 44.16680908203125, 23.134384155273438, -16.28118133544922, 79.75103759765625, 6.233856201171875, 5.576372146606445, 33.18194580078125, -36.52655029296875, 44.84661865234375, 40.211334228515625, 52.47332763671875, 54.05073547363281, 20.93212890625, 40.87225341796875, 10.55072021484375, 11.614532470703125, 42.34324645996094, 5.63543701171875, 41.15625, 8.191215515136719, 21.799102783203125, 45.60784912109375, 33.23478698730469, 44.25830078125, 61.2762451171875, 35.10137939453125, -3.3964080810546875, 54.46592712402344, 6.136993408203125, -19.451446533203125, 4.09521484375, 66.78330993652344, 4.206512451171875, 17.325714111328125, 8.530441284179688, 46.472198486328125, 18.886978149414062, 13.328628540039062, 46.302215576171875, -2.593170166015625, -17.73516273498535, 14.504623413085938, 1.1561546325683594, 59.98358154296875, 38.47772216796875, 40.958038330078125, 25.815521240234375, 55.240753173828125, 0.888702392578125, 1.9977455139160156, 47.84782409667969, 20.5518798828125, -9.460166931152344, 69.80805969238281, 21.597030639648438, -25.61492919921875, 45.075042724609375, 16.641769409179688, 40.434410095214844, -4.1689300537109375, 65.67991638183594, 5.1734466552734375, -9.702682495117188, 40.80103302001953, 0.089263916015625, -21.01544189453125, 60.51165771484375, 16.90277099609375, 80.25393676757812, 58.28814697265625, -7.50421142578125, 35.19694519042969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000184.npy"}
{"epoch": 0.38534031413612563, "step": 185, "batch_size": 128, "mean": 16.943790435791016, "std": 28.845203399658203, "min": -49.50592041015625, "p10": -18.128268432617187, "median": 13.559967041015625, "p90": 54.50457305908203, "max": 83.44964599609375, "pos_frac": 0.6640625, "sample": [36.556541442871094, -23.461822509765625, 41.89208984375, -12.988174438476562, -6.161346435546875, 61.574798583984375, 40.379486083984375, 54.28086853027344, 49.42242431640625, -6.3417510986328125, 22.740692138671875, 8.271644592285156, 49.488677978515625, 38.42873001098633, 10.388206481933594, 69.83004760742188, 34.903900146484375, -11.177978515625, 7.543220520019531, 51.203094482421875, 25.260818481445312, 5.14666748046875, 39.51175308227539, 36.06089782714844, 23.8050537109375, 12.81695556640625, 49.05157470703125, 6.737968444824219, 2.3309879302978516, 62.873077392578125, -49.50592041015625, 41.40410614013672, 1.4389114379882812, -5.035797119140625, 16.32970428466797, 36.712921142578125, 20.64312744140625, 25.041160583496094, 22.8701171875, 55.1553955078125, -6.941253662109375, 22.44573974609375, -17.29827880859375, 34.43701171875, 3.291961669921875, 63.196502685546875, -8.120025634765625, 30.29918098449707, 23.21734619140625, 49.4769287109375, -27.586288452148438, -22.959609985351562, -30.061126708984375, -13.410308837890625, 10.031997680664062, 0.318359375, 9.263320922851562, 0.0, 7.511085510253906, 39.167076110839844, -12.360763549804688, -17.742218017578125, 8.744667053222656, -3.2663650512695312, 28.171127319335938, 28.305320739746094, 19.291702270507812, 14.048095703125, 31.93438720703125, -2.6339893341064453, -7.84747314453125, 46.506011962890625, 42.597930908203125, 47.77006530761719, -7.236362457275391, -0.097625732421875, -1.744781494140625, 13.07183837890625, 15.843318939208984, 47.64717102050781, 83.44964599609375, -5.7112579345703125, 8.822280883789062, 0.0, -9.95770263671875, 39.5769157409668, 55.02655029296875, -19.029052734375, 69.89306640625, -20.866912841796875, 36.915565490722656, 33.764404296875, -11.342170715332031, -2.515096664428711, 38.46527099609375, 65.7322998046875, -29.234092712402344, 35.13377380371094, 22.950790405273438, 9.531494140625, -7.702812194824219, 22.418319702148438, 32.77244567871094, -12.535003662109375, 73.01190185546875, -39.947784423828125, -7.7373809814453125, 6.704109191894531, 77.762939453125, 80.9249267578125, 46.23321533203125, 11.780487060546875, 12.017898559570312, -6.1175537109375, 50.03656005859375, -43.792083740234375, 29.533050537109375, 68.58465576171875, 26.288055419921875, -23.11981201171875, 29.034072875976562, -8.782394409179688, -14.540084838867188, -12.458343505859375, -21.76448631286621, -32.417701721191406, 10.97637939453125, 18.331146240234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000185.npy"}
{"epoch": 0.387434554973822, "step": 186, "batch_size": 128, "mean": 16.574623107910156, "std": 26.357868194580078, "min": -53.2706298828125, "p10": -8.621977996826171, "median": 13.448543548583984, "p90": 55.619769287109364, "max": 76.03857421875, "pos_frac": 0.7265625, "sample": [21.676254272460938, -1.0991363525390625, 11.81085205078125, 31.93731689453125, 18.989501953125, 63.61871337890625, 23.65087890625, -0.7645263671875, 12.294158935546875, 11.727460861206055, 12.960624694824219, 11.827056884765625, 7.2735595703125, 34.611724853515625, -16.1661376953125, 27.39556884765625, 28.246551513671875, 1.7421836853027344, 59.888214111328125, 24.188888549804688, 11.120025634765625, -0.30828094482421875, -5.746063232421875, 3.2481536865234375, 3.7015914916992188, 40.00102233886719, 17.145751953125, 23.4158935546875, -1.173614501953125, 0.619537353515625, 7.001033782958984, 28.16168212890625, 72.93235778808594, 2.199554443359375, -7.2172393798828125, -39.739784240722656, 23.70904541015625, 64.32049560546875, 23.075927734375, 22.701141357421875, 0.9470291137695312, 46.7677001953125, 42.88395690917969, 35.99605941772461, 33.461483001708984, 36.20458984375, -25.7529296875, 4.3154296875, 37.40122985839844, 23.7752685546875, 15.7288818359375, 17.454376220703125, 40.17955017089844, 6.2803955078125, -11.873458862304688, -17.182373046875, 36.997344970703125, 17.345565795898438, -0.3912353515625, 3.6259765625, -5.319797515869141, -6.202117919921875, 60.8680419921875, -33.264739990234375, 54.09471130371094, -2.9725341796875, 26.225494384765625, 2.5003662109375, -2.715240478515625, 59.17823791503906, -3.7259140014648438, 16.61431884765625, 76.03857421875, 29.724273681640625, -8.519134521484375, 5.937774658203125, 42.117095947265625, -53.2706298828125, 1.5350208282470703, 62.6180419921875, 18.2301025390625, -3.2408447265625, 6.82318115234375, 74.71722412109375, 41.517822265625, -33.37432861328125, 25.271881103515625, -8.559600830078125, 6.820709228515625, -32.222900390625, 10.905838012695312, 14.6639404296875, 29.87303924560547, -4.086639404296875, 72.47174072265625, 69.75973510742188, 23.20459747314453, 13.93646240234375, -52.46356201171875, 1.6504096984863281, 28.767475128173828, 47.71558380126953, 3.4098663330078125, -5.48297119140625, 37.770843505859375, 41.4384765625, 22.314132690429688, -2.0049819946289062, 0.34096527099609375, 31.19510269165039, 35.27056121826172, -8.767524719238281, 26.9798583984375, 1.1160545349121094, 68.1190185546875, 0.0, 74.71673583984375, 35.153202056884766, -12.77154541015625, 34.7674560546875, -15.722171783447266, -3.622272491455078, 0.29441070556640625, 2.8372726440429688, 27.233306884765625, -7.287117004394531, 44.456268310546875, -3.18988037109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000186.npy"}
{"epoch": 0.38952879581151834, "step": 187, "batch_size": 128, "mean": 17.204362869262695, "std": 31.366138458251953, "min": -56.082763671875, "p10": -22.649703979492184, "median": 16.04633331298828, "p90": 59.12242736816406, "max": 85.2657470703125, "pos_frac": 0.6796875, "sample": [38.4677734375, 51.31878662109375, 25.6798095703125, 42.2430419921875, -46.5457763671875, 18.394256591796875, 68.46063232421875, 2.797800064086914, -27.21428680419922, 78.5396728515625, 58.978607177734375, 0.4306793212890625, -12.00994873046875, -2.2668304443359375, 53.72697448730469, 68.31021118164062, 40.06315612792969, 17.017311096191406, 54.16545104980469, 29.59844970703125, 32.652313232421875, -15.356536865234375, 35.605438232421875, -2.0245513916015625, -19.130615234375, -24.173583984375, 39.74493408203125, -26.62176513671875, -21.996612548828125, -39.60723876953125, -2.034637451171875, 57.50904846191406, 27.8759765625, 10.855316162109375, -19.754234313964844, 34.38020324707031, -9.657135009765625, -7.5413818359375, 33.18412780761719, 42.60760498046875, 17.571701049804688, -4.309295654296875, 70.31393432617188, 13.31005859375, -8.594482421875, 33.613555908203125, 63.342498779296875, 4.32733154296875, -14.100860595703125, 29.7564697265625, 0.460906982421875, 28.094528198242188, 14.883979797363281, 41.061248779296875, -56.082763671875, -10.555572509765625, 39.405426025390625, -19.127685546875, 41.50801467895508, 64.27409362792969, 32.34733581542969, -5.546875, 42.938934326171875, -39.886260986328125, 85.2657470703125, 54.11302185058594, -38.090675354003906, -43.91790771484375, 10.330535888671875, 30.78912353515625, 63.7996826171875, -37.13167953491211, -4.8336334228515625, -17.923004150390625, -9.28851318359375, -26.325469970703125, 49.69789123535156, 11.279197692871094, 57.01715087890625, 6.04473876953125, -0.8370361328125, 12.858505249023438, -33.074859619140625, 3.3198013305664062, 11.538528442382812, -1.9198684692382812, 15.27471923828125, 6.30438232421875, -40.61004638671875, 36.6446533203125, 26.024398803710938, 53.324981689453125, 51.63154602050781, 28.797653198242188, 29.147239685058594, 65.59083557128906, 1.0126953125, 45.539703369140625, 44.290679931640625, 31.069358825683594, -9.40728759765625, 36.02435302734375, 78.2249755859375, 8.29827880859375, 2.6573829650878906, 56.03619384765625, -0.601043701171875, 17.890167236328125, -17.129241943359375, 64.83840942382812, 42.9813232421875, 25.4970703125, 5.89404296875, 16.817947387695312, 7.583427429199219, 50.005157470703125, 62.74287414550781, 0.482696533203125, 1.830240249633789, 5.6772003173828125, -10.84490966796875, 33.78367233276367, 20.59880828857422, -6.959938049316406, 0.0, 59.4580078125, -11.127960205078125, 18.473831176757812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000187.npy"}
{"epoch": 0.39162303664921466, "step": 188, "batch_size": 128, "mean": 17.772464752197266, "std": 30.371076583862305, "min": -66.95223999023438, "p10": -16.516160583496095, "median": 15.658086776733398, "p90": 56.69785919189452, "max": 81.5469970703125, "pos_frac": 0.71875, "sample": [3.178285598754883, 26.634552001953125, 51.32366943359375, 0.0, 19.284992218017578, 4.612762451171875, 5.2640380859375, 62.784210205078125, -6.168598175048828, 38.213294982910156, 44.16448974609375, 1.2226638793945312, 71.13473510742188, -7.117023468017578, 23.568206787109375, -12.32025146484375, 58.1463623046875, 20.099761962890625, 11.449302673339844, 78.5694580078125, 63.931884765625, 42.639923095703125, -11.625640869140625, 0.3538398742675781, -7.9627227783203125, -16.5504150390625, -21.249618530273438, 9.089950561523438, -19.403854370117188, 41.82099151611328, 54.02008056640625, -10.171737670898438, 43.566162109375, 20.599609375, 3.9652557373046875, 31.98876953125, 40.38275146484375, 54.407440185546875, 47.14558792114258, 18.767822265625, 11.12384033203125, 47.163299560546875, 7.756767272949219, 44.846221923828125, 71.05841064453125, -53.7506103515625, 14.727813720703125, 46.73565673828125, 28.8096923828125, 24.755691528320312, 7.188690185546875, 41.80168151855469, -16.501480102539062, -3.21368408203125, 27.27581787109375, 67.88943481445312, 81.5469970703125, 6.603427886962891, 38.153045654296875, 70.882568359375, 35.18499755859375, 9.862258911132812, 30.532482147216797, 50.28497314453125, 6.162471771240234, 1.252685546875, 7.3727569580078125, 56.07707214355469, 20.807899475097656, 32.985687255859375, -3.91717529296875, 51.27508544921875, 77.36257934570312, -1.2226676940917969, -10.56512451171875, 1.6759490966796875, 48.47198486328125, 64.65008544921875, -6.4755401611328125, 12.135604858398438, -36.486968994140625, 42.3363037109375, 31.121910095214844, 52.523162841796875, 0.3117523193359375, -6.27349853515625, 27.802581787109375, 44.24762725830078, 8.747039794921875, 0.24664306640625, -66.95223999023438, 22.158370971679688, -37.02299499511719, -26.7313232421875, 21.745803833007812, -10.717987060546875, -12.893157958984375, -26.015472412109375, 32.68383026123047, 25.98149871826172, 23.370391845703125, 1.5294265747070312, -30.434341430664062, -62.350921630859375, 44.1041259765625, 29.989280700683594, 31.587608337402344, 47.87139892578125, 63.510101318359375, -6.408672332763672, -37.722076416015625, 0.9894371032714844, 21.881393432617188, 11.647537231445312, 55.500091552734375, -11.206756591796875, 62.624542236328125, -4.337162017822266, -0.3014984130859375, -8.621658325195312, -2.1316680908203125, 12.193878173828125, 0.0, 7.745208740234375, 23.389495849609375, 16.588359832763672, 3.1834487915039062, -34.65673828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000188.npy"}
{"epoch": 0.393717277486911, "step": 189, "batch_size": 128, "mean": 20.662933349609375, "std": 24.804555892944336, "min": -36.86541748046875, "p10": -6.8270391464233375, "median": 17.611507415771484, "p90": 55.26000442504882, "max": 78.73480224609375, "pos_frac": 0.78125, "sample": [36.646026611328125, 14.142974853515625, -28.425018310546875, 9.854171752929688, 9.29742431640625, 11.437362670898438, 32.28150177001953, 74.50509643554688, 34.572181701660156, 19.681884765625, 20.399620056152344, 37.591827392578125, 67.08016967773438, 30.73565673828125, 46.091217041015625, 36.13067626953125, 52.033660888671875, -2.722341537475586, -12.728668212890625, 12.93963623046875, 54.48870849609375, 48.97479248046875, -13.30035400390625, 70.44723510742188, 28.159759521484375, 12.836044311523438, 31.3365478515625, 31.196807861328125, 2.3708839416503906, -4.900627136230469, -22.57525634765625, 1.0326766967773438, 20.586021423339844, 32.690345764160156, 3.9254837036132812, 24.09882354736328, 14.286674499511719, 1.9114017486572266, 10.06512451171875, 19.348236083984375, 2.1035385131835938, 60.950439453125, 48.55230712890625, 40.698333740234375, 45.862548828125, 31.230926513671875, 17.61328125, -17.63653564453125, 52.23631286621094, -22.50323486328125, -26.6395263671875, -5.28125, -17.9158935546875, 36.97784423828125, 4.042724609375, 17.74591064453125, 35.8380126953125, 1.6373748779296875, 8.413818359375, 11.351318359375, 48.397979736328125, 19.758377075195312, 34.838409423828125, -3.6536865234375, 7.07757568359375, 13.402656555175781, 17.75128746032715, 6.870628356933594, 5.534934997558594, 10.740684509277344, 16.541610717773438, 7.944526672363281, -11.454410552978516, 43.62945556640625, -2.03485107421875, 0.9038963317871094, 13.727386474609375, 3.9500961303710938, -3.7060546875, 13.455169677734375, 24.288101196289062, 25.4254150390625, 17.567901611328125, -5.30133056640625, 30.579299926757812, 31.92664337158203, 64.44253540039062, -6.262908935546875, -1.3592109680175781, -1.0062217712402344, 5.842864990234375, 60.43658447265625, 18.822601318359375, -36.86541748046875, 1.281341552734375, 0.0, 78.73480224609375, 17.347824096679688, 26.490127563476562, 64.10450744628906, 48.91656494140625, 39.566314697265625, 54.60155487060547, 17.60973358154297, 23.67914581298828, 0.0, 13.766998291015625, 73.9969482421875, -2.433837890625, 56.79638671875, 39.29844665527344, 9.949600219726562, 40.013763427734375, 8.327590942382812, 21.975067138671875, -15.212417602539062, 39.19903564453125, 58.12181091308594, 44.582305908203125, 27.976303100585938, 23.72198486328125, 71.19427490234375, -9.238914489746094, 68.12100219726562, 48.9927978515625, -0.2166748046875, -4.2809906005859375, -8.143342971801758], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000189.npy"}
{"epoch": 0.3958115183246073, "step": 190, "batch_size": 128, "mean": 19.982906341552734, "std": 29.025943756103516, "min": -56.584930419921875, "p10": -13.44011535644531, "median": 19.999099731445312, "p90": 57.681356811523436, "max": 94.81207275390625, "pos_frac": 0.7578125, "sample": [-21.85785675048828, 13.7855224609375, 76.1466064453125, 31.56024169921875, -1.3100128173828125, 14.616853713989258, -40.31013488769531, -3.0640735626220703, -17.88275146484375, 1.297515869140625, 37.17005920410156, -41.781455993652344, -16.001541137695312, 57.1707763671875, 39.889373779296875, -1.1644439697265625, 6.41265869140625, -11.456298828125, 19.603485107421875, -44.40576171875, -7.15412712097168, -8.734443664550781, -25.495986938476562, 48.19844055175781, 42.07848358154297, 6.1291351318359375, 43.79595947265625, -1.31939697265625, 30.9649658203125, 63.919891357421875, 85.19491577148438, -12.342361450195312, 20.299560546875, 18.28974151611328, 42.511077880859375, -4.8353424072265625, 51.337738037109375, 11.188041687011719, 40.87377166748047, 27.55023193359375, 52.11773681640625, -56.584930419921875, 7.9193115234375, 75.82087707519531, 44.897178649902344, 23.0635986328125, 46.31239318847656, 15.823837280273438, 58.872711181640625, 71.64373779296875, 15.482528686523438, 4.182991027832031, 36.59101104736328, -1.6840438842773438, 23.604812622070312, 25.023162841796875, 4.878755569458008, -5.989166259765625, 7.842529296875, -7.510307312011719, -2.4375, 46.492576599121094, 24.1641845703125, 14.507797241210938, 34.93598937988281, 26.785888671875, 19.799072265625, -24.548553466796875, 17.317962646484375, 41.818328857421875, 39.35039520263672, 24.391109466552734, 33.9141845703125, 0.0, 52.507568359375, 55.04014587402344, 24.690799713134766, 28.099990844726562, 66.56796264648438, 14.466495513916016, 0.77044677734375, 8.004180908203125, 27.771072387695312, 33.206275939941406, 51.98602294921875, 19.30572509765625, 18.37580108642578, 32.40708923339844, 43.1966552734375, 6.234954833984375, 16.058181762695312, 94.81207275390625, 37.559967041015625, 60.93353271484375, -1.9806442260742188, -52.04498291015625, 11.184646606445312, -7.352935791015625, 53.19786834716797, -23.680206298828125, 40.49609375, 19.410552978515625, 25.791259765625, 59.286590576171875, 34.8033447265625, 34.86846923828125, 13.300666809082031, 7.086334228515625, 1.384124755859375, 28.722915649414062, 3.863496780395508, 72.78396606445312, 23.284772872924805, 33.97784423828125, 5.46824836730957, 21.807281494140625, -5.20684814453125, 35.66444778442383, -31.110565185546875, 20.199127197265625, 24.67169189453125, 0.6502132415771484, 4.506660461425781, 34.2882194519043, -46.410400390625, 61.04547119140625, -8.012237548828125, 60.20262145996094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000190.npy"}
{"epoch": 0.39790575916230364, "step": 191, "batch_size": 128, "mean": 16.84998321533203, "std": 29.941240310668945, "min": -81.971923828125, "p10": -17.214535331726072, "median": 11.289421081542969, "p90": 56.65752716064453, "max": 81.23336791992188, "pos_frac": 0.7578125, "sample": [57.22874450683594, 34.082122802734375, -16.995208740234375, 0.3767547607421875, 32.548179626464844, 69.56024169921875, 10.9732666015625, 13.866867065429688, 3.6533737182617188, -19.4901123046875, 5.012550354003906, 30.74591064453125, 16.403955459594727, 5.1342315673828125, 1.397125244140625, -7.10498046875, 20.521257400512695, 0.86627197265625, 1.4434185028076172, 17.463363647460938, 24.604652404785156, 52.016510009765625, 8.915996551513672, 1.6852493286132812, 0.12876510620117188, 11.12066650390625, -5.37103271484375, -50.54534912109375, 47.50274658203125, 33.00413513183594, 8.742134094238281, 2.9733638763427734, -10.7335205078125, 44.156890869140625, 4.1771240234375, -26.16864013671875, 34.51380157470703, -6.6470947265625, -81.971923828125, -15.322601318359375, 21.263511657714844, 50.205535888671875, 80.86842346191406, 11.458175659179688, -0.5930900573730469, 74.52799987792969, 11.75286865234375, -12.297576904296875, 4.852132797241211, 7.650177001953125, 5.94024658203125, 43.14385986328125, 78.72732543945312, 60.11235046386719, 10.993759155273438, 6.0089569091796875, 13.116382598876953, 59.10694885253906, 17.4219970703125, 9.910987854003906, 5.761724472045898, -3.919189453125, 26.456697463989258, -13.919456481933594, 51.907440185546875, 4.908823013305664, 74.0516357421875, 36.368064880371094, 18.0672607421875, 2.8172454833984375, 3.195404052734375, 31.821876525878906, 8.99609375, -23.5616455078125, 2.4391746520996094, -44.61505126953125, 21.742523193359375, 17.926513671875, 44.176239013671875, -16.649276733398438, 1.83355712890625, 49.7113037109375, 1.1915664672851562, 45.277557373046875, -29.501922607421875, 78.31689453125, 56.4127197265625, -24.35675048828125, 41.026885986328125, 62.3017578125, 42.0157470703125, 50.26275634765625, 10.663909912109375, 2.2392959594726562, -1.2476806640625, -12.590621948242188, 41.346221923828125, 25.992462158203125, 71.1114501953125, 25.35748291015625, 26.7738037109375, 8.280075073242188, 52.552371978759766, 32.039512634277344, 81.23336791992188, 50.795440673828125, 4.315391540527344, -4.5672607421875, -37.37786865234375, 14.972259521484375, -3.8510894775390625, -45.60154724121094, 20.64852523803711, 24.368881225585938, 43.12933349609375, 46.80023193359375, -2.602325439453125, -4.760982513427734, -20.37744140625, 55.7852783203125, 27.081268310546875, 12.448486328125, 12.402748107910156, -25.58935546875, 0.0, 42.80908203125, -17.72629737854004, 68.8392333984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000191.npy"}
{"epoch": 0.4, "step": 192, "batch_size": 128, "mean": 14.962003707885742, "std": 30.166566848754883, "min": -81.6358642578125, "p10": -21.012431335449214, "median": 12.377668380737305, "p90": 56.962274169921876, "max": 79.80340576171875, "pos_frac": 0.7109375, "sample": [13.396453857421875, 37.947967529296875, 58.61128234863281, -36.61883544921875, 55.763916015625, 13.555694580078125, -17.491973876953125, 12.788253784179688, -10.874198913574219, 62.76318359375, 29.15179443359375, 16.879791259765625, 41.05859375, 32.02435302734375, 57.0350341796875, -5.7180633544921875, -34.43994140625, 39.251708984375, -81.6358642578125, 64.06463623046875, -53.89482116699219, 56.93109130859375, -8.277359008789062, -39.839664459228516, 29.399505615234375, 47.706390380859375, 0.05606842041015625, 8.277542114257812, 3.1108932495117188, 40.39390563964844, -62.492919921875, 23.259315490722656, 36.76173400878906, 47.61895751953125, -0.523345947265625, 2.2454280853271484, -9.642044067382812, 44.040000915527344, 7.239898681640625, 11.497802734375, 3.7532501220703125, -17.739013671875, -9.509567260742188, 45.549224853515625, 52.158050537109375, -32.89673614501953, 10.5867919921875, -38.6131591796875, 0.0, 0.0, 2.2249832153320312, -2.156005859375, -33.37501907348633, 24.75146484375, -17.70770263671875, 41.462501525878906, 3.8882598876953125, 38.621612548828125, -8.270263671875, -9.95147705078125, 24.510581970214844, 19.760873794555664, -23.936920166015625, -4.99102783203125, 47.7630615234375, 46.68804931640625, 33.56608581542969, 57.91729736328125, -0.16437149047851562, 48.66700744628906, -18.64727783203125, 4.963623046875, 19.80560302734375, -4.34893798828125, 22.151634216308594, 33.328758239746094, 30.1376953125, 64.922607421875, 59.851837158203125, 8.137062072753906, 7.917362213134766, 55.358978271484375, -1.6804580688476562, -27.89208984375, -19.759078979492188, 13.996307373046875, 12.759639739990234, 1.15557861328125, 38.926910400390625, 31.624603271484375, 0.9014892578125, 15.93017578125, -31.361312866210938, 79.80340576171875, 8.493179321289062, 27.675201416015625, 0.734405517578125, -6.606864929199219, 63.648895263671875, 38.0379638671875, 9.478401184082031, 15.82861328125, 5.819305419921875, -8.189632415771484, 1.42138671875, 9.48822021484375, 35.826904296875, 8.568710327148438, 77.37619018554688, 28.491098403930664, 41.872032165527344, -6.2325439453125, 32.38965606689453, 15.35455322265625, 20.98333740234375, 9.322456359863281, 44.6829833984375, 9.714008331298828, -41.204376220703125, 0.54364013671875, 2.6114349365234375, 66.8499755859375, 27.555145263671875, 11.995697021484375, 14.961898803710938, 58.615142822265625, -3.406017303466797, 60.51130676269531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000192.npy"}
{"epoch": 0.40209424083769635, "step": 193, "batch_size": 128, "mean": 16.149559020996094, "std": 28.24808120727539, "min": -43.23828125, "p10": -22.80376358032226, "median": 16.549808502197266, "p90": 53.321652221679685, "max": 73.67935180664062, "pos_frac": 0.6953125, "sample": [50.24549865722656, -4.8162994384765625, -2.69635009765625, -4.6374664306640625, -8.275650024414062, 43.7437744140625, 21.00855255126953, 47.602630615234375, 18.33856201171875, 25.53411865234375, 22.371795654296875, 2.317293167114258, 15.1668701171875, 46.5035400390625, -7.27435302734375, 45.15007019042969, -3.149169921875, 6.9374542236328125, -0.09716796875, 3.2458419799804688, 50.0115966796875, -16.149169921875, 41.15019226074219, 7.687713623046875, 23.573074340820312, 27.039276123046875, 11.227386474609375, -31.734481811523438, 43.008544921875, -3.728515625, -24.545021057128906, 2.350128173828125, 29.15753173828125, 10.94964599609375, 38.866634368896484, -9.595550537109375, 62.170654296875, -14.4029541015625, 0.8082637786865234, 20.811180114746094, -35.91058349609375, -6.65924072265625, 17.535491943359375, 36.491424560546875, 3.7318649291992188, 19.54046630859375, 57.173736572265625, 27.60101318359375, 53.44989013671875, 48.26298522949219, -4.8745880126953125, 9.897979736328125, 42.781105041503906, 6.588901519775391, 41.6300048828125, 3.4375076293945312, 65.29935455322266, 25.3349609375, -12.80023193359375, 30.40234375, -4.931915283203125, 25.604576110839844, 42.27387237548828, -15.168685913085938, 73.67935180664062, -10.277587890625, 18.673141479492188, 36.628173828125, 17.233062744140625, -7.10052490234375, -41.04986572265625, -43.23828125, 17.603607177734375, 73.27960205078125, -28.083045959472656, 45.63812255859375, 53.266693115234375, 61.805938720703125, 17.3778076171875, 66.39794921875, 46.370086669921875, 23.51336669921875, 48.434326171875, -32.97743225097656, 67.4566650390625, 56.06085205078125, 8.642822265625, 1.6457061767578125, -1.5594635009765625, 17.67950439453125, 31.18798828125, -2.5504379272460938, 64.29989624023438, 46.05804443359375, 16.931678771972656, -1.200225830078125, 18.63416290283203, 7.119789123535156, 20.30841064453125, 46.46223449707031, 40.807891845703125, 59.01466369628906, 10.122110366821289, -25.240219116210938, 19.61724853515625, -19.492835998535156, 67.423828125, -25.831390380859375, -41.06358337402344, 3.42584228515625, -3.45562744140625, 8.3406982421875, -37.27044677734375, 27.2913818359375, 16.167938232421875, 13.711761474609375, 52.417724609375, -26.83990478515625, 12.7362060546875, -5.972808837890625, -3.875762939453125, 0.122772216796875, -37.41461181640625, 9.148162841796875, 30.721771240234375, -22.057510375976562, 1.61700439453125, 46.051361083984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000193.npy"}
{"epoch": 0.4041884816753927, "step": 194, "batch_size": 128, "mean": 12.385601043701172, "std": 27.979183197021484, "min": -50.48493957519531, "p10": -18.068737030029297, "median": 7.062713623046875, "p90": 52.44451141357422, "max": 86.81784057617188, "pos_frac": 0.6328125, "sample": [86.81784057617188, -0.6507568359375, 27.77520751953125, -34.64808654785156, 18.4862060546875, 66.522705078125, 25.972808837890625, -18.479156494140625, 1.05072021484375, -0.163055419921875, 45.06843566894531, -18.068496704101562, -2.857522964477539, 4.5762481689453125, 10.167083740234375, 0.7300167083740234, -4.591987609863281, 24.219985961914062, 59.448944091796875, -6.4067535400390625, 14.848663330078125, -4.516582489013672, 20.690753936767578, 3.18963623046875, 10.600616455078125, -0.8834991455078125, 1.469085693359375, 19.20880126953125, -18.417266845703125, -35.72979736328125, -10.019683837890625, 18.884254455566406, -8.210552215576172, 12.323638916015625, 5.630645751953125, 19.248855590820312, 70.04425048828125, 34.19140625, 6.069122314453125, 44.723445892333984, -11.923370361328125, 71.25225830078125, -5.5992431640625, -17.6090087890625, -8.991836547851562, 46.62205505371094, -8.18216323852539, 53.196678161621094, 48.158294677734375, 7.3009185791015625, -11.130958557128906, 20.0040283203125, 26.657196044921875, 29.557647705078125, 75.394775390625, 25.417922973632812, 9.694023132324219, 20.58282470703125, 17.025802612304688, -6.805511474609375, 13.57281494140625, 11.7833251953125, -3.45703125, 38.56626892089844, -37.0772705078125, -10.4947509765625, 1.458099365234375, -49.469390869140625, 24.393798828125, -12.69915771484375, 71.01373291015625, -3.463836669921875, -22.67559814453125, 63.36114501953125, -1.0717010498046875, 60.87518310546875, -1.6729583740234375, -16.116668701171875, 29.5394287109375, 44.41249084472656, -18.590850830078125, 11.4610595703125, 47.90399169921875, 52.122154235839844, -5.283786773681641, -39.7960205078125, 34.27745056152344, -1.3618698120117188, 44.86761474609375, 56.25518798828125, 30.700439453125, -35.518157958984375, 14.391326904296875, -7.364994049072266, 4.271942138671875, 8.06756591796875, 4.489437103271484, -50.48493957519531, 70.92608642578125, 6.2304840087890625, -4.673553466796875, 20.497802734375, 2.53631591796875, 8.257362365722656, -30.226043701171875, 11.454780578613281, -11.902099609375, 27.26568603515625, 3.6921234130859375, -12.591064453125, 14.398452758789062, 67.55038452148438, 37.719696044921875, 6.8245086669921875, 21.784530639648438, 21.068756103515625, 44.25762939453125, 2.0083255767822266, -18.069297790527344, 3.710205078125, -1.158203125, -5.718719482421875, 48.67913818359375, -16.1064453125, 2.0150489807128906, -17.2515869140625, 31.801788330078125, 30.250762939453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000194.npy"}
{"epoch": 0.406282722513089, "step": 195, "batch_size": 128, "mean": 11.845905303955078, "std": 28.3515682220459, "min": -60.639404296875, "p10": -17.673699951171862, "median": 5.8764190673828125, "p90": 49.435888671875, "max": 85.90277099609375, "pos_frac": 0.6015625, "sample": [-6.8916168212890625, -12.305992126464844, -1.3332672119140625, 46.69366455078125, 48.72650146484375, 39.63911437988281, -29.73223876953125, -0.57275390625, 36.29437255859375, 6.895915985107422, 30.782958984375, 9.015430450439453, 2.295520782470703, -6.855869293212891, -1.4879226684570312, 21.21844482421875, -6.62530517578125, -9.93743896484375, 11.675811767578125, 29.17974853515625, -8.909652709960938, 12.171630859375, 44.16102600097656, -6.1412200927734375, -14.5994873046875, -2.3983612060546875, 0.495452880859375, -0.39300537109375, -25.7161865234375, 4.504364013671875, -9.72821044921875, 39.03411865234375, -7.4462738037109375, -5.2916259765625, -0.1885986328125, 3.531768798828125, 57.92161560058594, 55.112213134765625, -0.628814697265625, 2.83013916015625, -24.84686279296875, -7.502593994140625, 11.098167419433594, 6.259490966796875, 49.52655029296875, 4.545501708984375, 66.4130859375, -39.442230224609375, -14.123687744140625, 12.782623291015625, -28.756393432617188, 28.525421142578125, 27.061447143554688, 35.63294982910156, -51.519317626953125, 17.82525634765625, -4.26007080078125, -46.207611083984375, 54.09759521484375, -3.665130615234375, 20.768814086914062, -60.639404296875, 2.1938934326171875, 37.708106994628906, -12.417167663574219, -36.637969970703125, -44.98640441894531, -11.509033203125, -2.134063720703125, 36.34381103515625, 50.751220703125, 26.84503173828125, 41.391326904296875, -2.8846282958984375, 27.10882568359375, 59.7659912109375, -2.8600921630859375, -2.12127685546875, 22.871307373046875, 12.589874267578125, 39.45172882080078, 41.372222900390625, 26.62884521484375, 33.9173583984375, -1.4798126220703125, -2.591461181640625, 2.3581771850585938, -31.639923095703125, -1.4676055908203125, 3.4833831787109375, -7.72589111328125, 85.90277099609375, 39.83447265625, 0.3613128662109375, 40.380516052246094, 25.413848876953125, 77.0494384765625, -3.570831298828125, 26.552108764648438, 66.20376586914062, -14.492485046386719, 30.312942504882812, 10.979225158691406, 28.680526733398438, 3.621490478515625, 8.344451904296875, 6.728057861328125, 46.573455810546875, 35.304901123046875, 51.8524169921875, 71.78762817382812, 15.0330810546875, -43.376953125, 13.41943359375, -13.333297729492188, 4.824424743652344, 42.949066162109375, -44.314117431640625, 5.49334716796875, 26.7022705078125, 11.07696533203125, -3.152374267578125, -0.3083953857421875, 31.106979370117188, 50.697242736816406, 36.617401123046875, 49.39703369140625, -7.271575927734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000195.npy"}
{"epoch": 0.4083769633507853, "step": 196, "batch_size": 128, "mean": 18.246681213378906, "std": 26.328556060791016, "min": -44.61700439453125, "p10": -11.705529022216798, "median": 14.061927795410156, "p90": 53.66029510498046, "max": 85.78546142578125, "pos_frac": 0.7578125, "sample": [1.0750999450683594, 7.440765380859375, 25.396873474121094, 1.3453369140625, 13.898483276367188, 8.132476806640625, -17.999725341796875, 10.413297653198242, 39.7486572265625, 0.3503875732421875, 3.4114761352539062, 56.808624267578125, 25.71728515625, 31.01996612548828, 69.57424926757812, -18.9171142578125, 4.776275634765625, 3.795989990234375, 11.937774658203125, 19.652618408203125, 21.467681884765625, 3.3146133422851562, 34.195831298828125, 48.24359130859375, 0.11532211303710938, 17.085250854492188, -2.5580596923828125, -0.7879314422607422, -31.787315368652344, 41.184722900390625, 80.13140869140625, 64.56285095214844, 37.110076904296875, 44.871337890625, 0.850677490234375, 34.27978515625, 5.881065368652344, -29.8017578125, 22.8004150390625, 29.866369247436523, -11.758438110351562, 62.4849853515625, 27.759429931640625, -8.920135498046875, 9.623653411865234, 9.089118957519531, 10.823333740234375, 71.99346923828125, 11.26483154296875, 25.1727294921875, -21.072616577148438, -0.8300914764404297, -0.6086273193359375, 2.6287803649902344, -0.6430511474609375, 55.720703125, 48.14544677734375, 3.256014823913574, 4.41644287109375, 20.02386474609375, 32.36328125, 8.137664794921875, 36.391876220703125, -0.281463623046875, 16.012161254882812, 55.63468933105469, -15.38241958618164, 18.385284423828125, 0.514434814453125, -13.071563720703125, 8.356048583984375, 15.773956298828125, 0.5859184265136719, 39.79407501220703, 27.948081970214844, 45.910736083984375, 1.0423583984375, 14.225372314453125, 34.945274353027344, 32.208709716796875, 19.137619018554688, -34.83641052246094, 20.981674194335938, -2.823680877685547, -12.0457763671875, 49.05096435546875, 39.4439697265625, 46.019287109375, 26.584869384765625, 2.04022216796875, 41.030914306640625, -44.61700439453125, 14.37530517578125, 75.20883178710938, 19.70892333984375, -5.53582763671875, -11.682853698730469, -8.78857421875, 53.2275390625, 51.91691589355469, 28.00482177734375, -10.302093505859375, 6.91790771484375, 85.78546142578125, 62.775177001953125, 5.810029983520508, -19.197818756103516, 45.915000915527344, 54.51026916503906, 48.811676025390625, 38.49885559082031, 73.2904052734375, 48.476318359375, 53.2960205078125, -5.20648193359375, -10.097427368164062, -3.426097869873047, 39.51336669921875, 23.386993408203125, 26.070159912109375, -1.242523193359375, -5.76807975769043, -34.533660888671875, 30.121627807617188, 7.880683898925781, 10.2222900390625, -2.2476806640625, 3.2441024780273438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000196.npy"}
{"epoch": 0.41047120418848165, "step": 197, "batch_size": 128, "mean": 23.795928955078125, "std": 28.237192153930664, "min": -53.4486083984375, "p10": -9.31586227416992, "median": 24.68834400177002, "p90": 57.60384826660156, "max": 84.69479370117188, "pos_frac": 0.8203125, "sample": [-31.94549560546875, 6.5230255126953125, -5.218505859375, 39.98265075683594, 47.283538818359375, -8.535881042480469, 41.99467468261719, 17.261825561523438, 7.189182281494141, 56.53546142578125, 50.02667236328125, 47.527069091796875, -11.135818481445312, 24.300994873046875, 39.195098876953125, -21.366043090820312, 84.69479370117188, 67.8296127319336, 23.417236328125, 68.04025268554688, 0.5882244110107422, 44.41803741455078, 65.98694610595703, 15.826995849609375, 25.703460693359375, 11.791839599609375, 51.24139404296875, 57.697265625, 58.473602294921875, 56.11238098144531, 48.334964752197266, 33.74638366699219, -31.144561767578125, 36.23604202270508, 24.06744384765625, 32.765342712402344, 43.54017639160156, 0.8927078247070312, 14.397933959960938, 63.64373779296875, 16.5380859375, 38.191162109375, -53.4486083984375, 57.563812255859375, -34.357025146484375, 19.33856201171875, 65.3782958984375, 0.7673568725585938, -3.77886962890625, -1.6834564208984375, 34.60664367675781, 23.837623596191406, 9.899139404296875, -13.255462646484375, 5.9441680908203125, 0.9312591552734375, 49.865447998046875, 10.428024291992188, 53.9124755859375, 44.541595458984375, 39.10148620605469, 82.43621826171875, 3.7593765258789062, 59.4373779296875, -29.272125244140625, -3.047893524169922, 40.63883972167969, 50.399986267089844, 25.075693130493164, 15.507537841796875, 15.630569458007812, 31.7333984375, -1.6283073425292969, 48.34660339355469, 2.5299835205078125, 36.61528015136719, 8.63775634765625, 19.41119384765625, -25.1080322265625, 3.51092529296875, 8.3619384765625, 10.741592407226562, 28.719757080078125, 40.63543701171875, -23.8824462890625, 46.24407958984375, 3.80810546875, 30.803207397460938, 38.99505615234375, 5.979156494140625, 37.309661865234375, -0.225433349609375, 55.880859375, 35.6949462890625, 39.99847412109375, 23.101776123046875, 19.56549072265625, -31.41046142578125, 9.521492004394531, -43.085845947265625, 8.95819091796875, 27.248489379882812, -4.7775115966796875, 45.41497802734375, 54.1258544921875, 30.691162109375, 25.8782958984375, 22.463546752929688, 54.56886291503906, 30.59002685546875, 79.47991943359375, 70.77980041503906, 7.31298828125, 3.8509521484375, 25.235069274902344, 15.996871948242188, 51.5557861328125, 0.6008567810058594, 45.85379409790039, 11.763504028320312, 0.0, 81.01165771484375, 20.396759033203125, 28.62506103515625, 29.819496154785156, -34.54095458984375, 33.508331298828125, -4.146400451660156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000197.npy"}
{"epoch": 0.41256544502617803, "step": 198, "batch_size": 128, "mean": 18.130714416503906, "std": 32.721893310546875, "min": -78.44331359863281, "p10": -15.766535186767578, "median": 16.161819458007812, "p90": 64.56896667480468, "max": 88.12188720703125, "pos_frac": 0.7421875, "sample": [47.138153076171875, 16.176544189453125, 7.23968505859375, -5.527008056640625, 39.21949768066406, 1.408416748046875, 43.780853271484375, 55.086669921875, 24.937591552734375, 16.1470947265625, -12.191970825195312, -8.286865234375, 28.823699951171875, 28.572113037109375, 9.775020599365234, -8.272705078125, 67.63833618164062, 39.17730712890625, 35.14593505859375, 8.978225708007812, 65.05673217773438, -60.056396484375, -41.48406982421875, 66.0794677734375, 81.31005859375, 88.12188720703125, -56.1297607421875, 11.03302001953125, 8.660934448242188, 16.29083251953125, 64.35992431640625, 45.84008026123047, 43.991455078125, 38.0125732421875, 8.7603759765625, 24.174758911132812, -27.677490234375, 83.15594482421875, -39.5615234375, 79.48812866210938, -1.799041748046875, 0.438690185546875, 46.433197021484375, 47.21567916870117, -16.018882751464844, 0.2446460723876953, 73.45501708984375, -18.527725219726562, -12.469146728515625, 6.760227203369141, 20.428115844726562, 24.115875244140625, -2.9282264709472656, 30.96490478515625, 16.766342163085938, 53.501251220703125, 60.25738525390625, 3.2941436767578125, 18.25311279296875, 45.421142578125, 11.482162475585938, 18.20843505859375, -3.81787109375, 51.250518798828125, 27.480819702148438, 19.154815673828125, 2.3712120056152344, 73.234619140625, 23.011215209960938, 15.844703674316406, 13.371185302734375, -47.12908935546875, -15.65838623046875, 23.726318359375, -78.44331359863281, -15.511199951171875, 15.31585693359375, 50.89666748046875, -12.356056213378906, 41.036865234375, 10.4814453125, 47.26179504394531, 33.3851318359375, -14.170379638671875, 40.63353729248047, 1.3858718872070312, 28.66375732421875, 57.07061767578125, -46.71757507324219, 12.787109375, 0.0, 52.16705322265625, 9.3763427734375, -12.58123779296875, -17.876708984375, 6.350067138671875, 1.6497478485107422, 28.427154541015625, 67.24505615234375, 8.26409912109375, 70.15838623046875, 1.6623764038085938, 50.29423522949219, 34.85200500488281, 22.395843505859375, 27.790390014648438, 26.384033203125, 3.3150901794433594, 1.770050048828125, -74.29791259765625, 72.33389282226562, -8.373016357421875, 2.307596206665039, -6.759056091308594, -3.4162120819091797, -5.299407958984375, 66.5137939453125, 22.75738525390625, 37.635284423828125, 7.320037841796875, 20.4517822265625, -19.1763916015625, 4.05548095703125, -14.425739288330078, 14.777080535888672, 57.340240478515625, 51.3304443359375, -0.4389495849609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000198.npy"}
{"epoch": 0.41465968586387436, "step": 199, "batch_size": 128, "mean": 18.90143394470215, "std": 28.875886917114258, "min": -43.039276123046875, "p10": -15.618050384521483, "median": 13.808319091796875, "p90": 59.75074081420898, "max": 88.7113037109375, "pos_frac": 0.7421875, "sample": [12.3868408203125, 29.271728515625, 47.2557373046875, 43.893775939941406, 66.52825164794922, 42.833396911621094, 41.41392517089844, 2.8156661987304688, 58.07867431640625, 16.7430419921875, 88.64990234375, 50.41529846191406, 30.787750244140625, 39.531341552734375, 75.22564697265625, -40.044921875, 42.80902099609375, 38.76390075683594, 6.966972351074219, -2.0513553619384766, 3.6145401000976562, 1.806549072265625, -1.2109870910644531, 1.8399887084960938, 4.87738037109375, 20.886619567871094, 12.874580383300781, -2.1096572875976562, 13.243743896484375, -2.82470703125, 34.172767639160156, 27.35302734375, 40.505165100097656, 7.5296783447265625, 23.10443878173828, 38.37776184082031, -21.774974822998047, -7.749046325683594, 45.26133728027344, 12.728851318359375, 7.706581115722656, 43.18194580078125, 10.10955810546875, 7.254373550415039, 3.5580902099609375, 60.83502197265625, 9.13836669921875, -5.3142242431640625, -8.313308715820312, -35.47711181640625, 42.037940979003906, 59.731842041015625, 8.52545166015625, 66.65902709960938, -7.3905792236328125, 21.240325927734375, 11.297393798828125, 10.107162475585938, 54.16508483886719, -22.89137077331543, -41.37109375, 51.14141845703125, 52.594757080078125, 21.614974975585938, -7.373073577880859, -26.83645248413086, 35.879112243652344, 9.8555908203125, -15.266105651855469, 24.799972534179688, 73.4664306640625, -0.540069580078125, 2.753429412841797, 67.69464111328125, 4.643402099609375, -2.0198822021484375, 28.288482666015625, 21.72088623046875, -0.6229381561279297, 18.414825439453125, 62.995697021484375, 3.7512054443359375, 58.90936279296875, -8.604629516601562, 60.26727294921875, 60.140106201171875, -41.957237243652344, 45.36676025390625, 51.05908203125, 20.46581268310547, -23.4931640625, -10.9453125, 32.271820068359375, 60.5546875, -7.51678466796875, 22.4669189453125, 0.13526153564453125, 10.14813232421875, 23.052200317382812, 29.904815673828125, 18.45770263671875, -16.439254760742188, 15.559539794921875, 9.84161376953125, 33.683685302734375, -43.039276123046875, -30.37982177734375, 0.7051849365234375, 9.077095031738281, 40.11700439453125, 14.495964050292969, 10.55755615234375, 50.701629638671875, -2.475921630859375, 46.21868133544922, 14.372894287109375, 59.794837951660156, -0.691558837890625, -0.05936431884765625, 40.394126892089844, 7.17694091796875, -39.99224853515625, 0.8627777099609375, -12.474052429199219, 88.7113037109375, 54.3289794921875, 19.278060913085938, -18.1541748046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000199.npy"}
{"epoch": 0.4167539267015707, "step": 200, "batch_size": 128, "mean": 16.24165916442871, "std": 34.364593505859375, "min": -67.93386840820312, "p10": -30.076156616210934, "median": 11.917232513427734, "p90": 62.25399169921874, "max": 108.43280029296875, "pos_frac": 0.6796875, "sample": [46.16888427734375, -33.613739013671875, 80.81521606445312, 13.904411315917969, -41.1246337890625, 10.808921813964844, -26.919254302978516, 40.2047119140625, -5.09552001953125, 4.009260177612305, -1.3614349365234375, 54.274208068847656, 19.815216064453125, 0.6667022705078125, -60.3673095703125, 59.81512451171875, 40.43377685546875, -12.114448547363281, 22.64794921875, 79.31182861328125, -41.07220458984375, -31.838226318359375, 72.61590576171875, 18.993759155273438, 70.86474609375, 49.633697509765625, -16.85643768310547, -4.668632507324219, 39.80267333984375, -0.8660888671875, -26.582275390625, -49.964599609375, 17.875328063964844, -29.32098388671875, 11.537078857421875, 18.27642822265625, -32.69860076904297, 48.17767333984375, 37.54829406738281, -17.83843994140625, 0.0, 31.02057647705078, 31.501678466796875, 17.203628540039062, -5.335357666015625, -52.70513916015625, 65.1719970703125, 6.001800537109375, 87.66232299804688, -2.957489013671875, 16.251953125, 11.383209228515625, 38.006011962890625, 3.7682342529296875, -8.589859008789062, -6.305023193359375, 17.64983367919922, -50.53166198730469, 69.08612060546875, 1.6195640563964844, 36.18400573730469, 6.90966796875, 0.5276966094970703, 25.22052001953125, 54.103607177734375, 42.9625129699707, 9.525558471679688, 61.00341796875, -32.83172607421875, 47.516326904296875, 68.74913024902344, 38.39495849609375, -22.344467163085938, 28.7386474609375, 37.91204833984375, 17.54278564453125, 2.461456298828125, -4.6596221923828125, 9.51910400390625, 51.232177734375, -41.94517517089844, 69.2613525390625, 108.43280029296875, -0.8476715087890625, 7.602752685546875, 33.32307434082031, -6.508544921875, 40.04536819458008, -67.93386840820312, 52.31011962890625, 8.376312255859375, 86.25738525390625, -10.947372436523438, 58.6566162109375, 45.13519287109375, 60.253814697265625, 4.447944641113281, 69.35687255859375, -15.397705078125, -26.253890991210938, 8.512481689453125, 20.142181396484375, 50.6488037109375, 27.1922607421875, -10.83819580078125, 51.62760925292969, 12.297386169433594, -33.57275390625, -7.562835693359375, 3.1231689453125, 8.73297119140625, 41.221046447753906, 30.84478759765625, 70.34490966796875, 25.81640625, 36.94403076171875, -5.553276062011719, -7.223014831542969, 41.94354248046875, 6.247966766357422, 9.064987182617188, -1.196157455444336, 14.009254455566406, 3.43756103515625, -0.9955062866210938, 13.606880187988281, 19.8785400390625, 2.1428985595703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000200.npy"}
{"epoch": 0.418848167539267, "step": 201, "batch_size": 128, "mean": 18.742576599121094, "std": 28.12017822265625, "min": -49.4676513671875, "p10": -17.906866455078124, "median": 18.528827667236328, "p90": 56.91074829101562, "max": 97.64324951171875, "pos_frac": 0.734375, "sample": [-26.233367919921875, -9.957183837890625, 33.868858337402344, 49.440185546875, 39.3670654296875, 26.3460693359375, 21.0224609375, 28.7471923828125, 17.3916015625, 40.02040100097656, 52.45896911621094, 6.049607276916504, -5.678062438964844, 19.909759521484375, -20.016143798828125, -29.175567626953125, 45.79853820800781, 62.74859619140625, 39.83140563964844, 18.025619506835938, 14.438583374023438, 38.24139404296875, 43.631080627441406, 0.0, 43.47423553466797, 19.311233520507812, 45.416015625, 51.351470947265625, 3.8087615966796875, 57.596221923828125, 63.66143798828125, 38.356781005859375, 14.024520874023438, -3.520233154296875, 0.0, -49.34814453125, 4.579803466796875, -2.076934814453125, 2.769989013671875, 5.456756591796875, 20.56939697265625, 33.25848388671875, 48.59376525878906, 46.392860412597656, -23.958511352539062, 9.698989868164062, 22.710548400878906, 38.52235412597656, -16.362205505371094, 29.024620056152344, -1.5249786376953125, 19.03203582763672, 20.87841796875, 36.2418212890625, -1.1417999267578125, 14.134529113769531, 26.03833770751953, 56.24595260620117, 22.069686889648438, 45.39100646972656, -2.0698318481445312, 40.428436279296875, 59.228424072265625, 10.328460693359375, 56.616973876953125, 43.841583251953125, 25.815643310546875, 0.0, 9.85150146484375, -49.4676513671875, 13.6033935546875, 11.455001831054688, 0.3460693359375, 65.87933349609375, 36.017608642578125, 11.9581298828125, 3.704448699951172, 7.320945739746094, -7.8726806640625, 20.844696044921875, 19.590606689453125, 22.75872802734375, -13.618024826049805, -10.020271301269531, 0.283660888671875, 61.572998046875, -5.599708557128906, 52.93902587890625, -25.5714111328125, 97.64324951171875, 62.64251708984375, 35.77581787109375, 36.616973876953125, -4.0067138671875, -24.737091064453125, -11.593505859375, 7.956398010253906, 56.145660400390625, 8.254558563232422, 58.004364013671875, 48.461700439453125, 43.78689193725586, -35.82037353515625, 42.6151123046875, 8.207405090332031, -7.1720733642578125, -19.0953369140625, 32.3929443359375, 61.890625, -33.7947998046875, 13.473175048828125, 0.349761962890625, 11.09100341796875, 9.2216796875, -2.1432647705078125, 72.61444091796875, 4.7301025390625, 20.245330810546875, 27.875097274780273, 59.015289306640625, 9.75955581665039, -36.724456787109375, 7.3598175048828125, 61.54524230957031, -4.519142150878906, 20.72100830078125, -17.39752197265625, -29.4619140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000201.npy"}
{"epoch": 0.42094240837696334, "step": 202, "batch_size": 128, "mean": 18.301023483276367, "std": 31.729175567626953, "min": -69.73574829101562, "p10": -16.713520812988282, "median": 14.088699340820312, "p90": 61.91770629882812, "max": 76.85250854492188, "pos_frac": 0.703125, "sample": [66.90863037109375, 14.514472961425781, 4.8140716552734375, 17.045867919921875, 59.149017333984375, 0.774688720703125, -13.882110595703125, 64.12617492675781, -36.335205078125, -7.84326171875, -2.484375, 64.71408081054688, 19.729537963867188, 53.377105712890625, 44.91203308105469, -1.2975997924804688, -16.282028198242188, 45.713844299316406, 3.9069137573242188, 67.08203125, 19.936885833740234, 40.463348388671875, -8.1405029296875, 10.431930541992188, -3.837005615234375, 7.563690185546875, 72.5777587890625, 1.1079998016357422, 38.77867126464844, 14.71685791015625, -23.80511474609375, 20.835662841796875, -17.344680786132812, 62.511962890625, 16.782630920410156, 47.23655700683594, 47.55718994140625, 73.928955078125, 7.692649841308594, -32.75239562988281, 3.7695846557617188, 67.96249389648438, 40.63587951660156, 45.08074951171875, 15.375839233398438, 42.42082214355469, 46.250030517578125, 13.662925720214844, 59.7200927734375, -2.3560638427734375, -48.989532470703125, 34.8975830078125, 55.939239501953125, -5.784004211425781, -49.822601318359375, 32.168235778808594, 36.45416259765625, -9.370361328125, 21.162796020507812, -12.381240844726562, 20.082351684570312, 40.780487060546875, 61.86407470703125, -14.364410400390625, -3.01165771484375, -16.443023681640625, 69.3685302734375, -2.6236114501953125, 32.1771240234375, 4.2173614501953125, 1.0267772674560547, 52.9508056640625, 52.038818359375, 2.2799129486083984, 39.347564697265625, 46.623046875, 49.009429931640625, 24.94671630859375, 56.028411865234375, 52.964202880859375, 0.0, 61.434783935546875, 1.1835403442382812, 6.84600830078125, 11.30877685546875, -2.9052047729492188, 0.22757720947265625, -11.613433837890625, 7.346229553222656, -23.712158203125, 40.029998779296875, 45.57611083984375, 3.644561767578125, 12.917800903320312, 67.60946655273438, 33.348976135253906, -9.801971435546875, -17.380096435546875, -12.0594482421875, -19.875, -69.73574829101562, 62.0428466796875, 22.044082641601562, 50.11720275878906, 26.78021240234375, -14.089324951171875, -9.50616455078125, 13.350799560546875, 24.611053466796875, 1.705148696899414, 27.274749755859375, 56.07733154296875, -29.288787841796875, 11.96685791015625, 35.459381103515625, -62.464752197265625, 4.762359619140625, 43.71730422973633, -37.33258056640625, -10.383277893066406, 0.09099388122558594, 57.65071105957031, 2.8407440185546875, -0.12006568908691406, 2.0076141357421875, 69.65483093261719, -8.628074645996094, 76.85250854492188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000202.npy"}
{"epoch": 0.42303664921465967, "step": 203, "batch_size": 128, "mean": 14.27060317993164, "std": 24.7705078125, "min": -50.77803039550781, "p10": -16.62677993774414, "median": 10.411762237548828, "p90": 46.11277084350586, "max": 82.41886901855469, "pos_frac": 0.6953125, "sample": [2.011964797973633, 5.620597839355469, 27.156158447265625, 29.622901916503906, 36.48626708984375, 36.84349060058594, 8.792678833007812, 19.902252197265625, 9.261581420898438, 47.33366394042969, 49.12091064453125, 21.77606201171875, 3.1116580963134766, 32.237518310546875, 30.223342895507812, 31.40509033203125, -5.612213134765625, -16.27745819091797, 7.336517333984375, 7.1656951904296875, 11.327041625976562, 0.0, -32.980194091796875, 39.165916442871094, -17.536865234375, 9.546016693115234, -17.441864013671875, 30.62249755859375, -4.46331787109375, -0.9795913696289062, 19.805023193359375, 34.806427001953125, 28.449310302734375, 50.033203125, 34.54119873046875, -4.884721755981445, 21.40081787109375, 75.6412353515625, 17.007843017578125, -9.492462158203125, 43.02886962890625, 58.60040283203125, 17.392120361328125, 34.18505096435547, 1.5502433776855469, -4.7696990966796875, -0.7292022705078125, 39.17982482910156, -2.6996002197265625, 38.293739318847656, 1.4978160858154297, 11.480987548828125, 18.55419921875, 24.15997314453125, 82.41886901855469, -34.297088623046875, 36.810150146484375, -3.171995162963867, 50.022613525390625, 57.75042724609375, 5.441596984863281, 26.318145751953125, 31.344940185546875, -4.658100128173828, -30.721771240234375, 57.2155647277832, -25.00958251953125, 55.17669677734375, 8.529960632324219, 3.98541259765625, 15.354583740234375, -1.4868087768554688, -50.77803039550781, 42.65130615234375, -7.498931884765625, -2.3443679809570312, -13.576019287109375, 2.744873046875, -23.482452392578125, -1.778564453125, 23.757057189941406, 7.3656463623046875, -40.769317626953125, 39.541778564453125, 45.58953094482422, 9.3663330078125, 27.132888793945312, 13.47564697265625, 63.787506103515625, 44.4798583984375, 10.408546447753906, -7.77020263671875, -5.639862060546875, 42.135345458984375, 5.016323089599609, -8.855972290039062, 27.9998779296875, -1.2293548583984375, 43.838287353515625, 8.5977783203125, 38.26277160644531, 4.113922119140625, -7.988494873046875, 7.4052734375, 7.265959739685059, -34.874908447265625, -2.441680908203125, -21.167449951171875, 12.649749755859375, 60.00140380859375, 0.0, 10.41497802734375, 16.45996856689453, 17.812896728515625, -0.016017913818359375, -2.052398681640625, 15.086692810058594, 49.41571044921875, 4.6452789306640625, 2.425079345703125, 29.475311279296875, -30.93310546875, 6.68133544921875, 39.963958740234375, 19.61749267578125, -17.99212646484375, 10.901769638061523, 18.5037841796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000203.npy"}
{"epoch": 0.42513089005235605, "step": 204, "batch_size": 128, "mean": 21.05264663696289, "std": 28.63489532470703, "min": -55.47918701171875, "p10": -12.880401611328121, "median": 15.634944915771484, "p90": 60.64233245849609, "max": 98.27197265625, "pos_frac": 0.7734375, "sample": [82.78753662109375, 41.91217041015625, 70.20156860351562, 7.186187744140625, 10.720726013183594, 56.83441162109375, 4.3403778076171875, 32.939414978027344, 3.0420942306518555, 71.38818359375, 11.25213623046875, 8.762420654296875, 47.5892333984375, -17.80596923828125, -30.00408935546875, -55.47918701171875, 13.583839416503906, -21.6302490234375, 10.976158142089844, 19.40911865234375, 59.549591064453125, 11.437705993652344, 4.7249755859375, 38.210350036621094, 19.205718994140625, 4.513671875, 15.205558776855469, -15.767654418945312, 13.950424194335938, 25.273895263671875, 23.17218017578125, 98.27197265625, 6.094024658203125, -19.143753051757812, 39.35125732421875, -12.137451171875, 9.202342987060547, 84.93951416015625, -1.3760986328125, 52.0849609375, 64.02018737792969, 4.928840637207031, 31.569503784179688, -6.273506164550781, 42.440155029296875, 46.4007568359375, 4.5860137939453125, 70.51385498046875, 47.975189208984375, 75.12722778320312, 44.51591491699219, 35.0675048828125, 6.27471923828125, 41.00004959106445, 18.625465393066406, -3.4940948486328125, 40.1600341796875, 24.979251861572266, 4.286491394042969, -18.187255859375, 23.175506591796875, 44.1832275390625, 21.369384765625, -11.19781494140625, -0.055084228515625, 70.46432495117188, 16.0643310546875, 60.87884521484375, -39.5853271484375, 44.0804443359375, -5.222257614135742, 54.173583984375, -1.1047821044921875, -7.8105010986328125, 4.253322601318359, 9.67022705078125, 24.254474639892578, 55.32891082763672, -4.459943771362305, 17.76934814453125, 7.2851715087890625, 1.773834228515625, -30.340347290039062, 12.990081787109375, 29.262115478515625, -4.24896240234375, 49.22663879394531, 16.702972412109375, 58.54998779296875, 44.557708740234375, 65.76397705078125, 11.098518371582031, -32.108245849609375, 47.1578369140625, -5.636327743530273, 4.9251708984375, 47.5032958984375, -20.67041015625, 10.874969482421875, 36.531494140625, 7.1351165771484375, -1.849212646484375, 7.44970703125, 26.714935302734375, 40.38951873779297, 64.56353759765625, 22.993797302246094, 17.37646484375, 41.229339599609375, 9.343345642089844, 33.24470520019531, -2.2625274658203125, 40.551422119140625, 44.35565185546875, 9.741897583007812, 21.57354736328125, 23.222671508789062, 2.9940567016601562, -1.3245887756347656, 13.104248046875, -15.540184020996094, 10.000053405761719, 60.54096984863281, 26.427146911621094, -14.61395263671875, 81.85784912109375, 0.15851593017578125, -1.3543701171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000204.npy"}
{"epoch": 0.4272251308900524, "step": 205, "batch_size": 128, "mean": 19.814178466796875, "std": 28.392303466796875, "min": -43.44578552246094, "p10": -14.934890747070309, "median": 16.388809204101562, "p90": 54.608619689941406, "max": 96.8250732421875, "pos_frac": 0.78125, "sample": [17.220794677734375, 0.0, 2.2551422119140625, 0.9841289520263672, 9.708831787109375, 49.491790771484375, -17.42498779296875, -13.616912841796875, -10.330276489257812, 1.0449962615966797, 96.8250732421875, 39.405303955078125, 4.636085510253906, 9.029693603515625, -30.50421142578125, 8.503604888916016, 67.66473388671875, 69.46405029296875, 46.718238830566406, 5.97637939453125, 4.0018463134765625, 10.343429565429688, -7.3809814453125, 54.64007568359375, 16.175079345703125, 6.483642578125, 51.177276611328125, 10.340103149414062, 62.43157958984375, 19.451873779296875, 20.157318115234375, 29.913040161132812, 2.4099578857421875, -1.616973876953125, 2.883026123046875, 5.757776260375977, 29.82263946533203, 10.98529052734375, 26.67455291748047, 8.224395751953125, 26.411239624023438, 0.611572265625, -37.75421142578125, 30.527938842773438, 8.100341796875, 25.541015625, 0.0, 24.46514892578125, -30.61284637451172, 19.7313232421875, 4.74017333984375, 15.61666488647461, 54.57757568359375, 56.8431396484375, -30.564773559570312, 20.090118408203125, 52.595481872558594, 22.4046630859375, 0.6708984375, 45.0826416015625, -43.44578552246094, -2.5489501953125, 6.2607421875, 14.1116943359375, 5.723358154296875, 16.6025390625, 57.776580810546875, -10.318649291992188, -21.1676025390625, 88.73614501953125, 40.853271484375, 47.1209716796875, -33.9718017578125, 31.205902099609375, -0.298675537109375, 45.09361267089844, 2.07208251953125, 42.864830017089844, 46.07045364379883, -2.01025390625, 22.505966186523438, 64.654052734375, 40.691009521484375, 56.52130126953125, 22.54449462890625, 12.144693374633789, -22.943695068359375, 1.384979248046875, 42.953407287597656, 51.39735794067383, 51.64190673828125, 45.653350830078125, 78.07763671875, -13.867706298828125, 29.13824462890625, 58.75492858886719, 52.97581481933594, 11.61865234375, 10.252996444702148, 2.34661865234375, 73.413330078125, 38.94134521484375, -4.549530029296875, -32.672271728515625, 32.75738525390625, 40.942138671875, 14.687568664550781, 35.478515625, 38.77397155761719, 29.82037353515625, 28.228286743164062, 7.1640625, -36.134307861328125, 9.21151351928711, 31.1971435546875, -18.361053466796875, -38.830509185791016, 54.59513854980469, -2.8184585571289062, 46.52375411987305, 41.94659423828125, 47.42765808105469, 0.2884521484375, 41.57000732421875, -1.2988739013671875, 53.102783203125, 0.0, 26.628028869628906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000205.npy"}
{"epoch": 0.4293193717277487, "step": 206, "batch_size": 128, "mean": 17.202072143554688, "std": 26.53757095336914, "min": -68.67877197265625, "p10": -14.199601745605468, "median": 11.824874877929688, "p90": 52.21150512695313, "max": 76.79037475585938, "pos_frac": 0.71875, "sample": [33.564117431640625, 13.279415130615234, -37.445770263671875, 47.953460693359375, 76.79037475585938, -2.46112060546875, -14.185882568359375, 15.249603271484375, 11.799285888671875, 52.67863464355469, 19.44109344482422, 36.26762390136719, 42.49530029296875, -11.274820327758789, -21.953125, 15.0919189453125, 41.9755859375, 0.18804931640625, -2.62603759765625, 55.6348876953125, 1.48114013671875, 51.998443603515625, -3.5499496459960938, -14.231613159179688, -3.10455322265625, 4.063140869140625, -20.55072021484375, 12.45562744140625, 33.44963836669922, 8.764007568359375, 1.92071533203125, -10.2122802734375, -19.30841827392578, 47.86286926269531, 51.375396728515625, -10.929828643798828, 73.3121337890625, 55.331939697265625, 55.40484619140625, 42.87095642089844, 5.268402099609375, 11.484764099121094, -2.97747802734375, 29.57281494140625, 31.860504150390625, 23.595916748046875, -18.405975341796875, 51.986236572265625, 41.9912109375, 52.16896057128906, 20.36798095703125, 39.63185501098633, 41.68207550048828, 35.30877685546875, 5.752964019775391, 26.77801513671875, 20.78875732421875, 68.86380004882812, -68.67877197265625, 10.978271484375, 25.40289306640625, -22.94512176513672, -6.0182647705078125, 46.41637420654297, 36.949302673339844, 4.955902099609375, -0.82373046875, -9.815673828125, -3.9818363189697266, 31.130218505859375, 8.473037719726562, 49.10650634765625, 31.219940185546875, -26.084136962890625, 46.1317138671875, 2.4929962158203125, -8.65667724609375, 9.203254699707031, 9.26611328125, 0.8907833099365234, -14.38946533203125, -11.425308227539062, 61.89466857910156, 52.31077575683594, 36.31300354003906, 5.267333984375, 18.104751586914062, -19.031150817871094, 8.630683898925781, -0.12914657592773438, 36.649566650390625, 44.701637268066406, 35.43011474609375, 4.764518737792969, -7.90289306640625, 34.772369384765625, 13.74169921875, -7.691131591796875, 4.57000732421875, -2.165281295776367, 43.989776611328125, -6.85028076171875, 57.9058837890625, 58.55125427246094, 3.64495849609375, 44.0103759765625, -29.446380615234375, -6.291046142578125, -15.923004150390625, 11.8504638671875, 1.6957244873046875, 65.46255493164062, 37.05558776855469, 5.587249755859375, 2.0925521850585938, 39.458404541015625, 40.66734313964844, 5.549171447753906, 53.787628173828125, 30.65997314453125, 10.163970947265625, 11.9647216796875, -6.4591064453125, 38.561187744140625, 28.744659423828125, 7.2037353515625, -10.8505859375, 2.4630889892578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000206.npy"}
{"epoch": 0.431413612565445, "step": 207, "batch_size": 128, "mean": 15.13114070892334, "std": 32.34260940551758, "min": -63.82843017578125, "p10": -21.637451171874993, "median": 11.892364501953125, "p90": 59.453790283203126, "max": 81.91816711425781, "pos_frac": 0.7109375, "sample": [52.85662841796875, 38.392791748046875, 12.15399169921875, -6.446002960205078, 12.625625610351562, 16.068832397460938, 20.470001220703125, 80.55490112304688, 15.276641845703125, -2.6509037017822266, 15.704248428344727, 72.99151611328125, -41.48565673828125, 2.5050125122070312, 0.638427734375, 50.621299743652344, 4.176383972167969, 13.431564331054688, 81.91816711425781, -6.4754638671875, 59.13006591796875, -8.13360595703125, -63.47039794921875, -9.67660903930664, 45.90315628051758, 19.314117431640625, -8.79644775390625, 12.89923095703125, -51.177703857421875, 38.77845764160156, 49.28399658203125, 12.67193603515625, 8.7762451171875, -25.85858154296875, 34.773590087890625, 48.77352523803711, 11.6307373046875, -2.650754928588867, 16.752315521240234, 58.133392333984375, -8.361190795898438, 8.453170776367188, 32.41021728515625, 4.170745849609375, 6.2327880859375, -16.666534423828125, 10.364263534545898, 3.26214599609375, -51.35272216796875, 11.4251708984375, -63.82843017578125, -17.28680419921875, -18.334991455078125, 23.8349609375, 55.43182373046875, 0.0, 7.6117095947265625, -25.79925537109375, 26.271453857421875, -6.11163330078125, 67.84381103515625, -6.870880126953125, -29.68207550048828, 21.205657958984375, 13.37274169921875, -63.2274169921875, 62.04512023925781, -40.071807861328125, 3.39056396484375, -4.945487976074219, 29.7884521484375, -9.543487548828125, 8.670272827148438, 59.416748046875, 62.53997802734375, 32.30682373046875, 36.072853088378906, -8.585464477539062, -4.98260498046875, 61.12779235839844, 2.333211898803711, -43.36492919921875, 7.5772705078125, 25.11496353149414, 4.704315185546875, 79.97381591796875, 16.706153869628906, 41.60125732421875, -15.95263671875, 1.5526504516601562, 6.67816162109375, 20.048751831054688, -19.85382080078125, 55.72789001464844, 17.027252197265625, 46.06654357910156, 28.261749267578125, 4.7354736328125, 16.89093780517578, -57.3294677734375, 73.26568603515625, 59.54022216796875, -12.095703125, 43.68495178222656, 5.287342071533203, 72.44650268554688, 22.903839111328125, 1.952301025390625, 9.901838302612305, 55.89215087890625, 5.663051605224609, 30.176849365234375, -1.2257614135742188, 60.087310791015625, 30.09014892578125, -7.381904602050781, 35.7181396484375, 55.30983352661133, 65.72467041015625, -13.394744873046875, 9.416023254394531, -25.866348266601562, 49.4166259765625, 50.983123779296875, 0.572845458984375, 13.708786010742188, 46.50627136230469, 8.0191650390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000207.npy"}
{"epoch": 0.43350785340314135, "step": 208, "batch_size": 128, "mean": 17.956958770751953, "std": 29.49012565612793, "min": -76.0986328125, "p10": -10.92095489501953, "median": 13.317153930664062, "p90": 54.890570068359374, "max": 103.84014892578125, "pos_frac": 0.765625, "sample": [3.101654052734375, 6.0603179931640625, -21.5821533203125, 13.673629760742188, -39.97398376464844, 3.2801437377929688, -5.118509292602539, 21.48419189453125, 3.8484725952148438, 38.9288330078125, -9.792009353637695, 11.3861083984375, 30.455886840820312, 31.053375244140625, 27.949432373046875, 81.47586059570312, 14.09796142578125, 8.988407135009766, 15.9742431640625, 82.98953247070312, -59.171356201171875, 54.57810974121094, -7.0154876708984375, 28.00090789794922, 10.4815673828125, 48.1883544921875, 80.20489501953125, 33.031097412109375, 11.191673278808594, -12.325653076171875, 103.84014892578125, 37.61149597167969, 36.633148193359375, 30.887939453125, -10.884475708007812, 52.29066467285156, -76.0986328125, 10.3505859375, 9.4696044921875, 3.831298828125, 13.618558883666992, -41.456573486328125, 55.61964416503906, -8.338176727294922, 48.14849853515625, 23.13568115234375, 21.157089233398438, 76.23260498046875, 0.0, 92.1357421875, -16.567840576171875, 0.5790462493896484, 7.3687896728515625, -9.473129272460938, 1.9406585693359375, 90.6466064453125, 11.274490356445312, 8.706161499023438, 16.057395935058594, 57.840972900390625, 14.494571685791016, 25.131607055664062, -3.7689743041992188, 36.3216552734375, -4.02001953125, 4.94757080078125, 1.637359619140625, 2.7414398193359375, 18.250503540039062, 29.439956665039062, 91.1142578125, 0.0, 18.890235900878906, 42.148529052734375, 6.090492248535156, 52.76031494140625, -4.917045593261719, 57.886199951171875, 11.1993408203125, 43.17887878417969, 8.750774383544922, 30.9588623046875, 44.24700927734375, 63.923431396484375, -5.594635009765625, 28.928810119628906, -11.006072998046875, -0.7756576538085938, 2.359405517578125, 28.80316162109375, -9.237274169921875, 17.018951416015625, 33.63427734375, 4.1181640625, 44.643829345703125, 28.6009521484375, 44.77124786376953, 9.826385498046875, 32.68438720703125, -7.013923645019531, 3.1186676025390625, 4.719121932983398, 2.9126434326171875, 12.359310150146484, 14.63677978515625, 11.111968994140625, -0.9639739990234375, 50.69403076171875, 11.571746826171875, 19.49627685546875, 13.371871948242188, -22.512542724609375, 41.413780212402344, 26.90069580078125, 7.0615234375, -25.6575927734375, 13.262435913085938, 0.7618122100830078, -23.062789916992188, 35.415130615234375, 42.046417236328125, 13.795379638671875, -11.891716003417969, 57.934120178222656, 18.466697692871094, -27.6953125, -4.24493408203125, 28.296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000208.npy"}
{"epoch": 0.4356020942408377, "step": 209, "batch_size": 128, "mean": 15.46480941772461, "std": 28.80724334716797, "min": -70.690185546875, "p10": -20.82522964477539, "median": 12.71670913696289, "p90": 51.96445922851562, "max": 88.83985900878906, "pos_frac": 0.640625, "sample": [79.44114685058594, -8.56964111328125, 46.8375358581543, 48.623069763183594, -9.5634765625, -3.408599853515625, 21.11676025390625, -22.8353271484375, -25.21112060546875, 17.818817138671875, 43.784759521484375, -6.727256774902344, 3.011627197265625, 6.643781661987305, 69.4822998046875, 52.18829345703125, 28.228256225585938, 4.249822616577148, 4.716766357421875, 40.42852783203125, 52.28082275390625, -4.012359619140625, 4.5811767578125, -41.299774169921875, 13.093017578125, 6.799468994140625, 14.734426498413086, 7.92681884765625, -21.81689453125, 56.09675598144531, 57.108123779296875, -23.075469970703125, -1.4113922119140625, -24.524017333984375, -0.741912841796875, -18.9190673828125, 78.43380737304688, 17.696990966796875, 0.0, 8.84661865234375, 6.934814453125, -2.561981201171875, 25.637374877929688, -26.3570556640625, 14.629287719726562, 27.861083984375, 30.606739044189453, 3.4225006103515625, 20.8597412109375, 71.80958557128906, 30.368682861328125, 23.1063232421875, 13.2559814453125, -2.032867431640625, 15.081817626953125, -0.9901294708251953, -6.156463623046875, 12.032951354980469, 46.177581787109375, -5.54608154296875, 39.633209228515625, -11.590896606445312, -3.0811767578125, 50.560821533203125, 55.46539306640625, 11.463043212890625, -9.933746337890625, 0.5830078125, 64.90252685546875, 5.160369873046875, 20.86309814453125, 48.6629638671875, 19.805145263671875, 20.818359375, -2.0073165893554688, 0.0, 35.72271728515625, 40.94744873046875, -20.400230407714844, 22.198974609375, -23.03564453125, -2.7034759521484375, 33.569114685058594, -0.116851806640625, 1.0263824462890625, 51.8685302734375, 41.459014892578125, -35.10601806640625, -12.1551513671875, 6.481536865234375, 17.79883575439453, -15.492897033691406, -16.072174072265625, 41.210750579833984, -1.6608238220214844, -0.82635498046875, 21.92144775390625, 29.866065979003906, 88.83985900878906, -70.690185546875, -9.039840698242188, 4.649620056152344, -28.31884765625, -4.863361358642578, -9.159530639648438, -51.42718505859375, -4.5644378662109375, 12.340400695800781, 37.1968994140625, 41.783905029296875, 36.09416198730469, 49.61790466308594, 32.57444763183594, 13.789993286132812, 34.475425720214844, 51.44476318359375, 31.060806274414062, 46.84613037109375, -2.83837890625, 74.1484375, 25.998565673828125, -25.7039794921875, 36.31864929199219, 30.96527099609375, 42.032623291015625, 49.213226318359375, -5.0779266357421875, 53.789215087890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000209.npy"}
{"epoch": 0.437696335078534, "step": 210, "batch_size": 128, "mean": 17.005859375, "std": 28.400800704956055, "min": -63.725311279296875, "p10": -12.220777893066405, "median": 15.197399139404297, "p90": 49.82264099121093, "max": 108.614013671875, "pos_frac": 0.703125, "sample": [-5.982231140136719, -1.3909072875976562, 29.0042724609375, -2.615478515625, 27.949813842773438, 11.548171997070312, 71.6815185546875, 64.79150390625, 30.1949462890625, 17.477493286132812, 46.323760986328125, 11.114315032958984, 3.34967041015625, 39.842132568359375, 31.5076904296875, -1.51763916015625, 37.599639892578125, -3.7777862548828125, 35.99360656738281, 2.8951416015625, 21.148693084716797, -12.066268920898438, -38.31451416015625, -2.1422882080078125, 35.66065979003906, 25.741531372070312, 8.150482177734375, 108.614013671875, -41.146820068359375, -4.9925994873046875, 34.875457763671875, 18.539093017578125, 11.013458251953125, -0.8306732177734375, 44.46502685546875, -15.322052001953125, 5.371009826660156, 35.6456298828125, 0.8274955749511719, 22.865478515625, 2.0247802734375, 30.347213745117188, 7.731845855712891, -5.846717834472656, 9.091072082519531, 38.975555419921875, 24.39593505859375, 1.977447509765625, 48.32177734375, 11.39940071105957, 18.419891357421875, -2.659637451171875, 31.34136962890625, 60.8447265625, 18.1156005859375, 26.381309509277344, 31.10394287109375, -12.057479858398438, 47.302490234375, -6.36676025390625, 46.3114013671875, 40.109405517578125, 50.760467529296875, 47.5910758972168, 15.75933837890625, 21.254364013671875, -52.671875, -6.613128662109375, 14.03814697265625, 23.11676025390625, -33.1610107421875, -20.952529907226562, 58.652587890625, 24.047210693359375, 66.57876586914062, 55.677154541015625, 0.030527114868164062, 6.921302795410156, 5.05645751953125, -63.725311279296875, 46.69270324707031, 14.635459899902344, -0.8586368560791016, 70.71875, 18.567825317382812, 49.42071533203125, 75.45123291015625, 31.736358642578125, -0.2745361328125, -2.6486968994140625, -3.407440185546875, 11.855701446533203, 2.802684783935547, -18.635543823242188, 18.821226119995117, 17.87841796875, -5.067169189453125, 73.903564453125, -12.581298828125, 31.149673461914062, 34.180633544921875, 77.0216064453125, 6.388284683227539, -2.125274658203125, 65.63137817382812, 12.762260437011719, 44.33753967285156, -2.0897598266601562, 8.591033935546875, -15.40771484375, 22.413665771484375, 18.6119384765625, -2.41558837890625, 11.31781005859375, 11.734848022460938, -48.79454040527344, 9.282890319824219, 30.815460205078125, 22.75665283203125, -45.988311767578125, 43.73877716064453, -7.62994384765625, 40.03636169433594, 33.6351318359375, -21.90679931640625, -2.8708343505859375, -0.608062744140625, 29.452438354492188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000210.npy"}
{"epoch": 0.4397905759162304, "step": 211, "batch_size": 128, "mean": 23.780736923217773, "std": 27.771923065185547, "min": -43.704376220703125, "p10": -7.570690155029295, "median": 18.34775161743164, "p90": 58.44875793457031, "max": 93.11819458007812, "pos_frac": 0.796875, "sample": [16.33837127685547, 10.477630615234375, 77.474609375, 34.4603271484375, 48.15721893310547, 57.93475341796875, 8.751394271850586, 6.522758483886719, -9.59356689453125, 40.57366943359375, 45.785736083984375, -43.704376220703125, 30.266357421875, 33.145904541015625, -10.881256103515625, 55.703094482421875, 2.3720245361328125, -1.7740707397460938, 1.26043701171875, -13.256057739257812, 6.7230377197265625, 38.10533142089844, 9.718116760253906, 14.92242431640625, 21.312210083007812, 18.965728759765625, 5.0525054931640625, -29.448638916015625, 26.375015258789062, -3.5919189453125, 16.817798614501953, 49.06536865234375, -0.31756591796875, 17.729774475097656, 12.228836059570312, 24.569976806640625, 10.21217155456543, -30.01490020751953, 20.838699340820312, 60.266815185546875, 10.96142578125, 41.87445068359375, 7.91778564453125, -1.55096435546875, -2.474334716796875, 83.86865234375, 57.733612060546875, 39.792877197265625, -2.5807342529296875, 9.798526763916016, 49.35540771484375, -5.0120697021484375, 4.849639892578125, -1.57568359375, 93.11819458007812, 41.64434814453125, 52.330841064453125, -13.891914367675781, 13.845779418945312, 47.42047119140625, -5.4793701171875, -6.973297119140625, -12.885910034179688, 79.26962280273438, -8.964607238769531, 4.17681884765625, -0.854034423828125, 45.02894592285156, 54.361572265625, 8.943519592285156, -16.597412109375, 1.16864013671875, -1.5981998443603516, 34.74803924560547, 4.233221054077148, 11.429722785949707, 46.668575286865234, 43.081695556640625, 46.49492645263672, 42.96873474121094, 29.902587890625, 23.8973388671875, 1.5446701049804688, 41.3199462890625, 50.735809326171875, 85.36090087890625, 9.8770751953125, 2.615325927734375, 42.25591278076172, 5.7601318359375, -9.2901611328125, 74.16543579101562, 71.66705322265625, -25.103607177734375, 67.20660400390625, 13.419609069824219, 51.24688720703125, 61.403289794921875, 10.339273452758789, 6.54229736328125, 42.832916259765625, 56.95892333984375, 32.96099853515625, 8.414630889892578, -41.29121017456055, 3.3963394165039062, 33.865631103515625, 36.269775390625, 65.33541870117188, 42.18975830078125, 35.490509033203125, 9.866334915161133, 54.76133728027344, 32.71392822265625, 2.7617263793945312, 27.127777099609375, 68.40255737304688, 59.648101806640625, 43.91297912597656, 16.241165161132812, 15.312301635742188, 25.91736602783203, -1.1376209259033203, 55.97568130493164, 28.20427703857422, 37.181060791015625, 4.4361419677734375, 55.12605285644531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000211.npy"}
{"epoch": 0.4418848167539267, "step": 212, "batch_size": 128, "mean": 16.44942283630371, "std": 27.770570755004883, "min": -66.94622802734375, "p10": -13.132977294921874, "median": 13.186897277832031, "p90": 51.8075927734375, "max": 84.06756591796875, "pos_frac": 0.7265625, "sample": [5.380786895751953, -3.332427978515625, -10.303314208984375, 14.843208312988281, -11.04132080078125, -12.89239501953125, 1.2303314208984375, 13.838119506835938, 1.6636505126953125, -1.0234375, 45.40338134765625, 0.9545745849609375, 75.973388671875, 3.202188491821289, 55.59814453125, -20.085548400878906, 6.486566543579102, 50.032989501953125, 51.90199279785156, 25.062835693359375, 42.393592834472656, 6.775146484375, 0.40496826171875, 15.61871337890625, 23.05267333984375, 9.33258056640625, 0.9741363525390625, -43.45362854003906, 16.665252685546875, 11.21484375, -17.039745330810547, -23.529617309570312, 4.9608306884765625, 5.411163330078125, 18.114280700683594, 32.39117431640625, 27.797454833984375, 63.04951477050781, 16.204315185546875, -0.8444061279296875, 34.0543212890625, -60.201385498046875, -1.689544677734375, -3.923635482788086, -10.18316650390625, 44.16371154785156, 21.852325439453125, 46.844329833984375, -19.479690551757812, -30.40704345703125, 16.18730926513672, 7.60498046875, -26.383697509765625, 58.5494384765625, 26.717178344726562, -5.28411865234375, -5.853858947753906, 43.57037353515625, 6.28070068359375, -6.404224395751953, -5.13238525390625, 3.8892784118652344, 47.78778839111328, -0.658111572265625, -5.706539154052734, 48.10255432128906, -8.584548950195312, 37.463287353515625, 36.56512451171875, 59.83959197998047, 29.439407348632812, 50.11988830566406, -23.446929931640625, 12.108062744140625, 51.76713562011719, 10.297821044921875, 31.018367767333984, 38.32635498046875, -1.5980453491210938, 69.90336608886719, 30.82305908203125, 43.215423583984375, 15.630508422851562, 45.97125244140625, -52.39085388183594, 19.12359619140625, 17.788604736328125, -13.6943359375, 10.455398559570312, 7.440418243408203, 32.437530517578125, 52.17230224609375, 5.229522705078125, 5.436428070068359, 82.58587646484375, 38.01995849609375, 21.32452392578125, 55.88196563720703, 35.41438293457031, 19.775146484375, 23.306190490722656, 84.06756591796875, 40.45109558105469, -0.0005550384521484375, 12.535675048828125, -66.94622802734375, 31.469024658203125, 45.0618896484375, -3.885345458984375, -9.974433898925781, 1.18463134765625, -26.849212646484375, 27.8350830078125, 7.555776596069336, 22.0277099609375, -0.7008819580078125, 26.727142333984375, 9.411758422851562, 34.07363510131836, -1.3418960571289062, 46.470184326171875, 8.389389038085938, 31.414993286132812, 57.16059112548828, 9.84918212890625, 21.0362548828125, 72.84246826171875, 9.8109130859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000212.npy"}
{"epoch": 0.44397905759162304, "step": 213, "batch_size": 128, "mean": 15.838903427124023, "std": 28.384260177612305, "min": -45.76947021484375, "p10": -16.289251708984374, "median": 11.980724334716797, "p90": 59.47715454101562, "max": 94.720703125, "pos_frac": 0.671875, "sample": [59.7677001953125, 40.10267639160156, -11.574935913085938, 56.673675537109375, -6.31683349609375, 32.56958770751953, 46.815425872802734, 13.978607177734375, -3.68890380859375, 29.43255615234375, 0.083160400390625, -4.970703125, 16.165908813476562, -12.319267272949219, 14.425905227661133, 86.35067749023438, -1.036529541015625, 18.605377197265625, 19.6533203125, 27.27362060546875, 16.56683349609375, -4.508026123046875, 76.46530151367188, -0.998291015625, 55.49684143066406, -7.299072265625, -1.317230224609375, 13.094635009765625, 14.889389038085938, 12.364128112792969, 6.3363189697265625, -16.4476318359375, -11.027969360351562, 15.238739013671875, 53.75553894042969, 3.0377349853515625, 46.905853271484375, -2.057861328125, 6.1943817138671875, -40.36187744140625, 51.75776672363281, 22.90289306640625, 15.586257934570312, 27.343505859375, -31.6962890625, 53.05389404296875, 34.9046630859375, 4.421783447265625, -23.502838134765625, 13.313629150390625, 5.209938049316406, 34.019134521484375, 39.307952880859375, 2.8970909118652344, 47.639862060546875, -17.320114135742188, -3.936920166015625, 45.805885314941406, -20.269378662109375, 67.35957336425781, 14.423187255859375, 14.77178955078125, 12.642349243164062, -20.453659057617188, 6.60955810546875, 11.085205078125, 70.55935668945312, -16.22137451171875, -10.509384155273438, -4.1527557373046875, 45.62953186035156, -17.503173828125, 19.415771484375, -45.76947021484375, 13.018341064453125, 4.360595703125, 28.400192260742188, 61.762237548828125, 6.0004730224609375, 25.86224365234375, -11.816978454589844, 6.402275085449219, -2.458240509033203, -2.52947998046875, -0.02587127685546875, 49.3980712890625, -24.528594970703125, -1.144561767578125, 11.597320556640625, 85.7730712890625, 40.07978057861328, 62.103057861328125, 3.8816051483154297, 8.315277099609375, 28.8387451171875, -4.800243377685547, 59.368682861328125, -10.868438720703125, -14.6190185546875, 94.720703125, 2.119434356689453, 26.65601348876953, 16.863967895507812, 61.746612548828125, -26.880462646484375, 80.69879150390625, 27.166259765625, -3.0078659057617188, 27.497314453125, 0.0, 9.301885604858398, 9.849380493164062, 16.05303955078125, 59.730255126953125, 9.158723831176758, -35.641998291015625, 22.087127685546875, 17.00095558166504, -2.7566986083984375, 16.75079345703125, 2.4722557067871094, 27.7581787109375, -16.642608642578125, 1.6573104858398438, 70.55810546875, 1.3278732299804688, -16.050216674804688, -2.8300552368164062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000213.npy"}
{"epoch": 0.44607329842931936, "step": 214, "batch_size": 128, "mean": 16.457096099853516, "std": 28.665414810180664, "min": -43.388946533203125, "p10": -17.66099510192871, "median": 14.767669677734375, "p90": 54.760067749023435, "max": 104.84701538085938, "pos_frac": 0.734375, "sample": [36.064449310302734, 33.95391845703125, -5.2035369873046875, 9.1463623046875, 10.581844329833984, 40.984405517578125, 15.045516967773438, -11.363151550292969, -30.914306640625, -13.203582763671875, 18.11883544921875, 33.304664611816406, -12.266433715820312, 25.587005615234375, 18.197509765625, 43.82342529296875, -42.150299072265625, 32.10435485839844, -5.086761474609375, 22.60015869140625, 5.745994567871094, -7.345947265625, 13.415771484375, 31.072219848632812, 30.6627197265625, 37.11848449707031, 20.446434020996094, -13.07696533203125, 18.9664306640625, 14.322792053222656, 4.645751953125, 7.2169036865234375, 5.525478363037109, 0.3355560302734375, 54.09321594238281, 15.929473876953125, 104.84701538085938, 19.751190185546875, 58.25001525878906, -33.63873291015625, 14.489822387695312, -15.8978271484375, 2.5102577209472656, 12.16412353515625, 8.247184753417969, 1.432891845703125, 1.6312255859375, 61.425323486328125, 4.505218505859375, -5.372518539428711, 42.64215850830078, 45.30108642578125, -16.162521362304688, 35.542137145996094, 2.6420364379882812, 52.76097106933594, 16.22235107421875, 36.418182373046875, -5.87603759765625, 46.1239013671875, 38.169586181640625, 79.66339111328125, 27.614883422851562, 4.353515625, 3.5420684814453125, -18.373031616210938, 2.766021728515625, -35.51091766357422, 7.2564697265625, 0.0, -8.6021728515625, -18.31463623046875, -2.9571533203125, 10.7655029296875, 47.879364013671875, -9.854461669921875, 27.342689514160156, -24.363162994384766, 9.442428588867188, 9.064851760864258, -9.83990478515625, -37.65576171875, 70.6573486328125, 31.638885498046875, 27.230499267578125, -23.843902587890625, 56.31605529785156, 22.84149932861328, 17.72216796875, 12.456680297851562, -41.943328857421875, 23.61846160888672, 22.00653076171875, 40.2613525390625, 64.23757934570312, 14.189804077148438, -4.19207763671875, 26.0557861328125, 1.998992919921875, 28.726119995117188, 52.81739807128906, 63.05438232421875, 22.466552734375, 28.1414794921875, 25.45916748046875, 59.823944091796875, 85.8380126953125, 24.283782958984375, -43.388946533203125, 25.551620483398438, 97.82928466796875, 40.787994384765625, 8.98773193359375, -19.48736572265625, -12.33258056640625, 56.33155822753906, 32.217742919921875, -29.429153442382812, -10.55096435546875, -14.32940673828125, 5.889373779296875, 29.89459991455078, 21.239227294921875, 17.41900634765625, -17.380863189697266, 66.439453125, 12.127265930175781, 36.08172607421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000214.npy"}
{"epoch": 0.4481675392670157, "step": 215, "batch_size": 128, "mean": 16.342525482177734, "std": 31.68714714050293, "min": -74.2446060180664, "p10": -23.5277587890625, "median": 12.379805564880371, "p90": 54.57498779296875, "max": 115.1766357421875, "pos_frac": 0.703125, "sample": [12.417898178100586, 17.763351440429688, 8.327362060546875, 42.702178955078125, 68.48529052734375, 40.103363037109375, 37.17333984375, 12.341712951660156, 17.54290771484375, -4.40185546875, 70.52627563476562, 68.0032958984375, 41.116546630859375, -16.7757568359375, 0.20697021484375, 21.5054931640625, -0.463470458984375, 88.865478515625, 90.17529296875, 46.174163818359375, 42.39971923828125, -4.432373046875, 0.0, -24.61236572265625, -2.7066574096679688, -13.167606353759766, 18.366470336914062, 19.411102294921875, 84.84115600585938, -7.858497619628906, 51.73458480834961, 1.012481689453125, -3.528411865234375, -33.55839538574219, 0.8527812957763672, 24.642501831054688, 50.99530029296875, 14.170669555664062, 32.77595520019531, 83.18679809570312, -5.531768798828125, 23.40972900390625, -23.8787841796875, -7.424468994140625, -0.29527854919433594, 0.1055908203125, -14.99517822265625, 32.60633850097656, 45.026153564453125, 37.41228485107422, 3.332244873046875, 115.1766357421875, 38.444000244140625, 18.6136474609375, -27.8988037109375, 2.5602760314941406, 8.395431518554688, -14.624725341796875, 0.0, 52.22650146484375, -1.3976211547851562, 2.393707275390625, 17.313812255859375, -16.813140869140625, 22.302886962890625, 22.182498931884766, -28.931777954101562, 1.8510589599609375, 59.91145324707031, 40.1900634765625, 14.465072631835938, 7.408121109008789, -2.2946701049804688, 54.140625, -44.482208251953125, 46.59173583984375, 7.346435546875, -74.2446060180664, 19.68669319152832, 34.307373046875, 11.943588256835938, 7.4278717041015625, 13.393280029296875, 3.618743896484375, -21.865394592285156, -3.4996719360351562, 45.20640563964844, 10.795562744140625, 58.5814208984375, -25.88427734375, -9.532295227050781, 15.04620361328125, 15.671455383300781, -3.208740234375, 4.4873809814453125, 11.825248718261719, -2.778127670288086, 20.945037841796875, 65.10675048828125, 108.97833251953125, -23.3773193359375, 20.043533325195312, -30.732681274414062, 0.5620670318603516, -45.73492431640625, -28.340240478515625, 28.904037475585938, 53.981712341308594, -31.07489013671875, 43.099090576171875, 4.2437744140625, -6.899271011352539, 22.221721649169922, 21.568939208984375, 8.344100952148438, 20.313812255859375, 14.451690673828125, 34.1712646484375, -35.30816650390625, 22.7716064453125, 10.981033325195312, 2.30157470703125, 7.15142822265625, 55.5885009765625, 39.9527587890625, 35.40441131591797, 8.8046875, 51.287689208984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000215.npy"}
{"epoch": 0.450261780104712, "step": 216, "batch_size": 128, "mean": 18.02680206298828, "std": 30.581655502319336, "min": -68.69146728515625, "p10": -20.932968139648438, "median": 18.120765686035156, "p90": 59.86656646728515, "max": 88.078369140625, "pos_frac": 0.75, "sample": [16.043045043945312, 10.600776672363281, 37.985435485839844, 23.771751403808594, 2.2760009765625, 6.12969970703125, -68.69146728515625, 6.672760009765625, 40.561798095703125, -2.946247100830078, 49.30883026123047, 52.119110107421875, 24.523193359375, -2.3803176879882812, -0.5102691650390625, 36.118255615234375, 88.078369140625, 7.484161376953125, 45.038177490234375, 63.66815185546875, 15.240325927734375, -52.58154296875, 27.546348571777344, 19.2569580078125, -13.770450592041016, 17.583465576171875, -58.03271484375, 16.95159149169922, 24.870187759399414, 35.77374267578125, -1.4488983154296875, -20.762908935546875, -37.81329345703125, -4.643501281738281, 35.711002349853516, 5.975214004516602, 40.39178466796875, -38.204986572265625, 71.12539672851562, 41.45093536376953, -12.388416290283203, 53.83674621582031, 62.0848388671875, 46.923728942871094, 39.254302978515625, 44.037994384765625, 30.18560791015625, 28.261566162109375, 3.7671585083007812, 0.6233139038085938, -14.349822998046875, -5.724456787109375, 65.91549682617188, 38.80940246582031, -21.32977294921875, 32.302337646484375, 2.481781005859375, 70.3121337890625, 22.34033203125, 4.899501800537109, 0.0, 26.266128540039062, -15.595916748046875, 18.113037109375, 59.439849853515625, 61.086181640625, 59.74755859375, -5.604217529296875, -6.0071258544921875, 42.79931640625, 45.94580078125, 50.04730224609375, -50.2388916015625, 12.5836181640625, 59.88569641113281, 65.79396057128906, 5.102386474609375, 33.752410888671875, 20.51177978515625, 10.55401611328125, 18.954666137695312, 1.9249038696289062, 27.17230224609375, 19.5238037109375, 65.97213745117188, 36.422027587890625, -35.01664733886719, -5.8841400146484375, 69.10882568359375, 83.755126953125, -30.118560791015625, -24.0784912109375, 4.7880859375, 52.20306396484375, -0.01906585693359375, 19.4063720703125, 12.99285888671875, -10.622955322265625, 9.219137191772461, 10.59228515625, 70.37002563476562, 9.88555908203125, 0.0, 23.851821899414062, 30.18085479736328, -36.96758270263672, 1.449798583984375, 37.218475341796875, 39.5482177734375, 16.682769775390625, -0.12935447692871094, 37.00506591796875, 20.425445556640625, 23.6168212890625, 16.963401794433594, 18.128494262695312, 24.4632568359375, 12.223388671875, 4.32659912109375, 59.858367919921875, 22.20184326171875, -51.463592529296875, 35.228973388671875, 34.23488235473633, 5.8719482421875, 1.60791015625, -25.510040283203125, 2.965057373046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000216.npy"}
{"epoch": 0.4523560209424084, "step": 217, "batch_size": 128, "mean": 24.114669799804688, "std": 33.26805877685547, "min": -65.583984375, "p10": -12.547698974609371, "median": 18.61627960205078, "p90": 62.839864349365236, "max": 121.5771484375, "pos_frac": 0.78125, "sample": [7.17138671875, -0.6438446044921875, -16.9801025390625, 25.237457275390625, 46.347686767578125, -40.53300476074219, 9.805816650390625, 59.002410888671875, -1.6858634948730469, -1.4566650390625, 63.74787902832031, 2.367431640625, 17.677528381347656, 34.51969909667969, 1.65423583984375, 60.83958435058594, 46.559871673583984, 51.2960205078125, 50.14424133300781, 1.07623291015625, 57.40040588378906, 17.0299072265625, 3.2466659545898438, 45.70814514160156, 121.5771484375, 21.9722900390625, 47.98394775390625, 46.062767028808594, 51.408111572265625, 62.683815002441406, 20.4737548828125, 44.859527587890625, -11.68243408203125, 88.618896484375, 43.55926513671875, 18.552459716796875, 85.12322998046875, 46.66204833984375, 60.949951171875, -7.21783447265625, -65.583984375, 56.140167236328125, 27.678436279296875, 105.128662109375, -23.885299682617188, 35.197601318359375, 10.481475830078125, -35.89167022705078, 29.38134765625, 14.00762939453125, 3.0387649536132812, 2.7480239868164062, -6.841400146484375, 29.482772827148438, 25.13953399658203, -35.644805908203125, 10.965301513671875, 52.98771667480469, 24.277183532714844, -5.342864990234375, 9.715629577636719, 54.773284912109375, -2.4838104248046875, 82.19677734375, 0.790191650390625, 9.343772888183594, 7.778602600097656, 66.46565246582031, 36.770347595214844, 3.9989776611328125, 46.91033172607422, 47.42565155029297, 22.859832763671875, -2.686573028564453, 62.09368896484375, -7.179443359375, -5.9314422607421875, -18.5015869140625, 10.0145263671875, 115.15032958984375, 8.657638549804688, 40.131011962890625, 16.134654998779297, -8.867267608642578, -58.003021240234375, 73.15325927734375, 63.2039794921875, 54.7918701171875, 7.488006591796875, -14.566650390625, 12.73443603515625, 48.70928955078125, -2.31976318359375, 48.063079833984375, 9.069442749023438, 74.3773193359375, 1.3437576293945312, 9.486785888671875, 7.0362091064453125, 56.73411560058594, -25.86322021484375, 0.0, 22.338241577148438, 10.916534423828125, -21.29296875, 91.26449584960938, 46.60255432128906, -15.029052734375, 37.1773681640625, 77.45588684082031, 16.115158081054688, 18.680099487304688, 19.55908203125, 61.3660888671875, 48.155517578125, 29.357406616210938, 23.528289794921875, 16.81854248046875, -25.5765380859375, 10.070327758789062, 39.313934326171875, 5.362066268920898, 28.733779907226562, 13.010726928710938, 14.714485168457031, -11.48876953125, 29.66455078125, 4.275470733642578], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000217.npy"}
{"epoch": 0.4544502617801047, "step": 218, "batch_size": 128, "mean": 19.525588989257812, "std": 29.767396926879883, "min": -51.131256103515625, "p10": -11.222460174560547, "median": 14.125812530517578, "p90": 58.298747253417964, "max": 93.00753784179688, "pos_frac": 0.7734375, "sample": [24.94952392578125, 12.903663635253906, 2.1759033203125, 55.777191162109375, 4.726318359375, -0.3504791259765625, -10.702056884765625, 67.61077880859375, 9.980987548828125, 8.795722961425781, -31.681549072265625, 8.931655883789062, 19.74462890625, 66.07913208007812, 57.82647705078125, -46.126861572265625, -10.62860107421875, 49.82449722290039, 48.01631164550781, 18.271148681640625, -5.639839172363281, 47.09486389160156, 42.160400390625, 53.87782287597656, 10.68560791015625, 13.363601684570312, -9.33624267578125, 6.28619384765625, -30.212913513183594, 37.558631896972656, 41.88475799560547, 77.91656494140625, 2.03228759765625, 56.547996520996094, -14.5498046875, 11.89764404296875, 49.533058166503906, -44.647430419921875, 17.174301147460938, 13.707763671875, 93.00753784179688, 17.196849822998047, 28.965423583984375, 1.2905597686767578, 3.5552520751953125, 54.98475646972656, 3.4415512084960938, -7.08074951171875, -19.71539306640625, 49.72554016113281, 48.54704284667969, -11.520576477050781, 39.04154968261719, 14.97930908203125, 27.741668701171875, 0.5417671203613281, 59.493438720703125, 44.2164306640625, 61.98622131347656, -17.9581298828125, 8.675796508789062, -3.9589996337890625, -24.41570281982422, 8.508163452148438, 0.0, 34.16314697265625, 11.3310546875, 2.2080078125, -11.094696044921875, 54.81964111328125, -5.912147521972656, 22.323654174804688, -33.18027877807617, 53.19346618652344, 28.51385498046875, 65.04327392578125, 14.543861389160156, 24.323577880859375, 20.29736328125, 0.7225189208984375, 73.71482849121094, 31.476043701171875, 75.11444091796875, 4.5755615234375, -0.26141357421875, 12.880462646484375, 8.518630981445312, 0.9673690795898438, 42.919647216796875, 26.86224365234375, 10.178665161132812, 35.30827331542969, 7.83642578125, -51.131256103515625, 22.715286254882812, 19.216705322265625, 57.46278381347656, 88.02685546875, -32.6900634765625, 31.03973388671875, 2.094696044921875, 27.8951416015625, -5.127811431884766, -1.1090087890625, 38.43011474609375, -9.915390014648438, -49.356590270996094, 59.40071105957031, 4.2789306640625, 19.4466552734375, 6.69183349609375, 43.05194091796875, 26.66400146484375, 34.453643798828125, -5.99908447265625, 13.07122802734375, 75.0128173828125, 0.031707763671875, 7.440895080566406, -10.234893798828125, 40.2235107421875, 36.06059265136719, 23.7598876953125, 10.82470703125, 32.27105712890625, 89.6173095703125, 10.024604797363281, 19.53564453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000218.npy"}
{"epoch": 0.45654450261780105, "step": 219, "batch_size": 128, "mean": 20.23830223083496, "std": 29.223297119140625, "min": -72.60443115234375, "p10": -6.861748504638672, "median": 16.715484619140625, "p90": 58.47901611328125, "max": 90.57757568359375, "pos_frac": 0.7578125, "sample": [-66.6912841796875, 0.447357177734375, 27.265098571777344, 6.5151824951171875, 53.11357116699219, 14.609054565429688, -11.455062866210938, 21.68157958984375, 47.018707275390625, 21.52264404296875, 28.123931884765625, -1.3617496490478516, 28.3701171875, -3.7027587890625, 2.217193603515625, 42.93486022949219, 49.86309814453125, -20.37750244140625, 42.12969970703125, 40.828304290771484, 31.04724884033203, 5.237028121948242, 29.44775390625, 81.96255493164062, 12.626426696777344, 38.28486633300781, 31.220916748046875, 4.1989898681640625, 46.59295654296875, -9.318267822265625, -6.427227020263672, -2.6927661895751953, -27.862701416015625, 33.276771545410156, 2.24688720703125, 56.566585540771484, -25.8251953125, 40.849735260009766, 26.9149169921875, 57.16868591308594, 72.10086059570312, 90.57757568359375, 5.742216110229492, 40.20051574707031, 6.99591064453125, -19.1217041015625, 5.666107177734375, 24.84539794921875, 23.551025390625, 58.14317321777344, 0.0, 16.309768676757812, 13.01214599609375, 21.577117919921875, -3.8104171752929688, 60.08782958984375, -15.155517578125, 61.968780517578125, 46.780487060546875, 9.896682739257812, 17.121200561523438, 18.467041015625, 2.216064453125, 1.512451171875, 1.8559293746948242, -6.4874267578125, -7.8651123046875, 8.52484130859375, 1.1932830810546875, 19.752838134765625, -7.049774169921875, 15.368316650390625, 8.237335205078125, 78.71566772460938, 7.776947021484375, -0.153167724609375, -37.19964599609375, -4.046234130859375, 13.83929443359375, 35.910125732421875, -0.9744052886962891, 44.84678649902344, 18.2611083984375, 21.567626953125, -4.90838623046875, 0.8575439453125, 72.92800903320312, 5.305135726928711, 71.75286865234375, -3.003509521484375, 51.34722900390625, 27.946853637695312, 37.88067626953125, 17.47344970703125, 35.633026123046875, 59.26264953613281, 10.847957611083984, 32.977081298828125, 55.63018798828125, 56.58424377441406, -6.781166076660156, 52.051422119140625, 45.41876220703125, -6.4046630859375, 59.866172790527344, 37.42657470703125, 42.02070617675781, 8.059906005859375, 23.112030029296875, 14.1722412109375, 3.262510299682617, -3.017242431640625, 57.65789794921875, -72.60443115234375, 76.17575073242188, 65.0916748046875, 6.3882598876953125, -4.2588348388671875, 26.667373657226562, 5.077674865722656, 4.2532958984375, 28.875259399414062, 47.8189697265625, 8.165733337402344, 78.09146118164062, -5.722248077392578, -5.421478271484375, -42.7833251953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000219.npy"}
{"epoch": 0.4586387434554974, "step": 220, "batch_size": 128, "mean": 21.249942779541016, "std": 30.151500701904297, "min": -86.09375, "p10": -15.544330978393553, "median": 22.992164611816406, "p90": 56.493603515625, "max": 98.36317443847656, "pos_frac": 0.75, "sample": [51.450775146484375, 68.52484130859375, 18.569305419921875, 40.13508605957031, 19.019147872924805, 45.1097412109375, 7.083892822265625, 46.06207275390625, 12.209365844726562, 13.78753662109375, -10.223644256591797, 4.75469970703125, 43.827232360839844, 0.25787353515625, 11.348854064941406, 41.158294677734375, 40.8870849609375, 43.4322509765625, -6.87359619140625, 23.064056396484375, 29.182876586914062, 12.744110107421875, 13.652656555175781, 2.4802284240722656, 0.43372154235839844, 98.36317443847656, -14.345611572265625, 48.4915771484375, -15.120559692382812, 40.89002990722656, 11.947845458984375, 4.0947723388671875, -6.70538330078125, -36.83929443359375, 21.314239501953125, 9.34808349609375, -0.28369140625, 24.740478515625, -0.7057781219482422, 76.6214599609375, 11.936256408691406, 25.80206298828125, 55.05072021484375, 25.689727783203125, 70.88510131835938, 48.098854064941406, -3.583606719970703, -86.09375, 32.81646728515625, 35.7353515625, 61.378662109375, 27.140731811523438, -7.348724365234375, 68.0079345703125, -20.145782470703125, 13.666854858398438, -25.46832275390625, 11.123176574707031, 4.578094482421875, 25.708450317382812, -0.41552734375, 37.019100189208984, 55.856414794921875, 36.583587646484375, 5.203483581542969, -9.351570129394531, 10.210723876953125, 37.73583984375, -20.854507446289062, 53.480316162109375, 34.54719543457031, 77.08932495117188, 22.920272827148438, 30.954345703125, 49.8253173828125, 44.414398193359375, -14.947265625, 38.301292419433594, 40.43915557861328, -12.786895751953125, -8.223236083984375, 8.46063232421875, 28.40213394165039, 2.772186279296875, 17.060928344726562, 76.95606231689453, -30.97479248046875, 37.648651123046875, 53.796417236328125, 83.95901489257812, -20.75604248046875, 15.547760009765625, -23.288604736328125, 0.2585029602050781, 9.460845947265625, 78.66828918457031, 39.8172607421875, 13.786483764648438, 28.75311279296875, -16.533130645751953, 51.8404541015625, 39.818572998046875, -28.1912841796875, -29.140228271484375, 0.6565933227539062, -7.8549652099609375, 44.13702392578125, -21.624862670898438, 57.980377197265625, 37.78302001953125, 39.41450500488281, 26.047332763671875, 39.4605712890625, -2.8646697998046875, 78.001708984375, -3.0509033203125, 47.474822998046875, 36.3988037109375, 9.838272094726562, 47.638153076171875, -35.245635986328125, 29.8450927734375, 44.564125061035156, 24.656625747680664, 24.801513671875, -8.560516357421875, 59.51103210449219, -4.00830078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000220.npy"}
{"epoch": 0.4607329842931937, "step": 221, "batch_size": 128, "mean": 16.185644149780273, "std": 29.637531280517578, "min": -61.93511962890625, "p10": -15.583410644531249, "median": 12.124862670898438, "p90": 55.307489013671876, "max": 78.97421264648438, "pos_frac": 0.71875, "sample": [27.30572509765625, 25.167724609375, -11.44818115234375, 12.67913818359375, 15.74853515625, 34.181610107421875, 12.428108215332031, 26.30303955078125, -6.35394287109375, -7.331384658813477, 10.685882568359375, 55.253509521484375, 0.496978759765625, 46.3524169921875, -23.598289489746094, 17.902587890625, 28.849029541015625, -13.786643981933594, 8.628454208374023, -10.728622436523438, 36.438385009765625, 17.307159423828125, 76.03057861328125, -10.782562255859375, 19.65826416015625, 45.502967834472656, 37.5706787109375, 34.22857666015625, 50.18714904785156, 1.50262451171875, 13.6357421875, 9.733642578125, 74.708251953125, -61.93511962890625, -10.848236083984375, 10.050994873046875, 7.53717041015625, 49.14495849609375, -3.4967803955078125, 42.76043701171875, -10.674819946289062, 6.093109130859375, -24.694183349609375, 19.57513427734375, -24.020401000976562, 3.01104736328125, 42.488433837890625, -10.24798583984375, 29.521995544433594, 20.232818603515625, -0.11949729919433594, 51.83580017089844, 12.2283935546875, -10.370330810546875, -6.576808929443359, 64.515869140625, 36.482757568359375, -40.0145263671875, -2.330352783203125, 11.395317077636719, 5.9381103515625, 34.90498352050781, 27.135406494140625, 3.0981597900390625, 27.65521240234375, 1.4593391418457031, 45.97552490234375, 37.88243103027344, -1.1138076782226562, 50.855857849121094, 37.236724853515625, -5.1897430419921875, 23.781005859375, 20.96410369873047, 52.01548767089844, 39.511962890625, 69.638671875, 14.990921020507812, -40.109657287597656, -29.69256591796875, 53.343841552734375, 68.3560791015625, -8.48126220703125, 4.916282653808594, -10.656982421875, 55.433441162109375, 11.575164794921875, -15.050445556640625, 44.31053161621094, 12.021331787109375, 68.9608154296875, 70.41970825195312, 0.7347602844238281, 4.957441329956055, -8.420867919921875, 0.24678802490234375, -18.97058868408203, 78.97421264648438, -57.135986328125, 8.059326171875, 4.3980560302734375, -43.029388427734375, -21.515892028808594, 69.44197082519531, 11.510589599609375, 1.77337646484375, 12.721946716308594, 68.89714050292969, 44.72694396972656, 10.232688903808594, 9.177154541015625, 43.2908935546875, -16.826995849609375, -14.630207061767578, 5.417686462402344, 3.7780075073242188, 59.73876953125, 26.3936767578125, 52.84490966796875, -8.533416748046875, 18.994216918945312, 46.235137939453125, -14.81939697265625, 63.117218017578125, 13.023956298828125, 8.495660781860352, -47.55810546875, 27.931732177734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000221.npy"}
{"epoch": 0.46282722513089003, "step": 222, "batch_size": 128, "mean": 19.888965606689453, "std": 28.327272415161133, "min": -69.9609375, "p10": -8.697505950927733, "median": 14.205612182617188, "p90": 59.604270935058594, "max": 88.39422607421875, "pos_frac": 0.7734375, "sample": [19.892669677734375, 11.970916748046875, 59.80021667480469, 46.2373046875, -29.97161865234375, -39.78678894042969, 50.715057373046875, 13.283281326293945, 14.915496826171875, 6.716346740722656, 0.243927001953125, 20.3924560546875, 51.99053955078125, 3.76318359375, 17.27801513671875, 12.478614807128906, 20.742469787597656, 77.12823486328125, 41.591064453125, 6.7346649169921875, 5.750205993652344, 0.8763580322265625, -13.17376708984375, -27.58758544921875, 10.136627197265625, 88.39422607421875, 12.609588623046875, -7.095489501953125, -43.30511474609375, 7.879642486572266, 3.493011474609375, 33.28739929199219, 55.180633544921875, -8.342018127441406, 14.263282775878906, 10.338497161865234, 18.11328125, 32.1348876953125, 45.44868469238281, 51.123321533203125, 5.5175933837890625, -7.6981964111328125, 20.21148681640625, 2.290008544921875, 48.9150390625, 6.779865264892578, -5.4878387451171875, 33.08148956298828, -69.9609375, 34.03094482421875, 64.08848571777344, 41.472530364990234, -0.7361068725585938, 60.11114501953125, -7.002101898193359, 58.79449462890625, 14.147941589355469, 68.7745361328125, 2.5007553100585938, 2.82830810546875, 14.713836669921875, 16.1474609375, 19.59844970703125, 42.52203369140625, 16.743804931640625, 11.4822998046875, 49.9002685546875, 77.33462524414062, 45.62445068359375, 16.703231811523438, 62.39837646484375, 47.14323425292969, 2.24542236328125, 43.59869384765625, -12.292724609375, -6.8211669921875, 13.821403503417969, 5.43505859375, 70.58905029296875, -9.5269775390625, 19.001312255859375, 10.081502914428711, -15.649835586547852, 11.57904052734375, -24.69439697265625, -0.2500762939453125, 49.12434387207031, -17.0841064453125, -6.5608062744140625, 20.0247802734375, 55.439849853515625, 43.56549072265625, 4.761966705322266, 20.685775756835938, 6.3889923095703125, -11.58642578125, 2.3754730224609375, 45.63511657714844, 22.107177734375, 68.66357421875, 63.7255859375, -6.509513854980469, -4.73370361328125, 23.703994750976562, 64.04750061035156, -1.4169692993164062, 55.554473876953125, 27.322479248046875, 6.164825439453125, 59.520294189453125, 29.584075927734375, 36.944580078125, -38.1368408203125, -1.604248046875, 7.24420166015625, -3.5458526611328125, 11.12664794921875, 40.085845947265625, -5.48870849609375, 47.27459716796875, 39.9439697265625, 10.69744873046875, 67.98970031738281, 49.02256774902344, 7.478271484375, -0.3430309295654297, 11.690277099609375, 49.174407958984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000222.npy"}
{"epoch": 0.4649214659685864, "step": 223, "batch_size": 128, "mean": 22.866228103637695, "std": 30.31200408935547, "min": -51.08275604248047, "p10": -18.5377326965332, "median": 22.790180206298828, "p90": 59.37333374023437, "max": 97.305908203125, "pos_frac": 0.7734375, "sample": [9.714530944824219, -16.872909545898438, 4.479866027832031, 58.8765869140625, 6.986732482910156, 10.047096252441406, -0.345672607421875, -14.196985244750977, 10.677743911743164, 60.53240966796875, 14.85076904296875, 57.122650146484375, 29.9249267578125, 10.653350830078125, 4.16827392578125, 86.74041748046875, 64.48941040039062, -41.41871643066406, 49.395721435546875, -19.254898071289062, 23.492889404296875, 51.3092041015625, 6.965576171875, 22.9207763671875, 14.6041259765625, -18.432861328125, 9.754104614257812, 44.736663818359375, 57.29638671875, -10.96185302734375, 44.40631103515625, 16.80633544921875, 38.46099853515625, 56.06065368652344, 45.06298828125, 57.05821228027344, 88.37860107421875, -33.689697265625, 38.317848205566406, -37.230255126953125, 10.814735412597656, 30.09271240234375, 5.776458740234375, 41.38095474243164, 63.48504638671875, 27.557159423828125, 71.899658203125, 42.623321533203125, 47.63128662109375, 38.63006591796875, -2.268310546875, 15.137969970703125, 53.800994873046875, 79.44195556640625, 50.93559265136719, 28.82122802734375, 1.55926513671875, 58.40948486328125, 15.085182189941406, 37.88177490234375, 33.974151611328125, 22.659584045410156, 24.15008544921875, 97.305908203125, 6.329109191894531, 49.55818176269531, 19.710174560546875, 42.43536376953125, 42.14056396484375, 4.3397979736328125, -1.5808868408203125, 27.321456909179688, 26.504531860351562, 61.03022766113281, -19.44103240966797, 12.660455703735352, -20.712989807128906, -22.268028259277344, 27.16387939453125, -11.777122497558594, 14.863685607910156, -21.540924072265625, 0.696533203125, 3.425294876098633, 50.68486022949219, 37.69744873046875, -8.804840087890625, -18.782432556152344, 41.5965576171875, -7.051723480224609, -51.08275604248047, 35.73899841308594, 15.878707885742188, 64.6820068359375, 55.95233154296875, 71.9986572265625, 58.872894287109375, 35.135963439941406, 33.63634490966797, 34.352294921875, -0.3230628967285156, 35.16438293457031, 16.8638916015625, 40.74995422363281, 42.4755859375, 7.697052001953125, 51.732177734375, 73.32302856445312, -2.4507522583007812, 0.3842620849609375, 66.11605834960938, -32.34113311767578, -7.5521240234375, -3.3885498046875, 58.680908203125, 29.61511993408203, -37.10398483276367, 3.8357391357421875, 15.603073120117188, 17.65869140625, -17.85314178466797, 50.809043884277344, 7.57415771484375, 29.94415283203125, 12.563224792480469, 10.10626220703125, -19.804290771484375, -13.240463256835938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000223.npy"}
{"epoch": 0.46701570680628274, "step": 224, "batch_size": 128, "mean": 21.96862030029297, "std": 28.216346740722656, "min": -53.90863037109375, "p10": -12.471447753906249, "median": 19.82593536376953, "p90": 61.69992370605468, "max": 90.2982177734375, "pos_frac": 0.8125, "sample": [76.42886352539062, 90.2982177734375, 19.074249267578125, 28.046722412109375, 2.2605152130126953, 4.1242218017578125, -28.941543579101562, 24.31878662109375, 51.224143981933594, 10.855339050292969, 6.54986572265625, 44.367759704589844, 28.097305297851562, 23.009414672851562, 24.850196838378906, 21.818115234375, 5.75067138671875, 46.47265625, 0.8332748413085938, -23.522781372070312, 27.628494262695312, -12.030731201171875, 79.40499877929688, 66.88555908203125, 36.52418518066406, 35.16803741455078, 42.658905029296875, 9.256820678710938, 18.36627197265625, 52.88655090332031, 62.151885986328125, 54.69465637207031, -53.90863037109375, 42.30989074707031, 19.856063842773438, 14.56103515625, 17.227584838867188, 38.512664794921875, 37.68305969238281, 22.801849365234375, 1.2881317138671875, -8.549346923828125, 51.165016174316406, 23.02999496459961, 38.38041687011719, 23.491622924804688, 37.253326416015625, -17.537452697753906, 56.79705810546875, -3.978395462036133, 27.526222229003906, 29.967620849609375, 0.0, 61.515289306640625, 52.192813873291016, 10.469913482666016, -4.10992431640625, -28.318313598632812, 4.687164306640625, 23.65350341796875, 9.51190185546875, 16.696990966796875, 3.9625816345214844, 44.628082275390625, 0.0953521728515625, 42.786590576171875, -13.499786376953125, -11.014434814453125, 10.081451416015625, -30.454635620117188, -0.160888671875, 72.9541015625, 7.2184295654296875, 45.538604736328125, 14.95941162109375, 14.928009033203125, -4.269889831542969, 38.2099609375, 8.777084350585938, 67.64517211914062, -10.950454711914062, -9.654876708984375, -28.063217163085938, 9.023868560791016, 12.917701721191406, 51.77459716796875, 31.48773193359375, -13.773101806640625, 62.581207275390625, -51.17432403564453, 57.44871520996094, 16.854278564453125, 45.2542724609375, 62.1307373046875, 20.13568115234375, 14.728302001953125, 36.622650146484375, 19.795806884765625, 32.79779052734375, 5.402191162109375, -15.863853454589844, 25.8892822265625, 4.0524139404296875, 13.866485595703125, 16.094619750976562, 17.438499450683594, 86.1754150390625, -23.635292053222656, 71.111083984375, 28.722991943359375, 18.77111053466797, -25.74786376953125, 4.322509765625, 3.95379638671875, -7.02117919921875, 37.31803894042969, 37.711151123046875, 5.2287750244140625, 42.197357177734375, 3.8421630859375, 5.015232086181641, 60.609832763671875, 43.295440673828125, 30.19500732421875, 62.433624267578125, 45.27293395996094, 4.288051605224609, 65.03244018554688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000224.npy"}
{"epoch": 0.46910994764397906, "step": 225, "batch_size": 128, "mean": 17.284034729003906, "std": 34.98202896118164, "min": -62.56243896484375, "p10": -29.42159309387207, "median": 12.926895141601562, "p90": 67.6424057006836, "max": 97.76632690429688, "pos_frac": 0.671875, "sample": [64.79486083984375, -29.90582275390625, 51.2049560546875, 56.352203369140625, 56.25547790527344, 7.90240478515625, -29.257991790771484, -40.07037353515625, 75.34786987304688, 70.05824279785156, 18.899612426757812, 57.296905517578125, 25.559906005859375, -5.8027801513671875, 77.73757934570312, 21.9976806640625, 78.35086059570312, 51.674835205078125, 5.0469818115234375, 8.835845947265625, -8.228240966796875, -11.009124755859375, 14.722152709960938, -3.3381271362304688, 68.75970458984375, -2.509918212890625, 72.64797973632812, -32.277496337890625, 3.22210693359375, -35.18939208984375, 0.6302642822265625, -44.304840087890625, 48.54252624511719, 40.9478759765625, 7.2500152587890625, 18.23645782470703, 46.209503173828125, -2.8956031799316406, 4.334625244140625, -10.85003662109375, 35.1009521484375, 75.617431640625, -23.01264190673828, -33.79124450683594, -42.02655029296875, -34.887603759765625, 38.95515441894531, 51.911102294921875, -8.292491912841797, 23.62432861328125, -20.6278076171875, 17.090087890625, 24.037399291992188, 56.6304931640625, -5.529266357421875, 15.631103515625, -25.0792236328125, 23.157958984375, 62.6300048828125, 39.31123352050781, 47.86451721191406, 68.34596252441406, 79.0489501953125, -9.35546875, 21.5831298828125, -8.752166748046875, -17.17779541015625, 45.5174560546875, 20.574569702148438, 69.84292602539062, 1.697174072265625, 60.3187255859375, 27.23699951171875, 46.39435577392578, -12.8427734375, 11.5615234375, 59.40138244628906, 61.4683837890625, -19.4525146484375, 45.49577331542969, -3.6177825927734375, -37.0091552734375, -15.609306335449219, 3.654359817504883, -42.30583190917969, -25.916778564453125, 20.55238914489746, 0.5443382263183594, 38.35923767089844, -62.56243896484375, 27.48162841796875, 3.5555419921875, -22.58868408203125, -4.327392578125, 35.75494384765625, -22.75982666015625, 1.0804443359375, 8.181846618652344, 97.76632690429688, 37.9407958984375, 2.7993545532226562, 43.970237731933594, 64.47714233398438, 67.34088134765625, -15.275115966796875, 38.86485290527344, -32.627479553222656, 12.560226440429688, 11.0484619140625, 86.58807373046875, -9.469436645507812, 27.328857421875, 50.159637451171875, 50.43641662597656, 1.0870361328125, 19.706077575683594, 8.070030212402344, -13.4891357421875, 22.020742416381836, -29.803329467773438, 7.0346832275390625, 1.5659141540527344, -17.03137969970703, 8.68133544921875, -13.233932495117188, 13.293563842773438, 23.633163452148438, 78.0438232421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000225.npy"}
{"epoch": 0.4712041884816754, "step": 226, "batch_size": 128, "mean": 15.30355453491211, "std": 29.928668975830078, "min": -62.887786865234375, "p10": -22.306690216064453, "median": 12.21255874633789, "p90": 55.89385223388671, "max": 84.66061401367188, "pos_frac": 0.7109375, "sample": [14.32879638671875, 84.66061401367188, -6.86138916015625, 33.520751953125, 58.861541748046875, 66.50088500976562, 8.173721313476562, -1.8248710632324219, 16.001754760742188, 52.258880615234375, 69.0956039428711, 33.32127380371094, 5.63140869140625, 68.95774841308594, 7.756195068359375, 8.878429412841797, 17.242599487304688, -19.04998779296875, 54.90815353393555, 62.092437744140625, 19.67475128173828, 0.1104888916015625, 44.310760498046875, 36.729759216308594, 0.3110809326171875, 15.035369873046875, -16.250032424926758, -47.73187255859375, 47.315834045410156, 20.01068115234375, 0.5630893707275391, 51.077484130859375, 38.66424560546875, 34.92864990234375, -8.125259399414062, 11.40118408203125, 55.48490905761719, -62.887786865234375, -0.86590576171875, 31.777450561523438, -36.93086242675781, 9.577476501464844, -23.216110229492188, -2.4988059997558594, 17.498504638671875, 30.634857177734375, 13.96856689453125, -37.02070617675781, -28.0787353515625, 35.634368896484375, 5.470985412597656, -35.75518798828125, 54.897369384765625, 3.5166893005371094, 7.769092559814453, -32.652740478515625, 3.9315567016601562, 20.6414794921875, 0.04296875, -1.7439041137695312, -11.812400817871094, 26.799835205078125, 13.023933410644531, -25.954544067382812, 26.60528564453125, -6.579521179199219, 7.8561553955078125, 59.079345703125, 3.4906044006347656, 5.40252685546875, -17.875411987304688, 5.073705673217773, 17.605316162109375, -11.700775146484375, 17.99767303466797, 2.58447265625, 5.65447998046875, 40.419891357421875, 6.420372009277344, 56.848052978515625, -43.87263488769531, 50.880653381347656, 60.898040771484375, -43.720458984375, 70.44430541992188, -1.889007568359375, 9.667953491210938, 13.1207275390625, -0.440216064453125, 52.87835693359375, 44.968048095703125, 54.5289306640625, 51.77313995361328, 20.04180908203125, 36.87066650390625, 22.023590087890625, -8.716522216796875, 7.595001220703125, 46.168174743652344, 9.561813354492188, 72.74363708496094, 14.59478759765625, 66.43789672851562, 1.11297607421875, 2.7894287109375, 14.941329956054688, 46.29212951660156, 19.750762939453125, 45.9912109375, -11.156097412109375, 25.847564697265625, -26.353546142578125, -0.6361923217773438, 44.1280517578125, 3.6055679321289062, -0.5192489624023438, 45.797637939453125, -11.33188247680664, -21.75689697265625, 68.6016845703125, 23.30206298828125, -21.91693878173828, -33.32725524902344, 19.041080474853516, -20.13629150390625, -1.9461669921875, 14.28485107421875, -6.7247467041015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000226.npy"}
{"epoch": 0.4732984293193717, "step": 227, "batch_size": 128, "mean": 21.18825340270996, "std": 30.974742889404297, "min": -67.2786865234375, "p10": -14.590191650390617, "median": 20.19432830810547, "p90": 63.14800109863281, "max": 91.77151489257812, "pos_frac": 0.7578125, "sample": [-34.708770751953125, 31.792633056640625, 10.862091064453125, 19.175582885742188, 42.501953125, 34.07209014892578, 32.94378662109375, 20.69506072998047, -2.6615142822265625, 24.039779663085938, 62.482513427734375, 37.72834777832031, 11.302932739257812, -6.044708251953125, 4.5651397705078125, 29.446870803833008, 0.0, 70.7899169921875, -67.2786865234375, -42.57535934448242, 14.385955810546875, -42.658935546875, 51.65010452270508, 68.70309448242188, 91.77151489257812, 91.08023071289062, 51.663665771484375, -25.856094360351562, 50.29352951049805, -12.235015869140625, -6.473876953125, 57.12956237792969, 12.279808044433594, 12.499191284179688, 25.6138916015625, -11.3255615234375, 18.031959533691406, 21.981712341308594, 20.952056884765625, 63.1383056640625, 63.170623779296875, 61.78277587890625, 9.981109619140625, 28.6724853515625, 26.472625732421875, -6.6773834228515625, 19.3056640625, 0.416717529296875, -8.484611511230469, 20.9271240234375, 69.43264770507812, 19.69359588623047, -8.40716552734375, -23.955352783203125, 21.38299560546875, 78.41679382324219, 5.2276611328125, 37.41432189941406, 69.24824523925781, 28.86893081665039, 73.08047485351562, -47.49656677246094, -0.060302734375, 47.755096435546875, -61.2930908203125, 19.31103515625, 15.531997680664062, 71.17050170898438, -21.205047607421875, 26.23291015625, 43.21043395996094, 7.4996337890625, 3.3072891235351562, 39.38653564453125, 61.164031982421875, 9.3485107421875, 32.47357177734375, 28.99725341796875, -31.809600830078125, 45.69920349121094, 68.55975341796875, 5.5287017822265625, 18.864395141601562, 61.889373779296875, -0.55633544921875, 37.94642639160156, 5.09210205078125, 22.914031982421875, 12.464553833007812, 13.964752197265625, -20.085601806640625, 38.55493927001953, 13.6077880859375, -20.648887634277344, 59.366058349609375, 32.887451171875, 47.52635192871094, 4.795692443847656, -10.742774963378906, 66.79366302490234, -1.87152099609375, 67.25677490234375, 4.187505722045898, 37.399658203125, -25.76128387451172, 49.02557373046875, -2.9525413513183594, -4.137176513671875, 1.1991653442382812, 10.033584594726562, 31.6552734375, -5.1101226806640625, 58.2830810546875, 52.49855422973633, 12.7886962890625, 23.322998046875, 28.33312225341797, 32.6519775390625, 32.877197265625, -2.829925537109375, -8.529449462890625, 12.812210083007812, 13.16705322265625, 24.846702575683594, 8.700363159179688, 7.362768173217773, 41.93170928955078, 47.28355407714844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000227.npy"}
{"epoch": 0.47539267015706804, "step": 228, "batch_size": 128, "mean": 22.008136749267578, "std": 29.71903419494629, "min": -50.39434814453125, "p10": -15.933514404296876, "median": 20.736434936523438, "p90": 59.54017181396484, "max": 89.5592041015625, "pos_frac": 0.7265625, "sample": [-18.600189208984375, -18.990509033203125, 52.396240234375, 56.445159912109375, 43.015716552734375, -4.792476654052734, 77.52276611328125, 23.948654174804688, 26.34600830078125, 60.564117431640625, 17.060287475585938, -32.96842956542969, -1.099395751953125, 73.81854248046875, 5.05792236328125, 28.383956909179688, -3.4272003173828125, 45.677833557128906, -19.711685180664062, 37.28093719482422, 55.500030517578125, 13.978729248046875, -29.417579650878906, -15.93096923828125, 11.845016479492188, 46.17811584472656, 20.611480712890625, 3.0387229919433594, -12.473907470703125, 10.071067810058594, 10.426788330078125, 16.316848754882812, -17.138656616210938, 34.3089599609375, 41.934112548828125, 56.655372619628906, 32.127197265625, 13.839736938476562, -2.80316162109375, -6.6746368408203125, 57.63441467285156, -4.585235595703125, -12.97900390625, -7.7843017578125, 35.365509033203125, 7.107330322265625, 0.126708984375, 15.270843505859375, 29.485931396484375, 49.25843811035156, 28.70721435546875, 81.50006103515625, 35.26609802246094, 10.0323486328125, 72.90841674804688, 29.65380859375, 44.7127685546875, 20.301666259765625, 47.52630615234375, 72.041259765625, 1.5869598388671875, -22.431533813476562, -4.6004638671875, -2.4854812622070312, 39.04734802246094, 0.250030517578125, 0.0, 23.475982666015625, 18.986572265625, 6.7323455810546875, 11.876235961914062, 71.18255615234375, 27.37181854248047, -0.884002685546875, -25.027923583984375, 52.370361328125, -14.812286376953125, 52.280517578125, -4.6470489501953125, 48.77093505859375, -12.45880126953125, -26.113914489746094, 12.280517578125, 44.50634765625, 58.80995178222656, 68.736572265625, 3.38543701171875, -13.74090576171875, 72.67298889160156, 23.64837646484375, -15.939453125, 59.43145751953125, 30.573455810546875, 55.764312744140625, 31.97502899169922, 43.559810638427734, -4.25311279296875, 3.9278717041015625, 11.753433227539062, 59.79383850097656, 48.080108642578125, -32.59022521972656, -7.723594665527344, 26.749197006225586, 45.165069580078125, 45.45343017578125, 49.6815185546875, -50.39434814453125, -14.02423095703125, -0.957794189453125, 12.604934692382812, 36.425933837890625, 7.390594482421875, 20.86138916015625, 32.20574951171875, 80.446533203125, 52.60699462890625, 33.04779815673828, 12.104452133178711, -31.26983642578125, 34.542633056640625, 47.65240478515625, 67.22317504882812, 28.395492553710938, 89.5592041015625, 8.832191467285156, 29.66735076904297, 16.077117919921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000228.npy"}
{"epoch": 0.4774869109947644, "step": 229, "batch_size": 128, "mean": 27.261943817138672, "std": 31.731237411499023, "min": -45.53704833984375, "p10": -13.961727905273436, "median": 25.177261352539062, "p90": 71.02596740722656, "max": 119.95126342773438, "pos_frac": 0.8125, "sample": [50.03936767578125, -4.43585205078125, 26.617950439453125, 46.96466064453125, -7.399200439453125, -36.67958068847656, 19.579849243164062, 41.94953918457031, 31.40936279296875, -27.77904510498047, 82.998779296875, 80.66944885253906, 23.18639373779297, -0.9059028625488281, 85.68313598632812, 56.25043487548828, 5.8490447998046875, 6.025390625, 16.96677017211914, -1.76275634765625, 25.100814819335938, 19.56591796875, 55.87994384765625, 8.9537353515625, 19.462692260742188, -15.473388671875, 9.256156921386719, 4.317558288574219, 68.77285766601562, 31.246978759765625, 11.709808349609375, 32.574676513671875, 78.13250732421875, 48.58587646484375, 33.07421875, 72.5296630859375, 24.910079956054688, 18.441558837890625, 87.38168334960938, -45.53704833984375, 9.2735595703125, -40.703338623046875, 33.30616760253906, 46.7418212890625, 25.607433319091797, 80.07513427734375, 119.95126342773438, 62.30799865722656, 40.19041442871094, 54.71412658691406, 25.253707885742188, -1.2424850463867188, 12.650352478027344, 9.74102783203125, 46.538352966308594, 48.99525451660156, 57.316070556640625, 8.00604248046875, 8.002487182617188, -31.550918579101562, 20.063230514526367, -38.93971633911133, 64.02249145507812, 43.9012451171875, -5.4660797119140625, -13.6907958984375, 35.77203369140625, 9.632217407226562, 46.8719482421875, 1.889404296875, 2.2582931518554688, 24.628311157226562, -0.14313316345214844, 13.107391357421875, 50.083221435546875, 27.89814567565918, 6.014793395996094, 58.43254852294922, 70.76336669921875, 38.55389404296875, 58.66355895996094, 45.187164306640625, 21.89837646484375, 52.97648620605469, 16.366714477539062, 8.5914306640625, -30.684112548828125, 69.20266723632812, 85.388427734375, 25.307029724121094, 81.60806274414062, 52.1107177734375, 6.438413619995117, 13.239166259765625, 11.626953125, 61.965576171875, 35.18927001953125, 65.06265258789062, 20.6561279296875, 15.278182983398438, 35.32963562011719, -11.433021545410156, 31.1309814453125, 71.95016479492188, -2.1040496826171875, 4.163665771484375, 27.573196411132812, -17.417037963867188, 10.805950164794922, 71.63870239257812, 35.95289611816406, -7.461143493652344, 20.5533447265625, 36.8172607421875, 18.877853393554688, 21.7789306640625, 48.767333984375, -14.593902587890625, 38.120391845703125, -23.380615234375, 74.72021484375, 15.659423828125, 28.162551879882812, 27.984710693359375, 50.5203857421875, -28.911144256591797, 68.137939453125, -14.830177307128906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000229.npy"}
{"epoch": 0.47958115183246075, "step": 230, "batch_size": 128, "mean": 23.39605140686035, "std": 32.774356842041016, "min": -98.68267822265625, "p10": -10.241714477539062, "median": 15.934955596923828, "p90": 65.02769927978515, "max": 105.13504028320312, "pos_frac": 0.8125, "sample": [63.77081298828125, 34.219635009765625, 35.78997802734375, 54.243675231933594, 105.13504028320312, 37.916473388671875, 78.96186828613281, -2.26141357421875, 71.61273193359375, 38.796966552734375, 14.529441833496094, 20.796234130859375, 12.325355529785156, -36.422298431396484, -98.68267822265625, 51.61637878417969, 56.93208312988281, -2.386547088623047, -4.825042724609375, -42.93499755859375, 2.21795654296875, 9.774852752685547, 76.96273803710938, -18.421615600585938, 0.0, 65.02490234375, -2.330535888671875, 8.395015716552734, 10.53741455078125, 48.12980651855469, 46.45640563964844, 61.24530029296875, 0.7679595947265625, 34.89723205566406, 0.08423995971679688, -24.754547119140625, 52.519287109375, 50.97465515136719, 30.47021484375, 14.915786743164062, 0.6591529846191406, -20.589111328125, 6.9318695068359375, 30.8265380859375, 6.338464736938477, 60.2344970703125, 32.64756774902344, 9.824462890625, 61.15509033203125, -10.61181640625, -3.651823043823242, 16.072616577148438, -23.201416015625, 38.792449951171875, 43.824371337890625, -16.649139404296875, 49.598541259765625, 4.51806640625, 55.69061279296875, 22.917022705078125, -15.106826782226562, 25.7431640625, 85.083984375, 48.140769958496094, 12.118614196777344, 4.987091064453125, -10.083099365234375, 57.1089973449707, 5.5146942138671875, 49.648345947265625, 34.25933837890625, 6.591278076171875, -6.20904541015625, -7.5210418701171875, 18.33721923828125, 14.2791748046875, 7.7266845703125, 15.797294616699219, 80.25222778320312, 1.9376068115234375, 3.571077346801758, 42.29248046875, 20.42755126953125, -5.757726669311523, 12.5067138671875, -77.21832275390625, 4.079505920410156, -11.611076354980469, 75.78106689453125, 50.380615234375, 63.997772216796875, 28.858814239501953, 11.997940063476562, 6.62371826171875, 77.73605346679688, 67.18844604492188, 1.6958427429199219, 7.38507080078125, 56.02699279785156, 42.83409118652344, 61.84745788574219, 0.7155685424804688, 40.795249938964844, 4.947233200073242, -22.157432556152344, 45.6693115234375, -0.6021728515625, 68.1405029296875, 37.94908142089844, 2.917724609375, 42.747867584228516, 65.03422546386719, 33.81559753417969, 2.5591812133789062, 35.90631103515625, 30.51209259033203, 5.241189956665039, 73.31700134277344, 98.5133056640625, 5.382545471191406, 1.4883270263671875, 21.80963134765625, 8.090911865234375, 3.814849853515625, 22.568359375, 47.675575256347656, 12.983108520507812, 13.280197143554688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000230.npy"}
{"epoch": 0.4816753926701571, "step": 231, "batch_size": 128, "mean": 18.71527099609375, "std": 28.23818016052246, "min": -54.60139465332031, "p10": -13.10525970458984, "median": 16.69886016845703, "p90": 55.7550048828125, "max": 89.08633422851562, "pos_frac": 0.75, "sample": [17.948318481445312, 17.11602783203125, 1.898345947265625, -18.48638916015625, 46.477455139160156, 55.43157958984375, 32.257080078125, -19.201919555664062, 37.1929931640625, -54.60139465332031, 59.093231201171875, -22.667221069335938, 45.792266845703125, 43.588623046875, -2.641571044921875, 39.69972229003906, 8.417396545410156, 5.912662506103516, 11.761642456054688, -4.515625, -20.984115600585938, -11.681488037109375, -5.6453857421875, 33.88726806640625, 13.0068359375, 50.397430419921875, 12.705047607421875, 44.758811950683594, 4.491485595703125, -34.6787109375, 45.80218505859375, 56.206573486328125, 33.008209228515625, 2.017557144165039, 49.75909423828125, -8.304153442382812, 74.64450073242188, 41.79583740234375, 1.7810211181640625, 30.7359619140625, 0.0801239013671875, 27.3612060546875, -1.329376220703125, 2.3446197509765625, 66.04702758789062, 31.4993896484375, -7.2830810546875, 20.168350219726562, 6.124267578125, 22.081771850585938, 55.5850830078125, -1.74432373046875, 8.289634704589844, 11.043060302734375, 5.9813232421875, 13.43096923828125, 39.38787841796875, -11.8865966796875, 16.641311645507812, 67.92156982421875, -11.511123657226562, 3.4241714477539062, -1.6462860107421875, 24.60205078125, 1.2667980194091797, 24.119384765625, 29.817840576171875, 26.714324951171875, 16.75640869140625, 22.41619873046875, 36.05369567871094, 26.0379638671875, 81.19183349609375, -33.1875, -46.4095458984375, -2.12554931640625, 12.193504333496094, -6.8692626953125, 48.22747802734375, 29.711395263671875, 14.3128662109375, 7.5390625, 23.497589111328125, 6.16778564453125, -0.47782135009765625, 9.16741943359375, -0.00872802734375, 67.55593872070312, 27.858932495117188, 13.342941284179688, 40.59222412109375, 42.8497314453125, 34.44575500488281, 43.63206481933594, -15.948806762695312, 31.156661987304688, 9.202335357666016, 28.8231201171875, 80.74755859375, 0.8515853881835938, -28.585830688476562, 21.44439697265625, 19.060501098632812, 89.08633422851562, 42.67164611816406, -2.669219970703125, 9.929847717285156, 56.1514892578125, 6.194580078125, 17.083145141601562, -4.0781402587890625, 7.527378082275391, 44.71998596191406, -25.15582275390625, 40.4627685546875, 72.33409118652344, 4.395843505859375, 66.89080810546875, 20.202621459960938, 31.023452758789062, -10.703262329101562, 7.07586669921875, 59.99226379394531, -40.90325927734375, 54.725250244140625, 0.0, -38.74562072753906, 43.410247802734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000231.npy"}
{"epoch": 0.4837696335078534, "step": 232, "batch_size": 128, "mean": 19.886363983154297, "std": 31.380245208740234, "min": -44.58558654785156, "p10": -17.341439819335935, "median": 12.210723876953125, "p90": 59.321495056152344, "max": 108.75384521484375, "pos_frac": 0.6953125, "sample": [-14.50360107421875, 59.35968017578125, 0.30783843994140625, 54.16168212890625, 21.668914794921875, 23.829254150390625, 50.761993408203125, 25.647003173828125, 13.08990478515625, -3.67913818359375, 24.770172119140625, 70.97769165039062, 27.666290283203125, -15.638381958007812, -15.361557006835938, 16.85710906982422, 4.776863098144531, 2.678741455078125, 11.33154296875, 17.6044921875, 17.06488037109375, 8.792884826660156, -44.58558654785156, 26.931365966796875, 53.771636962890625, 37.43328857421875, 34.10102081298828, 8.76348876953125, 0.7320175170898438, 71.92138671875, 10.597381591796875, -6.849891662597656, -27.082550048828125, 38.7725830078125, 49.6343994140625, 59.083587646484375, -17.697784423828125, 19.315765380859375, 5.801483154296875, 57.4537353515625, 10.567886352539062, -22.188934326171875, 0.2237873077392578, 7.369346618652344, 10.334930419921875, 1.656595230102539, 47.003997802734375, 6.913238525390625, 50.54121398925781, -1.2197113037109375, 22.5914306640625, -19.30499267578125, 56.04828643798828, 16.672744750976562, -1.5814056396484375, 41.81982421875, 20.093765258789062, 61.128562927246094, -0.010406494140625, 53.69551086425781, 7.738525390625, 87.61651611328125, 59.30513000488281, -21.6328125, 53.806640625, 48.67466735839844, -2.34429931640625, 2.103668212890625, -4.9158782958984375, -5.943748474121094, 8.986625671386719, 60.498382568359375, 8.91241455078125, -11.64715576171875, 79.23622131347656, 81.62545776367188, -3.274078369140625, -16.09405517578125, -0.279571533203125, 34.916046142578125, 1.8163471221923828, 3.881683349609375, -17.188720703125, 36.31707000732422, 45.7901611328125, -14.190227508544922, -4.6109619140625, 33.70771026611328, -3.463165283203125, -18.69146728515625, 54.43031311035156, 23.079620361328125, 0.0, 72.27276611328125, 41.44355010986328, 6.193553924560547, 50.707366943359375, 23.016212463378906, 41.521209716796875, 52.57891845703125, 53.35687255859375, 29.5289306640625, 85.89053344726562, -33.7313232421875, 48.70729064941406, 108.75384521484375, 8.74072265625, -13.52197265625, 45.714813232421875, 9.928985595703125, -4.423828125, 78.97540283203125, -39.76689147949219, -9.67095947265625, -12.723297119140625, -29.3675537109375, 33.44615936279297, -0.11529922485351562, 29.543365478515625, -27.871612548828125, 64.75045776367188, -2.1676025390625, -44.49964904785156, 55.000213623046875, 49.137542724609375, 39.71820831298828, 1.364166259765625, -17.76092529296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000232.npy"}
{"epoch": 0.48586387434554973, "step": 233, "batch_size": 128, "mean": 21.953235626220703, "std": 30.623516082763672, "min": -46.64909362792969, "p10": -14.596173095703124, "median": 16.613847732543945, "p90": 65.41869049072265, "max": 83.62542724609375, "pos_frac": 0.78125, "sample": [-35.94856262207031, 53.54566192626953, 0.9781494140625, -5.2557373046875, 4.665863037109375, 64.64250183105469, 6.797088623046875, 33.89567565917969, 7.835239410400391, 7.467002868652344, 32.789337158203125, 22.08627700805664, 51.1099853515625, 33.37701416015625, 16.65225601196289, 1.649139404296875, 46.804901123046875, 55.330322265625, 15.85638427734375, 73.90078735351562, 58.5465087890625, 59.10523223876953, 65.98202514648438, 45.68797302246094, 34.62531280517578, 69.19393157958984, 57.74920654296875, 23.2569580078125, 48.59617614746094, -24.076278686523438, 24.87255859375, -4.7578582763671875, 61.20610046386719, 2.52764892578125, 65.17726135253906, 20.5911865234375, 14.322662353515625, -34.3427734375, 38.10910415649414, 78.95805358886719, -5.811027526855469, -4.88416862487793, 10.585105895996094, 11.955810546875, -13.924652099609375, 71.13885498046875, 25.842483520507812, 16.01272201538086, 13.964691162109375, 3.6438827514648438, 75.3914794921875, 76.23180389404297, 15.832046508789062, -6.1861724853515625, 23.823577880859375, -3.81109619140625, 77.3013916015625, 4.611083984375, 53.21403503417969, 8.32513427734375, 25.3360595703125, 57.685546875, -36.30975341796875, 12.118976593017578, 16.575439453125, -14.26531982421875, -16.988723754882812, 3.987873077392578, -43.451904296875, -18.52386474609375, 54.24578857421875, 69.29461669921875, 53.48797607421875, 32.73445129394531, 79.679931640625, 36.7818603515625, 14.7447509765625, 4.53759765625, 12.22294807434082, 11.093765258789062, 6.640987396240234, -0.3206939697265625, 11.2144775390625, 44.773101806640625, -15.921432495117188, 4.514142990112305, 83.62542724609375, 23.81280517578125, -15.3681640625, 28.233551025390625, 19.311588287353516, -26.526870727539062, -42.1478271484375, 80.01693725585938, 14.415184020996094, -14.133056640625, 2.050067901611328, 49.1788330078125, -0.94915771484375, 59.4774169921875, -7.91021728515625, 5.249092102050781, -10.616859436035156, -46.64909362792969, 44.571258544921875, 47.174896240234375, 14.497161865234375, 40.92418670654297, 25.468292236328125, -32.64976501464844, 1.336273193359375, 35.608642578125, 56.65667724609375, -1.3941497802734375, 40.60723876953125, 23.090286254882812, -12.124404907226562, 71.15280151367188, 17.23302459716797, 25.27203369140625, 20.089685440063477, 4.757417678833008, 0.0244140625, 36.0267333984375, 53.08013916015625, 14.477386474609375, 13.244338989257812, 17.164154052734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000233.npy"}
{"epoch": 0.48795811518324606, "step": 234, "batch_size": 128, "mean": 22.740859985351562, "std": 35.033668518066406, "min": -58.234619140625, "p10": -18.294773864746094, "median": 16.804306030273438, "p90": 68.7071563720703, "max": 103.326171875, "pos_frac": 0.75, "sample": [-14.153915405273438, -50.989013671875, -8.6407470703125, 2.203777313232422, 38.646270751953125, 58.30311584472656, 69.765380859375, 35.87725830078125, -18.12933349609375, 74.36532592773438, 56.60356140136719, -53.09625244140625, 52.042388916015625, 45.02247619628906, -14.022216796875, 30.805328369140625, 23.33513641357422, 12.778228759765625, 6.842437744140625, 78.52545166015625, 46.45655822753906, 16.293548583984375, 31.831329345703125, 80.47158813476562, 11.301177978515625, 1.2354507446289062, 40.3995361328125, 50.78703308105469, 54.70021057128906, 30.232391357421875, 14.966842651367188, 41.74517822265625, 52.523529052734375, 88.98281860351562, 55.0123291015625, 14.80267333984375, 26.42510986328125, 4.015350341796875, 4.495506286621094, 17.3150634765625, 42.9124755859375, -0.2268829345703125, 7.700408935546875, -11.2237548828125, 15.710205078125, -8.4951171875, 39.898536682128906, 46.44286346435547, -6.9554290771484375, 66.96912384033203, 6.542640686035156, 7.76397705078125, 71.04840087890625, 28.161319732666016, -2.2892684936523438, 93.15463256835938, 55.00706481933594, 29.02703094482422, 48.180667877197266, 35.86572265625, 7.378452301025391, 6.765592575073242, 65.82705688476562, 9.179168701171875, 6.745420455932617, -3.2012939453125, -18.680801391601562, 0.0, 37.9595947265625, -6.83135986328125, 10.230384826660156, 24.261550903320312, -35.146751403808594, 9.981956481933594, 58.97746276855469, 87.67398071289062, 63.840240478515625, 15.882461547851562, -31.3724365234375, 39.483306884765625, 0.8304347991943359, 49.981597900390625, 68.25363159179688, -16.6600341796875, 36.82469177246094, 18.63800048828125, 47.51511001586914, 6.2137451171875, -56.892486572265625, 82.5379638671875, -32.802276611328125, 71.77850341796875, 11.939544677734375, 14.68701171875, 55.04191589355469, -1.0472259521484375, 49.636417388916016, 28.8170166015625, 0.38169097900390625, 6.7890625, 44.98370361328125, 0.0, 12.491844177246094, 10.15875244140625, 103.326171875, 77.37469482421875, -36.27110290527344, -8.427642822265625, 81.03887939453125, 45.7078857421875, 35.752662658691406, -5.347991943359375, -58.234619140625, 20.974609375, 44.42498779296875, 63.5225830078125, -24.62445068359375, -15.935211181640625, -30.31304931640625, 2.913818359375, 32.07855224609375, 1.5653839111328125, -41.3939208984375, 13.564857482910156, -45.286895751953125, 65.47946166992188, -13.152557373046875, 67.767822265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000234.npy"}
{"epoch": 0.4900523560209424, "step": 235, "batch_size": 128, "mean": 20.79585838317871, "std": 34.014461517333984, "min": -77.49786376953125, "p10": -19.723456573486327, "median": 21.32250213623047, "p90": 60.72545928955078, "max": 96.57412719726562, "pos_frac": 0.6953125, "sample": [13.190338134765625, 6.104316711425781, -3.3922119140625, 2.2640762329101562, 10.132831573486328, -44.87457275390625, -19.868309020996094, -5.318096160888672, 47.96510314941406, 43.435394287109375, -0.24102783203125, 83.454833984375, 57.751953125, 39.924346923828125, 33.60673522949219, 22.118370056152344, 52.794944763183594, 32.9097900390625, -1.655618667602539, -2.91375732421875, 46.32659149169922, 18.492431640625, 31.532623291015625, 39.85284423828125, 77.79011535644531, 56.393585205078125, 55.412506103515625, -15.590591430664062, 50.545494079589844, -8.943367004394531, 61.414794921875, -4.0780029296875, 35.336517333984375, 59.96351623535156, 22.1700439453125, 49.04943084716797, 75.58963012695312, 55.89105224609375, 6.869941711425781, 85.67718505859375, -6.561859130859375, 40.111083984375, -18.479942321777344, -25.08660888671875, 44.793365478515625, 84.96435546875, -53.48638916015625, -24.92470932006836, -40.826507568359375, 29.3653564453125, 54.492919921875, -10.774810791015625, -0.2664966583251953, -34.4161376953125, 31.5994873046875, 21.23980712890625, -13.047332763671875, 55.30303955078125, -9.535888671875, -29.984954833984375, 11.04062271118164, 14.96856689453125, 60.52894592285156, -11.359222412109375, 87.67733764648438, 21.443344116210938, -8.5313720703125, -58.58513259887695, 4.72479248046875, 37.246917724609375, -19.661376953125, 37.978607177734375, -9.395645141601562, 46.66810607910156, 6.543449401855469, 8.17694091796875, 6.51812744140625, 66.51496887207031, 3.34832763671875, 20.326141357421875, 23.45294189453125, 26.372817993164062, -46.791107177734375, 21.405197143554688, 15.25360107421875, -13.141490936279297, 50.72284698486328, 51.38382339477539, 48.991241455078125, 4.536376953125, 96.57412719726562, 46.369171142578125, 32.6630859375, 89.43405151367188, 61.183990478515625, 53.699432373046875, 17.780075073242188, 17.0869140625, 53.139183044433594, -8.66111946105957, -3.1142120361328125, 63.433624267578125, -23.390512466430664, 29.040985107421875, 0.3953857421875, 53.22987365722656, -5.341224670410156, 15.214508056640625, 50.862823486328125, -8.888153076171875, -1.860504150390625, 12.534202575683594, 6.862632751464844, -23.621826171875, 26.68517303466797, 8.036369323730469, 58.84100341796875, -77.49786376953125, -17.15734100341797, 1.7357635498046875, 42.61289978027344, 46.242530822753906, 21.460357666015625, 67.04934692382812, 40.013580322265625, 27.006141662597656, 25.489471435546875, -3.196247100830078], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000235.npy"}
{"epoch": 0.49214659685863876, "step": 236, "batch_size": 128, "mean": 23.340309143066406, "std": 32.06284713745117, "min": -69.8634033203125, "p10": -18.423725509643553, "median": 23.428192138671875, "p90": 62.701515197753906, "max": 102.4287109375, "pos_frac": 0.78125, "sample": [26.711700439453125, 21.410186767578125, -54.1185302734375, 20.183212280273438, 36.87415313720703, 36.459381103515625, 9.938400268554688, 56.2762451171875, -9.188613891601562, -27.2222900390625, -43.305908203125, 34.02839660644531, -17.4993896484375, 27.68695068359375, 49.3529052734375, 55.852691650390625, -0.1009521484375, 52.60627746582031, 41.67582702636719, 16.2222900390625, 17.635650634765625, 45.359703063964844, 29.088348388671875, -18.583934783935547, 30.248703002929688, 52.84196472167969, 39.399322509765625, 5.513481140136719, 18.186767578125, 56.54508972167969, 16.785690307617188, 29.5589599609375, 29.58240509033203, 22.14411163330078, 13.410491943359375, 20.282241821289062, 8.110595703125, 76.66128540039062, 53.21783447265625, 3.140625, 10.169906616210938, 53.865943908691406, 66.34817504882812, 25.2562255859375, 64.91958618164062, 2.4766979217529297, -22.176010131835938, 12.782135009765625, 8.7230224609375, 0.0, -0.91180419921875, 23.594650268554688, 51.6807861328125, 54.532554626464844, 8.167985916137695, -20.07122802734375, -55.380889892578125, 59.9105224609375, 9.78759765625, -16.09735107421875, 37.99277877807617, 62.61474609375, 45.4737548828125, 30.242027282714844, -0.771575927734375, -20.390121459960938, 48.42535400390625, -69.8634033203125, 33.404998779296875, 6.0570526123046875, -18.355064392089844, 102.4287109375, 20.839218139648438, 91.9991455078125, 1.9416656494140625, -8.031784057617188, -36.797607421875, -2.800384521484375, 62.00492858886719, 39.20208740234375, 14.592849731445312, 50.096405029296875, -13.88909912109375, 47.51907730102539, 29.344070434570312, 25.8359375, 67.01087951660156, 84.7412109375, 6.179351806640625, 43.63323974609375, 43.47503662109375, 54.00921630859375, 62.294830322265625, 9.728439331054688, 8.79058837890625, 67.66612243652344, 38.96490478515625, 26.413108825683594, 57.99322509765625, 65.22688293457031, 47.440765380859375, 24.0523681640625, -16.52642822265625, 16.897293090820312, 0.0, 20.13037109375, 6.62969970703125, 19.747161865234375, -55.24720764160156, 62.90397644042969, 28.121292114257812, -1.6981391906738281, -35.953460693359375, -8.861343383789062, 37.95428466796875, 54.826316833496094, 63.63346862792969, 12.750114440917969, 6.216346740722656, -18.691024780273438, 64.59706115722656, 49.754425048828125, 63.822418212890625, 15.76129150390625, 1.36468505859375, 57.310577392578125, 11.567874908447266, 23.261734008789062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000236.npy"}
{"epoch": 0.4942408376963351, "step": 237, "batch_size": 128, "mean": 20.06699562072754, "std": 31.989704132080078, "min": -69.2957763671875, "p10": -13.785542297363275, "median": 17.218297958374023, "p90": 65.58884963989257, "max": 100.96759033203125, "pos_frac": 0.734375, "sample": [22.65997314453125, 7.687164306640625, 25.782238006591797, 73.75326538085938, -26.90093994140625, 26.146865844726562, 18.3214111328125, 40.53874969482422, 34.390106201171875, 26.721298217773438, -3.62591552734375, 24.4805908203125, 66.36990356445312, -28.032981872558594, -3.0889053344726562, -3.5397262573242188, 17.13958740234375, 19.227561950683594, 100.96759033203125, 46.71758270263672, 38.07746887207031, 24.95758056640625, 67.63494873046875, 46.498390197753906, 3.3583602905273438, -0.458740234375, 38.709381103515625, 0.0, 38.69164276123047, 36.14373779296875, 6.065948486328125, 7.91436767578125, -27.405654907226562, 17.195777893066406, -11.776153564453125, 29.53765869140625, 11.5714111328125, 0.78045654296875, -69.2957763671875, -21.08354949951172, 2.2259445190429688, -5.256927490234375, -0.079376220703125, 47.75248718261719, 2.6787948608398438, -0.40679931640625, 64.6240234375, 22.22021484375, 40.90547180175781, 5.6317138671875, 6.8960113525390625, 87.452392578125, 31.908245086669922, 11.561882019042969, 6.7980499267578125, 46.218711853027344, 64.28890991210938, 35.91130065917969, 21.640884399414062, 21.117172241210938, 85.85250854492188, 26.66339874267578, 10.28204345703125, 27.1566162109375, 35.66419982910156, -4.1059417724609375, 69.17724609375, 55.278167724609375, -21.196929931640625, -49.29107666015625, 0.1503314971923828, 73.39520263671875, -44.78126525878906, -0.19671630859375, 56.69493103027344, 1.1949691772460938, -17.949813842773438, -2.38189697265625, 80.27595520019531, -6.5697021484375, 12.11407470703125, 21.000473022460938, 61.238189697265625, 18.111480712890625, 62.079254150390625, -27.1861572265625, -1.958160400390625, 4.437164306640625, 65.25411224365234, 21.843597412109375, 17.24081802368164, 96.12336730957031, -9.387397766113281, -2.14495849609375, -28.09130859375, 74.3748779296875, -20.03314208984375, 1.2745361328125, 47.56470489501953, 26.876476287841797, 7.4306640625, 5.234947204589844, 39.42315673828125, 0.8460311889648438, 8.522689819335938, 10.290731430053711, 53.43122863769531, -11.280731201171875, -12.0008544921875, -3.567657470703125, 0.793121337890625, 68.1717529296875, -4.465705871582031, 4.317512512207031, 28.71405792236328, 12.048614501953125, -66.69692993164062, 27.27587890625, 53.1910400390625, 68.5179443359375, 55.0150146484375, -10.162872314453125, 23.81982421875, 44.878631591796875, 14.32925033569336, 25.30859375, 16.400665283203125, 55.82275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000237.npy"}
{"epoch": 0.4963350785340314, "step": 238, "batch_size": 128, "mean": 22.763446807861328, "std": 34.0037956237793, "min": -46.47613525390625, "p10": -15.40256881713867, "median": 19.318601608276367, "p90": 65.9925033569336, "max": 107.23187255859375, "pos_frac": 0.703125, "sample": [-21.010589599609375, -1.8918685913085938, 4.618812561035156, 9.118515014648438, -34.53849792480469, 107.23187255859375, 53.76960754394531, 33.471099853515625, 39.6142578125, -33.185638427734375, -8.49700927734375, 28.175643920898438, -10.9935302734375, 5.18798828125, 0.7625045776367188, -26.37073516845703, 62.242431640625, 62.851104736328125, -5.680511474609375, 39.48382568359375, 55.280643463134766, 56.460662841796875, 25.23309326171875, 46.73851776123047, 87.8804931640625, 23.92223358154297, -10.38201904296875, 2.7339859008789062, 31.734130859375, -46.47613525390625, -38.53448486328125, -45.711463928222656, 88.336181640625, -16.813629150390625, -4.514373779296875, 79.12936401367188, 24.741455078125, 52.03533935546875, 2.5074234008789062, 52.78204345703125, 85.866943359375, 11.5733642578125, 35.32063293457031, 88.982177734375, 47.7525634765625, 58.10687255859375, 47.987640380859375, 46.103851318359375, 3.5196533203125, 31.847808837890625, 66.96907043457031, 99.38555908203125, 16.776947021484375, -14.797828674316406, 15.594085693359375, 15.059967041015625, -5.427001953125, 2.3880558013916016, -9.590591430664062, 36.27264404296875, -0.30303192138671875, 2.64312744140625, -16.890472412109375, -43.188232421875, 56.556243896484375, 36.38092041015625, 29.002716064453125, 60.330772399902344, 45.490234375, 64.34430694580078, -28.722747802734375, -7.18475341796875, 22.92131805419922, -0.254638671875, -3.5796279907226562, 37.3355712890625, 73.9371337890625, -9.255035400390625, 41.1212158203125, 51.72125244140625, 39.388427734375, 76.9266357421875, -4.264858245849609, -12.011741638183594, 3.65948486328125, 22.225799560546875, 49.04772186279297, -4.44746208190918, 19.858379364013672, 26.574859619140625, 2.660837173461914, -4.0852813720703125, -5.7720184326171875, 3.108306884765625, 104.77215576171875, -8.034652709960938, 29.849273681640625, 13.345542907714844, 13.571571350097656, 17.347965240478516, 20.835952758789062, 83.05999755859375, 61.406097412109375, 12.761962890625, 37.40921401977539, 11.5159912109375, -24.461181640625, 60.74143981933594, 25.533782958984375, -9.685333251953125, 15.452667236328125, 65.573974609375, 12.351875305175781, 11.389495849609375, 46.10845947265625, -8.918319702148438, 18.778823852539062, 37.97454833984375, -8.782928466796875, -10.4024658203125, 64.46504211425781, 69.66989135742188, 25.60491943359375, -21.08551025390625, 20.113353729248047, 4.3685302734375, 47.10047912597656, -2.42193603515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000238.npy"}
{"epoch": 0.49842931937172774, "step": 239, "batch_size": 128, "mean": 19.639163970947266, "std": 35.30225372314453, "min": -82.4002685546875, "p10": -21.469529724121095, "median": 15.096176147460938, "p90": 74.52535400390624, "max": 94.88653564453125, "pos_frac": 0.6953125, "sample": [-32.65107727050781, -20.044021606445312, 78.60602569580078, 53.053955078125, -10.131011962890625, -5.553497314453125, -82.4002685546875, 36.631935119628906, 82.87628173828125, 0.0649261474609375, 41.667816162109375, -79.55438232421875, -39.5245361328125, -3.3431396484375, 47.766441345214844, 38.183349609375, 51.031829833984375, 76.28099822998047, 20.134002685546875, -1.556396484375, -0.696868896484375, 49.6890869140625, 6.476755142211914, 17.887893676757812, 5.904304504394531, 26.34619140625, 5.595363616943359, 32.21045684814453, 19.8719482421875, -15.49468994140625, -32.93023681640625, 55.046966552734375, 45.49193572998047, 2.8464508056640625, 88.224609375, -8.63623046875, 20.566421508789062, 58.63360595703125, 89.58685302734375, 72.7431640625, 52.87248992919922, -5.0404052734375, 12.923873901367188, -4.152069091796875, 94.88653564453125, -8.627830505371094, -15.486572265625, 13.84930419921875, -32.2808837890625, 48.6065673828125, -32.48602294921875, 38.49287033081055, 8.466560363769531, 27.262298583984375, 29.56549072265625, -5.81964111328125, 38.939178466796875, 10.040702819824219, 8.255302429199219, -29.076385498046875, 13.746337890625, 74.0614013671875, 5.4228057861328125, 51.7191162109375, 18.557640075683594, 25.67279052734375, 49.337425231933594, -1.273956298828125, 7.1653900146484375, 80.3792724609375, 0.44762611389160156, 18.45752716064453, 8.421173095703125, 29.503570556640625, 51.795021057128906, -7.1785888671875, 41.43597412109375, 7.833717346191406, -21.64776611328125, 30.733657836914062, 1.32635498046875, 23.414718627929688, 79.28173828125, -18.43096923828125, -4.23236083984375, 71.99713134765625, -45.363677978515625, 21.595787048339844, 6.99835205078125, -36.60906982421875, 66.80615234375, -4.105213165283203, -4.252357482910156, 58.829017639160156, 2.099987030029297, 25.429141998291016, 13.603706359863281, -0.46392822265625, 75.60791015625, 27.027099609375, -0.8759117126464844, 83.35153198242188, 8.958236694335938, -11.048049926757812, 40.829864501953125, 4.250968933105469, -21.393142700195312, 21.872406005859375, 32.156158447265625, 80.54547119140625, 4.2071990966796875, 28.952972412109375, 41.830047607421875, 85.29922485351562, 81.43643188476562, 0.0, 34.33280944824219, 3.0139102935791016, 63.75413513183594, -4.699371337890625, -28.549163818359375, 42.13029479980469, 11.859344482421875, 16.343048095703125, 27.68951416015625, -37.618682861328125, -4.994911193847656, 22.934494018554688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000239.npy"}
{"epoch": 0.5005235602094241, "step": 240, "batch_size": 128, "mean": 24.304807662963867, "std": 31.55854034423828, "min": -59.002410888671875, "p10": -12.606707763671874, "median": 23.656936645507812, "p90": 63.51830368041992, "max": 112.16912841796875, "pos_frac": 0.7734375, "sample": [14.470600128173828, 53.99055480957031, 34.53955078125, 50.20509338378906, -12.623092651367188, -7.624198913574219, 35.93701171875, 32.58477783203125, 0.0, 83.69674682617188, 26.0491943359375, -21.42816162109375, -19.068069458007812, 6.1277008056640625, -13.604743957519531, 63.302955627441406, 33.47955322265625, 27.915496826171875, 54.084407806396484, 15.528640747070312, 12.754638671875, 3.82122802734375, 48.947845458984375, 56.396087646484375, 14.602447509765625, 49.11375427246094, 42.037872314453125, -22.088714599609375, 64.02078247070312, 68.99627685546875, 27.684539794921875, 14.0670166015625, 58.3402099609375, 45.31068420410156, 6.49755859375, -7.407257080078125, 66.66055297851562, 27.64305877685547, 48.77543640136719, 2.5350265502929688, 15.792938232421875, 38.13397216796875, 64.97677612304688, 1.1436538696289062, 52.5184326171875, 38.338775634765625, 51.5648193359375, 31.015289306640625, 27.810638427734375, 45.19300842285156, 54.836883544921875, 31.533416748046875, 0.0, 24.507835388183594, -59.002410888671875, 20.65655517578125, 0.80084228515625, 5.0522613525390625, -8.53692626953125, 7.059600830078125, 30.671722412109375, 52.58335876464844, -6.4442596435546875, -4.7159423828125, 69.99822998046875, 55.94981384277344, 36.07611083984375, 47.51471710205078, 62.12001037597656, 34.35198974609375, 80.97421264648438, -15.324310302734375, 0.0, 11.488555908203125, -11.584686279296875, -7.56048583984375, 90.2474365234375, 1.25, 15.435043334960938, 23.595672607421875, -12.599685668945312, 31.55645751953125, 42.432098388671875, 18.367813110351562, 46.44618225097656, 112.16912841796875, -25.402618408203125, 111.04046630859375, -54.28887939453125, 54.6417236328125, 14.475517272949219, -5.07928466796875, 6.632904052734375, 61.72520446777344, 50.37855529785156, 14.936843872070312, 5.5148468017578125, 43.549217224121094, 8.651086807250977, -11.635210037231445, -2.2053070068359375, 0.5348415374755859, 6.729213714599609, 8.611419677734375, 54.39476013183594, 26.5560302734375, 42.40699768066406, 71.70135498046875, 23.79705810546875, 6.2047119140625, 78.30258178710938, 21.479263305664062, 23.71820068359375, -42.83496856689453, 27.139144897460938, -2.758434295654297, -15.69091796875, 33.72047424316406, -16.2860107421875, 53.006378173828125, 0.0, 0.42360687255859375, 20.231201171875, 14.64208984375, 17.6387939453125, -40.38758850097656, 9.7674560546875, 80.36398315429688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000240.npy"}
{"epoch": 0.5026178010471204, "step": 241, "batch_size": 128, "mean": 18.825510025024414, "std": 34.65370178222656, "min": -68.7391357421875, "p10": -23.248683166503906, "median": 12.801788330078125, "p90": 65.05330505371093, "max": 110.98056030273438, "pos_frac": 0.703125, "sample": [2.1898193359375, 70.646728515625, 43.89665222167969, 37.527099609375, 19.44922637939453, -19.4791259765625, 16.47967529296875, 1.1450519561767578, 23.7218017578125, 50.76348876953125, 54.93756103515625, 0.0, -10.14080810546875, 62.976348876953125, 46.46983337402344, 84.6080322265625, 4.178497314453125, 0.1805267333984375, 13.1561279296875, 69.43609619140625, 36.41935348510742, 57.70490646362305, -9.051467895507812, -8.768707275390625, 56.68146514892578, 12.505325317382812, -27.70684814453125, 15.058624267578125, 54.148223876953125, -9.118942260742188, 61.20185852050781, -28.980926513671875, 60.64314270019531, 48.03578186035156, 13.098251342773438, 10.199127197265625, -8.84686279296875, -9.088394165039062, 4.9263153076171875, 64.62750244140625, -4.6881256103515625, 12.376419067382812, 44.751197814941406, 91.6199951171875, 4.33050537109375, 45.926300048828125, 54.501312255859375, -57.982666015625, 15.777084350585938, 18.80609130859375, 15.19500732421875, -2.3348617553710938, 4.698280334472656, 50.419158935546875, -22.19867706298828, -3.2789154052734375, 63.02674865722656, 7.5355377197265625, 4.109161376953125, -23.336898803710938, 19.555328369140625, 64.12094116210938, 29.588531494140625, 56.919708251953125, -3.8991241455078125, 31.7801513671875, -13.374603271484375, 32.93757629394531, 4.841957092285156, 21.3209228515625, -41.663360595703125, 3.511444091796875, 72.36845397949219, 36.636932373046875, -14.96502685546875, -31.9384765625, 71.98065185546875, 41.10334777832031, 3.80328369140625, -2.31439208984375, 4.206451416015625, -1.453033447265625, 73.85269165039062, 22.06500244140625, 32.11848449707031, 16.727005004882812, -32.9056396484375, 59.18748474121094, -8.030868530273438, 66.04684448242188, 7.781219482421875, -38.8062744140625, 0.10773849487304688, 17.86273193359375, 51.391357421875, 33.15284729003906, 3.1939735412597656, 1.8825225830078125, 73.92413330078125, -24.23779296875, 48.4573974609375, -23.21087646484375, 17.63897705078125, 0.03973388671875, 71.00347900390625, 57.573028564453125, -17.672012329101562, 6.27069091796875, 67.76034545898438, 39.50359344482422, -16.6551513671875, -68.7391357421875, -19.6083984375, -3.517303466796875, 50.828086853027344, -0.6833992004394531, 86.28076171875, 33.29689025878906, -33.02705383300781, 9.708953857421875, 110.98056030273438, 2.6156082153320312, 10.407989501953125, -45.879425048828125, 8.16180419921875, -19.351516723632812, 48.3277587890625, -36.310211181640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000241.npy"}
{"epoch": 0.5047120418848168, "step": 242, "batch_size": 128, "mean": 22.71630096435547, "std": 31.249670028686523, "min": -54.85145568847656, "p10": -15.930812072753906, "median": 21.131080627441406, "p90": 62.69515686035156, "max": 90.20538330078125, "pos_frac": 0.765625, "sample": [51.43894958496094, -0.06069183349609375, 70.94854736328125, 7.027256011962891, 10.26239013671875, 8.556938171386719, 33.164520263671875, 40.22492980957031, -19.29644775390625, 8.11578369140625, 73.243408203125, 51.89703369140625, 14.031845092773438, -4.0686187744140625, -49.120521545410156, 29.235260009765625, 9.821609497070312, 58.78727722167969, -54.85145568847656, 27.562297821044922, 7.4553070068359375, 78.85516357421875, 67.95159912109375, 39.61513137817383, 39.39166259765625, 78.28997802734375, 11.136245727539062, 78.83819580078125, -2.1586151123046875, -13.289255142211914, 34.12664794921875, 54.994140625, -0.91522216796875, 63.645843505859375, 26.8133544921875, 90.20538330078125, 75.15486145019531, 4.08796501159668, -19.59485626220703, 3.914306640625, 56.664154052734375, -18.5859375, -4.58935546875, 43.33270263671875, -16.085037231445312, -6.937700271606445, 23.786453247070312, 40.61381530761719, 56.79134750366211, 6.288444519042969, -7.4082794189453125, -9.943679809570312, -8.013198852539062, 42.20948791503906, -5.29730224609375, -9.757682800292969, 59.122039794921875, 14.163909912109375, -7.26080322265625, 49.65519714355469, 25.17510986328125, 1.3824844360351562, -15.864715576171875, 75.62600708007812, 6.1397247314453125, 2.6739578247070312, 81.2291259765625, 41.22357177734375, 21.333267211914062, 14.171981811523438, 21.136703491210938, 50.648468017578125, -24.170166015625, 7.8710784912109375, 53.3875732421875, 7.8618316650390625, 35.62877655029297, 27.87274932861328, 15.8580322265625, -28.038780212402344, 29.205307006835938, 50.890113830566406, 18.753662109375, 27.569068908691406, -32.105316162109375, 11.428298950195312, 7.278709411621094, 17.959754943847656, 33.507965087890625, 55.11705017089844, 84.30206298828125, 3.598297119140625, 18.032241821289062, 20.929492950439453, -1.4050750732421875, 15.378952026367188, 51.14442443847656, 57.040924072265625, 35.62074279785156, 14.22705078125, 3.880176544189453, 46.8521728515625, -28.033687591552734, -13.086944580078125, 41.52613830566406, 17.769393920898438, 21.70758056640625, -50.685791015625, 22.4053955078125, 23.3057861328125, 21.125457763671875, 4.4353179931640625, 54.856719970703125, 76.35026550292969, 57.083465576171875, 47.53462219238281, 62.2877197265625, 21.14665985107422, -30.64606475830078, 47.71734619140625, 1.934906005859375, 24.513877868652344, 11.066360473632812, -32.671112060546875, 39.83989715576172, 59.50770568847656, -12.658111572265625, 44.813873291015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000242.npy"}
{"epoch": 0.506806282722513, "step": 243, "batch_size": 128, "mean": 26.204782485961914, "std": 32.1352424621582, "min": -57.54118347167969, "p10": -12.076566314697265, "median": 24.735549926757812, "p90": 66.7070098876953, "max": 95.822998046875, "pos_frac": 0.7421875, "sample": [-47.296417236328125, -10.371124267578125, 26.602157592773438, 30.436737060546875, 51.926513671875, 36.029052734375, 26.09197998046875, 74.5908203125, -28.726226806640625, 56.44432830810547, 19.7896728515625, 89.2308349609375, 2.4642410278320312, -18.06878662109375, 44.53118896484375, -8.2855224609375, -22.928512573242188, 22.7415771484375, 79.2650146484375, -47.09318542480469, 15.01788330078125, 12.86627197265625, -25.25994873046875, -1.9442825317382812, 30.036331176757812, 65.43182373046875, 95.822998046875, 53.359619140625, -12.446609497070312, 38.4427490234375, 6.626943588256836, 49.13121032714844, 10.404525756835938, 29.80035400390625, 65.66046142578125, 2.760223388671875, 74.85482788085938, 12.918731689453125, 35.060211181640625, 60.805572509765625, -8.842300415039062, 54.20332336425781, 30.545654296875, 44.58917236328125, 28.878616333007812, 39.412506103515625, 72.332763671875, -6.772979736328125, 44.11077880859375, 28.04981231689453, 29.59307861328125, -57.54118347167969, -6.601917266845703, 31.720489501953125, 4.607421875, -0.87835693359375, 16.240928649902344, 24.630706787109375, 10.09747314453125, -7.7825927734375, 92.42472839355469, 59.13079833984375, 49.351341247558594, -8.8419189453125, -1.31890869140625, 53.89717483520508, 20.150924682617188, 51.98046875, 50.45915222167969, -3.55548095703125, 25.983688354492188, -22.691253662109375, -6.81951904296875, 58.791473388671875, 50.60296630859375, 11.59844970703125, 4.377960205078125, 47.82568359375, -1.826080322265625, 20.084556579589844, -4.4876861572265625, 51.417938232421875, -19.859275817871094, 24.577239990234375, -0.96588134765625, 0.0, 5.068576812744141, 23.533233642578125, 53.9410400390625, 64.7687759399414, 27.51537322998047, 41.45941162109375, 47.2196044921875, -4.852195739746094, 59.61883544921875, 56.73443603515625, 69.85903930664062, 16.062576293945312, 9.58468246459961, 18.802764892578125, 16.961669921875, 22.5029296875, -8.642829895019531, 22.54425048828125, -13.95745849609375, 66.12310791015625, 81.7880859375, 49.06829833984375, 4.772941589355469, 24.84039306640625, 19.129878997802734, 11.665870666503906, 67.03964233398438, -14.96319580078125, -11.917976379394531, 94.43276977539062, 16.63113021850586, 66.564453125, -4.1768798828125, 18.186126708984375, 33.41178894042969, 50.68669128417969, 87.51968383789062, 48.91162109375, 85.15959167480469, 38.89093017578125, 35.238677978515625, -13.122406005859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000243.npy"}
{"epoch": 0.5089005235602094, "step": 244, "batch_size": 128, "mean": 27.2258358001709, "std": 34.429466247558594, "min": -70.6854248046875, "p10": -9.415156745910643, "median": 27.035415649414062, "p90": 70.79482421875, "max": 132.49932861328125, "pos_frac": 0.765625, "sample": [-70.6854248046875, 12.2894287109375, 36.958160400390625, 0.491851806640625, 24.512039184570312, 5.744297027587891, 53.24530029296875, -16.424789428710938, 55.52093505859375, 39.698646545410156, 71.76956176757812, 62.023681640625, 12.93524169921875, -6.8820343017578125, 6.476127624511719, 21.420516967773438, 46.04620361328125, 52.11061096191406, 45.742828369140625, 41.780517578125, 10.85382080078125, 54.51051330566406, 2.2568893432617188, -1.94671630859375, 47.215965270996094, -1.04638671875, -22.19281005859375, -26.753196716308594, 64.47042083740234, -23.772491455078125, 6.06329345703125, -8.560371398925781, -4.0452728271484375, -5.903831481933594, -61.3763427734375, 13.572532653808594, 71.06903076171875, -0.252838134765625, 37.87908935546875, 48.40088653564453, -2.4349365234375, 53.619537353515625, 70.50927734375, 36.9223747253418, 91.11962890625, 51.83204650878906, 2.305023193359375, 18.638397216796875, 22.069229125976562, 35.08198547363281, 58.4234619140625, 58.626190185546875, 0.0, 55.0733642578125, -2.305145263671875, -31.615203857421875, 3.98150634765625, 52.20716857910156, -8.748687744140625, 2.7446212768554688, 72.35415649414062, -17.112594604492188, -0.195343017578125, 42.06935119628906, 3.717540740966797, 70.67730712890625, 65.7091064453125, 74.78692626953125, 132.49932861328125, 15.37762451171875, 34.05717468261719, 34.956878662109375, -24.48529052734375, 57.53753662109375, 28.885009765625, 7.5063629150390625, 84.22161865234375, -10.067413330078125, 81.164794921875, 47.155242919921875, 0.895294189453125, 64.51690673828125, 8.222335815429688, -42.70506286621094, 24.30194091796875, 37.18853759765625, -3.8724365234375, 52.203269958496094, 50.98680877685547, 9.63153076171875, 7.0810699462890625, -2.68017578125, 34.95794677734375, 53.6029052734375, 88.35446166992188, 75.87057495117188, 60.92607116699219, 2.5265426635742188, 87.76821899414062, 33.77043151855469, 5.861198425292969, 54.39398193359375, 33.59190368652344, 29.167808532714844, 1.072845458984375, 44.77061462402344, 17.923080444335938, 15.961380004882812, 71.54492950439453, 25.185821533203125, -20.349639892578125, 32.613311767578125, 4.60040283203125, 66.64523315429688, 61.534881591796875, 89.5125732421875, -9.135618209838867, 40.487953186035156, 40.68739318847656, -6.29351806640625, 17.7894287109375, 10.441131591796875, 59.75629425048828, -38.762969970703125, 23.0308837890625, 46.74359130859375, 59.216583251953125, -0.712646484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000244.npy"}
{"epoch": 0.5109947643979058, "step": 245, "batch_size": 128, "mean": 26.574052810668945, "std": 35.09322738647461, "min": -73.69145202636719, "p10": -14.390828895568845, "median": 25.31760025024414, "p90": 65.5848388671875, "max": 109.88473510742188, "pos_frac": 0.7578125, "sample": [21.3497314453125, -8.689193725585938, 21.108993530273438, 21.362464904785156, 6.028228759765625, 14.636688232421875, 48.67974853515625, 4.0911865234375, 1.380126953125, 51.900238037109375, 30.9124755859375, -0.42645263671875, 9.4976806640625, 78.32354736328125, 42.915435791015625, 12.8778076171875, 85.49209594726562, 88.53634643554688, 14.498611450195312, 102.8614501953125, -73.69145202636719, -16.019989013671875, -46.25035095214844, -19.772216796875, 35.21514892578125, 40.93505859375, 0.0, 22.19854736328125, 42.63892364501953, -6.759033203125, 49.09682846069336, 53.08619689941406, 37.49540710449219, 55.75543212890625, 1.1837158203125, 90.31036376953125, 2.7136688232421875, 14.768829345703125, 58.438270568847656, 28.102142333984375, 83.68478393554688, 13.282669067382812, 27.104644775390625, 45.938446044921875, 11.760650634765625, 52.66218566894531, 20.18267822265625, 35.708648681640625, 17.946393966674805, -3.9896697998046875, -44.067626953125, 21.854604721069336, -6.683197021484375, 58.02734375, 19.910736083984375, -5.018013000488281, 46.30488586425781, 36.436676025390625, 44.074066162109375, 62.69764709472656, 94.1142578125, -0.030879974365234375, 3.5439071655273438, 38.139617919921875, 33.43986511230469, 48.57136535644531, 23.951881408691406, -28.093948364257812, 48.586273193359375, -42.37603759765625, 15.824066162109375, 51.31121063232422, 46.09702682495117, 45.876739501953125, 60.65404510498047, -2.626678466796875, -33.708282470703125, 27.56190299987793, 106.45150756835938, 46.73199462890625, 6.323003768920898, 59.344947814941406, -16.902801513671875, 10.090858459472656, 9.674072265625, -37.96101379394531, 10.212615966796875, 65.31402587890625, 66.21673583984375, 55.545166015625, 32.20155334472656, -0.98480224609375, -0.7243728637695312, 96.30917358398438, 29.964691162109375, 93.8345947265625, 16.318222045898438, 18.68353271484375, 39.57573699951172, 54.17628479003906, -43.79443359375, 62.28355407714844, 40.280029296875, -13.588836669921875, -0.9769115447998047, 59.13665771484375, 18.134307861328125, 46.03076171875, 40.11146926879883, 43.47428894042969, 9.849609375, -19.56683349609375, 29.374908447265625, 26.683319091796875, 55.616294860839844, 12.37445068359375, -1.471466064453125, -12.172393798828125, 87.18913269042969, -2.3916053771972656, 55.997467041015625, 13.843582153320312, 57.747406005859375, 109.88473510742188, -39.283660888671875, -13.692617416381836, 0.0, 34.57232666015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000245.npy"}
{"epoch": 0.5130890052356021, "step": 246, "batch_size": 128, "mean": 17.543224334716797, "std": 36.566287994384766, "min": -66.3458251953125, "p10": -24.07075729370117, "median": 11.051200866699219, "p90": 72.42389678955077, "max": 107.5042724609375, "pos_frac": 0.6640625, "sample": [33.40721893310547, 49.512237548828125, 44.92633056640625, 3.970367431640625, 61.1536865234375, -8.703804016113281, 107.5042724609375, 6.852203369140625, 6.455661773681641, 30.555191040039062, -47.8164176940918, 44.25508117675781, 19.48919677734375, 13.569114685058594, -21.566070556640625, -4.5546417236328125, -10.251983642578125, 88.082763671875, 85.88546752929688, -34.333656311035156, 68.47698974609375, 83.32696533203125, 13.6976318359375, 57.205039978027344, 34.170928955078125, -5.270851135253906, 14.127685546875, -21.6474609375, 24.678466796875, -1.3455047607421875, 29.76019287109375, -27.422576904296875, 63.62139892578125, 5.932170867919922, 4.986114501953125, 2.933807373046875, -40.21745300292969, 74.51800537109375, 2.9736328125, 0.9886455535888672, 10.45989990234375, -58.935333251953125, -12.6717529296875, 74.26057434082031, -34.031158447265625, -14.2315673828125, -36.52727127075195, -26.256973266601562, 3.32135009765625, 30.973052978515625, 25.82257080078125, 85.6446533203125, -8.244064331054688, 14.0587158203125, 75.56317138671875, -14.906621932983398, 30.05413818359375, -14.401519775390625, -66.3458251953125, 24.6473388671875, 78.374755859375, 28.415771484375, 38.835182189941406, 23.8251953125, -40.2147216796875, 25.78685760498047, 84.04891967773438, 42.443450927734375, -15.446517944335938, 92.67364501953125, 71.63674926757812, 37.779388427734375, -16.575103759765625, 1.813140869140625, 15.043731689453125, 0.91424560546875, 60.60032272338867, 11.642501831054688, -15.556320190429688, -17.801467895507812, -45.177337646484375, -4.745330810546875, 55.39764404296875, 60.62353515625, -12.408073425292969, -9.410842895507812, -43.463470458984375, -23.75311279296875, 23.4476318359375, 6.640052795410156, 30.63958740234375, 64.48002624511719, 31.935043334960938, -22.179733276367188, 30.58161163330078, 20.75782012939453, 6.8089752197265625, -2.87640380859375, 0.0, 49.327880859375, 69.13288879394531, 37.0140380859375, -7.759071350097656, -21.68907928466797, 48.68475341796875, 2.3660736083984375, 0.7933349609375, 40.316741943359375, 9.045269012451172, 44.55989074707031, 17.714744567871094, -0.9410552978515625, 63.657196044921875, 39.003814697265625, 9.712615966796875, 75.33444213867188, -12.430694580078125, -24.811927795410156, -21.476593017578125, -17.18109130859375, 80.92291259765625, 18.97113037109375, 9.283843994140625, 43.65403747558594, 1.9199981689453125, 10.105133056640625, -4.08807373046875, 66.71295166015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000246.npy"}
{"epoch": 0.5151832460732985, "step": 247, "batch_size": 128, "mean": 21.96489715576172, "std": 32.534236907958984, "min": -64.40884399414062, "p10": -19.35573806762695, "median": 18.876361846923828, "p90": 69.93379669189453, "max": 92.56338500976562, "pos_frac": 0.7421875, "sample": [51.67623519897461, -21.6708984375, 47.62873840332031, 43.57801055908203, -5.4222412109375, 18.91461944580078, 69.68499755859375, 70.51432800292969, 46.59771728515625, 30.961923599243164, 0.0, 49.555076599121094, 71.3286361694336, 16.749557495117188, 4.1957550048828125, 29.970733642578125, -19.275718688964844, 0.767791748046875, -27.022064208984375, 45.36546325683594, -0.34969329833984375, 75.8470458984375, -19.15728759765625, -25.878501892089844, 41.873443603515625, 50.91783142089844, 35.618804931640625, 0.3149871826171875, 56.927215576171875, 70.8292236328125, 47.2242431640625, 57.395301818847656, 10.092056274414062, 87.9002685546875, 38.770843505859375, -23.196624755859375, 4.3191070556640625, 74.83064270019531, -7.7947998046875, 6.1210479736328125, -4.906288146972656, -13.395790100097656, 19.2457275390625, -14.298507690429688, 2.081390380859375, 75.9765625, -2.8253326416015625, 25.26715087890625, 3.6933212280273438, -54.9180908203125, -2.633394241333008, 33.929718017578125, 24.239723205566406, 34.433990478515625, -20.038330078125, 60.23656463623047, 92.56338500976562, 13.24200439453125, 2.0048370361328125, -8.771903991699219, -25.698593139648438, -10.118263244628906, -13.467338562011719, 81.3353271484375, 0.6526031494140625, 0.287200927734375, 49.576629638671875, -35.327178955078125, 9.498275756835938, -19.542449951171875, 13.847488403320312, 69.48193359375, -64.40884399414062, -2.1561927795410156, 47.130584716796875, 8.060203552246094, 21.86339569091797, 10.00360107421875, 19.033203125, 38.84538269042969, 22.980804443359375, 37.442237854003906, 65.29962158203125, 9.327423095703125, -22.473846435546875, -3.806396484375, 3.9305419921875, 32.95777893066406, 52.529937744140625, 18.056396484375, 21.36846923828125, 48.7213249206543, 21.448989868164062, 16.828857421875, 7.7348175048828125, 55.84306716918945, 74.6376953125, 58.84642028808594, 86.5562744140625, -13.214004516601562, 24.3631591796875, 11.193984985351562, 45.21929931640625, 34.11016845703125, 50.397186279296875, 14.122085571289062, 18.26165771484375, -27.416656494140625, 82.08575439453125, 14.633544921875, -42.7044677734375, 55.53143310546875, 34.104164123535156, 3.3189697265625, 21.835302352905273, -4.335947036743164, 42.151336669921875, 0.9867172241210938, -0.013916015625, -1.89825439453125, 50.506195068359375, 30.869293212890625, 78.43968200683594, 2.5611305236816406, 52.25885772705078, 5.151580810546875, 18.838104248046875, 31.192291259765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000247.npy"}
{"epoch": 0.5172774869109947, "step": 248, "batch_size": 128, "mean": 19.782489776611328, "std": 33.73414611816406, "min": -64.16009521484375, "p10": -22.298930358886718, "median": 15.830574035644531, "p90": 67.55740814208983, "max": 110.2144775390625, "pos_frac": 0.7734375, "sample": [9.013748168945312, 58.146484375, -12.590171813964844, 2.4625015258789062, 78.50839233398438, -16.83563232421875, 14.020538330078125, 11.39019775390625, 87.99517822265625, 16.27312469482422, 23.796310424804688, 19.396408081054688, 40.7828369140625, 14.490470886230469, 9.112884521484375, 27.136871337890625, 86.59100341796875, 19.412155151367188, 10.55096435546875, 45.36969757080078, -15.157875061035156, 15.407913208007812, 35.75653839111328, 36.274253845214844, 70.97613525390625, 22.866912841796875, 9.752960205078125, 26.214141845703125, 5.65966796875, 41.794769287109375, 11.872940063476562, -21.456459045410156, 78.78631591796875, -12.053207397460938, 4.268285751342773, 19.756927490234375, 56.170867919921875, -64.16009521484375, 22.213584899902344, -1.0668010711669922, -24.6151123046875, 29.619300842285156, 27.897850036621094, -14.579071044921875, -4.690830230712891, 30.2607421875, -11.62228012084961, -48.6619873046875, 14.82135009765625, 36.23036193847656, 13.817138671875, 80.06614685058594, 22.57318115234375, 17.21167755126953, 94.4814453125, -26.703643798828125, 60.851318359375, 0.468048095703125, 9.63861083984375, 77.38973999023438, 36.0487060546875, 51.795265197753906, -22.179779052734375, -22.576950073242188, 12.134727478027344, 2.43310546875, 110.2144775390625, 62.12538146972656, 37.39751434326172, 36.377349853515625, 44.65946960449219, -8.822525024414062, 2.9455928802490234, 43.80625534057617, 0.45184326171875, -38.80531311035156, 0.0, 25.28741455078125, 86.96856689453125, 70.24298095703125, 50.80870056152344, 40.844696044921875, 28.965072631835938, 37.678497314453125, 9.50729751586914, 42.781768798828125, -17.91741943359375, -34.5057373046875, 0.1861572265625, 13.22998046875, -25.524978637695312, 13.10430908203125, 70.91571044921875, 56.3096923828125, 66.40644836425781, -37.67938232421875, 25.251312255859375, 6.0533294677734375, 3.5646133422851562, 55.813385009765625, 0.7071762084960938, -32.285247802734375, 12.869857788085938, -8.31793212890625, 50.69842529296875, 29.76007080078125, 31.9967041015625, 6.419708251953125, 11.307647705078125, -59.15934753417969, 14.162811279296875, 16.25323486328125, 6.371894836425781, -7.0993194580078125, 93.34292602539062, 37.5001220703125, 40.686553955078125, 36.05696105957031, 21.26446533203125, -44.574188232421875, 10.157279968261719, -3.9485397338867188, 9.751785278320312, 18.185470581054688, 47.31049346923828, -53.08551025390625, 13.294967651367188, 22.8768310546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000248.npy"}
{"epoch": 0.5193717277486911, "step": 249, "batch_size": 128, "mean": 21.49321746826172, "std": 33.48618698120117, "min": -74.4248046875, "p10": -16.39290313720703, "median": 17.80885887145996, "p90": 64.38615417480469, "max": 111.70440673828125, "pos_frac": 0.7109375, "sample": [-4.22052001953125, 7.587795257568359, -39.16310119628906, -29.316497802734375, 31.30774688720703, 3.937286376953125, -13.758468627929688, 1.553985595703125, 9.83612060546875, 9.0570068359375, -6.211761474609375, -1.4459915161132812, 49.90635681152344, 64.65692138671875, -0.6056671142578125, 52.278656005859375, 28.21923828125, 56.947113037109375, 20.407623291015625, 62.61285400390625, 11.900619506835938, -74.4248046875, 53.966163635253906, 19.92376708984375, -18.305023193359375, 91.28375244140625, 59.566009521484375, -15.205322265625, 7.64373779296875, 25.055374145507812, 13.701377868652344, -10.583740234375, 31.14398193359375, -12.738899230957031, 12.24560546875, 12.285018920898438, 79.33285522460938, 13.126777648925781, 66.13052368164062, -35.21449279785156, 55.29319763183594, 35.513916015625, -6.95867919921875, 67.88438415527344, 93.15835571289062, 9.488723754882812, -56.18672180175781, 30.05535888671875, 44.164337158203125, 7.3105010986328125, -3.8387451171875, 19.180091857910156, 98.76779174804688, 40.40680694580078, 53.03924560546875, -2.07781982421875, 9.553756713867188, 36.68754577636719, -28.48974609375, 47.388397216796875, 27.065399169921875, -0.1290283203125, -0.544830322265625, 3.332763671875, 56.16046142578125, 9.663337707519531, 50.365509033203125, -0.41117286682128906, 40.951873779296875, 37.23161315917969, -17.627975463867188, -12.937911987304688, 13.982879638671875, 28.584228515625, 17.32994842529297, 47.76861572265625, 111.70440673828125, 79.982177734375, 56.21131134033203, 44.31195068359375, 94.2703857421875, 39.63726043701172, 10.913429260253906, -19.259613037109375, 47.43115234375, 80.11677551269531, -1.5342540740966797, -13.328094482421875, -3.192424774169922, 35.926666259765625, 24.080169677734375, -15.86358642578125, 52.158294677734375, 41.971038818359375, 28.46649169921875, 37.0174560546875, 39.8934326171875, 23.149337768554688, 8.070281982421875, -23.09161376953125, 69.64190673828125, 0.96246337890625, 20.169631958007812, 29.384796142578125, 30.521652221679688, 14.393798828125, -3.24346923828125, -53.22979736328125, -17.7828369140625, -11.88519287109375, -21.318206787109375, 4.511716842651367, 41.811279296875, 1.4151992797851562, 16.352691650390625, -0.2037353515625, 42.8885498046875, 32.580047607421875, 55.949371337890625, 5.919830322265625, 64.27011108398438, 48.29925537109375, 18.287769317626953, -12.48876953125, -4.7118682861328125, 24.347412109375, 9.591493606567383, 80.10792541503906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000249.npy"}
{"epoch": 0.5214659685863874, "step": 250, "batch_size": 128, "mean": 21.138418197631836, "std": 35.945335388183594, "min": -63.577362060546875, "p10": -26.16038818359375, "median": 23.203224182128906, "p90": 65.92899780273437, "max": 100.26727294921875, "pos_frac": 0.7109375, "sample": [21.470733642578125, 8.26873779296875, -48.47322082519531, 52.0960693359375, 26.561248779296875, 97.86837768554688, 100.26727294921875, 49.02191162109375, 22.645370483398438, -48.813880920410156, 42.374427795410156, 16.65995216369629, 63.27471923828125, 23.971710205078125, 13.5133056640625, 78.50126647949219, 32.30743408203125, 0.7012939453125, 64.95306396484375, 58.2762451171875, 71.7008056640625, 64.16162109375, 43.3658447265625, 40.65153503417969, -24.1689453125, 0.07220458984375, -63.577362060546875, 18.4764404296875, -0.569305419921875, 1.855224609375, 87.51947021484375, -44.05267333984375, 41.897216796875, 90.48770141601562, -9.653572082519531, 6.33282470703125, 13.43408203125, -5.254720687866211, 72.79559326171875, -60.383331298828125, -1.4768238067626953, 25.117584228515625, 68.2061767578125, 29.290069580078125, 45.416351318359375, 32.23712158203125, -28.2847900390625, -40.071990966796875, -1.6696834564208984, 37.20262145996094, 46.31938171386719, -12.712745666503906, 38.19598388671875, -4.285888671875, 23.761077880859375, 10.99932861328125, -9.198287963867188, -25.55474853515625, 34.84455108642578, 3.3778076171875, 42.16105651855469, 71.60527801513672, 52.250606536865234, 0.9093551635742188, 21.74211883544922, 26.66680908203125, 88.86599731445312, 53.39485168457031, -13.750778198242188, 27.798126220703125, 50.62212371826172, 50.233245849609375, -10.250289916992188, 55.24258041381836, 31.6492919921875, -16.381317138671875, 37.36058044433594, 59.73524475097656, 55.21917724609375, 21.235958099365234, 13.269441604614258, -12.8992919921875, 29.9925537109375, -13.834716796875, 63.8717041015625, 16.461708068847656, -4.3427886962890625, -49.40728759765625, 43.608489990234375, 89.44451904296875, -9.2650146484375, 69.39846801757812, -33.904510498046875, -19.82195281982422, -53.346282958984375, 3.45794677734375, 14.270256042480469, -35.326873779296875, -52.7401123046875, 53.75856018066406, 35.69783020019531, 48.06433868408203, 48.1072998046875, 26.60394287109375, 43.39226531982422, 36.277549743652344, -27.57354736328125, 19.71649169921875, 10.959331512451172, -2.1538028717041016, 7.259429931640625, 32.88759231567383, 78.23677062988281, -1.5714111328125, 29.81243896484375, 24.39320945739746, 54.40083312988281, 3.0611419677734375, -2.20751953125, -5.513053894042969, -9.790229797363281, 51.95454406738281, 33.286773681640625, 3.120025634765625, 5.255859375, 18.387969970703125, 37.70558166503906, -5.258647918701172], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000250.npy"}
{"epoch": 0.5235602094240838, "step": 251, "batch_size": 128, "mean": 22.16494369506836, "std": 32.82558059692383, "min": -79.88261413574219, "p10": -15.184896850585938, "median": 17.411727905273438, "p90": 67.01908416748046, "max": 118.18505859375, "pos_frac": 0.734375, "sample": [31.667007446289062, 24.764404296875, 22.712738037109375, -6.2006988525390625, -16.68450927734375, -20.34039306640625, -33.0078125, 6.6528167724609375, 27.637100219726562, 15.558258056640625, 11.237972259521484, 49.388641357421875, 37.8128662109375, 51.70994567871094, -3.2388458251953125, -3.3774795532226562, 14.12005615234375, 18.055755615234375, 3.9328136444091797, 72.04449462890625, 7.1177520751953125, -5.841217041015625, -15.159637451171875, 69.369140625, 37.51637268066406, 75.66513061523438, -26.64825439453125, 49.3404541015625, 64.348388671875, 20.028594970703125, 59.09674072265625, 8.205780029296875, -7.7872314453125, 26.0599365234375, 32.758270263671875, -10.537235260009766, 29.2896728515625, 0.8474693298339844, 2.9965362548828125, -19.084461212158203, 41.76228332519531, 33.96632385253906, 60.553192138671875, 72.13726806640625, 74.09318542480469, 2.297119140625, 0.761444091796875, -26.440414428710938, 1.9735965728759766, -1.6566009521484375, -27.140228271484375, 15.68060302734375, 118.18505859375, 26.087158203125, 45.7327880859375, 52.593894958496094, 7.89166259765625, 61.45280456542969, 13.657852172851562, 28.863014221191406, 50.802162170410156, -79.88261413574219, 26.976322174072266, 45.48699951171875, 15.768341064453125, -17.153968811035156, 9.18060302734375, -4.766845703125, 16.7677001953125, -4.329648971557617, 55.264404296875, 44.755615234375, -14.140289306640625, 3.4572410583496094, 53.328582763671875, -2.3976593017578125, 15.117416381835938, 23.32806396484375, 5.532550811767578, 41.05043029785156, -0.161834716796875, 46.15144348144531, 2.885483741760254, 13.222854614257812, 38.85343933105469, 37.368019104003906, 72.24432373046875, -39.398468017578125, -2.94219970703125, 1.78936767578125, 15.113426208496094, 66.0321044921875, 41.77252197265625, 29.869781494140625, 89.98306274414062, -5.297571182250977, 91.3431396484375, 61.9677734375, 49.253334045410156, -3.170055389404297, 36.24957275390625, 19.973724365234375, 54.37481689453125, 35.40667724609375, 61.941436767578125, 24.99542236328125, 69.14387512207031, -2.6212081909179688, 37.50787353515625, -42.745361328125, 1.651763916015625, 30.624786376953125, 9.258758544921875, -5.967552185058594, 3.2935943603515625, -37.35003662109375, -15.24383544921875, -13.674530029296875, 9.765174865722656, 83.66119384765625, 66.10845947265625, -11.503707885742188, 75.52052307128906, 76.36602783203125, -13.56573486328125, 5.640468597412109, 36.578758239746094, 46.21714782714844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000251.npy"}
{"epoch": 0.5256544502617801, "step": 252, "batch_size": 128, "mean": 22.438114166259766, "std": 36.73619079589844, "min": -84.60726928710938, "p10": -22.925704193115234, "median": 24.871482849121094, "p90": 70.18798370361328, "max": 118.6141357421875, "pos_frac": 0.7265625, "sample": [57.380279541015625, 80.55098724365234, 76.64471435546875, 41.44775390625, -37.27692794799805, 24.523818969726562, -27.8092041015625, 15.976181030273438, 11.117607116699219, -11.432037353515625, 6.385955810546875, -19.00634002685547, 14.067901611328125, 28.3511962890625, 99.26422119140625, 27.41290283203125, 69.49241638183594, 28.723175048828125, 64.24632263183594, 28.447052001953125, 1.584848403930664, 45.118408203125, -14.5675048828125, 76.23869323730469, -84.60726928710938, 0.79913330078125, 6.1185302734375, 30.902015686035156, -28.12579345703125, 2.3456878662109375, 3.9470367431640625, 70.05331420898438, 66.67327880859375, 42.73974609375, 43.730186462402344, 24.726486206054688, -43.3956298828125, -8.875589370727539, 13.83197021484375, 118.6141357421875, 43.606231689453125, -1.1713638305664062, 26.163116455078125, 20.9227294921875, 30.75463104248047, 1.89715576171875, 35.146240234375, -6.7629547119140625, 8.494140625, 10.713584899902344, 45.13056945800781, -1.27630615234375, 76.06088256835938, 47.62066650390625, -16.97918701171875, 0.75048828125, 45.347900390625, -26.09912109375, 11.249725341796875, 5.664955139160156, 46.12705993652344, -14.601799011230469, 25.0164794921875, 88.16094970703125, 22.81586456298828, 27.840957641601562, 63.661102294921875, 29.74005126953125, 44.04656982421875, 62.1337890625, -1.634979248046875, 55.97703552246094, 79.34869384765625, -1.9590301513671875, -23.26105499267578, 46.05908203125, -38.34429931640625, 6.148956298828125, 44.86126708984375, 43.695430755615234, 51.065155029296875, 3.86614990234375, 43.5321044921875, 51.36705780029297, -7.37872314453125, -0.18890380859375, 0.0, 47.608070373535156, -30.660736083984375, 61.78479766845703, -1.3325347900390625, 57.559326171875, 5.3863525390625, -17.6822509765625, 27.78509521484375, 78.50921630859375, -4.12335205078125, 78.39364624023438, 41.17156982421875, 70.50221252441406, -9.88287353515625, 27.478363037109375, 26.26827621459961, 2.087005615234375, 72.36959838867188, -2.3094482421875, 54.496826171875, -70.59375, 0.9193115234375, 8.501419067382812, -8.449310302734375, -22.781982421875, 27.27337646484375, -59.71929931640625, -43.245697021484375, 0.81378173828125, 49.595916748046875, 69.01678466796875, -39.2137451171875, 53.84869384765625, 37.332183837890625, 83.68458557128906, 48.09873962402344, -22.21051025390625, 5.466552734375, 42.70048522949219, 11.742450714111328, 62.20050048828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000252.npy"}
{"epoch": 0.5277486910994764, "step": 253, "batch_size": 128, "mean": 18.748004913330078, "std": 34.617549896240234, "min": -68.6385726928711, "p10": -23.35856018066406, "median": 12.442970275878906, "p90": 69.04674682617187, "max": 95.9366455078125, "pos_frac": 0.7265625, "sample": [85.68508911132812, -6.7919921875, 9.502166748046875, 22.58863067626953, 15.3907470703125, 27.14678955078125, 1.500558853149414, 73.00538635253906, 81.1654052734375, 23.24310302734375, 23.26947021484375, 70.81214141845703, 3.565521240234375, -5.82470703125, 44.47784423828125, -55.370452880859375, 30.576377868652344, -25.234580993652344, 58.12586975097656, -24.404266357421875, 2.473602294921875, 15.127105712890625, 13.572389602661133, 11.717744827270508, -2.4119873046875, -22.910400390625, 21.492095947265625, 69.98745727539062, 54.231422424316406, 51.743873596191406, -4.008819580078125, -5.292507171630859, -36.30021667480469, 11.679306030273438, 95.9366455078125, 1.627838134765625, -19.749740600585938, 50.53631591796875, -64.98727416992188, -22.284942626953125, 46.92388916015625, 12.891357421875, 66.734619140625, 20.08393096923828, 11.722137451171875, -27.86578369140625, 69.72183227539062, 1.754486083984375, -7.029449462890625, 17.812469482421875, 67.08824157714844, 24.46734619140625, 69.3843994140625, -15.141860961914062, -2.842437744140625, 61.9801025390625, 4.977806091308594, 8.97930908203125, -27.1414794921875, 49.455535888671875, 36.10931396484375, -32.258758544921875, -6.1100616455078125, 56.883689880371094, 17.435882568359375, 65.4544677734375, -28.832916259765625, -11.3817138671875, 48.53117370605469, -20.76947021484375, 11.48858642578125, -9.4224853515625, -68.6385726928711, 5.004608154296875, -60.533660888671875, 5.076852798461914, -35.24907302856445, 3.601259231567383, -5.273933410644531, -9.805145263671875, 40.63151550292969, 3.7781295776367188, 36.8089599609375, 51.338623046875, 25.348373413085938, -0.633209228515625, 29.81897735595703, 8.78497314453125, -39.7022705078125, 81.26431274414062, -3.9464073181152344, 6.570960998535156, 45.88299560546875, 4.855072021484375, 4.5342559814453125, 61.92088317871094, -22.168228149414062, 88.060302734375, 9.036735534667969, 35.37879943847656, 34.16033172607422, 9.754013061523438, -18.9500732421875, -0.34375, 31.8037109375, 68.90203857421875, 8.621017456054688, 74.76309204101562, 3.770721435546875, 14.132423400878906, 18.54229736328125, 82.2669677734375, 45.87910461425781, 3.5151901245117188, 54.975128173828125, 11.994583129882812, 11.35418701171875, 15.027511596679688, 19.227996826171875, 52.2833251953125, 3.480804443359375, 26.297821044921875, 74.63348388671875, 54.67903137207031, 8.46630859375, 46.93470764160156, 31.297622680664062, 54.835723876953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000253.npy"}
{"epoch": 0.5298429319371728, "step": 254, "batch_size": 128, "mean": 23.57862091064453, "std": 31.919889450073242, "min": -54.98779296875, "p10": -11.994179534912108, "median": 18.530487060546875, "p90": 66.45853729248046, "max": 97.25440979003906, "pos_frac": 0.7578125, "sample": [73.1270751953125, -3.4802093505859375, -42.40078353881836, 16.925384521484375, -5.370368957519531, -2.6271095275878906, 21.098846435546875, 10.391189575195312, -0.44725608825683594, 16.163772583007812, -5.875633239746094, -9.85980224609375, -0.72686767578125, -2.8922119140625, 5.119956970214844, 51.35052490234375, 16.0662841796875, 59.305999755859375, 38.5220947265625, -18.853302001953125, -24.264251708984375, 67.80670166015625, 4.9266815185546875, 7.99017333984375, -3.73651123046875, 10.39337158203125, 38.902069091796875, -27.757598876953125, 54.1649169921875, 37.9183349609375, 58.5447998046875, 78.76486206054688, -37.307525634765625, 37.607635498046875, 2.650716781616211, 12.270828247070312, -11.863792419433594, 10.716197967529297, 7.785896301269531, 5.0364532470703125, 34.30589294433594, 38.176666259765625, 44.71270751953125, 15.675270080566406, -51.27738571166992, 19.844635009765625, 35.7615966796875, 34.60205078125, 47.3529052734375, 56.05281066894531, 7.3887176513671875, 2.7370433807373047, 10.78036117553711, 59.00080871582031, 45.04296875, 18.28118896484375, 53.457820892333984, -20.8486328125, 16.734405517578125, 44.3616943359375, 9.520301818847656, 8.568901062011719, 83.90521240234375, 9.912742614746094, 43.49159240722656, 70.97687530517578, 33.39044189453125, 62.605743408203125, 55.062896728515625, 23.768783569335938, 9.240669250488281, 27.477813720703125, 74.2479248046875, 4.2539215087890625, 25.196090698242188, -15.864959716796875, -3.6500091552734375, 29.503753662109375, 1.9740486145019531, 5.476295471191406, -3.175323486328125, -6.502662658691406, 79.45440673828125, 63.505584716796875, 26.491764068603516, 89.2237548828125, 47.416778564453125, 57.39787292480469, 15.70111083984375, 68.80282592773438, 13.133001327514648, 62.592041015625, 57.71124267578125, 54.10792922973633, 62.87102508544922, 21.91949462890625, 1.867879867553711, 47.05017852783203, -5.434906005859375, -8.243637084960938, -18.903724670410156, 25.764984130859375, 19.012222290039062, 0.7011814117431641, 0.0, -12.298416137695312, 72.26547241210938, 50.25932312011719, 26.0589599609375, 27.9410400390625, 3.614837646484375, 11.001419067382812, 30.179458618164062, 65.88075256347656, 39.4381103515625, -0.0227203369140625, -54.98779296875, -7.684289932250977, 88.7626953125, 18.77978515625, -33.83648681640625, -26.70281982421875, 2.6476287841796875, 97.25440979003906, 87.52049255371094, 41.463592529296875, 45.160003662109375, 55.61293029785156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000254.npy"}
{"epoch": 0.5319371727748691, "step": 255, "batch_size": 128, "mean": 26.616350173950195, "std": 36.9672737121582, "min": -62.99235534667969, "p10": -14.817771911621092, "median": 17.79029083251953, "p90": 75.37382431030272, "max": 107.902099609375, "pos_frac": 0.765625, "sample": [5.499187469482422, -34.798065185546875, 90.85760498046875, 51.836151123046875, 4.5645599365234375, -52.359161376953125, 2.86053466796875, 68.80488586425781, -8.50262451171875, 53.521942138671875, -6.691558837890625, 11.815834045410156, 13.109657287597656, 2.613372802734375, 89.94741821289062, 82.204345703125, 4.2589111328125, 10.95428466796875, 81.00332641601562, 7.741424560546875, 64.05624389648438, -17.38885498046875, 1.1854248046875, 59.96891784667969, 29.68585205078125, 67.05477905273438, 38.79767608642578, 51.15313720703125, -14.43743896484375, 34.236114501953125, 44.956298828125, 14.767791748046875, 17.402145385742188, 8.1378173828125, 15.572235107421875, 17.234771728515625, 68.12158203125, 30.9404296875, -43.143524169921875, -2.32965087890625, 107.902099609375, 98.47039794921875, 74.11518859863281, 12.494056701660156, -42.01506805419922, -12.198190689086914, -36.900726318359375, 55.104888916015625, 50.42570495605469, 38.877716064453125, 64.81729125976562, 69.09060668945312, 32.41143798828125, 50.127044677734375, 89.54766845703125, -2.5783767700195312, 39.46478271484375, -41.1636962890625, -26.516616821289062, 55.41192626953125, 16.880645751953125, 69.77835083007812, 51.154937744140625, 50.385101318359375, 44.02496337890625, 39.09446716308594, 2.4852828979492188, -46.000709533691406, 73.34396362304688, 12.5653076171875, 76.9337158203125, 0.56475830078125, 17.281856536865234, -7.3358001708984375, 4.96673583984375, 6.66080379486084, 15.932052612304688, 16.069305419921875, 6.010955810546875, 25.880126953125, -5.3887939453125, 44.084381103515625, -62.99235534667969, 85.79644775390625, 74.02618408203125, 50.312843322753906, -6.30621337890625, 51.292152404785156, 18.2578125, 10.872024536132812, 82.6304931640625, 52.75494384765625, 29.8958740234375, 74.7052993774414, -12.78173828125, 95.8056640625, 73.299560546875, 10.114410400390625, -9.198806762695312, -15.705215454101562, -0.4521799087524414, -23.5953369140625, 31.130783081054688, 4.366668701171875, 81.66961669921875, 25.005294799804688, 5.504150390625, 57.19482421875, 8.09710693359375, 34.6873779296875, 14.77362060546875, 15.604530334472656, -9.812904357910156, 69.57525634765625, 16.107223510742188, 38.31263732910156, -5.2960357666015625, -10.836822509765625, 54.71038055419922, 27.699432373046875, 22.972427368164062, 59.81769561767578, -0.7129058837890625, 55.03411865234375, -35.352294921875, 93.50564575195312, 18.178436279296875, -1.2497024536132812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000255.npy"}
{"epoch": 0.5340314136125655, "step": 256, "batch_size": 128, "mean": 27.032861709594727, "std": 40.65485763549805, "min": -82.23275756835938, "p10": -26.631091690063474, "median": 25.047452926635742, "p90": 76.65574340820312, "max": 111.97940063476562, "pos_frac": 0.7578125, "sample": [2.5190887451171875, 65.04165649414062, 68.23989868164062, -25.9859619140625, 19.184371948242188, 16.768890380859375, 62.39088439941406, 60.72650146484375, 50.95751953125, -20.316497802734375, 49.96009826660156, -45.29124450683594, -19.842559814453125, -44.89361572265625, 42.6375846862793, 61.11531066894531, 85.93095397949219, -22.994140625, 38.233116149902344, 70.71099853515625, 1.0201339721679688, 64.68109130859375, -11.520858764648438, 111.97940063476562, 17.4375, 70.94151306152344, 60.21221923828125, -47.7926025390625, 74.74288940429688, 11.186279296875, 87.5068359375, 23.100597381591797, 20.68634033203125, 31.286834716796875, 26.994308471679688, 5.48992919921875, -33.29022216796875, -39.06927490234375, 75.07522583007812, 10.367584228515625, 90.25503540039062, 27.56817626953125, 33.590492248535156, 77.81610107421875, 29.143081665039062, 32.695068359375, -16.80712890625, 16.945655822753906, -23.1851806640625, 105.18707275390625, 4.84100341796875, -9.182098388671875, 86.19105529785156, 83.9422607421875, 6.3345947265625, 6.600061416625977, -82.23275756835938, -20.669921875, 67.51532745361328, 42.027557373046875, 58.01237487792969, 14.206878662109375, 44.413177490234375, 85.37429809570312, -0.6535940170288086, 46.6214599609375, 44.27177429199219, -48.50273132324219, 54.175140380859375, -0.4549560546875, -39.88763427734375, -52.377681732177734, 50.5439453125, 55.695587158203125, 41.1671142578125, 61.6177978515625, -7.067291259765625, 74.733642578125, 18.594482421875, 75.79586791992188, 93.33511352539062, 44.16667175292969, 54.497108459472656, 0.83367919921875, 1.88128662109375, 45.6549072265625, -40.62876892089844, 16.791580200195312, 9.927688598632812, 81.82244873046875, -21.728988647460938, 65.58065795898438, 54.945404052734375, 79.75927734375, 7.285736083984375, 59.18907928466797, 20.982421875, -6.2816162109375, 22.040008544921875, 20.58502197265625, 28.82208251953125, 18.618303298950195, 80.281494140625, 40.377685546875, 18.28729248046875, 39.80572509765625, 17.25628662109375, 66.90231323242188, -28.136394500732422, -78.48397827148438, 4.760986328125, 73.80722045898438, 76.158447265625, 14.303802490234375, 0.9817276000976562, -3.44879150390625, 75.04949951171875, -21.70233154296875, -7.981090545654297, -29.420318603515625, 52.92874526977539, -7.26690673828125, 18.57537078857422, 18.35253143310547, 14.04107666015625, 48.89912033081055, 42.29473876953125, 64.51921081542969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000256.npy"}
{"epoch": 0.5361256544502618, "step": 257, "batch_size": 128, "mean": 23.749109268188477, "std": 34.24029541015625, "min": -56.82081604003906, "p10": -15.517449951171875, "median": 19.595977783203125, "p90": 70.79675598144532, "max": 110.62673950195312, "pos_frac": 0.703125, "sample": [29.66168212890625, 37.170196533203125, 90.42655944824219, -3.5047760009765625, -1.9091033935546875, 86.34884643554688, 19.3914794921875, 70.44613647460938, 24.478271484375, 57.824790954589844, 20.207183837890625, 95.92684936523438, 52.414398193359375, 52.15081787109375, -1.0381660461425781, 23.158424377441406, 15.0557861328125, -30.183853149414062, -8.614097595214844, 43.33245849609375, 68.86962890625, -25.83264923095703, 47.45545196533203, 100.51181030273438, -16.1131591796875, -15.26214599609375, 14.334228515625, 19.066314697265625, 0.22637939453125, -21.606637954711914, -5.6958160400390625, -21.38275146484375, 69.4395751953125, 19.80047607421875, 32.266326904296875, 74.45367431640625, 86.58160400390625, 0.7985992431640625, 82.72528076171875, 52.130577087402344, -7.50555419921875, -0.08623504638671875, -4.9512176513671875, 62.0745849609375, 77.14620971679688, 0.052581787109375, -1.7736892700195312, 32.08378601074219, 14.301910400390625, -56.82081604003906, 44.904541015625, 85.25518798828125, 41.21991729736328, 33.252723693847656, 66.98788452148438, 35.167694091796875, 26.681396484375, 2.9643707275390625, 23.150924682617188, 11.90399169921875, 0.7291793823242188, 77.32026672363281, 49.82415771484375, 12.946025848388672, -19.784027099609375, -8.610565185546875, 28.376220703125, 5.827062606811523, 6.709556579589844, 35.41169738769531, -4.92193603515625, 54.1859130859375, 43.41954040527344, 27.931640625, 9.828710556030273, 9.99249267578125, -33.67384338378906, -1.7545242309570312, 54.15916442871094, 63.599945068359375, 9.158767700195312, 6.686532020568848, 42.58251953125, 23.92291259765625, 0.2545928955078125, -29.85638427734375, -6.10791015625, 62.5618896484375, 36.10595703125, -5.198822021484375, -11.74615478515625, 59.888671875, -12.285430908203125, 64.75308227539062, 34.87972640991211, 62.3018798828125, 49.94683837890625, 21.401580810546875, -12.209823608398438, 5.554039001464844, 5.84417724609375, 12.086395263671875, 71.6148681640625, 21.938446044921875, -4.224853515625, -10.109771728515625, -5.413543701171875, -41.41676330566406, -22.153045654296875, 72.74674987792969, 22.799575805664062, 58.98675537109375, -17.034820556640625, 110.62673950195312, -6.13641357421875, 1.9476318359375, 51.8861083984375, -18.009429931640625, 52.53843688964844, 28.23974609375, 59.22772216796875, -9.973983764648438, -5.9453582763671875, 15.4378662109375, 1.1021957397460938, 0.21697998046875, -2.0439910888671875, 59.475494384765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000257.npy"}
{"epoch": 0.5382198952879581, "step": 258, "batch_size": 128, "mean": 25.13190460205078, "std": 32.680633544921875, "min": -83.5396728515625, "p10": -14.285980987548827, "median": 22.780410766601562, "p90": 70.96072387695312, "max": 103.58157348632812, "pos_frac": 0.78125, "sample": [61.995758056640625, 8.930543899536133, 10.270881652832031, -5.073997497558594, 34.59858703613281, 15.332382202148438, 71.77633666992188, -11.0418701171875, 15.54157829284668, 82.12347412109375, -14.4967041015625, 31.58148193359375, 34.4403076171875, 6.50250244140625, 54.095489501953125, 31.775924682617188, -14.195671081542969, 16.28375244140625, 59.0220947265625, 49.06565856933594, 77.5643310546875, 24.71807098388672, -37.88409423828125, 9.61248779296875, 70.93984985351562, -5.5359954833984375, -83.5396728515625, 63.42341613769531, 45.2996826171875, 4.686058044433594, 66.81719970703125, 31.35089111328125, -15.735992431640625, 33.83686828613281, 44.0198974609375, 26.939849853515625, 2.7855567932128906, -19.034332275390625, 34.05741882324219, 52.03794860839844, 53.810028076171875, 21.190448760986328, 5.851295471191406, 29.092208862304688, 8.326126098632812, 15.805816650390625, 22.763763427734375, 19.054534912109375, 12.508514404296875, 8.8121337890625, 30.337684631347656, 5.124288558959961, 63.2730712890625, 18.193328857421875, -63.205108642578125, 46.120567321777344, 26.3939208984375, -3.218231201171875, -33.3323974609375, 14.626922607421875, -14.645599365234375, 14.156700134277344, 22.79705810546875, 30.925048828125, 72.6920166015625, -0.2998199462890625, 80.80203247070312, 55.218597412109375, -14.980682373046875, 61.255157470703125, 44.39161682128906, 50.68603515625, 0.30757904052734375, -1.20458984375, -0.06923770904541016, 40.609710693359375, 71.00942993164062, -32.07499694824219, 34.380531311035156, 20.2137451171875, -6.078680038452148, -20.433746337890625, 53.07554626464844, 62.95759963989258, -6.6918792724609375, 0.126800537109375, 46.76218795776367, 26.294036865234375, 61.9613037109375, -8.009552001953125, 11.46468734741211, 5.689910888671875, 18.341705322265625, 6.2933502197265625, 7.1229095458984375, 24.92376708984375, 52.634132385253906, 72.97377014160156, 9.753543853759766, -14.962158203125, 3.527008056640625, -4.3753204345703125, 35.258880615234375, 11.058074951171875, 103.58157348632812, -0.6324005126953125, 61.482757568359375, 81.34207153320312, 54.15643310546875, 42.92491149902344, 30.451263427734375, -8.898406982421875, 83.60089111328125, 49.10430908203125, 72.58395385742188, 81.84506225585938, 39.794410705566406, 9.663650512695312, 34.51519775390625, 39.917240142822266, -9.73028564453125, 62.874610900878906, 46.16072082519531, 85.56231689453125, 1.1146240234375, 16.393905639648438, 22.45843505859375, -35.63844299316406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000258.npy"}
{"epoch": 0.5403141361256545, "step": 259, "batch_size": 128, "mean": 24.2410945892334, "std": 35.55464172363281, "min": -68.49447631835938, "p10": -16.62283477783203, "median": 19.84392547607422, "p90": 71.14349670410157, "max": 121.79067993164062, "pos_frac": 0.75, "sample": [59.8721923828125, -21.281265258789062, 71.56106567382812, 8.419029235839844, -17.811508178710938, 7.463134765625, 9.541275024414062, 82.30220031738281, 67.59710693359375, -8.72454833984375, 44.21588134765625, 12.719711303710938, 28.577423095703125, 62.92375183105469, 9.822357177734375, 27.06646728515625, 58.811492919921875, -44.188934326171875, 42.89002990722656, 63.75178527832031, 51.53570556640625, 83.10882568359375, -16.096221923828125, 9.399028778076172, 76.62521362304688, 18.377716064453125, 31.656471252441406, 44.92729949951172, -40.123291015625, -1.4857177734375, 2.318511962890625, 16.26580810546875, -16.1134033203125, 70.96453857421875, 33.880615234375, 36.143829345703125, 30.85308837890625, 7.571006774902344, 15.55657958984375, 17.199066162109375, -19.68096923828125, 16.582122802734375, 64.05050659179688, 8.461071014404297, 5.6126708984375, -39.3043212890625, 40.56237030029297, 79.14462280273438, -1.7516059875488281, 37.06904602050781, 50.34445571899414, 32.816162109375, 65.42657470703125, 23.768310546875, 6.6208038330078125, 9.17111587524414, -0.25347900390625, 78.72845458984375, 57.32160949707031, 12.858856201171875, 100.1065673828125, 44.07080841064453, 2.43353271484375, 1.2487869262695312, -4.098419189453125, -52.95612335205078, 32.23388671875, -15.227691650390625, -28.45037078857422, 68.81201171875, 56.9072265625, -8.70196533203125, -5.9429168701171875, 43.11454772949219, 72.56074523925781, 2.310577392578125, 28.22430419921875, 0.0, -7.158721923828125, 73.82644653320312, 55.171630859375, 58.63702392578125, -7.147064208984375, 8.618499755859375, 36.247314453125, -0.2519683837890625, 84.32632446289062, -45.030067443847656, 1.7199783325195312, 37.864105224609375, -68.49447631835938, 15.153717041015625, 65.88526916503906, 121.79067993164062, 22.92333984375, 47.7283935546875, 35.69189453125, -14.778564453125, 48.89447784423828, 5.5579833984375, 18.699142456054688, 33.40362548828125, -14.955276489257812, 63.859710693359375, 8.77764892578125, 21.243499755859375, 73.58280944824219, -23.12872314453125, 31.885009765625, 61.16754150390625, 12.3499755859375, 64.009521484375, 6.0233306884765625, -8.570159912109375, 6.535436630249023, 8.966079711914062, 87.71038818359375, 5.28460693359375, 20.98870849609375, -15.72747802734375, 57.547935485839844, -51.701507568359375, -22.1888427734375, 43.598236083984375, -5.3047943115234375, 36.31103515625, 52.75193786621094, 50.27717971801758], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000259.npy"}
{"epoch": 0.5424083769633508, "step": 260, "batch_size": 128, "mean": 11.477216720581055, "std": 34.87553024291992, "min": -84.68402099609375, "p10": -27.649308776855467, "median": 8.914636611938477, "p90": 58.37164077758788, "max": 89.95172119140625, "pos_frac": 0.640625, "sample": [77.16119384765625, 1.4485321044921875, 50.249961853027344, 9.430007934570312, -48.8798828125, 41.35243225097656, 29.739593505859375, 63.70831298828125, 36.41393280029297, 14.180023193359375, -11.733779907226562, 16.47906494140625, 0.0, -14.26251220703125, 17.90875244140625, 4.641164779663086, 10.895782470703125, 89.95172119140625, 5.022605895996094, 14.605167388916016, 56.81282043457031, -84.68402099609375, -2.492412567138672, -12.70013427734375, 84.52264404296875, -42.614593505859375, 17.378482818603516, 44.85551071166992, 24.76129150390625, 13.608154296875, 27.187164306640625, -38.93943786621094, 20.66425323486328, 1.833444595336914, 4.131221771240234, 76.0377197265625, -26.585891723632812, -47.464508056640625, -8.304473876953125, 13.624885559082031, 8.462860107421875, 27.786041259765625, 31.8438720703125, -25.6134033203125, -20.679794311523438, 54.95920944213867, -30.130615234375, -9.364343643188477, -15.2835693359375, 13.789478302001953, 6.4765625, 5.4832763671875, 70.71661376953125, 71.30532836914062, 4.337730407714844, 10.2857666015625, 68.83203125, -74.45761108398438, 56.52500915527344, 42.77687072753906, -16.361480712890625, -3.199871063232422, 62.008888244628906, -2.45928955078125, 5.564537048339844, -22.997879028320312, 30.90194320678711, 47.776702880859375, -5.15985107421875, 66.43856811523438, 25.605712890625, 46.442352294921875, -22.212234497070312, 8.12725830078125, 5.507415771484375, -10.264144897460938, -9.42742919921875, 30.774765014648438, 9.65692138671875, 12.329345703125, -39.81488037109375, -21.575828552246094, 84.49127197265625, -3.0814666748046875, 0.420257568359375, 22.0794677734375, 18.332427978515625, -18.183013916015625, -9.407135009765625, -57.61700439453125, -24.648223876953125, 0.931793212890625, -47.38960266113281, 81.87588500976562, 10.105026245117188, 13.626220703125, -4.774505615234375, 9.366413116455078, -20.453094482421875, 78.1600341796875, 49.99847412109375, 29.96173095703125, 16.452774047851562, 33.266357421875, 28.78643798828125, 4.917572021484375, -3.0736846923828125, 54.29486083984375, -15.392112731933594, 56.01202392578125, -20.784423828125, 27.512802124023438, 52.039031982421875, 7.048187255859375, -31.825225830078125, -3.8171539306640625, -10.5611572265625, -65.58395385742188, -14.003204345703125, 7.0355377197265625, 37.57366943359375, -32.739013671875, 33.68070983886719, 11.825942993164062, 51.786346435546875, 0.7029495239257812, 27.27648162841797, -22.79986572265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000260.npy"}
{"epoch": 0.5445026178010471, "step": 261, "batch_size": 128, "mean": 20.69570541381836, "std": 37.37733459472656, "min": -61.047027587890625, "p10": -26.526268005371087, "median": 19.899879455566406, "p90": 68.45976791381835, "max": 128.92831420898438, "pos_frac": 0.703125, "sample": [-17.944259643554688, -61.047027587890625, 43.524566650390625, -15.788307189941406, 22.906707763671875, 15.154571533203125, 35.03009033203125, -13.052215576171875, -0.130340576171875, 63.67333221435547, 72.12376403808594, 0.0, -6.84235954284668, -33.36676025390625, 2.5312767028808594, 31.953857421875, -17.41180419921875, 28.435150146484375, 62.88298034667969, 29.4556884765625, 23.948211669921875, -44.49607467651367, 63.630531311035156, -56.0687255859375, 42.73490905761719, -23.701435089111328, -11.70025634765625, 63.14646911621094, -20.118812561035156, -48.910858154296875, 46.29583740234375, 57.06037902832031, 69.7606201171875, 22.027053833007812, 80.70431518554688, 43.020355224609375, -6.7288360595703125, 14.81899642944336, -31.837249755859375, 30.0567626953125, 10.830917358398438, 9.721282958984375, 80.49868774414062, -19.155426025390625, 7.821502685546875, 4.90863037109375, -0.18807220458984375, 67.90225982666016, -38.76252746582031, 44.44348907470703, 16.358673095703125, -23.373931884765625, 31.643768310546875, 82.71815490722656, 64.346923828125, 41.8697509765625, 20.9910888671875, -24.977218627929688, 65.98005676269531, 65.47618103027344, 65.50643920898438, 66.94281005859375, 48.21256637573242, 10.717880249023438, 28.970611572265625, -0.11203765869140625, 23.601593017578125, -13.356918334960938, 19.45623779296875, -2.5699310302734375, 38.99285888671875, 128.92831420898438, 1.632110595703125, 81.42105102539062, 40.7969970703125, 4.65838623046875, 2.153045654296875, 2.5242557525634766, 3.919586181640625, 7.966697692871094, 24.5263671875, -3.8554534912109375, 46.17195129394531, -52.81964111328125, 88.9390869140625, 11.98779296875, 29.618194580078125, 35.18096160888672, 18.31920623779297, -46.875274658203125, 14.430824279785156, 34.403167724609375, -55.237274169921875, 0.218231201171875, -36.534881591796875, 74.63743591308594, 61.689849853515625, 64.48388671875, 69.78831481933594, 20.343521118164062, -30.140716552734375, 17.935585021972656, -1.1438331604003906, 48.5994873046875, 29.741180419921875, 33.314453125, -21.504745483398438, 95.7896728515625, 78.5582275390625, -7.252372741699219, 43.870025634765625, -5.011268615722656, 29.22382354736328, -47.8095703125, 29.683074951171875, 7.757728576660156, 16.145172119140625, 32.03546142578125, 46.30265808105469, 43.64549255371094, 43.00189208984375, 86.87510681152344, 7.0595550537109375, -22.4287109375, -1.6252174377441406, 14.222198486328125, 7.6700286865234375, 49.971900939941406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000261.npy"}
{"epoch": 0.5465968586387434, "step": 262, "batch_size": 128, "mean": 24.984996795654297, "std": 33.39118957519531, "min": -54.48236083984375, "p10": -14.195312118530273, "median": 22.069091796875, "p90": 67.35239715576172, "max": 101.33978271484375, "pos_frac": 0.71875, "sample": [79.13763427734375, 43.09783935546875, 51.87005615234375, 11.106155395507812, 54.466522216796875, 2.3456954956054688, 81.98419189453125, 68.54428100585938, 17.70440673828125, 47.326568603515625, 93.86868286132812, -14.240234375, 18.291961669921875, 13.23968505859375, 21.26366424560547, 4.315725326538086, 54.91820526123047, 22.780181884765625, -4.861778259277344, 49.273590087890625, 65.88948059082031, -7.589447021484375, -14.17605972290039, 59.90782165527344, 29.905914306640625, 86.1861572265625, 12.989990234375, 21.5572509765625, 59.7672119140625, 60.930694580078125, 14.060150146484375, -50.21769714355469, 23.072586059570312, -2.387603759765625, -0.221282958984375, -28.741043090820312, -18.45001220703125, 24.83831787109375, -3.8948211669921875, 47.71067810058594, 14.747955322265625, -6.201154708862305, -7.936553955078125, -11.8145751953125, 26.359832763671875, 26.112266540527344, 43.46385192871094, -7.636505126953125, 13.785003662109375, -5.91839599609375, 46.959930419921875, 101.33978271484375, 58.18621826171875, 52.208168029785156, -18.66693115234375, -1.5077667236328125, 36.3609619140625, 33.583984375, -20.36572265625, 13.155517578125, 46.16009521484375, -12.26116943359375, -5.7469940185546875, 52.30010986328125, -5.25640869140625, 8.58819580078125, 77.02890014648438, 5.75543212890625, 78.45343017578125, 25.084854125976562, -6.1852264404296875, -0.5323028564453125, 77.67965698242188, -0.96697998046875, 9.4814453125, -18.97412109375, -14.258293151855469, 70.99530029296875, 0.67962646484375, -18.733428955078125, -54.48236083984375, -10.717784881591797, 0.0, 51.04279327392578, 97.92507934570312, 31.802169799804688, 8.068656921386719, 51.93232727050781, 6.801521301269531, 23.968917846679688, 62.27265167236328, 53.920684814453125, 62.3377685546875, 15.022186279296875, 67.22441101074219, 53.988037109375, -5.020233154296875, 23.434844970703125, 24.538116455078125, 47.715576171875, 2.346273422241211, -2.151885986328125, 38.792449951171875, 63.347503662109375, -18.800628662109375, 3.0658645629882812, 25.70916748046875, 22.5809326171875, 67.65103149414062, 54.425323486328125, -1.7064399719238281, 13.657569885253906, 90.8072509765625, 46.268714904785156, 51.29766845703125, 13.58135986328125, 67.218017578125, 32.51106262207031, 63.48847198486328, 60.60560607910156, 17.461807250976562, -51.333839416503906, 16.54010009765625, 34.524993896484375, 5.7613372802734375, 50.44047546386719, 0.96197509765625, -35.827392578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000262.npy"}
{"epoch": 0.5486910994764398, "step": 263, "batch_size": 128, "mean": 21.93096923828125, "std": 36.46741485595703, "min": -67.90576171875, "p10": -21.371408271789548, "median": 17.13414764404297, "p90": 69.171875, "max": 108.2760009765625, "pos_frac": 0.7109375, "sample": [66.87362670898438, 32.18916320800781, 71.02120971679688, 18.0665283203125, -1.8669204711914062, 38.35345458984375, -14.891021728515625, 7.1698455810546875, 43.85216522216797, 50.77886962890625, 5.023557662963867, 38.46949768066406, 23.511749267578125, 17.32275390625, -25.74237060546875, 3.491191864013672, 1.412588119506836, -42.338714599609375, 43.224761962890625, 23.26092529296875, 38.51519775390625, -3.590545654296875, 2.82757568359375, 42.74238586425781, -8.73370361328125, 65.27587890625, 26.313079833984375, 46.69593048095703, 0.3492431640625, -24.33270263671875, -11.451622009277344, 27.803192138671875, 68.61431884765625, 50.00079345703125, 89.0322265625, -11.611679077148438, -0.074493408203125, -4.390289306640625, -34.84625244140625, 70.47283935546875, 11.281951904296875, -3.697113037109375, -17.04412841796875, 74.92108154296875, 64.56484985351562, 5.743120193481445, 45.33607482910156, 5.84515380859375, 65.28106689453125, 16.252281188964844, 39.75225830078125, -30.708892822265625, -12.29327392578125, 56.0182991027832, 67.67233276367188, 51.2384033203125, -54.519012451171875, 7.3148193359375, 60.39129638671875, 67.78311157226562, 0.0, 21.990203857421875, -67.90576171875, 10.59417724609375, 34.4210205078125, 6.440185546875, -5.4708251953125, 1.4635467529296875, 5.992164611816406, 36.500396728515625, 61.326961517333984, 1.6662139892578125, -23.231307983398438, -20.574308395385742, 27.763126373291016, 8.53466796875, 35.327178955078125, 47.679901123046875, 53.9344482421875, 0.3652191162109375, 2.9507598876953125, 4.900718688964844, -38.1568603515625, 21.081939697265625, 34.01985168457031, 20.54861831665039, -42.97954559326172, 93.656005859375, 78.83148193359375, 3.0985183715820312, -15.22119140625, 78.85443115234375, 10.320518493652344, -50.051055908203125, -12.116775512695312, -1.864959716796875, -48.01759338378906, 70.9434814453125, 16.945541381835938, -44.169586181640625, 52.937400817871094, -12.888397216796875, 0.0, 4.1643524169921875, 44.62646484375, 41.390533447265625, 108.2760009765625, 40.521484375, 100.91143798828125, -3.5924758911132812, 32.11871337890625, 13.825340270996094, 71.46470642089844, 0.0, 74.5831298828125, 60.260276794433594, 16.193283081054688, 29.736663818359375, 62.206581115722656, 5.7601776123046875, 64.04779052734375, 68.39860534667969, 18.07086181640625, -5.553466796875, 104.62985229492188, -5.792354583740234, 53.28472900390625, -2.737213134765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000263.npy"}
{"epoch": 0.5507853403141362, "step": 264, "batch_size": 128, "mean": 26.259389877319336, "std": 35.27334976196289, "min": -70.14480590820312, "p10": -10.187431335449217, "median": 17.3853759765625, "p90": 77.4844482421875, "max": 110.62652587890625, "pos_frac": 0.765625, "sample": [-1.575124740600586, -6.8768310546875, 102.61843872070312, 14.365951538085938, 88.3753662109375, 58.421424865722656, 33.287620544433594, 51.438079833984375, 2.3799495697021484, 7.86334228515625, 4.916049957275391, 68.70172119140625, 33.96697998046875, 45.29374694824219, -17.60693359375, 64.46654510498047, 30.6268310546875, 64.76675415039062, -5.421625137329102, 0.533782958984375, 74.80128479003906, 64.32070922851562, -25.92059326171875, 41.8555908203125, -27.36682891845703, 91.4793701171875, 59.50323486328125, -1.3797111511230469, 32.09552001953125, 53.76300048828125, -22.053070068359375, 30.071212768554688, 64.49624633789062, 63.99937438964844, 94.48660278320312, 0.0, 17.86932373046875, 56.24822998046875, 38.48594665527344, 14.558601379394531, -16.724334716796875, 65.45236206054688, -28.833755493164062, 48.54638671875, 41.443443298339844, 85.496337890625, -11.226974487304688, 84.61428833007812, -47.95574951171875, 23.852798461914062, 3.32781982421875, 6.13916015625, 70.46942138671875, 48.63478088378906, 72.284912109375, -41.225494384765625, 73.54777526855469, 58.050254821777344, 10.201868057250977, 10.634300231933594, 38.60455322265625, 2.647775650024414, 10.348304748535156, 24.095382690429688, 0.0714263916015625, -12.3302001953125, -4.9398193359375, 9.860702514648438, 11.936599731445312, 65.75543212890625, -1.607635498046875, 6.3110198974609375, -70.14480590820312, -5.077537536621094, 27.125370025634766, 0.32942962646484375, 18.178585052490234, 8.073699951171875, 79.4894790649414, 77.60574340820312, 60.774749755859375, 0.28409576416015625, 2.2779102325439453, 0.4981689453125, -2.5622482299804688, 110.62652587890625, -0.5276031494140625, 0.0, 16.90142822265625, 9.911376953125, -29.265594482421875, 25.415328979492188, 60.635581970214844, 84.64427185058594, 16.4298095703125, 44.50498962402344, -3.350954055786133, 27.0216064453125, 36.9217529296875, 14.736083984375, 77.43246459960938, 37.906158447265625, -8.33929443359375, -9.741912841796875, 6.211822509765625, 3.364959716796875, 26.3448486328125, 7.905120849609375, -17.9927978515625, 55.06981658935547, 0.8880691528320312, 28.3282470703125, 31.70867919921875, 86.40402221679688, 12.27032470703125, 83.80645751953125, -1.876251220703125, 16.656288146972656, 84.73043823242188, -7.324882507324219, -8.925125122070312, 8.810195922851562, 32.1263313293457, 24.200714111328125, 65.2147216796875, 1.7945365905761719, 27.38812255859375, 16.04339599609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000264.npy"}
{"epoch": 0.5528795811518324, "step": 265, "batch_size": 128, "mean": 23.68561363220215, "std": 36.26947021484375, "min": -55.121063232421875, "p10": -22.03695373535156, "median": 24.009567260742188, "p90": 71.28536834716796, "max": 99.884765625, "pos_frac": 0.6953125, "sample": [4.420001983642578, 99.884765625, 60.70953369140625, -14.1744384765625, 28.71014404296875, 6.764739990234375, 47.84801483154297, 40.03648376464844, 28.31280517578125, -25.377567291259766, 5.21002197265625, 24.497955322265625, 52.910179138183594, -8.141845703125, 74.90301513671875, 69.67314147949219, 31.634368896484375, 79.87744140625, 61.97613525390625, 25.60667610168457, 89.06304931640625, 53.65106201171875, 90.7071533203125, 32.225685119628906, 94.08712768554688, 46.100555419921875, 45.4295654296875, 89.66262817382812, 70.48388671875, -1.1879425048828125, -37.704872131347656, 43.85357666015625, 87.13302612304688, 9.111114501953125, 56.02239990234375, 71.05148315429688, 37.5745849609375, 44.8023681640625, 54.261749267578125, 82.97006225585938, 4.351043701171875, 1.2664794921875, -33.2388916015625, 20.660263061523438, -2.5732955932617188, -21.189239501953125, -13.3397216796875, -30.978668212890625, -2.9945297241210938, 50.25956726074219, -0.463653564453125, -8.971847534179688, -19.66552734375, 3.1368255615234375, -13.865585327148438, 67.17791748046875, 57.91485595703125, 9.288055419921875, 29.202972412109375, -55.121063232421875, 13.548831939697266, 59.89175796508789, 7.682403564453125, 52.2943115234375, 54.77716064453125, -33.25311279296875, 71.83110046386719, 13.886962890625, 25.56524658203125, -19.881805419921875, -26.441864013671875, 72.28628540039062, 57.920684814453125, 10.05145263671875, 0.0, -6.960460662841797, -48.493072509765625, -52.48065185546875, 64.85726928710938, 66.23564147949219, 28.8646240234375, 4.39886474609375, 47.28575134277344, -2.7771453857421875, 45.80988311767578, 25.30096435546875, 0.0, -1.1294021606445312, -16.8336181640625, 19.871490478515625, 44.71728515625, -24.01495361328125, 1.470062255859375, -2.6133880615234375, 43.62884521484375, 6.684059143066406, 22.9393310546875, 96.91766357421875, 30.9373779296875, -18.9949951171875, 16.35034942626953, 29.00665283203125, 37.56488037109375, 35.381099700927734, 6.34393310546875, 41.15419006347656, -38.146331787109375, 70.8282470703125, 43.462249755859375, -43.356292724609375, -2.8436660766601562, 6.29217529296875, 7.5919189453125, -5.218292236328125, 33.63067626953125, -20.003509521484375, -0.8037567138671875, 82.00341796875, 46.1063232421875, 49.986793518066406, 51.72276306152344, -4.014007568359375, 15.900794982910156, 23.52117919921875, 18.400354385375977, -30.6328125, -7.2993316650390625, 37.611907958984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000265.npy"}
{"epoch": 0.5549738219895288, "step": 266, "batch_size": 128, "mean": 25.816532135009766, "std": 34.02170181274414, "min": -70.75323486328125, "p10": -9.60776138305664, "median": 22.809490203857422, "p90": 69.93244018554687, "max": 96.27935791015625, "pos_frac": 0.828125, "sample": [38.64617919921875, 50.2412109375, 86.27850341796875, 14.972885131835938, -2.419525146484375, -9.465385437011719, 54.83753967285156, 1.4805908203125, -12.092430114746094, 1.8867034912109375, 9.39105224609375, -0.216949462890625, 0.0540313720703125, 39.41656494140625, 15.333633422851562, -8.910003662109375, -9.939971923828125, 21.224395751953125, 21.102602005004883, -37.15202331542969, 67.95744323730469, 34.79444885253906, 5.14349365234375, 27.41845703125, 79.49490356445312, 26.8076171875, 0.240570068359375, 24.114532470703125, 84.73101806640625, 32.553466796875, -0.7649497985839844, 96.27935791015625, -1.5277099609375, 33.018157958984375, 7.179119110107422, 64.61962890625, -31.82342529296875, 6.161922454833984, 32.010772705078125, 42.88568115234375, 5.280220031738281, 19.26611328125, 27.896102905273438, 2.7999343872070312, 72.54198455810547, 62.93255615234375, 53.72624206542969, 62.77631378173828, 9.647560119628906, 24.288360595703125, 9.2144775390625, 53.43096923828125, -48.2982177734375, -61.91127014160156, 1.2151947021484375, -40.39927673339844, 5.217990875244141, 25.5491943359375, 25.1650390625, 9.156005859375, 68.88787841796875, 49.71781921386719, 13.486541748046875, 15.833831787109375, 17.206220626831055, 68.55197143554688, 44.15846252441406, 26.22015380859375, 6.44049072265625, 19.812255859375, 30.258544921875, 83.41795349121094, 41.9085693359375, 33.943389892578125, 62.787635803222656, 94.31942749023438, 30.50909423828125, -1.6274566650390625, -41.293609619140625, -49.349098205566406, -70.75323486328125, 81.50205993652344, 20.646240234375, 63.06114196777344, -11.91607666015625, 40.098236083984375, 29.239715576171875, 16.974273681640625, 72.3697509765625, 53.87873840332031, -36.4945068359375, 4.72430419921875, -11.102592468261719, 7.69580078125, 56.877777099609375, 51.72906494140625, 36.71223449707031, 57.48492431640625, 8.78228759765625, 9.500656127929688, 13.56158447265625, 33.052978515625, 13.428802490234375, 95.64356994628906, 10.278425216674805, 5.688751220703125, -8.844791412353516, 7.340431213378906, 23.316490173339844, -1.507476806640625, 75.0108642578125, 44.683349609375, 89.63594055175781, 7.490348815917969, 67.55387878417969, 22.302490234375, 44.9144287109375, 45.066871643066406, 9.044261932373047, 19.155670166015625, 91.27838134765625, 8.133132934570312, 61.91175842285156, 6.761417388916016, 57.966796875, 42.44595718383789, 32.656982421875, 64.88417053222656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000266.npy"}
{"epoch": 0.5570680628272251, "step": 267, "batch_size": 128, "mean": 20.769792556762695, "std": 37.307430267333984, "min": -73.49481201171875, "p10": -27.706032180786128, "median": 21.83740234375, "p90": 66.02391204833984, "max": 112.65176391601562, "pos_frac": 0.71875, "sample": [5.024343490600586, 48.18986511230469, 41.01982116699219, 53.312042236328125, 98.31304931640625, 31.339988708496094, 5.496452331542969, 22.0333251953125, -40.139892578125, 8.059356689453125, -0.1071014404296875, 20.91497802734375, -7.96881103515625, 41.807373046875, 31.00543785095215, -16.541473388671875, 0.6589889526367188, 83.51177978515625, 73.30659484863281, 26.065017700195312, -43.7369384765625, 88.9617919921875, 61.84028625488281, 3.9975967407226562, 12.807437896728516, -18.226715087890625, 39.69464111328125, 53.39091491699219, 49.79412841796875, -73.49481201171875, 34.836669921875, 16.262466430664062, 21.6414794921875, -10.66845703125, 9.88031005859375, 47.562774658203125, -7.059539794921875, 76.9873046875, 60.42724609375, 47.317344665527344, -32.67633819580078, 54.730865478515625, 29.397430419921875, 32.99822998046875, 10.207405090332031, 98.6614990234375, 6.061767578125, -14.233642578125, 64.80924987792969, 71.09211730957031, -54.69544982910156, -49.107513427734375, 37.7694091796875, 43.75707244873047, 7.1935882568359375, 112.65176391601562, 34.700584411621094, 48.36505126953125, 78.70111083984375, 7.614385604858398, 45.56243896484375, 49.310951232910156, -46.4903564453125, 11.38189697265625, 23.777755737304688, -0.2145843505859375, 48.8292236328125, -31.018863677978516, 57.1773681640625, 54.8468017578125, 9.825347900390625, 18.91265869140625, 44.26026916503906, -65.25503540039062, 35.08064270019531, 57.19696044921875, 0.288116455078125, 45.1075439453125, 1.333892822265625, 0.4159278869628906, -26.28624725341797, 30.1627197265625, -2.7830734252929688, 30.50909423828125, -3.0156097412109375, 13.647598266601562, -23.03228759765625, 54.92070007324219, 16.40789794921875, 45.94319152832031, -8.338197708129883, -38.06886291503906, 15.629501342773438, -10.623382568359375, 3.3393096923828125, 60.31450653076172, 28.235305786132812, 45.586544036865234, 31.91527557373047, -7.7415771484375, 68.85812377929688, 23.364776611328125, -17.784210205078125, 36.6051025390625, -9.854057312011719, 22.73541259765625, 23.847305297851562, 55.19395446777344, -5.347480773925781, 98.3973388671875, -18.893112182617188, -50.51881408691406, 52.337303161621094, 16.790313720703125, 51.62200927734375, 0.6736354827880859, 79.42303466796875, 59.537353515625, -12.506561279296875, -18.4742431640625, 71.14338684082031, -40.410980224609375, -5.5870361328125, 3.1979503631591797, 25.881385803222656, -44.07623291015625, -14.944961547851562, 4.7557373046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000267.npy"}
{"epoch": 0.5591623036649215, "step": 268, "batch_size": 128, "mean": 25.098426818847656, "std": 35.84693145751953, "min": -60.64042663574219, "p10": -9.073309326171874, "median": 18.70197296142578, "p90": 75.02173233032227, "max": 112.18572998046875, "pos_frac": 0.7421875, "sample": [19.25811767578125, 3.9007186889648438, 40.50072479248047, 53.27667236328125, 75.03628540039062, 4.393035888671875, -0.2711067199707031, -4.5075225830078125, -18.763273239135742, -33.23985290527344, -2.587493896484375, 92.40322875976562, 14.68963623046875, 75.01549530029297, -0.12286376953125, 42.386260986328125, 112.18572998046875, -6.260408401489258, -0.06402587890625, 54.35911560058594, 60.88135528564453, 58.36064147949219, 19.765487670898438, -60.35804748535156, 54.43109130859375, 33.238433837890625, -0.36548805236816406, 27.36920166015625, 28.8831787109375, 9.171356201171875, 4.409521102905273, 54.27558898925781, 27.0125732421875, 58.32414245605469, 17.318817138671875, 1.8088798522949219, -60.64042663574219, -42.708587646484375, -9.682968139648438, 98.174560546875, 48.401702880859375, 24.538291931152344, 16.62371826171875, 79.94708251953125, 2.540740966796875, 12.849868774414062, 38.851776123046875, -54.174468994140625, -23.61541748046875, -3.29351806640625, 19.57208251953125, 55.541839599609375, 39.11285400390625, 73.41781616210938, 4.5832366943359375, 31.77484130859375, 85.38304138183594, -8.152519226074219, 30.57830810546875, 6.973052978515625, 102.7818603515625, 14.033462524414062, 67.15225982666016, -5.408203125, 0.3267822265625, 16.511260986328125, 70.6480712890625, 25.693023681640625, -24.474441528320312, -6.3751220703125, 50.53514099121094, 11.39447021484375, 64.12936401367188, 54.127899169921875, 17.933135986328125, -6.507637023925781, 39.96551513671875, 93.36929321289062, 81.97003173828125, 34.35186767578125, 32.312225341796875, 45.72917175292969, 55.0533447265625, -8.812026977539062, 45.707855224609375, 61.739105224609375, -31.734695434570312, 37.770416259765625, 30.1585693359375, 7.907135009765625, 13.547027587890625, 48.45721435546875, 88.99710083007812, 38.873626708984375, 3.0699462890625, 71.23468017578125, 11.742446899414062, 0.5149078369140625, 15.257415771484375, 33.100830078125, -4.813568115234375, 29.832435607910156, 22.0052490234375, -0.6099090576171875, 4.385610580444336, 64.99868774414062, 10.898345947265625, -24.4591064453125, -4.0382080078125, 27.255306243896484, 98.17945861816406, 9.831451416015625, 49.73419189453125, 59.74812316894531, -0.61572265625, 76.02001953125, -44.13392639160156, -7.0864715576171875, 102.68328857421875, 11.144378662109375, 3.172271728515625, -20.0543212890625, 12.740280151367188, 26.822677612304688, 8.68353271484375, 18.145828247070312, -3.367431640625, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000268.npy"}
{"epoch": 0.5612565445026177, "step": 269, "batch_size": 128, "mean": 21.542972564697266, "std": 36.897064208984375, "min": -112.61605834960938, "p10": -20.988686370849607, "median": 22.427886962890625, "p90": 69.7389778137207, "max": 103.89453125, "pos_frac": 0.703125, "sample": [-47.01249694824219, 28.317848205566406, 61.192047119140625, -26.84636688232422, 32.736351013183594, -3.585845947265625, 38.7630615234375, 38.2791748046875, 20.776702880859375, -43.024383544921875, 46.728294372558594, 11.710121154785156, -6.934989929199219, 53.726646423339844, -20.2034912109375, 60.248565673828125, 2.39117431640625, -4.944915771484375, 70.16809844970703, 17.559974670410156, 103.89453125, 83.25, 16.509552001953125, 77.24737548828125, 71.7093505859375, -14.691154479980469, 22.599151611328125, 35.515560150146484, -0.46832275390625, 40.7642822265625, 2.2158966064453125, 50.86041259765625, 60.2872314453125, -41.667144775390625, 65.46426391601562, 0.0, 24.94268798828125, -11.219236373901367, 32.36284637451172, 50.309326171875, -20.54065704345703, 16.039657592773438, 82.6458969116211, 44.49403381347656, 6.8429718017578125, 36.426326751708984, 78.12448120117188, 0.0, 41.12287902832031, 46.452056884765625, 55.87947082519531, -12.047821044921875, 69.55506896972656, -53.56256103515625, -42.115966796875, 67.96470642089844, 53.313865661621094, -8.386451721191406, -27.454925537109375, -2.056140899658203, 74.75091552734375, 27.281005859375, 34.4971923828125, 16.3260498046875, 52.020179748535156, -25.40728759765625, 11.568389892578125, 36.47747802734375, 63.174774169921875, 11.389694213867188, 43.0133056640625, 26.91241455078125, 0.0, 69.17381286621094, 11.856101989746094, 17.882904052734375, -72.104736328125, 2.1749229431152344, 30.895355224609375, 5.3921661376953125, -9.65667724609375, -4.5318603515625, 77.39263916015625, 13.95721435546875, 43.06975555419922, 20.8328857421875, 26.60968017578125, 41.420196533203125, 28.819740295410156, 75.378662109375, 42.14398193359375, 22.256622314453125, 10.17987060546875, -15.745635986328125, 62.30003356933594, 9.08648681640625, -7.5170745849609375, 44.96501159667969, 7.0797576904296875, 28.2186279296875, 49.84632873535156, 22.686492919921875, 2.860445022583008, -16.148292541503906, 0.9934825897216797, -52.589691162109375, 40.9085693359375, 84.26443481445312, -6.8768463134765625, 51.29185485839844, -22.034088134765625, 13.364875793457031, 45.795867919921875, -20.320114135742188, 55.23179626464844, -26.85495376586914, -15.26422119140625, -10.162544250488281, 30.266250610351562, -4.5731964111328125, 12.70550537109375, 77.6434326171875, 32.20726013183594, -112.61605834960938, -13.028472900390625, 8.764305114746094, 60.088218688964844, 82.88616943359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000269.npy"}
{"epoch": 0.5633507853403141, "step": 270, "batch_size": 128, "mean": 25.52783203125, "std": 38.70653533935547, "min": -99.87454223632812, "p10": -24.385662460327147, "median": 23.43834686279297, "p90": 77.86731567382812, "max": 88.28350830078125, "pos_frac": 0.75, "sample": [-33.391845703125, 39.298675537109375, 55.962890625, 49.007713317871094, -23.91653823852539, 85.18928527832031, -6.6012115478515625, 19.292236328125, 15.119903564453125, -37.4183349609375, 66.72200012207031, 13.881744384765625, 15.82391357421875, 14.592498779296875, 12.569931030273438, 54.21044921875, -12.118743896484375, 83.09147644042969, 20.613975524902344, -34.99253845214844, -42.095489501953125, 29.658065795898438, 3.322052001953125, 41.5723876953125, 10.945379257202148, 4.60223388671875, 66.67167663574219, 66.41893768310547, -6.99859619140625, 0.2767677307128906, 31.51177978515625, 40.67109680175781, 29.10888671875, 4.530963897705078, 4.7964324951171875, 6.7734832763671875, -27.6019287109375, 5.6929779052734375, -54.4886474609375, 25.751449584960938, -10.21295166015625, 88.28350830078125, 23.808212280273438, 11.591041564941406, 45.7447509765625, 3.3939285278320312, -4.5379791259765625, 77.58758544921875, 49.50120544433594, -64.55999755859375, -28.133026123046875, 63.28265380859375, 72.89608764648438, 69.17718505859375, 23.0684814453125, 44.27728271484375, 84.02114868164062, -22.12261962890625, 66.994140625, 2.1870880126953125, 16.35198974609375, -13.646560668945312, 9.22357177734375, 11.443798065185547, 71.67971801757812, 68.86061096191406, 81.66847229003906, 3.3882598876953125, 69.617919921875, 25.830467224121094, 53.84687805175781, -4.90362548828125, -58.17572021484375, 74.37793731689453, -6.55181884765625, 81.19085693359375, 31.472793579101562, 23.058349609375, 48.89740753173828, 27.31005859375, -16.869720458984375, 6.430145263671875, 19.95379638671875, 58.598480224609375, 32.53435134887695, -4.578285217285156, -3.27423095703125, 12.598052978515625, 1.6368026733398438, -1.404266357421875, -28.28033447265625, -33.002105712890625, 68.5645751953125, 78.52001953125, -6.998687744140625, 15.973274230957031, 63.230316162109375, 85.77542114257812, 38.737091064453125, -25.48028564453125, 68.540283203125, 84.25787353515625, 41.002166748046875, 42.466026306152344, 25.762832641601562, -99.87454223632812, -12.255889892578125, -8.62481689453125, 65.0921630859375, 17.3177490234375, 31.244720458984375, 44.1392822265625, 25.042984008789062, 46.378204345703125, 85.12832641601562, 84.785888671875, 74.74609375, 20.297203063964844, 9.67120361328125, 81.2117919921875, -14.63995361328125, 86.908935546875, 61.15533447265625, 73.45515441894531, 33.702232360839844, 65.1475830078125, 43.78741455078125, -20.1943359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000270.npy"}
{"epoch": 0.5654450261780105, "step": 271, "batch_size": 128, "mean": 24.81477928161621, "std": 36.254722595214844, "min": -51.84466552734375, "p10": -16.967056274414062, "median": 17.1453800201416, "p90": 69.3121208190918, "max": 123.28033447265625, "pos_frac": 0.734375, "sample": [58.33396911621094, -49.44031524658203, -5.979888916015625, 19.428314208984375, 9.075469970703125, -1.47821044921875, 8.364585876464844, -9.669391632080078, 28.19281005859375, -3.8963851928710938, -6.11395263671875, 16.314590454101562, 69.27577209472656, 4.5331878662109375, 30.629287719726562, 24.848052978515625, 35.80548095703125, 16.65567398071289, 18.594345092773438, 18.996444702148438, -1.2509994506835938, 5.593147277832031, 68.20230102539062, 65.08320617675781, 1.4175033569335938, 2.3859691619873047, -47.6285400390625, 62.74786376953125, 8.087337493896484, -31.824249267578125, 65.65383911132812, 84.76953887939453, -3.1416015625, 26.4161376953125, 5.5832366943359375, 20.5791015625, 25.002410888671875, 10.776229858398438, 33.366241455078125, -29.27783203125, 63.53289794921875, 9.852340698242188, -1.543731689453125, 65.89694213867188, 56.5444450378418, 8.671356201171875, 37.1285400390625, -2.3894729614257812, 37.319183349609375, 16.390716552734375, -5.5350341796875, 54.06109619140625, 74.90721130371094, 35.90904998779297, 123.28033447265625, 94.63079833984375, 0.8677234649658203, 79.75506591796875, 23.155181884765625, 51.801753997802734, 50.65623474121094, 5.65350341796875, -4.440711975097656, 117.94976806640625, 49.34912109375, -51.84466552734375, -17.151763916015625, 74.49478149414062, -10.4912109375, -10.349365234375, 68.54266357421875, 108.95147705078125, -16.94879150390625, 44.0462760925293, -8.0103759765625, 15.74920654296875, -26.328933715820312, 20.585205078125, 29.98394775390625, 49.38267517089844, -17.009674072265625, -30.13848876953125, 58.296875, -8.809005737304688, 103.83432006835938, 50.321075439453125, -1.729461669921875, -30.376861572265625, 23.990283966064453, 13.909698486328125, 3.902130126953125, 77.29095458984375, 53.871734619140625, -5.7489166259765625, 2.3073883056640625, 57.54527282714844, 4.402412414550781, 61.628822326660156, 60.500701904296875, 15.146736145019531, 8.220657348632812, 17.984634399414062, -5.0047607421875, 55.47621154785156, 86.59674072265625, 15.710784912109375, -33.688629150390625, 12.20703125, 13.657608032226562, 48.2265625, 28.26416015625, 62.294036865234375, 69.39693450927734, 7.8575897216796875, 51.24980163574219, 9.76287841796875, 40.84385681152344, 57.33648681640625, -1.590972900390625, -7.647796630859375, 63.18325424194336, -38.35771179199219, 62.8646240234375, 81.52285766601562, 14.780708312988281, 6.3729248046875, 17.635086059570312, -31.025848388671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000271.npy"}
{"epoch": 0.5675392670157068, "step": 272, "batch_size": 128, "mean": 22.67195701599121, "std": 36.57318878173828, "min": -69.69891357421875, "p10": -17.50415802001953, "median": 18.950244903564453, "p90": 68.75636596679686, "max": 117.92718505859375, "pos_frac": 0.7578125, "sample": [22.00048828125, 0.0, -22.473388671875, 34.32806396484375, -39.26835632324219, 46.104736328125, 88.421875, 9.7508544921875, 55.57591247558594, -5.0785369873046875, -24.4384765625, 11.251104354858398, 38.182525634765625, 67.6632080078125, 48.345458984375, 95.00616455078125, 95.84942626953125, 57.48602294921875, 28.27203369140625, 14.52093505859375, 64.94076538085938, 7.2128143310546875, -8.4356689453125, -6.055023193359375, 54.092742919921875, -6.53509521484375, 3.430173873901367, 94.5615234375, 2.770925521850586, 7.1124267578125, 71.30706787109375, 41.25508117675781, 48.183990478515625, 5.924694061279297, 80.26360321044922, 45.427398681640625, 89.34255981445312, 75.79193115234375, 10.824554443359375, -4.023773193359375, 39.26683044433594, -37.85772705078125, -18.428878784179688, 49.67250061035156, -12.44818115234375, 20.81591796875, 53.582332611083984, -33.974395751953125, 5.1837158203125, 3.3868885040283203, 83.7982177734375, 63.939308166503906, -10.302680969238281, 18.869728088378906, 47.44163513183594, 58.39007568359375, 35.49201965332031, 38.43902587890625, 53.190887451171875, -0.23347854614257812, -3.1832523345947266, 9.210952758789062, 6.9580535888671875, 10.424392700195312, 26.786922454833984, 9.829002380371094, 117.92718505859375, 14.79638671875, 61.81817626953125, 11.121826171875, 67.46340942382812, -5.705268859863281, 43.0008544921875, 8.790603637695312, -69.69891357421875, -20.745376586914062, -4.374481201171875, 19.895126342773438, 2.293243408203125, 38.779571533203125, 27.548873901367188, 36.67706298828125, -5.3668212890625, 89.85490417480469, 1.67578125, -69.69058227539062, 20.809480667114258, 12.929447174072266, 21.942657470703125, 1.9776725769042969, 24.7041015625, 48.83885955810547, 5.811187744140625, -53.246337890625, 46.42805480957031, 31.546875, 37.2471923828125, 58.49687957763672, -8.381195068359375, 3.5406227111816406, 0.0, -17.10784912109375, 7.349403381347656, -45.37367248535156, -13.684234619140625, 19.03076171875, 28.0521240234375, 1.3456878662109375, 0.9523544311523438, 9.57733154296875, 52.35311508178711, 117.45745849609375, -66.7364501953125, 26.67322540283203, 65.051513671875, -32.12225341796875, 14.670097351074219, 2.2095394134521484, 7.915483474731445, -3.133453369140625, 53.706787109375, 30.17487335205078, 23.60448455810547, 9.0311279296875, 43.672119140625, 73.267333984375, 28.774246215820312, 29.44986343383789], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000272.npy"}
{"epoch": 0.5696335078534032, "step": 273, "batch_size": 128, "mean": 25.53744888305664, "std": 38.10216522216797, "min": -88.71023559570312, "p10": -19.54493179321289, "median": 24.154067993164062, "p90": 72.58376770019531, "max": 127.05535888671875, "pos_frac": 0.6640625, "sample": [24.37115478515625, 94.20428466796875, 88.67034912109375, 12.844877243041992, 101.63927459716797, -3.5745849609375, 107.3089599609375, 18.037841796875, 19.76922607421875, 5.92449951171875, -20.990234375, 21.27080535888672, -4.658039093017578, 33.295589447021484, 127.05535888671875, -52.90728759765625, 43.14642333984375, 93.77969360351562, 9.84478759765625, 28.5208740234375, -26.320587158203125, 45.63475799560547, 64.74632263183594, 26.47918128967285, 72.42352294921875, 20.17535400390625, -47.149871826171875, 25.393310546875, 41.18408203125, -4.542266845703125, 18.984878540039062, -23.209503173828125, -2.4807205200195312, 65.51396179199219, -14.2855224609375, -0.0283203125, 6.630462646484375, 13.14764404296875, -6.599273681640625, 51.14239501953125, -5.811044692993164, -1.600128173828125, -1.4552536010742188, 33.755584716796875, 36.85792541503906, 19.068939208984375, 56.964744567871094, -0.1742725372314453, -6.952964782714844, 8.409576416015625, 49.590118408203125, -22.283172607421875, 83.439208984375, -15.447418212890625, 36.07704162597656, -4.17657470703125, -1.361907958984375, 34.251251220703125, 4.17242431640625, 54.92064666748047, -36.64794921875, 56.584632873535156, -1.745025634765625, 51.65565490722656, -12.880340576171875, 75.03948974609375, 49.72209167480469, -0.5464401245117188, 51.93586730957031, -3.449798583984375, 64.78567504882812, -0.8036651611328125, -35.22949981689453, -14.6690673828125, 26.20147705078125, 74.8353271484375, 15.691116333007812, 67.94876098632812, 59.96327209472656, -11.1134033203125, 55.94239807128906, 63.625953674316406, 16.275253295898438, -19.41222381591797, 41.7708740234375, 63.11280822753906, 25.024078369140625, 25.12946319580078, 78.43179321289062, 29.774185180664062, 23.936981201171875, -27.469810485839844, -4.978668212890625, 53.637176513671875, -88.71023559570312, 28.471420288085938, -55.84210205078125, 58.526824951171875, 71.9775390625, 9.734603881835938, 85.19024658203125, 12.896965026855469, 58.1832275390625, 69.44998168945312, 71.39326477050781, -2.85141658782959, -34.8875732421875, 14.16790771484375, 70.01943969726562, 54.750160217285156, 41.20320129394531, 44.45831298828125, -3.4089508056640625, 53.1102294921875, 8.74847412109375, 49.987274169921875, -14.401641845703125, 72.95767211914062, 14.5625, -4.2386322021484375, -5.718776702880859, 80.031982421875, 30.6566162109375, 59.70612335205078, 63.018341064453125, -19.854583740234375, 53.615234375, -12.827072143554688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000273.npy"}
{"epoch": 0.5717277486910994, "step": 274, "batch_size": 128, "mean": 28.23946189880371, "std": 37.67235565185547, "min": -83.06414794921875, "p10": -12.825904846191403, "median": 21.03337860107422, "p90": 70.9920913696289, "max": 106.0657958984375, "pos_frac": 0.8046875, "sample": [68.3191909790039, 70.436279296875, -35.6690673828125, 55.06037139892578, -2.4957427978515625, 55.58740234375, 35.002471923828125, 25.470001220703125, 49.532745361328125, 5.204399108886719, 31.803802490234375, 66.4297866821289, 89.10623168945312, 20.14759063720703, 3.0287094116210938, 59.895362854003906, 106.0657958984375, 88.88368225097656, -49.1739501953125, 4.944995880126953, 10.95928955078125, 65.5228271484375, 65.389404296875, 35.424224853515625, -5.6132659912109375, -5.3236083984375, -38.87914276123047, -27.483734130859375, 8.032182693481445, 12.8572998046875, 21.05450439453125, 20.477874755859375, 58.67729187011719, 33.60626220703125, 65.5423583984375, 21.538055419921875, 7.959259033203125, 61.48411560058594, 21.934158325195312, 42.499420166015625, 86.99325561523438, -4.204433441162109, -3.11285400390625, 7.220512390136719, -14.747177124023438, 72.28898620605469, 19.7130126953125, 13.490158081054688, 28.983917236328125, 98.27578735351562, -9.247230529785156, 8.34930419921875, 68.02239990234375, 7.3133697509765625, -70.498046875, 16.80230712890625, -12.00250244140625, 1.731109619140625, 0.8490982055664062, 10.433929443359375, 20.798118591308594, 7.685474395751953, -17.35443878173828, 43.80883026123047, -17.091705322265625, -1.9692630767822266, 51.85865783691406, 0.3089618682861328, 58.744110107421875, 64.12940979003906, 11.079879760742188, -71.13272094726562, 80.00634765625, 21.012252807617188, 47.035400390625, 39.18194580078125, -3.943206787109375, 31.487960815429688, 93.32464599609375, 8.967620849609375, 104.36651611328125, 65.1966323852539, 13.1351318359375, 63.0657958984375, -1.32427978515625, 50.13153076171875, 66.83358764648438, 69.6441650390625, 54.455810546875, 77.25684356689453, 39.46776580810547, 18.043121337890625, -21.190399169921875, 48.306884765625, 61.745513916015625, 60.248802185058594, 8.251022338867188, 78.3043212890625, 95.92868041992188, 20.546630859375, -47.746002197265625, 15.5863037109375, 12.5443115234375, -15.428558349609375, 33.54408264160156, 64.67023468017578, 3.4567794799804688, 68.67097473144531, 13.7919921875, 37.3544921875, 53.0001220703125, 57.15083312988281, 6.6329345703125, 5.8324127197265625, 35.25569152832031, -6.989715576171875, 4.282196044921875, 7.917388916015625, 8.742385864257812, 34.90826416015625, 6.203876495361328, -83.06414794921875, 62.3448486328125, -1.200399398803711, 17.90469741821289, 58.008880615234375, 66.25857543945312, 104.77151489257812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000274.npy"}
{"epoch": 0.5738219895287958, "step": 275, "batch_size": 128, "mean": 31.430099487304688, "std": 35.81398391723633, "min": -61.835174560546875, "p10": -6.55969314575195, "median": 25.385741233825684, "p90": 81.41845703125, "max": 110.27325439453125, "pos_frac": 0.796875, "sample": [110.27325439453125, -4.1605224609375, -2.075592041015625, -41.28216552734375, 85.13172912597656, 26.16046142578125, 78.19326782226562, 22.6221923828125, 23.076812744140625, -61.835174560546875, 24.96471405029297, 90.73028564453125, -27.1627197265625, 49.802642822265625, 0.0, 8.05712890625, 11.560073852539062, 46.9161376953125, 4.288423538208008, 3.764556884765625, 0.0, 0.0, 61.642494201660156, 67.45807647705078, 52.286590576171875, 3.1611175537109375, 88.14151000976562, 54.85696029663086, 16.5107421875, 46.75608825683594, 51.956695556640625, 4.069103240966797, 45.07630157470703, 77.23211669921875, 81.447509765625, 67.04193115234375, 1.1366825103759766, -0.03131103515625, -43.361053466796875, 38.29396057128906, 81.406005859375, 24.306793212890625, 40.47740936279297, 21.770233154296875, 10.6529541015625, -4.014068603515625, 5.8340301513671875, 70.77659606933594, 83.72409057617188, 9.96356201171875, 81.0140380859375, -27.379302978515625, 2.110706329345703, 15.832046508789062, 49.406925201416016, 12.327789306640625, 51.687652587890625, 17.05560302734375, 73.679443359375, 0.0, 15.233642578125, 17.171730041503906, 31.635986328125, 60.485809326171875, -5.700614929199219, 76.53436279296875, 87.44526672363281, 48.433135986328125, 11.649993896484375, 0.0, 3.538167953491211, 82.52203369140625, 15.295738220214844, 62.074920654296875, -1.3175735473632812, 27.53192138671875, 80.08880615234375, 92.0908203125, 34.63017272949219, 4.061389923095703, 62.39875411987305, 72.31816101074219, 42.34490966796875, 46.054595947265625, 31.412063598632812, 2.1318817138671875, 17.572616577148438, 57.1963005065918, 43.886566162109375, 27.15521240234375, -14.514434814453125, 25.458051681518555, 8.496292114257812, 84.2999267578125, 16.32769775390625, 65.43499755859375, -11.087020874023438, 0.97747802734375, 106.6641845703125, 6.9839935302734375, 25.313430786132812, 15.512603759765625, 55.409210205078125, -10.449806213378906, 78.20213317871094, 67.99956512451172, 12.163841247558594, -27.765869140625, -5.066436767578125, -8.564208984375, 85.53225708007812, -33.50035095214844, 63.480960845947266, 0.23443603515625, 28.26934814453125, 3.0676517486572266, 12.627410888671875, 46.67803955078125, 89.02766418457031, 68.6533432006836, -13.805900573730469, 40.0269775390625, 40.22465515136719, -4.021720886230469, 51.56278991699219, -15.954833984375, 74.19854736328125, 65.74542236328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000275.npy"}
{"epoch": 0.5759162303664922, "step": 276, "batch_size": 128, "mean": 28.397563934326172, "std": 35.32380676269531, "min": -56.59747314453125, "p10": -14.260079765319825, "median": 24.479110717773438, "p90": 79.39844970703125, "max": 125.7750244140625, "pos_frac": 0.765625, "sample": [81.71893310546875, 19.66845703125, 60.554344177246094, -4.2675018310546875, 15.82373046875, -0.41851806640625, 14.237030029296875, 34.739044189453125, 23.334228515625, 25.839797973632812, 29.599102020263672, 1.909088134765625, 11.134429931640625, 2.118581771850586, 3.2143402099609375, 67.64663696289062, 33.201629638671875, 20.126800537109375, -25.26165771484375, 35.43206787109375, 81.8109130859375, 11.298141479492188, 7.802248001098633, 85.51589965820312, 47.317352294921875, -17.8104248046875, -14.1934814453125, -12.735595703125, -6.6878662109375, 64.16131591796875, 27.3072509765625, 32.247955322265625, 34.64776611328125, 40.911216735839844, 38.22544860839844, 11.225128173828125, 24.53033447265625, -0.9813404083251953, 34.99458312988281, 27.107864379882812, 17.507720947265625, 19.894332885742188, 76.23202514648438, 22.763946533203125, 4.518451690673828, 69.15093994140625, 35.826904296875, 81.35372924804688, 12.373779296875, -5.07794189453125, 4.903228759765625, 40.623992919921875, 56.31591796875, 21.55963134765625, 66.7020492553711, 8.764396667480469, 45.22550964355469, 27.002899169921875, 72.31448364257812, -26.88018798828125, 0.0, -20.599609375, 4.5492706298828125, -38.71589660644531, 5.4727783203125, -17.993438720703125, 27.266204833984375, 90.60775756835938, -2.385986328125, -7.7414093017578125, 27.783069610595703, -8.85687255859375, 59.74542236328125, 69.658935546875, 63.215431213378906, 79.75430297851562, 63.216583251953125, -19.254104614257812, 29.346343994140625, 39.189453125, 64.98661804199219, 62.43341064453125, 59.12043762207031, 33.89605712890625, 10.081001281738281, 31.197067260742188, -56.59747314453125, 51.352272033691406, -5.0384521484375, 24.427886962890625, -4.34307861328125, 125.7750244140625, 37.2662353515625, 15.998138427734375, 7.709075927734375, 3.8170013427734375, -8.318954467773438, 93.64990234375, 9.673919677734375, 82.54864501953125, 42.380126953125, 20.927932739257812, 79.24594116210938, -2.2178955078125, -0.255340576171875, 59.71949768066406, -14.415475845336914, -34.038387298583984, 21.1064453125, 62.631439208984375, 1.815185546875, 56.442291259765625, 15.663841247558594, 93.2496337890625, 72.32766723632812, -29.554168701171875, 67.77996826171875, 20.257461547851562, 52.90730285644531, -36.0330810546875, 87.30911254882812, 23.470123291015625, 56.90260314941406, -29.409454345703125, -11.668937683105469, 80.3690185546875, 109.53961181640625, 66.42164611816406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000276.npy"}
{"epoch": 0.5780104712041885, "step": 277, "batch_size": 128, "mean": 20.62718391418457, "std": 39.42424392700195, "min": -92.20751953125, "p10": -24.975061225891107, "median": 22.63011932373047, "p90": 68.31638183593749, "max": 122.3465576171875, "pos_frac": 0.671875, "sample": [-0.459564208984375, 5.811553955078125, -4.525547027587891, 60.54217529296875, 27.9820556640625, -14.824798583984375, -33.36309814453125, -6.650115966796875, 30.015655517578125, 13.47418212890625, 25.68603515625, 29.289566040039062, 61.32671356201172, 66.23312377929688, -3.7159271240234375, 41.830108642578125, 46.11012649536133, -52.9300537109375, 45.674468994140625, -0.1712799072265625, 23.109176635742188, 6.852909088134766, 51.53987121582031, 37.3311767578125, 46.63398742675781, 52.66455078125, -0.42011070251464844, -1.2794647216796875, 48.1646728515625, 23.78972625732422, 41.05158233642578, -76.203125, -11.842899322509766, -28.883018493652344, -3.0254745483398438, 6.67578125, 59.37852478027344, 6.351127624511719, 41.891029357910156, 17.82159423828125, -92.20751953125, 3.582275390625, -13.728431701660156, -15.603973388671875, 9.082275390625, 35.209014892578125, 26.929840087890625, 61.171852111816406, 2.0489959716796875, 28.4072265625, -2.9766845703125, -13.126532554626465, 1.0262775421142578, 122.3465576171875, 74.514892578125, 0.0, -1.429962158203125, 56.00769805908203, 69.98155212402344, 22.15106201171875, 92.537841796875, 63.6004638671875, -58.46142578125, -3.0424861907958984, -0.0015869140625, 7.63116455078125, 47.69171905517578, -40.06402587890625, 83.8631591796875, 106.83987426757812, -43.367679595947266, 16.131805419921875, 89.23251342773438, 40.10337829589844, 5.673675537109375, -83.09490966796875, 34.72624206542969, -3.23956298828125, -18.6412353515625, 20.15049934387207, 15.66900634765625, 50.32422637939453, 24.0477294921875, -6.10284423828125, 27.379150390625, 49.00608825683594, 59.662208557128906, 66.02438354492188, -43.854637145996094, 72.8880615234375, 0.9273891448974609, 5.3567352294921875, -10.687149047851562, 28.01734161376953, 71.43869018554688, -4.6895751953125, 32.101844787597656, 43.135894775390625, 38.80479431152344, 73.87948608398438, 32.38893127441406, 12.397926330566406, 64.30291748046875, 51.410797119140625, 27.76097869873047, 67.60273742675781, 47.402130126953125, 77.29840087890625, -22.943389892578125, 0.0, 72.27996826171875, 111.53143310546875, 6.998420715332031, 27.685394287109375, -29.92071533203125, -8.489967346191406, -60.73167419433594, 65.503662109375, 10.18353271484375, 30.73674774169922, 46.89790344238281, -52.314544677734375, -0.9754142761230469, 0.33774566650390625, -5.598796844482422, 40.02339172363281, 47.889488220214844, -23.300222396850586], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000277.npy"}
{"epoch": 0.5801047120418849, "step": 278, "batch_size": 128, "mean": 19.940826416015625, "std": 41.610530853271484, "min": -111.03598022460938, "p10": -39.03274841308593, "median": 16.33513641357422, "p90": 72.29239807128906, "max": 122.35592651367188, "pos_frac": 0.7265625, "sample": [63.84912109375, 50.969112396240234, -45.61151123046875, 11.822517395019531, 4.4298858642578125, 34.03900146484375, 39.058868408203125, 11.048675537109375, 13.967489242553711, 77.55863952636719, 34.626556396484375, 61.070892333984375, 51.79878616333008, 40.76348876953125, 46.23274230957031, 41.761444091796875, -65.00044250488281, -10.59661865234375, 20.948898315429688, -0.530029296875, 13.388015747070312, 66.17510986328125, -2.575042724609375, -11.83935546875, 16.338958740234375, 72.27499389648438, 83.93731689453125, 53.33174133300781, 0.8994140625, 7.050048828125, -111.03598022460938, 109.08856201171875, -48.15428924560547, 8.789154052734375, 38.42279052734375, 25.377120971679688, 27.37016487121582, -38.6993408203125, 88.4681396484375, 16.13068389892578, 75.19367218017578, 82.63299560546875, 33.574920654296875, 9.0838623046875, -3.5953140258789062, 35.47808837890625, -1.7847137451171875, 89.9683837890625, 43.37262725830078, 29.056541442871094, 7.561250686645508, 2.70751953125, 3.2012481689453125, -46.122802734375, -12.82635498046875, -63.421295166015625, 2.985231399536133, 92.7158203125, -31.68499755859375, -48.348426818847656, 16.331314086914062, 43.737548828125, 21.771026611328125, 3.643798828125, 17.595458984375, 68.37158203125, -26.936859130859375, 61.56584167480469, 66.228515625, 25.481033325195312, 102.565185546875, 122.35592651367188, -8.317447662353516, -14.881881713867188, 55.92584228515625, -51.7437858581543, 64.10175323486328, 6.389984130859375, 50.450836181640625, 19.247833251953125, -68.34544372558594, 14.438011169433594, 2.235769271850586, 28.001129150390625, 56.227500915527344, -9.104537963867188, 20.274658203125, 45.322105407714844, 48.26068115234375, 59.212501525878906, 4.946220397949219, -8.10662841796875, 75.31462860107422, 106.44500732421875, 2.94244384765625, 72.3330078125, -39.810699462890625, 21.652175903320312, 22.380165100097656, -1.293060302734375, -7.7783355712890625, 5.270721435546875, 20.8897705078125, -50.933326721191406, -2.93792724609375, 16.2713623046875, 70.21432495117188, -10.925216674804688, 9.009552001953125, 20.274669647216797, 62.20215606689453, 4.444400787353516, -5.383636474609375, 55.683837890625, 1.3244857788085938, -22.678192138671875, 20.677099227905273, -64.48635864257812, 46.1548957824707, -16.608963012695312, 8.920242309570312, 5.7104339599609375, 11.47039794921875, 65.1455078125, -55.635589599609375, 33.3228759765625, -15.565757751464844, 54.46916198730469], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000278.npy"}
{"epoch": 0.5821989528795811, "step": 279, "batch_size": 128, "mean": 30.35663414001465, "std": 38.44716262817383, "min": -73.1579818725586, "p10": -10.249781799316406, "median": 27.150711059570312, "p90": 81.03783721923828, "max": 127.34246826171875, "pos_frac": 0.78125, "sample": [31.08258056640625, 66.47950744628906, 48.51310729980469, -20.918617248535156, 7.4966888427734375, 24.2320556640625, 14.698638916015625, 31.346832275390625, 3.6125335693359375, 62.75897216796875, 32.34747314453125, 98.08151245117188, 31.454833984375, -5.101776123046875, -1.827392578125, 27.101654052734375, 48.245574951171875, 56.59966278076172, 26.4466552734375, 27.19976806640625, 10.815750122070312, 31.71479034423828, -5.6890869140625, 29.76226806640625, 14.623344421386719, 24.040069580078125, 12.169597625732422, -13.856048583984375, -59.36180114746094, -3.8602676391601562, 29.5609130859375, 111.49725341796875, 79.0545654296875, 66.65618896484375, 42.97761535644531, -14.203338623046875, 3.8041458129882812, 56.404788970947266, -15.126358032226562, 8.252365112304688, 58.98674011230469, 62.32856750488281, 57.73388671875, 87.90406799316406, 2.8540267944335938, 77.53352355957031, -0.68963623046875, 7.6349639892578125, -6.7748260498046875, 45.36858367919922, 19.1363525390625, 42.28729248046875, -0.285919189453125, 23.54470443725586, 43.05023193359375, -11.093147277832031, 127.34246826171875, -73.1579818725586, 50.29803466796875, 94.1837158203125, -8.90887451171875, 11.3936767578125, 11.352783203125, 26.1627197265625, -3.0556182861328125, -10.218338012695312, -1.523956298828125, 68.39283752441406, -2.9089393615722656, 100.50579833984375, -10.323150634765625, 114.39678955078125, 27.029495239257812, 40.36515808105469, 5.736721038818359, -68.38722229003906, 14.18377685546875, 62.9573974609375, 102.10580444335938, -53.739227294921875, 6.84051513671875, 21.257980346679688, 57.661865234375, 62.391845703125, 38.66485595703125, 35.447601318359375, 20.474273681640625, 0.49977874755859375, 80.93754577636719, 1.55462646484375, 56.308441162109375, -15.164657592773438, 62.4896240234375, 31.21295166015625, 22.964372634887695, 7.8319244384765625, 81.2718505859375, 4.245204925537109, 34.087310791015625, 89.68350219726562, 36.63555908203125, 77.0503158569336, -6.4805145263671875, 62.63214111328125, 89.03570556640625, 72.62185668945312, 7.189300537109375, 9.491744995117188, 8.32525634765625, 75.52058410644531, -6.649932861328125, 28.84588623046875, 33.64837646484375, 112.486328125, 70.09138488769531, 43.23077392578125, 23.342498779296875, 32.66241455078125, -49.196533203125, -12.802108764648438, 91.7017822265625, -5.49774169921875, 66.703857421875, 6.781608581542969, 69.78271484375, 14.489288330078125, 29.61077880859375, 50.944061279296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000279.npy"}
{"epoch": 0.5842931937172775, "step": 280, "batch_size": 128, "mean": 27.162830352783203, "std": 38.26719284057617, "min": -77.58053588867188, "p10": -17.406283569335937, "median": 20.951148986816406, "p90": 83.14176177978516, "max": 113.56564331054688, "pos_frac": 0.765625, "sample": [59.54225158691406, 61.9431266784668, 1.040985107421875, 58.16387939453125, -4.29656982421875, -60.209503173828125, 72.97811889648438, 77.74615478515625, 91.23147583007812, 36.01319885253906, -1.1990509033203125, 40.70832061767578, 25.847267150878906, 50.21038818359375, 14.22296142578125, -24.979713439941406, 63.4725341796875, 53.32403564453125, 80.46737670898438, 1.9621429443359375, 9.63525390625, 90.12891387939453, -41.14595031738281, 84.03977966308594, -18.479888916015625, 59.73919677734375, 20.154617309570312, 13.103759765625, -0.2180328369140625, 67.77037048339844, 0.0, 13.793853759765625, 16.887496948242188, -16.9461669921875, 24.776092529296875, 113.56564331054688, -15.872573852539062, 85.5609130859375, 14.209785461425781, -9.47808837890625, 3.0535888671875, 31.36420440673828, 8.385047912597656, 16.540618896484375, 45.18616485595703, 22.991989135742188, 56.947509765625, -37.1939697265625, 89.44918823242188, 15.009185791015625, 23.088653564453125, 6.701545715332031, 6.4261474609375, 10.806489944458008, 58.47398376464844, 97.86700439453125, 22.757080078125, -8.042831420898438, 7.04510498046875, 66.04025268554688, 70.10549926757812, 24.473770141601562, 20.91368865966797, -33.001731872558594, -5.8318328857421875, 7.7437591552734375, -38.179901123046875, 88.70907592773438, 6.256195068359375, 45.490997314453125, 74.75152587890625, 29.474365234375, 99.3153076171875, 86.89285278320312, 98.92938232421875, -14.78509521484375, 57.03049850463867, 32.24298095703125, 21.31817626953125, 36.4337158203125, 40.079376220703125, 55.492401123046875, -15.421539306640625, 61.54454040527344, 82.75689697265625, 3.3014144897460938, -1.549835205078125, 14.057815551757812, -21.260955810546875, -0.79119873046875, 47.071868896484375, 12.82000732421875, 37.090782165527344, -40.338897705078125, 31.299144744873047, 77.8411636352539, -5.3195343017578125, 52.2880859375, 48.25018310546875, 18.1441650390625, 70.33279418945312, 18.265052795410156, 72.84346008300781, -16.053787231445312, 9.142364501953125, 9.052810668945312, 0.43251800537109375, 2.2050018310546875, 87.98560333251953, 0.8379287719726562, 20.988609313964844, 29.411178588867188, 3.6619300842285156, 49.288848876953125, -20.600799560546875, 53.79627227783203, 66.47100830078125, 99.63259887695312, -3.0626220703125, -77.58053588867188, -12.43878173828125, 8.490283966064453, -20.253616333007812, 6.380504608154297, 33.74803161621094, 0.9433231353759766, 50.345855712890625, -25.374176025390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000280.npy"}
{"epoch": 0.5863874345549738, "step": 281, "batch_size": 128, "mean": 25.695493698120117, "std": 37.11967086791992, "min": -68.53131103515625, "p10": -13.899577713012695, "median": 19.31505584716797, "p90": 78.9754425048828, "max": 151.72900390625, "pos_frac": 0.734375, "sample": [44.789306640625, 28.486663818359375, 0.44287109375, -2.4419784545898438, 30.205810546875, 7.59906005859375, 41.702972412109375, 36.63384246826172, 3.692474365234375, 87.8009033203125, 82.20831298828125, -40.09014892578125, -57.740447998046875, 83.33172607421875, 14.376277923583984, 33.6033935546875, 55.24302673339844, 33.99723815917969, 0.845184326171875, 6.282173156738281, 80.39012145996094, 4.48236083984375, 1.2991180419921875, -7.7552642822265625, 39.92987060546875, 0.5615234375, -27.533416748046875, 78.86181640625, 10.75579833984375, 23.724929809570312, 0.0, 27.425689697265625, 3.93255615234375, 44.63543701171875, 14.205333709716797, 65.68547058105469, 33.33062744140625, -29.310806274414062, 83.07145690917969, 31.23834228515625, -68.53131103515625, 19.056243896484375, 73.6502685546875, -10.8310546875, 0.0, 30.57177734375, 12.987762451171875, 46.101531982421875, -20.171142578125, 20.61602783203125, 18.733184814453125, -20.296356201171875, 26.78207015991211, 27.33380126953125, 1.197967529296875, 94.00979614257812, -7.39208984375, 23.588394165039062, -4.8026123046875, 37.24061584472656, 12.675323486328125, 20.780242919921875, -18.6451416015625, -1.6436843872070312, 61.16571044921875, 47.990081787109375, 15.25762939453125, -5.5045166015625, -3.200958251953125, 4.76507568359375, 14.33892822265625, 74.55851745605469, -5.8388671875, 79.24057006835938, 72.15602111816406, 26.798187255859375, 76.45271301269531, 5.87078857421875, 31.61407470703125, 26.67278289794922, 13.214729309082031, 49.65788269042969, 151.72900390625, 75.69619750976562, 65.77713012695312, -14.050697326660156, -0.1570892333984375, 61.85844039916992, 12.396194458007812, 1.5851516723632812, 96.47817993164062, 30.54241943359375, 53.61091613769531, -13.83481216430664, 19.573867797851562, -30.65240478515625, -4.5848388671875, 89.79412841796875, 70.482177734375, -3.179443359375, -18.39874267578125, 33.59124755859375, 0.0, 86.628173828125, 9.148933410644531, -11.92755126953125, 50.68762969970703, -27.04071044921875, 90.82321166992188, 97.78384399414062, 72.3943099975586, 72.96549224853516, 67.49612426757812, -2.7646484375, 73.12753295898438, 11.169586181640625, -7.3936920166015625, 32.703094482421875, 14.521461486816406, 21.789031982421875, 17.143692016601562, -40.12591552734375, -7.110832214355469, 73.22244262695312, 22.225433349609375, 6.6098175048828125, 18.70050048828125, -0.13153839111328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000281.npy"}
{"epoch": 0.5884816753926702, "step": 282, "batch_size": 128, "mean": 25.07317352294922, "std": 41.089393615722656, "min": -85.46737670898438, "p10": -24.742826080322263, "median": 20.09050750732422, "p90": 78.0416748046875, "max": 130.58148193359375, "pos_frac": 0.7578125, "sample": [71.08203125, -35.17097473144531, 26.021163940429688, 130.58148193359375, 33.710693359375, 4.768013000488281, -85.46737670898438, 78.514892578125, 37.543434143066406, 9.279327392578125, 6.227256774902344, 46.520084381103516, 22.87615966796875, -23.874969482421875, 6.38543701171875, 6.3536376953125, 72.73892211914062, -20.348297119140625, 68.23046875, 1.8062362670898438, -25.399871826171875, 19.614593505859375, -40.418338775634766, 38.36016845703125, 7.312122344970703, 82.04458618164062, 73.71449279785156, 44.95477294921875, -6.09197998046875, 7.34869384765625, -17.20684814453125, 12.702346801757812, 72.4130859375, 38.81767272949219, -5.7493896484375, 25.49298095703125, 50.06993865966797, 9.070087432861328, 29.77239990234375, 15.2177734375, 30.107711791992188, 39.96131896972656, 7.520477294921875, -35.9239501953125, 76.7799072265625, -24.015701293945312, -49.174346923828125, 32.31086730957031, 85.5460205078125, 5.660026550292969, 23.3360595703125, 13.843231201171875, -23.346435546875, 9.926521301269531, -20.405029296875, 98.25653076171875, 25.775711059570312, 17.19647216796875, 0.09751510620117188, -26.85589599609375, 7.84130859375, 7.08612060546875, -10.913238525390625, 56.38458251953125, 84.93267822265625, 55.42705535888672, 59.144317626953125, -8.434982299804688, -2.279388427734375, 65.86815643310547, -50.6737060546875, -2.55615234375, 71.05256652832031, 6.607452392578125, 8.781585693359375, 100.63681030273438, 66.62521362304688, -24.46123504638672, 71.503662109375, -17.735260009765625, -7.8408203125, 7.3154296875, 5.1990966796875, 87.98164367675781, 33.683197021484375, -29.589576721191406, 33.401153564453125, 11.181198120117188, 62.64862060546875, 74.6900634765625, 95.85293579101562, 30.221038818359375, 8.38128662109375, 40.497802734375, -55.70806884765625, 20.566421508789062, 56.61201477050781, 82.26518249511719, 6.935403823852539, -3.9619140625, 76.71197509765625, 129.96759033203125, 64.78983306884766, 10.050350189208984, 68.303466796875, 4.4425811767578125, -73.50369262695312, 12.398834228515625, 45.615478515625, 24.537166595458984, 72.81085205078125, 24.085159301757812, 38.25726318359375, 35.966453552246094, 5.126594543457031, 77.8388671875, 84.0653076171875, 26.118728637695312, 81.19570922851562, 44.951629638671875, 19.13348388671875, -29.57284927368164, -1.5852432250976562, -38.5023193359375, 73.55023193359375, -13.503448486328125, 10.804733276367188, 53.7279052734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000282.npy"}
{"epoch": 0.5905759162303665, "step": 283, "batch_size": 128, "mean": 25.48038673400879, "std": 38.34830093383789, "min": -61.41816711425781, "p10": -24.952801513671876, "median": 23.716461181640625, "p90": 72.77844696044922, "max": 122.11819458007812, "pos_frac": 0.734375, "sample": [-7.4804229736328125, 26.224868774414062, 72.84852600097656, -23.028961181640625, 12.345609664916992, -37.740447998046875, 63.36492919921875, -33.234474182128906, 38.8275146484375, -0.69085693359375, 38.19036865234375, 75.78411865234375, 40.558448791503906, 45.84246063232422, 0.5194721221923828, -30.401657104492188, -7.688751220703125, 47.67607879638672, 11.289443969726562, -23.26043701171875, 109.67120361328125, 93.11044311523438, 28.141632080078125, -23.68939208984375, 78.25552368164062, 62.530975341796875, 42.63792419433594, -6.4259796142578125, 39.09416198730469, 10.099899291992188, 26.437591552734375, 31.947525024414062, 32.776641845703125, 61.20374298095703, 30.425819396972656, 52.394447326660156, -5.4937286376953125, -0.7176723480224609, 34.88653564453125, 26.581085205078125, -8.11663818359375, 61.99395751953125, -17.229217529296875, -41.58938217163086, 54.061798095703125, 44.98066711425781, 5.299896240234375, -8.356048583984375, -50.38818359375, 56.450958251953125, 47.4271240234375, 75.53097534179688, 57.509033203125, 46.17839050292969, 89.96812438964844, 3.4354629516601562, 9.283369064331055, 48.960723876953125, 26.82257080078125, 52.792510986328125, 21.69219970703125, 52.4967041015625, 91.3453369140625, 87.8203125, 65.0064697265625, -2.95623779296875, 70.04022216796875, 22.63720703125, 24.088424682617188, 11.350326538085938, -28.85022735595703, 73.51805877685547, -44.34864807128906, 59.8970947265625, 20.733078002929688, 8.301910400390625, 63.754119873046875, 54.94508361816406, 0.0, -2.4678955078125, -24.745361328125, 11.742446899414062, 58.953651428222656, -52.63026428222656, 70.28057861328125, 49.943359375, 44.08246612548828, 8.685211181640625, -25.43682861328125, 21.521347045898438, 10.736068725585938, 2.6534500122070312, 3.3796844482421875, -16.01416015625, 22.8043212890625, 72.7484130859375, 59.91619873046875, 41.846771240234375, -60.86578369140625, 64.73956298828125, 1.609598159790039, 22.4703369140625, 119.31207275390625, 5.122547149658203, 26.959686279296875, 18.071868896484375, 61.7459716796875, -61.41816711425781, 15.428436279296875, 8.302413940429688, 87.12542724609375, 19.235595703125, -0.8510608673095703, -0.49056053161621094, 0.0, 38.31146240234375, 56.231285095214844, 122.11819458007812, 19.816062927246094, 2.4313831329345703, 26.22895050048828, 64.76385498046875, 23.344497680664062, 19.218917846679688, -32.4930419921875, 57.516021728515625, -37.4100341796875, -23.38721466064453], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000283.npy"}
{"epoch": 0.5926701570680628, "step": 284, "batch_size": 128, "mean": 25.24908447265625, "std": 43.75712203979492, "min": -97.99798583984375, "p10": -34.15974502563476, "median": 21.625595092773438, "p90": 84.47662658691405, "max": 119.8419189453125, "pos_frac": 0.734375, "sample": [-27.000167846679688, 43.36798095703125, -36.85894775390625, -97.99798583984375, 32.11119079589844, 28.759613037109375, 57.134429931640625, 88.18475341796875, -34.55888366699219, 21.448486328125, 45.627410888671875, 86.46389770507812, 40.046165466308594, 11.207015991210938, 21.445526123046875, 46.81349182128906, 80.2852783203125, 57.347900390625, 83.62493896484375, 6.1666259765625, 79.86187744140625, 89.43505859375, -12.216827392578125, 45.7099609375, -5.785070419311523, 10.848907470703125, 21.1053466796875, -36.85614013671875, 10.841011047363281, -33.289093017578125, -43.49491882324219, 27.22296142578125, 34.52044677734375, -18.241058349609375, 7.9783935546875, 29.78643798828125, 54.760406494140625, 30.39434814453125, 42.65979766845703, 2.6833953857421875, -69.08207702636719, 9.86767578125, 49.32704162597656, 23.887725830078125, -4.207511901855469, -4.192901611328125, -33.988685607910156, 38.2413330078125, 13.681976318359375, 110.55950927734375, 75.541259765625, -47.16748046875, 31.93304443359375, 20.234619140625, 81.56549072265625, 53.85662841796875, -1.587493896484375, -1.8982620239257812, 88.19314575195312, 42.937835693359375, 39.93316650390625, 102.06556701660156, 16.956954956054688, -6.3607177734375, 58.17497253417969, 41.09710693359375, 105.30145263671875, 20.999755859375, 43.93743133544922, 9.870447158813477, 57.508636474609375, 4.830535888671875, -8.498516082763672, 28.72540283203125, 4.133356094360352, 100.50399780273438, 1.4052734375, 6.7561187744140625, 13.861297607421875, -6.404205322265625, 96.89033508300781, 81.99858093261719, 53.862972259521484, 90.81716918945312, -10.944732666015625, -83.21426391601562, 9.019445419311523, -8.678131103515625, 31.238967895507812, -29.440216064453125, 58.09324645996094, 16.74456787109375, 55.68596267700195, 73.48013305664062, 1.1249656677246094, -2.5044097900390625, 22.77001953125, -58.74729919433594, 1.5626068115234375, 41.38075256347656, 79.3243408203125, 12.604965209960938, 54.65068054199219, 11.189208984375, -1.3845710754394531, 69.35964965820312, 12.35333251953125, 21.802703857421875, 19.07012939453125, 72.07463073730469, 75.58617401123047, 119.8419189453125, 27.82958984375, -19.087799072265625, -50.61143493652344, -59.11053466796875, 54.300689697265625, -43.58404541015625, 5.962005615234375, -14.41851806640625, 96.30694580078125, 18.268966674804688, 57.0037841796875, 109.39144897460938, 58.493797302246094, -3.3649024963378906, -47.49395751953125, 50.33782958984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000284.npy"}
{"epoch": 0.5947643979057592, "step": 285, "batch_size": 128, "mean": 20.33536720275879, "std": 36.73875045776367, "min": -67.086181640625, "p10": -26.208322143554685, "median": 16.886367797851562, "p90": 71.33976593017579, "max": 97.76577758789062, "pos_frac": 0.6953125, "sample": [10.218536376953125, 49.098968505859375, 47.4927978515625, 23.77392578125, 69.48513793945312, 80.55496215820312, 68.9541015625, -52.53002166748047, 63.229705810546875, 24.3399658203125, 13.218292236328125, -14.410308837890625, 9.80291748046875, -48.5374755859375, -12.64434814453125, 24.504928588867188, 16.144912719726562, -37.06329345703125, -17.3236083984375, -4.120147705078125, -4.73687744140625, 52.306640625, -45.98973846435547, 13.782455444335938, 31.206626892089844, 40.573272705078125, 82.35812377929688, 41.434783935546875, 31.609664916992188, 28.744415283203125, 97.76577758789062, -28.39837646484375, 17.627822875976562, -2.110198974609375, 79.64517211914062, 82.58187866210938, 65.31640625, -14.428466796875, -38.14830017089844, 4.5155029296875, -33.763427734375, -22.847412109375, -10.39593505859375, -4.929107666015625, -0.41911888122558594, 42.869140625, 19.69732666015625, 34.226226806640625, 0.42032623291015625, 3.2074432373046875, 71.30665588378906, 10.130271911621094, 3.2831192016601562, -11.132331848144531, 84.18853759765625, 2.9120407104492188, -4.282661437988281, -5.870368957519531, 58.89945983886719, 11.240486145019531, 9.9090576171875, 49.3905029296875, 59.987220764160156, 37.31376647949219, 31.6011962890625, 23.57806396484375, -21.294586181640625, 7.790538787841797, -21.600296020507812, 35.4708251953125, -44.4720458984375, 59.629150390625, 18.581222534179688, 13.8516845703125, 55.287376403808594, 4.1327362060546875, 34.745880126953125, 26.14285659790039, 22.863067626953125, 13.627899169921875, 43.7811279296875, 65.72979736328125, 5.138031005859375, -1.6515655517578125, -9.526565551757812, 5.9444580078125, 19.98681640625, 70.92329406738281, 79.9256591796875, 61.89332962036133, 55.07625198364258, 63.969940185546875, -2.7471160888671875, -16.52496337890625, -20.206146240234375, 17.814300537109375, 29.333251953125, 39.47625732421875, -18.431747436523438, 9.3067626953125, 54.16550827026367, -30.557823181152344, 41.358802795410156, -29.760223388671875, -4.75128173828125, -19.162399291992188, 6.8424072265625, -11.335628509521484, 90.68212890625, 52.156890869140625, -67.086181640625, 13.20998764038086, 30.81683349609375, -27.37579345703125, 7.804542541503906, -59.2080078125, 67.35564422607422, 3.4420700073242188, 73.43832397460938, 22.699432373046875, 13.34075927734375, 44.785247802734375, 32.5548095703125, 73.6996841430664, -25.707977294921875, 71.41702270507812, 85.32522583007812, 78.44244384765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000285.npy"}
{"epoch": 0.5968586387434555, "step": 286, "batch_size": 128, "mean": 22.45984649658203, "std": 41.10821533203125, "min": -75.72923278808594, "p10": -32.68833618164062, "median": 22.879165649414062, "p90": 74.30549621582031, "max": 122.53851318359375, "pos_frac": 0.6875, "sample": [-7.087921142578125, -13.67926025390625, 25.636566162109375, 37.67787170410156, 15.466880798339844, 35.64375305175781, 11.01129150390625, 1.2188339233398438, 11.7818603515625, 28.680450439453125, 41.15674591064453, 12.605224609375, -8.3258056640625, 38.08378601074219, -54.33367919921875, 90.71476745605469, 8.403228759765625, 79.84062194824219, 59.023887634277344, 71.9718017578125, -25.66748046875, 30.598358154296875, 22.90594482421875, 17.796051025390625, 12.61260986328125, 18.86090087890625, -6.850311279296875, -4.949501037597656, 35.64556121826172, -5.143035888671875, -75.72923278808594, 1.6431121826171875, 71.2489242553711, 65.47514343261719, -15.747528076171875, 33.593772888183594, 5.39288330078125, 31.009552001953125, 6.345867156982422, 66.87898254394531, -10.604278564453125, 65.03009033203125, 63.640899658203125, -42.687599182128906, 51.7120361328125, -41.302490234375, 25.825439453125, 122.53851318359375, 31.1470947265625, -10.813690185546875, 73.76388549804688, -63.80717468261719, 83.00448608398438, 58.234222412109375, 22.852386474609375, 80.67706298828125, 0.0065155029296875, 79.32073974609375, -30.00580596923828, 94.75509643554688, -3.4561614990234375, 0.0, -8.553466796875, -55.325408935546875, 26.319591522216797, 40.27349853515625, -4.195987701416016, -32.439239501953125, -16.916046142578125, 17.8369140625, -33.269561767578125, 52.557159423828125, 61.69783020019531, 35.742401123046875, -18.62531280517578, 19.8394775390625, 71.58935546875, 32.24421691894531, -35.51871109008789, 14.084484100341797, 34.4677734375, 17.119243621826172, 7.9190216064453125, 109.38238525390625, 74.61044311523438, 29.278564453125, 24.112579345703125, -43.591033935546875, 62.860565185546875, -61.01434326171875, 44.77794647216797, 73.23284912109375, 26.970458984375, 42.538665771484375, 74.1748046875, 62.81298828125, 36.896392822265625, 43.6898193359375, 57.98267364501953, 53.97320556640625, 97.22845458984375, 16.792938232421875, 74.66827392578125, -10.648902893066406, 33.558349609375, -27.5899658203125, 96.586669921875, 42.363426208496094, -3.01715087890625, -26.798587799072266, -33.820098876953125, 69.26971435546875, -0.16698455810546875, -29.230873107910156, -50.91926574707031, 113.2730712890625, -19.088470458984375, -34.29087829589844, 22.64183807373047, 13.5738525390625, 24.361587524414062, 46.255859375, -2.856372833251953, 17.206710815429688, 36.179931640625, -16.946578979492188, 62.804290771484375, 2.6905441284179688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000286.npy"}
{"epoch": 0.5989528795811518, "step": 287, "batch_size": 128, "mean": 23.744396209716797, "std": 38.34843063354492, "min": -85.94366455078125, "p10": -23.601524353027337, "median": 24.887502670288086, "p90": 69.117138671875, "max": 111.67262268066406, "pos_frac": 0.7265625, "sample": [58.76893615722656, -14.283035278320312, -31.720306396484375, -9.186128616333008, -17.233734130859375, -27.783798217773438, 69.20553588867188, 64.80000305175781, 50.054786682128906, 75.38409423828125, 60.944976806640625, 11.146560668945312, 9.584829330444336, 76.30145263671875, 5.797607421875, 11.312164306640625, -18.881759643554688, -33.055328369140625, -7.025413513183594, 18.146095275878906, -16.983978271484375, 111.67262268066406, 65.40644836425781, 11.049835205078125, 33.30426025390625, -8.04498291015625, -8.424270629882812, 33.2608642578125, 51.0736083984375, 32.876739501953125, 5.08941650390625, 65.24984741210938, 48.21064758300781, 50.07318115234375, -3.4163818359375, 4.0283355712890625, 34.03102111816406, 60.137603759765625, -0.5659847259521484, 67.4637222290039, 65.78459930419922, 46.979461669921875, 7.18865966796875, 22.8043212890625, 80.68019104003906, 0.0067424774169921875, 0.1710052490234375, 61.566680908203125, 30.459793090820312, 94.30126953125, 24.272476196289062, 68.53805541992188, 47.81634521484375, 41.37489318847656, 13.79888916015625, 31.085845947265625, 23.90410614013672, -12.07989501953125, 49.2127685546875, -10.371826171875, 69.07925415039062, 41.34880065917969, 33.02235412597656, 25.50252914428711, 1.7547435760498047, 54.263580322265625, 62.7349853515625, 56.54522705078125, 50.2681884765625, 41.355010986328125, -67.4781494140625, -1.7447319030761719, 53.3629150390625, 80.15308380126953, 37.74542236328125, 52.0452880859375, 53.42803955078125, 55.49131774902344, 17.1278076171875, 52.313724517822266, 96.28274536132812, -32.9072265625, -46.28376770019531, 1.3013763427734375, -33.162445068359375, 34.37332534790039, -21.220718383789062, 35.58399963378906, -21.95867919921875, 13.257659912109375, 58.11351013183594, 78.38189697265625, -85.94366455078125, -12.206298828125, 28.739334106445312, -10.75732421875, 39.7558479309082, 7.804590225219727, -63.882904052734375, 28.864883422851562, 5.315582275390625, 14.0447998046875, 3.9432373046875, 19.737640380859375, -20.442047119140625, -27.434829711914062, 63.25623321533203, -68.1878662109375, 83.70639038085938, -5.2478790283203125, 80.1541748046875, -2.9200439453125, 65.59390258789062, 42.03644561767578, 76.08859252929688, 6.685302734375, 11.912704467773438, 51.552734375, 42.751495361328125, -4.269390106201172, 14.77220344543457, 5.603507995605469, -45.64862060546875, -36.06463623046875, 12.090667724609375, 76.49751281738281, -6.0676422119140625, 42.10258483886719], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000287.npy"}
{"epoch": 0.6010471204188481, "step": 288, "batch_size": 128, "mean": 22.937244415283203, "std": 38.27833557128906, "min": -76.68594360351562, "p10": -22.51708526611328, "median": 17.588958740234375, "p90": 71.32326812744141, "max": 123.63580322265625, "pos_frac": 0.71875, "sample": [11.465179443359375, 0.9820938110351562, -12.868804931640625, -2.051422119140625, 45.51229476928711, -1.312255859375, -6.9231719970703125, -22.425262451171875, 8.667938232421875, 60.33349609375, 23.4288330078125, 100.89236450195312, 1.8187713623046875, 69.05296325683594, 37.804779052734375, -14.714706420898438, -5.790275573730469, 41.28489685058594, 11.1021728515625, 0.0, 18.0394287109375, -10.90185546875, 21.7239990234375, 7.064662933349609, 31.114891052246094, -3.6353759765625, -23.464691162109375, 2.9259033203125, -36.971405029296875, 18.571533203125, 104.632568359375, 40.6627197265625, 24.80724334716797, 48.70979309082031, 8.802947998046875, 0.0, -4.4994354248046875, 9.830329895019531, 10.73748779296875, 16.8580322265625, -17.840011596679688, 53.73371124267578, -48.1776123046875, -46.43980407714844, 10.05728530883789, 63.48289489746094, 92.62687683105469, 76.00750732421875, -76.68594360351562, 1.169729232788086, 66.64138793945312, -5.973533630371094, 39.80816650390625, 13.346389770507812, 37.79486083984375, -60.126617431640625, -3.5374679565429688, 32.8369140625, 58.8231201171875, -44.470245361328125, 123.63580322265625, -43.427528381347656, -5.648529052734375, 49.81681823730469, 107.93109130859375, 11.005950927734375, -12.224868774414062, -31.328155517578125, 70.37106323242188, 15.351348876953125, 65.136962890625, 18.76190185546875, 0.1263427734375, 33.20170593261719, 10.85943603515625, 14.831207275390625, 11.290283203125, 24.368362426757812, -20.59124755859375, 24.98739242553711, 47.167266845703125, 33.38615417480469, -17.13232421875, -19.501220703125, -13.11822509765625, -30.937850952148438, 13.176525115966797, 86.18804931640625, 71.0494384765625, 12.056900024414062, 35.64129638671875, 9.482406616210938, 47.04205322265625, -1.8328475952148438, 58.255157470703125, 57.23393630981445, 51.35083770751953, -31.472442626953125, 79.984619140625, 17.13848876953125, 39.740509033203125, 34.78717041015625, -38.00225830078125, 99.90179443359375, 9.130348205566406, 27.8677978515625, 18.977901458740234, 23.6165771484375, 32.29669189453125, -1.3858795166015625, -22.731338500976562, 45.632720947265625, 42.50096130371094, 65.32672119140625, 41.797576904296875, 71.32357788085938, 27.90240478515625, 9.10272216796875, 26.08392333984375, 67.3134765625, 16.584148406982422, 117.97946166992188, 44.584197998046875, 78.33599853515625, 71.32313537597656, 5.9786376953125, 73.79074096679688, 58.247840881347656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000288.npy"}
{"epoch": 0.6031413612565445, "step": 289, "batch_size": 128, "mean": 17.4959774017334, "std": 38.638851165771484, "min": -70.48272705078125, "p10": -32.761834716796876, "median": 13.03948974609375, "p90": 71.72653656005859, "max": 97.96063232421875, "pos_frac": 0.65625, "sample": [-61.175323486328125, -10.29388427734375, 15.4775390625, -37.202903747558594, -7.842529296875, 1.618377685546875, 11.026275634765625, -14.34228515625, 36.5965576171875, -34.082794189453125, 71.55889892578125, 0.060302734375, 2.1761951446533203, 38.7093505859375, -15.743576049804688, 50.36590576171875, -21.8265380859375, 61.967376708984375, 10.130523681640625, -40.942596435546875, 95.83258056640625, 75.08544921875, 62.55908203125, 14.19012451171875, 0.90802001953125, -9.713851928710938, -8.17626953125, 4.6966400146484375, 10.431884765625, 44.448211669921875, -28.024368286132812, 51.0924072265625, -4.993438720703125, 22.89385986328125, -0.7816390991210938, 0.0, 5.5931243896484375, 42.9063720703125, -11.984954833984375, 12.50897216796875, 22.176284790039062, -13.37939453125, -42.00688171386719, -70.10589599609375, 53.47021484375, -21.41204833984375, 39.394775390625, 51.0927734375, 3.577789306640625, 1.316802978515625, -0.5323028564453125, 37.393341064453125, -70.48272705078125, -32.195709228515625, 9.73187255859375, 20.973365783691406, -4.344682693481445, -4.384590148925781, 21.930627822875977, 91.88583374023438, -42.366943359375, 72.20286560058594, 23.877914428710938, 35.061004638671875, 42.9930419921875, 70.0826187133789, -13.656951904296875, -64.80146789550781, -14.09075927734375, 4.9337005615234375, 72.32847595214844, -20.567031860351562, 11.00299072265625, 58.97336196899414, 22.212257385253906, 15.2669677734375, 6.1209716796875, 45.2335205078125, 29.728134155273438, 7.720123291015625, 43.677093505859375, -17.1461181640625, -53.890899658203125, 78.01715087890625, -13.186203002929688, 44.54938507080078, -13.69158935546875, 72.11769104003906, -17.243057250976562, 77.40475463867188, -8.634368896484375, 88.26605224609375, 66.00720977783203, 7.05316162109375, 20.829513549804688, 26.33416748046875, 60.4627685546875, 11.429611206054688, 45.19445037841797, 39.66270446777344, 75.38487243652344, 47.08025360107422, 78.4779052734375, 55.178565979003906, -15.7882080078125, 48.28594970703125, 14.771331787109375, -3.7535247802734375, 60.195770263671875, 94.1796875, -47.06793212890625, 60.314208984375, 69.16708374023438, -12.615005493164062, 70.43235778808594, -36.93182373046875, -36.82257080078125, 13.57000732421875, -18.368515014648438, 67.60729217529297, 24.8897705078125, 24.468292236328125, 97.96063232421875, 2.9478912353515625, 14.21649169921875, -6.2394561767578125, 31.5987548828125, 17.0703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000289.npy"}
{"epoch": 0.6052356020942409, "step": 290, "batch_size": 128, "mean": 22.00309944152832, "std": 36.94159698486328, "min": -70.59149169921875, "p10": -23.098876953125, "median": 15.07586669921875, "p90": 74.11491088867187, "max": 107.24114990234375, "pos_frac": 0.6875, "sample": [81.17303466796875, 9.42669677734375, -3.88995361328125, 54.40397644042969, 42.2991943359375, -64.60574340820312, 1.8445587158203125, 57.56304931640625, -4.95361328125, 12.56390380859375, 100.19598388671875, 71.86416625976562, 32.14056396484375, -10.251327514648438, 68.7769775390625, 40.40766906738281, 9.352588653564453, 30.158294677734375, 78.38088989257812, 51.69303894042969, 70.33758544921875, 2.2896575927734375, 0.696258544921875, 16.338821411132812, 107.24114990234375, 5.5396728515625, -45.88621520996094, -62.91033935546875, -0.5817718505859375, 3.305938720703125, 86.56005859375, 3.216930389404297, -4.1207275390625, 38.53125, 78.50897216796875, 87.2081298828125, 65.24357604980469, 32.953163146972656, -3.684326171875, -31.678619384765625, 58.83937072753906, 25.210500717163086, -2.743194580078125, 62.129974365234375, 27.06756591796875, -0.8776130676269531, 1.82537841796875, -8.294891357421875, 23.375526428222656, 33.91522216796875, -7.718719482421875, -1.093017578125, 32.27824401855469, 45.04114532470703, 31.026206970214844, -23.866668701171875, 3.235321044921875, 51.81175231933594, 8.551239013671875, 13.812911987304688, 7.262298583984375, 13.794929504394531, -2.9245223999023438, 66.498291015625, 24.6202392578125, 22.492279052734375, 16.95899200439453, 60.8616943359375, 18.081802368164062, 71.121826171875, -22.898101806640625, 24.092758178710938, -4.00830078125, 0.438323974609375, -3.361785888671875, 76.34603881835938, -7.751764297485352, -35.231964111328125, 2.727264404296875, 44.87164306640625, -2.484344482421875, -12.956962585449219, -22.686203002929688, 72.86471557617188, 50.752716064453125, 61.228607177734375, 52.696434020996094, -7.6419677734375, -70.59149169921875, 4.3019561767578125, 54.617218017578125, -32.11824035644531, 35.91961669921875, -0.18597412109375, -28.152435302734375, 5.743949890136719, -10.059417724609375, -5.8083648681640625, -28.649246215820312, 60.18511962890625, 6.766851425170898, -8.745635986328125, 33.115234375, 66.68988037109375, 78.00732421875, 40.601837158203125, 74.06402587890625, 76.47686767578125, 81.8175048828125, 10.908348083496094, -22.701187133789062, 17.52655029296875, 63.13726806640625, 33.09576416015625, 13.46337890625, -25.069137573242188, -34.494911193847656, 20.12103271484375, -23.567352294921875, 64.75093078613281, 2.84197998046875, -4.1517791748046875, 11.579116821289062, 37.28206253051758, -10.225234985351562, 28.983657836914062, 77.74172973632812, 74.233642578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000290.npy"}
{"epoch": 0.6073298429319371, "step": 291, "batch_size": 128, "mean": 29.35886001586914, "std": 40.09138488769531, "min": -91.07525634765625, "p10": -17.075526428222656, "median": 25.623233795166016, "p90": 81.23993225097657, "max": 118.11297607421875, "pos_frac": 0.7421875, "sample": [79.67007446289062, -0.3190155029296875, 102.71688842773438, 4.4934234619140625, 102.67143249511719, 24.51507568359375, 92.67262268066406, 69.538330078125, 25.0704345703125, -20.224456787109375, 5.51507568359375, 58.255645751953125, 11.385917663574219, -5.30523681640625, 84.95346069335938, 8.241744995117188, 83.22366333007812, -14.224822998046875, 50.296661376953125, 18.49226188659668, 49.328460693359375, 27.359725952148438, 54.101959228515625, 32.794342041015625, -3.968597412109375, 62.2213134765625, 36.98561096191406, 1.88421630859375, 50.473663330078125, -19.330078125, -74.57331848144531, -4.792201995849609, -22.36480712890625, 13.09225845336914, -91.07525634765625, 17.2197265625, 59.783935546875, 31.663818359375, 26.66082763671875, 14.173248291015625, 85.02874755859375, -19.026016235351562, -30.103981018066406, 26.17603302001953, 81.19093322753906, 17.360443115234375, 1.4757843017578125, 44.6080322265625, 51.563720703125, 4.38031005859375, 64.50836181640625, 38.97843551635742, 48.21277618408203, -14.856231689453125, 68.91238403320312, 66.11363220214844, 15.92449951171875, -15.644485473632812, -28.782180786132812, 67.45816040039062, 88.8719482421875, 8.0997314453125, -4.78009033203125, -7.68597412109375, 49.42266845703125, 9.12054443359375, 8.8638916015625, 37.30232238769531, -26.580902099609375, 75.55923461914062, -16.937606811523438, 17.436431884765625, 18.974239349365234, 118.11297607421875, 33.52543640136719, 23.751861572265625, 22.862876892089844, 68.51318359375, 76.33078002929688, -5.29864501953125, 54.443115234375, 0.39785003662109375, -6.4425048828125, -16.39117431640625, 55.23899841308594, 90.36868286132812, 45.08750534057617, -32.84747314453125, 0.0, -8.228958129882812, -17.3973388671875, 69.86419677734375, 13.6259765625, -65.29204559326172, 110.66561889648438, 57.56705856323242, 56.190399169921875, 26.70086669921875, 80.86726379394531, -9.991851806640625, 57.388519287109375, 71.10580444335938, 21.63946533203125, 69.53030395507812, -0.314727783203125, 48.19573974609375, 13.77392578125, 14.8447265625, 112.9556884765625, -10.9542236328125, -39.565887451171875, 69.90493774414062, 17.925148010253906, 26.64467430114746, 58.2713508605957, 30.35772705078125, 61.69921875, -1.3765411376953125, 90.6153564453125, 75.15431213378906, 20.70263671875, 11.370071411132812, 1.03948974609375, -10.545520782470703, 81.35426330566406, 45.363677978515625, 50.50846862792969, 53.69694519042969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000291.npy"}
{"epoch": 0.6094240837696335, "step": 292, "batch_size": 128, "mean": 24.067129135131836, "std": 37.882598876953125, "min": -106.83383178710938, "p10": -21.675896263122556, "median": 20.73554229736328, "p90": 72.91465454101562, "max": 134.8402099609375, "pos_frac": 0.7578125, "sample": [14.48785400390625, 3.051605224609375, 57.6146240234375, -37.1326904296875, 65.89559173583984, 47.51874542236328, 72.78713989257812, 87.53314208984375, 18.329254150390625, 69.7259521484375, 108.1129150390625, 57.353118896484375, -22.653076171875, -0.253570556640625, 37.009368896484375, 20.551025390625, -21.257104873657227, -6.41847038269043, 4.02252197265625, -4.27459716796875, 14.080734252929688, 9.161972045898438, 25.510345458984375, 49.498260498046875, 57.139747619628906, 34.09344482421875, 55.83705139160156, 13.48992919921875, 18.5902099609375, -3.889528274536133, 53.615142822265625, 55.54444885253906, 1.0563812255859375, 33.3326416015625, 56.40289306640625, 15.84206771850586, 18.42620849609375, -25.4361572265625, -72.06167602539062, 10.794853210449219, 40.30805969238281, 22.20964813232422, 32.159637451171875, 22.76483917236328, -28.737884521484375, 5.6949310302734375, 26.478912353515625, -45.1463623046875, -4.1390838623046875, 63.667816162109375, -18.89202880859375, 88.69354248046875, -16.35595703125, -12.302291870117188, -6.225982666015625, 22.762741088867188, 5.159881591796875, 3.2979736328125, 11.10760498046875, 10.385269165039062, 60.38178253173828, 19.3887939453125, 74.9056396484375, -19.940963745117188, -5.31341552734375, 8.955535888671875, 16.758514404296875, 45.1209716796875, 134.8402099609375, 36.963623046875, 5.79547119140625, 31.620086669921875, 29.76519775390625, 27.14459228515625, 54.92497253417969, 95.91510009765625, 13.975894927978516, 77.10072326660156, 20.281375885009766, -1.7137737274169922, -34.097442626953125, -106.83383178710938, 88.0074462890625, 37.16084289550781, 71.11592864990234, 49.3590087890625, 53.80279541015625, 20.920059204101562, 4.643871307373047, 56.38232421875, -27.52606201171875, 5.553192138671875, -8.135147094726562, 17.63409423828125, 26.71721649169922, -30.11334228515625, -3.299957275390625, -18.881072998046875, 20.986337661743164, 73.21218872070312, 46.87310791015625, 44.423728942871094, 9.569992065429688, 10.30038070678711, 23.121360778808594, 58.10833740234375, 47.68896484375, -0.35198974609375, 9.230361938476562, 20.368026733398438, -28.007949829101562, 49.667694091796875, 38.2216796875, 0.276275634765625, 74.58709716796875, 73.45556640625, 99.54312133789062, -42.477569580078125, 47.00177001953125, 11.83917236328125, 90.7205810546875, -51.72314453125, -16.419158935546875, 45.230010986328125, 59.63568115234375, 62.46336364746094, 31.7255859375, 26.11821746826172], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000292.npy"}
{"epoch": 0.6115183246073298, "step": 293, "batch_size": 128, "mean": 24.149169921875, "std": 37.535579681396484, "min": -73.66535949707031, "p10": -20.212284088134762, "median": 18.35549545288086, "p90": 72.44558944702148, "max": 118.488037109375, "pos_frac": 0.734375, "sample": [62.755279541015625, -12.179107666015625, 74.64022827148438, 76.21810913085938, 16.503875732421875, -37.72296142578125, 15.373748779296875, 52.44970703125, 5.042938232421875, 16.90831756591797, -9.083587646484375, 48.558631896972656, 26.08892822265625, 19.407546997070312, 28.22587776184082, 9.769737243652344, -19.38188934326172, -0.141326904296875, -24.984817504882812, -22.149871826171875, 27.281044006347656, -2.338470458984375, 80.958740234375, 71.93060302734375, 10.680267333984375, 0.0, 65.34295654296875, 40.46128845214844, 86.482421875, 17.406959533691406, 61.368896484375, -25.60223388671875, -8.788772583007812, 9.972824096679688, 17.2703857421875, -11.722549438476562, 10.6090087890625, 60.21692657470703, 1.938629150390625, 12.055389404296875, 61.49949645996094, 0.07737159729003906, 65.061767578125, 63.85443115234375, 14.8436279296875, -61.776458740234375, 58.154083251953125, 71.29490661621094, 92.18136596679688, 26.95073699951172, 35.93769836425781, 34.64520263671875, 67.18440246582031, 55.3990478515625, -3.7132415771484375, 58.32127380371094, 24.33770751953125, -11.71563720703125, 8.943717956542969, 16.428985595703125, -2.4086456298828125, 79.29252624511719, 43.75634765625, -28.870450973510742, -43.048561096191406, 13.931686401367188, 8.354106903076172, -17.8045654296875, 26.224899291992188, -6.836311340332031, -73.66535949707031, -24.27593994140625, 31.576950073242188, 108.15850830078125, -46.059661865234375, 4.609466552734375, -6.18902587890625, 61.629608154296875, -16.24908447265625, 27.36773681640625, 15.342376708984375, -0.298004150390625, 32.865447998046875, 10.590774536132812, 69.3970947265625, 12.40985107421875, 38.066558837890625, 67.87893676757812, 83.50631713867188, -16.6429443359375, -13.01568603515625, 94.0921630859375, 0.5038604736328125, 69.5900650024414, 35.79736328125, 33.7623291015625, 20.281593322753906, -64.07707214355469, 20.896286010742188, 15.91208267211914, 37.052001953125, 57.663055419921875, 85.71751403808594, 63.070770263671875, 46.969482421875, 73.64722442626953, 17.82360076904297, 25.094284057617188, -26.182098388671875, 1.4676284790039062, 3.5166549682617188, -5.308502197265625, -2.05810546875, 51.58758544921875, 111.65933227539062, 118.488037109375, 18.88739013671875, 20.830856323242188, 61.693443298339844, -39.631439208984375, 40.051544189453125, -1.5662918090820312, 10.34173583984375, 38.172607421875, 11.712547302246094, 23.766571044921875, 33.227691650390625, 17.31097412109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000293.npy"}
{"epoch": 0.6136125654450262, "step": 294, "batch_size": 128, "mean": 21.839691162109375, "std": 37.79030990600586, "min": -67.10418701171875, "p10": -22.81719741821289, "median": 14.774726867675781, "p90": 72.27126388549804, "max": 120.07733154296875, "pos_frac": 0.7109375, "sample": [61.361083984375, 67.27490234375, 66.51052856445312, -0.135772705078125, -9.769790649414062, -18.85381317138672, 2.6412525177001953, 7.598480224609375, 17.94744110107422, 71.542236328125, 35.83123779296875, 63.67596435546875, 64.50421142578125, -13.563827514648438, -9.070556640625, 0.5142536163330078, 70.0937271118164, 90.12452697753906, 71.30892181396484, 10.371551513671875, 19.892364501953125, 62.944549560546875, 34.936431884765625, -22.39543914794922, 4.3413238525390625, 63.96197509765625, 13.47491455078125, -7.3987274169921875, -2.9693603515625, 13.725379943847656, 88.90234375, 5.28436279296875, 7.2466583251953125, 51.73431396484375, -9.605194091796875, 16.763824462890625, -6.190948486328125, -42.974884033203125, 48.517059326171875, 6.285125732421875, 35.81951904296875, 93.82281494140625, -30.1568603515625, -26.104690551757812, 82.61581420898438, 26.05724334716797, 48.64591979980469, -4.97906494140625, -2.1375045776367188, 24.993438720703125, 87.05125427246094, -7.3437042236328125, 83.03081512451172, 77.35065460205078, 25.925060272216797, 71.42030334472656, 76.13616943359375, -19.913726806640625, 10.8392333984375, 10.724456787109375, 36.95269775390625, 17.882232666015625, 18.1175537109375, 66.46969604492188, -60.749359130859375, 50.46922302246094, 63.0733528137207, 34.223907470703125, 15.481307983398438, 2.558837890625, 4.728973388671875, 93.21122741699219, 42.674346923828125, 0.0, 73.97232818603516, 120.07733154296875, -26.346359252929688, 1.7900238037109375, -5.68963623046875, 36.823577880859375, -6.623327255249023, 65.32086181640625, 7.270973205566406, -67.10418701171875, -15.1446533203125, 52.009735107421875, -21.330078125, 3.481964111328125, 50.314361572265625, -40.56159973144531, 20.944015502929688, 12.981605529785156, 7.481651306152344, -52.45722961425781, 6.156364440917969, 36.55351257324219, 15.869903564453125, -0.0023040771484375, 15.248260498046875, -8.4012451171875, 14.301193237304688, 28.72010040283203, 13.071868896484375, 13.455167770385742, 52.99949645996094, 38.06315612792969, -43.402618408203125, -43.60833740234375, 17.54248046875, -0.4174308776855469, -26.30670928955078, 47.23329162597656, 47.63764953613281, 12.66351318359375, 53.184593200683594, 26.73712158203125, 0.8144912719726562, -13.194129943847656, 51.76486587524414, -12.538681030273438, 13.41864013671875, -23.801300048828125, 39.518226623535156, 2.3383140563964844, 78.94979858398438, -61.25572204589844, 57.924598693847656, 81.75720977783203], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000294.npy"}
{"epoch": 0.6157068062827226, "step": 295, "batch_size": 128, "mean": 25.41050910949707, "std": 40.96420669555664, "min": -74.63238525390625, "p10": -20.62531013488769, "median": 18.010581970214844, "p90": 78.53002700805663, "max": 143.95166015625, "pos_frac": 0.75, "sample": [85.27008056640625, 16.98358154296875, -27.674835205078125, -50.236541748046875, -7.2242889404296875, 35.77330017089844, -12.791961669921875, 69.570556640625, -49.550018310546875, 4.3863677978515625, -11.910919189453125, 23.314842224121094, 77.36643981933594, 15.93341064453125, -9.752960205078125, -6.5800933837890625, -63.3304443359375, 50.554412841796875, -48.113067626953125, 22.787948608398438, 35.57359313964844, 109.25762939453125, -14.36553955078125, -74.63238525390625, -7.62994384765625, 53.84502410888672, 12.449722290039062, -14.793609619140625, 37.16033935546875, -0.804107666015625, -2.6119384765625, 18.067642211914062, 40.71917724609375, 58.12603759765625, 57.95379638671875, 10.445037841796875, 64.4012451171875, 7.878082275390625, 40.74981689453125, 40.65557861328125, 106.16465759277344, -10.885440826416016, 43.2684326171875, 51.109344482421875, 24.068115234375, 40.850311279296875, 8.16033935546875, 103.02655029296875, 72.5753173828125, 57.88629150390625, -59.211944580078125, -49.232688903808594, 57.106170654296875, 143.95166015625, 75.670166015625, 111.7020263671875, -13.390792846679688, 52.58458709716797, 38.9161376953125, 8.363170623779297, -5.1510009765625, 0.748260498046875, 13.163528442382812, 2.4308929443359375, 59.8077392578125, 23.85442352294922, 21.90880584716797, 36.3570556640625, 65.90626525878906, 12.621841430664062, 2.8835906982421875, 9.816390991210938, 1.4638042449951172, 73.513671875, -23.798095703125, 28.74951171875, 74.80400085449219, 59.11769104003906, 11.286422729492188, 52.60821533203125, 5.8070831298828125, 25.209083557128906, 13.784553527832031, 22.938560485839844, 8.675064086914062, 26.07696533203125, 0.0, 51.02360153198242, 15.050010681152344, 11.111282348632812, 53.23756408691406, 97.81729125976562, 29.532684326171875, 7.20904541015625, 81.78036499023438, 5.905193328857422, -35.107696533203125, 92.3028564453125, 30.887775421142578, 79.68235778808594, 48.0531005859375, -3.0700225830078125, 10.254241943359375, 14.2081298828125, 17.953521728515625, -48.808258056640625, 66.04129791259766, 5.663852691650391, 1.3472442626953125, 35.71171569824219, -19.265544891357422, 70.01402282714844, 85.57830810546875, 78.03617095947266, 3.1743316650390625, 96.73855590820312, 6.582971572875977, 6.684837341308594, 17.65283203125, -35.1192626953125, -30.7691650390625, -1.2357177734375, 90.177734375, -9.63525390625, 62.5987548828125, 72.51451110839844, 20.0626220703125, -3.52056884765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000295.npy"}
{"epoch": 0.6178010471204188, "step": 296, "batch_size": 128, "mean": 34.602783203125, "std": 41.45797348022461, "min": -88.13333129882812, "p10": -17.592183685302736, "median": 32.10423278808594, "p90": 89.48428344726562, "max": 109.83013916015625, "pos_frac": 0.796875, "sample": [59.85802459716797, 25.874126434326172, 16.042236328125, 59.754112243652344, 67.56576538085938, -11.659095764160156, -0.1440887451171875, 0.23958206176757812, 34.669273376464844, 7.272216796875, 42.274658203125, -18.742218017578125, 22.266632080078125, 27.3294677734375, 12.350311279296875, 74.0123291015625, 67.05998229980469, 49.437835693359375, 63.723899841308594, -3.47357177734375, 77.33412170410156, 37.028167724609375, 4.0595703125, 62.61552429199219, 45.05320739746094, 85.46524047851562, 0.5109157562255859, 13.50787353515625, 87.68350219726562, -29.091888427734375, 10.6455078125, 24.424827575683594, 89.87158203125, 2.3150596618652344, -67.94927978515625, 95.86529541015625, 44.01141357421875, 59.40306091308594, 82.03268432617188, 23.106170654296875, -19.057571411132812, 30.392486572265625, 92.9658203125, 75.1288833618164, 6.803794860839844, -17.5548095703125, 26.7740478515625, -26.862472534179688, 107.1370849609375, 19.358596801757812, 99.22470092773438, 28.36188507080078, 17.915008544921875, 87.4716796875, 89.31829833984375, 102.90267944335938, 8.440261840820312, 101.95377349853516, 10.997711181640625, 44.91631317138672, -12.2279052734375, -17.67938995361328, 36.666778564453125, -27.5306396484375, 46.3636474609375, 48.93977355957031, 25.622344970703125, 88.34405517578125, -57.67585754394531, 60.68121337890625, 77.8173828125, -7.5216522216796875, -29.94842529296875, 64.99578094482422, 46.318565368652344, 24.474929809570312, 103.49893188476562, 105.39016723632812, -5.118938446044922, 55.4718017578125, 18.638671875, 78.83233642578125, 14.885482788085938, -88.13333129882812, 89.99295043945312, 55.17353820800781, 25.870651245117188, -27.857666015625, 76.41787719726562, -7.712432861328125, 64.44929504394531, 12.338058471679688, 46.13983154296875, -31.48553466796875, 48.51312255859375, 0.0, 59.981842041015625, 48.34834289550781, 61.35589599609375, 43.4974365234375, 18.231353759765625, 72.33599853515625, 54.77479553222656, 70.2066650390625, 16.673980712890625, 33.81597900390625, 3.569244384765625, 7.622833251953125, 27.8502197265625, 65.10791015625, 93.2728271484375, 45.834014892578125, 21.44454002380371, -9.98602294921875, -4.4349212646484375, 74.99688720703125, 74.05419921875, -6.8084716796875, 0.1895751953125, 91.14949035644531, 18.032028198242188, 46.34967041015625, 17.0137939453125, 89.10684204101562, 26.928253173828125, 109.83013916015625, -72.7490234375, -1.5779571533203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000296.npy"}
{"epoch": 0.6198952879581152, "step": 297, "batch_size": 128, "mean": 25.743038177490234, "std": 40.197513580322266, "min": -86.46896362304688, "p10": -18.23394775390625, "median": 17.847801208496094, "p90": 85.0648681640625, "max": 118.9622802734375, "pos_frac": 0.7265625, "sample": [100.7332763671875, 5.641971588134766, 73.14630126953125, 85.4210205078125, -8.883453369140625, 2.0479965209960938, 74.1572265625, 42.05220031738281, 85.8935546875, -7.427803039550781, 0.9410552978515625, -2.694183349609375, -18.18450927734375, 5.1881561279296875, 88.92453002929688, -36.66220474243164, -8.966278076171875, 26.06134033203125, 53.93437194824219, 49.5, 9.065040588378906, 25.532363891601562, -10.910377502441406, 26.884521484375, -36.963409423828125, -25.0152587890625, -19.965240478515625, -86.46896362304688, 16.173583984375, -8.707378387451172, -23.270004272460938, 56.62316131591797, 14.213111877441406, 40.215354919433594, -14.463241577148438, 3.295318603515625, 82.6409912109375, 10.72467041015625, 118.9622802734375, 6.432373046875, -0.7213592529296875, 19.600475311279297, 38.602684020996094, -8.216552734375, 8.942840576171875, -16.8328857421875, 2.021209716796875, 14.36639404296875, -58.222015380859375, 63.7701416015625, 2.3314666748046875, -17.782745361328125, 3.235443115234375, -18.34930419921875, 0.02960205078125, 104.11126708984375, 75.45899200439453, 62.71659469604492, -6.798330307006836, -19.244033813476562, 56.97565460205078, 51.074859619140625, -63.55328369140625, 24.699134826660156, 87.02096557617188, 24.19207763671875, 28.016357421875, -27.6563720703125, 89.16168212890625, -8.5294189453125, 26.14996337890625, 57.41864013671875, 87.68114471435547, 78.64433288574219, 99.77154541015625, -39.298095703125, 30.221881866455078, 14.688560485839844, 41.97216796875, -5.7894744873046875, -11.737823486328125, 48.45405578613281, 43.802886962890625, -4.760627746582031, 2.2446422576904297, 17.767440795898438, 21.203155517578125, -4.2531890869140625, 69.50100708007812, 8.016632080078125, 66.71112060546875, 3.9412841796875, 32.66764831542969, 65.0863037109375, 32.51751708984375, 44.57481384277344, 6.736907958984375, 75.96749877929688, -5.023826599121094, 96.6533203125, 67.1387939453125, 58.79478454589844, 33.03926086425781, 74.37339782714844, -8.873390197753906, 45.678985595703125, 4.475982666015625, 17.92816162109375, 52.03166961669922, 45.985198974609375, 100.6883544921875, 66.60678100585938, 2.116943359375, 9.856414794921875, 1.37744140625, -6.6963958740234375, 0.62591552734375, 16.99591064453125, 68.45587158203125, 19.525634765625, 12.486839294433594, 66.62838745117188, 90.26760864257812, -35.302490234375, -6.433013916015625, 33.000762939453125, 71.64625549316406, 84.9122314453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000297.npy"}
{"epoch": 0.6219895287958115, "step": 298, "batch_size": 128, "mean": 29.4274845123291, "std": 42.33559036254883, "min": -69.67202758789062, "p10": -20.8248779296875, "median": 24.978370666503906, "p90": 84.70737152099609, "max": 124.41159057617188, "pos_frac": 0.71875, "sample": [57.79539489746094, 66.375244140625, -14.743316650390625, 20.505813598632812, -21.1630859375, -35.611114501953125, -12.759490966796875, -11.977020263671875, -18.3994140625, 98.30618286132812, 25.042617797851562, -3.7900238037109375, 7.2773895263671875, 20.58929443359375, 56.52653503417969, -21.2703857421875, 53.349395751953125, 31.333240509033203, 20.479415893554688, 46.79625701904297, 80.57693481445312, 66.66311645507812, 34.73382568359375, 123.991455078125, -3.4863128662109375, -25.9036865234375, 70.693603515625, -47.71173095703125, 82.0982666015625, 25.871063232421875, 52.282135009765625, -24.85467529296875, 68.85570526123047, 45.47019958496094, 24.91412353515625, -37.357269287109375, 6.879707336425781, 45.13629150390625, 80.06616973876953, -18.78948974609375, 62.18157958984375, -12.92498779296875, 69.8427734375, -16.705352783203125, 34.918731689453125, 86.65090942382812, 30.4088134765625, 20.335052490234375, -49.31736755371094, 3.0786361694335938, 52.58644104003906, 78.06500244140625, 14.0216064453125, 84.6712646484375, -46.67706298828125, 23.618133544921875, 63.02610778808594, 81.76559448242188, 2.966888427734375, 56.4039306640625, 101.5350341796875, -69.67202758789062, 65.83734130859375, -14.706207275390625, -19.963455200195312, 117.60675048828125, 87.82711791992188, 38.254852294921875, -5.99615478515625, 0.0, 4.5395050048828125, 5.418601989746094, 112.955810546875, 30.862579345703125, 23.914390563964844, -2.0051422119140625, 59.28765869140625, 18.38543701171875, 4.635444641113281, 64.0120849609375, -16.678726196289062, -11.598167419433594, 9.567314147949219, 53.23362731933594, 62.308433532714844, 7.735748291015625, 84.75267028808594, 70.22662353515625, 73.0679931640625, 3.253173828125, 124.41159057617188, 26.739288330078125, -31.26251983642578, 1.42108154296875, 6.314781188964844, 27.38055419921875, 36.49256896972656, 14.64434814453125, -5.869636535644531, -2.06707763671875, -2.3809814453125, 64.87930297851562, 26.045066833496094, 68.0775146484375, 106.2017822265625, 84.68795776367188, -9.957244873046875, 52.42580795288086, 113.75164794921875, 35.122589111328125, 83.181884765625, 29.408447265625, -1.878957748413086, 15.00738525390625, 92.02841186523438, 10.4906005859375, -11.650360107421875, 66.07382202148438, 19.893310546875, 98.41571044921875, 2.1880970001220703, 28.24169921875, 10.772903442382812, 8.302520751953125, -24.38482666015625, -20.679931640625, 39.27136993408203, -27.329879760742188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000298.npy"}
{"epoch": 0.6240837696335079, "step": 299, "batch_size": 128, "mean": 23.52667236328125, "std": 35.88572692871094, "min": -69.87196350097656, "p10": -19.23724479675293, "median": 16.73922348022461, "p90": 74.2975715637207, "max": 107.22787475585938, "pos_frac": 0.75, "sample": [46.71900939941406, 76.35830688476562, 14.193817138671875, 29.451683044433594, 16.599136352539062, 7.131988525390625, 60.01280212402344, -32.0401611328125, 47.75323486328125, -67.20114135742188, 89.981201171875, 35.82658386230469, 18.53325653076172, 60.641571044921875, -24.188720703125, 79.10507202148438, 3.149871826171875, 17.16242218017578, 9.39080810546875, -69.87196350097656, 73.79375457763672, -3.330169677734375, 40.970367431640625, -14.70709228515625, 84.90805053710938, -0.529815673828125, -42.34391784667969, -16.6888427734375, 56.484222412109375, 12.780481338500977, 35.39959716796875, -6.9030914306640625, 80.86338806152344, 3.582244873046875, -41.7821044921875, 23.791484832763672, 64.31951904296875, -5.0391998291015625, -28.893638610839844, 65.723876953125, 9.1632080078125, 89.74833679199219, 16.693626403808594, 7.665666580200195, -7.048088073730469, -2.43603515625, 91.24176025390625, -5.125478744506836, 1.6180572509765625, -28.64068603515625, -36.051025390625, 1.5352020263671875, 3.713094711303711, 15.9666748046875, 57.75128173828125, -23.116424560546875, 84.29217529296875, -3.183074951171875, 89.64703369140625, 21.603675842285156, 38.55123519897461, 9.473388671875, 2.147308349609375, 40.256195068359375, -8.916751861572266, 107.22787475585938, -0.2679443359375, 3.421009063720703, 39.958251953125, 58.16487121582031, 55.46826171875, 6.214468002319336, 43.4647216796875, 8.030181884765625, 63.83329772949219, 54.35171127319336, 6.970680236816406, -19.579578399658203, -0.6127471923828125, 71.82952880859375, 6.7817230224609375, 35.281494140625, 34.8988037109375, 16.784820556640625, 19.89173126220703, -12.771530151367188, 75.47314453125, 40.8084716796875, 32.09730529785156, 32.09016418457031, 7.415679931640625, 42.957672119140625, 70.49488830566406, 4.353851318359375, 5.051685333251953, -5.635101318359375, 27.92400360107422, 2.887420654296875, 61.49567413330078, -20.687705993652344, -26.63818359375, 33.362152099609375, 15.429092407226562, -14.94439697265625, 58.09272766113281, 98.8641357421875, 12.750961303710938, 72.52397918701172, 17.568145751953125, 1.16986083984375, 13.236114501953125, 1.7640323638916016, 64.20661163330078, -19.090530395507812, 4.106689453125, 7.829681396484375, -3.647552490234375, 23.46044921875, 27.207992553710938, 34.77349853515625, -9.385177612304688, 35.230255126953125, 76.14036560058594, 70.38055419921875, 38.0738525390625, 63.450408935546875, 22.2159423828125, 49.555389404296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000299.npy"}
{"epoch": 0.6261780104712041, "step": 300, "batch_size": 128, "mean": 24.343521118164062, "std": 39.78641128540039, "min": -62.9483642578125, "p10": -24.193966674804688, "median": 28.053016662597656, "p90": 75.25386047363281, "max": 106.6378173828125, "pos_frac": 0.7109375, "sample": [42.6318359375, 50.54817199707031, 16.132671356201172, -48.02006530761719, 25.347885131835938, -13.520172119140625, 28.8831787109375, 47.00572204589844, 21.268033981323242, -24.6712646484375, 69.1639404296875, 71.55453491210938, 71.53359985351562, 3.102954864501953, 5.196846008300781, 75.86322021484375, -42.24908447265625, 5.17803955078125, 12.183212280273438, -5.644622802734375, 81.92251586914062, 34.631591796875, 33.83599853515625, 43.65155029296875, 43.06184387207031, 13.63861083984375, 33.093505859375, 2.6190185546875, 10.6295166015625, -19.99567413330078, 57.72637939453125, 51.05609130859375, 78.33494567871094, 40.49334716796875, 67.94244384765625, -23.593765258789062, -50.44749450683594, 22.807464599609375, 3.6896629333496094, 106.6378173828125, 44.77525329589844, -15.62969970703125, -23.989410400390625, 1.84539794921875, 56.044830322265625, 65.30474853515625, -13.273422241210938, 70.66461944580078, 63.461181640625, 97.51638793945312, 27.560226440429688, -46.73857116699219, -58.19148254394531, 28.545806884765625, 33.69717025756836, -6.714529037475586, 36.474578857421875, 37.799468994140625, 60.62445068359375, -9.316459655761719, 71.58557891845703, 66.41485595703125, -36.32000732421875, 91.73057556152344, 35.50396728515625, 4.38970947265625, 2.9635009765625, 42.128997802734375, 46.42149353027344, 74.99270629882812, 43.43011474609375, 46.3450927734375, 0.18079185485839844, 11.206878662109375, -15.295822143554688, -55.17919921875, 94.82652282714844, 50.02374267578125, -33.45612335205078, 76.475341796875, 71.59268188476562, 10.957141876220703, 41.062652587890625, 4.94677734375, 43.630767822265625, 16.66754150390625, -12.33856201171875, 0.0, -0.72259521484375, 19.695587158203125, 90.37548828125, -12.779052734375, -62.9483642578125, 31.023101806640625, -38.708740234375, 74.46585083007812, 32.310577392578125, -3.3726806640625, 67.09002685546875, 52.6236572265625, 15.67535400390625, 76.68609619140625, 4.328758239746094, 51.755062103271484, -18.552261352539062, 59.937042236328125, 59.20135498046875, -55.56060791015625, -27.1126708984375, -15.468414306640625, 55.890960693359375, 2.6689815521240234, 21.354461669921875, -19.14996337890625, 61.0162353515625, 65.0789566040039, 30.78680419921875, 32.11564636230469, -15.639633178710938, -4.899658203125, 101.305908203125, -11.968597412109375, -23.08050537109375, -18.153579711914062, 77.16912841796875, 77.16336059570312, -7.965904235839844, 3.76318359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000300.npy"}
{"epoch": 0.6282722513089005, "step": 301, "batch_size": 128, "mean": 24.26255226135254, "std": 40.96747589111328, "min": -94.12908935546875, "p10": -29.933235168457028, "median": 21.794133186340332, "p90": 77.03924179077146, "max": 139.26708984375, "pos_frac": 0.734375, "sample": [14.670272827148438, 96.06134033203125, 71.57421875, 37.07684326171875, 18.39196014404297, 65.19183349609375, 42.30717086791992, -32.26670837402344, 40.18251037597656, 6.767555236816406, 4.561958312988281, -0.45452880859375, 62.93450927734375, 14.452545166015625, 5.761474609375, -1.6200027465820312, -44.26893615722656, -32.556976318359375, 71.53903198242188, 99.07470703125, 17.544219970703125, 17.997695922851562, 40.36962890625, 22.01702880859375, 21.4261474609375, 15.15667724609375, 0.154510498046875, 40.7596435546875, 11.730880737304688, 39.56842041015625, 42.69976806640625, -3.29327392578125, 0.0, 97.4942626953125, 59.451202392578125, 47.157615661621094, -10.453643798828125, 29.29949951171875, -25.663299560546875, 61.84895324707031, -21.951580047607422, 68.76764678955078, 91.30870056152344, 25.0618896484375, -89.89422607421875, 3.572174072265625, -37.26898193359375, 60.729522705078125, -23.50732421875, 23.3587646484375, 11.455146789550781, -11.83489990234375, -29.2467041015625, 64.05928039550781, 95.24969482421875, 20.599403381347656, 22.482177734375, 84.3175048828125, 2.026336669921875, 139.26708984375, 15.647232055664062, 60.99957275390625, 21.469512939453125, 32.587425231933594, -34.010223388671875, -47.79219436645508, 10.29290771484375, -94.12908935546875, -45.880218505859375, 44.779693603515625, 32.39378356933594, 29.69305419921875, -8.9122314453125, 8.37893295288086, 36.205322265625, 20.481170654296875, 83.11082458496094, 6.0831298828125, 82.97088623046875, -40.95379638671875, -14.2625732421875, -15.15960693359375, 99.89128112792969, 58.126251220703125, 69.25762939453125, 24.0426025390625, -5.946264266967773, -31.535140991210938, 63.942779541015625, -39.50440979003906, 72.93867492675781, 20.4326171875, -8.994384765625, 6.2112579345703125, 63.1488037109375, -13.20587158203125, 29.591384887695312, 30.8021240234375, 41.42671203613281, 65.8275146484375, 52.771026611328125, 74.49710845947266, 90.11538696289062, 0.19024658203125, 22.67638397216797, 73.85319519042969, 34.85491943359375, -13.328582763671875, 93.95086669921875, -9.212745666503906, 19.742477416992188, 25.154029846191406, -0.51708984375, 28.55560302734375, 2.4536209106445312, 52.32067108154297, -19.4180908203125, 21.571237564086914, 1.262847900390625, 70.13894653320312, 87.16151428222656, -37.734619140625, -16.6041259765625, 35.276100158691406, 24.575927734375, 33.88385009765625, 54.04185104370117, 9.7288818359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000301.npy"}
{"epoch": 0.6303664921465969, "step": 302, "batch_size": 128, "mean": 28.31484031677246, "std": 40.06153869628906, "min": -82.01641845703125, "p10": -15.274374389648433, "median": 28.05181121826172, "p90": 81.17417602539062, "max": 109.35302734375, "pos_frac": 0.765625, "sample": [88.49739074707031, 42.505706787109375, 52.659358978271484, 60.962677001953125, 19.845016479492188, 49.662498474121094, 0.0, 6.2703399658203125, -3.69110107421875, -22.027328491210938, -71.79940795898438, 58.172698974609375, 48.160362243652344, 81.88336181640625, -8.540802001953125, 3.1707305908203125, 27.21990966796875, 42.01611328125, -13.133346557617188, -18.018020629882812, 78.84384155273438, 62.0086669921875, 98.28338623046875, -10.57537841796875, 65.3572998046875, 77.67855834960938, 109.35302734375, 6.30181884765625, -56.20941162109375, 56.27771759033203, 9.660873413085938, 62.457977294921875, 65.88187408447266, 53.573333740234375, 32.62083053588867, 52.05326843261719, 101.62503051757812, 35.54466247558594, 36.419647216796875, 35.38641357421875, 40.094696044921875, 1.2152328491210938, -76.43853759765625, 0.288299560546875, 40.10406494140625, 38.74653625488281, 95.75225830078125, 35.0386962890625, 96.86227416992188, 71.31060791015625, -5.000030517578125, 66.07987976074219, 5.376899719238281, 1.4039459228515625, 13.794685363769531, -25.75677490234375, 43.895538330078125, 26.59747314453125, 53.583251953125, 28.208847045898438, 48.46533203125, 29.019432067871094, 29.66802978515625, 10.578201293945312, 30.90350341796875, -2.754058837890625, -7.6822509765625, 7.653127670288086, -4.871406555175781, -39.90052032470703, 66.6572036743164, 87.98829650878906, -20.05718994140625, -37.70416259765625, 76.90692138671875, 43.9993896484375, 2.0322952270507812, 50.4769287109375, 71.96815490722656, 27.894775390625, 18.696502685546875, 20.4501953125, 72.94082641601562, -52.1214599609375, 82.18215942382812, -3.84039306640625, -82.01641845703125, 12.949661254882812, -14.098526000976562, -34.88829803466797, 9.919282913208008, -36.78924560546875, -1.07354736328125, 16.2674560546875, 12.323875427246094, 36.827789306640625, 42.19160461425781, 81.8851318359375, 80.8702392578125, -12.474342346191406, 6.603302001953125, 12.248382568359375, 22.65521240234375, -9.833953857421875, 56.13092803955078, 0.7854633331298828, 17.3055419921875, 88.48690795898438, 23.68436050415039, 63.77217102050781, 3.7454681396484375, 29.02501678466797, 35.78253173828125, -10.293930053710938, 59.432891845703125, 20.87279510498047, 104.55377197265625, 3.285531997680664, 67.56771850585938, 9.72381591796875, 18.93304443359375, 106.7843017578125, 26.46282958984375, 70.95225524902344, 51.36430358886719, -5.741180419921875, 70.32160949707031, -9.273422241210938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000302.npy"}
{"epoch": 0.6324607329842932, "step": 303, "batch_size": 128, "mean": 26.003910064697266, "std": 41.425315856933594, "min": -91.421875, "p10": -23.764736938476563, "median": 28.3505859375, "p90": 74.42593994140626, "max": 123.980712890625, "pos_frac": 0.75, "sample": [12.0101318359375, 18.776405334472656, -67.59103393554688, -59.906219482421875, 60.7777099609375, -9.312545776367188, 51.18231201171875, 47.624755859375, 67.734375, 43.87824249267578, -22.31353759765625, 61.87794494628906, 86.0408935546875, 69.9185791015625, 37.026641845703125, -35.440704345703125, -0.726531982421875, 59.02829360961914, -7.33665657043457, 86.52032470703125, 42.921722412109375, 25.7386474609375, 60.39141082763672, 25.316909790039062, 23.80072021484375, -23.61346435546875, 36.00126647949219, -8.629180908203125, 92.41375732421875, 5.189262390136719, 72.2314453125, 72.3089828491211, -20.842697143554688, 77.96675109863281, -3.8066864013671875, 9.493560791015625, -2.8583221435546875, 40.35137939453125, 0.0, 16.005706787109375, -38.05549621582031, 13.899505615234375, -8.383186340332031, 1.392120361328125, 6.084892272949219, 2.38226318359375, 35.11094665527344, 28.197357177734375, 45.77520751953125, 63.21330261230469, 73.53561401367188, -52.38311767578125, 69.73794555664062, -72.61456298828125, 80.35043334960938, 74.70095825195312, 75.16403198242188, 35.02613830566406, 19.3619384765625, 63.54644775390625, 1.6466255187988281, -2.8659210205078125, 11.044418334960938, -8.9532470703125, -48.26837158203125, 3.98907470703125, 52.5152587890625, 38.62554931640625, 51.71733093261719, 24.274730682373047, 91.60104370117188, -26.064926147460938, 21.594024658203125, 1.45465087890625, 69.93600463867188, 65.41535949707031, 0.7120819091796875, 44.153045654296875, 75.57051849365234, 112.9237060546875, 59.32196807861328, -74.785888671875, 46.47565460205078, -56.82754135131836, 0.33876800537109375, 4.91943359375, 68.70266723632812, 23.266876220703125, 25.44512939453125, 60.84814453125, 4.607696533203125, -14.7305908203125, 81.39810180664062, -1.44940185546875, 60.1441650390625, 1.7473297119140625, -0.8480072021484375, 34.605796813964844, -20.930450439453125, 31.608856201171875, -24.117706298828125, -91.421875, 39.820068359375, -0.16681671142578125, 3.1432037353515625, 51.297935485839844, 74.30807495117188, 44.42884826660156, 73.72804260253906, 50.214088439941406, 32.092041015625, 45.2706298828125, 28.63355827331543, 70.93768310546875, -3.8662872314453125, 123.980712890625, 31.843063354492188, -47.420135498046875, 60.47613525390625, 8.616287231445312, 94.614501953125, 13.4429931640625, 73.6549072265625, 6.52752685546875, 28.503814697265625, 6.223297119140625, 32.820953369140625, 29.841690063476562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000303.npy"}
{"epoch": 0.6345549738219896, "step": 304, "batch_size": 128, "mean": 23.96483039855957, "std": 38.331756591796875, "min": -78.21833801269531, "p10": -21.556729125976563, "median": 17.254928588867188, "p90": 80.10788269042969, "max": 109.3790283203125, "pos_frac": 0.734375, "sample": [0.0, 25.23638916015625, -2.814849853515625, 22.451873779296875, -11.958404541015625, 0.49957275390625, -78.21833801269531, 49.23883056640625, 50.71588134765625, -2.70166015625, 42.942962646484375, 67.6937255859375, 33.70013427734375, 39.026702880859375, -41.343238830566406, 30.40582275390625, 9.05401611328125, 78.47296905517578, 29.416778564453125, 11.71258544921875, 10.185470581054688, 18.099411010742188, 8.401260375976562, -2.974700927734375, 63.57490539550781, 69.5614013671875, 65.14230346679688, 12.274497985839844, 13.841842651367188, 4.315032958984375, -46.526885986328125, -3.3968887329101562, 14.2879638671875, 0.220306396484375, 14.990524291992188, 10.948272705078125, 23.503692626953125, 2.17333984375, 58.324615478515625, 41.74591064453125, 3.87799072265625, 48.95896911621094, 60.54054260253906, -26.235198974609375, 94.75830078125, 23.024337768554688, -1.1069717407226562, 43.789520263671875, -19.9910888671875, 27.740234375, 29.179893493652344, -17.7032470703125, 16.410446166992188, -35.635589599609375, 0.0, 81.45294189453125, -41.07475280761719, 28.115737915039062, 3.7570648193359375, -7.066497802734375, -22.23419189453125, 15.110260009765625, -35.822235107421875, 60.597412109375, 4.979736328125, 14.171852111816406, -21.266387939453125, 34.1058349609375, 109.3790283203125, 63.320343017578125, -60.447784423828125, 23.8763427734375, 21.693527221679688, -23.34576416015625, -34.206573486328125, 85.93637084960938, 83.83464050292969, -4.989570617675781, 82.92660522460938, -38.745880126953125, 72.42431640625, 64.05526733398438, -5.870201110839844, -4.965442657470703, 85.289794921875, 16.395660400390625, 82.38906860351562, -9.767303466796875, 70.41015625, -63.4332275390625, 7.8321533203125, 3.270355224609375, 21.4600830078125, 86.23501586914062, 89.02790832519531, 12.014551162719727, -4.969879150390625, 62.769500732421875, 79.66917419433594, 75.15013122558594, 24.2850341796875, 22.19580078125, 27.165313720703125, 3.255035400390625, 29.734657287597656, 79.91043090820312, 15.642105102539062, 94.84619140625, 60.26335144042969, 45.884857177734375, 54.542449951171875, 35.15130615234375, -14.657562255859375, -10.01611328125, 39.82366943359375, 13.771484375, 0.43994140625, 7.4757843017578125, 53.82038879394531, 18.423355102539062, 15.92974853515625, 61.7235107421875, -3.35504150390625, 75.58270263671875, -16.360015869140625, 80.568603515625, 15.72747802734375, 92.44635009765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000304.npy"}
{"epoch": 0.6366492146596858, "step": 305, "batch_size": 128, "mean": 34.53980255126953, "std": 40.8027229309082, "min": -102.50033569335938, "p10": -4.302338027954098, "median": 32.05286407470703, "p90": 86.65567626953126, "max": 125.39312744140625, "pos_frac": 0.84375, "sample": [5.8272705078125, 23.396953582763672, -2.8124656677246094, 120.298095703125, 19.08660888671875, 4.5269775390625, 54.34276580810547, -19.100875854492188, 60.61529541015625, 22.903846740722656, 33.756649017333984, 21.071563720703125, 5.5901336669921875, 30.444412231445312, 23.132217407226562, -36.801063537597656, -33.22674560546875, 19.4901123046875, 36.67230224609375, 0.6991157531738281, 39.06622314453125, 75.70916748046875, 5.082366943359375, 32.85777282714844, 7.0775299072265625, 16.136138916015625, -56.739654541015625, -3.221343994140625, 20.654281616210938, 57.14045715332031, 16.241485595703125, 61.57537841796875, 54.22395324707031, 32.74778747558594, 88.06280517578125, 36.07237243652344, 0.6489486694335938, -2.4143600463867188, 72.19818878173828, 1.6645965576171875, -71.39933776855469, 43.48020935058594, -3.4488525390625, 51.68492889404297, 87.13031005859375, 70.57249450683594, 38.64617919921875, 78.06321716308594, 76.57272338867188, 24.910003662109375, -2.8937549591064453, -59.15782165527344, 97.83465576171875, 91.30270385742188, 21.335662841796875, 5.897735595703125, 125.39312744140625, 31.671539306640625, 111.46295166015625, 57.85418701171875, 24.066543579101562, 8.244613647460938, 0.7716064453125, 86.35379028320312, 33.811004638671875, 56.37107849121094, 86.56069946289062, 58.2193489074707, 68.54144287109375, 66.67645263671875, -9.505058288574219, 62.29424285888672, -6.293804168701172, 29.042648315429688, 71.22967529296875, -102.50033569335938, 11.505523681640625, 41.88145446777344, 3.58795166015625, 69.69015502929688, 39.59419250488281, 83.8385009765625, 96.96835327148438, 6.273124694824219, 31.7862548828125, 58.7705078125, -20.721893310546875, 98.96267700195312, 1.6880950927734375, 3.1007232666015625, 86.87728881835938, 56.1475830078125, 37.186920166015625, 32.31947326660156, 24.377349853515625, 61.89239501953125, 32.9254150390625, 81.6273193359375, 100.65191650390625, 37.50012969970703, 29.984344482421875, -1.39239501953125, 65.86489868164062, 61.9300537109375, 108.47064208984375, 14.235549926757812, 27.38409423828125, 22.51300048828125, 26.64875030517578, 23.44891357421875, 66.77645874023438, 0.5434036254882812, 48.19683837890625, 3.42120361328125, 11.157440185546875, -39.52373504638672, 80.16519165039062, 3.8719406127929688, 11.468704223632812, -17.335159301757812, 85.05984497070312, 36.21539306640625, 117.17855834960938, 63.63604736328125, 76.14129638671875, -16.344741821289062, -2.51446533203125, 79.96682739257812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000305.npy"}
{"epoch": 0.6387434554973822, "step": 306, "batch_size": 128, "mean": 28.746726989746094, "std": 40.56861114501953, "min": -81.99618530273438, "p10": -15.112609481811521, "median": 27.561687469482422, "p90": 82.59475860595703, "max": 110.29067993164062, "pos_frac": 0.765625, "sample": [5.9258575439453125, -45.984771728515625, -40.219337463378906, 103.85855102539062, -55.327301025390625, 98.12814331054688, -1.5842914581298828, 15.04990005493164, -5.68035888671875, 79.73095703125, 4.504671096801758, 15.5782470703125, 77.20164489746094, 28.50164794921875, 19.818191528320312, 8.23590087890625, -23.9024658203125, 64.76541900634766, -3.0241241455078125, 16.79241943359375, 37.038238525390625, 21.468414306640625, -35.90045166015625, 37.08306884765625, -81.99618530273438, 106.7099609375, 99.94537353515625, 73.75247192382812, 62.677215576171875, 0.14388275146484375, 47.477508544921875, 43.648529052734375, 15.36865234375, 58.18194580078125, 88.21733093261719, 58.89082336425781, 95.66412353515625, 68.26800537109375, 35.793212890625, 23.3017578125, 4.5490570068359375, 66.51242065429688, -2.45037841796875, 9.860671997070312, 32.231597900390625, 8.776351928710938, 19.732574462890625, 47.06158447265625, 44.094390869140625, -50.12457275390625, -9.353271484375, 71.42573547363281, -14.67166519165039, 104.97442626953125, 38.98919677734375, -76.12197875976562, 18.764625549316406, 55.5355224609375, -11.30914306640625, 87.0936279296875, -2.5670166015625, 9.34735107421875, 46.64501953125, 34.022422790527344, 70.02411651611328, 2.5382843017578125, 9.547042846679688, 2.3372669219970703, 54.53631591796875, -22.076461791992188, 38.92315673828125, -29.490020751953125, 63.15864562988281, 12.90966796875, 109.1234130859375, 0.5871639251708984, -5.2659912109375, 19.217178344726562, 14.82666015625, 40.199066162109375, 15.67181396484375, 36.4989013671875, 61.147796630859375, 36.388031005859375, 83.1424560546875, -16.236541748046875, -12.6739501953125, 44.0240478515625, 40.538970947265625, 110.29067993164062, 51.02508544921875, 31.03985595703125, -11.993942260742188, -16.1414794921875, 1.80242919921875, 33.912628173828125, 66.38693237304688, -9.9180908203125, 60.682891845703125, -13.33027458190918, 77.6692123413086, 48.236846923828125, 71.91523742675781, 70.88827514648438, 40.558799743652344, 26.621726989746094, 30.820877075195312, 8.055572509765625, 7.8692474365234375, -3.06707763671875, -6.490039825439453, 7.351318359375, 72.7159423828125, 67.21383666992188, 9.890380859375, 105.8115234375, 10.677389144897461, -1.5279083251953125, -6.50775146484375, 4.282989501953125, 53.189178466796875, 63.122352600097656, 0.6725311279296875, -44.78387451171875, 30.30577850341797, 31.979461669921875, 82.36003112792969, 95.27401733398438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000306.npy"}
{"epoch": 0.6408376963350786, "step": 307, "batch_size": 128, "mean": 31.36455535888672, "std": 40.60696792602539, "min": -99.71052551269531, "p10": -13.04530334472656, "median": 30.930435180664062, "p90": 80.59758071899414, "max": 121.3350830078125, "pos_frac": 0.7734375, "sample": [77.607177734375, 73.05531311035156, 3.5133819580078125, 81.74655151367188, -41.00213623046875, -4.796119689941406, 50.637062072753906, 48.628814697265625, 38.27996826171875, 73.04006958007812, -99.71052551269531, 0.9243030548095703, 32.06825256347656, 26.474403381347656, -1.3436279296875, 57.353050231933594, 66.5711669921875, 59.84996032714844, -0.51800537109375, 12.831268310546875, 40.107452392578125, -24.26934814453125, 52.93153381347656, 8.304418563842773, 96.989013671875, -15.694610595703125, 31.93817138671875, 4.35980224609375, 65.91468811035156, 103.34222412109375, 56.43214416503906, 16.055068969726562, 16.2135009765625, 36.268707275390625, 30.131301879882812, -7.8737030029296875, 70.07354736328125, 53.504241943359375, 9.6417236328125, 3.0725479125976562, 15.628570556640625, 18.049205780029297, 69.21148681640625, -8.73394775390625, 43.48291015625, 49.092620849609375, 80.39385986328125, -12.45849609375, -25.390350341796875, 20.29931640625, 40.40052795410156, 3.067270278930664, 83.4190673828125, 6.218141555786133, 18.189529418945312, 69.21543884277344, 63.81268310546875, 87.75865173339844, 11.467391967773438, 4.140037536621094, 18.047607421875, 47.775718688964844, -46.683807373046875, 65.5975341796875, -2.0750732421875, -4.4851226806640625, 68.48550415039062, 46.390586853027344, -4.562782287597656, 6.305706024169922, -2.7574005126953125, 81.07292938232422, -23.51068115234375, -14.414520263671875, 23.8642578125, -8.1939697265625, 15.054534912109375, 74.17030334472656, 61.8155517578125, 14.60723876953125, 71.82395935058594, -11.5360107421875, 38.347843170166016, 98.35610961914062, 96.14443969726562, 72.83062744140625, -7.674480438232422, 64.9430160522461, 19.136249542236328, 63.770782470703125, 16.83063507080078, -25.767745971679688, 43.26782989501953, 104.4105224609375, 51.63938903808594, 6.045074462890625, 65.95034790039062, -74.40036010742188, -17.93877410888672, 67.515625, 1.436004638671875, 107.75198364257812, 23.359954833984375, 81.447021484375, -95.85598754882812, 13.1658935546875, -0.624481201171875, 61.42347717285156, 20.4638671875, 47.746063232421875, -18.52105712890625, 25.7677001953125, 121.3350830078125, 51.967201232910156, 11.494491577148438, 52.306549072265625, -1.885284423828125, -10.95623779296875, 31.729568481445312, 75.64263916015625, 55.279815673828125, 7.552120208740234, 63.818336486816406, 83.47886657714844, 63.449676513671875, 9.531841278076172, 75.41888427734375, 56.82339096069336], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000307.npy"}
{"epoch": 0.6429319371727749, "step": 308, "batch_size": 128, "mean": 23.126930236816406, "std": 39.312503814697266, "min": -81.66867065429688, "p10": -23.145021820068354, "median": 14.904281616210938, "p90": 73.25659637451172, "max": 110.13623046875, "pos_frac": 0.765625, "sample": [83.65425109863281, 33.953468322753906, -26.92193603515625, 77.76295471191406, 59.000274658203125, -41.76524353027344, -35.40228271484375, 34.03692626953125, 6.5555419921875, 21.369205474853516, 8.624267578125, 63.4510498046875, -3.3942031860351562, -58.544158935546875, -1.041748046875, 4.368194580078125, 70.46878051757812, -0.9044647216796875, 67.68960571289062, 56.60540771484375, 65.20630645751953, 50.716583251953125, 15.595596313476562, -29.26903533935547, 6.82659912109375, 72.64662170410156, -80.17694091796875, -21.526344299316406, 14.600677490234375, 83.01266479492188, 42.43394470214844, 5.269073486328125, 7.89312744140625, 35.87358093261719, -10.51531982421875, 21.91259765625, 60.379241943359375, 7.7478790283203125, 58.10356903076172, 14.0545654296875, 1.1940841674804688, 22.254371643066406, 110.13623046875, -2.22760009765625, -15.04901123046875, -3.5333251953125, 1.35791015625, 87.15457153320312, -1.8460235595703125, 66.82670593261719, 9.573524475097656, 6.172019958496094, 3.1558380126953125, 50.00428771972656, 1.3100814819335938, 25.874725341796875, 72.34967041015625, 21.12030792236328, -63.880340576171875, 74.67987060546875, 26.659347534179688, 33.284523010253906, -4.8643798828125, -8.603118896484375, 106.56353759765625, 66.20338439941406, -3.9197254180908203, -55.01695251464844, 14.031448364257812, 1.9293670654296875, 56.65092468261719, -43.28449630737305, 68.9712142944336, 40.96832275390625, 8.240875244140625, 14.04193115234375, 80.80899047851562, 105.29693603515625, 88.55511474609375, -0.336456298828125, 55.793434143066406, 11.05377197265625, 4.103996276855469, 5.627777099609375, 52.516357421875, -81.66867065429688, 11.425392150878906, 10.835968017578125, -12.478057861328125, 50.8246955871582, 15.2078857421875, 5.96783447265625, 35.76708984375, 8.458358764648438, 65.74601745605469, 57.533546447753906, 76.32894897460938, 76.26923370361328, 24.878189086914062, 42.23211669921875, 27.729583740234375, 56.9500732421875, 107.27737426757812, -16.243606567382812, 18.30780792236328, 10.05645751953125, 10.89068603515625, -16.025012969970703, 61.23243713378906, 59.719573974609375, -42.273040771484375, 12.5552978515625, -18.596832275390625, 1.9444046020507812, 2.672454833984375, 55.872802734375, 5.7971343994140625, 19.321014404296875, 10.64337158203125, 4.286895751953125, -33.842529296875, 51.63676452636719, 38.202911376953125, 34.503997802734375, -38.50751495361328, 25.8011474609375, 70.81637573242188, 19.9334716796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000308.npy"}
{"epoch": 0.6450261780104712, "step": 309, "batch_size": 128, "mean": 30.45040512084961, "std": 41.644344329833984, "min": -62.84808349609375, "p10": -20.67374267578125, "median": 24.200817108154297, "p90": 84.74107360839844, "max": 121.95730590820312, "pos_frac": 0.75, "sample": [16.586273193359375, 72.60147094726562, 68.32608032226562, 16.297943115234375, 1.2425689697265625, -30.110382080078125, -2.43505859375, -49.49772644042969, 75.48236083984375, 35.305511474609375, 100.66519165039062, 5.666709899902344, -8.2457275390625, 5.6740264892578125, -15.795272827148438, 45.858558654785156, 10.72271728515625, 17.374855041503906, -8.055343627929688, 23.77374267578125, -7.3392181396484375, 59.300811767578125, -0.280303955078125, -43.60072326660156, 9.2952880859375, 48.8975830078125, -1.416717529296875, 94.61520385742188, 43.984375, 23.572052001953125, 3.5775604248046875, 83.23651123046875, 61.41615295410156, 5.0652313232421875, 0.964080810546875, -5.9443817138671875, 48.10539245605469, -15.619781494140625, 121.95730590820312, -62.69561767578125, 77.23110961914062, -21.765228271484375, -4.509227752685547, 17.518081665039062, 66.10832214355469, -24.23084259033203, 2.5048828125, -37.028594970703125, -39.364532470703125, 27.17864990234375, 73.68856811523438, -9.641597747802734, 14.45635986328125, -62.84808349609375, 64.36029052734375, 73.69760131835938, 61.30863952636719, 22.4981689453125, 29.80804443359375, 13.964599609375, 91.03662109375, 3.102630615234375, 31.54706573486328, 95.72998046875, -7.716156005859375, 85.8800048828125, 7.849822998046875, 66.36817932128906, 21.60663604736328, 26.82470703125, -0.856353759765625, 21.877349853515625, 13.251914978027344, 106.90789794921875, 4.290138244628906, 73.93760681152344, 74.96405029296875, 50.90473937988281, 9.921226501464844, 94.46990966796875, 58.14421081542969, 70.31092071533203, 3.2860183715820312, 9.968551635742188, 68.5191879272461, 83.21002197265625, -3.2021026611328125, 24.627891540527344, 19.76318359375, -31.700477600097656, 53.179901123046875, -37.7825927734375, 35.52677917480469, -12.984819412231445, 79.47975158691406, 40.864776611328125, 15.826431274414062, 71.08343505859375, 48.39642333984375, 8.820175170898438, -32.98046875, 46.24076843261719, 71.3270263671875, 96.608154296875, 84.25296020507812, -2.55230712890625, 115.21087646484375, 20.7205810546875, 18.412551879882812, -57.604736328125, 86.46392822265625, 47.71014404296875, 37.000335693359375, 77.57113647460938, 43.84063720703125, 36.4871826171875, 38.34930419921875, 98.93695068359375, -20.205963134765625, 78.00631713867188, 95.0966796875, 65.92298126220703, 56.41911315917969, -18.32318115234375, 58.4779052734375, 30.744964599609375, -3.0106544494628906, 57.828521728515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000309.npy"}
{"epoch": 0.6471204188481675, "step": 310, "batch_size": 128, "mean": 26.258644104003906, "std": 38.14013671875, "min": -55.572509765625, "p10": -20.290054321289062, "median": 22.457504272460938, "p90": 75.77562255859374, "max": 117.653564453125, "pos_frac": 0.7265625, "sample": [109.85906982421875, 9.911468505859375, 2.6915130615234375, 112.36065673828125, 31.49481201171875, 25.52508544921875, 11.779998779296875, 8.826240539550781, 68.93879699707031, 26.531463623046875, 61.19548034667969, 106.71630859375, 41.571895599365234, 60.911163330078125, 17.9818115234375, 25.075653076171875, 11.201873779296875, -0.5937252044677734, 26.469970703125, -26.085098266601562, -20.20660400390625, 43.675628662109375, 14.182754516601562, 35.442718505859375, 27.78258514404297, -44.85858154296875, 43.13761901855469, -30.808349609375, 65.31538391113281, -7.71282958984375, -6.777553558349609, -3.9857864379882812, -11.83953857421875, 73.51953125, -0.034942626953125, -2.708404541015625, 53.906005859375, 102.823974609375, -55.572509765625, -37.64007568359375, 60.29500198364258, -2.6230926513671875, 6.952484130859375, 17.623138427734375, 37.77674865722656, -1.861328125, 56.67633056640625, 27.230819702148438, 14.947647094726562, -4.0473175048828125, -5.118133544921875, 25.329666137695312, 83.47894287109375, 59.440277099609375, 37.82061767578125, 71.69110107421875, 99.87506103515625, 37.14878845214844, -2.64501953125, -46.9847412109375, -18.680252075195312, 86.6153564453125, 41.848594665527344, 22.97491455078125, 4.019378662109375, -42.22154998779297, 8.218338012695312, 21.940093994140625, 75.02166748046875, 3.623706817626953, -9.975532531738281, 14.690643310546875, 2.7360687255859375, 26.078033447265625, 48.807159423828125, 50.107025146484375, 62.739410400390625, 99.4915771484375, 45.47686767578125, -1.0254974365234375, -25.335002899169922, 42.88545227050781, 56.497772216796875, 13.863052368164062, -12.669830322265625, 67.22087097167969, 54.56523895263672, 4.3411102294921875, 63.486656188964844, -1.16583251953125, 4.974151611328125, -0.34375, 47.53216552734375, -37.6767578125, 40.98881912231445, 0.14810943603515625, 6.4667816162109375, 51.413848876953125, 38.35789489746094, 15.047195434570312, 32.926368713378906, 85.82272338867188, 7.06170654296875, -20.484771728515625, 73.4285888671875, -3.312652587890625, 1.77801513671875, -32.881500244140625, -35.695037841796875, 45.73710632324219, -15.136772155761719, 80.27896118164062, -11.786453247070312, 50.1810302734375, 14.027679443359375, -32.70880126953125, 12.7044677734375, 15.243064880371094, 8.316169738769531, 46.38356018066406, 35.68212890625, 5.239982604980469, 73.6123046875, 78.72119140625, 117.653564453125, 66.80117797851562, 77.53485107421875, 67.88142395019531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000310.npy"}
{"epoch": 0.6492146596858639, "step": 311, "batch_size": 128, "mean": 27.831504821777344, "std": 43.77439498901367, "min": -106.804931640625, "p10": -19.231268692016602, "median": 25.931198120117188, "p90": 78.90520706176757, "max": 136.69723510742188, "pos_frac": 0.7265625, "sample": [8.950149536132812, -4.9320831298828125, 69.76274108886719, 58.52326202392578, 77.54318237304688, 7.726774215698242, 12.584396362304688, -1.70166015625, 78.1929931640625, 61.53363037109375, 73.26162719726562, 4.918018341064453, -14.534078598022461, -17.874923706054688, 25.788589477539062, 4.89422607421875, 68.12533569335938, 20.97553825378418, -26.340578079223633, 21.824371337890625, 111.12057495117188, 74.89046478271484, 7.6237335205078125, 59.45466613769531, -6.133693695068359, 12.58172607421875, 85.90689086914062, 67.92459106445312, 6.931148529052734, -2.974884033203125, -42.14728546142578, -106.804931640625, -78.20211791992188, 58.43292236328125, -7.847564697265625, 85.00640869140625, -60.0242919921875, -18.97357177734375, 69.86868286132812, 38.091217041015625, -41.64488220214844, 34.2772331237793, 26.073806762695312, 9.76531982421875, -11.89832878112793, 36.58544921875, -13.693756103515625, 34.635498046875, -50.01017761230469, 14.081489562988281, 43.828392028808594, 20.50543212890625, -4.797882080078125, 23.554092407226562, 58.065093994140625, -6.6524505615234375, 34.26617431640625, 2.39117431640625, -16.52850341796875, 72.43927001953125, 91.11920166015625, 78.19754791259766, 77.25344848632812, 61.048072814941406, 69.2196044921875, -28.514358520507812, 26.077239990234375, 2.1643218994140625, 22.29938507080078, 38.208518981933594, -52.73981475830078, -9.033130645751953, 82.02059936523438, 47.567718505859375, 14.041854858398438, 21.35633087158203, 74.37200927734375, 44.3758544921875, 31.60816192626953, 58.5631103515625, 136.69723510742188, 8.0673828125, -47.26564025878906, 16.219871520996094, -0.14522552490234375, -10.419189453125, 96.07466125488281, 69.2491455078125, 71.42571258544922, 28.800140380859375, 52.49021911621094, 52.86517333984375, 55.63035583496094, 26.32171630859375, 97.09063720703125, -49.883270263671875, 100.0177001953125, -4.030059814453125, -64.34968566894531, 59.20037841796875, 6.536012649536133, 81.51156616210938, -19.832561492919922, 38.178436279296875, 12.299530029296875, -4.471183776855469, 85.97838592529297, 77.69879150390625, 9.038080215454102, 32.482177734375, 61.968017578125, 54.9315185546875, 66.83990478515625, 17.684539794921875, -18.111404418945312, 10.916160583496094, 50.67618179321289, 80.55641174316406, 74.79917907714844, 2.7247543334960938, -12.786762237548828, -0.9906005859375, 2.267242431640625, 120.54840087890625, 39.47979736328125, 64.67166137695312, 53.0341796875, -18.647201538085938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000311.npy"}
{"epoch": 0.6513089005235602, "step": 312, "batch_size": 128, "mean": 27.501510620117188, "std": 40.3631477355957, "min": -89.162109375, "p10": -19.749788665771483, "median": 25.828723907470703, "p90": 77.04231567382813, "max": 117.66525268554688, "pos_frac": 0.75, "sample": [40.499603271484375, -1.457122802734375, 8.005767822265625, 67.79008483886719, 62.2486572265625, -89.162109375, 69.21377563476562, 70.75167846679688, -35.89220428466797, -16.045867919921875, -15.44354248046875, -22.398948669433594, 49.026824951171875, 49.264892578125, 27.22943115234375, 87.59384155273438, 31.506912231445312, 74.36317443847656, -17.688125610351562, -30.42828369140625, 7.69866943359375, 62.53042984008789, 71.786376953125, 2.0106201171875, 9.692314147949219, 11.003494262695312, 27.96197509765625, 117.66525268554688, -32.262847900390625, -13.31229019165039, 60.391448974609375, -3.177631378173828, 8.8272705078125, 91.75537109375, 31.03118896484375, 21.366317749023438, 38.219635009765625, -0.7726459503173828, 38.76424026489258, 34.004302978515625, 94.01573944091797, 79.16206359863281, 24.428016662597656, 75.43011474609375, 38.87090301513672, 51.478858947753906, 37.83099365234375, 65.64736938476562, 69.49209594726562, 21.993621826171875, 76.95175170898438, 69.85061645507812, 19.921615600585938, 6.408561706542969, -6.58734130859375, 13.46136474609375, 27.380889892578125, 23.970535278320312, -1.509368896484375, 77.25363159179688, 4.830787658691406, -16.17584228515625, 71.02890014648438, -48.4967041015625, 18.748336791992188, -27.4200439453125, 97.69644165039062, 50.896514892578125, -13.872650146484375, 57.28790283203125, 19.32794189453125, 73.44049072265625, 23.433591842651367, 1.0488643646240234, 51.690582275390625, 67.72344970703125, 5.310691833496094, 79.23919677734375, -9.151435852050781, 60.62103271484375, 49.910919189453125, 93.93582153320312, -48.322540283203125, 68.34204864501953, 1.6565704345703125, 0.0, 43.2720947265625, -20.245399475097656, -59.949005126953125, -85.79081726074219, 13.132568359375, 43.749053955078125, 11.06201171875, 40.73173522949219, 67.36602783203125, 0.5469436645507812, -47.23695373535156, 14.52016830444336, 62.53033447265625, 5.0388336181640625, 37.0491943359375, 3.3722991943359375, 57.320098876953125, 50.93335723876953, -11.0369873046875, 34.57496643066406, 49.817840576171875, 6.8077392578125, 98.495849609375, 19.92279052734375, 82.99569702148438, 5.7421875, 37.739044189453125, -4.5211944580078125, -17.438980102539062, 45.330230712890625, 41.451416015625, 23.693023681640625, -17.028594970703125, 92.2684326171875, -30.27288818359375, 95.25119018554688, -19.537384033203125, 70.57997131347656, 46.74859619140625, 19.294647216796875, 23.725522994995117, -4.156982421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000312.npy"}
{"epoch": 0.6534031413612565, "step": 313, "batch_size": 128, "mean": 26.955429077148438, "std": 45.27848434448242, "min": -91.65798950195312, "p10": -27.404518127441406, "median": 22.316604614257812, "p90": 82.24221496582031, "max": 138.6368408203125, "pos_frac": 0.7421875, "sample": [92.84121704101562, 74.00421142578125, -25.5111083984375, -6.415496826171875, -23.384765625, -6.333038330078125, 59.29931640625, 74.74191284179688, 79.45672607421875, -10.018463134765625, -14.206253051757812, 77.04568481445312, -63.92437744140625, 77.95942687988281, 45.079002380371094, 81.89332580566406, 73.41531372070312, 27.16314697265625, -15.710723876953125, -15.759033203125, 4.919921875, 10.7181396484375, 3.1337623596191406, 81.77130126953125, 56.99078369140625, -4.451995849609375, -4.42242431640625, 41.888671875, 49.898193359375, 110.11740112304688, 106.0814208984375, 128.4940185546875, 34.417274475097656, 28.1993408203125, -31.737091064453125, 45.90167236328125, 27.978065490722656, 36.282623291015625, 102.63558959960938, 33.99212646484375, -17.58721923828125, 107.84991455078125, 83.05628967285156, 9.4954833984375, -47.961578369140625, 21.211273193359375, 13.889251708984375, -36.027313232421875, 58.303955078125, 74.0433349609375, 33.69181823730469, 33.895538330078125, -59.802825927734375, 7.521085739135742, 5.0328369140625, -5.2330780029296875, 74.44010925292969, 64.94483947753906, 13.441421508789062, 6.255767822265625, 67.50505828857422, 3.9688720703125, 68.05022430419922, 10.518875122070312, 13.853828430175781, 34.541046142578125, -17.586502075195312, 53.283172607421875, -24.028221130371094, -27.048583984375, 93.07707214355469, -8.725494384765625, 23.41876983642578, 75.16067504882812, 16.95159912109375, -2.26116943359375, 17.770675659179688, 84.25172424316406, -39.76507568359375, 55.91719436645508, 11.549636840820312, -62.582061767578125, 48.27677917480469, -28.235031127929688, 27.06329345703125, 68.93975830078125, 17.955902099609375, -10.883941650390625, 13.006002426147461, 6.189964294433594, 20.28515625, 138.6368408203125, 58.02626419067383, 43.0120849609375, 65.5660400390625, 79.31399536132812, 18.991775512695312, 94.8800048828125, -23.77752685546875, -79.54751586914062, 45.124267578125, 46.21215057373047, 7.428808212280273, 81.3980712890625, 19.83978271484375, -40.328369140625, 1.138845443725586, 5.6278533935546875, 74.52914428710938, 71.95623779296875, 11.385040283203125, 46.0399169921875, 23.542022705078125, 1.16436767578125, 1.5480594635009766, -4.8332366943359375, 51.930908203125, 16.871429443359375, -53.76141357421875, -57.719879150390625, 1.343109130859375, -91.65798950195312, 54.317901611328125, 21.82342529296875, 26.02520751953125, 89.763427734375, 22.809783935546875, 86.34916687011719], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000313.npy"}
{"epoch": 0.6554973821989529, "step": 314, "batch_size": 128, "mean": 26.702152252197266, "std": 42.136959075927734, "min": -88.193603515625, "p10": -25.975325775146484, "median": 27.291534423828125, "p90": 74.25968475341797, "max": 115.13967895507812, "pos_frac": 0.7421875, "sample": [19.174179077148438, 28.133514404296875, 62.89210510253906, 33.880279541015625, -14.679000854492188, 23.32672119140625, -37.86854553222656, 55.30268859863281, -26.03766632080078, -57.82749938964844, -7.8967742919921875, 40.67372131347656, 63.89788818359375, 74.01716613769531, 94.9632568359375, -38.646331787109375, -38.974029541015625, 99.33734130859375, 66.22803497314453, 82.89614868164062, 29.56683349609375, -7.343658447265625, 37.821990966796875, 51.806243896484375, 9.231796264648438, 20.246070861816406, 3.2894515991210938, 91.31976318359375, 3.975067138671875, 96.59133911132812, 11.926820755004883, -13.111778259277344, 5.5587158203125, -30.346572875976562, 71.98446655273438, 46.415496826171875, 45.000335693359375, 66.68940734863281, 25.3614501953125, 84.05465698242188, -21.421371459960938, 56.80244445800781, -25.101593017578125, 61.984344482421875, 3.735595703125, 19.094863891601562, 99.12179565429688, 30.740272521972656, 19.293701171875, 9.533203125, 55.2303466796875, -1.01251220703125, 6.6939849853515625, 64.82720947265625, -9.47845458984375, 33.849761962890625, 26.449554443359375, 16.60986328125, 0.3260040283203125, -38.28089904785156, 73.31318664550781, 61.82244873046875, 61.171348571777344, 0.0, -6.794586181640625, -45.89900207519531, -85.42486572265625, 69.74847412109375, 20.961395263671875, -9.04901123046875, 2.035675048828125, 19.635589599609375, 37.27557373046875, 67.50588989257812, 73.68070983886719, -22.964866638183594, 3.1711349487304688, 99.293701171875, -25.9486083984375, -88.193603515625, 59.89582061767578, 31.483673095703125, 69.4091796875, 59.494171142578125, 0.0, 67.12353515625, -10.72979736328125, 61.48108673095703, 74.8255615234375, 28.610595703125, 22.960594177246094, 69.20599365234375, 64.26448059082031, 29.3253173828125, 23.789276123046875, 79.21307373046875, 38.754974365234375, 21.34014892578125, 30.98175048828125, 10.18646240234375, 7.812244415283203, -18.431442260742188, 0.09857177734375, -33.614959716796875, 34.324188232421875, -10.544677734375, 55.92115783691406, 56.57157897949219, 40.3350830078125, 58.68736267089844, 95.71780395507812, 15.68585205078125, -47.948944091796875, 5.4031219482421875, 2.41534423828125, 115.13967895507812, 63.9530029296875, -9.316047668457031, 9.516082763671875, 36.5888671875, 72.842529296875, 48.72277069091797, -16.393844604492188, 49.265106201171875, -81.56610107421875, 105.07553100585938, 50.38047790527344, -7.55052375793457], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000314.npy"}
{"epoch": 0.6575916230366492, "step": 315, "batch_size": 128, "mean": 30.639928817749023, "std": 41.82595443725586, "min": -90.492431640625, "p10": -13.177175140380859, "median": 23.847721099853516, "p90": 86.50396270751952, "max": 114.2657470703125, "pos_frac": 0.8046875, "sample": [4.7869415283203125, 35.76446533203125, 3.7682247161865234, 2.4471893310546875, 35.49645233154297, -22.5361328125, -19.2933349609375, 9.41644287109375, 73.99136352539062, -12.743377685546875, 23.921127319335938, -7.133209228515625, 53.68230438232422, 17.738616943359375, 60.796173095703125, -12.93170166015625, 50.0299072265625, 79.70932006835938, 35.659576416015625, -9.868568420410156, -36.11137008666992, -13.749946594238281, 1.5074920654296875, 4.992973327636719, 106.52755737304688, 7.2042236328125, 58.680908203125, 94.33139038085938, 106.14483642578125, 0.293975830078125, 23.774314880371094, 32.06097412109375, 89.07810974121094, 85.4007568359375, 44.680503845214844, 114.2657470703125, 95.46026611328125, 75.63572692871094, 51.49351501464844, -3.721099853515625, 4.694854736328125, -25.390838623046875, 19.36798095703125, 74.51446533203125, -5.1952056884765625, 13.568008422851562, 6.119903564453125, 10.581329345703125, 13.91009521484375, 3.63092041015625, 45.932342529296875, -8.43838119506836, 76.12286376953125, 97.36398315429688, 79.0565185546875, 79.64289855957031, 12.706451416015625, -4.511619567871094, 18.69256591796875, 25.746723175048828, 7.8885040283203125, 3.764495849609375, -8.91659164428711, 36.56361389160156, 70.05630493164062, 73.47071838378906, 38.130615234375, 60.46837615966797, 15.110595703125, 75.92523193359375, -43.646514892578125, 24.811843872070312, 56.010955810546875, 24.954315185546875, 70.61103820800781, -26.40106201171875, -90.492431640625, 100.68658447265625, 3.6991443634033203, 76.74310302734375, 92.64788818359375, -0.28497314453125, -3.5738067626953125, 68.11923217773438, 104.90374755859375, 1.8593826293945312, 26.13653564453125, 66.61790466308594, 14.697845458984375, 105.724365234375, 9.877639770507812, 18.93267822265625, -75.51251220703125, -19.764251708984375, 80.25790405273438, 90.75534057617188, 50.68841552734375, 11.282878875732422, -66.55036926269531, 69.689697265625, 19.335693359375, 7.62066650390625, 0.465972900390625, 26.90906524658203, 0.4983673095703125, -61.599609375, -32.225433349609375, 44.9061279296875, 12.873945236206055, 9.68402099609375, 84.03042602539062, 9.857040405273438, 65.41719055175781, 93.90412902832031, 14.66131591796875, 55.0115966796875, 77.9268798828125, 9.58294677734375, 10.635833740234375, 14.969131469726562, 50.418853759765625, -5.303779602050781, 75.28782653808594, 29.511558532714844, 66.42609405517578, 69.31988525390625, 44.183990478515625, 32.88819122314453], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000315.npy"}
{"epoch": 0.6596858638743456, "step": 316, "batch_size": 128, "mean": 32.503456115722656, "std": 44.0900993347168, "min": -69.48188781738281, "p10": -22.007107543945313, "median": 28.676979064941406, "p90": 93.95662078857421, "max": 130.9765625, "pos_frac": 0.7421875, "sample": [95.806884765625, 9.008110046386719, 23.0419921875, 31.5577392578125, -6.70947265625, -21.610336303710938, -4.2496337890625, 100.44805908203125, 16.19561767578125, 99.92059326171875, 22.565109252929688, -21.95379638671875, 80.07765197753906, 16.86449432373047, -21.633926391601562, 27.180871963500977, 55.35386657714844, -23.847938537597656, 54.97833251953125, 18.048980712890625, 57.33758544921875, 86.23921203613281, -22.131500244140625, 57.85365295410156, 71.73345947265625, 107.01663208007812, 32.155548095703125, 72.9293212890625, 16.76788330078125, 108.16056823730469, -3.5178604125976562, 77.02427673339844, -17.40509033203125, 62.84483337402344, 84.32363891601562, 1.3942794799804688, 45.603843688964844, 82.08148193359375, 95.58183288574219, 29.000640869140625, -4.073150634765625, 49.44352722167969, 12.30999755859375, 50.077728271484375, 7.78076171875, 52.58954620361328, -26.40097427368164, 89.341552734375, 9.57830810546875, 1.244384765625, 22.278778076171875, 6.485504150390625, -69.48188781738281, -23.6954345703125, -19.581573486328125, 87.55267333984375, 61.88467025756836, 46.3970947265625, 6.335205078125, -0.2799072265625, -6.648643493652344, 32.79278564453125, 114.2607421875, 77.52027893066406, 30.5887451171875, 23.7933349609375, -34.23101806640625, 130.9765625, 26.60931396484375, 61.6343994140625, 7.9972686767578125, 0.4447174072265625, -23.593994140625, -28.001983642578125, 0.4375457763671875, 73.21038055419922, 29.24737548828125, 79.68486022949219, 7.8724517822265625, 77.92230224609375, 66.50347900390625, 67.91124725341797, 102.36138916015625, -5.50872802734375, -2.0038890838623047, 36.67413330078125, 38.086578369140625, 108.14871215820312, -11.330635070800781, 85.992919921875, -60.156585693359375, 50.80322265625, 0.0, -49.7864990234375, 28.353317260742188, 90.72085571289062, 102.06890869140625, 9.223167419433594, -2.88848876953125, 116.36843872070312, 48.342559814453125, 13.578750610351562, -49.94638442993164, 9.390579223632812, -23.2791748046875, 72.4732666015625, 86.64450073242188, 0.284423828125, 38.13287353515625, 47.2427978515625, 106.1087646484375, 9.47344970703125, 93.26010131835938, -15.711334228515625, 18.96246337890625, 36.505615234375, -8.165863037109375, -66.99061584472656, 22.411056518554688, -1.73773193359375, 55.299896240234375, 45.422454833984375, 64.44387817382812, 32.471282958984375, 47.00999069213867, 7.7649383544921875, 66.99285125732422, -5.8267669677734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000316.npy"}
{"epoch": 0.6617801047120419, "step": 317, "batch_size": 128, "mean": 28.647232055664062, "std": 40.642765045166016, "min": -76.80740356445312, "p10": -14.815997314453124, "median": 22.3787899017334, "p90": 80.44165649414063, "max": 145.05987548828125, "pos_frac": 0.7421875, "sample": [63.09730529785156, -61.51890182495117, 0.0, 3.7385025024414062, 107.34613037109375, -14.745758056640625, 92.53427124023438, 43.888092041015625, 8.996021270751953, 88.368896484375, 11.268402099609375, 78.06295776367188, 34.509918212890625, -10.44140625, 68.72384643554688, -28.7503662109375, -5.3846893310546875, 72.21994018554688, 59.09381103515625, 6.089363098144531, 42.26324462890625, 10.70654296875, 59.76788330078125, 56.55955505371094, 5.861091613769531, 77.70369720458984, 0.0799102783203125, 1.003326416015625, -4.896240234375, 19.768951416015625, -7.923675537109375, -26.563385009765625, 18.316696166992188, -4.6161651611328125, -14.356285095214844, 47.48933410644531, 69.49405670166016, 73.40440368652344, 58.175621032714844, -23.8690185546875, -10.750335693359375, 27.370285034179688, 88.6689453125, -62.26055145263672, -60.02947998046875, 80.4268798828125, 110.97613525390625, 64.91242980957031, 12.706596374511719, 0.0, 40.57811737060547, 37.4310302734375, 37.95201873779297, 111.31668090820312, -28.844024658203125, -13.372604370117188, 6.927001953125, 23.069549560546875, 0.1731109619140625, 0.059478759765625, 8.0743408203125, 88.3419189453125, 56.00608825683594, 52.984031677246094, 0.0, 0.34222412109375, 22.2943115234375, 14.116607666015625, 18.682903289794922, 52.2286376953125, 66.06350708007812, -5.5352325439453125, -5.83599853515625, -76.80740356445312, 96.34701538085938, 68.97500610351562, 80.47613525390625, -46.2122802734375, 40.07081604003906, -9.8516845703125, 8.491500854492188, 63.22385025024414, 59.40834045410156, -16.713966369628906, 38.64459228515625, 41.691436767578125, 24.105621337890625, -18.90435791015625, 61.72823715209961, 74.83515167236328, -1.0068893432617188, 17.563690185546875, 97.20223999023438, 8.550239562988281, 12.172866821289062, -4.944990158081055, 13.236312866210938, 60.44073486328125, 46.4786491394043, 145.05987548828125, 7.299552917480469, 60.0244026184082, 47.948822021484375, -14.979888916015625, -9.610443115234375, 28.665069580078125, 16.074859619140625, 27.060699462890625, 0.2676677703857422, 25.588287353515625, 59.98060607910156, 4.703987121582031, 78.00244140625, 5.169162750244141, 37.7923583984375, 56.465057373046875, 60.445159912109375, 20.226104736328125, 92.20350646972656, 22.463268280029297, -4.449489593505859, 32.01863098144531, -21.77581787109375, 90.21160888671875, 21.922348022460938, 74.11673736572266, -4.215980529785156, 58.425872802734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000317.npy"}
{"epoch": 0.6638743455497382, "step": 318, "batch_size": 128, "mean": 28.17367172241211, "std": 39.8231086730957, "min": -88.248046875, "p10": -16.329550170898433, "median": 27.63592529296875, "p90": 81.59539794921875, "max": 116.24981689453125, "pos_frac": 0.7578125, "sample": [31.470504760742188, 16.209701538085938, 81.4713363647461, 70.27947235107422, 12.948638916015625, 1.689300537109375, 27.73858642578125, 22.800872802734375, 71.29132080078125, 34.38221740722656, 0.609710693359375, -7.8118743896484375, 88.84738159179688, 11.793367385864258, 98.6693115234375, 6.5240478515625, 63.879425048828125, -9.885391235351562, 41.06117248535156, 14.816085815429688, -15.352325439453125, -18.6097412109375, 52.04736328125, 90.991943359375, -12.014190673828125, 41.14326477050781, 57.0550537109375, 71.47837829589844, -55.20074462890625, 18.229644775390625, 2.3727340698242188, -19.852584838867188, 19.85821533203125, 18.3282470703125, 23.487655639648438, -8.94610595703125, 5.863468170166016, 24.608123779296875, 58.874420166015625, 114.42825317382812, 25.3251953125, 58.317291259765625, 36.072509765625, 60.2005615234375, 1.6642913818359375, 4.780426025390625, -63.11445617675781, 50.218040466308594, 0.258514404296875, 68.44439697265625, -88.248046875, 13.98007583618164, 35.051292419433594, 97.90362548828125, 65.09850311279297, 30.255645751953125, -13.553153991699219, 66.20114135742188, 75.87405395507812, -8.4273681640625, 43.43144607543945, 56.303192138671875, 92.3773193359375, 29.261276245117188, -40.242862701416016, 36.3052978515625, -11.721588134765625, 4.54315185546875, 81.62445068359375, -3.2545928955078125, 65.31184387207031, 29.892105102539062, 9.1251220703125, 36.38225555419922, 62.352664947509766, 58.404937744140625, 31.5477294921875, 66.7541275024414, 34.801483154296875, -8.750091552734375, 20.177818298339844, 24.0836181640625, 27.53326416015625, 77.79519653320312, 31.09539794921875, -19.42333984375, 23.646270751953125, 61.878570556640625, -38.88701629638672, -10.751541137695312, -2.17315673828125, 39.06903076171875, 23.29425048828125, 49.03514099121094, 51.8323974609375, 31.849700927734375, 47.774871826171875, -9.816055297851562, 60.33392333984375, -9.24560546875, 63.58088684082031, 116.24981689453125, 85.85260009765625, 24.090866088867188, -34.4036865234375, 42.70404052734375, 33.412078857421875, 0.6985321044921875, -1.0832977294921875, -28.105575561523438, 86.08425903320312, 22.87872314453125, 39.44305419921875, 96.68972778320312, 25.068077087402344, 82.83807373046875, 39.905521392822266, 81.58294677734375, -22.831878662109375, 44.822235107421875, -54.831146240234375, 25.899261474609375, -3.8953475952148438, 20.9990234375, 89.65116882324219, -2.0333480834960938, -86.46723937988281, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000318.npy"}
{"epoch": 0.6659685863874345, "step": 319, "batch_size": 128, "mean": 30.636043548583984, "std": 37.23179244995117, "min": -91.37278747558594, "p10": -13.965631103515625, "median": 25.26223373413086, "p90": 81.52777252197265, "max": 120.79443359375, "pos_frac": 0.8046875, "sample": [52.450294494628906, 2.5250625610351562, 32.32373046875, 29.270339965820312, 10.734550476074219, 111.92852783203125, 53.5771484375, -34.16265869140625, 44.40238952636719, 13.1285400390625, -14.56988525390625, 17.043315887451172, 36.275352478027344, 85.640625, 70.92672729492188, 85.29315185546875, -1.42010498046875, 0.0, 20.456390380859375, 52.44427490234375, 7.607765197753906, 18.288253784179688, 50.65496826171875, 4.713836669921875, 20.87811279296875, 61.044036865234375, 4.547149658203125, 19.825714111328125, -13.7066650390625, 41.714500427246094, -21.421600341796875, 5.392967224121094, 53.0521240234375, 10.921669006347656, 96.2824935913086, 20.549331665039062, 5.088348388671875, 68.39556884765625, 74.98379516601562, 21.523239135742188, 4.163084030151367, 69.92898559570312, 55.1402587890625, -53.52508544921875, -2.3511886596679688, 37.796234130859375, 79.92413330078125, 3.07696533203125, -37.443267822265625, 85.7335205078125, 67.00048828125, 60.255706787109375, 9.803741455078125, 15.214752197265625, 48.86651611328125, 22.8271484375, 19.043975830078125, -26.027000427246094, 19.75463104248047, 75.90753173828125, -91.37278747558594, 27.69731903076172, 52.062095642089844, -31.031356811523438, 55.35887908935547, -11.616539001464844, 20.687118530273438, 14.778852462768555, 6.9329833984375, 44.21826171875, 7.872772216796875, 44.98529815673828, 63.30729675292969, 83.66917419433594, -0.7418975830078125, -13.105224609375, 63.835693359375, 66.16998291015625, 86.81486511230469, -18.490310668945312, 9.675422668457031, 57.849700927734375, 67.2823715209961, 19.479644775390625, 14.22442626953125, -0.3601837158203125, 64.21658325195312, 120.79443359375, 43.1431884765625, 8.108062744140625, -10.978973388671875, 45.970306396484375, 28.50958251953125, -2.07269287109375, 60.511749267578125, 13.083541870117188, 15.950942993164062, 84.90142822265625, -29.497528076171875, 99.28369140625, 37.9510498046875, 8.409027099609375, 83.38604736328125, 13.469131469726562, -35.435577392578125, 10.494873046875, 56.6904296875, 43.89091491699219, 80.73136901855469, -5.284893035888672, 30.782440185546875, 28.31987762451172, -16.026262283325195, 73.98567962646484, -9.3807373046875, 54.9197998046875, 55.56306838989258, 99.67822265625, -18.42577362060547, 33.827789306640625, 86.321044921875, 5.000701904296875, 39.895591735839844, 7.977165222167969, 59.05712890625, 46.0340576171875, 17.75336456298828, 76.02566528320312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000319.npy"}
{"epoch": 0.6680628272251309, "step": 320, "batch_size": 128, "mean": 29.64186668395996, "std": 45.03295135498047, "min": -84.088623046875, "p10": -29.643872070312497, "median": 28.115921020507812, "p90": 88.4937255859375, "max": 106.84115600585938, "pos_frac": 0.75, "sample": [48.492610931396484, 25.595190048217773, -28.8355712890625, -72.07748413085938, 106.84115600585938, 84.53036499023438, -31.5299072265625, 0.15850830078125, 76.39637756347656, 86.13685607910156, 30.43121337890625, 67.8609619140625, 28.45758056640625, 5.48187255859375, 78.5390625, 73.9334716796875, -17.9951171875, -0.9824409484863281, 14.1083984375, 19.207862854003906, 8.794685363769531, 80.6628646850586, 45.16949462890625, 18.948928833007812, 4.0041351318359375, -35.0948486328125, 105.68301391601562, 58.98579406738281, 19.974441528320312, 29.649166107177734, 42.11613464355469, 63.7349853515625, 89.57247924804688, 35.821693420410156, 37.07080078125, 68.2474365234375, -12.455078125, 62.37413024902344, 11.609466552734375, -0.8732070922851562, 6.112051010131836, 25.373519897460938, 57.0096435546875, -15.852241516113281, 91.21891784667969, 104.28451538085938, 49.31341552734375, 11.344085693359375, 38.309326171875, 35.34797668457031, -21.037643432617188, 9.73565673828125, -48.28367614746094, -0.826080322265625, 79.41769409179688, -8.819549560546875, 10.871368408203125, 7.120452880859375, 42.63745880126953, 81.93148803710938, 72.68597412109375, 92.85357666015625, 20.984893798828125, 88.7904052734375, 83.90402221679688, -17.84075927734375, 79.0746841430664, -13.189208984375, 91.896240234375, 75.5793228149414, 33.15276336669922, -12.20111083984375, 21.74060821533203, 6.068878173828125, -84.088623046875, 33.2628288269043, 10.62353515625, 86.011474609375, 10.14276123046875, 65.03239440917969, -58.335479736328125, 22.672698974609375, 14.800949096679688, -7.2261962890625, -27.124191284179688, 38.298065185546875, 38.3653564453125, 64.800537109375, 34.62388610839844, 2.6115798950195312, -41.54872131347656, 27.774261474609375, 100.63140869140625, -0.26190185546875, 62.3350830078125, -18.35879135131836, 9.45635986328125, 47.31166076660156, -17.415451049804688, 93.7080078125, 61.58740234375, -33.296142578125, 61.03045654296875, 88.3665771484375, 69.172607421875, 102.69122314453125, 75.68887329101562, -40.767974853515625, 46.03279113769531, -62.4141845703125, 18.636871337890625, 64.21807861328125, -64.97509765625, -1.152069091796875, 96.10784912109375, 21.24444580078125, 79.6796646118164, 18.027053833007812, 53.21711730957031, 54.15455627441406, -10.057687759399414, 11.030181884765625, 19.712203979492188, -67.13011932373047, 101.85208129882812, 22.171066284179688, -65.20420837402344, 58.97569274902344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000320.npy"}
{"epoch": 0.6701570680628273, "step": 321, "batch_size": 128, "mean": 25.292766571044922, "std": 40.33306884765625, "min": -81.47726440429688, "p10": -24.339294433593746, "median": 21.018810272216797, "p90": 84.48123321533203, "max": 111.248291015625, "pos_frac": 0.7109375, "sample": [76.79080200195312, -38.79490661621094, 39.992523193359375, -76.96987915039062, 19.435577392578125, 94.9241943359375, 13.09088134765625, 1.53631591796875, -32.28633117675781, 105.37139892578125, 66.02665710449219, 33.46165466308594, -39.006248474121094, 38.434104919433594, 41.42817687988281, 5.2877197265625, -0.7036666870117188, 70.0108413696289, 96.46023559570312, 8.94659423828125, 32.92596435546875, -11.132568359375, 7.020973205566406, -2.75616455078125, 87.94515991210938, 48.16387939453125, -0.7116851806640625, 12.890045166015625, 43.7535400390625, -0.008941650390625, 65.99872589111328, 43.287513732910156, 37.37629318237305, 7.37200927734375, 28.7928466796875, -31.257904052734375, -26.6357421875, 31.51616668701172, 7.6673126220703125, 53.03558349609375, 33.129642486572266, 24.393096923828125, -4.8064727783203125, 49.16943359375, 9.34869384765625, 95.18940734863281, -4.748626708984375, 75.14703369140625, 111.248291015625, 33.914886474609375, -8.758064270019531, 32.13691711425781, -13.320079803466797, 35.18072509765625, -5.089988708496094, 0.1435089111328125, 84.26495361328125, -3.1558151245117188, 71.66854858398438, -31.10382080078125, 4.292472839355469, 53.00347137451172, 21.778717041015625, 58.32530975341797, 43.50770568847656, 84.98588562011719, -10.920501708984375, 40.2042236328125, -31.409461975097656, -5.0262603759765625, -5.464508056640625, 41.09857177734375, -14.63836669921875, 93.30780029296875, -81.47726440429688, 62.24766540527344, 92.25509643554688, 104.30319213867188, 16.904632568359375, 53.2073974609375, 18.7398681640625, -23.3551025390625, 34.640380859375, 54.92912292480469, 9.839897155761719, 77.38343811035156, -19.652328491210938, 89.80216979980469, -47.606014251708984, 72.43023681640625, 26.58069610595703, 9.224584579467773, 53.116302490234375, 9.947929382324219, 15.020877838134766, 92.91824340820312, -2.64544677734375, 34.00897216796875, 26.301788330078125, 59.988189697265625, 56.95140075683594, -8.72369384765625, 45.244293212890625, 9.504493713378906, 4.161041259765625, 45.616790771484375, -2.188365936279297, 0.2637977600097656, 4.427419662475586, -0.44425201416015625, -9.081298828125, -35.16351318359375, 8.71522331237793, -14.01422119140625, 17.167724609375, 108.53570556640625, -62.281829833984375, 70.68057250976562, 29.280670166015625, 20.25890350341797, 7.230072021484375, 42.648193359375, 23.267837524414062, 7.833732604980469, 73.24588012695312, 73.8297119140625, 0.0, -34.2237548828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000321.npy"}
{"epoch": 0.6722513089005235, "step": 322, "batch_size": 128, "mean": 31.4697265625, "std": 41.51901626586914, "min": -94.3101806640625, "p10": -20.663211059570312, "median": 31.391983032226562, "p90": 86.82245483398437, "max": 116.02067565917969, "pos_frac": 0.7578125, "sample": [-39.366668701171875, 10.76397705078125, -45.3006591796875, 46.380455017089844, 24.587844848632812, 9.551864624023438, 34.00132751464844, -26.7003173828125, 72.78433227539062, 50.389404296875, -8.708755493164062, -20.58001708984375, 57.537139892578125, 12.5096435546875, 65.95858001708984, 59.20295715332031, 31.96697998046875, -3.42486572265625, 54.86744689941406, 44.738861083984375, 33.13262939453125, -11.319915771484375, 103.476806640625, 1.7107391357421875, 5.19586181640625, 23.923927307128906, -17.296295166015625, 58.37895965576172, 6.4268035888671875, 59.17767333984375, 28.718650817871094, 74.22564697265625, -4.645347595214844, -15.61431884765625, 69.15193176269531, 25.77392578125, 23.39917755126953, 81.56000518798828, -35.378387451171875, 65.01407623291016, 76.19241333007812, 66.64653778076172, -23.455642700195312, 21.284698486328125, 3.7340087890625, 9.753173828125, 46.794944763183594, -21.325149536132812, 68.55941772460938, -59.21128463745117, 64.93177795410156, 6.7853240966796875, 69.46109008789062, 101.83981323242188, -6.194061279296875, 77.28068542480469, 50.10469055175781, -17.651260375976562, 34.75294494628906, 63.85670471191406, 101.371826171875, 0.8589019775390625, 107.2996826171875, 109.68222045898438, -14.118377685546875, 55.86035919189453, 48.94874572753906, 36.939605712890625, 61.5275993347168, 11.721542358398438, 102.5955810546875, 89.669677734375, 55.6710205078125, 42.825469970703125, 8.437042236328125, 0.0, -42.629547119140625, 45.000030517578125, 44.21002197265625, -20.857330322265625, 88.5693359375, 58.2591552734375, 73.58880615234375, 85.42282104492188, -7.7397308349609375, 28.279388427734375, 44.21600341796875, -31.442138671875, -2.39453125, 34.395477294921875, 93.54119873046875, -4.52642822265625, 5.636016845703125, -25.24267578125, -2.969482421875, 1.9077720642089844, 30.816986083984375, 2.7886199951171875, -94.3101806640625, 55.76678466796875, 58.39784240722656, 14.923843383789062, 50.475921630859375, 86.07379150390625, 47.96484375, -11.227653503417969, 3.7915725708007812, 116.02067565917969, 108.32598876953125, 5.180164337158203, -11.729766845703125, 62.75508117675781, 22.40423583984375, 18.215530395507812, 6.4734649658203125, 9.628894805908203, -28.777610778808594, 10.883697509765625, 27.22900390625, 53.28755187988281, 64.06906127929688, 106.45462036132812, 63.35633850097656, -2.54095458984375, 14.0220947265625, 88.97286987304688, 39.870697021484375, 83.73284912109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000322.npy"}
{"epoch": 0.6743455497382199, "step": 323, "batch_size": 128, "mean": 21.645793914794922, "std": 44.040679931640625, "min": -85.24209594726562, "p10": -28.50391693115234, "median": 15.515365600585938, "p90": 77.54732055664061, "max": 113.45941162109375, "pos_frac": 0.671875, "sample": [4.32806396484375, 7.619140625, -65.99430847167969, -60.833892822265625, 66.0202407836914, 67.64886474609375, -19.477310180664062, 29.138160705566406, -0.63690185546875, 95.89468383789062, -27.425613403320312, 68.02395629882812, 17.69580078125, -16.71710205078125, 6.6622467041015625, 51.08643341064453, 62.932098388671875, -9.160293579101562, 32.540924072265625, -14.29034423828125, 111.6734619140625, 57.35858154296875, 17.03594970703125, 63.339599609375, 14.902069091796875, 93.30751037597656, 13.5235595703125, 76.79084777832031, -38.33941650390625, 71.77175903320312, -13.59954833984375, -74.0498046875, -16.558372497558594, 26.7119140625, 0.11142158508300781, 51.07086181640625, 14.069053649902344, 13.651439666748047, 46.93635559082031, 23.576749801635742, 13.74395751953125, 67.97958374023438, -42.778594970703125, 48.7274169921875, 72.16433715820312, -11.522735595703125, -9.719741821289062, -70.9774169921875, 70.6318359375, 23.740493774414062, 5.857378005981445, 21.758014678955078, -15.77081298828125, 5.969371795654297, 22.9161376953125, 49.000640869140625, 36.038055419921875, 75.92984771728516, 82.50138854980469, -16.01561737060547, 61.43475341796875, 64.86810302734375, 49.11651611328125, -19.70849609375, 96.84967041015625, 15.730072021484375, 64.56318664550781, -18.608901977539062, -31.01995849609375, 61.30731201171875, -8.989990234375, 3.489645004272461, -15.138168334960938, 1.9700927734375, 67.53822326660156, 20.975311279296875, 103.11981201171875, -6.508941650390625, 88.51414489746094, 5.51544189453125, 66.21916961669922, -55.46026611328125, 12.92831039428711, 15.19122314453125, 12.723388671875, 101.0579833984375, -43.20576477050781, 84.94631958007812, -60.08343505859375, 34.70975875854492, 75.34808349609375, 31.629966735839844, -16.731033325195312, 50.780845642089844, 1.0298309326171875, -8.349769592285156, -27.08435821533203, -16.20000457763672, 3.1782379150390625, -4.2784271240234375, 26.5020751953125, -41.379974365234375, -7.835205078125, -3.9714813232421875, 30.28448486328125, 35.91387939453125, 10.91461181640625, -8.303840637207031, 68.9505615234375, 104.833251953125, -5.894340515136719, 29.922256469726562, 79.31242370605469, 74.25932312011719, -64.89205169677734, -85.24209594726562, 6.5627288818359375, 15.3006591796875, 45.03641891479492, 27.440963745117188, 16.293365478515625, -5.18896484375, 45.803192138671875, 82.16096496582031, 58.493804931640625, -7.166717529296875, -12.788204193115234, 113.45941162109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000323.npy"}
{"epoch": 0.6764397905759162, "step": 324, "batch_size": 128, "mean": 34.33634948730469, "std": 38.75347900390625, "min": -38.265106201171875, "p10": -10.51304416656494, "median": 25.914566040039062, "p90": 86.54451293945313, "max": 122.22650146484375, "pos_frac": 0.7734375, "sample": [29.917526245117188, 96.4674072265625, 16.32647705078125, 72.25895690917969, 19.4718017578125, 61.0113525390625, 3.832672119140625, 1.4243106842041016, 82.27633666992188, 18.436790466308594, 20.07525634765625, -0.4076118469238281, -21.630584716796875, 97.10821533203125, -7.471435546875, 89.81198120117188, 63.642234802246094, 11.165142059326172, -12.8544921875, 0.0, 14.32147216796875, -5.789344787597656, 70.51184844970703, 1.2950592041015625, -0.8910675048828125, 24.137847900390625, 58.91584777832031, -25.164636611938477, 63.776214599609375, -32.28185272216797, 8.291479110717773, 65.23273468017578, 73.33149719238281, 94.27093505859375, 64.60276794433594, 62.684608459472656, 4.32794189453125, 38.961334228515625, 59.4276123046875, 0.0, -1.4006366729736328, 81.56475830078125, 87.527587890625, 68.24491882324219, 47.822479248046875, 68.3088150024414, 0.0, -3.181396484375, 54.2210693359375, -0.820556640625, -27.393692016601562, 23.364227294921875, 43.35911560058594, 15.916748046875, 60.14031982421875, 33.290626525878906, 3.5439453125, 69.32603454589844, 89.52584838867188, 19.432373046875, -34.295654296875, 81.28854370117188, 7.6107940673828125, 66.87884521484375, 56.909812927246094, -11.633474349975586, 38.79817581176758, 56.88958740234375, 22.2625732421875, -38.265106201171875, 94.72463989257812, 36.076332092285156, -37.653541564941406, 68.02790832519531, 18.38538360595703, -2.097259521484375, 23.325820922851562, 21.125503540039062, 7.1269378662109375, 122.22650146484375, 111.90179443359375, 4.487644195556641, -7.787666320800781, 36.357696533203125, 107.62667846679688, 78.2840576171875, 16.372879028320312, 15.0404052734375, -16.127899169921875, 11.81536865234375, 72.41557312011719, 27.6912841796875, 38.63591766357422, -0.95257568359375, -18.27325439453125, 47.72259521484375, 28.384014129638672, 87.833984375, 66.37364196777344, 5.703369140625, -27.169517517089844, -6.912591934204102, 7.922607421875, 33.631011962890625, -28.464019775390625, 86.210693359375, 109.87368774414062, 71.82424926757812, 14.837753295898438, 4.529632568359375, 85.39059448242188, 40.143795013427734, -10.032859802246094, 67.72451782226562, 13.873748779296875, 67.21670532226562, 87.32342529296875, 69.4451904296875, -0.2592926025390625, 81.32852172851562, 70.27554321289062, 2.7478485107421875, 66.26817321777344, 76.09676361083984, 74.54627990722656, 8.014862060546875, 2.2589111328125, 1.5717010498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000324.npy"}
{"epoch": 0.6785340314136126, "step": 325, "batch_size": 128, "mean": 29.982200622558594, "std": 48.64344787597656, "min": -71.72250366210938, "p10": -28.743994903564452, "median": 18.77410125732422, "p90": 97.60994415283203, "max": 147.7928466796875, "pos_frac": 0.6875, "sample": [57.307525634765625, 78.15087890625, -41.52885437011719, -7.191162109375, 37.90289306640625, 19.651123046875, 50.2559814453125, 83.03448486328125, 35.24528503417969, -2.8372802734375, -5.34796142578125, 15.433013916015625, 76.68572235107422, 81.86041259765625, 0.99078369140625, 3.274873733520508, 90.93508911132812, 65.5262451171875, 54.617156982421875, 90.31198120117188, 85.02572631835938, 99.60487365722656, 54.40144348144531, -8.735137939453125, 56.975830078125, 8.775436401367188, 12.170032501220703, 51.563018798828125, 69.96675109863281, 10.237884521484375, -37.60926818847656, 6.4131011962890625, 31.287601470947266, 97.52799987792969, 87.27362060546875, 29.163284301757812, 16.15584373474121, 93.12882995605469, 17.897079467773438, -28.27294921875, 60.724609375, -69.2420654296875, 83.10337829589844, 3.6995849609375, -42.01295471191406, 0.109039306640625, 132.134521484375, -67.26174926757812, 103.64698791503906, -2.006256103515625, -8.122085571289062, 21.690277099609375, 97.8011474609375, 8.437835693359375, 3.9152069091796875, -38.327964782714844, -6.62689208984375, -57.78179168701172, -8.629180908203125, -32.332611083984375, 10.83148193359375, 68.9503173828125, -0.6381607055664062, -6.13604736328125, 41.458038330078125, 36.62617492675781, -11.842094421386719, -56.1219482421875, 126.266845703125, 6.878662109375, 44.574371337890625, 21.896636962890625, 147.7928466796875, 2.991241455078125, -1.17742919921875, 0.8778533935546875, -10.113449096679688, -12.3416748046875, 36.993621826171875, 71.33203125, -29.843101501464844, -9.847076416015625, -13.57977294921875, 111.84930419921875, -31.635345458984375, 49.2574462890625, 70.59722900390625, 37.19183349609375, 26.7386474609375, -17.410430908203125, 1.2575836181640625, 62.62208557128906, 103.40176391601562, -5.572174072265625, 3.3986473083496094, -71.72250366210938, -12.096046447753906, 95.05499267578125, 34.38038635253906, 63.27418899536133, 105.88894653320312, 16.343154907226562, 55.53666687011719, 121.1600341796875, 85.05581665039062, -31.281707763671875, 7.242240905761719, 43.42133331298828, 14.721389770507812, -2.84783935546875, -17.644805908203125, 79.4256591796875, 27.315780639648438, 73.43560028076172, 4.449821472167969, 68.75357818603516, 61.1595458984375, 104.97491455078125, 6.0760498046875, -5.358131408691406, -8.676025390625, -23.3675537109375, 132.02935791015625, -20.99652099609375, -12.699356079101562, 115.68801879882812, 53.251312255859375, 78.09909057617188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000325.npy"}
{"epoch": 0.680628272251309, "step": 326, "batch_size": 128, "mean": 28.50141716003418, "std": 43.34861373901367, "min": -65.69772338867188, "p10": -26.952536010742186, "median": 29.6717472076416, "p90": 83.4464630126953, "max": 107.43426513671875, "pos_frac": 0.7265625, "sample": [63.7254638671875, 61.562286376953125, 70.11909484863281, 90.80949401855469, 35.08467102050781, 38.730865478515625, -24.617279052734375, -8.995941162109375, 73.70658111572266, -27.950881958007812, 53.259002685546875, -16.491241455078125, 73.50982666015625, 76.83271789550781, 15.882614135742188, 2.3333282470703125, 7.97137451171875, 83.22344970703125, 105.57987976074219, 6.9013671875, 30.537567138671875, 85.39511108398438, 5.1343841552734375, -9.152740478515625, 42.116180419921875, 41.29460144042969, 67.52693176269531, 55.0609130859375, 22.191654205322266, 13.788368225097656, -63.84010314941406, 81.78076171875, 1.94415283203125, 33.923316955566406, -59.72224426269531, -43.854156494140625, 32.374725341796875, 32.6912841796875, -16.78106689453125, 66.06802368164062, 38.706756591796875, 29.23455810546875, -6.96820068359375, 81.73028564453125, -13.969390869140625, 46.664764404296875, -15.003509521484375, 78.03056335449219, -22.060211181640625, 88.99873352050781, 30.36285400390625, 36.331024169921875, 79.83010864257812, 13.762603759765625, 84.03515625, -65.69772338867188, -26.524673461914062, 13.2021484375, 64.51036071777344, -7.568607330322266, 71.13662719726562, 66.61432647705078, 15.3302001953125, 64.01789855957031, 0.0, -59.821441650390625, 8.936994552612305, 5.35003662109375, 61.25071716308594, -13.664794921875, 7.05908203125, 91.25820922851562, 107.43426513671875, 46.86859130859375, 24.86060333251953, -35.72625732421875, 25.243698120117188, -30.800216674804688, 29.507644653320312, 73.3274917602539, 0.415252685546875, 96.50869750976562, -8.557418823242188, 11.818313598632812, 3.9285049438476562, 72.01962280273438, -25.65289306640625, -31.819732666015625, 53.34172439575195, 6.7877349853515625, 71.54082489013672, 18.635772705078125, 99.81732177734375, -8.98101806640625, 45.166290283203125, 17.55572509765625, 34.398780822753906, 82.24835205078125, 20.814300537109375, 61.50860595703125, 85.03662109375, 0.0, 67.50213623046875, 29.542770385742188, 47.87738037109375, 73.39433288574219, -19.50311279296875, 82.04817199707031, 51.15498352050781, 0.0, 58.15948486328125, -64.93841552734375, -40.585968017578125, -49.789161682128906, 1.74224853515625, -61.83209228515625, 83.96682739257812, -10.49517822265625, 75.17086791992188, 29.800724029541016, 42.614585876464844, 10.22039794921875, -6.724853515625, -13.789718627929688, 76.49183654785156, 86.17378997802734, 21.703033447265625, 86.29922485351562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000326.npy"}
{"epoch": 0.6827225130890052, "step": 327, "batch_size": 128, "mean": 33.075462341308594, "std": 45.217403411865234, "min": -86.51382446289062, "p10": -16.355322265625, "median": 26.312679290771484, "p90": 92.42030029296875, "max": 129.0516357421875, "pos_frac": 0.765625, "sample": [15.665855407714844, 71.38533020019531, 53.203521728515625, 17.648475646972656, 129.0516357421875, -68.93789672851562, 54.1962890625, 81.40876007080078, 7.4788818359375, 62.04684829711914, -4.848020553588867, 117.48394775390625, -8.047119140625, -74.59158325195312, 35.91712188720703, -0.5399169921875, -34.4193115234375, -1.9557037353515625, 99.86611938476562, 98.36302185058594, 51.56941223144531, 20.12371063232422, 14.940643310546875, 15.191741943359375, 50.682403564453125, -26.531097412109375, 13.455032348632812, 61.77418518066406, 14.182159423828125, 68.44155883789062, 41.163726806640625, 64.01620483398438, -14.690866470336914, -10.332992553710938, 32.33551025390625, 65.150634765625, 89.58212280273438, 6.746063232421875, 124.030517578125, 48.310394287109375, -19.19439697265625, -14.609472274780273, 2.9980850219726562, 65.71359252929688, 26.41905975341797, -17.131072998046875, 93.83465576171875, 27.124114990234375, 58.96336364746094, 12.669281005859375, 87.4520263671875, 4.793304443359375, -5.798095703125, -86.51382446289062, 68.8636245727539, -0.78314208984375, 72.62859344482422, 57.95664978027344, 111.82867431640625, 25.799530029296875, 28.830230712890625, 66.04582214355469, 57.87541198730469, 15.03814697265625, 93.0592041015625, 80.4041748046875, 88.45986938476562, 11.582275390625, 34.61822509765625, -4.492828369140625, 23.614501953125, 14.26092529296875, -10.5574951171875, -18.253639221191406, 46.385498046875, 21.02130889892578, 22.378707885742188, 32.7662353515625, 105.12710571289062, 5.650909423828125, -16.022857666015625, 7.276275634765625, 26.206298828125, 73.95950317382812, 13.41107177734375, 98.446533203125, 12.672378540039062, 0.0, 82.36245727539062, 7.501029968261719, 22.935546875, 96.7213134765625, 112.54080200195312, 21.699676513671875, 65.13409423828125, 33.443572998046875, 69.25653076171875, 24.15985107421875, 74.22703552246094, 92.146484375, 57.318695068359375, -0.5254669189453125, -86.16815185546875, 12.939323425292969, 42.6109619140625, 59.809417724609375, -47.76861572265625, 22.948883056640625, 52.97271728515625, 74.8333740234375, -67.3694839477539, 53.8568115234375, 90.20025634765625, 4.86737060546875, -26.714874267578125, 58.639068603515625, -1.369781494140625, 95.6031494140625, 0.00725555419921875, -6.1824493408203125, 79.76171875, 66.8114242553711, 83.44294738769531, 6.0550384521484375, -56.23919677734375, 68.47421264648438, 17.972457885742188, -8.551849365234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000327.npy"}
{"epoch": 0.6848167539267016, "step": 328, "batch_size": 128, "mean": 27.90850830078125, "std": 41.24835205078125, "min": -78.873046875, "p10": -15.16874084472656, "median": 20.37579345703125, "p90": 79.77774047851562, "max": 123.97955322265625, "pos_frac": 0.7578125, "sample": [12.969566345214844, 52.22018814086914, 90.83238220214844, 20.784759521484375, 76.96221923828125, 23.91644287109375, 0.411346435546875, 60.403961181640625, 52.224090576171875, 59.12583923339844, 50.971038818359375, 35.4986686706543, 12.14453125, 61.03643798828125, 7.582969665527344, -3.861522674560547, 7.81787109375, -11.742279052734375, 30.56683349609375, 23.288650512695312, 33.46794128417969, 58.19487762451172, 22.58502197265625, -8.725971221923828, 109.49899291992188, 20.3511962890625, 8.782363891601562, 22.371826171875, 20.18609619140625, 66.36226654052734, -0.5955963134765625, 71.79878234863281, 27.966217041015625, 89.83447265625, 123.97955322265625, 58.55860900878906, 90.06852722167969, 4.4541168212890625, 94.157958984375, -4.388093948364258, 8.616455078125, 85.59751892089844, 66.1021728515625, 7.876426696777344, 8.87841796875, -10.251832962036133, -17.291046142578125, 20.400390625, -14.548004150390625, 68.35171508789062, 36.853790283203125, 13.751708984375, -14.1109619140625, 93.40679931640625, 12.317977905273438, 28.0665283203125, 11.08734130859375, 5.279624938964844, 14.187019348144531, -25.517417907714844, 4.804603576660156, 13.877548217773438, 79.12103271484375, 81.31005859375, 51.56199645996094, 95.53054809570312, 66.5982894897461, -53.11004638671875, 60.83350372314453, -49.41822052001953, 12.524856567382812, 52.9459228515625, 67.4362564086914, 64.5274658203125, 8.43133544921875, 69.68661499023438, 105.48648071289062, -0.481170654296875, -16.61712646484375, 70.97286987304688, -77.88458251953125, 7.768379211425781, 81.82174682617188, 19.960342407226562, -54.785400390625, 2.675262451171875, 66.76411437988281, 72.59635162353516, 62.44929504394531, -9.177581787109375, -2.1572265625, 78.7415771484375, 12.619731903076172, -1.342437744140625, 7.00286865234375, -16.64385986328125, -6.4976654052734375, 14.563812255859375, -58.5989990234375, 28.340116500854492, 33.381866455078125, 24.54290771484375, 65.54854583740234, -4.448944091796875, -5.05914306640625, 70.51953125, 77.762451171875, -78.873046875, -4.832611083984375, -23.311859130859375, 32.3369140625, 11.4615478515625, 77.70227813720703, 4.8875732421875, 17.642074584960938, 1.684906005859375, 2.86395263671875, 71.10025024414062, -3.807535171508789, -70.89952087402344, 7.4392547607421875, 68.25533294677734, 42.03715515136719, -39.29780578613281, 55.184814453125, 90.4775390625, 64.01924133300781, -1.385162353515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000328.npy"}
{"epoch": 0.6869109947643979, "step": 329, "batch_size": 128, "mean": 25.17148208618164, "std": 47.83449935913086, "min": -95.8447265625, "p10": -31.68970489501953, "median": 17.11908721923828, "p90": 85.91419677734373, "max": 138.82666015625, "pos_frac": 0.671875, "sample": [-14.5203857421875, 2.42041015625, -2.4896240234375, 36.854217529296875, -1.1013259887695312, -5.022003173828125, 10.780471801757812, 17.495285034179688, 3.6289596557617188, 25.280426025390625, 83.91163635253906, -17.647586822509766, -61.5965576171875, 36.152008056640625, 42.28599548339844, 111.08740234375, 5.198089599609375, -50.769195556640625, 70.00518798828125, 47.44105529785156, 32.10589599609375, 7.11395263671875, 74.47396850585938, -11.40283203125, 70.55133056640625, 83.34681701660156, -70.46807861328125, 59.223236083984375, 96.51312255859375, 41.49261474609375, -0.618865966796875, 7.963531494140625, 4.8046875, 81.12265014648438, -31.20257568359375, 34.80322265625, 56.79881286621094, 30.10101318359375, 16.131729125976562, 59.7645263671875, 11.7767333984375, 101.8851318359375, -3.995330810546875, -14.5535888671875, 61.025604248046875, -27.5543212890625, -1.781219482421875, 89.43508911132812, -6.523956298828125, -95.8447265625, -1.9789161682128906, 104.0147705078125, -25.395660400390625, -6.7001953125, 66.5321044921875, 31.7535400390625, 46.84086608886719, 33.34733581542969, 38.15863037109375, -51.910247802734375, 9.378875732421875, 84.40524291992188, -0.8193359375, -32.82633972167969, 0.5283737182617188, -3.3215370178222656, 0.2584953308105469, -25.79608154296875, 111.37188720703125, 11.725425720214844, 36.36366271972656, 1.350860595703125, 49.10400390625, 125.0516357421875, 14.682147979736328, 62.280792236328125, 49.211090087890625, -13.90179443359375, -45.971473693847656, 81.93080139160156, 77.88270568847656, 0.543060302734375, -86.737060546875, 59.1407470703125, -7.55755615234375, 102.65052795410156, 37.405113220214844, 4.881072998046875, 73.0972900390625, 3.1040115356445312, 112.68878173828125, 3.8597564697265625, -42.69544982910156, 106.84420776367188, 73.43991088867188, -64.32354736328125, 90.03962707519531, 78.84521484375, -1.6144981384277344, 53.27423095703125, -10.518539428710938, 10.249420166015625, 54.804718017578125, 31.245101928710938, -2.7360916137695312, -11.19989013671875, 127.410400390625, 35.81890869140625, 82.48960876464844, 138.82666015625, -2.80609130859375, 40.885406494140625, 76.78562927246094, -10.879730224609375, 32.89263916015625, -46.92078399658203, 16.742889404296875, -14.445098876953125, 39.06822204589844, -9.302093505859375, 23.275733947753906, 59.697357177734375, 2.8869075775146484, 48.08025360107422, 71.97892761230469, 27.571136474609375, -47.42059326171875, -62.850791931152344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000329.npy"}
{"epoch": 0.6890052356020943, "step": 330, "batch_size": 128, "mean": 22.858707427978516, "std": 43.46964645385742, "min": -73.62373352050781, "p10": -28.350509643554677, "median": 22.271209716796875, "p90": 82.32240142822265, "max": 136.56927490234375, "pos_frac": 0.71875, "sample": [22.5184326171875, 79.79734802246094, 14.60638427734375, -68.947998046875, 35.65332794189453, 53.440338134765625, 0.017608642578125, 88.91485595703125, 61.803802490234375, 25.940673828125, -0.218017578125, 30.2601318359375, -4.706268310546875, -17.53857421875, 24.86529541015625, -7.8132781982421875, 86.92892456054688, 24.874359130859375, 22.43701171875, -1.981536865234375, 70.09234619140625, -45.66053771972656, -12.229888916015625, 3.045440673828125, 0.82415771484375, 61.76812744140625, 18.90216064453125, -17.930068969726562, 2.9054183959960938, -10.334579467773438, 41.46192932128906, 33.523193359375, 110.36705017089844, 44.432342529296875, 17.00458526611328, -2.939727783203125, 0.35382080078125, -43.738067626953125, -19.985595703125, 15.689910888671875, 39.52520751953125, 94.74038696289062, 31.563446044921875, 32.383453369140625, 32.24803161621094, 33.925140380859375, 19.89474868774414, 117.52935791015625, 6.474220275878906, 45.940673828125, 5.557491302490234, 80.82998657226562, 69.19241333007812, -62.680816650390625, 9.0804443359375, 22.10540771484375, -60.831947326660156, 20.8170166015625, 32.53912353515625, 28.03980255126953, 4.4631195068359375, 95.42585754394531, 11.752338409423828, 45.323516845703125, -73.62373352050781, 2.007720947265625, 71.44534301757812, -40.8687744140625, 4.855846405029297, 23.930282592773438, 33.172607421875, -34.35491943359375, 101.099365234375, 27.863723754882812, -0.6336669921875, 78.57867431640625, -61.86859130859375, 112.05743408203125, -16.078033447265625, -0.629425048828125, 20.796630859375, 68.52647399902344, -2.267486572265625, -8.556732177734375, 22.596466064453125, 0.985321044921875, 26.4873046875, 85.80470275878906, 51.92718505859375, 33.35804748535156, 105.0604248046875, 8.9183349609375, 38.79612731933594, 21.093414306640625, 0.803558349609375, -21.711807250976562, 7.685150146484375, 29.601787567138672, 38.95630645751953, 66.81219482421875, 101.0048828125, 24.60296630859375, -51.9276123046875, 43.068634033203125, -58.949188232421875, 78.59603881835938, -9.88232421875, 16.228309631347656, -25.777191162109375, 25.347625732421875, 67.50189208984375, -57.03901672363281, -5.845298767089844, -53.402740478515625, -0.6444091796875, 113.8126220703125, 31.674453735351562, 29.647018432617188, 3.07220458984375, 78.52784729003906, -19.97955322265625, 136.56927490234375, 36.718780517578125, -6.5411376953125, -19.94732666015625, 77.24652099609375, 37.847381591796875, 17.513427734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000330.npy"}
{"epoch": 0.6910994764397905, "step": 331, "batch_size": 128, "mean": 28.5369930267334, "std": 42.60688781738281, "min": -89.20610046386719, "p10": -15.733715820312495, "median": 22.445667266845703, "p90": 86.69044036865233, "max": 125.32049560546875, "pos_frac": 0.7890625, "sample": [8.41070556640625, 109.62371826171875, 56.684906005859375, -32.93998718261719, 125.32049560546875, 39.43310546875, -10.031097412109375, 2.90081787109375, 37.6990966796875, 50.659820556640625, 4.0625152587890625, -11.0203857421875, 40.169677734375, -29.011993408203125, 106.45965576171875, -14.44915771484375, -4.73895263671875, 3.390317916870117, -5.6805877685546875, -18.73101806640625, 10.367485046386719, 30.30645751953125, -29.880203247070312, 28.238632202148438, 34.278926849365234, 89.94564056396484, 71.86834716796875, 39.901268005371094, 8.166397094726562, 73.59230041503906, 11.13482666015625, 29.95452880859375, 4.88092041015625, 25.7230224609375, 32.30535125732422, 83.76461791992188, 22.297378540039062, 37.96636962890625, 99.80615234375, 70.63472747802734, -20.72173309326172, 0.9273681640625, -0.8997802734375, -89.20610046386719, 4.724798202514648, 9.084545135498047, 89.00514221191406, 50.6788330078125, 34.8018798828125, 0.0, 28.8365478515625, -2.8682708740234375, 46.410888671875, -4.151824951171875, 7.2493896484375, 34.5028076171875, 20.36749267578125, 77.99518585205078, 32.109283447265625, 39.8377685546875, 12.21112060546875, 40.49261474609375, -33.6368408203125, -6.242584228515625, 116.66973876953125, 36.945823669433594, 99.86380004882812, 48.410762786865234, -9.360923767089844, 47.772605895996094, 2.4072265625, 12.733810424804688, 85.69842529296875, 4.8167724609375, 32.541053771972656, 13.641141891479492, -22.077667236328125, 2.412109375, 11.529655456542969, -19.634132385253906, 15.98736572265625, 27.657928466796875, 74.04135131835938, 15.856796264648438, -87.14637756347656, 101.15890502929688, 6.34112548828125, 80.90750122070312, -11.419769287109375, 25.4515380859375, 109.42446899414062, 48.41357421875, 60.66992950439453, 105.82357788085938, 20.936767578125, 78.68844604492188, 67.57392883300781, 13.563369750976562, -11.0262451171875, 3.3781890869140625, 0.88446044921875, -1.4139404296875, 85.0814208984375, 47.30061340332031, -78.97488403320312, 22.593955993652344, 116.302978515625, 2.750375747680664, 64.15493774414062, 1.8094100952148438, 7.059417724609375, 15.307823181152344, 74.62527465820312, 10.618499755859375, 66.52458190917969, 41.781829833984375, 5.12353515625, 62.64404296875, 14.527143478393555, 68.2152099609375, -80.51141357421875, 16.28533172607422, 106.298095703125, 63.38169860839844, 40.920654296875, -27.241607666015625, 71.3690414428711, 19.690628051757812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000331.npy"}
{"epoch": 0.6931937172774869, "step": 332, "batch_size": 128, "mean": 31.842941284179688, "std": 45.13701248168945, "min": -100.33256530761719, "p10": -18.76429634094238, "median": 27.908416748046875, "p90": 89.47648010253906, "max": 121.47616577148438, "pos_frac": 0.7578125, "sample": [23.22719955444336, 27.876708984375, 2.3739089965820312, 108.21875, 64.39448547363281, 76.50199890136719, 40.86915588378906, 39.879737854003906, 114.993408203125, 73.86907196044922, 10.269668579101562, 10.428850173950195, 3.069011688232422, -9.020782470703125, 6.19842529296875, 59.90264892578125, -16.348861694335938, 12.6187744140625, 86.85739135742188, -19.588237762451172, 56.370208740234375, 5.638885498046875, 27.94012451171875, 30.58777618408203, 0.278533935546875, -4.62677001953125, -24.57867431640625, 3.8061676025390625, -1.1141624450683594, 2.653411865234375, 71.6756591796875, 97.0220947265625, -0.85394287109375, 3.546630859375, 90.29403686523438, 57.79914855957031, 76.5963134765625, 82.98420715332031, 25.497406005859375, 39.70222473144531, 55.491180419921875, 12.8050537109375, 61.170806884765625, -4.289031982421875, 69.60833740234375, 36.27558898925781, 92.07009887695312, 68.33123779296875, 0.148529052734375, 20.214492797851562, 88.11373901367188, 110.85403442382812, -57.696868896484375, -38.57623291015625, 103.031982421875, 65.02741241455078, -45.3262939453125, 24.695114135742188, 89.1260986328125, -70.97900390625, 24.594993591308594, 81.753173828125, 65.53099060058594, 3.3700790405273438, 11.607780456542969, 7.171112060546875, 38.66570281982422, 19.67987060546875, 12.82640266418457, -26.155731201171875, -13.044891357421875, 72.91116333007812, 41.58222961425781, -0.349395751953125, 3.8387908935546875, 103.97415161132812, 55.94049072265625, 43.51361083984375, -18.411178588867188, 55.87556457519531, -16.852127075195312, 42.885772705078125, -14.823654174804688, 9.718414306640625, 74.6322021484375, -45.763092041015625, 20.155838012695312, 15.766632080078125, 10.1292724609375, 67.30464172363281, 87.13800048828125, 56.661659240722656, -43.00409698486328, 65.27865600585938, 53.9384765625, -4.3089599609375, 70.9525146484375, -16.27088165283203, -16.415969848632812, 35.567100524902344, 90.83648681640625, 51.429344177246094, -65.61920166015625, -38.6309814453125, 121.47616577148438, 117.03363037109375, 67.62132263183594, 19.102310180664062, 4.7747802734375, 36.18907165527344, 78.77668762207031, 43.163726806640625, 62.1549072265625, 111.6600341796875, 49.369659423828125, 73.47286987304688, -16.205337524414062, 83.52392578125, -10.44094467163086, 64.4162826538086, 58.97216796875, 107.28657531738281, -7.407661437988281, -28.637527465820312, 0.0, 6.573699951171875, -100.33256530761719, 19.864959716796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000332.npy"}
{"epoch": 0.6952879581151833, "step": 333, "batch_size": 128, "mean": 25.139057159423828, "std": 39.55549621582031, "min": -62.533233642578125, "p10": -28.93512916564941, "median": 26.517807006835938, "p90": 76.82732543945312, "max": 106.78271484375, "pos_frac": 0.71875, "sample": [51.80744171142578, -42.952239990234375, -30.403667449951172, 74.89044189453125, 81.19601440429688, -25.18292236328125, 34.75830078125, 16.998809814453125, 28.837356567382812, 67.143798828125, -34.86259460449219, 10.43768310546875, -9.853302001953125, 82.13188934326172, 78.03772735595703, 74.55502319335938, 70.26470947265625, 11.502593994140625, 54.25370788574219, 26.251251220703125, -19.020309448242188, 36.281585693359375, -0.0655670166015625, 82.38702392578125, 73.56854248046875, 30.636077880859375, 38.212982177734375, 37.6455078125, 76.86526489257812, 5.847625732421875, 17.058868408203125, 92.328857421875, -16.4322509765625, 60.170654296875, -50.662017822265625, 7.3131256103515625, 27.83197021484375, -15.133796691894531, 75.11688232421875, 52.86346435546875, 39.599090576171875, 16.66379165649414, 5.0860595703125, 3.5277252197265625, 29.955337524414062, 77.97735595703125, -4.95136833190918, 90.15725708007812, 22.325439453125, -45.48149108886719, -46.7325439453125, 56.155731201171875, 0.0, -5.18304443359375, 80.26530456542969, 16.87860107421875, 63.2979736328125, 91.00103759765625, 72.98009490966797, 52.67613220214844, 1.733154296875, 41.05304718017578, -50.940460205078125, 76.81106567382812, 41.02069091796875, -22.090469360351562, 45.83403015136719, 106.78271484375, 34.18516540527344, 26.78436279296875, -55.399871826171875, 51.061065673828125, 32.570343017578125, -43.610595703125, 8.297698974609375, 62.601348876953125, -1.772735595703125, -1.2156982421875, 20.772689819335938, 63.09095764160156, 23.74786376953125, 29.452346801757812, 41.477752685546875, 83.90011596679688, -28.305755615234375, 66.74174499511719, 70.73443603515625, -7.99029541015625, 4.60107421875, 85.27665710449219, 16.478717803955078, 53.433197021484375, -8.608619689941406, 48.10155487060547, -24.202003479003906, -15.96600341796875, 23.680908203125, 9.069168090820312, -58.43243408203125, 57.926448822021484, -1.6488037109375, 63.78135681152344, -41.907440185546875, -2.97747802734375, 41.86724853515625, 66.36859130859375, 0.920684814453125, 35.753326416015625, 76.77679443359375, 16.056427001953125, 67.72196197509766, 3.245145797729492, 64.60748291015625, 6.454315185546875, 31.49724006652832, -7.90234375, -40.93800354003906, 57.33275604248047, -8.3468017578125, 28.913360595703125, -5.259063720703125, 15.88336181640625, -62.533233642578125, 10.756797790527344, -5.913543701171875, 28.547409057617188, 19.360107421875, 1.8713436126708984], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000333.npy"}
{"epoch": 0.6973821989528796, "step": 334, "batch_size": 128, "mean": 29.347991943359375, "std": 45.76959991455078, "min": -103.81391906738281, "p10": -25.45014953613281, "median": 28.542251586914062, "p90": 92.18456115722655, "max": 131.1869354248047, "pos_frac": 0.734375, "sample": [53.088462829589844, 64.6866455078125, -16.815032958984375, 0.523193359375, 0.0, -2.783355712890625, 7.8654937744140625, -16.03656005859375, 18.678342819213867, 1.8298149108886719, 24.54656982421875, 31.043746948242188, 28.268341064453125, 66.90922546386719, -38.635040283203125, 57.919044494628906, -1.4121551513671875, 97.81283569335938, 35.127960205078125, -5.78814697265625, 55.98279571533203, 46.503509521484375, 131.1869354248047, 96.76539611816406, -28.1678466796875, 31.49884033203125, 17.633468627929688, 17.8216552734375, -0.9588394165039062, 15.27105712890625, 55.3970947265625, -9.000656127929688, 0.6927623748779297, 64.6981201171875, 73.28366088867188, 28.94427490234375, -4.109992980957031, 62.557769775390625, -70.73175048828125, 57.06201171875, -4.073921203613281, 5.600372314453125, 19.11602783203125, 17.7252197265625, 26.69373321533203, 39.49835205078125, 5.1314697265625, -42.19915771484375, 74.09521484375, -6.67340087890625, 74.33734130859375, -52.736358642578125, 84.13957214355469, 46.85107421875, 117.69601440429688, 72.42294311523438, -4.549493789672852, 42.32275390625, 8.108016967773438, 24.966339111328125, 23.761611938476562, 5.732841491699219, 122.12811279296875, 46.995330810546875, -43.06575012207031, -60.059417724609375, 34.716094970703125, -60.757720947265625, 16.557373046875, -24.866180419921875, 91.3631591796875, 1.8387908935546875, -13.801239013671875, 109.69833374023438, 76.6429443359375, 70.66622924804688, 33.51780700683594, 20.37726593017578, 11.7921142578125, 44.60041046142578, 31.745635986328125, 34.95821762084961, -7.65814208984375, 70.54864501953125, 64.70065307617188, 104.28706359863281, 34.10728454589844, 41.918609619140625, 57.47003173828125, 28.816162109375, 60.340362548828125, 67.05865478515625, 76.76695251464844, -8.3204345703125, 1.8898754119873047, 12.526382446289062, 1.6134796142578125, -13.51483154296875, 82.11111450195312, 98.8416519165039, 97.01822662353516, 9.316314697265625, -7.0549774169921875, -44.07818603515625, 90.98114776611328, -26.812744140625, 40.24485778808594, 66.5587158203125, 97.47474670410156, 79.60426330566406, -103.81391906738281, 105.72259521484375, 83.78512573242188, 0.3040790557861328, 52.782684326171875, -85.43045043945312, -3.6563262939453125, 48.4554443359375, -18.9783935546875, 24.518844604492188, 27.7354736328125, 97.65701293945312, 74.40655517578125, 31.043212890625, 33.620269775390625, -9.156387329101562, -45.48406982421875, 94.10116577148438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000334.npy"}
{"epoch": 0.6994764397905759, "step": 335, "batch_size": 128, "mean": 25.787185668945312, "std": 47.22347640991211, "min": -108.8724365234375, "p10": -37.699319458007814, "median": 20.307220458984375, "p90": 89.30942077636718, "max": 123.24911499023438, "pos_frac": 0.7109375, "sample": [-41.50213623046875, -80.7716064453125, 8.161247253417969, -25.476455688476562, 58.51678466796875, -38.528839111328125, 94.96026611328125, -11.091766357421875, 24.767593383789062, 73.90606689453125, 50.861968994140625, -70.24502563476562, 27.81292724609375, -2.0594234466552734, 0.9144973754882812, 12.16845703125, 1.796173095703125, 1.1155357360839844, -18.56353759765625, -37.34381103515625, -12.916114807128906, 97.7349853515625, -1.37884521484375, 12.390274047851562, 11.13519287109375, 4.4688720703125, 34.22773742675781, 40.17701721191406, -30.354583740234375, 77.18475341796875, -108.8724365234375, 123.24911499023438, -13.9700927734375, 51.92156982421875, 35.2109375, 112.78762817382812, 11.245231628417969, 55.181243896484375, 78.35904693603516, 9.370155334472656, 94.07113647460938, 73.20248413085938, 69.8929443359375, 60.790252685546875, 113.92938232421875, 87.00048828125, 37.44184875488281, 74.97742462158203, 65.57940673828125, 91.12857055664062, -8.252853393554688, 55.65403747558594, 4.481424331665039, 54.81560516357422, 68.109130859375, 67.69277954101562, 115.45150756835938, 46.4168701171875, 2.7875137329101562, 31.69146728515625, 20.401397705078125, -51.0567626953125, -1.8043060302734375, -16.681365966796875, -51.590301513671875, 77.09536743164062, 34.46764373779297, 88.52978515625, 43.768646240234375, 18.526519775390625, 19.90863037109375, -1.4869308471679688, 4.253326416015625, 61.8236083984375, 117.3818359375, 80.6470947265625, 6.424125671386719, -58.34532165527344, 6.169189453125, 15.869888305664062, 28.155807495117188, 46.35791015625, -8.1534423828125, 61.4901123046875, 95.77140808105469, -8.489509582519531, -32.369384765625, 1.3692779541015625, 47.02618408203125, 3.0675048828125, -27.2923583984375, 1.936737060546875, 54.2435302734375, 5.680091857910156, -3.9738235473632812, -6.246368408203125, 20.213043212890625, 13.199134826660156, 53.468841552734375, -3.0319366455078125, 96.83364868164062, -46.3021240234375, 26.20709228515625, 27.755401611328125, 75.53160095214844, 44.84403991699219, 0.0, 87.23562622070312, 98.436279296875, 74.6996078491211, -50.51470947265625, 48.08265686035156, 70.43896484375, -65.18510437011719, -39.31170654296875, 120.43569946289062, -1.2766914367675781, -1.9688491821289062, 14.660024642944336, -14.393768310546875, 12.128219604492188, 44.935630798339844, 12.099151611328125, 51.280330657958984, -56.90080261230469, 21.129409790039062, 23.321197509765625, 46.422332763671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000335.npy"}
{"epoch": 0.7015706806282722, "step": 336, "batch_size": 128, "mean": 25.532835006713867, "std": 42.76279067993164, "min": -92.98272705078125, "p10": -22.066181945800775, "median": 22.371450424194336, "p90": 83.3638313293457, "max": 155.8258056640625, "pos_frac": 0.7421875, "sample": [43.122955322265625, 15.690582275390625, 58.713836669921875, -15.639923095703125, 66.33940887451172, 34.06121826171875, -8.625640869140625, 73.59927368164062, 73.64874267578125, 48.261444091796875, 5.730232238769531, 24.03271484375, 27.535064697265625, 22.78600311279297, 2.8353309631347656, 22.84844970703125, 1.07672119140625, 103.29944610595703, 4.6812744140625, 83.43062591552734, 94.85292053222656, 22.01953125, 53.726593017578125, 70.60005187988281, -42.74639892578125, -12.622943878173828, 8.849334716796875, 7.828132629394531, 15.798934936523438, 31.1134033203125, 79.3316650390625, 27.91344451904297, 1.5327835083007812, 79.21820068359375, 66.02911376953125, 80.82669067382812, 42.19085693359375, 122.22238159179688, 11.860511779785156, 7.4027099609375, -44.47419738769531, 74.49508666992188, -28.48553466796875, 83.335205078125, 52.5928955078125, 0.25254058837890625, -0.2612495422363281, 56.9693603515625, -25.940567016601562, 22.723369598388672, 49.777198791503906, 1.823028564453125, 18.780548095703125, 45.538726806640625, 28.173324584960938, -33.30633544921875, -7.670166015625, -13.835174560546875, 30.36725616455078, -11.069313049316406, 47.082183837890625, 30.1572265625, -56.40728759765625, 47.020965576171875, -20.405731201171875, 50.131019592285156, 28.55560302734375, -35.164794921875, 44.56407165527344, 84.6033706665039, 2.372650146484375, -3.3472442626953125, 78.579833984375, 27.616790771484375, 15.421304702758789, -31.62884521484375, 2.991607666015625, 8.695541381835938, 12.531494140625, -13.634994506835938, 4.134859085083008, 32.835113525390625, 20.381454467773438, -53.34954833984375, 70.84968566894531, 26.98895263671875, 56.383453369140625, 59.295013427734375, 15.277168273925781, 77.33558654785156, -7.642669677734375, -57.726043701171875, 95.81614685058594, 90.31781005859375, 55.8209228515625, 40.052093505859375, 96.54281616210938, 7.784904479980469, 0.9803466796875, 35.54736328125, -3.5086021423339844, 90.79763793945312, 4.397552490234375, -13.251678466796875, -35.532684326171875, 155.8258056640625, 8.95599365234375, 70.09202575683594, 120.13320922851562, -92.98272705078125, 5.7734832763671875, 0.180908203125, -1.7959747314453125, 83.87284851074219, 26.607940673828125, 9.071327209472656, 55.294677734375, -9.74444580078125, -70.300537109375, 0.0, -3.85797119140625, 18.113616943359375, 27.815750122070312, 98.49112701416016, -18.551177978515625, -7.26690673828125, -15.970809936523438, 23.020538330078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000336.npy"}
{"epoch": 0.7036649214659686, "step": 337, "batch_size": 128, "mean": 28.759422302246094, "std": 43.6176643371582, "min": -74.94760131835938, "p10": -25.97733764648437, "median": 26.534698486328125, "p90": 88.35489501953124, "max": 122.367919921875, "pos_frac": 0.7265625, "sample": [25.457733154296875, -24.491912841796875, 6.0183868408203125, 7.443256378173828, 9.878570556640625, -2.416229248046875, -9.041961669921875, 107.58270263671875, 42.91864013671875, 122.367919921875, 0.7150115966796875, 45.71964645385742, 18.638046264648438, 41.0684814453125, 104.265380859375, 89.3564453125, -31.413238525390625, 7.794456481933594, 0.52911376953125, 11.923477172851562, -29.443328857421875, -37.84637451171875, -9.986465454101562, -37.3575439453125, 42.75745391845703, 1.846923828125, 9.30657958984375, 77.15821838378906, 1.0482673645019531, 81.5260009765625, 11.642578125, 80.41586303710938, 22.5625, 42.346099853515625, 74.82122802734375, 45.72709655761719, 29.31597900390625, 2.2450599670410156, 44.775146484375, 57.342620849609375, -8.264503479003906, -17.65758514404297, -12.356353759765625, 82.35696411132812, 50.496246337890625, 76.8211669921875, -3.9736328125, -50.9515380859375, -8.6201171875, 27.611663818359375, 9.430770874023438, 72.70748138427734, -0.514739990234375, 47.47874069213867, 36.945953369140625, -35.6781005859375, 13.329658508300781, 33.685302734375, 32.848907470703125, 18.76123046875, 1.1108245849609375, 6.418064117431641, 39.443817138671875, 94.8607406616211, 68.88068389892578, -18.331024169921875, 67.15824890136719, 54.51222229003906, -74.94760131835938, 16.226776123046875, 66.7529296875, 95.66439819335938, 24.339385986328125, 29.664703369140625, 59.84483337402344, 4.6667633056640625, -0.9521484375, -16.35565948486328, 91.03187561035156, 58.94684600830078, 12.529197692871094, 71.58734130859375, 36.000762939453125, 85.10906982421875, 70.7892837524414, 87.919189453125, -16.082172393798828, 52.747650146484375, -19.5518798828125, -43.066368103027344, 42.98052978515625, 80.38764953613281, 24.233718872070312, 108.8365707397461, -48.182899475097656, 97.51266479492188, 0.625640869140625, -2.369476318359375, 55.24860382080078, 28.974212646484375, 43.498634338378906, -10.71112060546875, -52.22454833984375, 97.06411743164062, -22.110565185546875, -36.195220947265625, 110.550048828125, 38.64161682128906, -0.4814720153808594, 0.9407596588134766, -18.415542602539062, 30.455345153808594, 84.703369140625, 87.9256591796875, -45.664581298828125, 100.78727722167969, -0.23468017578125, 3.9960556030273438, -21.0692138671875, 79.9002685546875, 61.34716033935547, 52.28073501586914, -33.972076416015625, 30.923080444335938, 10.19659423828125, 90.6312026977539, 86.89167785644531, 67.410400390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000337.npy"}
{"epoch": 0.7057591623036649, "step": 338, "batch_size": 128, "mean": 32.770042419433594, "std": 43.935264587402344, "min": -66.78219604492188, "p10": -19.663852691650387, "median": 25.405715942382812, "p90": 96.0266845703125, "max": 133.2611083984375, "pos_frac": 0.78125, "sample": [90.36257934570312, -18.921348571777344, 5.827545166015625, 56.73687744140625, 86.82034301757812, 95.90521240234375, 67.56853485107422, 56.839019775390625, -4.141487121582031, -25.364227294921875, 15.709354400634766, 76.12879943847656, 25.62005615234375, -1.3152923583984375, 81.79190063476562, 12.684951782226562, 67.80044555664062, 26.74945068359375, -16.092422485351562, -11.195587158203125, 94.05905151367188, 41.75196838378906, 1.923543930053711, 105.12667846679688, 14.38232421875, 39.09626770019531, 79.6502685546875, 50.538047790527344, -7.7778472900390625, 0.9074897766113281, 40.19342041015625, 6.96136474609375, -5.2308349609375, 16.524002075195312, 38.05287170410156, 67.68929290771484, 26.0693359375, 96.8861083984375, 19.034622192382812, 23.2327880859375, 42.004425048828125, 22.5526123046875, 34.828094482421875, 85.96670532226562, 101.6458740234375, 21.171920776367188, -24.642791748046875, 82.38504791259766, 44.5294189453125, 25.191375732421875, 28.170757293701172, -23.758270263671875, 99.34185791015625, -63.60334777832031, 21.436309814453125, -1.1736030578613281, -0.9798431396484375, 97.62240600585938, -32.00303649902344, 98.90731811523438, 75.452880859375, 133.2611083984375, 75.07968139648438, 40.437103271484375, 48.85186767578125, 19.174468994140625, 15.777793884277344, -7.9974212646484375, -66.78219604492188, 5.7796173095703125, 10.706661224365234, 57.63819885253906, 84.18106079101562, 5.318603515625, -21.3963623046875, -0.9131107330322266, 14.963672637939453, 86.15805053710938, 78.13914489746094, -6.453937530517578, 52.84844970703125, 1.1771221160888672, 51.13197326660156, -2.0951156616210938, 69.33694458007812, 19.0240478515625, -12.146598815917969, 3.610687255859375, 35.03021240234375, 71.24978637695312, 45.37159729003906, 104.235107421875, 14.818450927734375, 3.974456787109375, -58.60151672363281, 80.85726928710938, 17.250091552734375, 121.90255737304688, -64.06736755371094, 24.967384338378906, 15.708465576171875, -40.17134094238281, 11.738525390625, 45.796875, 13.725608825683594, 95.8831787109375, 58.78852844238281, 20.3948974609375, -31.497764587402344, 113.2401123046875, 35.71142578125, 11.94329833984375, 0.6002025604248047, 32.9033203125, -6.41326904296875, 96.31011962890625, 20.788970947265625, 49.6983642578125, 52.517486572265625, 28.152099609375, -60.459686279296875, 3.8280792236328125, 26.62786865234375, 0.059326171875, 109.3056640625, 116.63014221191406, -29.435028076171875, 76.75849914550781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000338.npy"}
{"epoch": 0.7078534031413612, "step": 339, "batch_size": 128, "mean": 33.387630462646484, "std": 45.087589263916016, "min": -99.16830444335938, "p10": -26.055696105957033, "median": 36.13612365722656, "p90": 86.32215576171875, "max": 147.10855102539062, "pos_frac": 0.7734375, "sample": [45.8521728515625, -28.211029052734375, 60.689613342285156, 68.5916748046875, 51.318626403808594, -3.996612548828125, 26.662017822265625, -13.069915771484375, 62.4044189453125, -3.200225830078125, 101.15289306640625, 14.22601318359375, 2.37506103515625, 2.4958038330078125, 59.923919677734375, -3.10107421875, 48.779327392578125, 42.744049072265625, 35.56268310546875, -48.135711669921875, 68.42851257324219, 7.816864013671875, 37.857330322265625, 61.61688232421875, 0.0, 69.9544677734375, -24.2440185546875, 29.092315673828125, 0.0, 38.568294525146484, 38.68354797363281, -27.52893829345703, 86.87774658203125, 75.0746078491211, 1.777069091796875, -70.85859680175781, 73.21990966796875, 93.07672119140625, 30.021713256835938, 10.993682861328125, 47.954742431640625, 80.52313232421875, -22.50640869140625, 62.946685791015625, 81.01251220703125, 38.48211669921875, 60.353271484375, 2.2911758422851562, 1.8206558227539062, 79.11419677734375, 61.8785400390625, 14.29949951171875, 128.890869140625, 51.86445617675781, 67.88992309570312, 73.72235107421875, 23.758453369140625, 2.6921768188476562, 78.94084167480469, 71.65411376953125, 62.03729248046875, 1.1557083129882812, -20.1588134765625, -50.51116943359375, -1.6717052459716797, 104.51388549804688, 9.107574462890625, 67.6280517578125, 74.4337158203125, 27.7572021484375, 31.751113891601562, 11.454193115234375, -58.01167297363281, 47.449371337890625, 68.95706176757812, 20.42156982421875, 91.04541015625, 77.05657958984375, -44.577301025390625, 61.214141845703125, 1.9092941284179688, 63.15166473388672, 94.15643310546875, -7.70318603515625, 86.8687744140625, 147.10855102539062, 35.00848388671875, 0.0, 48.791046142578125, 52.258758544921875, 56.98602294921875, 75.960205078125, 69.09309387207031, 89.38421630859375, 70.8587646484375, -7.500701904296875, 17.670455932617188, -4.618621826171875, 13.198211669921875, 86.087890625, 82.27105712890625, -26.117347717285156, 23.51141357421875, 16.90478515625, 6.653594970703125, 76.30982971191406, 65.25569915771484, 2.152587890625, -26.667282104492188, -26.029273986816406, 90.31356048583984, 33.79608154296875, -99.16830444335938, 66.81062316894531, 22.822967529296875, 12.415771484375, 12.699333190917969, -5.0236358642578125, 93.3377685546875, -46.721435546875, -80.80809020996094, 104.46319580078125, 36.709564208984375, 24.91973876953125, 79.9656982421875, 0.8047027587890625, -44.602622985839844, 75.8404541015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000339.npy"}
{"epoch": 0.7099476439790576, "step": 340, "batch_size": 128, "mean": 28.88920783996582, "std": 45.0467529296875, "min": -77.96893310546875, "p10": -24.286209106445312, "median": 24.85833740234375, "p90": 84.08467178344726, "max": 134.56402587890625, "pos_frac": 0.71875, "sample": [120.47947692871094, 36.55811309814453, -20.568374633789062, 78.05708312988281, 1.7209320068359375, 79.15739440917969, -24.253021240234375, -14.057525634765625, -41.14752197265625, 89.89387512207031, -45.49957275390625, 66.63746643066406, -53.489105224609375, 5.017799377441406, 0.0, 71.6800765991211, 17.878814697265625, -24.022186279296875, 61.358978271484375, 97.1165771484375, 84.37181854248047, 19.722564697265625, -14.118392944335938, 65.36611938476562, -34.77549743652344, -15.840507507324219, -20.63531494140625, 72.5542221069336, -2.326416015625, 83.96160888671875, 116.36651611328125, 3.0346603393554688, 47.52587890625, 86.81790161132812, 24.72332763671875, -6.60357666015625, 1.313669204711914, -48.197479248046875, 43.932647705078125, 10.731658935546875, 79.71746826171875, 74.17599487304688, -45.77484130859375, 30.49932861328125, 26.050674438476562, 83.57188415527344, -15.825592041015625, 37.38569641113281, 10.592926025390625, -13.531082153320312, 52.8387451171875, 103.49407958984375, -6.518209457397461, -53.2760009765625, 7.321630477905273, -5.187042236328125, 72.91679382324219, 68.093994140625, 26.937835693359375, -12.861190795898438, 9.944183349609375, -77.96893310546875, 93.90129089355469, 0.0, 114.88064575195312, 1.451934814453125, 44.907958984375, 10.372688293457031, 78.943603515625, 30.35699462890625, 4.767242431640625, -2.112751007080078, 74.28945922851562, -47.47850036621094, 75.21546936035156, 28.167236328125, 44.538848876953125, -45.712432861328125, 93.89581298828125, 12.18389892578125, 69.72415161132812, 134.56402587890625, -44.4423828125, 18.774505615234375, -10.25927734375, 52.412811279296875, 41.4146728515625, 11.599868774414062, 58.882293701171875, 20.251312255859375, 44.02581787109375, -7.666543960571289, 37.78350067138672, -9.2374267578125, 68.744140625, 60.549072265625, -10.648361206054688, 4.50164794921875, 33.50958251953125, 42.947601318359375, 132.583740234375, 6.732032775878906, 77.18624877929688, 5.5629425048828125, 56.24317932128906, -24.3636474609375, 13.515022277832031, -40.615478515625, 78.73117065429688, 66.42190551757812, 74.4740982055664, 31.343048095703125, 5.729732513427734, 54.52653503417969, 92.51165771484375, 83.50550842285156, 8.978240966796875, 18.55097198486328, 12.825286865234375, 58.40899658203125, 41.19010925292969, 24.99334716796875, -13.487548828125, 21.14468765258789, 29.971054077148438, 12.180694580078125, 75.95614624023438, -10.02457046508789], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000340.npy"}
{"epoch": 0.7120418848167539, "step": 341, "batch_size": 128, "mean": 23.485198974609375, "std": 47.08413314819336, "min": -79.69271850585938, "p10": -36.21983337402343, "median": 19.640132904052734, "p90": 83.5292251586914, "max": 139.6060791015625, "pos_frac": 0.6875, "sample": [4.3807373046875, -1.72601318359375, 38.841522216796875, 78.11036682128906, 61.10389709472656, -21.508941650390625, 52.808135986328125, 15.21527099609375, 12.014572143554688, 0.345458984375, 68.16552734375, 28.552444458007812, 83.31210327148438, 10.872669219970703, 108.77008056640625, 95.50390625, -32.022979736328125, 25.670318603515625, 19.57879638671875, 39.487884521484375, 72.1014404296875, 25.3922119140625, 70.78175354003906, -58.64744567871094, 103.52923583984375, 32.64097595214844, 5.237464904785156, 11.0396728515625, 7.81732177734375, -6.038330078125, 9.53118896484375, 25.50390625, -9.097564697265625, -3.478973388671875, -61.1483154296875, 14.719024658203125, -46.42486572265625, -38.96510314941406, 0.1417369842529297, 31.773834228515625, 16.48895263671875, 39.168338775634766, 79.82070922851562, 35.612945556640625, 25.916656494140625, 6.32647705078125, 81.66596984863281, 48.262840270996094, -2.8661975860595703, 63.19721603393555, 51.82322692871094, 84.03584289550781, 115.45611572265625, -17.5203857421875, 76.75643920898438, 55.0399169921875, 57.655120849609375, 20.98455810546875, 18.287368774414062, -14.77294921875, 32.643798828125, -59.991668701171875, 139.6060791015625, -6.720060348510742, 76.09228515625, -61.05033874511719, -65.91824340820312, 75.56086730957031, 90.94047546386719, 0.758575439453125, 29.01641845703125, 72.90396118164062, 65.01324462890625, -5.2503814697265625, 59.78410339355469, 82.11536407470703, 46.4349365234375, 4.2513885498046875, 25.775238037109375, -71.68255615234375, 41.922760009765625, -5.620819091796875, -9.433609008789062, -10.305694580078125, 117.97711181640625, 22.1737060546875, -11.543670654296875, 39.140594482421875, 2.3519439697265625, 91.10506439208984, -28.151336669921875, 48.67626953125, 86.91793823242188, 65.4437484741211, -46.049400329589844, -9.65838623046875, -27.627883911132812, -0.7329311370849609, -9.44580078125, 16.26580810546875, -12.84356689453125, 73.53607177734375, 21.945068359375, 2.4801025390625, -64.67161560058594, 20.81464385986328, 24.363922119140625, -35.04328918457031, -79.69271850585938, -24.527999877929688, 4.83013916015625, -48.93035888671875, -17.83868408203125, 95.068603515625, -78.80415344238281, -3.766021728515625, 77.30477142333984, 88.8465576171875, 75.35049438476562, 53.09327697753906, 81.2203369140625, 3.2996826171875, 19.70146942138672, -19.0994873046875, 117.13970947265625, 12.4847412109375, -9.399322509765625, 10.330057144165039], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000341.npy"}
{"epoch": 0.7141361256544503, "step": 342, "batch_size": 128, "mean": 28.615371704101562, "std": 41.86682891845703, "min": -111.43881225585938, "p10": -16.985128784179686, "median": 19.829612731933594, "p90": 84.5992950439453, "max": 124.81610107421875, "pos_frac": 0.7421875, "sample": [-7.48248291015625, 13.392738342285156, 18.516571044921875, 15.158416748046875, 44.0242919921875, 116.0223388671875, 9.830970764160156, 30.007736206054688, -0.5382156372070312, 33.330047607421875, -21.796371459960938, 1.695037841796875, 1.9479522705078125, 70.08793640136719, 51.75018310546875, 74.83477783203125, -4.074516296386719, 37.6019287109375, -27.89287567138672, 3.818939208984375, 43.34907531738281, 7.75946044921875, 9.51416015625, 21.989425659179688, -25.080276489257812, 14.5462646484375, 10.312744140625, -36.82373046875, 37.522552490234375, 69.025634765625, 3.9280929565429688, 58.74658203125, 43.77550506591797, 53.666717529296875, -3.99346923828125, 11.891265869140625, 15.677909851074219, -8.308801651000977, -17.909881591796875, 24.424285888671875, 39.43890380859375, 106.76034545898438, 7.919342041015625, -17.11163330078125, -13.934982299804688, -15.18060302734375, -27.866424560546875, 44.0128173828125, 33.7496337890625, 7.383880615234375, 61.82527542114258, 0.8650054931640625, -17.0150146484375, 87.91595458984375, 11.454879760742188, 92.23989868164062, 84.08963012695312, -0.39935302734375, 52.09808349609375, 65.09124755859375, 94.56434631347656, 85.78851318359375, 59.283416748046875, 69.2791976928711, 56.79560852050781, -60.13688278198242, -6.6495208740234375, 0.0, -9.85205078125, 27.06005859375, 100.6761474609375, 4.086875915527344, 83.58683776855469, 4.433737754821777, 31.452178955078125, 51.2855224609375, 81.69631958007812, 30.361732482910156, 88.61083984375, 3.3529529571533203, -9.71282958984375, 69.3651123046875, 15.570960998535156, 52.059173583984375, -57.43548583984375, -5.008247375488281, -111.43881225585938, -14.577667236328125, 55.031829833984375, 54.159637451171875, 75.4925537109375, 69.33143615722656, 2.8664627075195312, 9.187713623046875, 8.99273681640625, 64.30026245117188, 55.56927490234375, 21.142654418945312, 26.521209716796875, -32.789031982421875, 12.37225341796875, 69.58193969726562, -16.972320556640625, 124.81610107421875, 10.289642333984375, 69.69866943359375, -10.966552734375, 13.03289794921875, 65.34417724609375, -46.82630157470703, 91.75616455078125, 57.203521728515625, 26.821609497070312, 73.90255737304688, 58.50982666015625, 85.93072509765625, 18.160018920898438, 15.074066162109375, 67.81423950195312, 80.68435668945312, -9.432472229003906, -9.782943725585938, -16.884902954101562, 111.515869140625, 65.89608764648438, -9.119270324707031, 115.7825927734375, 6.67462158203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000342.npy"}
{"epoch": 0.7162303664921466, "step": 343, "batch_size": 128, "mean": 26.74979591369629, "std": 44.154823303222656, "min": -87.570556640625, "p10": -25.154318237304686, "median": 20.54275131225586, "p90": 85.8417236328125, "max": 127.84689331054688, "pos_frac": 0.71875, "sample": [-21.590606689453125, 77.05072021484375, 75.39913940429688, 2.609516143798828, 32.127593994140625, 28.25762939453125, 33.634674072265625, 18.14532470703125, -26.99169921875, 22.165771484375, 86.61529541015625, -28.33233642578125, -12.76080322265625, -23.845916748046875, 47.64056396484375, 88.72498321533203, 83.10128784179688, 33.5545654296875, -7.840461730957031, -6.277704238891602, 96.38510131835938, 13.135711669921875, 93.27474975585938, 82.67069244384766, 50.62548065185547, 49.52116394042969, 6.275291442871094, 57.302001953125, 14.63555908203125, 68.95039367675781, 37.68084716796875, -57.39014434814453, 19.999908447265625, -6.374786376953125, -38.56077575683594, 17.515548706054688, 8.690948486328125, 81.08865356445312, -24.6732177734375, 0.569580078125, -18.19891357421875, 100.94363403320312, 28.12127685546875, 32.16339111328125, 0.0, -48.26953125, 23.29461669921875, 78.99494934082031, 40.40052032470703, 120.76531982421875, -71.00152587890625, 6.228324890136719, 30.286184310913086, 13.90194320678711, 79.175048828125, -28.822509765625, 59.34547424316406, 53.605987548828125, -0.6213150024414062, 31.2510986328125, 68.37738800048828, 2.3901214599609375, -4.679573059082031, 3.573516845703125, 67.00935363769531, 82.16352844238281, -11.751861572265625, 78.75778198242188, 11.87158203125, 45.25823974609375, 41.972320556640625, -14.246231079101562, 41.44158935546875, -8.108978271484375, 0.5792388916015625, 30.33403778076172, -6.053924560546875, 126.9052734375, 70.61306762695312, 78.40721130371094, 21.085594177246094, 26.833419799804688, -15.728179931640625, 3.731201171875, 67.90077209472656, 31.801864624023438, 3.342437744140625, 28.672943115234375, 7.90087890625, -22.80157470703125, 7.599506378173828, 85.51019287109375, 80.36911010742188, 127.84689331054688, 13.5806884765625, -11.537540435791016, 71.01176452636719, 39.879730224609375, 77.0601806640625, 14.282440185546875, 91.57109069824219, 83.42936706542969, 7.481903076171875, 59.83319091796875, -87.570556640625, 88.27294921875, -13.670440673828125, -4.528419494628906, -26.276885986328125, -34.135345458984375, 6.461368560791016, 67.11831665039062, 87.22281646728516, 3.39947509765625, -38.2705078125, -18.643218994140625, -5.672088623046875, -34.129486083984375, 23.224472045898438, 103.19564819335938, 41.60173797607422, 0.8985366821289062, 17.452911376953125, 11.459457397460938, 6.62255859375, 92.95352172851562, -5.9610595703125, -66.79771423339844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000343.npy"}
{"epoch": 0.7183246073298429, "step": 344, "batch_size": 128, "mean": 30.148412704467773, "std": 44.622047424316406, "min": -83.9786376953125, "p10": -17.386546707153318, "median": 22.869007110595703, "p90": 94.06401596069335, "max": 121.24261474609375, "pos_frac": 0.7421875, "sample": [21.694580078125, -4.3782806396484375, 69.51004028320312, 32.81500244140625, -5.662506103515625, 81.20138549804688, 80.11557006835938, 22.228179931640625, 75.9688720703125, 5.91204833984375, 63.0635986328125, 9.067550659179688, 62.27267837524414, -19.48385238647461, 52.85076904296875, 94.0340576171875, 23.024932861328125, 32.39276123046875, -83.9786376953125, 63.399658203125, -36.404266357421875, 9.752975463867188, 4.1002197265625, -12.9420166015625, 114.66937255859375, 28.175872802734375, -11.581012725830078, 102.48019409179688, 81.05241394042969, 11.654083251953125, 83.67201232910156, -25.93011474609375, 7.973175048828125, 67.25045776367188, -73.37014770507812, 36.4290771484375, 73.02412414550781, 62.861480712890625, 9.816802978515625, -12.452934265136719, -82.22566223144531, 5.986907958984375, 10.786392211914062, -15.15191650390625, -3.6500244140625, 11.982742309570312, -43.51031494140625, 45.37629699707031, 95.45748901367188, 33.458953857421875, -16.487701416015625, -16.040328979492188, -0.4215087890625, -32.46863555908203, 41.31982421875, 32.979408264160156, 94.13391876220703, -4.302154541015625, 121.24261474609375, 101.53907012939453, 80.31405639648438, 80.85005187988281, 110.85195922851562, 66.4193115234375, 88.31658935546875, -2.4557266235351562, 3.340118408203125, 75.94644165039062, -4.65985107421875, -10.585693359375, -36.704132080078125, 13.39190673828125, 13.3328857421875, 116.19168090820312, 34.528167724609375, 43.81980895996094, 21.68572998046875, 44.80828857421875, 9.04046630859375, 50.99659729003906, 37.46111297607422, 12.563873291015625, 32.8138427734375, 80.21934509277344, 7.4779052734375, 23.38934326171875, -3.234344482421875, 81.43869018554688, 11.914230346679688, 22.71308135986328, 119.75213623046875, 38.281036376953125, 14.768470764160156, -21.675491333007812, -5.4926910400390625, 97.388427734375, 44.77264404296875, -1.522796630859375, 0.049957275390625, -23.885116577148438, -5.094255447387695, 26.620498657226562, 14.461273193359375, 110.8037109375, -9.16876220703125, -31.5880126953125, 83.62013244628906, 15.959075927734375, 10.409507751464844, 14.339645385742188, 2.1629638671875, 49.65272521972656, 3.544677734375, 89.72854614257812, 50.30130386352539, 32.73468017578125, 35.706298828125, 9.705703735351562, 27.63226318359375, 94.92276000976562, 70.56964111328125, 21.343338012695312, 45.928375244140625, 91.99048614501953, 43.11767578125, -11.968002319335938, 96.27456665039062, -69.621826171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000344.npy"}
{"epoch": 0.7204188481675393, "step": 345, "batch_size": 128, "mean": 27.368812561035156, "std": 37.76881408691406, "min": -77.35882568359375, "p10": -13.955107116699217, "median": 18.939193725585938, "p90": 79.03674774169922, "max": 107.56903076171875, "pos_frac": 0.7578125, "sample": [28.2318115234375, 62.76799774169922, 55.366455078125, -39.27326965332031, 104.17266845703125, 62.267181396484375, 5.15496826171875, -32.56280517578125, 12.766937255859375, -2.4595260620117188, 51.733673095703125, 8.013969421386719, -2.6832504272460938, 1.0973968505859375, 35.95501708984375, 26.893035888671875, 87.33575439453125, 11.839996337890625, 9.586883544921875, -4.448272705078125, 20.57318115234375, -17.0699462890625, 60.44586944580078, 78.60786437988281, 2.5254783630371094, 46.6328125, 74.93606567382812, 17.649932861328125, -57.738739013671875, 2.8592529296875, -5.633323669433594, 9.220794677734375, -3.2221755981445312, 22.242050170898438, -2.8580322265625, 42.0992431640625, 46.21484375, 3.485992431640625, 16.0523681640625, 20.019744873046875, 28.391082763671875, -6.351776123046875, -24.233867645263672, 70.81634521484375, 9.1043701171875, 66.96199035644531, 19.139923095703125, 0.3170642852783203, 34.375144958496094, 107.56903076171875, 74.10340118408203, 59.0006103515625, 35.6170654296875, 60.99113464355469, 95.90843200683594, 18.73846435546875, 21.5877685546875, -15.106460571289062, 10.982177734375, -5.109672546386719, 92.87322998046875, 36.174835205078125, 25.513107299804688, 22.037948608398438, 17.150375366210938, -30.663925170898438, -77.35882568359375, 9.233139038085938, -22.65802001953125, -21.97607421875, 94.96305847167969, 106.529541015625, 80.61433410644531, 17.964019775390625, 29.812660217285156, -10.391151428222656, 106.00775146484375, -20.847640991210938, 17.531524658203125, -2.0380706787109375, 77.86468505859375, 83.85934448242188, 47.021018981933594, -49.053863525390625, 61.853546142578125, 20.369293212890625, -24.828292846679688, 97.96096801757812, 48.3839111328125, 38.361045837402344, 92.41987609863281, 66.74679565429688, 13.5523681640625, 10.27386474609375, 45.54100036621094, 10.791900634765625, 45.536781311035156, 71.38663482666016, 71.04678344726562, -11.98126220703125, 80.0374755859375, -13.461669921875, 72.01396942138672, 12.080245971679688, 9.245361328125, 31.36810302734375, 0.8449592590332031, 15.337066650390625, 10.83447265625, -6.4129486083984375, -4.014923095703125, 65.93035125732422, 15.48114013671875, 28.87078857421875, 75.98663330078125, 72.1435546875, 9.698272705078125, 29.694957733154297, 0.0, 14.175796508789062, -2.8907241821289062, -6.798980712890625, 7.514350891113281, -0.29000282287597656, 74.85856628417969, 5.915771484375, 55.90849685668945, 39.95643615722656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000345.npy"}
{"epoch": 0.7225130890052356, "step": 346, "batch_size": 128, "mean": 29.614782333374023, "std": 41.91163635253906, "min": -93.80303955078125, "p10": -15.236588287353515, "median": 29.0616397857666, "p90": 85.11564636230467, "max": 118.69241333007812, "pos_frac": 0.75, "sample": [-33.616943359375, -29.036767959594727, -1.0464038848876953, -53.06758117675781, 32.22636413574219, 40.1199951171875, 92.933837890625, 99.7269287109375, 49.192955017089844, 5.1579437255859375, 74.66139221191406, -14.844772338867188, -12.55487060546875, 28.47900390625, 78.08660888671875, -51.275054931640625, -14.204193115234375, 9.51727294921875, 23.05169677734375, 88.87771606445312, 12.338653564453125, 13.134292602539062, 11.317916870117188, 8.7978515625, 29.644275665283203, 68.1055908203125, 8.456085205078125, 115.693603515625, -10.4013671875, -9.860687255859375, 22.125030517578125, 20.045997619628906, -46.5577392578125, 16.531097412109375, 1.094085693359375, 90.5730209350586, 31.494903564453125, 87.97445678710938, 11.171607971191406, 22.17413330078125, 52.71966552734375, 15.110916137695312, -75.78558349609375, 4.286376953125, -19.139541625976562, 33.585205078125, 3.910614013671875, -7.9609527587890625, -1.2464752197265625, 53.10560989379883, 61.434417724609375, 14.971405029296875, 70.82666015625, -13.383983612060547, 17.828948974609375, 49.21412658691406, 100.99942016601562, 32.26490783691406, -3.6147384643554688, 58.413848876953125, 99.8428955078125, 98.2774658203125, 32.29827880859375, 5.925846099853516, 22.676803588867188, 22.67536163330078, 99.83172607421875, -8.341873168945312, 55.531097412109375, 53.789031982421875, 67.6112060546875, 30.16217041015625, 70.93731689453125, 104.29913330078125, 73.36669921875, 66.49819946289062, -37.90919494628906, 75.28277587890625, 43.97808837890625, 34.7603759765625, 40.229736328125, -55.61163330078125, 88.60342407226562, 8.104034423828125, -3.9496688842773438, -6.118846893310547, -16.15082550048828, 2.1208324432373047, -37.007835388183594, 60.17254638671875, 52.879425048828125, 31.0615234375, -8.437797546386719, 83.89044189453125, 0.29126739501953125, -5.217704772949219, 63.54469299316406, -93.80303955078125, 0.2122344970703125, 76.63580322265625, 4.005331039428711, 6.9677734375, 58.063232421875, 55.55891418457031, 78.24217224121094, 41.092987060546875, 56.45831298828125, 46.746124267578125, 22.593368530273438, 50.54144287109375, 0.0, 31.297348022460938, -4.229248046875, 24.02642822265625, 75.23892211914062, 61.28034973144531, -0.2376556396484375, 80.395263671875, 31.932449340820312, -22.416305541992188, 68.53987121582031, 31.344329833984375, -7.7122955322265625, 67.18499755859375, 81.64051818847656, 66.94151306152344, 9.782941818237305, 118.69241333007812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000346.npy"}
{"epoch": 0.724607329842932, "step": 347, "batch_size": 128, "mean": 24.324626922607422, "std": 41.67538833618164, "min": -107.46853637695312, "p10": -24.849569702148436, "median": 19.703716278076172, "p90": 83.05052032470702, "max": 102.74847412109375, "pos_frac": 0.7265625, "sample": [9.195709228515625, 56.74444580078125, 64.47271728515625, 9.567657470703125, 20.42620849609375, 26.90850067138672, 3.2769927978515625, 87.58264923095703, 8.227134704589844, 55.85438537597656, -107.46853637695312, 22.728759765625, 75.01390075683594, 30.612838745117188, 97.10737609863281, -39.18479919433594, -16.29852294921875, -2.370025634765625, 73.68812561035156, 91.90657043457031, 50.980224609375, -28.74609375, 46.48297119140625, 7.497016906738281, 11.125335693359375, 30.5887451171875, 82.07720947265625, 19.28778076171875, 19.70630645751953, -1.1940841674804688, -1.203369140625, 35.659393310546875, 94.88539123535156, 9.727828979492188, 77.50091552734375, 22.208362579345703, 74.21241760253906, -3.6199111938476562, 37.47882080078125, 7.086601257324219, 40.697174072265625, 77.68072509765625, 101.671630859375, 19.701126098632812, 17.24871826171875, 32.28948974609375, -9.625808715820312, 13.66400146484375, 16.4661865234375, -39.147979736328125, -2.691375732421875, 4.045806884765625, 37.77171325683594, -32.17005920410156, 30.432830810546875, 43.892578125, -3.58099365234375, 76.60839080810547, 51.723548889160156, -25.849998474121094, 49.23712158203125, 32.814422607421875, -1.0173797607421875, -5.907217025756836, 19.712181091308594, 55.54399490356445, 57.45782470703125, 33.12413024902344, 99.79708862304688, 23.673084259033203, 6.239227294921875, -73.25497436523438, -37.8660888671875, 4.344097137451172, 61.852630615234375, -8.808622360229492, -11.550590515136719, -8.290760040283203, -14.571548461914062, 5.885009765625, 5.582862854003906, 44.1163330078125, 88.07443237304688, -12.96661376953125, 11.737041473388672, -82.08415985107422, 56.981414794921875, 102.74847412109375, 25.67962646484375, 33.11236572265625, -8.620391845703125, 55.774932861328125, -42.453857421875, 71.1905746459961, 12.744789123535156, -3.9039306640625, 49.76936340332031, -4.919374465942383, 68.176513671875, -30.5472412109375, -103.76531982421875, -24.621978759765625, 85.6390380859375, 87.47482299804688, 0.0, 11.99481201171875, 9.170745849609375, 27.13739013671875, 45.56097412109375, 90.91253662109375, 19.10125732421875, 8.519432067871094, 70.38009643554688, 67.45257568359375, -8.7655029296875, 12.315948486328125, 14.501983642578125, -8.35760498046875, 69.97280883789062, 0.199798583984375, 30.03759765625, 99.8009033203125, 85.32157897949219, 41.124176025390625, 2.4754257202148438, -25.380615234375, 53.32806396484375, 12.852561950683594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000347.npy"}
{"epoch": 0.7267015706806282, "step": 348, "batch_size": 128, "mean": 22.718242645263672, "std": 40.482669830322266, "min": -106.15371704101562, "p10": -17.537298583984374, "median": 10.546875, "p90": 82.90274047851562, "max": 112.28919982910156, "pos_frac": 0.640625, "sample": [15.46466064453125, -0.7042388916015625, -2.233989715576172, 4.8692626953125, -1.92181396484375, 2.962860107421875, 74.29554748535156, -7.006725311279297, 38.741943359375, 66.64801788330078, -2.961578369140625, 79.51136779785156, 22.45880126953125, 61.430755615234375, 23.582229614257812, 10.33404541015625, 106.44064331054688, 71.02581787109375, -24.68383026123047, 0.0, 23.3497314453125, 36.131683349609375, 62.614898681640625, 14.440877914428711, -22.215911865234375, 100.60054016113281, -2.5114212036132812, -7.167205810546875, -5.8909454345703125, 51.662109375, 36.54710388183594, 43.568302154541016, -10.264175415039062, -5.9513702392578125, 53.127197265625, -14.74371337890625, 88.03643798828125, -6.442045211791992, 61.254905700683594, 105.66424560546875, 5.876630783081055, -11.0089111328125, -54.80316162109375, 22.205596923828125, -2.448944091796875, 82.28009033203125, 25.564361572265625, -1.38970947265625, 19.519317626953125, 52.248008728027344, 19.190059661865234, 51.448028564453125, -6.990631103515625, 0.6713027954101562, -7.518707275390625, 15.951751708984375, 63.049774169921875, 105.15380859375, 91.17941284179688, -48.24794006347656, -5.79150390625, -6.7247314453125, -28.724090576171875, 63.12664794921875, 1.4315052032470703, 47.087921142578125, 21.003570556640625, -0.6013526916503906, 3.7299041748046875, 31.049224853515625, 69.08519744873047, -39.9404296875, -63.363250732421875, -4.04315185546875, -27.240875244140625, -11.3814697265625, -17.421905517578125, 19.998748779296875, 50.410186767578125, 6.72344970703125, 88.97140502929688, 5.05078125, 53.342742919921875, 112.28919982910156, 0.3330078125, 95.54718017578125, 0.5405426025390625, -17.91326904296875, -20.2205810546875, 18.26934814453125, -8.569219589233398, 80.04612731933594, 67.009765625, -8.537002563476562, 110.56979370117188, -17.806549072265625, 15.885292053222656, -0.9854907989501953, -8.310104370117188, 7.11749267578125, 69.49203491210938, 9.080337524414062, 58.860107421875, -8.692403793334961, 5.894813537597656, -1.1311187744140625, 84.3555908203125, 42.058319091796875, 5.735626220703125, 85.84469604492188, 39.87409973144531, -11.99481201171875, 32.509315490722656, -26.543243408203125, 6.1484375, -4.3460693359375, 88.08641052246094, 27.5858154296875, 64.25199127197266, 28.782211303710938, 64.466064453125, 3.6527252197265625, -7.832435607910156, 10.75970458984375, 9.93316650390625, 19.981491088867188, -106.15371704101562, 74.23651123046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000348.npy"}
{"epoch": 0.7287958115183246, "step": 349, "batch_size": 128, "mean": 31.453781127929688, "std": 44.0126838684082, "min": -73.6014404296875, "p10": -21.252514648437497, "median": 27.242767333984375, "p90": 88.26206359863279, "max": 129.13525390625, "pos_frac": 0.7734375, "sample": [44.66777038574219, 63.76214599609375, 2.390167236328125, -66.33212280273438, -44.46270751953125, 1.3970146179199219, 129.13525390625, 45.768890380859375, -33.71244812011719, 45.8760986328125, 33.32728576660156, -44.57078552246094, 52.44007873535156, 92.19677734375, 33.227508544921875, 84.60995483398438, 96.59426879882812, 4.8570098876953125, 38.381683349609375, 102.03839111328125, 53.06365966796875, 1.05389404296875, -15.4384765625, -20.889892578125, -3.8398895263671875, 19.570449829101562, -28.0408935546875, -2.741567611694336, 103.75604248046875, 85.56629943847656, 30.119384765625, 74.35035705566406, 71.03231811523438, -30.12884521484375, 27.723419189453125, 23.709091186523438, 107.31834411621094, 4.0906982421875, 11.77484130859375, 38.541046142578125, 10.876152038574219, -1.0250511169433594, 61.14520263671875, -0.69775390625, 15.668289184570312, 51.726318359375, -9.515869140625, 126.83743286132812, 37.902618408203125, 9.589630126953125, -31.776763916015625, -2.988321304321289, -19.14031982421875, 75.13644409179688, 79.90748596191406, 46.36506652832031, 4.1478118896484375, 13.4521484375, -49.93487548828125, 43.76557922363281, 0.0, 111.2322998046875, -6.85198974609375, 44.18487548828125, 76.66799926757812, 26.70550537109375, 75.56658935546875, 97.28857421875, 71.85189819335938, -0.2662334442138672, 75.9251708984375, -22.0986328125, -55.453529357910156, 125.98681640625, 2.0198974609375, 26.762115478515625, 3.97845458984375, 63.497314453125, 7.279655456542969, 9.129341125488281, 15.36395263671875, 48.198997497558594, 78.9866943359375, 65.68145751953125, 100.721923828125, 7.816162109375, 31.862045288085938, 7.32392692565918, 81.8013916015625, 35.07452392578125, 40.91560363769531, 29.6004638671875, 80.4132080078125, 4.500064849853516, -17.77941131591797, 4.8837127685546875, 81.91459655761719, 55.49724578857422, 17.143951416015625, -59.40447998046875, 25.453369140625, 31.010284423828125, 20.866806030273438, 5.455169677734375, -73.6014404296875, 10.21954345703125, -2.973052978515625, 4.542375564575195, 15.260429382324219, 70.7587890625, 9.68060302734375, 73.76895141601562, 17.108428955078125, 69.32444763183594, 77.01634216308594, 86.65411376953125, 92.01394653320312, -10.7890625, 48.207733154296875, 104.0601806640625, 7.81591796875, 29.738433837890625, 62.619781494140625, 68.65338134765625, -36.55780029296875, -2.9573974609375, 10.65850830078125, 74.52937316894531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000349.npy"}
{"epoch": 0.7308900523560209, "step": 350, "batch_size": 128, "mean": 28.658565521240234, "std": 45.63983154296875, "min": -87.86767578125, "p10": -32.424890136718744, "median": 27.331382751464844, "p90": 90.19956970214842, "max": 118.44952392578125, "pos_frac": 0.78125, "sample": [89.07809448242188, 22.523147583007812, -0.8789520263671875, -0.4496307373046875, 8.453739166259766, 32.73643493652344, 10.035934448242188, 116.87075805664062, 3.4101829528808594, 62.14190673828125, 28.24005126953125, 11.489669799804688, 80.57138061523438, -1.115142822265625, 19.849151611328125, 85.78390502929688, 92.6373291015625, 92.70199584960938, 16.512855529785156, -28.622220993041992, 82.83644104003906, 13.310707092285156, 2.61260986328125, 58.04035949707031, 55.80677032470703, 75.05474853515625, 34.96046447753906, 27.203948974609375, 118.44952392578125, 99.02584838867188, 48.831634521484375, 35.50018310546875, 11.335956573486328, 13.293722152709961, -30.594772338867188, 59.95420837402344, 10.7061767578125, 11.103630065917969, 48.99737548828125, 35.5537109375, 14.19085693359375, 47.57643127441406, 46.267242431640625, -50.015350341796875, 85.16986083984375, -5.802886962890625, 6.229766845703125, 28.631576538085938, -87.86767578125, 9.692794799804688, 28.781417846679688, 64.1051025390625, 114.49407958984375, 52.55756378173828, 17.24322509765625, 29.369491577148438, 87.70208740234375, 71.34457397460938, 36.682647705078125, 110.64682006835938, 56.30860900878906, -31.1083984375, 93.94331359863281, 5.23236083984375, -4.833709716796875, -19.538848876953125, 29.765060424804688, 28.99810791015625, -72.04931640625, 6.838142395019531, 59.50645446777344, 32.7559814453125, 2.9465408325195312, -51.150115966796875, -6.4521484375, 61.04608154296875, -5.586158752441406, 16.93730926513672, 12.92919921875, 42.65528869628906, 108.34442138671875, 80.17015075683594, 37.936927795410156, 20.165863037109375, 24.788238525390625, 63.33821105957031, 16.38092041015625, 99.64935302734375, -44.05613708496094, 5.603973388671875, 97.89382934570312, 27.458816528320312, 18.588043212890625, 25.943328857421875, 82.13668823242188, 58.38243865966797, 19.471038818359375, 83.89371490478516, 115.42800903320312, 80.57122802734375, 76.70657348632812, -46.3421630859375, 40.4993896484375, 28.347721099853516, -63.44622802734375, -51.499244689941406, -35.4967041015625, 4.5306854248046875, -61.291839599609375, 31.412078857421875, -78.74197387695312, -43.482452392578125, 13.230072021484375, 85.83743286132812, 31.475311279296875, 30.64117431640625, -1.672454833984375, 3.839263916015625, 97.70242309570312, -49.34112548828125, 1.910400390625, 25.621475219726562, -12.0013427734375, -7.5347900390625, 89.15481567382812, -20.99829864501953, 33.336517333984375, 3.7137222290039062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000350.npy"}
{"epoch": 0.7329842931937173, "step": 351, "batch_size": 128, "mean": 21.681129455566406, "std": 46.77170944213867, "min": -107.22329711914062, "p10": -23.96990051269531, "median": 12.209686279296875, "p90": 87.18606262207031, "max": 132.62124633789062, "pos_frac": 0.6484375, "sample": [0.0, 69.0810546875, 9.4080810546875, 117.42178344726562, -1.56561279296875, 103.92568969726562, 11.7762451171875, -20.495752334594727, 28.149688720703125, -0.82208251953125, 26.489486694335938, -23.895111083984375, 10.420501708984375, 2.1189117431640625, 83.27718353271484, -36.72688293457031, 36.8159065246582, 55.96977996826172, -26.395034790039062, 15.2247314453125, 16.929039001464844, 69.634033203125, -13.00714111328125, 120.3486328125, 74.00779724121094, 90.73947143554688, 5.215723037719727, -12.6864013671875, 14.200347900390625, 5.2640380859375, 89.00898742675781, 98.92295837402344, 28.514198303222656, 8.599559783935547, -0.00067138671875, -39.37554931640625, -1.113260269165039, 76.161865234375, -11.956809997558594, -32.77096176147461, -50.409698486328125, 62.981964111328125, 8.475860595703125, 39.5889892578125, -17.552490234375, 55.87721252441406, 5.374961853027344, 17.80804443359375, -14.506011962890625, 18.906158447265625, -22.096084594726562, -15.7044677734375, 12.64312744140625, -5.68316650390625, 1.5167236328125, -1.571197509765625, 81.7012939453125, 0.013427734375, -6.73260498046875, 14.95025634765625, 93.32534790039062, 101.88839721679688, 14.253755569458008, 125.1341552734375, 66.62181091308594, -13.26068115234375, 55.474578857421875, 6.930145263671875, 81.3817138671875, -39.52583312988281, 1.5230560302734375, 70.51293182373047, -18.324188232421875, 3.6112518310546875, 96.3569107055664, -97.51107025146484, 37.41583251953125, -11.39190673828125, 24.253036499023438, 36.50537109375, 0.0, -5.29571533203125, 17.140228271484375, 80.30062866210938, 77.62132263183594, -0.5595903396606445, -0.313262939453125, 35.354637145996094, 58.8514404296875, -4.31488037109375, 22.250762939453125, 4.49151611328125, 50.040252685546875, -9.542034149169922, 0.0, 86.83740234375, 79.9027099609375, -3.79132080078125, 14.182182312011719, 9.5074462890625, 6.2918548583984375, -107.22329711914062, 56.23096466064453, 74.89140319824219, 54.99005126953125, -12.089202880859375, 132.62124633789062, 18.9842529296875, 85.94418334960938, -3.2866363525390625, 0.8695354461669922, 21.932113647460938, -10.2315673828125, 24.325241088867188, 10.33928108215332, 87.99960327148438, 14.348312377929688, -38.0953369140625, -76.16510009765625, -47.301025390625, 14.169052124023438, 52.9678955078125, -99.9000244140625, 22.236846923828125, -24.1444091796875, -18.950298309326172, 130.563232421875, 22.631484985351562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000351.npy"}
{"epoch": 0.7350785340314137, "step": 352, "batch_size": 128, "mean": 31.588491439819336, "std": 45.49630355834961, "min": -89.66668701171875, "p10": -24.70885009765625, "median": 31.091602325439453, "p90": 91.54573822021483, "max": 117.22817993164062, "pos_frac": 0.75, "sample": [61.8111572265625, 25.338775634765625, 107.93341064453125, -30.7379150390625, 7.66961669921875, 109.88449096679688, 53.490013122558594, 36.47722244262695, 66.13398742675781, 0.58837890625, -16.269775390625, 71.29463195800781, -66.92303466796875, 16.669342041015625, 82.37254333496094, 4.197822570800781, 19.582611083984375, 98.42218017578125, 79.2667236328125, 55.048583984375, 117.22817993164062, -29.866971969604492, -25.8028564453125, 96.70211791992188, 40.7491455078125, -89.66668701171875, -9.104888916015625, 43.70585632324219, 16.9180908203125, 64.33463287353516, -1.3847770690917969, 2.5340576171875, 100.36993408203125, 11.75860595703125, 75.61138916015625, -45.469635009765625, 5.049339294433594, 51.61749267578125, -10.143402099609375, 1.01715087890625, 36.58056640625, -70.5234375, -10.19024658203125, 56.908050537109375, -16.560348510742188, 26.21929931640625, 0.0, 49.87298583984375, 29.767425537109375, 88.47355651855469, 4.435638427734375, 106.25442504882812, 60.3792724609375, 20.45733642578125, 85.50214385986328, 53.600372314453125, 11.113784790039062, 0.0, 42.633056640625, 66.5015869140625, 57.45844268798828, 1.223907470703125, 20.658248901367188, 55.166015625, -70.82003784179688, -20.75653076171875, 13.427143096923828, 28.3475341796875, -34.631591796875, 8.080596923828125, 52.02226257324219, 102.09768676757812, 46.390960693359375, 20.01240348815918, 11.204296112060547, -68.419677734375, 104.08355712890625, 35.527530670166016, -60.410186767578125, -35.3184814453125, 103.19607543945312, 97.42997741699219, 89.33586120605469, 11.305976867675781, -8.614837646484375, 82.398193359375, 11.405143737792969, -24.239990234375, -2.0687637329101562, 73.88461303710938, 18.07513427734375, -5.4334564208984375, 42.438385009765625, 17.889314651489258, 40.631134033203125, 38.03759765625, 78.85675048828125, 68.47449493408203, -4.2762298583984375, 107.37466430664062, 36.454833984375, 46.378814697265625, 72.90602111816406, 48.173614501953125, 62.67787170410156, -11.33050537109375, 66.22500610351562, 76.066650390625, 18.788818359375, -6.019111633300781, 26.593536376953125, 52.99884033203125, 87.989013671875, 3.8780136108398438, -16.439910888671875, -17.9599609375, 31.864276885986328, 85.36122131347656, 72.65560150146484, 105.09431457519531, -22.206451416015625, -26.566131591796875, 81.87875366210938, 71.65237426757812, 21.189598083496094, 46.11595916748047, 30.318927764892578, 57.309906005859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000352.npy"}
{"epoch": 0.7371727748691099, "step": 353, "batch_size": 128, "mean": 31.256126403808594, "std": 43.53288269042969, "min": -91.00180053710938, "p10": -17.826225280761715, "median": 24.586891174316406, "p90": 85.79145812988281, "max": 139.61447143554688, "pos_frac": 0.78125, "sample": [14.34195327758789, 80.3868408203125, 10.21126937866211, 4.9641571044921875, 15.098602294921875, 97.6612548828125, -49.052734375, 74.44024658203125, 47.492645263671875, 11.977554321289062, 26.591888427734375, 6.209136962890625, 8.918426513671875, 56.81170654296875, -7.258506774902344, 10.34613037109375, -16.343307495117188, -19.31640625, 32.554290771484375, -4.303253173828125, 23.059890747070312, 55.3773307800293, 6.734130859375, -91.00180053710938, 75.73942565917969, 4.7130126953125, 53.490692138671875, 49.422607421875, 137.22540283203125, -2.054515838623047, 90.750732421875, 67.61751556396484, 67.47598266601562, -63.49798583984375, 67.63803100585938, -47.27813720703125, 109.1151123046875, 69.94384765625, 50.843994140625, 139.61447143554688, 7.3573150634765625, -31.08905029296875, 27.85296630859375, 90.07672119140625, 15.66497802734375, -0.5791015625, 85.99942016601562, 127.28021240234375, 108.07894897460938, 63.82386016845703, 76.88447570800781, 36.70359802246094, -22.162002563476562, 58.6865234375, 11.296951293945312, 112.28329467773438, 5.812349319458008, -7.1730804443359375, 62.916282653808594, 31.11431884765625, 110.37840270996094, 35.805938720703125, -17.187576293945312, -47.79205322265625, 3.9801177978515625, 23.085617065429688, -49.37232971191406, -12.52716064453125, 22.69525146484375, 23.201202392578125, 53.39729309082031, 61.56687927246094, 66.50099182128906, 74.0509262084961, 77.69757080078125, 67.82923889160156, 9.675163269042969, 74.55979919433594, 43.65754699707031, 8.796463012695312, 72.82806396484375, 10.815216064453125, 10.125213623046875, 18.59320068359375, 41.3267822265625, 85.70233154296875, 25.715728759765625, 70.5128173828125, 61.2060546875, 3.8095436096191406, -6.230123519897461, -37.19464111328125, 69.89979553222656, 30.090789794921875, -5.6776580810546875, -15.74224853515625, -3.1980419158935547, 19.412124633789062, 17.680152893066406, -9.7279052734375, 53.6473388671875, 3.6964950561523438, 25.64166259765625, 5.325374603271484, 88.1856689453125, 31.57281494140625, 58.079833984375, 45.6390380859375, 75.60734558105469, -19.499771118164062, 23.532119750976562, 1.3617134094238281, 15.775867462158203, 46.888031005859375, 4.628929138183594, 0.0, 109.29061889648438, 36.32696533203125, -31.444793701171875, 0.7433013916015625, 2.71636962890625, -45.754180908203125, -9.210037231445312, 13.171112060546875, 79.67793273925781, 38.34661865234375, 68.9161148071289, 60.48443603515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000353.npy"}
{"epoch": 0.7392670157068063, "step": 354, "batch_size": 128, "mean": 18.72800064086914, "std": 47.50568771362305, "min": -97.958251953125, "p10": -39.55433349609374, "median": 12.821067810058594, "p90": 88.58064575195313, "max": 123.85525512695312, "pos_frac": 0.625, "sample": [3.0591278076171875, -27.734207153320312, -72.25617980957031, 79.20980834960938, 33.58613586425781, 117.19662475585938, 14.39947509765625, 7.849376678466797, 11.995281219482422, -0.3608970642089844, 22.303176879882812, 33.20556640625, -30.983489990234375, 13.007583618164062, 90.45654296875, -6.5245208740234375, -14.664085388183594, 0.08284759521484375, 1.09002685546875, -3.86944580078125, 48.98149108886719, 44.70463562011719, 68.31610107421875, -70.44268798828125, 19.294082641601562, 14.424568176269531, -6.237186431884766, -2.079803466796875, 80.96131896972656, 118.66720581054688, -27.889144897460938, -1.560516357421875, 0.3819427490234375, 40.11578369140625, 101.38674926757812, 12.634552001953125, 10.81668472290039, -1.585906982421875, 89.06048583984375, 78.2215805053711, 2.3175201416015625, -44.3717041015625, 27.250091552734375, -16.3271484375, 74.14392852783203, 69.07343292236328, 2.514617919921875, 4.91961669921875, -69.78030395507812, 7.5684814453125, 18.18560791015625, 123.85525512695312, 31.816375732421875, 100.1556396484375, 92.11471557617188, 93.28683471679688, 67.6594467163086, 42.74005126953125, 42.77880859375, 30.724822998046875, 23.323028564453125, -7.0326690673828125, -10.127525329589844, -31.878082275390625, 3.73516845703125, 44.792144775390625, 13.674358367919922, 120.1158447265625, -4.5435028076171875, -1.5965423583984375, 92.00506591796875, -31.1790771484375, 54.64471435546875, 6.828851699829102, 115.9410400390625, 30.935806274414062, 64.2834243774414, -54.04332733154297, -9.1878662109375, -0.0808868408203125, -32.23443603515625, -37.48974609375, 88.375, 32.616363525390625, -14.570236206054688, 32.711151123046875, -2.208221435546875, 77.38937377929688, -18.084693908691406, -44.495758056640625, 23.7615966796875, -33.54932403564453, 79.42643737792969, 4.341888427734375, 44.82560729980469, 20.374710083007812, -7.6923370361328125, -13.16973876953125, -44.76637649536133, -62.9906005859375, 33.73106384277344, -4.11895751953125, -5.6633148193359375, 40.65216827392578, 58.5091552734375, 34.792266845703125, -2.779052734375, -4.1754150390625, 16.503433227539062, 108.72747802734375, 70.23898315429688, 25.660873413085938, -36.75133514404297, 79.6890869140625, -53.70414733886719, 52.041778564453125, -97.958251953125, 46.13005065917969, -84.779296875, -25.643966674804688, 44.09864807128906, 17.055084228515625, -61.411285400390625, 28.711395263671875, -0.21038818359375, -54.79405212402344, 59.5548095703125, 10.0499267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000354.npy"}
{"epoch": 0.7413612565445026, "step": 355, "batch_size": 128, "mean": 31.806377410888672, "std": 45.99357223510742, "min": -93.33718872070312, "p10": -19.77781982421875, "median": 29.33025550842285, "p90": 91.48908996582031, "max": 121.73040771484375, "pos_frac": 0.7578125, "sample": [-0.948150634765625, -80.23284912109375, 61.6168212890625, 40.093170166015625, 1.940093994140625, 7.665153503417969, 32.06817626953125, 3.5609130859375, 53.988067626953125, 15.20465087890625, 84.56459045410156, 8.593082427978516, -10.369270324707031, 66.77560424804688, -0.029916763305664062, 75.94752502441406, 99.0501708984375, 86.90846252441406, -3.1135215759277344, -21.080322265625, -41.45709228515625, 7.96441650390625, 20.59454345703125, 104.263671875, 8.06817626953125, 98.07501220703125, -8.71673583984375, 1.2127685546875, 27.05841064453125, 39.25531005859375, 15.04296875, 41.65142822265625, 45.90875244140625, 107.109375, 108.6607666015625, 31.115264892578125, 110.67645263671875, 28.463855743408203, 6.064178466796875, -43.223876953125, 59.70592498779297, 77.91645812988281, 47.75799560546875, 49.09382629394531, -47.359893798828125, 6.86065673828125, -3.731842041015625, 19.153236389160156, 0.518402099609375, 24.2462158203125, 61.175933837890625, 65.08052062988281, 67.63714599609375, 10.230873107910156, 2.372344970703125, 59.787353515625, -0.9320945739746094, 22.798295974731445, 33.8834228515625, -10.76961898803711, 91.86239624023438, 8.130090713500977, -19.2196044921875, -12.581283569335938, 63.932228088378906, -93.33718872070312, 17.774658203125, 121.73040771484375, 65.82720947265625, 72.56344604492188, 2.32159423828125, 71.33718872070312, -89.48602294921875, 82.63681030273438, 1.235870361328125, 20.129135131835938, 72.05593872070312, 29.039295196533203, 106.73233032226562, 53.41251754760742, 51.10148620605469, 79.46116638183594, -1.168060302734375, 1.9524078369140625, 17.205322265625, -44.905517578125, 30.399505615234375, -29.2647705078125, 19.42748260498047, 26.910369873046875, 72.419921875, -10.40850830078125, -78.08657836914062, -3.8662109375, -4.7466278076171875, 29.6212158203125, 36.01640319824219, 103.29415130615234, -6.474822998046875, 91.3291015625, 32.983612060546875, -27.824790954589844, 112.7554931640625, 59.10041809082031, 104.09140014648438, 74.95912170410156, 90.22337341308594, 62.35162353515625, -36.27569580078125, -4.62799072265625, 39.594024658203125, 25.367568969726562, -10.547042846679688, 83.68081665039062, 67.71209716796875, 52.210411071777344, 83.14964294433594, 67.8321304321289, 81.81864929199219, 60.29320526123047, 6.8689117431640625, -1.34783935546875, 9.35723876953125, 103.1724853515625, 49.334014892578125, -53.8907470703125, 85.22914123535156, 31.915374755859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000355.npy"}
{"epoch": 0.743455497382199, "step": 356, "batch_size": 128, "mean": 24.913833618164062, "std": 40.58464813232422, "min": -92.40104675292969, "p10": -24.288007354736326, "median": 15.503231048583984, "p90": 80.11787719726563, "max": 108.2852783203125, "pos_frac": 0.703125, "sample": [-2.72723388671875, -3.508392333984375, 81.69583129882812, -3.545684814453125, 63.592864990234375, -18.07421875, 14.5806884765625, -23.343215942382812, 0.9412612915039062, 6.6025848388671875, -5.2636566162109375, 6.9308319091796875, 18.4237060546875, 59.731842041015625, -6.502265930175781, 2.1387062072753906, 20.890838623046875, -36.648040771484375, -36.342041015625, 11.599212646484375, -25.218963623046875, 80.63267517089844, 33.91559600830078, 30.037567138671875, -1.0523834228515625, 108.2852783203125, -50.16737365722656, 55.42078399658203, 3.9034423828125, 56.709228515625, 24.569305419921875, 51.252197265625, 8.6236572265625, 49.78821563720703, 53.08502197265625, 79.00486755371094, 52.80657958984375, 64.01106262207031, 105.48912048339844, 59.25764465332031, -34.78910827636719, 80.32363891601562, 4.003852844238281, 3.2835769653320312, 15.465522766113281, 12.73822021484375, -4.0281982421875, 101.7088623046875, 58.000213623046875, -3.589366912841797, 13.941757202148438, 94.67745971679688, 89.84030151367188, 91.95098876953125, -1.6866493225097656, 26.58758544921875, -41.928192138671875, 77.8302001953125, 56.66883087158203, 31.15582275390625, 70.49041748046875, -73.42670440673828, 50.007816314697266, 52.78363037109375, 62.88682556152344, 64.22452545166016, 43.9256591796875, -5.463951110839844, -2.1401519775390625, 3.711334228515625, -14.295700073242188, 19.263381958007812, 50.5013427734375, -24.07727813720703, 5.798095703125, 50.298553466796875, 58.121429443359375, 2.8782424926757812, 79.02626037597656, -24.779708862304688, 30.753150939941406, -6.250423431396484, 102.38088989257812, 23.827224731445312, -13.303985595703125, 71.12661743164062, 70.91075134277344, 53.91465759277344, 23.14168930053711, -2.097949981689453, -41.907325744628906, -2.315837860107422, 7.7620086669921875, -92.40104675292969, 15.540939331054688, 11.275634765625, 34.20611572265625, 0.0, 9.838851928710938, 17.48809814453125, 34.75025939941406, 12.205780029296875, -4.681818008422852, -18.2427978515625, 0.6716423034667969, 13.875946044921875, 15.355823516845703, 53.75257873535156, -2.074249267578125, -38.72128677368164, 80.02969360351562, 99.01591491699219, 75.42015075683594, 57.93988037109375, -8.513557434082031, 85.96026611328125, 59.70587158203125, 4.287960052490234, 49.79498291015625, 69.98532104492188, -24.92889404296875, 47.33653259277344, 0.1324920654296875, -47.2655029296875, 39.38671875, -12.829086303710938, 13.657485961914062, 85.65589904785156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000356.npy"}
{"epoch": 0.7455497382198953, "step": 357, "batch_size": 128, "mean": 28.784751892089844, "std": 46.805267333984375, "min": -101.05427551269531, "p10": -20.849798583984374, "median": 17.293346405029297, "p90": 94.3267333984375, "max": 150.40997314453125, "pos_frac": 0.734375, "sample": [45.382843017578125, 87.32077026367188, -6.257438659667969, -0.068023681640625, 83.50387573242188, 32.066253662109375, 0.0, -4.650146484375, 32.2535400390625, 2.450042724609375, 70.79866027832031, -53.22935485839844, -3.2406444549560547, 150.40997314453125, 116.28289794921875, 117.8221435546875, 32.34444808959961, 4.0058746337890625, 55.93312072753906, 17.479248046875, 12.210647583007812, 8.3426513671875, 31.903671264648438, 87.87614440917969, 97.06741333007812, -42.94964599609375, 131.68496704101562, 17.107444763183594, 22.498844146728516, 110.51385498046875, -6.236663818359375, 11.978363037109375, -14.637451171875, 7.054689407348633, 127.8167724609375, 19.67911720275879, -11.964447021484375, -35.904930114746094, 1.45697021484375, 40.20172882080078, 51.68809509277344, 123.554931640625, 12.620208740234375, 56.17469787597656, 28.705917358398438, -24.493255615234375, 5.0623779296875, -70.03103637695312, 90.76495361328125, 14.389556884765625, -8.705856323242188, 27.04859161376953, 2.8776092529296875, 35.5618896484375, 37.99853515625, 62.03746032714844, 97.7730712890625, 75.02896118164062, 21.81146240234375, -20.652099609375, 83.1748046875, 9.430191040039062, 15.654449462890625, -21.31109619140625, 70.63751220703125, 1.0003128051757812, 80.92378234863281, 50.479034423828125, 31.590240478515625, 57.90899658203125, 19.892974853515625, -18.67127227783203, -14.254043579101562, 1.23785400390625, 3.1178207397460938, 79.02487182617188, 49.46308898925781, -45.877349853515625, 13.772979736328125, -30.114227294921875, -36.48101806640625, 16.465866088867188, 13.097366333007812, 15.3929443359375, 96.10809326171875, 19.8060302734375, 90.30999755859375, -24.470611572265625, 3.9951629638671875, -0.5441741943359375, 3.4951858520507812, 16.02752685546875, 23.897628784179688, -19.667373657226562, 33.06683349609375, 77.08660888671875, 13.74127197265625, -52.058799743652344, 0.0, 98.51171875, -65.88035583496094, 0.0, 25.195068359375, -18.7718505859375, 93.56329345703125, 5.427419662475586, 15.337821960449219, 69.21595764160156, 85.56649780273438, 36.9732666015625, 103.32275390625, -101.05427551269531, 91.32449340820312, 41.531219482421875, 3.6912059783935547, 54.42427062988281, -6.787750244140625, 68.10098266601562, 43.95094299316406, 81.56915283203125, 115.34481811523438, 13.251136779785156, -4.530479431152344, -13.768844604492188, 42.07188415527344, -1.2017059326171875, 1.2155990600585938, 64.98023986816406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000357.npy"}
{"epoch": 0.7476439790575916, "step": 358, "batch_size": 128, "mean": 32.13968276977539, "std": 47.847896575927734, "min": -87.59317016601562, "p10": -21.144137573242187, "median": 25.620868682861328, "p90": 93.53640670776367, "max": 127.9007568359375, "pos_frac": 0.765625, "sample": [2.62908935546875, 2.60809326171875, 53.07823944091797, -26.533355712890625, -2.327117919921875, 19.422571182250977, -20.5802001953125, 63.935882568359375, 95.02716064453125, 127.9007568359375, 5.224033355712891, 1.8526802062988281, 49.44889831542969, 79.856689453125, 93.29220581054688, 79.0300064086914, 23.069976806640625, 92.04934692382812, 97.41511535644531, 32.580352783203125, -83.48004150390625, 80.05266571044922, -11.190872192382812, -59.318275451660156, 78.13058471679688, 12.454444885253906, 77.02029418945312, 0.06756591796875, 101.53510284423828, -7.27569580078125, 7.2999114990234375, 36.3006591796875, 2.7130584716796875, 65.99716186523438, 33.06573486328125, 10.325408935546875, 125.96319580078125, -43.23468017578125, 31.58837890625, 66.07298278808594, -61.3482666015625, 57.014678955078125, 86.2294921875, 101.28048706054688, 30.255271911621094, 83.54879760742188, 67.375732421875, 88.39302062988281, 94.10620880126953, 9.336669921875, -35.05303955078125, -7.729408264160156, -3.41790771484375, 18.785926818847656, 40.83369445800781, -12.098663330078125, 73.03302001953125, 87.19593811035156, -2.0652313232421875, 118.004150390625, -11.21240234375, -87.59317016601562, 109.6517333984375, -10.586898803710938, 5.8581695556640625, 81.32879638671875, 54.688629150390625, 13.79483413696289, -4.752471923828125, 6.7242431640625, 24.651771545410156, 11.098793029785156, 10.5697021484375, 9.655952453613281, 3.481414794921875, -63.31280517578125, 73.89302062988281, 62.6010627746582, -3.1989517211914062, 107.45623779296875, 22.479459762573242, -41.4913330078125, 76.99826049804688, 39.82794189453125, 89.10613250732422, 26.5899658203125, 71.254638671875, 31.11944580078125, 19.081287384033203, 86.27499389648438, 6.434989929199219, 99.927490234375, 68.69038391113281, 61.96885681152344, 46.383636474609375, 83.94387817382812, -22.459991455078125, 9.6146240234375, 0.77935791015625, 50.06948471069336, 120.486572265625, 5.4534912109375, 35.76591491699219, 97.00885009765625, 50.95269775390625, 87.02276611328125, 87.5589599609375, -72.93563842773438, -12.174629211425781, -70.34341430664062, -30.588031768798828, 5.295658111572266, 66.34339904785156, 7.883697509765625, 36.51294708251953, -16.25804901123047, 10.656867980957031, 13.028409957885742, -0.3873291015625, 22.330810546875, 0.0, 33.241729736328125, 64.95077514648438, 29.132110595703125, 84.31861114501953, -14.872810363769531, 24.37286376953125, 0.982513427734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000358.npy"}
{"epoch": 0.749738219895288, "step": 359, "batch_size": 128, "mean": 29.711278915405273, "std": 41.64919662475586, "min": -97.03546142578125, "p10": -17.357565307617186, "median": 28.056915283203125, "p90": 84.01702194213867, "max": 128.29806518554688, "pos_frac": 0.765625, "sample": [27.348876953125, -32.1280517578125, 71.08961486816406, 91.30438232421875, 65.10159301757812, -10.51934814453125, 31.892616271972656, 10.677444458007812, 28.06536865234375, 28.94891357421875, 24.656883239746094, 42.79229736328125, 19.751178741455078, 20.05402374267578, 11.171173095703125, -11.6898193359375, 58.97804260253906, 33.12474060058594, 18.284515380859375, 17.848739624023438, 57.20204162597656, -4.33319091796875, 85.11502838134766, 103.94012451171875, -16.91546630859375, 77.62133026123047, 3.582061767578125, 28.0484619140625, 55.33197021484375, -88.67459106445312, -2.34271240234375, 120.580810546875, 3.53802490234375, 12.50738525390625, 2.670543670654297, 49.296783447265625, -5.463672637939453, 73.48196411132812, 75.8140869140625, 61.998291015625, -18.389129638671875, -58.93644714355469, 83.54644775390625, 3.3982467651367188, 64.6773681640625, 54.132354736328125, 37.18385314941406, 40.05718994140625, 67.46490478515625, 64.31009674072266, 34.931671142578125, 5.076873779296875, 128.29806518554688, -10.332748413085938, 43.08001708984375, 0.8152008056640625, 38.467529296875, -95.34417724609375, 68.86579132080078, 36.271240234375, 110.411376953125, 36.567718505859375, -97.03546142578125, 53.48974609375, -1.157318115234375, -1.164031982421875, -6.6680908203125, 34.153656005859375, 71.57376098632812, 64.4041748046875, -1.370635986328125, 20.952239990234375, -24.722206115722656, 28.860565185546875, 47.060203552246094, 50.80604553222656, -31.657196044921875, 68.65139770507812, 11.102935791015625, 78.06143951416016, 47.203643798828125, -4.20037841796875, 8.932464599609375, 50.293975830078125, 38.57756042480469, 8.486221313476562, 9.534393310546875, 71.56649780273438, 17.689849853515625, 26.44757080078125, 87.63386535644531, 12.157119750976562, 52.056640625, 68.65812683105469, 22.60797119140625, -20.252349853515625, -12.058334350585938, 45.416473388671875, -21.102855682373047, -0.5074748992919922, 20.53643798828125, 90.94781494140625, 33.30950927734375, -26.137832641601562, 16.770248413085938, -28.302444458007812, 96.00448608398438, -13.067474365234375, 41.781890869140625, 86.18746948242188, 68.96543884277344, 23.199684143066406, 17.485595703125, 24.649497985839844, -14.852569580078125, 104.27349853515625, 61.9482421875, 28.3585205078125, 58.660919189453125, 2.4276504516601562, -5.2977142333984375, 95.84344482421875, 15.2176513671875, 7.528963088989258, 105.098388671875, 8.399154663085938, 58.98685073852539, -24.667877197265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000359.npy"}
{"epoch": 0.7518324607329843, "step": 360, "batch_size": 128, "mean": 25.821168899536133, "std": 41.160675048828125, "min": -119.69561767578125, "p10": -12.695252990722647, "median": 18.723220825195312, "p90": 84.20125579833984, "max": 134.32000732421875, "pos_frac": 0.75, "sample": [53.680694580078125, 38.270347595214844, -24.137916564941406, 88.8824462890625, -72.1258544921875, -7.728036880493164, -60.248626708984375, 40.48828125, 134.32000732421875, -28.887069702148438, 28.27142333984375, 33.98388671875, -18.763504028320312, 75.02706909179688, 51.94537353515625, 51.69480895996094, 14.023269653320312, 51.7530517578125, 17.24725341796875, 40.07908630371094, 83.02740478515625, 0.9320831298828125, 12.73529052734375, 63.67390441894531, 65.42192840576172, -1.859405517578125, 2.4808578491210938, 17.03466796875, -9.980438232421875, 67.04080200195312, 28.12652587890625, 103.73692321777344, 20.675048828125, 21.07080078125, 83.64157104492188, -0.24365234375, 8.913467407226562, 74.4609375, -27.97344970703125, 32.29986572265625, 89.946533203125, 33.95159912109375, 40.2921142578125, 33.942840576171875, 18.84368896484375, 9.66872787475586, 11.880462646484375, 75.44876098632812, -2.56536865234375, 27.466827392578125, -9.6007080078125, 56.4970703125, 25.71148681640625, 36.97265625, 105.24856567382812, -6.5662841796875, 92.54890441894531, -10.094573974609375, 85.50718688964844, 3.2710494995117188, 18.602752685546875, -5.336944580078125, 82.57232666015625, 26.526548385620117, 12.879547119140625, 1.8831939697265625, 93.41494750976562, 12.590545654296875, 52.83265686035156, -66.70477294921875, 5.7254180908203125, 67.89697265625, 1.3521461486816406, 54.419586181640625, -8.218414306640625, 31.896656036376953, 33.192138671875, 98.76925659179688, 80.9705810546875, 75.2342300415039, -9.985733032226562, 66.63397216796875, 6.394075393676758, -0.28476715087890625, -119.69561767578125, 70.24649047851562, 39.553955078125, 17.808837890625, -20.22955322265625, 4.660491943359375, 89.53231811523438, 4.308319091796875, 87.65426635742188, 9.795860290527344, -0.9067306518554688, 21.710205078125, 3.77618408203125, -50.20843505859375, -6.1343841552734375, 4.516544342041016, 75.54267883300781, 0.0, 6.759796142578125, 13.827392578125, 90.34230041503906, 0.5681228637695312, 53.0308837890625, 29.9769287109375, 10.368377685546875, 4.454912185668945, 8.907928466796875, -4.540374755859375, -40.16576385498047, 69.44761657714844, 7.884284973144531, 27.40270233154297, -24.515403747558594, -7.8863525390625, 0.0, 7.75067138671875, 19.226699829101562, 25.381885528564453, 25.735214233398438, -26.87572479248047, -6.6562652587890625, 16.421005249023438, 95.78384399414062, 65.92784118652344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000360.npy"}
{"epoch": 0.7539267015706806, "step": 361, "batch_size": 128, "mean": 34.16133117675781, "std": 46.132598876953125, "min": -89.86224365234375, "p10": -20.51419830322265, "median": 31.04417610168457, "p90": 95.03441772460937, "max": 136.1883544921875, "pos_frac": 0.7265625, "sample": [39.35297393798828, -7.340606689453125, -33.620025634765625, 26.736785888671875, 21.771331787109375, -5.399169921875, 0.2904510498046875, -27.1964111328125, 68.15017700195312, 28.798797607421875, 58.04808807373047, 11.8382568359375, 2.4400291442871094, 88.55697631835938, -1.7227706909179688, 88.41575622558594, -0.6025924682617188, 10.368179321289062, -18.609466552734375, 2.5355758666992188, 80.63058471679688, 105.66375732421875, -17.565780639648438, 82.8372802734375, -2.0092239379882812, 102.97146606445312, 57.40751647949219, 74.85671997070312, 6.483966827392578, 117.43698120117188, 12.906585693359375, -2.8182220458984375, 64.64014434814453, 24.22705078125, 11.204296112060547, 77.56568908691406, -48.64227294921875, 71.63372802734375, 63.51145935058594, 21.37530517578125, 100.0115966796875, 96.33743286132812, 20.392715454101562, 41.639591217041016, 125.64297485351562, 13.85565185546875, -50.13299560546875, -5.36737060546875, 12.84979248046875, -9.539913177490234, -5.684469223022461, 56.92231750488281, 40.95350646972656, 0.0, 25.874404907226562, 10.173919677734375, 77.73736572265625, 40.672119140625, 4.968147277832031, -2.3990631103515625, 42.19920349121094, 71.24830627441406, -9.111360549926758, 100.24725341796875, 40.279998779296875, -35.76593017578125, 25.16790771484375, 51.423927307128906, 39.24432373046875, -33.24896240234375, 56.37451171875, -6.060546875, -15.2154541015625, 8.9307861328125, 31.393264770507812, 0.0, 62.02789306640625, -40.77406311035156, 33.98115539550781, 2.357391357421875, 125.594970703125, 87.66971588134766, 77.27204895019531, 42.848541259765625, -89.86224365234375, 78.79203033447266, -3.694793701171875, -29.8262939453125, 136.1883544921875, 76.77664184570312, -33.47975158691406, 37.257965087890625, -67.8516845703125, 77.06087493896484, 99.78463745117188, 69.91650390625, 43.966880798339844, 5.4584197998046875, 90.13238525390625, 62.34663772583008, 94.47598266601562, -10.106985092163086, 114.08372497558594, 37.14628601074219, 86.45741271972656, 85.71847534179688, 31.858001708984375, 15.467681884765625, 48.899322509765625, 61.67657470703125, 91.34161376953125, 50.09636688232422, -9.144927978515625, 11.03509521484375, 101.48333740234375, 83.76969909667969, -4.499664306640625, 86.51058959960938, 96.39663696289062, 25.813446044921875, 19.518829345703125, 54.38493347167969, -24.958572387695312, 30.695087432861328, -11.774024963378906, 9.692642211914062, 87.77153778076172, -54.247100830078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000361.npy"}
{"epoch": 0.7560209424083769, "step": 362, "batch_size": 128, "mean": 27.073198318481445, "std": 41.82011413574219, "min": -95.23788452148438, "p10": -24.522124099731442, "median": 16.844802856445312, "p90": 79.38445816040038, "max": 167.3797607421875, "pos_frac": 0.78125, "sample": [3.0974044799804688, 57.23492431640625, 6.264484405517578, 31.7825927734375, 55.19390869140625, 8.463607788085938, 67.57513427734375, 16.264083862304688, 0.8072509765625, 6.9498291015625, 25.325531005859375, -95.23788452148438, -44.62518310546875, -25.74404525756836, 69.88998413085938, 13.294466018676758, 90.10881042480469, -31.34625244140625, 13.098625183105469, 50.4918098449707, 54.524566650390625, 81.55329132080078, 167.3797607421875, 37.603759765625, 47.605224609375, 84.37745666503906, 55.50030517578125, 12.473220825195312, -55.32543182373047, 19.228546142578125, -9.308975219726562, -23.841995239257812, 54.15119934082031, -6.321296691894531, -21.645896911621094, 73.02700805664062, 10.804634094238281, -9.895633697509766, 65.65185546875, 13.045799255371094, 119.670166015625, 15.238250732421875, 99.90823364257812, 53.98017883300781, 50.65325927734375, 82.91944122314453, 79.08492279052734, -4.577850341796875, -10.41259765625, 16.104598999023438, -26.9661865234375, 46.34772872924805, 49.6160888671875, -23.998443603515625, 2.0849342346191406, 7.351287841796875, 20.8450927734375, 78.69732666015625, 7.0020751953125, 74.53350830078125, 51.05500793457031, -13.42706298828125, 25.742156982421875, 68.4959716796875, 38.272857666015625, 47.5731201171875, 8.341033935546875, 15.869247436523438, 58.71454620361328, 41.9786376953125, 93.95635986328125, 41.176666259765625, 27.831863403320312, 1.531494140625, 0.7909202575683594, 118.8677978515625, 38.77787780761719, 6.079536437988281, 60.09918212890625, 65.44831848144531, -42.713775634765625, 45.84039306640625, -31.275386810302734, 12.482162475585938, -32.539581298828125, -38.442535400390625, -10.525615692138672, 91.67475891113281, 6.274784088134766, 7.3378448486328125, 20.368408203125, 6.886711120605469, -5.89251708984375, 9.891738891601562, 38.13983154296875, 50.04405212402344, 64.1444091796875, 16.501052856445312, 14.660614013671875, -3.423828125, -4.9565887451171875, 68.2943344116211, 80.0833740234375, 16.861358642578125, 3.797943115234375, 16.2110595703125, -83.2789306640625, 16.8282470703125, 40.6912841796875, 12.860153198242188, 64.14961242675781, 12.625104904174805, 63.579654693603516, 12.802902221679688, 19.1983642578125, -9.10809326171875, 50.538787841796875, 94.30126953125, 10.71875, 16.328506469726562, -36.939208984375, 64.22879791259766, 62.289703369140625, 87.96932983398438, -11.120849609375, -43.79075622558594, 66.5562744140625, 49.481407165527344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000362.npy"}
{"epoch": 0.7581151832460733, "step": 363, "batch_size": 128, "mean": 31.38487434387207, "std": 43.97297668457031, "min": -114.37445068359375, "p10": -14.30253677368164, "median": 28.446014404296875, "p90": 85.2889450073242, "max": 137.67608642578125, "pos_frac": 0.7890625, "sample": [64.03777313232422, -2.7637786865234375, -9.34765625, 94.8358154296875, 79.8992919921875, 65.1798095703125, -0.4734020233154297, -49.43548583984375, 14.8944091796875, 66.81107330322266, -27.012542724609375, 6.494636535644531, 80.17761993408203, 55.569541931152344, 0.8494415283203125, 21.68060302734375, 48.314849853515625, 110.77992248535156, 0.45220947265625, -16.15656089782715, 16.129241943359375, 48.682220458984375, 101.2374267578125, 37.94775390625, 18.23369598388672, 104.6719970703125, 108.33415222167969, 137.67608642578125, 8.267486572265625, 3.4138412475585938, 81.05244445800781, -3.9074249267578125, -8.777633666992188, -3.735992431640625, 66.725341796875, 1.4572086334228516, -3.5790138244628906, 68.114013671875, 82.91181945800781, 0.9498291015625, -5.712059020996094, 15.729583740234375, 7.3800048828125, 59.458946228027344, 56.54469299316406, 49.5943603515625, 8.463485717773438, 48.17669677734375, 67.21176147460938, 1.84075927734375, 65.6883544921875, 3.533487319946289, 81.73828125, 81.5745849609375, -65.10504150390625, 53.2117919921875, 1.0551605224609375, 53.24945068359375, 25.13238525390625, 34.69586181640625, 20.732147216796875, 28.956192016601562, 59.664642333984375, 70.29307556152344, 72.38726806640625, -3.9600296020507812, 92.718017578125, 35.331939697265625, 31.7337646484375, 2.7644805908203125, 70.22206115722656, 5.4040985107421875, 107.15911865234375, 136.9814453125, 4.728385925292969, 90.8355712890625, -32.878509521484375, -0.3179473876953125, 18.569202423095703, 39.42877197265625, 4.3064117431640625, 16.710037231445312, -27.549896240234375, 5.6295166015625, 31.996185302734375, -14.006668090820312, -1.3629302978515625, -3.473907470703125, -52.75982666015625, 35.151092529296875, 46.90216064453125, -70.51516723632812, 58.956146240234375, 7.2411651611328125, 71.40692901611328, 72.60781860351562, 69.97650146484375, 18.827316284179688, 63.49627685546875, 55.99482727050781, 17.908843994140625, 34.52064514160156, -77.01960754394531, 62.34490966796875, 66.44912719726562, 51.751556396484375, 21.25279998779297, 44.74076843261719, 96.03092956542969, 46.21484375, -114.37445068359375, -31.592681884765625, -21.377853393554688, 0.12190628051757812, 13.04486083984375, 28.0614013671875, 8.88018798828125, 7.620269775390625, 67.69424438476562, 60.43701171875, 93.43798828125, 72.84530639648438, 8.27593994140625, 91.06146240234375, -14.992897033691406, 28.83062744140625, 14.356689453125, -13.6031494140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000363.npy"}
{"epoch": 0.7602094240837697, "step": 364, "batch_size": 128, "mean": 25.057167053222656, "std": 46.89210510253906, "min": -121.6624755859375, "p10": -17.938076019287102, "median": 16.28638458251953, "p90": 86.98873901367185, "max": 159.6336669921875, "pos_frac": 0.7109375, "sample": [48.230438232421875, -2.6810989379882812, 64.24658203125, 16.48318862915039, 26.905914306640625, 9.61590576171875, 72.50509643554688, 84.72784423828125, 69.64866638183594, 9.497093200683594, 10.403076171875, 41.43019104003906, -5.9803466796875, 16.132965087890625, -8.4892578125, 23.901519775390625, 39.306915283203125, -12.223388671875, 14.20526123046875, 28.0443115234375, -84.02192687988281, 1.2062530517578125, 31.021427154541016, -87.460693359375, 17.682861328125, -51.522613525390625, 115.36636352539062, 78.9121322631836, -13.802711486816406, 72.96292114257812, 6.97344970703125, 13.551826477050781, 26.87506103515625, 6.527387619018555, -23.2025146484375, 5.0351104736328125, -0.6014862060546875, 69.14598083496094, 33.57080078125, 79.85601806640625, 61.12322998046875, 7.23785400390625, -103.17193603515625, 50.98860168457031, 71.00736999511719, 95.08184814453125, -16.08258819580078, 23.109405517578125, -11.4766845703125, -3.86920166015625, 56.60105895996094, 82.691650390625, 1.78009033203125, 55.4410400390625, -9.388236999511719, -2.38812255859375, 104.139404296875, -1.0105838775634766, -121.6624755859375, 127.33839416503906, 19.820220947265625, -7.564208984375, -9.132568359375, -12.1053466796875, 26.7000732421875, 63.68531799316406, -27.903823852539062, 93.71395874023438, -4.15667724609375, 6.684539794921875, 34.4073486328125, 83.18449401855469, 62.88629150390625, -10.993354797363281, 66.23014831542969, 8.596923828125, 92.26416015625, 46.42644500732422, 20.901107788085938, 77.80940246582031, 67.86109924316406, -3.9395294189453125, 18.581016540527344, -5.925140380859375, 1.197906494140625, 16.7322998046875, 13.908615112304688, 7.5784912109375, 21.620010375976562, 96.47454833984375, 68.25457763671875, -0.3667144775390625, -10.7913818359375, 4.152130126953125, 67.76124572753906, 66.5496826171875, 4.430599212646484, -15.28676986694336, -14.576507568359375, 159.6336669921875, 30.19647216796875, 98.1353759765625, -37.44146728515625, 126.77786254882812, 73.54876708984375, -43.525054931640625, 32.797119140625, 43.9608154296875, 1.01416015625, 4.430511474609375, 17.788223266601562, 10.431396484375, -2.2157440185546875, 104.08015441894531, 12.555549621582031, 16.439804077148438, 4.054718017578125, -41.726165771484375, 72.544921875, -39.04251480102539, 0.7634506225585938, 36.70685577392578, 11.823646545410156, -22.267547607421875, -27.296478271484375, 92.88787841796875, 113.13954162597656, 9.97418212890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000364.npy"}
{"epoch": 0.762303664921466, "step": 365, "batch_size": 128, "mean": 30.71148681640625, "std": 44.34214401245117, "min": -72.87849426269531, "p10": -16.443231964111327, "median": 19.13046646118164, "p90": 95.48636627197266, "max": 128.93618774414062, "pos_frac": 0.7578125, "sample": [0.0, 1.9152565002441406, 70.22781372070312, 6.9195098876953125, 90.74520874023438, 103.11669921875, 27.83843994140625, 43.462554931640625, 0.0, 20.86566162109375, 6.552490234375, 5.324493408203125, 24.414093017578125, 100.34356689453125, 6.874000549316406, 8.335212707519531, 12.7459716796875, 8.613128662109375, 16.884109497070312, 88.82298278808594, 32.555511474609375, 60.887451171875, 0.43377685546875, 14.182456970214844, 25.892807006835938, 17.089340209960938, -9.66607666015625, 76.20211791992188, 70.2408447265625, 80.34512329101562, 86.82196807861328, 0.18004608154296875, -23.33172607421875, 55.10718536376953, -16.90936279296875, -48.2049560546875, 69.62188720703125, 50.322242736816406, 53.376556396484375, 34.601898193359375, -7.1753387451171875, 77.0909423828125, 105.7791748046875, 40.88285827636719, -45.80120849609375, 110.20614624023438, -28.57830810546875, 47.77400207519531, 30.32119369506836, 97.17916870117188, 67.54515838623047, -15.656570434570312, 9.38531494140625, -8.75860595703125, 8.751846313476562, 3.200958251953125, 60.98527526855469, 95.83729553222656, 77.55865478515625, 2.1444854736328125, 15.92352294921875, 78.56289672851562, -24.675201416015625, 2.10552978515625, 4.605121612548828, 17.90771484375, 72.08760070800781, 78.61503601074219, 20.00811767578125, 66.6851806640625, 116.28604125976562, -5.115020751953125, 118.72708129882812, 1.2351531982421875, -4.9239501953125, 123.44259643554688, 51.672210693359375, -65.74679565429688, 89.71206665039062, 3.3123855590820312, 105.56375122070312, 1.3230514526367188, 95.33596801757812, -16.24346160888672, 12.99542236328125, -8.39208984375, 77.58877563476562, 59.15367126464844, -2.557422637939453, -9.412734985351562, 13.42831802368164, 13.291055679321289, 19.894752502441406, 99.06332397460938, -2.1426925659179688, 7.6169586181640625, 17.706268310546875, -14.496414184570312, 69.09851837158203, -4.64508056640625, 21.387649536132812, -72.87849426269531, -13.103179931640625, -3.763458251953125, 69.94876098632812, 0.0, 7.157629013061523, 4.2733917236328125, 53.4592399597168, 112.44833374023438, 43.25689697265625, 32.44218444824219, 11.41433334350586, -17.4771728515625, -28.581573486328125, -60.02703857421875, 58.298309326171875, 81.2437744140625, -20.620712280273438, 25.145294189453125, 25.1534423828125, -59.68231201171875, 128.93618774414062, 55.672637939453125, 18.366180419921875, 82.04997253417969, 39.626548767089844, 39.903411865234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000365.npy"}
{"epoch": 0.7643979057591623, "step": 366, "batch_size": 128, "mean": 31.318374633789062, "std": 43.98028564453125, "min": -111.20343017578125, "p10": -16.74901123046874, "median": 26.828388214111328, "p90": 88.34202041625976, "max": 130.01300048828125, "pos_frac": 0.765625, "sample": [8.595375061035156, 30.823593139648438, 130.01300048828125, 64.15852355957031, 67.48934173583984, 70.53341674804688, 100.9136962890625, 36.23980712890625, 23.795005798339844, 85.5616455078125, 97.02503967285156, 67.24845886230469, 44.65918731689453, -35.025634765625, 12.762723922729492, 58.32892990112305, 19.118057250976562, 29.353515625, -39.32659912109375, 32.135406494140625, -111.20343017578125, 110.86279296875, -74.2786636352539, 14.662784576416016, 32.38165283203125, 76.61373901367188, 34.050743103027344, 59.588623046875, 86.92343139648438, 56.242523193359375, -9.405731201171875, 66.89544677734375, 16.981704711914062, -24.295074462890625, 85.93017578125, -66.67263793945312, 0.52203369140625, 45.825408935546875, 20.044082641601562, 28.26513671875, 5.247844696044922, -28.759002685546875, -6.12164306640625, -4.629386901855469, -4.4434814453125, 13.900527954101562, 81.66085815429688, 25.21343994140625, 63.32000732421875, -9.10260009765625, 42.645294189453125, 25.857925415039062, 7.536224365234375, 35.15418243408203, 42.89665222167969, -2.748931884765625, 27.216766357421875, 12.471885681152344, 42.70330810546875, 26.44001007080078, 3.57794189453125, 86.18480682373047, 21.974807739257812, 0.8252201080322266, 0.9772186279296875, 92.9552001953125, 23.247617721557617, 68.6826171875, 54.92189025878906, -47.1253662109375, 89.16180419921875, 94.898681640625, -59.46199035644531, 0.0, -13.514984130859375, 89.61054992675781, 49.662109375, 49.97587585449219, 70.88116455078125, -5.196525573730469, -1.5298538208007812, 32.57257080078125, 86.75491333007812, 37.831390380859375, -4.47796630859375, 34.13478088378906, 24.0697021484375, 120.82122802734375, 3.0645217895507812, -5.8818511962890625, -4.6198272705078125, 16.68670654296875, 67.24264526367188, 18.761688232421875, 24.632999420166016, 71.21647644042969, 22.074371337890625, -5.671630859375, 28.8201904296875, 93.45733642578125, 74.71786499023438, 14.323806762695312, 3.1040496826171875, 112.66348266601562, 69.00599670410156, 9.116561889648438, 67.54570007324219, -9.602977752685547, 119.85662841796875, 8.183710098266602, 1.2208251953125, 64.90770721435547, 78.5074462890625, 17.16253662109375, -3.2277297973632812, -24.5966796875, 87.99068450927734, -41.206565856933594, 68.66204833984375, 27.69915771484375, 78.19491577148438, 18.46343994140625, 2.504791259765625, 63.27717590332031, -34.030303955078125, -7.564399719238281, 90.9959716796875, -32.1202392578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000366.npy"}
{"epoch": 0.7664921465968586, "step": 367, "batch_size": 128, "mean": 32.46257400512695, "std": 40.19624328613281, "min": -64.20883178710938, "p10": -15.7504638671875, "median": 24.45625400543213, "p90": 89.16094665527343, "max": 119.54791259765625, "pos_frac": 0.8046875, "sample": [72.88406372070312, 29.634201049804688, -8.96563720703125, 89.51583862304688, -3.0780086517333984, 14.836318969726562, 21.21350860595703, -50.97161865234375, -10.78369140625, 23.305374145507812, 77.70953369140625, 20.950923919677734, 23.98974609375, -2.8419036865234375, 66.36183166503906, 105.1275634765625, 16.061538696289062, 89.00885009765625, -1.81494140625, 42.116363525390625, 1.787994384765625, 111.18634033203125, 61.06903076171875, -33.39569091796875, 37.892677307128906, 99.72348022460938, -64.20883178710938, -9.2789306640625, 31.799530029296875, -8.082351684570312, -19.822372436523438, 40.263580322265625, 1.771230697631836, 0.5978584289550781, 79.74588012695312, -2.17474365234375, 29.897369384765625, 64.15586853027344, 34.432830810546875, -42.20050048828125, 87.93865966796875, 17.397048950195312, 5.84025764465332, 63.72528076171875, 57.9453125, 5.471244812011719, -40.1492919921875, 53.27970886230469, 1.28228759765625, -15.8056640625, 34.0311279296875, 20.295394897460938, 19.527809143066406, 16.73419189453125, 75.18389892578125, 88.04989624023438, 65.8349609375, 33.85862731933594, 21.698623657226562, -17.95074462890625, -33.2239990234375, -20.50897216796875, 108.038330078125, 71.20294189453125, 9.467864990234375, 96.51199340820312, 6.1880645751953125, 48.059539794921875, 119.54791259765625, 46.156551361083984, 45.774261474609375, 11.384552001953125, 15.622909545898438, 40.649200439453125, 80.5472412109375, 85.34587097167969, 10.69927978515625, 22.198348999023438, 14.557464599609375, -36.468841552734375, 29.32794189453125, 90.94459533691406, 87.63300323486328, 94.4798583984375, 12.1292724609375, 54.09783935546875, 57.755340576171875, 5.53790283203125, -15.726806640625, 44.466949462890625, 63.70442199707031, 68.96987915039062, 18.08990478515625, 28.243438720703125, 24.922761917114258, -4.4533233642578125, -41.102577209472656, 89.6304931640625, 18.088897705078125, 98.28411865234375, 3.3803977966308594, 104.36834716796875, 82.28981018066406, 40.28753662109375, -9.592880249023438, 37.16041564941406, 41.4609375, 75.35394287109375, 9.54034423828125, 53.59930419921875, 17.773971557617188, 40.201683044433594, 21.6968994140625, -32.87470245361328, 59.671417236328125, 2.381744384765625, 4.7984619140625, 78.1046142578125, 45.4093017578125, 70.05097961425781, 10.701444625854492, 18.632171630859375, 8.77386474609375, 9.8104248046875, 107.47845458984375, 56.82085037231445, -15.557159423828125, 23.10376739501953], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000367.npy"}
{"epoch": 0.768586387434555, "step": 368, "batch_size": 128, "mean": 37.08063888549805, "std": 41.987518310546875, "min": -54.05029296875, "p10": -11.01550140380859, "median": 36.971595764160156, "p90": 91.59492111206053, "max": 123.57513427734375, "pos_frac": 0.8046875, "sample": [7.134857177734375, 106.06707763671875, 39.83180236816406, 90.69366455078125, 13.768585205078125, 83.95721435546875, 7.281639099121094, 5.702606201171875, 3.51690673828125, -4.66949462890625, 87.79934692382812, -22.068603515625, 57.09498596191406, 114.46453857421875, 43.927032470703125, 43.67523193359375, 121.864990234375, 9.068206787109375, 59.596405029296875, 52.363548278808594, -22.2664794921875, -2.7342529296875, 32.4071044921875, -26.668846130371094, 5.4453125, 18.425445556640625, -8.973114013671875, 3.6125259399414062, -3.9616546630859375, 97.60748291015625, 0.5542831420898438, 55.36578369140625, -13.450996398925781, 55.68330383300781, 8.905303955078125, -54.05029296875, -0.81976318359375, 119.65963745117188, 70.11427307128906, 70.30776977539062, 18.701416015625, 97.78834533691406, 87.54730224609375, 70.29220581054688, 22.1370849609375, 15.721160888671875, 74.21998596191406, -5.5043182373046875, 13.636056900024414, -37.786399841308594, 4.708221435546875, 46.41039276123047, 7.809900283813477, 82.2066421508789, -7.194267272949219, 3.4053192138671875, 105.30697631835938, 66.25182342529297, -18.28533935546875, 51.60215759277344, 38.27006530761719, 47.431396484375, 47.837432861328125, 82.64582824707031, 94.64892578125, 52.10003662109375, -32.0068359375, 30.547203063964844, 27.759674072265625, -2.540557861328125, 28.368072509765625, 76.97918701171875, 86.55795288085938, 24.855010986328125, 9.950515747070312, 69.32478332519531, 74.78076171875, 50.129547119140625, 82.09463500976562, 42.159759521484375, 6.49420166015625, 62.40155029296875, 88.59046936035156, 39.9117431640625, -7.979957580566406, 86.00090026855469, 9.627655029296875, 28.60309600830078, -0.9852142333984375, 35.673126220703125, 0.4872703552246094, 7.896293640136719, 93.6978530883789, 19.67127227783203, 46.821434020996094, 22.40363311767578, 31.01026153564453, 66.113037109375, -9.971717834472656, 56.12055206298828, 98.29202270507812, -34.81230926513672, 88.68840026855469, 72.56048583984375, -30.2706298828125, 22.617919921875, -7.284515380859375, 60.108612060546875, 123.57513427734375, 76.27798461914062, 21.4305419921875, 74.9717788696289, 22.294525146484375, 6.543510437011719, 11.404930114746094, 77.2344970703125, 104.6993408203125, 86.41751098632812, 38.4857177734375, 39.93829345703125, 42.8760986328125, -41.64457702636719, -53.319366455078125, 3.691497802734375, 69.28968048095703, 97.49519348144531, -41.327545166015625, 76.39825439453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000368.npy"}
{"epoch": 0.7706806282722513, "step": 369, "batch_size": 128, "mean": 26.298860549926758, "std": 44.197547912597656, "min": -82.88552856445312, "p10": -28.205464935302732, "median": 18.128692626953125, "p90": 83.53645248413085, "max": 133.68609619140625, "pos_frac": 0.734375, "sample": [49.57316589355469, 37.32341766357422, 63.99806213378906, -23.225921630859375, 21.93170928955078, 0.936065673828125, 9.011543273925781, 31.45440673828125, 37.90532684326172, 5.193603515625, 39.03765869140625, 30.462646484375, -8.902688980102539, 51.855621337890625, 109.94317626953125, 110.8193359375, 62.99580383300781, 67.97554016113281, -1.4541015625, 63.393096923828125, 84.72463989257812, 2.8792991638183594, 32.426849365234375, 17.413909912109375, -26.5654296875, -75.38616943359375, 55.3795166015625, -11.989013671875, 58.356468200683594, 4.1380462646484375, 73.33283996582031, 17.90740966796875, 78.58404541015625, 63.195068359375, 23.258544921875, 83.02722930908203, -5.0475311279296875, 33.10357666015625, 76.09022521972656, 55.17808532714844, 6.71661376953125, 18.996551513671875, 88.82235717773438, -27.851318359375, 4.122314453125, -59.7021484375, -6.826967239379883, 88.98350524902344, -50.960052490234375, 105.93572998046875, -9.126968383789062, -58.13043975830078, 27.209976196289062, -44.72412109375, 80.976806640625, 64.64154815673828, -29.03180694580078, -11.405311584472656, 72.35235595703125, 5.7381591796875, -51.22357940673828, 5.794586181640625, -39.992218017578125, 62.92884826660156, 95.91989135742188, 21.455055236816406, 58.0118408203125, -0.25189208984375, 41.02618408203125, -8.264724731445312, 133.68609619140625, -7.7657928466796875, 11.61331558227539, 12.391204833984375, -1.758026123046875, 118.32025146484375, 60.6259765625, 18.3499755859375, -53.00574493408203, 67.02256774902344, 2.3309593200683594, -82.88552856445312, 1.9157791137695312, -14.9927978515625, 19.363616943359375, 66.86072540283203, 80.821044921875, 50.5224609375, -38.693603515625, 0.0, 18.598281860351562, 76.34852600097656, 6.031982421875, 59.785247802734375, -40.01069641113281, 3.41943359375, 70.60299682617188, -16.880630493164062, 32.07173156738281, 86.15629577636719, 8.919082641601562, 86.58938598632812, -3.7225341796875, 13.87451171875, 41.574859619140625, -18.174257278442383, 73.25161743164062, 11.827239990234375, 54.21099853515625, -39.93798828125, 43.018272399902344, 4.92279052734375, 10.783782958984375, 76.94792175292969, 79.86239624023438, 2.44134521484375, 3.509033203125, 92.9224853515625, -5.6235809326171875, 15.826568603515625, 8.509984970092773, 103.74520874023438, 14.20855712890625, 10.929122924804688, 17.88811492919922, -2.0376739501953125, 60.124267578125, 6.641143798828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000369.npy"}
{"epoch": 0.7727748691099476, "step": 370, "batch_size": 128, "mean": 32.54434585571289, "std": 44.54856872558594, "min": -115.9476318359375, "p10": -9.372452545166015, "median": 27.195404052734375, "p90": 88.09686050415038, "max": 127.7562255859375, "pos_frac": 0.8046875, "sample": [29.996994018554688, -69.49246215820312, 86.23434448242188, 24.350311279296875, 51.852203369140625, 4.6046600341796875, 1.6446685791015625, 2.24835205078125, 40.30095672607422, 23.511550903320312, 47.39864730834961, 33.00006103515625, -9.128196716308594, 19.40484619140625, 36.03663635253906, 16.24383544921875, 1.554779052734375, 95.7579345703125, -2.7169876098632812, 37.400146484375, 79.07012939453125, 28.279699325561523, 57.0029296875, 75.52912902832031, 19.74456787109375, 0.685882568359375, -4.952968597412109, 66.92361450195312, 2.4956092834472656, 26.2003173828125, 19.62982177734375, -17.6549072265625, 65.92042541503906, 18.14590835571289, 34.76286315917969, 3.4675750732421875, 127.7562255859375, 46.70463562011719, 3.699420928955078, 104.96847534179688, 39.481651306152344, -83.00672912597656, 73.39459228515625, 7.064496994018555, 59.785545349121094, 51.874603271484375, 61.76646423339844, 26.768936157226562, 24.09149169921875, 0.30988311767578125, 69.17565155029297, 13.797271728515625, 9.544898986816406, 52.39851379394531, 60.889892578125, -2.10357666015625, 72.43135070800781, -5.94549560546875, 93.63422393798828, -59.40459442138672, -14.747482299804688, 86.96981811523438, -4.7086181640625, -31.01513671875, 3.7036705017089844, 80.90538024902344, 119.05828857421875, 39.38011169433594, 60.96647644042969, -22.983076095581055, 1.43603515625, -5.914276123046875, 37.95703125, 92.76571655273438, 70.15216064453125, -8.518310546875, 97.484375, 94.46218872070312, 26.90869140625, 90.60749816894531, 87.80242919921875, 52.39964294433594, 10.953765869140625, 125.48199462890625, 75.80583953857422, 83.39323425292969, -15.775199890136719, 7.782257080078125, -72.46507263183594, 88.78386688232422, 58.41761016845703, 14.704069137573242, -44.005859375, 27.48211669921875, 121.11691284179688, -7.317127227783203, 26.750381469726562, 14.636917114257812, 46.69773864746094, 86.593017578125, 81.38996887207031, 80.156005859375, 35.4244384765625, 74.541259765625, -7.596282958984375, 55.87415313720703, 47.75279235839844, 60.4327278137207, 23.63629150390625, 50.4073486328125, 112.99319458007812, -115.9476318359375, 8.576126098632812, -9.9423828125, 1.2126846313476562, 86.39717102050781, 3.8011474609375, 6.929103851318359, 52.3194580078125, 13.327102661132812, 8.379270553588867, -47.7369384765625, 15.060340881347656, 81.41757202148438, 4.437065124511719, -1.000335693359375, 53.566810607910156, -8.77325439453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000370.npy"}
{"epoch": 0.774869109947644, "step": 371, "batch_size": 128, "mean": 26.336973190307617, "std": 41.977909088134766, "min": -92.49169921875, "p10": -20.086799621582024, "median": 21.76895523071289, "p90": 85.97867279052734, "max": 126.5108642578125, "pos_frac": 0.7421875, "sample": [9.816291809082031, 50.06029510498047, 35.79693603515625, 72.39891815185547, -10.11545181274414, -0.049896240234375, 97.6353759765625, 12.38177490234375, 3.8724212646484375, 14.138946533203125, 2.2830276489257812, -82.70895385742188, 22.63848876953125, 4.729156494140625, 17.96649169921875, -5.6806640625, -35.39691162109375, 48.395111083984375, 4.233551025390625, 9.82391357421875, -6.347412109375, 34.46238708496094, -9.044113159179688, 99.4913330078125, 14.130462646484375, 56.442626953125, -0.393798828125, 43.96685791015625, -48.15672302246094, 33.19087219238281, 58.070709228515625, 87.69938659667969, -18.377578735351562, 55.26710510253906, -92.49169921875, 34.58624267578125, -24.074981689453125, -0.8891677856445312, 57.9908447265625, 28.968505859375, 88.18040466308594, 6.7799530029296875, -31.388511657714844, 1.070770263671875, 18.010498046875, 10.93768310546875, 7.76165771484375, 0.0, -14.462905883789062, -2.0032730102539062, 97.49850463867188, 88.70796203613281, 48.3970947265625, -57.15313720703125, -72.62689208984375, -4.3817138671875, 46.38575744628906, 107.3939208984375, 83.59188842773438, 9.004173278808594, 35.70979309082031, 54.54058837890625, 5.1051483154296875, 120.22036743164062, 34.30609130859375, 19.3779296875, 32.49658203125, 47.74281311035156, 89.15784454345703, -30.775619506835938, 29.525684356689453, 62.65277099609375, 74.63754272460938, -48.65386962890625, 24.563148498535156, 70.1871109008789, 64.78843688964844, 37.23661804199219, 52.953857421875, 8.3751220703125, -9.26019287109375, -39.588714599609375, 16.801483154296875, 126.5108642578125, 25.64755630493164, -27.177154541015625, -37.94115447998047, 65.32018280029297, 42.64898681640625, 19.891334533691406, 2.62493896484375, -11.51983642578125, 58.399139404296875, 34.67164611816406, 22.571456909179688, 1.0994796752929688, 68.61636352539062, 37.5455322265625, 20.966453552246094, 104.838623046875, 15.562288284301758, -4.156442642211914, 45.9888916015625, 18.922866821289062, 78.99673461914062, 74.194580078125, -2.2778778076171875, 22.96722412109375, 20.13763427734375, 27.449920654296875, 36.20304870605469, -17.760299682617188, 31.782684326171875, 30.967390060424805, -7.21722412109375, 85.53390502929688, 87.01646423339844, 2.278900146484375, 79.32254028320312, 82.25372314453125, 12.235458374023438, -13.980010986328125, 2.42437744140625, -2.4946823120117188, 18.77477264404297, 69.02264404296875, 62.93780517578125, 98.81571960449219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000371.npy"}
{"epoch": 0.7769633507853403, "step": 372, "batch_size": 128, "mean": 38.65019226074219, "std": 43.91731643676758, "min": -82.4130859375, "p10": -13.503044128417969, "median": 38.069732666015625, "p90": 95.6415786743164, "max": 125.20654296875, "pos_frac": 0.7890625, "sample": [84.65576171875, -2.647247314453125, 60.871124267578125, 44.14130783081055, -1.5360240936279297, 78.4810791015625, 68.25520324707031, -2.604766845703125, 57.34785461425781, 95.03684997558594, 33.289520263671875, 55.66481018066406, 75.3922119140625, 61.820465087890625, -13.338470458984375, 64.62014770507812, 15.983261108398438, 103.949951171875, -0.2958412170410156, 13.590560913085938, 60.418487548828125, 30.642189025878906, 84.13851928710938, 16.891204833984375, -14.826324462890625, 58.317169189453125, 90.31192016601562, 42.24565124511719, 78.14515686035156, 76.43875122070312, -13.340240478515625, -16.0528564453125, 101.36453247070312, -19.20135498046875, 57.7880859375, 55.42286682128906, 114.80389404296875, -10.940628051757812, 31.13359832763672, 55.749412536621094, -12.355262756347656, 103.53387451171875, 28.3785400390625, 46.99822235107422, 109.485595703125, -56.39422607421875, 84.524169921875, 65.88946533203125, 6.151527404785156, 46.421539306640625, 7.1982879638671875, 25.311782836914062, 45.41162109375, 0.155975341796875, 2.289031982421875, 27.084503173828125, 60.6754150390625, 86.91299438476562, 56.66712951660156, 27.68218994140625, 65.66656494140625, 24.070449829101562, 63.02806091308594, 13.970611572265625, 5.307273864746094, 67.89390563964844, 79.1668701171875, 12.95172119140625, 97.0526123046875, -0.8239593505859375, -2.113311767578125, 38.391937255859375, 10.610992431640625, 80.91192626953125, 98.60400390625, 113.7774658203125, -44.062835693359375, 108.61016845703125, 34.70257568359375, -82.4130859375, 78.64710998535156, 35.318939208984375, 34.38671875, 53.23162841796875, -48.598419189453125, 26.482711791992188, 42.062957763671875, 32.96604919433594, -6.385591506958008, 82.98310852050781, -14.501075744628906, 3.8520355224609375, 77.94364929199219, 42.491180419921875, 35.9080810546875, -68.75485229492188, -20.578079223632812, 30.668914794921875, 66.94691467285156, -13.882919311523438, 84.8746337890625, -0.4869842529296875, 7.128997802734375, 13.764404296875, 82.74476623535156, 66.71273803710938, -77.90960693359375, 99.9090576171875, 125.20654296875, 101.36669921875, 14.2979736328125, 114.6259765625, 91.64306640625, 0.6033859252929688, 42.250396728515625, 94.74273681640625, -1.6677303314208984, 32.8653564453125, -41.56414794921875, 76.16328430175781, 37.747528076171875, 73.16378784179688, 12.276464462280273, 6.630401611328125, -0.13496017456054688, 35.2625732421875, 45.130645751953125, 19.233673095703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000372.npy"}
{"epoch": 0.7790575916230367, "step": 373, "batch_size": 128, "mean": 32.19346618652344, "std": 43.90812301635742, "min": -122.1951904296875, "p10": -13.985373306274413, "median": 25.39719009399414, "p90": 93.14443664550781, "max": 122.11453247070312, "pos_frac": 0.796875, "sample": [7.5513916015625, 24.451271057128906, 86.81048583984375, 33.17448425292969, 0.6151371002197266, 6.1541595458984375, 16.08251953125, 107.67263793945312, 14.2244873046875, 66.49697875976562, 77.6729736328125, -9.190155029296875, 18.54608154296875, 93.23860168457031, 64.98834228515625, 42.977935791015625, 0.5407257080078125, 33.28900146484375, -22.209381103515625, 52.2587890625, 22.044921875, 5.299352645874023, 41.45069885253906, 38.136474609375, -31.0037841796875, 20.5758056640625, -2.9507369995117188, -8.973464965820312, 28.331298828125, 114.28897094726562, 4.483604431152344, -11.597335815429688, 41.954833984375, 17.46575927734375, 24.0728759765625, 62.48663330078125, 89.44329833984375, -9.878997802734375, 80.98345947265625, 49.707305908203125, 33.41741943359375, 74.73368072509766, 44.59002685546875, 24.06726837158203, 6.50860595703125, 65.10821533203125, 5.6705474853515625, -19.77252197265625, 32.8070068359375, 108.68035888671875, -24.085208892822266, 58.6270751953125, -63.48497772216797, -23.66656494140625, 49.357177734375, -13.78411865234375, 119.64715576171875, -5.229400634765625, 26.343109130859375, 33.509796142578125, 9.051673889160156, -122.1951904296875, 122.11453247070312, 36.474395751953125, 71.85836791992188, 10.01910400390625, -28.371627807617188, 91.03207397460938, 6.925537109375, 69.32592010498047, 69.00314331054688, 64.31836700439453, 8.502700805664062, 27.192657470703125, 54.1884765625, 24.120803833007812, 51.91363525390625, 0.7825927734375, 34.10044860839844, -9.076644897460938, -59.89190673828125, -8.876983642578125, 99.515869140625, 13.309471130371094, 31.123016357421875, 95.695556640625, 63.057830810546875, 88.7177734375, 104.73422241210938, 42.573333740234375, -8.446044921875, 93.10408020019531, 2.094482421875, 12.41510009765625, 85.92106628417969, 97.835693359375, 18.365142822265625, 99.908447265625, -41.34450912475586, -51.97267150878906, 39.99122619628906, 89.01691436767578, 97.40072631835938, 83.30841064453125, 13.225860595703125, 11.995086669921875, 66.09660339355469, -9.9822998046875, 110.83828735351562, 12.469635009765625, 9.496295928955078, 4.184059143066406, 5.594329833984375, 15.717994689941406, -18.96628189086914, 3.6492156982421875, 53.359580993652344, 2.3251495361328125, -12.164724349975586, 72.13236999511719, -2.8631362915039062, 85.32940673828125, -14.454967498779297, 11.98681640625, 91.14480590820312, 71.81671142578125, 31.39929962158203, 2.88427734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000373.npy"}
{"epoch": 0.7811518324607329, "step": 374, "batch_size": 128, "mean": 25.736183166503906, "std": 44.21866989135742, "min": -105.21902465820312, "p10": -25.334074401855467, "median": 18.07769775390625, "p90": 86.06190643310546, "max": 132.63995361328125, "pos_frac": 0.765625, "sample": [128.6466064453125, 0.0, 3.193603515625, 132.63995361328125, 4.048583984375, -43.574432373046875, -18.684349060058594, 18.749298095703125, 98.99555969238281, 59.88933563232422, 15.821304321289062, 121.36386108398438, 0.226318359375, -22.069915771484375, 19.190155029296875, 28.215576171875, 39.00294494628906, 64.2913818359375, 18.373687744140625, 9.531219482421875, -56.57502746582031, -31.056304931640625, 50.93952178955078, -33.57171630859375, 88.72218322753906, 37.460899353027344, 10.291839599609375, 9.575927734375, 3.1912460327148438, -19.9876708984375, 10.363021850585938, 18.417800903320312, -10.742279052734375, 29.0430908203125, -10.08575439453125, 21.726287841796875, 4.216522216796875, -14.44598388671875, 101.810791015625, 81.17333984375, 32.0986328125, 9.09521484375, 118.97000122070312, 2.673675537109375, 49.509735107421875, 2.043548583984375, 1.891876220703125, 33.7037353515625, -105.21902465820312, -11.775962829589844, 31.459426879882812, -26.381393432617188, 59.212799072265625, 11.323287963867188, 10.150146484375, 38.80670166015625, 18.0001220703125, 46.42523193359375, 65.15983581542969, 0.4084587097167969, 8.55670166015625, 107.61376953125, 100.92495727539062, 4.1962890625, -1.8922576904296875, 23.928184509277344, 47.48980712890625, 43.80377197265625, -16.793182373046875, 66.30146789550781, 15.736968994140625, 41.00732421875, 71.47162628173828, 55.40771484375, -60.0384521484375, 21.69720458984375, -34.7149658203125, 18.3858642578125, 52.689453125, 78.30760192871094, 66.01911163330078, 6.79437255859375, 18.1552734375, 77.10470581054688, 0.0, 86.98515319824219, 4.722675323486328, -14.3037109375, -24.885223388671875, -39.91357421875, 43.870361328125, 43.96113586425781, 50.409027099609375, 60.51655578613281, 13.603729248046875, -12.982086181640625, 115.61050415039062, 13.678375244140625, 109.83541870117188, 40.86309814453125, 15.11669921875, -33.28932189941406, -42.6337890625, 41.199005126953125, 17.41994285583496, 3.9341087341308594, 56.29840087890625, 85.66622924804688, 8.260848999023438, -29.924209594726562, 4.067535400390625, 9.0330810546875, 4.0157623291015625, 16.218231201171875, 81.5142822265625, -1.033538818359375, 36.606475830078125, 13.03082275390625, -20.672210693359375, 60.156524658203125, 128.48049926757812, 77.43789672851562, 84.1639404296875, 23.853302001953125, 19.723388671875, -3.3273353576660156, 49.539581298828125, -70.62371826171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000374.npy"}
{"epoch": 0.7832460732984293, "step": 375, "batch_size": 128, "mean": 29.269716262817383, "std": 45.48247146606445, "min": -78.57058715820312, "p10": -20.16631164550781, "median": 20.03759002685547, "p90": 93.06687622070312, "max": 136.6878662109375, "pos_frac": 0.75, "sample": [42.30610656738281, 5.328590393066406, 66.97926330566406, -2.7339935302734375, 29.36669921875, 3.8800888061523438, 16.472747802734375, -51.93141174316406, 5.80865478515625, 22.482452392578125, 99.99139404296875, 0.9821605682373047, 79.19285583496094, 83.58197021484375, 6.58404541015625, -11.874755859375, 47.497711181640625, 60.42230224609375, 7.803630828857422, -0.26647186279296875, 95.72298431396484, 94.80203247070312, -10.100936889648438, 4.2721099853515625, 28.827598571777344, 70.04637145996094, -21.488616943359375, 5.499755859375, -11.20428466796875, 49.70458984375, 12.768074035644531, 21.296573638916016, -3.7301025390625, -62.198089599609375, 0.4816093444824219, 84.54534912109375, 44.201171875, 85.04483032226562, 21.98548126220703, 9.837493896484375, 15.864578247070312, 9.3140869140625, -40.41224670410156, -2.4344482421875, -78.57058715820312, 78.69398498535156, 70.51889038085938, 52.819091796875, 25.282875061035156, -66.45599365234375, -59.76930236816406, -7.50341796875, 55.08758544921875, -0.4523792266845703, 45.35614776611328, 0.4052276611328125, 74.6040267944336, -8.25408935546875, -21.547027587890625, -54.057098388671875, 75.19027709960938, 93.02349853515625, -12.29901123046875, 122.37335205078125, 11.101978302001953, 23.006103515625, -11.350723266601562, 9.173088073730469, 129.80523681640625, -2.358875274658203, 20.536788940429688, 8.292230606079102, 10.405715942382812, 53.614654541015625, 36.54541015625, 110.302001953125, 40.1060791015625, -28.706390380859375, 32.46636962890625, 33.899444580078125, 44.4580078125, 44.220001220703125, -11.87689208984375, -13.163070678710938, 1.917144775390625, 50.845977783203125, 16.40826416015625, 74.82821655273438, 20.755096435546875, 87.31707763671875, -63.334312438964844, 97.00198364257812, 117.79156494140625, -3.9271697998046875, 109.255126953125, 136.6878662109375, -32.410980224609375, 84.97096252441406, 55.047969818115234, 86.97714233398438, -5.987308502197266, -19.599609375, 94.53620910644531, 3.3477935791015625, 80.1029052734375, 3.471986770629883, 64.94332885742188, 93.1680908203125, 21.02569580078125, 2.53717041015625, -2.6156768798828125, 109.34210205078125, 50.800384521484375, 43.325157165527344, 87.34841918945312, -22.75818634033203, 2.8910350799560547, 88.43505096435547, 2.0938720703125, 19.040069580078125, 19.53839111328125, 12.926872253417969, 63.546966552734375, 49.98542785644531, 7.924041748046875, 71.6246337890625, 6.447113037109375, 19.506515502929688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000375.npy"}
{"epoch": 0.7853403141361257, "step": 376, "batch_size": 128, "mean": 27.95077133178711, "std": 45.82789993286133, "min": -81.32162475585938, "p10": -30.041677093505857, "median": 27.175979614257812, "p90": 92.7414794921875, "max": 143.14544677734375, "pos_frac": 0.71875, "sample": [32.84075927734375, 27.171722412109375, 94.7259521484375, 126.99380493164062, 65.6175537109375, 143.14544677734375, 27.18023681640625, 28.359661102294922, 40.69227600097656, -30.66278076171875, 71.62518310546875, -22.413131713867188, -41.203094482421875, 18.175872802734375, -9.680633544921875, 28.965103149414062, -1.3041839599609375, 34.509002685546875, 62.72052001953125, 83.38833618164062, 51.805450439453125, -77.42160034179688, 10.350624084472656, 14.23480224609375, 46.2919921875, 12.656024932861328, -17.1710205078125, 96.61785888671875, -6.4200286865234375, 48.75634765625, 44.937255859375, 5.6640167236328125, 86.9295883178711, 30.515594482421875, -2.792877197265625, -20.94085693359375, 63.757537841796875, 31.369285583496094, 101.09625244140625, 1.768310546875, 0.0, 16.882972717285156, 14.184173583984375, 58.471214294433594, -35.619224548339844, 109.68893432617188, -37.468902587890625, 68.20526123046875, 23.80111312866211, 41.3121337890625, -65.34666442871094, 53.16168212890625, 32.43267822265625, 7.607421875, 81.1923828125, 3.31982421875, 6.163604736328125, 36.713775634765625, -20.947357177734375, -10.60504150390625, 65.30950164794922, -1.402984619140625, 109.86752319335938, 96.59646606445312, 16.632965087890625, 120.6600341796875, 100.04803466796875, 17.99443817138672, -20.161590576171875, -23.539581298828125, 67.66461181640625, -16.50384521484375, 24.078582763671875, -17.108360290527344, -4.54383659362793, -1.8111572265625, -46.7142333984375, 8.641777038574219, -56.28631591796875, 93.36666870117188, 57.345001220703125, 54.8740234375, 91.81106567382812, 22.16070556640625, 94.91770935058594, 60.7099609375, 5.69066047668457, -81.32162475585938, 11.717010498046875, -47.10902404785156, 39.5802001953125, 1.9640731811523438, 74.1446304321289, -8.980709075927734, 0.8800106048583984, 73.68512725830078, 44.30145263671875, 0.659942626953125, -32.694061279296875, 92.47354125976562, -3.279052734375, 66.57904052734375, -29.775489807128906, 73.58793640136719, 34.95640563964844, 10.63266372680664, 47.150421142578125, 41.2320556640625, 37.55351257324219, 81.59906005859375, -11.0654296875, -32.887237548828125, -20.34195327758789, 27.277923583984375, 68.80056762695312, 1.4116706848144531, 26.75714111328125, -10.239936828613281, 43.42852783203125, 1.1897811889648438, 4.726423263549805, 35.44544982910156, 39.575531005859375, 33.1485595703125, 86.95448303222656, -48.42578125, 116.70956420898438, 79.396240234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000376.npy"}
{"epoch": 0.787434554973822, "step": 377, "batch_size": 128, "mean": 26.03246307373047, "std": 43.96390914916992, "min": -72.35394287109375, "p10": -23.140042114257813, "median": 16.776874542236328, "p90": 84.83597946166991, "max": 127.30169677734375, "pos_frac": 0.7109375, "sample": [103.442138671875, 47.84855651855469, 65.04801940917969, -9.554153442382812, 27.74749755859375, 12.242782592773438, 5.6930389404296875, 51.8052978515625, 60.86616516113281, -2.023590087890625, 15.80377197265625, 14.659423828125, 39.303009033203125, -39.7587890625, 67.97247314453125, 58.24769592285156, 65.80776977539062, -16.7265625, 0.8040428161621094, -19.232208251953125, 79.71896362304688, 0.2464141845703125, 4.219085693359375, 16.288936614990234, -11.997129440307617, -27.26903533935547, 40.883331298828125, -32.67950439453125, 88.45693969726562, 46.471717834472656, 56.232887268066406, -12.64178466796875, 59.89410400390625, -72.35394287109375, 11.696075439453125, 63.63188171386719, 4.432548522949219, -1.75616455078125, -6.787628173828125, 22.864349365234375, 24.8170166015625, 62.90525436401367, 15.405509948730469, -50.12994384765625, 55.5718994140625, 7.92242431640625, 114.757080078125, 46.9254150390625, -30.258621215820312, -51.664306640625, 7.56597900390625, 107.54605102539062, -12.042228698730469, 0.084808349609375, 104.52667236328125, 2.9027481079101562, -18.086807250976562, -5.3190155029296875, 49.3613395690918, 83.68036651611328, 87.53240966796875, 32.784912109375, 79.64105224609375, 59.0087890625, -23.474822998046875, 81.51141357421875, 64.55888366699219, 0.0, -23.080413818359375, 34.96583557128906, 59.020599365234375, -4.5301513671875, 38.63175582885742, 71.10000610351562, -3.4640121459960938, 2.88153076171875, 7.348541259765625, 8.203338623046875, -0.3186759948730469, 72.74224853515625, 17.12542724609375, 122.96859741210938, 4.29827880859375, -21.77996826171875, -12.665496826171875, -14.153495788574219, 20.826171875, 16.428321838378906, 55.04296112060547, 19.57025146484375, 32.72613525390625, 23.107452392578125, 7.59320068359375, 68.12812805175781, -23.2791748046875, 100.76219177246094, 11.698318481445312, 18.01617431640625, 36.60284423828125, 71.70524597167969, 52.2061767578125, -55.24171447753906, -9.460540771484375, 27.318862915039062, 75.63519287109375, -14.335891723632812, 123.83294677734375, 23.37286376953125, 5.15240478515625, 79.8643798828125, 9.929801940917969, 9.17572021484375, 23.894866943359375, 27.9427490234375, 94.99127197265625, 73.7325439453125, 93.35404968261719, -70.742919921875, -10.4923095703125, 127.30169677734375, 13.907402038574219, -2.51409912109375, -44.52879333496094, 112.104736328125, -21.928802490234375, 4.3707275390625, -61.6202392578125, 49.127349853515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000377.npy"}
{"epoch": 0.7895287958115184, "step": 378, "batch_size": 128, "mean": 25.161815643310547, "std": 41.79407501220703, "min": -67.45767211914062, "p10": -24.02000732421875, "median": 19.501102447509766, "p90": 81.14633255004883, "max": 124.4466552734375, "pos_frac": 0.703125, "sample": [58.972808837890625, 6.88934326171875, 10.375030517578125, 64.20648193359375, -14.030548095703125, 4.186040878295898, -23.91314697265625, 5.7242431640625, 69.08270263671875, 48.03868103027344, -5.4788818359375, 19.875526428222656, 19.126678466796875, 72.38959503173828, 69.8499755859375, -37.142120361328125, -67.45767211914062, 107.3441162109375, 111.48489379882812, -2.289398193359375, 27.139923095703125, 32.26542663574219, 8.276321411132812, -30.36579132080078, 61.701416015625, 57.7298583984375, 120.22470092773438, 81.2684555053711, 21.264923095703125, 64.78266906738281, -2.2196426391601562, 46.29963684082031, 57.48020935058594, 27.273040771484375, 11.6566162109375, -58.131011962890625, 1.365732192993164, 2.495025634765625, 38.35792541503906, 67.38796997070312, 0.0, 88.99304962158203, -7.27166748046875, 75.72451782226562, 95.14297485351562, -41.82562255859375, -3.064006805419922, 124.4466552734375, -10.743278503417969, 82.7978515625, 29.072998046875, 58.686092376708984, -0.0782623291015625, 13.41162109375, -0.63580322265625, 10.521697998046875, -18.87030029296875, 54.400726318359375, -16.784576416015625, 72.93270874023438, 58.070068359375, -1.5581836700439453, 75.43494415283203, 67.17034912109375, 41.852203369140625, 37.3074951171875, -0.19020843505859375, -37.157562255859375, 80.14347839355469, 31.326950073242188, 25.585479736328125, 11.49139404296875, -2.24053955078125, -2.267974853515625, -57.535400390625, 81.31695556640625, -9.154922485351562, -27.556854248046875, 25.304168701171875, -0.4312896728515625, 21.164276123046875, 0.4230804443359375, 81.093994140625, 33.046329498291016, 3.54766845703125, 23.699798583984375, 66.0494384765625, 3.3564910888671875, 11.859039306640625, 21.746002197265625, 66.24848937988281, 57.827392578125, -12.394683837890625, 10.412734985351562, 40.16064453125, -17.24427032470703, -66.0147705078125, 10.2864990234375, 11.07366943359375, -27.416458129882812, -20.145263671875, 97.08795166015625, -8.950439453125, 0.278472900390625, -63.49922180175781, -0.4459075927734375, 21.064788818359375, 58.42668914794922, -0.42852783203125, 22.5081787109375, 22.242706298828125, 36.145172119140625, 3.8997154235839844, 13.645416259765625, 112.42724609375, 75.67205810546875, 91.95892333984375, 62.51860046386719, 91.28556060791016, 14.984786987304688, -39.26885986328125, 31.375572204589844, 69.72366333007812, 4.09283447265625, 45.70188903808594, -24.26934814453125, 6.2039794921875, 0.29486083984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000378.npy"}
{"epoch": 0.7916230366492146, "step": 379, "batch_size": 128, "mean": 28.241214752197266, "std": 44.91086196899414, "min": -86.90179443359375, "p10": -28.562707519531248, "median": 24.183025360107422, "p90": 90.55835266113282, "max": 121.69537353515625, "pos_frac": 0.765625, "sample": [-6.511960983276367, 7.193382263183594, 16.2513427734375, 30.278472900390625, 2.2499465942382812, -1.6246337890625, 2.9377212524414062, 82.10791015625, 16.427658081054688, 2.1159744262695312, 84.17257690429688, 52.61346435546875, 45.0452880859375, 2.3417510986328125, 8.354248046875, 6.8972930908203125, 25.380027770996094, -46.4462890625, 84.77603149414062, 43.51451110839844, 62.239044189453125, 40.29547119140625, 88.02012634277344, -27.73895263671875, 7.4341888427734375, 4.4398345947265625, 81.00068664550781, 0.58056640625, 47.80047607421875, -2.2953414916992188, 26.179840087890625, 86.25816345214844, 100.22445678710938, 44.094482421875, 34.13238525390625, 6.41094970703125, 21.48621368408203, 96.10183715820312, 77.40655517578125, 37.14604949951172, -16.582962036132812, -20.50872802734375, 19.7276611328125, 14.598392486572266, 7.263336181640625, 110.43203735351562, 11.561286926269531, -60.601776123046875, -7.446807861328125, 10.058502197265625, 52.0843505859375, -6.023201942443848, 25.81158447265625, -60.570343017578125, 2.848663330078125, 22.98602294921875, 96.97332763671875, 83.73367309570312, 66.41458129882812, 56.243011474609375, 33.23077392578125, -30.48480224609375, 16.070892333984375, 41.278289794921875, 78.83700561523438, 121.69537353515625, 9.86480712890625, 92.94488525390625, 5.149078369140625, -5.5023956298828125, -19.115234375, 0.0, 6.6074981689453125, 14.435638427734375, 14.78485107421875, 94.328369140625, 27.3670654296875, 78.17608642578125, 95.47161865234375, 79.56570434570312, 0.5213623046875, 64.99115753173828, 45.306793212890625, -0.9032516479492188, 52.657623291015625, 38.986053466796875, -5.9052886962890625, 49.86531448364258, -16.856689453125, 60.21165466308594, 29.12298583984375, -2.4682693481445312, 105.94509887695312, 4.008161544799805, 57.36834716796875, -6.60955810546875, 3.047262191772461, 36.938507080078125, -58.94416809082031, 120.95782470703125, 84.84268188476562, -86.90179443359375, 16.915618896484375, 32.860382080078125, 94.949462890625, 59.88031005859375, 38.225830078125, -31.2530517578125, 53.142059326171875, 90.50216674804688, 97.6461181640625, 36.05084228515625, 83.68997192382812, 31.350677490234375, -61.052772521972656, -63.90260314941406, 15.252693176269531, 2.5468292236328125, -12.999908447265625, 87.8096923828125, -33.48980712890625, 58.86528778076172, -65.64288330078125, -38.069400787353516, -52.3468017578125, 18.02037239074707, 36.053466796875, 90.689453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000379.npy"}
{"epoch": 0.793717277486911, "step": 380, "batch_size": 128, "mean": 29.314685821533203, "std": 45.12446594238281, "min": -64.3028564453125, "p10": -26.8042121887207, "median": 23.75115203857422, "p90": 91.54317016601561, "max": 163.7901611328125, "pos_frac": 0.7421875, "sample": [72.77273559570312, 40.68965148925781, 49.629058837890625, 11.859237670898438, 54.78851318359375, 13.43341064453125, -53.12242126464844, 2.6605224609375, 1.2497692108154297, -64.3028564453125, 37.72430419921875, 50.863525390625, 71.74285888671875, -51.555030822753906, 17.374267578125, 13.517654418945312, 76.46090698242188, 35.25579833984375, 72.88463592529297, -4.888275146484375, 0.0, 47.2183837890625, 12.067909240722656, 18.51165771484375, 26.782562255859375, 80.47695922851562, 9.661590576171875, 50.8880615234375, -0.835479736328125, 5.95831298828125, 33.501220703125, 112.47015380859375, 11.708847045898438, 42.24847412109375, 10.547821044921875, -28.908348083496094, 68.82817077636719, 14.837139129638672, 13.885292053222656, -16.390560150146484, 35.732025146484375, 36.61967468261719, 92.97576904296875, -27.072998046875, -15.159370422363281, -7.26531982421875, 28.847885131835938, 90.92919921875, 105.45703125, -28.530120849609375, -28.558746337890625, 31.461151123046875, 38.525848388671875, 1.902618408203125, 17.58685302734375, -23.66952133178711, -16.07175064086914, 9.26751708984375, 107.42129516601562, 36.525840759277344, 18.49224853515625, 82.53048706054688, 48.36065673828125, 26.692413330078125, 71.5765380859375, 103.33782958984375, -50.10962677001953, -6.165130615234375, 14.8289794921875, 7.705181121826172, 74.50888061523438, 113.0133056640625, 3.1606216430664062, 55.087425231933594, 95.80927276611328, 2.5564117431640625, 20.586700439453125, -30.99237060546875, 45.67828369140625, -55.05556106567383, 36.13362121582031, 163.7901611328125, -15.474372863769531, 36.511810302734375, 48.183685302734375, 22.202865600585938, 82.71878051757812, 69.24356079101562, 87.64892578125, 50.5322265625, 26.037384033203125, 135.87322998046875, -53.71014404296875, 35.58026123046875, -1.2410926818847656, 38.09809875488281, 1.68389892578125, -29.11273193359375, 109.33154296875, 133.52291870117188, 30.538043975830078, 36.92436218261719, 29.10375213623047, 82.08853149414062, 19.496673583984375, 18.01490020751953, 20.58197021484375, 23.006698608398438, -17.99346923828125, -33.28770446777344, -1.25335693359375, -20.979156494140625, 27.21246337890625, 9.341049194335938, 19.208511352539062, 63.907440185546875, -18.101348876953125, 55.515228271484375, 50.70603942871094, 142.78662109375, 24.49560546875, -11.86773681640625, -14.32861328125, 64.9248046875, -5.7614898681640625, -26.68901824951172, 119.09329223632812, -2.9824066162109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000380.npy"}
{"epoch": 0.7958115183246073, "step": 381, "batch_size": 128, "mean": 29.889774322509766, "std": 43.034053802490234, "min": -63.948822021484375, "p10": -17.982231903076173, "median": 18.487255096435547, "p90": 90.51167297363281, "max": 117.78550720214844, "pos_frac": 0.7265625, "sample": [66.21566772460938, 111.9144287109375, 87.47515869140625, 15.192222595214844, 70.86393737792969, -12.385528564453125, 90.51882934570312, 97.74122619628906, 57.06068420410156, 3.9376964569091797, 112.20455932617188, 24.84136962890625, 7.706268310546875, 80.59686279296875, 69.61032104492188, 82.98483276367188, 0.0, 10.686920166015625, 13.85443115234375, 36.41947937011719, 67.99247741699219, 47.896728515625, 4.5989532470703125, 117.78550720214844, 38.7298583984375, 8.698196411132812, 53.829124450683594, -63.38294982910156, 0.0, -63.948822021484375, 58.52899169921875, 50.44182586669922, -16.6656494140625, 67.28253173828125, -5.529594421386719, 15.549690246582031, 8.150299072265625, 38.006256103515625, 9.32611083984375, 52.9993896484375, 106.5726318359375, 36.04609680175781, -9.8116455078125, 61.93450927734375, 92.8701171875, 8.20465087890625, 92.69247436523438, -4.408313751220703, -48.068809509277344, 63.320465087890625, 82.70372009277344, 82.78846740722656, 32.85838317871094, -3.5802154541015625, 54.760223388671875, 66.4312744140625, 9.38458251953125, 71.11302185058594, 94.47259521484375, 33.89244079589844, -38.70954895019531, -1.13922119140625, 26.553192138671875, 8.230316162109375, 3.585052490234375, 16.15521812438965, 85.04845428466797, 1.0927734375, -8.310684204101562, -0.579559326171875, -10.146553039550781, 7.609334945678711, -10.9923095703125, 75.32699584960938, -12.84381103515625, 28.333526611328125, -2.9523544311523438, 24.141189575195312, 22.739944458007812, -18.556060791015625, 90.00326538085938, -17.736305236816406, 72.50611877441406, 28.8260498046875, 78.75482177734375, -5.0460205078125, -24.53460693359375, -5.121360778808594, 92.98915100097656, 10.07476806640625, -5.06890869140625, -36.740020751953125, 79.01496124267578, 66.14317321777344, 58.83917236328125, -36.61004638671875, 0.08431625366210938, 3.70196533203125, 7.8133544921875, -41.47272491455078, 47.015647888183594, 5.521553039550781, -55.8502197265625, 49.00011444091797, -26.835968017578125, 69.20185852050781, -10.753173828125, 6.326801300048828, -18.656845092773438, 72.15011596679688, -0.54547119140625, 105.91258239746094, 10.921051025390625, 112.77532958984375, 18.639190673828125, 18.33531951904297, 14.081413269042969, 7.476341247558594, 69.21878051757812, 31.798019409179688, 90.50860595703125, 93.6036376953125, 46.522071838378906, -15.037933349609375, -42.068084716796875, 52.2567138671875, 15.148040771484375, 6.313446044921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000381.npy"}
{"epoch": 0.7979057591623037, "step": 382, "batch_size": 128, "mean": 36.08700942993164, "std": 44.44522476196289, "min": -60.926513671875, "p10": -19.57288436889648, "median": 37.298099517822266, "p90": 93.25289154052734, "max": 142.59133911132812, "pos_frac": 0.765625, "sample": [-32.97515869140625, 9.960868835449219, 48.911163330078125, 23.385223388671875, 68.78936767578125, 83.16896057128906, 53.29345703125, -1.887969970703125, 5.563268661499023, 64.6623764038086, 41.1318359375, -30.94342041015625, 66.0211181640625, 9.659721374511719, 110.76727294921875, 42.510223388671875, 36.57801055908203, 87.31257629394531, 78.3897705078125, -10.157257080078125, 36.37811279296875, -56.13536071777344, 84.96174621582031, -18.212677001953125, -9.053329467773438, 80.29653930664062, 24.60980224609375, -19.076744079589844, 34.963958740234375, 47.861778259277344, 2.4613037109375, 41.05323791503906, 19.2803955078125, 25.203956604003906, 63.57720947265625, 40.91614532470703, -3.1937179565429688, 33.969383239746094, -6.71866512298584, 94.86184692382812, 25.063804626464844, 35.326385498046875, 98.35281372070312, 15.143585205078125, 96.387939453125, 88.33831787109375, 93.52189636230469, 38.0181884765625, 17.106201171875, 5.000495910644531, 92.52194213867188, 36.31678771972656, 10.573257446289062, 49.0787353515625, 3.77349853515625, 52.86175537109375, 142.59133911132812, 111.87173461914062, -20.730545043945312, -4.5570831298828125, -15.483734130859375, -58.96196746826172, 138.77313232421875, 23.74578857421875, -1.2569122314453125, 59.7472038269043, 103.44143676757812, 27.0958251953125, 93.13760375976562, 19.631622314453125, 103.58877563476562, 11.309799194335938, 11.442031860351562, -28.66558837890625, 82.81814575195312, 56.197296142578125, 72.21517944335938, 12.185348510742188, 31.430343627929688, 87.92230224609375, 25.1007080078125, -23.8021240234375, 39.781158447265625, 67.89234924316406, -21.274032592773438, 12.797332763671875, -54.544769287109375, 88.7435302734375, -11.774627685546875, 74.73480224609375, 78.315185546875, -16.761138916015625, 9.743484497070312, 96.08897399902344, 89.67117309570312, 29.1197509765625, 61.90729522705078, -60.926513671875, 1.698944091796875, -12.518310546875, 30.13323974609375, 48.5628662109375, 41.95872497558594, 78.8673095703125, 43.607513427734375, -39.954559326171875, 19.86469268798828, 63.51115798950195, -57.10260772705078, 57.25993347167969, 52.0640869140625, -0.0614013671875, 62.06329345703125, 52.619232177734375, 40.14759063720703, 43.23173522949219, -44.22314453125, 40.541748046875, -14.198966979980469, 63.74593734741211, 97.03607177734375, -13.645313262939453, 111.17373657226562, 52.71260070800781, 91.53289794921875, 56.46234130859375, 87.01148986816406, -8.7989501953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000382.npy"}
{"epoch": 0.8, "step": 383, "batch_size": 128, "mean": 29.232160568237305, "std": 42.54815673828125, "min": -85.30723571777344, "p10": -19.76662673950195, "median": 21.431320190429688, "p90": 87.01075897216795, "max": 126.32673645019531, "pos_frac": 0.734375, "sample": [7.2087554931640625, 47.15716552734375, 73.70086669921875, 71.06640625, -22.95172882080078, 21.221771240234375, -2.3387451171875, 17.3612060546875, -1.69927978515625, 25.374862670898438, 66.53170776367188, -64.06477355957031, 73.86946105957031, 22.17974853515625, 0.0, -1.6339797973632812, 4.301414489746094, -39.610595703125, 40.642303466796875, -33.16644287109375, 12.77459716796875, 102.01458740234375, 62.86885070800781, 53.50834655761719, -38.71697998046875, 102.85429382324219, -22.161544799804688, -2.3248558044433594, -21.544349670410156, 9.16253662109375, 41.40248489379883, 15.95068359375, -13.573783874511719, -17.128616333007812, 57.94252014160156, 0.1640625, 103.41012573242188, 92.6270751953125, 33.35908508300781, 70.61885070800781, 31.765241622924805, 5.211692810058594, 40.47021484375, 126.32673645019531, -85.30723571777344, -1.85791015625, 36.87675476074219, 85.0655517578125, 63.097747802734375, 7.74456787109375, 83.33447265625, 2.248748779296875, -1.7269344329833984, 79.74252319335938, 10.485366821289062, 21.85748291015625, 6.166908264160156, -51.63987731933594, -0.01047515869140625, -30.19232177734375, 63.56306457519531, 50.853515625, 5.36482048034668, 92.73685455322266, -15.879180908203125, 11.31939697265625, 99.16741943359375, 0.25614166259765625, -4.3757171630859375, 9.200042724609375, 30.942398071289062, 91.54957580566406, 77.04830932617188, 75.201904296875, 70.04388427734375, 23.905048370361328, -28.80865478515625, -1.6377410888671875, 18.25115966796875, 73.03022766113281, 3.9197540283203125, 59.63307189941406, 67.28385925292969, 50.23539733886719, 66.10784912109375, 26.98638916015625, 21.640869140625, 13.543174743652344, 21.94866943359375, -62.59075927734375, 91.98721313476562, 10.243722915649414, 39.254302978515625, 52.65680694580078, 36.84234619140625, 6.202932357788086, 15.4051513671875, 92.29621887207031, 68.470947265625, 58.434967041015625, 80.423095703125, -17.01837158203125, 125.87176513671875, -19.004745483398438, -8.9586181640625, 14.583969116210938, 20.111526489257812, 16.766448974609375, 0.0, 65.31013488769531, 71.61229705810547, -2.1991806030273438, 59.66815185546875, 48.48687744140625, 19.07879638671875, -17.738113403320312, -17.966949462890625, 8.5491943359375, 95.33950805664062, 63.70849609375, 12.666213989257812, 72.39400482177734, 93.84982299804688, -47.165672302246094, 11.62310791015625, -3.5079193115234375, 79.51004028320312, 55.470062255859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000383.npy"}
{"epoch": 0.8020942408376963, "step": 384, "batch_size": 128, "mean": 23.091142654418945, "std": 44.688385009765625, "min": -94.62640380859375, "p10": -28.12866287231445, "median": 15.951011657714844, "p90": 83.39967269897461, "max": 125.30657958984375, "pos_frac": 0.6953125, "sample": [112.29208374023438, -94.62640380859375, 18.451934814453125, -12.23931884765625, -13.267120361328125, 60.992523193359375, 30.83721923828125, 125.30657958984375, 12.845718383789062, 9.781341552734375, 55.094810485839844, -23.23312759399414, -13.783233642578125, -44.41773986816406, 21.04638671875, -59.25042724609375, 86.45379638671875, -2.2006607055664062, 101.81219482421875, -6.3841552734375, 3.2954750061035156, 94.56027221679688, 71.20431518554688, 75.42051696777344, 3.935577392578125, 57.132232666015625, 19.175460815429688, 51.185333251953125, 10.961669921875, -4.232757568359375, 2.392732620239258, -22.20269775390625, 4.23974609375, 25.027557373046875, -31.2149658203125, 74.2406997680664, 7.9266510009765625, 24.052459716796875, 58.407501220703125, -26.80596160888672, 4.4170074462890625, -56.13909912109375, 3.5643310546875, 17.901695251464844, 78.51727294921875, 81.73918151855469, 87.78738403320312, -76.97113800048828, 3.144775390625, 16.217300415039062, 66.42156982421875, 110.84112548828125, -20.98979949951172, 10.93902587890625, -44.74139404296875, 33.21904754638672, 49.87205505371094, 48.924560546875, -42.89567565917969, -8.946258544921875, 47.95860290527344, -61.428680419921875, 67.5767822265625, 60.82330322265625, 106.88348388671875, 35.46099853515625, 17.96136474609375, 46.0025634765625, 49.25439453125, 8.975200653076172, -5.318031311035156, 28.587364196777344, 83.25192260742188, 8.842941284179688, 52.06976318359375, 70.57290649414062, -17.774139404296875, 15.684722900390625, -6.749042510986328, 55.5196533203125, 22.610702514648438, -18.712875366210938, 53.4449577331543, 43.1763916015625, 80.23960876464844, 106.9527587890625, 20.06787109375, 7.426849365234375, 28.912567138671875, 34.437713623046875, 105.578369140625, -3.3813858032226562, 83.74442291259766, 62.942626953125, -4.1800384521484375, 0.4763336181640625, 13.35546875, 88.11866760253906, 14.527191162109375, -8.403083801269531, -4.0287628173828125, -34.49566650390625, 8.05377197265625, -36.802581787109375, -74.451416015625, 6.9174041748046875, -1.664215087890625, 26.12209129333496, 12.442352294921875, 28.03125, 101.24514770507812, -52.716552734375, 49.667694091796875, -1.6817474365234375, -2.8434200286865234, -7.531036376953125, 82.78717041015625, 62.713417053222656, -15.717056274414062, 66.92308807373047, 76.67903900146484, -18.745040893554688, 1.1181068420410156, 15.332275390625, 4.723236083984375, 36.64605712890625, 26.88311767578125, -22.469879150390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000384.npy"}
{"epoch": 0.8041884816753927, "step": 385, "batch_size": 128, "mean": 28.43819808959961, "std": 43.195926666259766, "min": -74.68780517578125, "p10": -23.339244079589843, "median": 22.611534118652344, "p90": 87.97037277221679, "max": 135.44775390625, "pos_frac": 0.734375, "sample": [-29.633453369140625, 35.73137283325195, 87.74124145507812, 74.66156005859375, 29.07940673828125, 22.139892578125, 88.50501251220703, 19.503997802734375, 38.354583740234375, 76.22361755371094, 20.01983642578125, 3.4629669189453125, -33.07441711425781, 17.911941528320312, 23.677032470703125, 9.636459350585938, 49.75933837890625, 65.32137298583984, -35.14990234375, -2.31256103515625, 68.0467529296875, -35.50018310546875, 57.064422607421875, -26.95147705078125, 48.05449676513672, 85.49977111816406, 61.953712463378906, 10.094963073730469, 65.56488037109375, 22.79840087890625, 0.380279541015625, 131.12344360351562, 68.67406463623047, -10.228759765625, 91.95315551757812, 135.44775390625, 53.748321533203125, 24.229217529296875, 52.531044006347656, 120.4820556640625, 4.868072509765625, -23.717666625976562, 39.4896240234375, 41.15032958984375, 43.5906982421875, -58.915679931640625, 68.16334533691406, 72.40428161621094, -66.42913818359375, 52.2664794921875, 112.00076293945312, 100.49938201904297, 130.324462890625, -4.350887298583984, 62.04608917236328, 15.609893798828125, 21.839599609375, 2.7679214477539062, -74.68780517578125, 10.831382751464844, 23.1385498046875, 96.85617065429688, -13.1937255859375, 89.94711303710938, 21.74432373046875, -3.430328369140625, -23.17706298828125, 40.67100524902344, -8.567092895507812, 85.9659423828125, 61.364715576171875, 2.9604415893554688, -18.3526611328125, 6.141338348388672, 22.424667358398438, -4.9596710205078125, 47.708797454833984, 69.55374145507812, 38.99693298339844, -34.06317138671875, 0.6740875244140625, 27.497575759887695, 75.74768829345703, 37.61167907714844, 38.38702392578125, 10.276351928710938, -6.626312255859375, 26.308876037597656, -59.6702880859375, 6.844196319580078, 33.469268798828125, -2.7790870666503906, -1.4005584716796875, 22.070777893066406, 60.4141845703125, 32.23388671875, 0.0, 31.508499145507812, 54.251434326171875, 17.342620849609375, -5.460700988769531, 22.377784729003906, 39.4930419921875, -21.67840576171875, 106.93875122070312, -14.152610778808594, -45.25421142578125, 56.15034103393555, -12.246429443359375, 101.66943359375, 39.3114013671875, 0.0, -7.314788818359375, -32.482810974121094, 68.06180572509766, 2.577117919921875, 50.911651611328125, -21.640480041503906, 11.231475830078125, 9.265350341796875, -9.811408996582031, 6.3431243896484375, 20.64838409423828, 32.964080810546875, 66.40054321289062, 16.94482421875, 107.72152709960938, 8.95172119140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000385.npy"}
{"epoch": 0.806282722513089, "step": 386, "batch_size": 128, "mean": 34.69988250732422, "std": 45.30290603637695, "min": -114.66909790039062, "p10": -11.84096870422363, "median": 30.88085174560547, "p90": 91.68871612548828, "max": 138.61456298828125, "pos_frac": 0.8046875, "sample": [23.486572265625, 2.1576004028320312, 1.57806396484375, 61.69530487060547, 50.17791748046875, 14.36907958984375, 62.3232421875, 96.51715087890625, 52.120697021484375, 48.25154113769531, 84.81816864013672, 119.3145751953125, -28.47381591796875, 31.3446044921875, 91.20712280273438, 18.25440216064453, 22.22509765625, 20.479736328125, 1.107778549194336, 61.326416015625, 37.65327453613281, 82.51335144042969, 21.8031005859375, 44.4949951171875, -9.71014404296875, -74.97421264648438, 32.19390869140625, 30.417098999023438, 83.06787109375, 35.426605224609375, 28.4368896484375, -41.7861328125, 86.55816650390625, 41.29362487792969, -10.605819702148438, 15.91156005859375, 19.478187561035156, 12.851898193359375, 36.58876037597656, 88.80059814453125, 138.61456298828125, 43.96405029296875, 39.39434814453125, 17.915740966796875, 39.694610595703125, 43.576904296875, 102.67376708984375, -40.35479736328125, 31.794021606445312, 109.13461303710938, 28.7132568359375, -11.093597412109375, -43.88812255859375, -28.247901916503906, 110.75180053710938, 52.71453094482422, -4.841651916503906, 0.0, -24.239776611328125, -13.584835052490234, -48.34844970703125, -26.92034912109375, 31.968017578125, 32.357269287109375, 0.6617012023925781, -10.342437744140625, 32.281982421875, -3.690420150756836, 88.88014221191406, -84.37557983398438, 93.42449951171875, 2.6165771484375, 61.730194091796875, 119.26116943359375, 113.13616943359375, 88.7434310913086, 7.399330139160156, -17.757888793945312, 76.24372863769531, 74.15614318847656, 85.02078247070312, 6.2086181640625, -114.66909790039062, 36.33074951171875, 1.228790283203125, 94.24508666992188, 80.95864868164062, 8.60247802734375, 72.19406127929688, 15.981697082519531, 23.84222412109375, 13.71249771118164, -4.2568359375, 0.0, 103.95497131347656, 54.43046569824219, -5.6578369140625, 4.46929931640625, 51.81585693359375, 61.58599853515625, 81.31155395507812, 90.83419036865234, 20.137786865234375, 85.70574951171875, 7.778076171875, 91.25567626953125, 13.36798095703125, -9.332290649414062, 1.12908935546875, 81.48016357421875, 104.80377197265625, 37.85113525390625, 90.57606506347656, 28.286598205566406, 8.945266723632812, 56.762481689453125, 22.19659423828125, 8.783050537109375, 19.85048484802246, 83.67697143554688, 28.82449722290039, 92.69914245605469, 10.397590637207031, 4.5872802734375, 75.39047241210938, 0.0, 68.0015869140625, 27.473724365234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000386.npy"}
{"epoch": 0.8083769633507853, "step": 387, "batch_size": 128, "mean": 32.05836486816406, "std": 46.49814224243164, "min": -82.44561767578125, "p10": -23.322966003417967, "median": 21.032440185546875, "p90": 93.70886764526367, "max": 123.78271484375, "pos_frac": 0.71875, "sample": [19.12677764892578, 14.022636413574219, 76.80828857421875, 17.837852478027344, 4.091888427734375, 101.1754150390625, 100.14833068847656, -9.565338134765625, -73.70263671875, -24.966720581054688, 0.0, 92.99716186523438, -6.018890380859375, 78.902099609375, 51.381927490234375, 79.97509765625, 6.093170166015625, -40.108489990234375, -11.378646850585938, 26.94281005859375, -3.6687164306640625, 3.094390869140625, 83.11250305175781, -38.034393310546875, 14.562828063964844, 93.46041870117188, -47.875518798828125, -6.138519287109375, 92.46415710449219, 0.6416759490966797, 85.47647094726562, -4.269805908203125, 83.62538146972656, 40.030792236328125, 72.24365234375, 73.31692504882812, -12.1148681640625, 15.603302001953125, 49.49491882324219, -8.340301513671875, -26.646942138671875, 75.43609619140625, -10.887435913085938, -9.115753173828125, 104.6646728515625, 22.819076538085938, 103.93594360351562, 19.246368408203125, -10.708494186401367, -0.2776641845703125, 7.680900573730469, 103.95213317871094, -29.183685302734375, -9.548629760742188, 12.07794189453125, 16.838638305664062, 8.390899658203125, -40.56752014160156, 76.50161743164062, 77.29658508300781, 55.4613037109375, 29.88763427734375, 41.950897216796875, 20.1427001953125, 81.39739990234375, -82.44561767578125, 25.889434814453125, 41.2938232421875, 8.570556640625, -10.65652084350586, 11.844284057617188, -10.65716552734375, 39.1824951171875, 87.62646484375, 79.91648864746094, 46.40626525878906, 41.00032043457031, 30.098159790039062, -6.9700775146484375, 28.42327880859375, 15.240959167480469, -1.9121246337890625, 87.97193908691406, 13.0609130859375, 41.850494384765625, 98.67770385742188, 63.68636703491211, 4.404327392578125, 61.20599365234375, 78.54633331298828, 15.977714538574219, 94.28858184814453, 20.112838745117188, 31.167266845703125, 123.78271484375, 65.26025390625, 75.09628295898438, -26.851043701171875, 82.14208984375, 93.09196472167969, 6.882366180419922, 20.457275390625, -22.618499755859375, 78.15291595458984, -6.834682464599609, -46.12347412109375, -74.6737060546875, -11.690790176391602, 76.53742980957031, 119.66229248046875, 10.58111572265625, 7.4827728271484375, 32.275123596191406, 54.69073486328125, 104.44766235351562, 74.97210693359375, 86.22628784179688, 21.60760498046875, 109.24591064453125, 16.543190002441406, -3.1636199951171875, -40.799407958984375, 101.57748413085938, 97.94140625, 2.8127059936523438, 83.36906433105469, -17.700634002685547, 44.093505859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000387.npy"}
{"epoch": 0.8104712041884817, "step": 388, "batch_size": 128, "mean": 23.88802719116211, "std": 44.797462463378906, "min": -105.976318359375, "p10": -27.856752014160158, "median": 13.720096588134766, "p90": 88.11040649414062, "max": 126.0733642578125, "pos_frac": 0.6953125, "sample": [38.0562744140625, -41.21551513671875, 80.61813354492188, 0.0, 80.5946044921875, 76.42927551269531, -3.768524169921875, 98.34024047851562, 9.346641540527344, 93.08920288085938, -35.7769775390625, 77.98895263671875, -45.998565673828125, 0.9104938507080078, 4.330718994140625, 15.6962890625, 8.053054809570312, 50.0965576171875, 114.9429931640625, -15.878448486328125, 70.80711364746094, 126.0733642578125, -44.186614990234375, -7.6142578125, -5.47515869140625, -21.404281616210938, 26.703811645507812, 56.567779541015625, 95.39962768554688, 93.05859375, 12.692306518554688, 13.886299133300781, 0.0, -0.2177734375, 13.55389404296875, -17.62762451171875, -11.72802734375, 11.495758056640625, 45.76171875, 42.43318176269531, -9.069503784179688, 13.026542663574219, 58.60865783691406, 78.63990783691406, -16.952651977539062, -27.939178466796875, 88.00186157226562, -6.867210388183594, -6.464061737060547, 0.798583984375, 46.32926940917969, 6.638397216796875, 57.17572021484375, -27.821426391601562, 9.656082153320312, 65.77552795410156, -15.736099243164062, 90.55245971679688, 30.32550048828125, -20.713241577148438, -19.751724243164062, 27.516403198242188, 77.35411071777344, 16.977325439453125, 12.118083953857422, -36.305419921875, 47.25526428222656, -101.83322143554688, 108.45257568359375, -37.092140197753906, 27.5584716796875, 82.80772399902344, 59.304840087890625, 26.97565460205078, 19.415496826171875, -3.96954345703125, 2.05615234375, 92.32403564453125, 5.006217956542969, -74.56644439697266, -2.978302001953125, 73.99580383300781, 92.12481689453125, -11.996932983398438, 20.171710968017578, 4.538726806640625, 37.91099548339844, 64.85884094238281, 38.68017578125, 21.072017669677734, -26.585205078125, 13.906158447265625, 31.181365966796875, -105.976318359375, 3.9119911193847656, 98.47697448730469, 88.36367797851562, 62.042457580566406, 17.262428283691406, 66.05072021484375, 7.082374572753906, 65.3294677734375, 79.3485107421875, 108.81549072265625, -14.9002685546875, 10.263450622558594, 40.2215576171875, 8.47930908203125, -28.35174560546875, 1.6092548370361328, 74.27554321289062, 10.328704833984375, -5.242584228515625, 12.441162109375, 9.510353088378906, -11.127059936523438, 60.584808349609375, 6.823726654052734, 74.66748046875, 21.17724609375, -39.132774353027344, -32.01567077636719, 29.712860107421875, 6.298095703125, 21.458648681640625, 77.33155822753906, 33.15129852294922, -9.089469909667969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000388.npy"}
{"epoch": 0.812565445026178, "step": 389, "batch_size": 128, "mean": 31.303802490234375, "std": 43.11943054199219, "min": -77.88449096679688, "p10": -19.053482055664062, "median": 28.451709747314453, "p90": 87.51980209350586, "max": 118.71923828125, "pos_frac": 0.75, "sample": [-14.059921264648438, -30.97705078125, -4.982276916503906, 58.72308349609375, 47.434844970703125, 79.8253173828125, 21.370086669921875, 20.98919677734375, -1.67620849609375, 92.7337646484375, 36.245086669921875, 89.76649475097656, 18.13787841796875, 118.71923828125, 15.810462951660156, 83.17399597167969, -0.352294921875, 79.179443359375, 35.95343017578125, 109.53750610351562, -30.4364013671875, 56.39710235595703, 80.018798828125, 55.57464599609375, 1.1213455200195312, 72.00473022460938, -52.417083740234375, 78.28057861328125, 25.899948120117188, 94.57675170898438, 22.693603515625, -62.96441650390625, 67.03912353515625, -22.517921447753906, -7.24041748046875, -4.836273193359375, -28.051334381103516, 57.046302795410156, 99.8736572265625, 58.752471923828125, 81.32662963867188, 0.0, -44.86112976074219, -6.9825439453125, 63.12884521484375, -20.05548095703125, 86.60243225097656, 67.869140625, 83.42167663574219, 13.381805419921875, 5.768165588378906, 86.8787841796875, 17.86395263671875, 5.552036285400391, 61.87726974487305, 30.12005615234375, 7.3936004638671875, 2.246002197265625, -20.22283935546875, -0.2583770751953125, 23.943328857421875, 13.725940704345703, 55.79576110839844, 71.15988159179688, -18.4764404296875, 69.09146881103516, 61.650421142578125, 68.9749755859375, 59.36628723144531, 21.523025512695312, 79.70999145507812, 8.12835693359375, 34.23016357421875, 89.01551055908203, 5.20733642578125, 108.2271728515625, 43.03961181640625, 43.7799072265625, -22.63848876953125, -1.697601318359375, 47.594940185546875, 71.9051513671875, 9.915626525878906, 52.433197021484375, 9.607551574707031, 91.462890625, -75.81964111328125, 70.53085327148438, 33.31610107421875, -12.774444580078125, 102.5054931640625, 30.88336181640625, 7.17437744140625, 2.501392364501953, 18.65857696533203, 39.00849151611328, 6.544586181640625, -77.88449096679688, -18.624053955078125, 98.2685546875, -7.9178314208984375, 57.736328125, -11.27923583984375, -17.218761444091797, 56.66412353515625, 90.72433471679688, 0.0, 26.783363342285156, 44.30206298828125, 103.49903869628906, 5.259906768798828, 7.19580078125, -5.450897216796875, 14.147705078125, 57.25096130371094, 83.71815490722656, 56.48114013671875, 10.340606689453125, -17.202545166015625, 7.936798095703125, 30.5758056640625, 54.481658935546875, -76.45985412597656, 2.0198516845703125, 50.59742736816406, 8.552032470703125, 54.01304626464844, 61.75323486328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000389.npy"}
{"epoch": 0.8146596858638744, "step": 390, "batch_size": 128, "mean": 30.83005714416504, "std": 46.42650604248047, "min": -80.76669311523438, "p10": -29.630599975585938, "median": 30.977996826171875, "p90": 93.40864562988281, "max": 117.30905151367188, "pos_frac": 0.765625, "sample": [93.6749267578125, 2.1611175537109375, 25.910186767578125, 8.769683837890625, 20.4925537109375, 68.38961791992188, 2.9944095611572266, 85.49443054199219, 55.69976806640625, 34.86097717285156, -25.29498291015625, 10.732772827148438, 3.3524856567382812, -12.60498046875, 25.134033203125, 75.02767944335938, 8.752592086791992, 82.47314453125, 80.99424743652344, 59.697357177734375, 21.30755615234375, -23.673553466796875, 98.08949279785156, -1.1713104248046875, 44.032806396484375, 34.806060791015625, -61.900177001953125, 70.09585571289062, 79.44064331054688, 43.24122619628906, 89.56502532958984, 46.45805358886719, -51.21284484863281, 21.939552307128906, 41.050048828125, 34.957244873046875, 33.781593322753906, 15.6903076171875, 3.6114654541015625, 51.15461730957031, -52.1224365234375, 52.80519104003906, -43.877349853515625, 59.41387939453125, 31.50652313232422, -30.150238037109375, -10.472444534301758, -78.13406372070312, 73.18154907226562, -3.60406494140625, 43.554725646972656, -11.6968994140625, -65.77894592285156, 25.437957763671875, 110.90390014648438, 9.201030731201172, 23.09332275390625, 89.14073181152344, 64.568115234375, -80.76669311523438, -22.3663330078125, 64.41854858398438, 89.41361999511719, 24.312820434570312, 105.47381591796875, -65.7364501953125, 117.30905151367188, 56.635223388671875, -10.74444580078125, 74.16748046875, 44.48060607910156, 100.90879821777344, 51.2489013671875, 0.0, 2.5514068603515625, -11.726409912109375, 93.29452514648438, 4.032196044921875, 67.25288391113281, 87.06723022460938, -6.884407043457031, 55.369415283203125, 72.5218505859375, -63.64373779296875, -16.25128173828125, 52.77325439453125, 106.01422119140625, 81.01191711425781, 3.222900390625, 5.251619338989258, 19.198501586914062, 96.0791015625, 22.28924560546875, 42.3453369140625, 17.861007690429688, 53.96436309814453, -14.034431457519531, 102.314453125, 20.268020629882812, -39.734130859375, 0.9917488098144531, -35.490631103515625, 74.38021850585938, 55.67608642578125, -12.842918395996094, 33.3721923828125, 98.55172729492188, 7.66900634765625, 6.354900360107422, 5.71417236328125, -29.40789794921875, 3.2758636474609375, 14.220405578613281, 66.05938720703125, 49.11121368408203, 24.553924560546875, 95.15997314453125, 83.36278533935547, 35.128753662109375, 90.6807861328125, 71.38986206054688, 30.70361328125, -1.04345703125, 100.55770874023438, 102.1046142578125, -50.0263671875, 31.25238037109375, 8.677299499511719], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000390.npy"}
{"epoch": 0.8167539267015707, "step": 391, "batch_size": 128, "mean": 30.095077514648438, "std": 42.12250518798828, "min": -87.90482330322266, "p10": -15.16367034912109, "median": 22.7574462890625, "p90": 88.6987533569336, "max": 134.8880615234375, "pos_frac": 0.7421875, "sample": [27.116798400878906, -25.569259643554688, -0.38463783264160156, 18.622093200683594, 5.305959701538086, 6.0516357421875, -2.47998046875, 5.945096969604492, 7.260467529296875, 52.98169708251953, 5.674736022949219, 10.452423095703125, -20.03551483154297, 53.026397705078125, 100.49542236328125, 26.117828369140625, 17.13835906982422, 36.513893127441406, -45.70109558105469, 2.0561141967773438, -13.955169677734375, 68.2868423461914, 53.427459716796875, 71.68716430664062, -17.983505249023438, 64.52197265625, 83.76568603515625, 38.420257568359375, 1.8852005004882812, 12.85455322265625, -0.2888031005859375, -20.61907958984375, 50.465850830078125, -9.628982543945312, 13.1373291015625, 44.96783447265625, 54.58204650878906, 7.10595703125, -7.07672119140625, 26.6490478515625, 13.301639556884766, 73.07708740234375, 83.98182678222656, 127.03173828125, 63.661712646484375, -9.25213623046875, 96.3593978881836, 4.0111083984375, -9.475322723388672, -0.4102783203125, 110.07876586914062, 54.07379913330078, 0.8764114379882812, 43.23455810546875, -5.3079986572265625, 68.95730590820312, 40.30548095703125, 80.25270080566406, 73.965087890625, -23.05194091796875, 34.1702880859375, -36.452003479003906, 78.86424255371094, 25.591625213623047, 24.7921142578125, 6.724761962890625, 69.179931640625, 6.869709014892578, 41.87554168701172, 2.4098968505859375, -1.074676513671875, 96.91921997070312, 18.393341064453125, 45.546165466308594, 64.64546203613281, -9.82049560546875, 88.0826416015625, 40.56500244140625, -5.3398284912109375, 73.74191284179688, -4.068817138671875, 24.868637084960938, 62.11114501953125, 134.8880615234375, 102.42333984375, 31.43536376953125, -4.468353271484375, 6.8739471435546875, 89.41043853759766, 7.821014404296875, 17.9598388671875, 10.451181411743164, -87.90482330322266, 92.110107421875, 36.196685791015625, 41.17926025390625, -4.214752197265625, 64.1046142578125, 99.7869873046875, 97.544189453125, -34.31060791015625, -25.453369140625, -0.3127899169921875, -12.483154296875, 4.702117919921875, -5.461944580078125, 14.443649291992188, 68.55409240722656, 74.7896728515625, 55.783599853515625, 13.84979248046875, 45.788124084472656, 75.49939727783203, -3.3433990478515625, 22.9434814453125, 46.481842041015625, 101.747314453125, 75.80300903320312, 125.0562744140625, 1.7829742431640625, -19.9559326171875, 7.9246063232421875, 27.47393798828125, 5.602470397949219, -18.834449768066406, 22.5714111328125, -79.52313232421875, 88.39374542236328], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000391.npy"}
{"epoch": 0.818848167539267, "step": 392, "batch_size": 128, "mean": 38.7894172668457, "std": 44.42434310913086, "min": -83.61883544921875, "p10": -9.487059783935546, "median": 32.657920837402344, "p90": 102.71127777099609, "max": 134.43927001953125, "pos_frac": 0.78125, "sample": [85.67758178710938, 67.15133666992188, 57.92347717285156, 33.18133544921875, 1.505126953125, 5.070964813232422, 10.54669189453125, 67.2677001953125, 57.899169921875, 65.64724731445312, 34.62396240234375, 50.84269714355469, 50.5277099609375, -3.23992919921875, -8.996002197265625, 0.8812255859375, -83.61883544921875, 134.43927001953125, 22.54620361328125, -19.944107055664062, 96.7200927734375, -13.422962188720703, 0.94122314453125, 83.49197387695312, 79.14053344726562, 79.58847045898438, 0.0, -20.488250732421875, 53.173828125, 72.63623046875, 5.4757537841796875, -10.887542724609375, 116.48114013671875, -5.8109130859375, 30.467227935791016, 110.85687255859375, 102.95118713378906, 92.22325897216797, 82.01702880859375, 36.6409912109375, 73.97219848632812, 97.2650146484375, 6.117439270019531, 14.2025146484375, -7.182403564453125, 14.949756622314453, -9.645851135253906, -0.67242431640625, 62.81355285644531, 83.6504135131836, 62.410614013671875, 93.77513122558594, 115.84249877929688, 32.88615417480469, 103.69845581054688, 71.78536224365234, 5.478118896484375, 43.1724853515625, 97.80038452148438, -2.972198486328125, 3.6406478881835938, 51.744834899902344, -29.790313720703125, 105.4588623046875, 27.578739166259766, 24.39989471435547, -19.824417114257812, 70.19891357421875, 116.97833251953125, -0.5201416015625, 36.9493408203125, 102.54385375976562, 102.0487060546875, 41.259605407714844, 23.91900634765625, 108.10000610351562, 30.0736083984375, -18.366661071777344, -4.808479309082031, 5.013408660888672, 10.198341369628906, 85.14036560058594, 1.6048583984375, 18.46844482421875, 14.940765380859375, -16.356170654296875, -9.41900634765625, 3.193450927734375, 14.511123657226562, -16.177719116210938, 1.680419921875, 0.0, 75.98226928710938, 17.4515380859375, 95.4154052734375, 105.13333129882812, 42.4901123046875, 46.755950927734375, 118.36273193359375, 6.53125, 102.60845947265625, -0.28333091735839844, 8.01641845703125, 1.3581619262695312, 12.6092529296875, 127.31918334960938, -1.3099441528320312, 109.44061279296875, 27.239700317382812, 45.57640838623047, -50.913238525390625, 68.54342651367188, 2.2634124755859375, 32.4296875, 76.14642333984375, 21.46124267578125, 55.801849365234375, -26.829986572265625, 37.97894287109375, 74.72332763671875, 6.018699645996094, 76.86859130859375, -8.390884399414062, 77.02572631835938, -5.33306884765625, 15.529518127441406, 71.71318054199219, 33.45208740234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000392.npy"}
{"epoch": 0.8209424083769633, "step": 393, "batch_size": 128, "mean": 28.382566452026367, "std": 45.876949310302734, "min": -111.39120483398438, "p10": -21.541894531249994, "median": 20.610137939453125, "p90": 86.94472961425781, "max": 156.4151611328125, "pos_frac": 0.734375, "sample": [31.10274887084961, -6.3895111083984375, 4.052558898925781, 18.606597900390625, 43.98430633544922, -43.302978515625, 24.93968963623047, 20.61810302734375, 80.71807861328125, 8.43536376953125, 11.73193359375, -6.989776611328125, 13.76190185546875, -5.8036651611328125, 77.86357879638672, 20.6021728515625, -62.93120193481445, 37.51848220825195, 15.931854248046875, 60.97705078125, 73.88345336914062, 36.993865966796875, 1.67999267578125, 25.037445068359375, -37.0482177734375, 94.49175262451172, 84.42991638183594, 75.72522735595703, 31.711257934570312, 17.329452514648438, 72.7327880859375, 6.5816192626953125, 10.633209228515625, 89.61279296875, 43.252227783203125, 48.64143371582031, -2.3984375, 78.82421875, 20.04962158203125, 56.94950866699219, 20.5272216796875, 5.580718994140625, 61.237030029296875, 87.5828857421875, -39.336090087890625, 89.29290771484375, -13.832588195800781, -19.756195068359375, -44.71331787109375, 52.714508056640625, 156.4151611328125, 28.044239044189453, 9.9056396484375, 13.063613891601562, 16.51898956298828, 39.1688232421875, 0.5696430206298828, -18.363327026367188, 91.8836669921875, 51.671478271484375, -94.12991333007812, 18.2562255859375, 13.36627197265625, -12.077362060546875, 29.707054138183594, 85.38426208496094, 54.91307067871094, 85.94363403320312, -35.83331298828125, -9.979583740234375, -25.708526611328125, -3.768238067626953, 0.6979522705078125, 63.13908386230469, 10.38918685913086, 53.285858154296875, 110.62591552734375, -18.381698608398438, 70.98068237304688, -8.288536071777344, 6.075752258300781, -6.4832305908203125, 72.13864135742188, -85.64552307128906, 18.142995834350586, 39.191139221191406, 82.50469970703125, 62.47968673706055, 77.0670166015625, 89.38496398925781, -15.141510009765625, 93.44674682617188, 63.04522705078125, 95.27098083496094, 89.25715637207031, 27.963653564453125, 21.82342529296875, -27.48541259765625, 75.48497009277344, -111.39120483398438, 9.209774017333984, 24.608489990234375, 9.94232177734375, -3.2174224853515625, 123.74462890625, 32.06024169921875, 42.202392578125, 0.0, 9.645523071289062, 34.6439208984375, 123.998046875, -28.618927001953125, 4.062004089355469, 7.2145843505859375, -9.34600830078125, 60.40361022949219, 84.55300903320312, -7.517200469970703, -32.1031494140625, -7.166839599609375, 2.8410987854003906, -13.988143920898438, 76.90223693847656, 86.67123413085938, -1.2939300537109375, 39.9234619140625, 63.572845458984375, 81.6529769897461], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000393.npy"}
{"epoch": 0.8230366492146597, "step": 394, "batch_size": 128, "mean": 26.372745513916016, "std": 42.75120544433594, "min": -57.52116394042969, "p10": -22.44956283569336, "median": 18.065589904785156, "p90": 91.73713684082031, "max": 130.42318725585938, "pos_frac": 0.6796875, "sample": [95.91168212890625, 5.581787109375, 99.19442749023438, 50.11004638671875, 40.04144287109375, 75.80696105957031, -0.6900634765625, -2.8856239318847656, 14.737869262695312, 91.70724487304688, 55.691162109375, -0.4151153564453125, 27.1314697265625, 50.29705810546875, -6.521697998046875, -1.6966819763183594, 79.78982543945312, -20.384422302246094, 69.81512451171875, -22.340187072753906, 12.599822998046875, 83.32965087890625, 117.0272216796875, -6.5765228271484375, 0.6670875549316406, -56.01536560058594, 10.70556640625, 33.24590301513672, 34.2730712890625, 22.8543701171875, -18.733123779296875, 0.0, -27.062644958496094, -42.242706298828125, 99.83837890625, 10.052474975585938, 17.992095947265625, 57.19237518310547, 6.17926025390625, 33.380104064941406, 41.366546630859375, 61.45433044433594, 35.209930419921875, -57.52116394042969, 19.157699584960938, 59.98834228515625, 80.46436309814453, 35.27682113647461, -0.1542816162109375, 22.4898681640625, 12.259307861328125, -26.2579345703125, 41.39666748046875, -13.67596435546875, -19.1170654296875, 99.03919982910156, 11.732833862304688, 11.57354736328125, 5.1336212158203125, 41.941253662109375, 33.61053466796875, -35.66326904296875, -0.25339508056640625, -4.2257843017578125, 89.65261840820312, 4.282011032104492, 1.5983963012695312, 68.32608032226562, 10.518699645996094, 32.456939697265625, 64.52630615234375, 76.33552551269531, 97.00360107421875, 6.534244537353516, 118.26239013671875, -5.53076171875, -46.352203369140625, -19.22747802734375, -6.011085510253906, 92.55007934570312, 25.31955337524414, -57.00238037109375, 75.65847778320312, -15.12091064453125, -27.95269775390625, 18.139083862304688, 32.2857666015625, 24.702529907226562, 6.171085357666016, 30.852081298828125, -1.275543212890625, -4.082786560058594, 12.504180908203125, 10.11480712890625, -20.011932373046875, -22.70477294921875, -4.434417724609375, 33.55352783203125, 18.8233642578125, 37.37030029296875, -41.76341247558594, 80.82818603515625, 13.509796142578125, -20.651748657226562, 85.27627563476562, 130.42318725585938, -6.7598114013671875, 23.99359130859375, 55.378570556640625, 95.415771484375, 63.06416320800781, 65.32317352294922, 77.83529663085938, 91.806884765625, 98.10476684570312, -31.0523681640625, 94.5916748046875, 13.959228515625, -16.262115478515625, 54.16217041015625, 4.908203125, 59.92852783203125, -22.22760009765625, 63.76802062988281, -13.318214416503906, 0.3444061279296875, -24.3963623046875, 76.87091064453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000394.npy"}
{"epoch": 0.8251308900523561, "step": 395, "batch_size": 128, "mean": 28.87946319580078, "std": 41.59687423706055, "min": -59.37506103515625, "p10": -20.40373840332031, "median": 22.898361206054688, "p90": 94.83162689208984, "max": 128.597412109375, "pos_frac": 0.734375, "sample": [36.60882568359375, 12.246246337890625, -0.28199005126953125, -2.1883163452148438, -5.470947265625, -12.1312255859375, -1.69183349609375, 114.6273193359375, 111.3076171875, 25.233673095703125, 15.149818420410156, 42.035919189453125, 22.7537841796875, -20.783111572265625, 1.6017322540283203, 12.594589233398438, 5.2886505126953125, -14.06689453125, 11.147811889648438, -4.84027099609375, 14.065689086914062, 81.50314331054688, -13.660957336425781, -23.133941650390625, 101.3826904296875, 35.194000244140625, 9.531295776367188, -3.872325897216797, 44.25157165527344, 54.62196350097656, 69.8314437866211, 39.263671875, 54.11170959472656, 44.5719108581543, 18.463134765625, 79.37692260742188, 23.0010986328125, 30.16876220703125, 3.100006103515625, -6.784000396728516, 19.114425659179688, 10.234386444091797, -11.34552001953125, -0.551177978515625, 64.89453125, 41.09739685058594, -3.3447723388671875, -29.945911407470703, 67.4826889038086, 96.9705810546875, 110.8848876953125, -59.37506103515625, 41.543426513671875, 91.0030517578125, 99.71435546875, 38.269287109375, 54.426055908203125, 99.20891571044922, 10.18035888671875, 47.83116149902344, -21.06170654296875, -34.855316162109375, 13.165115356445312, 0.99786376953125, 76.54057312011719, -20.24114990234375, 28.69719696044922, 30.274986267089844, -7.161285400390625, 7.3553009033203125, -13.424713134765625, 24.002159118652344, -25.000946044921875, -23.81854248046875, 22.94073486328125, 63.580474853515625, 39.23933410644531, -3.4461212158203125, 26.997100830078125, 29.1092529296875, 19.2159423828125, 4.571296691894531, 128.597412109375, 90.47940063476562, 79.95947265625, -18.1767578125, -0.3530120849609375, -41.72137451171875, 5.1606597900390625, 38.68603515625, 5.3538818359375, 27.685775756835938, -46.604949951171875, 47.145591735839844, -2.69091796875, 18.295120239257812, 106.04266357421875, 83.5751953125, 68.69589233398438, 96.79168701171875, -3.5985946655273438, 48.727691650390625, 17.511199951171875, 17.815017700195312, 61.1319580078125, 28.656982421875, 89.77522277832031, 6.756927490234375, 86.88523864746094, -45.46351623535156, 29.197662353515625, 79.64962768554688, 98.97564697265625, -43.7335205078125, 2.412811279296875, -42.744293212890625, 25.55767822265625, 6.559978485107422, 91.55203247070312, 96.9619140625, 96.0511474609375, 94.30897521972656, 22.855987548828125, 5.0186004638671875, 29.738998413085938, 25.781421661376953, 6.168670654296875, 47.03814697265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000395.npy"}
{"epoch": 0.8272251308900523, "step": 396, "batch_size": 128, "mean": 30.351566314697266, "std": 43.82816696166992, "min": -89.907958984375, "p10": -28.615017700195313, "median": 25.776687622070312, "p90": 89.51140213012695, "max": 132.18621826171875, "pos_frac": 0.75, "sample": [70.70563507080078, 23.30633544921875, -6.49365234375, 87.65060424804688, 66.08377075195312, 2.5578460693359375, -6.924533843994141, 11.953994750976562, -36.254150390625, 99.95706176757812, 48.43377685546875, -28.36395263671875, 20.052154541015625, 31.292816162109375, 6.2873382568359375, 22.816131591796875, 109.76104736328125, -29.0921630859375, 7.185646057128906, 58.397705078125, -31.356521606445312, -17.246047973632812, -2.4468994140625, 132.18621826171875, -7.5251617431640625, 4.379119873046875, 52.57159423828125, 74.95181274414062, 42.62330627441406, 34.22527313232422, 37.5465087890625, -45.874298095703125, 47.347869873046875, 74.7274169921875, -57.68772888183594, 87.97013092041016, 16.700531005859375, 3.4351806640625, -6.122402191162109, 37.86262512207031, 20.70220947265625, 76.3816146850586, 53.094024658203125, 73.12849426269531, 105.77587890625, 15.8045654296875, 4.013664245605469, 10.998794555664062, -2.826171875, 49.435768127441406, 109.224609375, 98.61097717285156, 95.21112060546875, 4.978302001953125, 79.61859893798828, 30.5516357421875, 89.45613861083984, -0.321563720703125, 14.14569091796875, 3.294830322265625, 6.7920989990234375, -12.290679931640625, -27.58025360107422, -28.63922119140625, 44.3590087890625, -30.563430786132812, 52.696807861328125, -64.07048034667969, 13.417922973632812, 13.597953796386719, 75.87098693847656, -6.992584228515625, -19.16119384765625, 32.364158630371094, 79.62088012695312, 58.5792236328125, 44.8843994140625, 56.75968933105469, 21.441497802734375, -51.17950439453125, 44.77520751953125, 61.825538635253906, 37.85533142089844, 66.8082275390625, 46.552490234375, 75.94818115234375, 20.18906021118164, 88.28877258300781, -28.604644775390625, 67.71347045898438, 12.613616943359375, 90.5294189453125, -26.297943115234375, 17.61261749267578, 19.169723510742188, 28.354812622070312, 70.69146728515625, 33.57069396972656, 9.165191650390625, 34.091064453125, 89.64035034179688, 79.21748352050781, 1.4537353515625, -89.907958984375, -1.386444091796875, 92.37324523925781, -6.779182434082031, 96.11883544921875, 93.3990478515625, 101.00222778320312, 0.0, 23.7984619140625, 81.24594116210938, 24.53973388671875, 15.800048828125, 7.073982238769531, 0.0, 39.089019775390625, 53.76910400390625, 9.023529052734375, 27.013641357421875, -39.3289794921875, 42.884735107421875, 80.40090942382812, 75.25129699707031, -35.83349609375, -42.860870361328125, 68.38130187988281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000396.npy"}
{"epoch": 0.8293193717277487, "step": 397, "batch_size": 128, "mean": 28.71177101135254, "std": 45.65945816040039, "min": -72.57723999023438, "p10": -32.68124847412109, "median": 25.06122589111328, "p90": 90.68843994140624, "max": 121.61248779296875, "pos_frac": 0.703125, "sample": [-9.729446411132812, 7.064544677734375, -10.019302368164062, 7.011619567871094, 104.0987548828125, 20.527587890625, -39.02184295654297, 34.565185546875, -6.336313247680664, 24.82562255859375, 7.9026031494140625, 13.58905029296875, 25.445724487304688, 6.33221435546875, 34.30461120605469, 97.2017822265625, 24.379261016845703, 46.3692626953125, 108.74420166015625, 71.51764678955078, -38.38197326660156, 19.670166015625, -20.1689453125, 62.53033447265625, 63.322120666503906, 72.05410766601562, 0.0, -7.07550048828125, 69.72189331054688, 36.1427001953125, 0.0, -6.21258544921875, 62.898834228515625, -33.13414001464844, -0.9864501953125, 2.382539749145508, 16.65631103515625, 75.05715942382812, -50.45294189453125, 74.77203369140625, 98.29385375976562, 9.323249816894531, 83.26080322265625, 93.63998413085938, -21.873077392578125, 12.868850708007812, 66.345947265625, 86.87701416015625, 118.62567138671875, -22.91796875, 121.61248779296875, -48.780609130859375, 74.50958251953125, -66.06021118164062, -22.02947998046875, -34.8447265625, 43.88177490234375, -69.166259765625, -6.808460235595703, -72.57723999023438, -21.2264404296875, 41.35835266113281, 95.57177734375, 71.23768615722656, 24.669845581054688, 25.296829223632812, 69.34121704101562, -0.3620758056640625, 13.575103759765625, -1.0671844482421875, 41.94773864746094, 108.66949462890625, 63.393402099609375, 59.36859893798828, 76.46338653564453, 62.252288818359375, 76.3762435913086, -42.99090576171875, 39.100982666015625, 30.344039916992188, 76.65777587890625, 24.4609375, 97.14956665039062, 93.5950927734375, 31.24578857421875, 1.1975784301757812, 75.68167114257812, 1.03961181640625, 11.660736083984375, 47.421112060546875, 73.06597900390625, 10.182327270507812, 49.39735412597656, 6.946617126464844, 89.63616943359375, 79.91580963134766, 25.339187622070312, 14.399604797363281, 86.57125854492188, 57.203155517578125, 11.46917724609375, -41.75567626953125, 31.95330810546875, -3.7942581176757812, 95.07998657226562, 88.63748168945312, 61.21759033203125, -2.6307830810546875, -9.714313507080078, -14.960700988769531, -19.470993041992188, 93.14373779296875, 68.79135131835938, 9.24530029296875, 28.39227294921875, 34.23486328125, -51.6636962890625, -12.751510620117188, 73.71492004394531, 14.401458740234375, -32.487152099609375, -29.404342651367188, 28.2359619140625, 23.997543334960938, -3.7099609375, -63.90826416015625, 52.1241455078125, 48.8818359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000397.npy"}
{"epoch": 0.831413612565445, "step": 398, "batch_size": 128, "mean": 22.57021141052246, "std": 50.08265686035156, "min": -102.58111572265625, "p10": -38.9311767578125, "median": 13.747611999511719, "p90": 89.01706008911133, "max": 150.11517333984375, "pos_frac": 0.671875, "sample": [11.35858154296875, 27.882965087890625, 83.02720642089844, -4.400390625, 150.11517333984375, 4.558868408203125, 37.083892822265625, -21.58660888671875, 24.60467529296875, 85.185791015625, -46.804534912109375, -40.59185791015625, -81.661376953125, 9.629486083984375, 1.8582839965820312, 5.6734619140625, 22.32568359375, -3.9995040893554688, 86.1290283203125, 94.09695434570312, 89.88755798339844, -68.90402221679688, -48.083984375, 69.92120361328125, 30.55767822265625, 58.9432373046875, 8.76214599609375, 61.906219482421875, 79.94400024414062, 5.1675262451171875, -39.341888427734375, 9.717132568359375, 27.682464599609375, -102.58111572265625, -40.273101806640625, 35.87261962890625, 22.24029541015625, 77.41184997558594, 12.93475341796875, -34.15575408935547, 1.2364768981933594, -30.19921875, 23.485374450683594, 67.5574951171875, -0.3231029510498047, -38.755157470703125, -3.598968505859375, 83.87967681884766, 14.88702392578125, -21.880996704101562, 39.32366943359375, -17.7274169921875, 36.086761474609375, -7.5933380126953125, 62.41316223144531, 31.2576904296875, 6.497455596923828, -61.122650146484375, 33.49688720703125, -8.635177612304688, 19.83038330078125, 27.687255859375, 91.29946899414062, 2.605194091796875, 64.59234619140625, 31.31903076171875, 8.686500549316406, -60.18170166015625, 81.48825073242188, 88.64398956298828, 38.23779296875, 116.21148681640625, -0.707122802734375, 117.87588500976562, -29.513412475585938, -15.43792724609375, 7.639614105224609, 101.4930419921875, -15.0194091796875, -6.450653076171875, 94.77896118164062, 103.23870849609375, 3.607654571533203, 38.24177551269531, 14.560470581054688, 1.6893463134765625, -36.54789733886719, 75.11004638671875, 25.04364013671875, -29.368133544921875, 58.802520751953125, 104.3028564453125, -8.297332763671875, -71.02926635742188, 9.51806640625, -3.28704833984375, 75.53472900390625, -16.382659912109375, -4.86474609375, 59.12248229980469, -13.307891845703125, 1.00616455078125, 78.5567626953125, 45.14326477050781, 7.54046630859375, 0.9908676147460938, 64.57791137695312, 87.32951354980469, 92.86135864257812, -15.12799072265625, -80.71135711669922, 81.27359008789062, 84.02325439453125, -33.991249084472656, 76.20965576171875, -8.474868774414062, -0.41143798828125, -91.11923217773438, 11.001914978027344, 108.00814819335938, 39.47589111328125, -5.013916015625, 7.3880462646484375, 119.05422973632812, 29.765533447265625, 20.521484375, 69.8145751953125, 36.14801025390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000398.npy"}
{"epoch": 0.8335078534031414, "step": 399, "batch_size": 128, "mean": 26.47659683227539, "std": 43.28916549682617, "min": -95.74301147460938, "p10": -28.218692016601562, "median": 23.930654525756836, "p90": 87.11602630615232, "max": 120.31829833984375, "pos_frac": 0.734375, "sample": [-10.855758666992188, 76.77256774902344, 79.171142578125, 68.87367248535156, 10.682952880859375, -13.709285736083984, 14.037307739257812, 25.16436767578125, 2.038787841796875, 97.57208251953125, 50.23057556152344, -33.70314025878906, -47.12939453125, 80.96104431152344, -3.9179534912109375, 31.850128173828125, 50.099761962890625, 20.650039672851562, 95.84971618652344, 26.074295043945312, 6.667205810546875, -47.570526123046875, -21.183929443359375, 83.59716796875, 14.229461669921875, -2.9456787109375, 11.438690185546875, -59.516212463378906, -77.3316650390625, 22.926040649414062, 1.003875732421875, 65.58282470703125, 11.58334732055664, 98.9814453125, 12.151870727539062, -8.44989013671875, 75.26687622070312, -14.327445983886719, 5.455352783203125, 100.60382080078125, 22.115478515625, 58.04627990722656, 120.31829833984375, 85.35462951660156, 35.7220458984375, -5.803764343261719, 73.42666625976562, -41.50115966796875, 9.80204963684082, 76.46522521972656, -28.736236572265625, 64.04580688476562, 30.945648193359375, 27.504241943359375, 108.69967651367188, -7.1102294921875, 5.399974822998047, 11.09521484375, -3.0281829833984375, -13.863285064697266, 33.23827362060547, 91.2259521484375, 42.6995849609375, 10.164710998535156, 48.52678680419922, 11.180919647216797, 74.46736145019531, 23.662227630615234, 28.14655303955078, 34.07086181640625, 55.94419860839844, 37.995391845703125, -10.246612548828125, -54.954498291015625, -95.74301147460938, -48.65373992919922, -9.623779296875, 24.199081420898438, 100.28021240234375, 92.57443237304688, 31.96874237060547, 78.26727294921875, 41.7674560546875, 57.385345458984375, 64.16336059570312, 42.108428955078125, 14.24932861328125, -14.825714111328125, 43.910888671875, 43.868072509765625, 13.07937240600586, 1.895599365234375, 8.674068450927734, 43.1978759765625, 18.536453247070312, 31.39850616455078, -46.52983093261719, -1.4368553161621094, 96.56002807617188, 6.707366943359375, 56.56817626953125, 8.1427001953125, 31.68408203125, -35.016265869140625, 45.5238037109375, 95.10714721679688, -11.84210205078125, 96.83883666992188, 63.805084228515625, 14.296340942382812, 96.9462890625, -54.086090087890625, -24.45098876953125, 25.6951904296875, 19.2330265045166, 22.47021484375, 47.99913024902344, 72.54425048828125, -8.823257446289062, 69.82244873046875, -27.99688720703125, 34.845306396484375, 60.54780578613281, 53.423255920410156, -15.118316650390625, -2.905975341796875, 44.64210510253906, 17.23443603515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000399.npy"}
{"epoch": 0.8356020942408376, "step": 400, "batch_size": 128, "mean": 30.028249740600586, "std": 43.15439224243164, "min": -85.27192687988281, "p10": -22.605632019042968, "median": 23.884902954101562, "p90": 87.10122985839844, "max": 112.4683837890625, "pos_frac": 0.7734375, "sample": [11.24652099609375, -8.230239868164062, 6.0653839111328125, 27.14202880859375, 34.27294921875, 9.133766174316406, 51.08533477783203, 58.0372314453125, 11.6064453125, -45.60389709472656, 8.546497344970703, 60.97119140625, -85.27192687988281, 16.27715301513672, 94.7037353515625, 12.399429321289062, 4.739383697509766, 74.47969055175781, 33.799896240234375, 46.48162841796875, 112.4683837890625, 6.938262939453125, 30.292526245117188, 42.40277099609375, 91.11849975585938, 54.405059814453125, -38.843109130859375, 40.47412109375, -11.273193359375, 51.90118408203125, 29.80767822265625, 4.534400939941406, 9.917488098144531, 64.41093444824219, 30.39984130859375, 84.14837646484375, 92.18450927734375, 86.61309814453125, 18.522811889648438, 22.73736572265625, -70.28038024902344, 71.17792510986328, 68.15104675292969, 87.64483642578125, 18.957733154296875, 67.82000732421875, 15.490997314453125, 28.58221435546875, 16.806976318359375, 9.18292236328125, 17.1822509765625, 73.66696166992188, 94.1491470336914, -54.08135986328125, 2.6522216796875, -21.607833862304688, 9.731304168701172, 14.416458129882812, 98.781005859375, 1.5452690124511719, -1.2540016174316406, 63.745819091796875, 61.904510498046875, 69.56719970703125, 11.447113037109375, -13.1279296875, 13.606109619140625, 79.46309661865234, 69.5145263671875, 101.44526672363281, 0.131072998046875, 9.166603088378906, -13.62762451171875, 7.201751708984375, -23.26763916015625, -35.946258544921875, 93.6505126953125, 98.41999816894531, 11.119644165039062, -30.211456298828125, 39.758567810058594, 57.666778564453125, 71.34722900390625, 40.80157470703125, -3.1027908325195312, 97.37857055664062, 72.20541381835938, -27.1649169921875, 42.636962890625, 28.103370666503906, 36.31452941894531, -14.015380859375, -17.788040161132812, 1.0023841857910156, -0.108978271484375, 76.107177734375, -7.1550750732421875, 64.54962158203125, 2.663397789001465, -26.80224609375, 29.8369140625, 86.86825561523438, 76.05852508544922, 25.032440185546875, -22.321914672851562, 13.2012939453125, 69.78141784667969, -0.6839141845703125, 4.8422088623046875, 68.48124694824219, 79.88670349121094, 5.542030334472656, 79.629150390625, -5.98980712890625, 64.41497802734375, 59.78059387207031, 80.95875549316406, -21.775726318359375, 111.41253662109375, 86.81753540039062, 94.75999450683594, 78.10720825195312, 0.0, -52.14898681640625, -26.86297607421875, 4.466064453125, -67.45623779296875, 18.618324279785156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000400.npy"}
{"epoch": 0.837696335078534, "step": 401, "batch_size": 128, "mean": 35.891639709472656, "std": 43.17746353149414, "min": -86.08567810058594, "p10": -12.033549118041986, "median": 37.57749938964844, "p90": 90.05572052001953, "max": 139.35986328125, "pos_frac": 0.75, "sample": [-10.48602294921875, -9.817535400390625, 32.7767333984375, 2.699615478515625, 16.9581298828125, 60.41424560546875, 51.1553955078125, 56.677215576171875, 44.844512939453125, -3.69720458984375, 71.35812377929688, 16.445831298828125, 30.012969970703125, 89.75277709960938, 37.895111083984375, 29.039810180664062, 7.312744140625, 61.15592956542969, -5.3924560546875, 69.57331848144531, -4.515556335449219, 72.698974609375, -15.581745147705078, 9.47650146484375, -8.53289794921875, 86.00067138671875, 50.797760009765625, 2.2057037353515625, 62.0572509765625, 22.23486328125, 101.64517211914062, 127.33953857421875, 78.77845001220703, 46.50274658203125, 9.51751708984375, 10.45220947265625, 39.35200500488281, 73.60026550292969, 12.749076843261719, 77.11553192138672, 139.35986328125, -7.39898681640625, 89.3179931640625, 94.02215576171875, 114.45144653320312, 50.14215087890625, 26.663625717163086, -1.8592529296875, 79.48974609375, 90.76258850097656, -23.322235107421875, 77.71414184570312, 43.56134033203125, 13.754753112792969, -77.6005859375, 78.75335693359375, 11.67474365234375, 66.17584228515625, 45.11309814453125, -10.512893676757812, -21.036346435546875, 64.88687133789062, 77.17898559570312, 75.17465209960938, 44.85614013671875, 46.757904052734375, -50.62762451171875, -27.464065551757812, 77.92080688476562, -3.879913330078125, -2.687206268310547, 3.5138015747070312, 107.9512939453125, 22.76129150390625, 27.854080200195312, 2.2413711547851562, 79.68382263183594, 65.251220703125, 37.84521484375, 44.97429656982422, 48.0369873046875, -20.15423583984375, 83.10359954833984, 98.14019775390625, 53.43511962890625, 69.2427978515625, 40.229705810546875, -4.049591064453125, 48.02960205078125, -18.305328369140625, 37.309783935546875, 5.5206451416015625, -1.5252685546875, -17.36175537109375, 39.26650619506836, -9.06463623046875, -17.854949951171875, 105.27008056640625, 113.28376770019531, 24.64874267578125, 13.88653564453125, -0.12176513671875, 96.37008666992188, 79.59754943847656, 97.12491607666016, 13.52227783203125, -3.9100341796875, 11.59661865234375, -15.934539794921875, 59.850982666015625, 5.64923095703125, -1.40130615234375, 21.292160034179688, 4.281642913818359, -7.817516326904297, 0.48368072509765625, 20.73699951171875, 75.89631652832031, 76.54124450683594, -8.405525207519531, 71.80964660644531, 71.91612243652344, 64.52056884765625, 0.0943450927734375, 117.13661193847656, -86.08567810058594, -24.896835327148438, 87.201171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000401.npy"}
{"epoch": 0.8397905759162304, "step": 402, "batch_size": 128, "mean": 31.10666275024414, "std": 45.8369026184082, "min": -88.27908325195312, "p10": -19.537356948852537, "median": 27.547286987304688, "p90": 88.06527709960936, "max": 136.6561279296875, "pos_frac": 0.7578125, "sample": [-7.784477233886719, 3.2638626098632812, -53.086395263671875, 31.148345947265625, 86.02880859375, -6.4585723876953125, -53.95166015625, -4.2288818359375, 38.18107604980469, -76.9552001953125, -70.2894287109375, 73.85699462890625, 15.539947509765625, -19.25551986694336, 4.8037567138671875, 127.86483764648438, 62.80548095703125, 65.17212677001953, 34.54966735839844, -16.23870849609375, 64.84742736816406, 1.5574188232421875, 35.21678924560547, 99.12298583984375, -68.53425598144531, 2.7737674713134766, -6.421173095703125, 83.72339630126953, -88.27908325195312, 26.999237060546875, 57.462890625, 17.650787353515625, 19.744964599609375, 72.0465087890625, 11.83233642578125, 0.0, 131.51141357421875, 72.53680419921875, -21.166168212890625, 76.65497589111328, 20.77743148803711, 23.517120361328125, 35.33087158203125, 7.005706787109375, 44.43634033203125, 84.91387176513672, -20.194976806640625, -5.364532470703125, 110.6201171875, 64.20429992675781, 18.6495361328125, -20.66778564453125, -16.52728271484375, 44.655975341796875, 39.636077880859375, 11.3941650390625, 80.73268127441406, 29.530593872070312, 1.6411590576171875, 36.52864074707031, 9.221115112304688, 114.965576171875, 90.46148681640625, 13.036422729492188, 62.62964630126953, -12.646133422851562, -27.002975463867188, 52.39469909667969, 34.09272766113281, -72.91038513183594, 70.47805786132812, 72.69051361083984, 20.500869750976562, 0.98651123046875, 73.36714172363281, -2.0471343994140625, 136.6561279296875, 14.755603790283203, 51.108154296875, 96.49929809570312, 85.82015991210938, 87.038330078125, 18.51580810546875, 61.60028839111328, 10.328384399414062, 27.56695556640625, 98.880615234375, 58.58349609375, -17.708343505859375, 78.54345703125, 33.67900085449219, 22.571792602539062, 12.029640197753906, -0.299468994140625, 34.9366455078125, 67.0052490234375, 6.141761779785156, 62.388763427734375, -6.84088134765625, 93.21115112304688, 92.58309936523438, 83.662841796875, 49.146240234375, -38.686004638671875, 119.0618896484375, -18.804885864257812, 34.36309814453125, 12.47119140625, 23.32147216796875, 81.94059753417969, 46.22142791748047, 75.9029541015625, 59.61383056640625, 61.353515625, -25.69195556640625, 5.55511474609375, -5.0879669189453125, 51.814910888671875, 30.748336791992188, 6.408966064453125, 100.62554931640625, 0.7351398468017578, 2.1664981842041016, -0.451385498046875, -0.06903076171875, 27.527618408203125, 56.39863204956055, 0.5241451263427734], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000402.npy"}
{"epoch": 0.8418848167539267, "step": 403, "batch_size": 128, "mean": 40.91115951538086, "std": 42.12593460083008, "min": -60.98052978515625, "p10": -6.608390808105466, "median": 40.478084564208984, "p90": 95.83323516845704, "max": 135.39141845703125, "pos_frac": 0.828125, "sample": [-10.831405639648438, 67.25961303710938, 0.0589599609375, 80.43257141113281, 73.0604248046875, 48.96405029296875, 95.78817749023438, 41.894744873046875, 27.209228515625, 16.431549072265625, 95.2476806640625, 73.90838623046875, 35.452911376953125, 19.9193115234375, 6.8820343017578125, 50.470672607421875, 20.085227966308594, 86.70578002929688, 27.330169677734375, 2.8149871826171875, 95.93836975097656, -12.116931915283203, -44.670654296875, 34.95294189453125, 79.92373657226562, 52.583221435546875, 83.38748168945312, 2.0794677734375, 53.8541259765625, 99.16549682617188, 122.1085205078125, 44.464324951171875, 0.355560302734375, 56.24577331542969, -4.78021240234375, 50.32048797607422, 10.816192626953125, -0.20758056640625, 0.0, -10.28436279296875, -0.7742443084716797, 89.30046081542969, 117.48089599609375, 48.0180778503418, 84.38687133789062, 88.6632080078125, 88.8951416015625, 14.58642578125, 65.30833435058594, 129.07843017578125, 6.67889404296875, 19.947654724121094, 12.82415771484375, -3.0682029724121094, 102.94192504882812, -3.833343505859375, 32.025299072265625, -36.37005615234375, 52.4525146484375, 3.9426116943359375, -8.598663330078125, 32.476226806640625, -1.6683158874511719, 103.03466796875, 82.2462158203125, 9.481101989746094, 126.35841369628906, 0.8601837158203125, 40.52770233154297, 21.9322509765625, 21.48321533203125, 26.6915283203125, 14.287979125976562, 65.75360107421875, -11.027191162109375, 46.30583190917969, 60.3463134765625, 98.5250244140625, 6.012199401855469, 42.163909912109375, 1.5147590637207031, 46.5301513671875, 105.97953796386719, 14.59893798828125, 82.36561584472656, -16.420440673828125, 4.2341766357421875, 78.67636108398438, 66.36920166015625, 50.39375305175781, 135.39141845703125, 40.428466796875, 135.05340576171875, 81.9176025390625, 18.473060607910156, 86.92472839355469, 5.524566650390625, 64.86880493164062, 78.21798706054688, 45.44752502441406, 60.549381256103516, 34.668968200683594, 81.65774536132812, 31.976654052734375, -60.98052978515625, 99.66703796386719, -18.1072998046875, 68.70623779296875, -1.2784004211425781, 76.69087219238281, -43.24658203125, 22.52539825439453, 34.620819091796875, 27.7335205078125, -27.04638671875, 47.65449523925781, 49.396026611328125, -5.7554168701171875, 52.752891540527344, 1.8757247924804688, -58.8543701171875, 81.94873809814453, 76.30322265625, 34.372711181640625, 39.77919006347656, 50.39369201660156, 60.978546142578125, 28.25775146484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000403.npy"}
{"epoch": 0.8439790575916231, "step": 404, "batch_size": 128, "mean": 26.749439239501953, "std": 47.313106536865234, "min": -114.72119140625, "p10": -31.658929443359373, "median": 21.07799530029297, "p90": 92.84896850585938, "max": 139.97161865234375, "pos_frac": 0.7265625, "sample": [94.12867736816406, 99.73248291015625, 25.60162353515625, 44.99870300292969, -58.11134338378906, 74.18229675292969, 95.70108032226562, 33.45088195800781, -58.29943084716797, 0.0, -84.60205078125, 37.105995178222656, 19.73682403564453, -5.66668701171875, -74.30722045898438, 49.51231384277344, 17.4542236328125, 93.346923828125, -0.8268661499023438, -36.8812255859375, 91.04483032226562, 14.712234497070312, 18.193389892578125, -14.7279052734375, -16.5321044921875, 65.3843994140625, -114.72119140625, 84.20523834228516, -21.056076049804688, 22.288291931152344, 5.112548828125, -33.362701416015625, 11.6781005859375, 7.792938232421875, 45.76577377319336, -5.416923522949219, 2.715595245361328, 11.4244384765625, 139.97161865234375, 41.81683349609375, 92.81106567382812, 0.0700836181640625, 27.93902587890625, -38.410430908203125, 71.73951721191406, 33.46874237060547, 46.151123046875, 24.359786987304688, 4.68476676940918, 16.269989013671875, 44.73005676269531, 6.021453857421875, 79.26492309570312, -33.4595947265625, -9.37554931640625, 110.88046264648438, -70.90087890625, 14.763809204101562, -6.316497802734375, 18.229965209960938, 122.05401611328125, 111.04377746582031, -30.928741455078125, 53.3856201171875, 23.7489013671875, 84.98408508300781, 81.35296630859375, 19.113433837890625, 1.26715087890625, 83.8165283203125, 52.72534942626953, 12.010129928588867, 6.632415771484375, -0.6417369842529297, -1.006683349609375, 59.36993408203125, 42.313446044921875, -3.0665283203125, 101.60098266601562, 27.362411499023438, 19.79571533203125, 16.429641723632812, 10.795013427734375, 52.9169921875, 12.056396484375, -24.942047119140625, -14.401641845703125, -5.586639404296875, 70.14776611328125, -80.41656494140625, -4.439720153808594, 30.069976806640625, 44.4095458984375, 45.6214599609375, -64.35511779785156, -20.297149658203125, 62.77044677734375, 7.7648773193359375, 94.4644546508789, 52.25921630859375, -4.397195816040039, 6.2626953125, 60.2181396484375, 43.097686767578125, 85.25392150878906, 26.134414672851562, 21.376380920410156, 47.61918640136719, -50.151771545410156, 93.39810180664062, 19.953338623046875, 92.93740844726562, 56.760284423828125, 36.79060363769531, 1.37664794921875, -0.573577880859375, 20.77960968017578, -7.076023101806641, 68.00619506835938, 75.01937866210938, 22.637908935546875, -7.636772155761719, 6.675666809082031, 73.6888427734375, 74.01123046875, 59.81089782714844, 84.54891967773438, 109.7374267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000404.npy"}
{"epoch": 0.8460732984293193, "step": 405, "batch_size": 128, "mean": 32.93724822998047, "std": 42.80260467529297, "min": -73.782470703125, "p10": -16.88765411376953, "median": 27.147140502929688, "p90": 91.43786010742187, "max": 121.84637451171875, "pos_frac": 0.78125, "sample": [10.70428466796875, 12.54541015625, 3.181060791015625, 0.0, -25.1917724609375, 3.800811767578125, 5.3719482421875, 87.28265380859375, 65.63394165039062, 54.3408203125, 0.9783172607421875, 13.434417724609375, 63.99409484863281, 20.190811157226562, 79.08663177490234, -19.11229705810547, -73.782470703125, 20.140731811523438, 30.0479736328125, 113.484375, 31.903667449951172, 91.04833984375, 4.609996795654297, -16.465980529785156, 76.99012756347656, 41.12725830078125, -14.22833251953125, 39.2401123046875, 95.83036804199219, 32.93567657470703, 62.88350296020508, 82.16553497314453, 4.80645751953125, -18.140289306640625, 63.688682556152344, 77.91332244873047, 12.998519897460938, -15.442718505859375, -25.91695213317871, 66.17054748535156, -2.049898147583008, 108.17610168457031, 57.79729461669922, 0.9968185424804688, 0.3870849609375, 117.40289306640625, 71.37326049804688, 39.20964813232422, 65.26870727539062, 6.7919921875, -2.571685791015625, 58.178863525390625, 94.1712646484375, 58.107208251953125, 8.942947387695312, 1.79071044921875, 72.63644409179688, 73.6063232421875, 63.72578430175781, -12.188568115234375, -56.54503631591797, 48.95707702636719, 48.76361083984375, -15.96112060546875, 14.814178466796875, -4.589698791503906, 38.27983093261719, -3.784576416015625, 18.1119384765625, 121.84637451171875, 15.668045043945312, 65.21173095703125, 80.99551391601562, 51.5250244140625, 92.34674072265625, 113.97882080078125, -21.819671630859375, 55.96192932128906, 13.146720886230469, 23.843231201171875, 99.84429931640625, 6.269001007080078, -25.334228515625, -70.36917114257812, 55.299346923828125, 23.211566925048828, 1.8900299072265625, 119.22366333007812, 17.648757934570312, 59.19264221191406, 72.77699279785156, 32.36864471435547, -21.839584350585938, 7.614677429199219, 84.74142456054688, 76.97149658203125, 10.085807800292969, 12.612045288085938, 75.41973876953125, 48.50733184814453, -44.28266906738281, 71.67080688476562, 3.420379638671875, 85.13677978515625, 41.629425048828125, 27.336578369140625, 7.542407989501953, 49.849700927734375, 43.814971923828125, 35.36517333984375, 26.95770263671875, 0.5450859069824219, 105.016357421875, -4.754180908203125, 1.8568878173828125, -12.117172241210938, -5.53460693359375, 7.560871124267578, -22.471267700195312, 1.8972530364990234, -8.234588623046875, 105.8863525390625, 88.4443359375, -6.84490966796875, -17.871559143066406, 98.77215576171875, 44.93096923828125, 61.556396484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000405.npy"}
{"epoch": 0.8481675392670157, "step": 406, "batch_size": 128, "mean": 25.51388168334961, "std": 45.17200469970703, "min": -85.3985595703125, "p10": -28.354767608642575, "median": 17.78527069091797, "p90": 86.85011749267578, "max": 120.65957641601562, "pos_frac": 0.703125, "sample": [79.97329711914062, -7.004432678222656, -26.365768432617188, 19.529296875, -48.58612060546875, 114.38287353515625, 66.82374572753906, 88.13180541992188, 52.636802673339844, 75.10423278808594, -2.8523483276367188, 0.5652790069580078, -22.20269775390625, 87.5877685546875, 42.4129638671875, -7.9795074462890625, 83.04763793945312, -14.843971252441406, 12.477203369140625, 16.8018798828125, -59.199798583984375, -26.32171630859375, 5.886749267578125, 45.32861328125, 70.91561889648438, 5.58770751953125, 29.470741271972656, 90.2080078125, 13.001922607421875, 49.46990966796875, 36.814117431640625, 84.27078247070312, -83.47039794921875, -6.48480224609375, 46.58953857421875, -55.07537841796875, -31.31341552734375, -7.628440856933594, -5.881706237792969, 79.89093017578125, -24.333465576171875, -32.2984619140625, -38.8629150390625, -11.320648193359375, 11.956092834472656, 71.53265380859375, 13.984375, 70.89398193359375, 62.256378173828125, 67.54821014404297, -6.91351318359375, 15.424713134765625, 63.92658996582031, 40.409034729003906, -85.3985595703125, -10.499359130859375, 54.71539306640625, 93.80593872070312, 49.80573272705078, 18.11663818359375, 3.85711669921875, 90.70394897460938, -27.495140075683594, 97.554931640625, 12.84698486328125, 26.10455322265625, 9.587860107421875, 3.651580810546875, -0.247528076171875, 81.90155029296875, 110.19009399414062, 71.2293701171875, 9.000930786132812, 29.2589111328125, 93.39066314697266, 38.264617919921875, 5.848167419433594, 73.2625732421875, 12.212486267089844, 93.76873779296875, 80.30868530273438, -7.753026962280273, -79.26024627685547, 9.121959686279297, 86.53398132324219, 73.96400451660156, 54.943206787109375, -5.306831359863281, 20.724822998046875, 25.050201416015625, 6.2928924560546875, -9.340465545654297, 54.71290588378906, 67.778564453125, -40.019691467285156, 120.25039672851562, 56.04182434082031, -1.2143096923828125, 70.83512115478516, 20.7122802734375, 29.455520629882812, 56.86564636230469, 29.11529541015625, 78.39421081542969, -17.510284423828125, 17.5830078125, -36.115814208984375, 107.13552856445312, -14.423294067382812, 5.476165771484375, 9.663330078125, 44.25506591796875, -9.877899169921875, 41.09173583984375, 120.65957641601562, 3.0177841186523438, 1.9399337768554688, 21.755172729492188, -30.360565185546875, 0.0138092041015625, -25.858245849609375, 14.399932861328125, 7.531005859375, -10.372406005859375, 55.871856689453125, -33.80731201171875, 26.173965454101562, 17.987533569335938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000406.npy"}
{"epoch": 0.8502617801047121, "step": 407, "batch_size": 128, "mean": 31.009235382080078, "std": 44.87074661254883, "min": -68.12335205078125, "p10": -21.18206329345703, "median": 24.997276306152344, "p90": 91.65643615722657, "max": 128.57044982910156, "pos_frac": 0.703125, "sample": [-22.05865478515625, 20.142578125, 35.3306884765625, -9.270706176757812, 83.5084228515625, 12.823741912841797, 83.14492797851562, 67.93333435058594, -23.2396240234375, 14.141702651977539, -1.9542579650878906, -7.5360107421875, -60.914581298828125, -5.644927978515625, -60.356414794921875, 63.601829528808594, 3.452545166015625, 0.4877128601074219, 16.28912353515625, -20.56365966796875, 53.606361389160156, -3.959625244140625, 10.165542602539062, 28.21611785888672, 89.237060546875, -2.5649261474609375, 35.48188018798828, 53.526092529296875, -15.539970397949219, -10.309539794921875, -33.244422912597656, 45.87797546386719, -30.83892822265625, 28.733856201171875, 61.45915222167969, 84.4259033203125, 74.794189453125, 91.89328002929688, -18.71014404296875, -3.403045654296875, 49.566497802734375, 22.488418579101562, 53.02630615234375, 3.4278030395507812, -28.501953125, 107.22061157226562, 12.241966247558594, 66.10956573486328, 119.003662109375, 16.36138916015625, 26.009735107421875, 69.10494995117188, 8.652679443359375, 114.477783203125, 27.455230712890625, 83.51950073242188, -13.346305847167969, -5.9284210205078125, 28.829605102539062, -11.88714599609375, 69.2198257446289, 23.984817504882812, 3.243682861328125, 22.354007720947266, 92.2591552734375, 33.51667785644531, -16.432342529296875, -8.918563842773438, 46.99980163574219, 93.17959594726562, 15.790924072265625, -0.5565643310546875, 75.10458374023438, 97.71408081054688, 91.554931640625, 23.287513732910156, -2.7402191162109375, 22.467727661132812, 0.0, 128.57044982910156, 89.52224731445312, 0.175323486328125, 41.159423828125, -25.9852294921875, 64.87533569335938, -62.4053955078125, 105.92233276367188, 37.05061721801758, 10.025604248046875, 73.0255126953125, 85.18157958984375, 0.2335205078125, 6.9473876953125, 63.65927505493164, 57.356201171875, 8.684600830078125, -26.843475341796875, 31.868980407714844, 33.733978271484375, 66.956787109375, 83.40005493164062, -10.240951538085938, 117.38949584960938, -68.12335205078125, 43.661407470703125, -27.7236328125, 67.52828979492188, 40.70610046386719, 93.45475769042969, -1.15203857421875, 85.8215103149414, 90.24496459960938, 41.52239990234375, -52.138763427734375, 104.38150024414062, -20.806381225585938, 43.01026916503906, -12.317249298095703, 6.581809997558594, 76.1488037109375, 77.53459167480469, 74.66204833984375, -8.0152587890625, 5.234926223754883, 0.906341552734375, 104.6842269897461, 65.08502197265625, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000407.npy"}
{"epoch": 0.8523560209424084, "step": 408, "batch_size": 128, "mean": 30.344974517822266, "std": 43.74425506591797, "min": -100.46588134765625, "p10": -22.65495147705078, "median": 27.608840942382812, "p90": 88.76779785156249, "max": 114.44891357421875, "pos_frac": 0.75, "sample": [-3.3629074096679688, 11.72161865234375, -34.76036834716797, 44.1085205078125, 7.8568115234375, 88.29241943359375, 56.87909698486328, 104.185302734375, 75.35916137695312, 85.4770278930664, -0.9670562744140625, 8.088447570800781, 52.09742736816406, 66.21160888671875, -33.778289794921875, 78.11591339111328, 33.89886474609375, 0.438446044921875, 1.7632598876953125, 26.056013107299805, 4.6971588134765625, 15.946868896484375, -12.157882690429688, -5.27325439453125, 107.628173828125, 77.05298614501953, 108.45916748046875, 92.240478515625, -25.666168212890625, -2.9132080078125, -17.194366455078125, 89.87701416015625, 67.23439025878906, 37.93769836425781, -18.8016357421875, 28.319438934326172, 114.44891357421875, 65.68777465820312, 90.23320770263672, 34.561126708984375, 0.0194244384765625, -69.80810546875, -100.46588134765625, 13.903022766113281, 68.98458862304688, 73.05947875976562, 4.915519714355469, 16.03960418701172, -30.4515380859375, 85.82574462890625, 13.650421142578125, 43.20660400390625, -8.378250122070312, -9.325485229492188, 25.832595825195312, 45.06083679199219, 43.07794189453125, 77.520263671875, -19.4598388671875, 17.438140869140625, 93.78077697753906, 75.94996643066406, 98.35247802734375, 7.309547424316406, -16.73876953125, 87.97381591796875, -34.327972412109375, 13.721023559570312, 70.8458251953125, -38.30461120605469, 104.23834228515625, -27.582565307617188, 76.26824951171875, 14.10546875, 47.079410552978516, 93.99310302734375, -11.16278076171875, 27.380020141601562, 0.083038330078125, 31.730239868164062, 73.98089599609375, 47.125511169433594, -12.447929382324219, 28.189849853515625, -23.8150634765625, 6.760498046875, 64.86666870117188, 49.950958251953125, 30.052658081054688, -44.84071350097656, 4.985565185546875, 4.327415466308594, 42.295989990234375, 74.6091079711914, 78.91378784179688, -0.4662933349609375, -65.12603759765625, 46.62139892578125, 97.17654418945312, 64.06915283203125, -22.157760620117188, 81.69761657714844, 13.07794189453125, 25.064895629882812, 67.67146301269531, -6.3541259765625, 10.598396301269531, 13.20050048828125, 16.7874755859375, 42.65496826171875, 31.193138122558594, 7.9249267578125, 24.723220825195312, 0.0, 3.48468017578125, 85.04901123046875, 27.837661743164062, -0.7476520538330078, 49.824981689453125, -18.514984130859375, -55.036354064941406, 25.17840576171875, 93.95417022705078, 35.205596923828125, 61.8817138671875, 68.41204833984375, 31.620525360107422, 77.35546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000408.npy"}
{"epoch": 0.8544502617801047, "step": 409, "batch_size": 128, "mean": 32.772281646728516, "std": 48.40333938598633, "min": -92.60586547851562, "p10": -27.678026580810542, "median": 25.06082534790039, "p90": 98.55469207763672, "max": 153.540771484375, "pos_frac": 0.7578125, "sample": [62.78167724609375, 59.18610382080078, 70.33808898925781, 5.070159912109375, 105.5374755859375, -53.032958984375, 63.423583984375, 103.84539794921875, 6.9046630859375, 19.7933349609375, 27.80023956298828, 5.220405578613281, 12.739700317382812, 2.32879638671875, 12.762969970703125, 115.45819091796875, 43.191993713378906, 17.395660400390625, 80.91986846923828, -59.043548583984375, -5.436859130859375, -26.59955596923828, -1.3748741149902344, 98.69441223144531, 74.32565307617188, -23.2266845703125, 21.775909423828125, -44.481964111328125, 80.08732604980469, -92.60586547851562, 100.12593078613281, 74.3203125, -47.072601318359375, 29.303817749023438, -2.88800048828125, 116.7352294921875, 14.409515380859375, 29.2530517578125, 107.6531982421875, 98.49481201171875, 60.8480224609375, 94.65963745117188, -54.10899353027344, 17.799087524414062, 2.4741363525390625, 40.73504638671875, 2.62896728515625, 89.17703247070312, 41.878021240234375, 80.43780517578125, 53.1922607421875, 72.06619262695312, 41.6693115234375, 61.317840576171875, 153.540771484375, -30.1944580078125, -35.563270568847656, 52.359161376953125, 13.728485107421875, 3.2168617248535156, 115.27438354492188, 59.6424560546875, 10.134780883789062, 18.9853515625, 79.50640869140625, 24.288619995117188, 21.932708740234375, -19.396881103515625, -2.705902099609375, -26.307723999023438, 88.43339538574219, -12.052474975585938, -54.06488037109375, 107.37687683105469, -0.3214378356933594, 96.64352416992188, 103.91094970703125, 73.12762451171875, 14.21014404296875, 23.060394287109375, 9.138248443603516, -8.411914825439453, 47.9342041015625, 0.0, 25.833030700683594, 129.5548095703125, 43.71241760253906, 101.39337158203125, 71.66348266601562, -32.5938720703125, -16.360328674316406, 90.62464141845703, 48.798583984375, 8.932464599609375, 3.1640777587890625, -6.754568099975586, 73.97218322753906, 90.70144653320312, 77.10357666015625, 81.85353088378906, 14.337203979492188, 26.338722229003906, 51.6976318359375, 56.43743896484375, 32.327362060546875, -3.9232254028320312, 16.875320434570312, -32.224853515625, -13.619186401367188, -8.709648132324219, 17.80198097229004, 2.567811965942383, -77.06587219238281, 30.24553108215332, 48.069793701171875, 27.181640625, 8.288276672363281, 88.0299072265625, 20.210289001464844, -43.26043701171875, 83.88735961914062, 3.8418807983398438, 7.549320220947266, 61.680084228515625, -25.628082275390625, 76.48091888427734, 84.61416625976562, 12.90673828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000409.npy"}
{"epoch": 0.856544502617801, "step": 410, "batch_size": 128, "mean": 28.909753799438477, "std": 47.384037017822266, "min": -77.93450927734375, "p10": -25.79454650878906, "median": 24.88751220703125, "p90": 89.55203094482422, "max": 140.54034423828125, "pos_frac": 0.7265625, "sample": [39.0238037109375, -17.086227416992188, 9.925193786621094, -25.553955078125, -7.620658874511719, 87.19827270507812, 9.570796966552734, 23.470916748046875, -44.247222900390625, 81.45014953613281, 121.83810424804688, -31.53472900390625, 19.20818519592285, -28.1646728515625, 61.79572296142578, -62.894805908203125, -19.12933349609375, 13.930633544921875, 92.98175048828125, 31.73724365234375, 93.56982421875, -68.11833190917969, 28.392578125, 10.485679626464844, 90.70086669921875, 3.58477783203125, 52.716064453125, 75.7044677734375, 59.046478271484375, 79.38917541503906, 31.29736328125, 41.877105712890625, 23.3175048828125, 65.88240814208984, -1.17376708984375, 52.03919219970703, 14.093475341796875, 70.38592529296875, 50.143402099609375, 89.48097229003906, 91.49395751953125, 26.304107666015625, -5.10809326171875, 89.71783447265625, -67.91438293457031, 56.1571044921875, -77.93450927734375, 21.1651611328125, -5.6400146484375, -6.2062835693359375, 68.53765869140625, 83.66940307617188, 48.135459899902344, -22.222747802734375, -12.509307861328125, 48.710693359375, -3.54736328125, 86.61216735839844, 35.705413818359375, 37.893890380859375, -25.340389251708984, 97.45721435546875, 38.6976318359375, 88.9056396484375, 0.5366249084472656, 4.582427978515625, -20.639190673828125, 0.78570556640625, 36.104583740234375, 8.680694580078125, 64.08460998535156, 73.11251068115234, 38.23409652709961, -8.85894775390625, -26.355926513671875, 0.0, 49.08826446533203, 12.56494140625, 80.31124877929688, 59.198516845703125, 129.94888305664062, 7.386566162109375, 11.8045654296875, -14.265060424804688, -22.48877716064453, 86.5965576171875, 21.63123321533203, 26.4180908203125, -18.084030151367188, 18.154884338378906, -76.75173950195312, 16.906982421875, -37.9561767578125, 51.17726135253906, -18.67181396484375, 115.199951171875, -75.30197143554688, 78.49221801757812, 12.941139221191406, 19.557815551757812, 20.554550170898438, 14.952857971191406, 32.70668029785156, -60.93682861328125, 32.207122802734375, 12.456466674804688, 64.10420227050781, 70.63482666015625, 128.65359497070312, 60.042762756347656, 73.440673828125, -17.698974609375, 62.62413024902344, 20.28232192993164, 0.459625244140625, 70.755126953125, 87.92239379882812, -35.466461181640625, 41.11628723144531, 9.336593627929688, 3.9044876098632812, 91.91696166992188, 111.13873291015625, 37.870330810546875, 35.18782043457031, -21.117294311523438, 140.54034423828125, -2.7222747802734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000410.npy"}
{"epoch": 0.8586387434554974, "step": 411, "batch_size": 128, "mean": 33.185630798339844, "std": 44.51335525512695, "min": -87.3184814453125, "p10": -15.894064331054686, "median": 25.302257537841797, "p90": 90.70686569213866, "max": 133.68682861328125, "pos_frac": 0.7734375, "sample": [56.958274841308594, 91.78411102294922, 30.198974609375, 73.1707763671875, 16.532958984375, 2.9292449951171875, -1.0089797973632812, 23.6912841796875, 25.417503356933594, 77.64825439453125, 76.64947509765625, 55.94464111328125, -2.4844398498535156, -15.511749267578125, -45.44134521484375, -31.467010498046875, 114.8504638671875, -7.9589996337890625, 79.500244140625, 24.97589111328125, 33.757110595703125, 45.607025146484375, 71.69633483886719, 37.4862060546875, 66.47074890136719, 56.95654296875, 30.5733642578125, 133.68682861328125, 54.077423095703125, -10.811431884765625, 101.95321655273438, 6.40211296081543, 43.50843048095703, 20.437210083007812, 74.45315551757812, -11.681396484375, 104.968994140625, 73.11343383789062, 25.12957763671875, 14.688507080078125, 3.624542236328125, 24.166473388671875, 2.6148834228515625, 56.79109191894531, 12.599075317382812, -44.015525817871094, 11.64813232421875, 10.421844482421875, -11.87200927734375, 68.59266662597656, 80.2286148071289, 20.829071044921875, -3.9494781494140625, 2.849231719970703, 11.51632022857666, 91.30303955078125, -14.312431335449219, 59.45831298828125, 17.8878173828125, 8.885276794433594, -0.1450023651123047, 15.578170776367188, -3.01983642578125, -87.3184814453125, 73.60345458984375, 61.485687255859375, 29.00457763671875, 83.37890625, 7.5430908203125, 107.27313232421875, 6.634424209594727, 70.24264526367188, 12.998184204101562, 89.27752685546875, 43.424888610839844, 105.60926818847656, 11.736373901367188, 8.959869384765625, 90.45136260986328, 72.41619873046875, 10.254165649414062, -80.37347412109375, 99.65286254882812, 4.624017715454102, 12.3328857421875, -56.68116760253906, 21.69342041015625, 1.207265853881836, 75.31134033203125, 81.59933471679688, 55.7607421875, -16.7861328125, 25.18701171875, 30.821502685546875, 17.1834716796875, 20.497268676757812, 31.03668212890625, 125.01591491699219, 54.089881896972656, -41.572662353515625, 61.296356201171875, -27.237808227539062, -44.016357421875, -14.995330810546875, 9.453876495361328, -4.94439697265625, 35.08827209472656, 28.163818359375, 49.12059020996094, 66.43135070800781, -21.605920791625977, 48.636474609375, 90.08427429199219, -5.009796142578125, 9.57373046875, -18.435928344726562, 42.18489074707031, 130.80908203125, 75.98870849609375, 75.86502075195312, 69.03472900390625, 112.77003479003906, 63.60015869140625, 122.70693969726562, 27.53759002685547, -4.3135223388671875, -22.382843017578125, -5.7518310546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000411.npy"}
{"epoch": 0.8607329842931937, "step": 412, "batch_size": 128, "mean": 30.089038848876953, "std": 47.61115646362305, "min": -110.0745849609375, "p10": -22.359368896484376, "median": 22.687110900878906, "p90": 97.73153076171874, "max": 152.40542602539062, "pos_frac": 0.6953125, "sample": [-60.405364990234375, 98.69129943847656, 5.396488189697266, 98.37103271484375, 18.485595703125, 43.84765625, 23.160476684570312, 75.10809326171875, -17.911209106445312, -19.557159423828125, 22.62371826171875, 29.974609375, -2.6136245727539062, -19.86534309387207, 23.189971923828125, -110.0745849609375, -0.2841339111328125, -17.866424560546875, -22.4150390625, 18.401390075683594, 0.417449951171875, 2.069305419921875, 15.02197265625, 65.95166015625, 91.45756530761719, 78.55315399169922, 99.00382995605469, 20.90021514892578, -11.287994384765625, 45.244850158691406, 66.17483520507812, 61.6689453125, -6.933319091796875, 99.09228515625, 103.30450439453125, 47.684967041015625, 97.45745849609375, 14.593818664550781, 55.41883850097656, 49.979522705078125, 61.95982360839844, -19.662872314453125, 83.21578979492188, -43.5458984375, 38.11067199707031, -0.948486328125, -2.8119468688964844, 5.17333984375, -57.433074951171875, 23.727386474609375, 52.30169677734375, 99.06008911132812, 109.79937744140625, -15.100372314453125, 98.69722747802734, -2.2875213623046875, -11.492324829101562, 45.97984313964844, -31.090675354003906, -7.039825439453125, -22.33551025390625, -4.5886688232421875, 115.19854736328125, 47.02130126953125, -5.0901641845703125, 93.24002075195312, 67.61427307128906, 152.40542602539062, 60.76971435546875, 52.8194580078125, -40.13226318359375, 71.35841369628906, -31.273193359375, -12.09442138671875, 34.702362060546875, 29.635116577148438, 103.98472595214844, 92.85456848144531, 7.111541748046875, -76.86212158203125, 33.698028564453125, 40.818458557128906, 79.63369750976562, 15.697494506835938, 82.62349700927734, 10.075088500976562, 89.96987915039062, 83.40058898925781, 105.20111083984375, -34.62330627441406, 12.715408325195312, 91.55868530273438, 72.39521789550781, -47.981658935546875, -2.362762451171875, 78.78573608398438, -10.729904174804688, 85.01547241210938, 7.0946044921875, 69.25315856933594, 1.79156494140625, 0.5615520477294922, 108.6575927734375, 4.486610412597656, 57.9691162109375, 5.110370635986328, -32.91996765136719, 46.40742492675781, 5.5448455810546875, 12.764190673828125, 83.32754516601562, 22.750503540039062, 0.0, 58.13957214355469, -9.72149658203125, 62.466278076171875, -1.6174163818359375, 8.10955810546875, -43.15400695800781, 3.415708541870117, 9.907417297363281, 48.5810546875, -19.362777709960938, 53.40264892578125, -0.3520050048828125, 21.96697998046875, 47.167144775390625, 88.77597045898438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000412.npy"}
{"epoch": 0.86282722513089, "step": 413, "batch_size": 128, "mean": 26.518516540527344, "std": 40.756072998046875, "min": -68.24368286132812, "p10": -19.27244720458984, "median": 18.46945095062256, "p90": 83.85923690795899, "max": 138.2369384765625, "pos_frac": 0.734375, "sample": [18.892118453979492, -6.5755615234375, -0.8950634002685547, -21.635475158691406, -1.40692138671875, 77.47982788085938, 26.4241943359375, 18.046783447265625, -55.69769287109375, -14.732315063476562, 79.614990234375, 33.36572265625, 83.35865783691406, -36.277923583984375, 16.16046142578125, 9.61398696899414, 17.37957763671875, 97.10836791992188, 41.028411865234375, -24.9573974609375, 34.093994140625, 17.026226043701172, 64.47721862792969, 12.528656005859375, 21.331558227539062, 4.157127380371094, 6.219032287597656, 111.35302734375, 7.703519821166992, 47.49901580810547, 45.371551513671875, -2.19366455078125, -20.4700927734375, 54.1353759765625, -7.310394287109375, 83.66266632080078, 69.61509704589844, 78.95254516601562, 84.82081604003906, 2.2598648071289062, 54.97288513183594, 1.9384002685546875, -39.004852294921875, 1.786407470703125, 24.21337890625, -19.095489501953125, -15.630653381347656, 47.05189514160156, 57.20855712890625, 44.01220703125, -23.86785888671875, -3.896240234375, -17.763778686523438, 21.467803955078125, 21.01727294921875, 64.93840789794922, 91.3509521484375, -19.685348510742188, 29.047767639160156, 37.02819061279297, 94.97659301757812, 10.40338134765625, 12.411331176757812, 10.381080627441406, 25.12542724609375, 125.0938720703125, -9.193920135498047, 54.335693359375, 85.7372817993164, -29.31671142578125, 60.96575927734375, 42.79946517944336, -18.19732666015625, 138.2369384765625, 19.02880096435547, -2.7263870239257812, 4.1835174560546875, 23.106796264648438, -7.353389739990234, 51.37806701660156, 16.009902954101562, 71.41546630859375, -35.39447021484375, 52.56695556640625, 78.56291961669922, -8.835906982421875, 11.72314453125, 47.529052734375, 3.7716331481933594, -37.1495361328125, 85.70437622070312, 14.070302963256836, -12.46673583984375, 50.8389892578125, 22.9273681640625, -17.05948257446289, -11.305458068847656, -10.333740234375, 78.41464233398438, 11.772333145141602, 4.734302520751953, -7.769073486328125, 6.8545989990234375, 75.02899169921875, 72.8775634765625, 55.48748779296875, 4.717729568481445, 8.292877197265625, 43.268707275390625, 7.3004913330078125, 23.97222900390625, 8.37200927734375, 29.200950622558594, 78.78829956054688, -10.2784423828125, -44.284027099609375, 35.531494140625, 78.90228271484375, -68.24368286132812, 128.4569091796875, 28.389556884765625, 84.85366821289062, 9.093338012695312, 0.172332763671875, 91.46441650390625, 84.31790161132812, 5.855621337890625, 26.2557373046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000413.npy"}
{"epoch": 0.8649214659685864, "step": 414, "batch_size": 128, "mean": 30.826454162597656, "std": 48.12659454345703, "min": -106.382080078125, "p10": -22.10957336425781, "median": 19.777667999267578, "p90": 95.2318344116211, "max": 144.91006469726562, "pos_frac": 0.71875, "sample": [36.456504821777344, -65.7720947265625, 48.415672302246094, 62.739166259765625, 2.822296142578125, 80.4927978515625, 12.22564697265625, 51.686767578125, 8.94757080078125, -37.748291015625, -1.2032012939453125, 6.352668762207031, 99.70074462890625, 16.570281982421875, -4.6809844970703125, -1.3195381164550781, -14.791473388671875, -32.6007080078125, 80.50772094726562, -17.441890716552734, -106.382080078125, -15.64404296875, 95.13824462890625, 6.296875, 62.9593505859375, -38.115814208984375, -21.567169189453125, 1.8995018005371094, 66.89791870117188, 76.97738647460938, -14.314552307128906, -10.27838134765625, 42.94804382324219, 81.29603576660156, -44.06353759765625, 69.09254455566406, 81.3916015625, 15.381607055664062, 115.0943603515625, 6.3563690185546875, 31.953720092773438, 92.09334564208984, 12.7188720703125, -0.747161865234375, 60.21498107910156, 79.41162109375, 4.7330322265625, 12.227127075195312, 34.239501953125, 110.46435546875, -6.0611114501953125, -11.787483215332031, -1.3369808197021484, 120.77517700195312, 64.71533203125, 4.203216552734375, 77.09085083007812, 87.70301818847656, 104.22259521484375, -47.2462158203125, 31.2745361328125, 5.232564926147461, -1.1343002319335938, 79.14901733398438, 5.423095703125, 53.102630615234375, 48.57831573486328, 56.53386306762695, 49.16656494140625, -3.5480499267578125, 64.212158203125, -6.886283874511719, -7.98828125, 114.5653076171875, 74.85977935791016, 76.13272094726562, 83.05073547363281, 5.22406005859375, 77.72796630859375, 70.59186553955078, 24.652023315429688, -6.08056640625, -52.30272674560547, 113.37210083007812, 126.66766357421875, 95.45021057128906, 104.8011474609375, 19.922813415527344, -3.2717819213867188, 74.89096069335938, -85.69021606445312, 6.561492919921875, 20.67974853515625, 68.7451171875, 5.8199005126953125, 86.12014770507812, -17.036300659179688, 1.799835205078125, -1.3713226318359375, 20.395095825195312, -15.409622192382812, 47.9647216796875, 144.91006469726562, 13.022064208984375, 10.389228820800781, 5.71893310546875, 79.84334564208984, 19.632522583007812, 53.41502380371094, 23.118377685546875, -23.37518310546875, 34.489776611328125, 109.9105224609375, 28.859161376953125, -2.8486156463623047, 18.52276611328125, 62.23821258544922, 103.8243408203125, 18.391143798828125, 25.253036499023438, -25.010848999023438, 0.8673095703125, -53.4432373046875, -25.19976806640625, 19.5035400390625, 71.28836059570312, 86.52093505859375, 1.6845245361328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000414.npy"}
{"epoch": 0.8670157068062827, "step": 415, "batch_size": 128, "mean": 35.140228271484375, "std": 45.533878326416016, "min": -86.2144775390625, "p10": -18.70803909301758, "median": 30.472042083740234, "p90": 97.60929260253906, "max": 132.2901611328125, "pos_frac": 0.7421875, "sample": [14.42426586151123, 55.92927551269531, 12.272453308105469, 13.463510513305664, -25.347442626953125, 0.617462158203125, 74.43473052978516, -9.842498779296875, 76.7974624633789, 10.343582153320312, 22.39104461669922, -4.825260162353516, 77.17178344726562, 50.71318054199219, -18.970298767089844, 4.557212829589844, 85.72584533691406, 3.532012939453125, 68.96035766601562, 18.500991821289062, 77.95166015625, 12.36993408203125, 96.54287719726562, -1.056488037109375, 52.47151184082031, 16.514404296875, 50.65740203857422, 100.09759521484375, 88.4699935913086, 42.2196044921875, -38.57743835449219, 104.50762939453125, 79.08319091796875, 0.74884033203125, -34.01091003417969, 17.798095703125, 52.87310791015625, 100.8701171875, -3.486083984375, 43.69203186035156, 41.46904754638672, 8.74078369140625, -1.2273406982421875, 17.590972900390625, 7.2705078125, -12.728134155273438, 94.93576049804688, 44.6983642578125, 109.1572265625, 76.68418884277344, -9.764259338378906, 24.880157470703125, -44.50396728515625, 50.21795654296875, 70.90713500976562, 88.52462768554688, -13.93951416015625, 58.21221923828125, 24.3092041015625, -25.30316162109375, 34.666900634765625, 78.888916015625, -19.298294067382812, 35.90589904785156, 132.2901611328125, -20.15887451171875, 14.065912246704102, 29.28282928466797, 1.229522705078125, -0.7034149169921875, -7.4550323486328125, 107.66378784179688, 55.3602294921875, 26.287521362304688, -2.45623779296875, -3.8413238525390625, 84.11056518554688, 116.4161376953125, 61.23309326171875, -34.60308837890625, 56.6365966796875, -5.334381103515625, 15.033126831054688, -38.7535400390625, 90.31982421875, -13.916961669921875, -68.54100799560547, 38.71270751953125, 51.37371826171875, -18.59564208984375, -14.00616455078125, 0.8642578125, 50.35713195800781, 91.709716796875, 80.52520751953125, -11.6295166015625, 42.74658203125, 29.702911376953125, 73.63284301757812, -86.2144775390625, 117.6148681640625, 20.004371643066406, 44.00885009765625, 41.623016357421875, 25.262451171875, 106.0595703125, 31.6390380859375, -16.293777465820312, 89.7445068359375, 52.06159973144531, 85.10368347167969, 31.241172790527344, 80.20231628417969, 53.79438018798828, -35.10460662841797, 120.62176513671875, 13.348403930664062, 82.99095153808594, 101.33529663085938, -6.732368469238281, 5.6469268798828125, 15.15911865234375, 83.3050537109375, 18.47735595703125, 101.70834350585938, 68.08876037597656, -12.469833374023438, 125.281494140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000415.npy"}
{"epoch": 0.8691099476439791, "step": 416, "batch_size": 128, "mean": 35.16900634765625, "std": 47.73186492919922, "min": -104.434814453125, "p10": -19.3599853515625, "median": 30.755165100097656, "p90": 95.34873657226562, "max": 135.77349853515625, "pos_frac": 0.7578125, "sample": [92.24740600585938, 129.63803100585938, 101.74432373046875, 24.146728515625, 1.3825607299804688, 0.892547607421875, -26.538330078125, 39.80305480957031, -5.902313232421875, 69.62478637695312, -31.8350830078125, 135.77349853515625, 27.640472412109375, 10.815383911132812, 7.795867919921875, 12.629692077636719, -7.09930419921875, 103.3883056640625, 129.216796875, 28.7845458984375, 44.085609436035156, 50.13115692138672, 31.483261108398438, 102.03578186035156, -16.0791015625, 27.723861694335938, -28.355247497558594, 73.935791015625, -31.754547119140625, 12.119186401367188, 63.44184875488281, 86.03137969970703, 83.04545593261719, 0.5352783203125, -1.3377838134765625, 107.22918701171875, -62.26091003417969, 74.33853149414062, 17.721588134765625, 14.450042724609375, 15.55859375, 13.88552474975586, -34.7177734375, -104.434814453125, -10.089946746826172, 70.97775268554688, 24.497238159179688, 72.01063537597656, 72.63168334960938, 78.96087646484375, 35.622589111328125, 94.99778747558594, 13.906524658203125, 5.1858062744140625, -58.725067138671875, 94.32117462158203, 86.20355224609375, 65.64802551269531, -3.4774169921875, 2.295501708984375, 44.73651123046875, 53.89790344238281, -64.05326080322266, -19.20123291015625, 111.42376708984375, 37.40313720703125, -9.803245544433594, 69.43182373046875, -11.07891845703125, 76.39671325683594, -0.7199859619140625, 15.140762329101562, 34.10980224609375, 70.50994110107422, 133.69223022460938, 66.74925231933594, 41.4659423828125, 76.3970947265625, -0.6651153564453125, 9.808151245117188, 23.62646484375, 27.1541748046875, 86.0921630859375, -18.338829040527344, -21.05987548828125, 45.24546813964844, 16.07891845703125, 58.580322265625, -1.4292755126953125, 21.72820281982422, 21.753578186035156, 65.43806457519531, 33.859642028808594, 62.32965087890625, -59.306427001953125, 89.236572265625, 45.909210205078125, 0.50848388671875, 74.80963134765625, 96.16761779785156, -43.805450439453125, 51.34986877441406, 37.310455322265625, 3.2528839111328125, 49.867645263671875, -8.03546142578125, 115.75250244140625, 54.642120361328125, 120.6751708984375, -18.348785400390625, 2.2595596313476562, 127.21539306640625, 37.666259765625, 9.482666015625, -8.67340087890625, 94.50935363769531, 45.594146728515625, 61.384735107421875, 26.60193634033203, -4.936756134033203, 56.08586120605469, -19.73040771484375, 86.023193359375, -11.17886734008789, 89.70510864257812, 3.6587677001953125, 30.027069091796875, 81.3243408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000416.npy"}
{"epoch": 0.8712041884816754, "step": 417, "batch_size": 128, "mean": 30.762866973876953, "std": 45.85350799560547, "min": -86.29434204101562, "p10": -20.71733169555664, "median": 31.94023895263672, "p90": 92.37588500976562, "max": 136.8662109375, "pos_frac": 0.734375, "sample": [2.539691925048828, 68.37078857421875, 17.843490600585938, 35.336639404296875, -20.021469116210938, 77.61517333984375, 16.14532470703125, 18.407516479492188, 104.22442626953125, 115.7086181640625, 3.8562469482421875, 55.6688232421875, 6.393157958984375, 54.09511184692383, -66.60494232177734, 5.588876724243164, 9.7799072265625, -8.091888427734375, -14.845458984375, 80.01931762695312, 93.00564575195312, -15.256301879882812, 100.54025268554688, -30.364044189453125, -21.09516143798828, -71.01314544677734, 28.6544189453125, 4.206535339355469, -20.555404663085938, 115.07958984375, 48.00627899169922, -8.262237548828125, 13.9193115234375, 47.40522003173828, 43.61320495605469, 87.85699462890625, 14.489105224609375, -37.03070068359375, 61.08482360839844, 67.49147033691406, 83.15585327148438, 15.31982421875, 36.77754211425781, -8.919952392578125, 42.93145751953125, 57.319854736328125, 13.605926513671875, 63.6748046875, 62.092315673828125, 47.04585266113281, 40.911590576171875, -0.13301658630371094, 6.2570037841796875, -38.180145263671875, -43.7391357421875, 40.63707733154297, -10.53656005859375, 92.10598754882812, 1.6598358154296875, -11.75921630859375, 19.10272216796875, 32.23785400390625, 2.3887195587158203, 12.647727966308594, 69.28619384765625, 15.833572387695312, 64.11328887939453, -24.6446533203125, -86.29434204101562, 1.640899658203125, -15.022552490234375, 64.88409423828125, 67.21588134765625, 31.642623901367188, 37.7061767578125, -18.24749755859375, 0.2525177001953125, 94.85443115234375, 0.0, -1.733123779296875, 58.562477111816406, 0.0, 40.873138427734375, 70.84295654296875, 58.33890151977539, 46.328338623046875, 21.263015747070312, 5.5672760009765625, 75.39654541015625, -15.60443115234375, -16.034942626953125, 85.15225219726562, 55.361610412597656, 86.8704833984375, 89.5751953125, 40.41094970703125, -60.19390869140625, 42.895111083984375, 131.9586181640625, 6.670501708984375, 60.8638916015625, -53.64142608642578, -47.4473876953125, 40.64840316772461, 60.6370849609375, 86.21119689941406, 114.97976684570312, -13.364410400390625, -28.897003173828125, 8.274078369140625, 44.89263916015625, -6.5253753662109375, 10.303857803344727, 101.23797607421875, 108.98464965820312, 67.91595458984375, -10.68994140625, 37.16996765136719, 102.64559936523438, 5.944974899291992, 55.008758544921875, 27.457473754882812, 41.85687255859375, -0.496002197265625, 68.70938110351562, 44.13726806640625, 136.8662109375, 107.83197021484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000417.npy"}
{"epoch": 0.8732984293193717, "step": 418, "batch_size": 128, "mean": 27.95363998413086, "std": 45.68138885498047, "min": -82.21273803710938, "p10": -27.055971908569337, "median": 28.32791519165039, "p90": 90.19371109008789, "max": 135.25738525390625, "pos_frac": 0.7109375, "sample": [-50.746429443359375, 79.48543548583984, 25.434539794921875, 36.452392578125, -24.543365478515625, -38.871795654296875, 1.4049072265625, 106.06912231445312, -26.96832275390625, -82.21273803710938, 37.55914306640625, 42.438751220703125, 36.563232421875, -49.484344482421875, 67.41165161132812, 22.213104248046875, 49.785621643066406, -25.15008544921875, 66.280517578125, 4.3154754638671875, 75.36517333984375, 58.35844421386719, -9.8245849609375, 0.11545372009277344, 1.591796875, 6.080291748046875, -36.10577392578125, 75.66741943359375, 31.1851806640625, 111.89849853515625, 85.3843765258789, 36.6322021484375, 94.689453125, 68.78811645507812, 14.989120483398438, 68.34823608398438, -21.043182373046875, 74.67599487304688, -31.204010009765625, -8.926856994628906, 2.9292449951171875, 45.9749755859375, 13.171875, 38.570770263671875, -0.6152496337890625, 29.238861083984375, -6.424625396728516, 36.842552185058594, 26.080902099609375, -14.150131225585938, -0.7255859375, 94.16546630859375, 68.53723907470703, 24.820556640625, 40.370330810546875, -6.326751708984375, 6.62890625, 60.787811279296875, 39.96168518066406, 27.970123291015625, -70.0030517578125, 57.630828857421875, 82.92977905273438, 41.79960632324219, 30.384994506835938, 8.894744873046875, 69.57086944580078, -17.35662078857422, -35.184539794921875, 0.7661514282226562, -26.348052978515625, 11.4658203125, 69.49667358398438, 19.46356201171875, 53.30561828613281, 103.24021911621094, 89.5106201171875, -4.8012847900390625, 59.693023681640625, -39.885467529296875, 34.73541259765625, 74.82839965820312, -5.938507080078125, 42.364959716796875, 31.59429931640625, 90.09717559814453, 84.42478942871094, -15.798259735107422, 8.87054443359375, -54.41619873046875, 6.981903076171875, -19.235321044921875, -15.28515625, 68.4552001953125, 77.21345520019531, 9.228996276855469, 102.49801635742188, 54.233795166015625, 9.07208251953125, 86.15499877929688, -22.128936767578125, 1.016693115234375, 11.824869155883789, 30.15119171142578, -58.58343505859375, 20.407928466796875, 52.947021484375, 72.88973236083984, -60.36370849609375, 96.64208984375, 14.861600875854492, 13.22711181640625, 38.37611389160156, -0.7653579711914062, 90.41896057128906, 28.685707092285156, 135.25738525390625, 95.10182189941406, -9.029243469238281, 134.75924682617188, 32.651519775390625, -27.260486602783203, -7.72833251953125, 48.88298034667969, -17.24049186706543, 96.60359191894531, 92.88546752929688, -2.9900054931640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000418.npy"}
{"epoch": 0.875392670157068, "step": 419, "batch_size": 128, "mean": 36.83320617675781, "std": 49.65074920654297, "min": -119.45622253417969, "p10": -24.75962753295898, "median": 35.00062561035156, "p90": 99.75530700683593, "max": 138.72528076171875, "pos_frac": 0.765625, "sample": [37.285400390625, 15.007583618164062, 6.694305419921875, 96.05426025390625, 20.856910705566406, 21.49786376953125, 6.981132507324219, 16.806655883789062, 51.61048889160156, -43.70574951171875, -56.559410095214844, 60.99163818359375, -61.0828857421875, 79.99311828613281, 11.890861511230469, 32.835235595703125, 122.88491821289062, 9.328788757324219, 47.89350891113281, 90.39157104492188, 81.16517639160156, 66.7347412109375, 100.42807006835938, 23.086868286132812, 104.626220703125, 9.1634521484375, 37.166015625, -1.0008544921875, -29.080963134765625, -46.1990966796875, 92.87276458740234, 85.95144653320312, 69.60652160644531, -2.8073806762695312, 25.7008056640625, -0.556182861328125, 62.289581298828125, 0.5670032501220703, 76.48809814453125, -55.74156188964844, 80.11203002929688, 9.290596008300781, 89.98239135742188, 57.386260986328125, 75.23846435546875, 5.752906799316406, 113.65364074707031, -37.30305480957031, 44.2940673828125, 106.42059326171875, 58.789520263671875, 62.9381103515625, -17.262855529785156, 54.02649688720703, 59.90330505371094, 75.2484130859375, 124.87139892578125, 99.46697998046875, 20.582275390625, -13.2022705078125, -23.267303466796875, 10.202423095703125, -22.595741271972656, 2.524700164794922, 123.59454345703125, -23.4228515625, 84.21588134765625, 102.69258117675781, -0.826568603515625, 32.621826171875, -23.50818634033203, 69.21575927734375, 67.941162109375, 18.689163208007812, 44.36968994140625, 0.0, -48.19099426269531, 13.715179443359375, 7.1473846435546875, 0.0, 21.04803466796875, 78.20429992675781, 81.2393798828125, 94.023681640625, -17.113422393798828, 80.3951416015625, 10.651504516601562, 101.12435913085938, 46.775177001953125, 10.862716674804688, 105.81478881835938, 22.41455078125, 101.60595703125, 47.327484130859375, -1.4588470458984375, 8.1439208984375, 73.08265686035156, 32.53021240234375, 88.00650024414062, 99.44354248046875, 91.12167358398438, 14.41253662109375, 138.72528076171875, 8.962020874023438, 78.95191955566406, -27.679656982421875, 79.22151184082031, 83.04605102539062, 44.89410400390625, 74.34523010253906, -41.62103271484375, -39.062225341796875, 109.0233154296875, 83.38925170898438, -119.45622253417969, 39.028594970703125, -71.93660736083984, -9.2344970703125, 55.62773132324219, -18.526275634765625, 48.477237701416016, 77.07736206054688, -13.93145751953125, 17.549407958984375, 16.240829467773438, 30.280187606811523, 83.76263427734375, 24.415542602539062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000419.npy"}
{"epoch": 0.8774869109947644, "step": 420, "batch_size": 128, "mean": 34.891143798828125, "std": 47.19925308227539, "min": -108.10786437988281, "p10": -21.775660705566406, "median": 35.539306640625, "p90": 94.99041137695312, "max": 142.76959228515625, "pos_frac": 0.7578125, "sample": [-40.37116241455078, 19.150482177734375, 21.42169189453125, -97.64801025390625, -19.169418334960938, -18.57545280456543, 23.58740234375, 88.74307250976562, 4.249267578125, 55.02978515625, 87.47052764892578, -3.37359619140625, -44.36346435546875, 42.345489501953125, 72.00613403320312, 5.682212829589844, 79.29950714111328, -45.045326232910156, 81.4677734375, 62.018150329589844, 33.995574951171875, 38.235443115234375, -7.553855895996094, 33.4869384765625, 2.3997650146484375, 96.15890502929688, 20.106414794921875, 45.218467712402344, 38.056663513183594, 10.188507080078125, -2.33709716796875, 92.05670166015625, 52.68280029296875, 56.05609130859375, 8.0855712890625, -0.5401763916015625, 69.90254974365234, 90.6268310546875, 111.05075073242188, 4.59332275390625, 99.9453125, 37.695953369140625, 37.17339324951172, 85.24307250976562, 12.73748779296875, 32.48419189453125, 37.083038330078125, -1.281707763671875, -2.0619354248046875, 51.877593994140625, -12.563980102539062, 59.4381103515625, 56.3551025390625, 95.5635986328125, 72.066650390625, 48.1739501953125, 0.0, 23.694229125976562, -23.71795654296875, -63.417449951171875, 71.20140838623047, 74.44306182861328, 8.085693359375, 14.753768920898438, 18.4449462890625, 26.552444458007812, 0.4781494140625, 102.84304809570312, 123.12176513671875, 78.27255249023438, 70.90512084960938, -21.533737182617188, 21.86944580078125, 66.29188537597656, -31.651885986328125, 100.62332153320312, 37.58757019042969, 14.275421142578125, 96.38800811767578, 31.669708251953125, -3.6004714965820312, 8.431549072265625, 142.76959228515625, -32.93133544921875, 87.84822082519531, 51.66900634765625, -14.117462158203125, -14.756385803222656, 90.0604248046875, 94.81045532226562, 43.77491760253906, 17.128692626953125, 97.763427734375, 95.41030883789062, 7.667816162109375, 29.17779541015625, 20.188655853271484, 48.578765869140625, 122.38604736328125, -3.07568359375, 113.72201538085938, 60.46311950683594, -108.10786437988281, -5.4677734375, 26.496261596679688, 85.39768981933594, 73.61038208007812, 7.32427978515625, -53.133148193359375, -29.60476303100586, 66.84951782226562, 71.00897216796875, 58.46989059448242, 90.52178955078125, 79.13360595703125, 20.04229736328125, 50.0675048828125, -57.22625732421875, 16.061614990234375, 74.00286865234375, 38.826629638671875, 91.75509643554688, -22.34014892578125, -9.574413299560547, 0.051116943359375, 77.8765869140625, -9.2547607421875, 52.404266357421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000420.npy"}
{"epoch": 0.8795811518324608, "step": 421, "batch_size": 128, "mean": 33.87506866455078, "std": 42.25980758666992, "min": -73.82699584960938, "p10": -17.16359100341797, "median": 32.26742935180664, "p90": 90.24558715820312, "max": 126.0670166015625, "pos_frac": 0.734375, "sample": [126.0670166015625, 29.480209350585938, 56.36457061767578, 23.7178955078125, 49.34300231933594, 52.30101013183594, 61.34552001953125, 7.847076416015625, -73.82699584960938, -0.6852951049804688, 80.33943176269531, -4.56671142578125, -5.597015380859375, 67.20556640625, 28.848133087158203, 41.004371643066406, 26.44622802734375, -61.332794189453125, 31.80474853515625, 36.830963134765625, -8.691841125488281, -0.767852783203125, 16.464553833007812, 31.91075897216797, 65.65279388427734, 114.15554809570312, 56.673248291015625, 33.639495849609375, 7.446186065673828, 48.11864471435547, 0.0, -62.77978515625, 56.615142822265625, 46.77758026123047, 7.36041259765625, 34.17156982421875, 94.83013916015625, 73.84588623046875, 25.425033569335938, 89.2396240234375, 94.87948608398438, 80.06404876708984, 33.7926025390625, 15.024515151977539, -1.1433372497558594, 8.603378295898438, 69.82735443115234, -18.320831298828125, 93.75225830078125, 65.55621337890625, 72.45989990234375, -8.64080810546875, 23.993789672851562, 40.159423828125, 16.10858154296875, 92.59283447265625, -35.90240478515625, 86.75301361083984, 13.105499267578125, 49.72279357910156, 29.7796630859375, 21.926300048828125, 116.64703369140625, 98.63711547851562, 66.07827758789062, 20.723163604736328, 60.49871826171875, 84.83187866210938, 87.4036865234375, 75.92872619628906, 82.50321960449219, 95.13836669921875, 45.604736328125, 76.75820922851562, 34.990882873535156, 46.159149169921875, -17.22454833984375, 23.5792236328125, 83.18037414550781, 8.521385192871094, 52.074249267578125, -17.137466430664062, 35.985687255859375, 105.80133056640625, -3.3646240234375, 14.536834716796875, 38.031829833984375, -17.130859375, 73.39898681640625, -45.43053436279297, 53.648902893066406, 60.651214599609375, 102.40884399414062, 24.316070556640625, -24.071895599365234, 30.036376953125, 45.118133544921875, -24.653270721435547, -53.22991943359375, 30.56188201904297, -26.3397216796875, -10.392242431640625, 30.51605224609375, -8.990478515625, 34.207366943359375, 85.17288208007812, 6.8008270263671875, -5.4554443359375, 60.319183349609375, -35.84271240234375, -20.534744262695312, 67.37570190429688, 32.62409973144531, 63.807403564453125, -13.204269409179688, 65.66307067871094, 22.832792282104492, 0.0, -16.037017822265625, -10.469947814941406, 114.21926879882812, 59.73046875, 21.814163208007812, 0.0, -1.690582275390625, -4.59271240234375, 105.45867919921875, 0.387054443359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000421.npy"}
{"epoch": 0.881675392670157, "step": 422, "batch_size": 128, "mean": 27.877517700195312, "std": 46.959720611572266, "min": -98.8424072265625, "p10": -26.995573806762692, "median": 26.770477294921875, "p90": 84.55679321289061, "max": 118.13247680664062, "pos_frac": 0.703125, "sample": [10.719146728515625, -11.709915161132812, 78.56980895996094, -39.61015319824219, 69.11752319335938, -13.692863464355469, 86.35015869140625, -69.218017578125, 40.678375244140625, -0.44775390625, 65.53573608398438, -1.05999755859375, 57.6522216796875, 81.6763916015625, 37.28448486328125, -94.47181701660156, 6.14654541015625, 9.666839599609375, -19.060699462890625, 70.76091766357422, 49.813934326171875, 33.475006103515625, 56.51564025878906, 15.856338500976562, -26.43902587890625, 77.1046142578125, -74.6270751953125, 58.390438079833984, 72.47586059570312, 108.74188232421875, 44.60972595214844, 1.6032257080078125, -23.445816040039062, 65.20358276367188, 7.1778564453125, 73.33502197265625, -3.242462158203125, -52.028526306152344, 95.13945007324219, 5.0529937744140625, 25.443023681640625, 78.767333984375, -23.4737548828125, 34.989349365234375, 37.42926025390625, -98.8424072265625, -9.6917724609375, 117.00704956054688, 43.15362548828125, 78.76643371582031, 36.7626953125, 74.20562744140625, 49.580596923828125, 19.952842712402344, 36.135162353515625, 3.393157958984375, -11.06488037109375, 23.8617000579834, 41.198974609375, 35.222625732421875, 74.91841125488281, -14.787525177001953, -16.518096923828125, -64.29779052734375, 93.47010803222656, 3.6817626953125, 32.77861785888672, 10.371627807617188, 2.1241836547851562, 110.05380249023438, 74.033447265625, -79.54327392578125, 80.2747802734375, -1.786407470703125, 50.680328369140625, -28.294185638427734, 79.6121826171875, -41.19451141357422, 105.09500122070312, 8.259315490722656, 103.084716796875, -2.7074661254882812, 44.285491943359375, 22.969520568847656, 101.70062255859375, 28.097930908203125, -1.044281005859375, 16.75811767578125, 83.7882080078125, -1.6233863830566406, -10.303413391113281, -21.4644775390625, 42.209320068359375, -12.99957275390625, 79.25678253173828, 52.93611145019531, -3.83612060546875, 4.210487365722656, 65.31329345703125, 28.455795288085938, 64.14688110351562, 9.816238403320312, 105.498291015625, 90.92205810546875, -34.5404052734375, 67.3026123046875, 40.6995849609375, 0.39569091796875, -16.880462646484375, 67.28448486328125, 0.0, -31.565338134765625, 80.71342468261719, 77.02020263671875, -31.718170166015625, 31.63604736328125, 7.87615966796875, 71.56987762451172, 23.135683059692383, -0.439605712890625, 16.5877685546875, 40.57067108154297, 2.7048110961914062, 100.2183837890625, 6.71331787109375, 20.68170166015625, -0.5756874084472656, 118.13247680664062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000422.npy"}
{"epoch": 0.8837696335078534, "step": 423, "batch_size": 128, "mean": 36.29489517211914, "std": 46.972381591796875, "min": -79.25, "p10": -20.03707275390625, "median": 27.254074096679688, "p90": 99.49750366210937, "max": 129.38287353515625, "pos_frac": 0.75, "sample": [5.5977935791015625, 104.37782287597656, -7.6066741943359375, 43.51020812988281, 87.73892211914062, 87.805908203125, 25.755653381347656, -12.5198974609375, 1.5121383666992188, 79.39836883544922, 27.39569091796875, 41.74114990234375, 27.112457275390625, 65.63291931152344, -23.3094482421875, -10.67926025390625, 86.0078125, -20.1104736328125, 79.64697265625, 3.92681884765625, 6.263402938842773, -20.795318603515625, 17.252212524414062, 73.20639038085938, -51.3272705078125, 61.63606262207031, 64.06935119628906, 9.395706176757812, 87.96479797363281, 108.90139770507812, 5.6110076904296875, 16.9454345703125, 26.7322998046875, 14.967315673828125, 59.3775634765625, 68.64773559570312, 65.0474853515625, -1.0975189208984375, -1.712249755859375, 45.34381103515625, 68.94608306884766, 98.58868408203125, -16.019943237304688, 60.05699157714844, -5.9906768798828125, 7.97357177734375, 4.443023681640625, 117.1856689453125, -29.2032470703125, -11.392486572265625, 110.9649658203125, -23.378372192382812, -15.739410400390625, 39.134979248046875, 78.43072509765625, -27.361785888671875, -46.27996826171875, -42.625091552734375, 106.74761962890625, 101.87942504882812, 15.040565490722656, -1.7107391357421875, 41.5416259765625, 12.687118530273438, -9.710136413574219, 65.33006286621094, 2.936788558959961, 74.88031005859375, 2.9745712280273438, 26.0882568359375, 58.207916259765625, 91.14505767822266, 51.41973876953125, 71.93673706054688, 33.01191711425781, 45.086822509765625, -20.005615234375, 85.58223724365234, 129.38287353515625, -18.741973876953125, 13.264495849609375, 98.75918579101562, 6.2044219970703125, 87.45086669921875, 118.88079833984375, 39.28913879394531, 54.8494873046875, 12.966827392578125, 12.131240844726562, 106.4705810546875, 77.95869445800781, 77.97702026367188, 92.91644287109375, 1.49310302734375, 116.41168212890625, 77.6146240234375, -0.08077430725097656, 5.736602783203125, 101.22024536132812, -7.29742431640625, 8.876388549804688, 50.919586181640625, 2.440338134765625, 108.01596069335938, 111.50665283203125, -17.0089111328125, -5.1390228271484375, 70.33552551269531, 71.45059204101562, -21.58612060546875, -66.33161926269531, 12.78196907043457, -18.328155517578125, 32.292236328125, 11.801822662353516, 91.4554443359375, 72.44747924804688, 78.33021545410156, 98.33735656738281, 18.079345703125, 74.00018310546875, 80.89794921875, 93.71692657470703, -31.6728515625, 9.157501220703125, 13.564620971679688, -2.3715877532958984, -79.25], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000423.npy"}
{"epoch": 0.8858638743455497, "step": 424, "batch_size": 128, "mean": 34.202178955078125, "std": 43.36287307739258, "min": -86.55245971679688, "p10": -17.297467041015622, "median": 33.86024475097656, "p90": 87.3489776611328, "max": 131.55087280273438, "pos_frac": 0.7578125, "sample": [53.813446044921875, 40.05474853515625, -0.8147430419921875, -8.32403564453125, 23.8421630859375, 16.998779296875, -2.8511962890625, -1.32421875, 26.2818603515625, -4.49664306640625, 5.646514892578125, -19.340667724609375, 131.55087280273438, 42.745147705078125, 0.85223388671875, 64.10718536376953, 4.9342041015625, 106.91455078125, 99.26661682128906, 81.7890396118164, -6.049541473388672, 6.0634765625, 112.60342407226562, 73.09771728515625, 59.14056396484375, 44.84894561767578, -3.9186630249023438, 19.632232666015625, 63.1383056640625, 67.07434844970703, 38.02857208251953, 93.1011962890625, 91.5599365234375, -25.346641540527344, 70.48736572265625, 53.06208801269531, 16.423187255859375, 75.92913818359375, 85.1572265625, -11.7882080078125, -10.02569580078125, 81.05194091796875, 38.71075439453125, -86.55245971679688, -4.186180114746094, 121.5062255859375, 105.76718139648438, 57.929412841796875, -77.59584045410156, 31.20684814453125, -19.086719512939453, 1.19580078125, 0.77960205078125, 85.01316833496094, 45.19422912597656, 87.71792602539062, -2.850616455078125, 83.84518432617188, 78.60751342773438, 42.447845458984375, 24.626373291015625, -5.434661865234375, 66.37530517578125, 32.51800537109375, 109.891357421875, 69.6595230102539, 35.202484130859375, 31.173919677734375, -42.93475341796875, -30.101837158203125, -25.542633056640625, 53.6324462890625, 70.6700439453125, 111.58575439453125, 87.19085693359375, 90.0849609375, 7.868965148925781, 56.30670928955078, 41.999725341796875, 16.2625732421875, 46.56935119628906, 47.268402099609375, -1.3744277954101562, 1.4508514404296875, 82.47991943359375, 19.329044342041016, -41.776123046875, 0.0, -26.535842895507812, 26.472137451171875, 9.99456787109375, 58.496360778808594, 69.17994689941406, -4.643951416015625, 36.611446380615234, 75.79267883300781, 66.31979370117188, 71.47543334960938, 78.08302307128906, 43.80016326904297, 9.925212860107422, -52.426666259765625, 50.174896240234375, 70.32364654541016, 39.352203369140625, 10.501253128051758, -12.01323127746582, 9.44378662109375, -18.863922119140625, 4.840272903442383, 62.43817138671875, -47.4346923828125, 75.144775390625, 84.01605224609375, -16.626129150390625, 4.919513702392578, 8.48084831237793, 25.434803009033203, 22.1322021484375, 46.0611572265625, 21.637847900390625, 59.14698028564453, -13.81048583984375, 25.434616088867188, 83.2801513671875, 9.406730651855469, 8.056285858154297, 104.3084716796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000424.npy"}
{"epoch": 0.8879581151832461, "step": 425, "batch_size": 128, "mean": 30.00723648071289, "std": 46.17476272583008, "min": -108.96038818359375, "p10": -22.26564178466796, "median": 23.517623901367188, "p90": 91.41291046142578, "max": 141.50442504882812, "pos_frac": 0.7421875, "sample": [43.20172119140625, 70.37039184570312, 54.04888916015625, 30.44294548034668, -13.6094970703125, 20.145843505859375, 34.65875244140625, 65.50161743164062, 10.07611083984375, -19.917007446289062, 88.5064697265625, 19.09857177734375, 92.17709350585938, 53.947174072265625, 70.05823516845703, 2.296550750732422, 56.36195755004883, 89.13995361328125, 1.3393898010253906, 17.84490966796875, 28.39593505859375, 61.300811767578125, 47.68170166015625, -5.296787261962891, 83.0977783203125, 41.47882843017578, 3.5588436126708984, 17.466796875, -55.41108703613281, -2.4072742462158203, 39.850608825683594, -1.562255859375, -10.977325439453125, 25.068130493164062, 94.70806884765625, -10.349485397338867, 6.0394287109375, 102.58340454101562, 5.529998779296875, 8.323051452636719, -13.329132080078125, 104.85403442382812, 68.9970474243164, 21.367279052734375, 1.0784912109375, 7.652091979980469, -18.953933715820312, -2.7515869140625, -50.938621520996094, 85.52667236328125, 59.47998046875, 67.43411254882812, -52.10888671875, 41.011688232421875, 93.95571899414062, 16.085861206054688, 84.17109680175781, 141.50442504882812, 32.74371337890625, 32.68597412109375, -19.057449340820312, 29.640869140625, 6.177341461181641, 28.744461059570312, 108.0167236328125, -29.145111083984375, 32.093994140625, -3.5138912200927734, 91.08540344238281, 53.41602325439453, 69.25271606445312, 7.20428466796875, 106.59619140625, 2.632200241088867, -17.379180908203125, -2.596221923828125, -8.007625579833984, 20.9893798828125, -5.6388092041015625, 47.594642639160156, 0.51776123046875, 93.33108520507812, 17.504150390625, 56.40641784667969, 101.66166687011719, -52.792388916015625, 20.261642456054688, -27.74578857421875, 38.534332275390625, -30.26641845703125, 14.568359375, -29.2343692779541, 14.952003479003906, -75.70851135253906, -5.0372314453125, -49.12457275390625, 63.257720947265625, 13.115425109863281, 68.72918701171875, -31.09735107421875, 31.002151489257812, 93.5555419921875, 34.536285400390625, 87.52040100097656, 27.24420166015625, 32.30607223510742, 88.25729370117188, 13.638351440429688, 2.3576507568359375, 83.6212158203125, 0.0, -19.60943603515625, 90.23193359375, 86.36569213867188, 127.53341674804688, -7.243749618530273, 9.955684661865234, 88.57072448730469, -108.96038818359375, 21.967117309570312, 74.11895751953125, 102.23393249511719, 63.584930419921875, 80.56712341308594, 69.6568603515625, -52.045616149902344, 5.862648010253906, 8.922584533691406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000425.npy"}
{"epoch": 0.8900523560209425, "step": 426, "batch_size": 128, "mean": 33.38637161254883, "std": 44.667999267578125, "min": -92.17056274414062, "p10": -14.06604537963867, "median": 22.839859008789062, "p90": 94.57042541503907, "max": 130.1754150390625, "pos_frac": 0.7734375, "sample": [82.04297637939453, -1.622802734375, 1.104766845703125, -27.837249755859375, 75.282470703125, 14.326858520507812, 70.75796508789062, 2.825897216796875, -55.5634765625, 8.917816162109375, -92.17056274414062, -31.175048828125, -31.430084228515625, 58.84614562988281, 87.90097045898438, 92.31655883789062, 81.658203125, 20.351959228515625, 77.11216735839844, 10.465377807617188, -37.89776611328125, 47.544952392578125, 106.42997741699219, 12.952674865722656, 1.2490310668945312, 18.479644775390625, 93.33126831054688, 102.89190673828125, 80.20626831054688, 94.52484130859375, 44.813148498535156, 1.6409034729003906, 67.6956787109375, 44.82176208496094, 77.70416259765625, 4.339689254760742, 2.0905685424804688, 16.26555633544922, 94.67678833007812, 102.94270324707031, 21.69921875, 97.84532165527344, 5.585975646972656, -4.964599609375, 82.44573974609375, 84.04156494140625, -0.8160400390625, 76.33311462402344, 9.231826782226562, -4.44921875, -28.800750732421875, 22.824981689453125, 19.72400665283203, 10.09783935546875, 22.116710662841797, 88.19503784179688, -72.4295654296875, 27.72245979309082, 90.0126953125, 59.756263732910156, 63.05583190917969, 66.65420532226562, -15.889739990234375, -2.4886245727539062, -3.7337188720703125, 66.95565795898438, 39.86512756347656, 56.61083984375, 0.758758544921875, -1.16259765625, 56.2459716796875, 14.240882873535156, -1.05413818359375, -3.718719482421875, 1.7386245727539062, 130.1754150390625, -10.0677490234375, 31.720458984375, 22.880081176757812, 44.35920715332031, 1.397308349609375, 85.30111694335938, 56.9237060546875, 53.164154052734375, 21.66131591796875, 23.153533935546875, 22.854736328125, 10.188278198242188, 6.886922836303711, 73.16278076171875, 101.549560546875, 11.933746337890625, -1.2621383666992188, 8.384185791015625, 100.21072387695312, 47.67083740234375, 44.7421875, 96.17648315429688, -25.626220703125, -13.284461975097656, 28.309295654296875, 14.4970703125, 0.0, 10.627983093261719, -9.261337280273438, 116.2952880859375, 50.62994384765625, 11.948974609375, 11.46337890625, 36.821563720703125, 76.6832275390625, 30.8663330078125, -51.40728759765625, 8.817779541015625, -5.099082946777344, 72.58843994140625, -19.573211669921875, 94.37872314453125, 86.40463256835938, 102.19891357421875, 65.643310546875, 4.405908584594727, -57.530609130859375, 51.858551025390625, 107.25433349609375, 38.721092224121094, -9.4161376953125, 96.00700378417969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000426.npy"}
{"epoch": 0.8921465968586387, "step": 427, "batch_size": 128, "mean": 43.68552780151367, "std": 44.86367416381836, "min": -82.5367431640625, "p10": -13.030965042114257, "median": 44.433719635009766, "p90": 100.97380371093749, "max": 135.1103515625, "pos_frac": 0.8203125, "sample": [15.743812561035156, 90.4180908203125, 14.1124267578125, -19.1632080078125, -13.002330780029297, 7.5850830078125, 30.676982879638672, 63.10125732421875, 53.03253173828125, 24.987396240234375, 58.437530517578125, 29.923263549804688, 108.25863647460938, 40.51971435546875, 84.91644287109375, 80.36965942382812, 67.549560546875, 6.778186798095703, 71.68997192382812, 115.51577758789062, 122.2601318359375, 41.04701232910156, -25.166671752929688, 37.105804443359375, 86.9144287109375, 48.5078125, 76.94061279296875, 85.9635009765625, 31.735485076904297, 45.790931701660156, 108.59857177734375, 60.93243408203125, 40.28167724609375, 27.446533203125, 2.405548095703125, 0.44101715087890625, 65.84677124023438, 12.8851318359375, -2.9935302734375, -24.423065185546875, 97.45191955566406, 15.6939697265625, -54.82635498046875, 42.993804931640625, -51.835906982421875, 37.59788513183594, -1.3319091796875, 102.54515075683594, -9.113838195800781, 77.74333953857422, 86.3184814453125, 15.67633056640625, 81.39508056640625, 64.7320556640625, 119.17236328125, 106.6572265625, 134.18408203125, -5.293479919433594, 19.097244262695312, -6.886436462402344, 59.12352752685547, 66.6787109375, 91.82980346679688, 16.89532470703125, 17.24847412109375, 31.179428100585938, -13.0977783203125, 78.04064178466797, 78.24685668945312, 4.662483215332031, -82.5367431640625, 94.06967163085938, 59.71015930175781, -14.923736572265625, 73.9139404296875, 15.692842483520508, -25.392852783203125, 79.64497375488281, 52.68608093261719, 25.01764678955078, 103.80755615234375, -53.673004150390625, 77.34233856201172, -29.8033447265625, 51.85809326171875, 75.12014770507812, 63.787025451660156, 56.175048828125, 43.076507568359375, 109.32562255859375, 74.83236694335938, 98.0633544921875, -20.7359619140625, 5.187225341796875, 89.33935546875, 69.95572662353516, 104.14801025390625, 52.369140625, 7.305763244628906, 56.563255310058594, -0.1266632080078125, 13.180404663085938, 36.78764343261719, -21.976028442382812, 15.05206298828125, 135.1103515625, -6.0220489501953125, 85.3909912109375, 12.472671508789062, 90.52777099609375, 97.4915771484375, 60.98847961425781, 61.457000732421875, 13.765411376953125, 32.32121276855469, 14.857009887695312, 78.07379150390625, 56.84088134765625, 2.303558349609375, 4.88018798828125, 93.09254455566406, 23.4129638671875, -1.4839019775390625, -3.9849853515625, 17.839202880859375, 71.22393798828125, 119.29570007324219, 100.30036926269531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000427.npy"}
{"epoch": 0.8942408376963351, "step": 428, "batch_size": 128, "mean": 41.03675079345703, "std": 43.75675964355469, "min": -73.96133422851562, "p10": -10.469613647460937, "median": 44.476375579833984, "p90": 90.0255859375, "max": 159.02841186523438, "pos_frac": 0.8125, "sample": [16.69629669189453, -7.02099609375, -25.700912475585938, 91.93524169921875, 84.28506469726562, -10.396270751953125, 42.542266845703125, 35.02789306640625, -10.6407470703125, 12.931976318359375, 159.02841186523438, 19.5745849609375, 75.84320068359375, 82.461181640625, -4.5928955078125, 102.02780151367188, 15.9608154296875, -4.9234619140625, 11.441566467285156, 0.17095947265625, 61.123931884765625, 19.394500732421875, -50.471771240234375, 120.79812622070312, 78.0919189453125, 88.868896484375, 65.00701904296875, 77.99015045166016, 36.11360168457031, 125.0908203125, 44.6021728515625, 102.56353759765625, 1.2815608978271484, 16.229515075683594, -11.384307861328125, 13.546630859375, -0.7903289794921875, 55.726959228515625, 27.747161865234375, 62.5360107421875, 42.590545654296875, -3.1051788330078125, -73.96133422851562, 55.584896087646484, 90.52275085449219, 130.79220581054688, -17.017066955566406, 44.35057830810547, 7.1341705322265625, 9.102645874023438, 79.18016052246094, 79.91932678222656, 71.04742431640625, 53.89900207519531, 21.764617919921875, 58.106414794921875, 84.63841247558594, 47.490570068359375, -41.285888671875, 128.6177978515625, -65.76070404052734, 107.79751586914062, 80.27275085449219, 33.00605773925781, 68.03155517578125, 7.1976470947265625, 76.03985595703125, 49.57615661621094, 56.78173828125, 39.46502685546875, 13.669052124023438, 35.198020935058594, 75.90459442138672, 76.69586181640625, 85.75558471679688, 9.875091552734375, 46.53302001953125, 0.174835205078125, 44.89970397949219, 20.938827514648438, 56.54718780517578, 24.61993408203125, 53.897705078125, 46.22222900390625, -9.358551025390625, 9.06195068359375, -14.332313537597656, 13.10797119140625, 17.166534423828125, 62.81581115722656, 140.564697265625, 97.93800354003906, 69.63546752929688, -4.3944091796875, 66.3251953125, -58.693756103515625, -20.19378662109375, 100.08404541015625, 80.94757080078125, 0.0, 9.99786376953125, 89.81251525878906, 79.30703735351562, 14.985595703125, 87.00387573242188, 56.21009063720703, -17.458328247070312, 66.76708984375, 52.009033203125, 61.56793212890625, 30.219329833984375, 65.40711975097656, 17.662567138671875, 14.025917053222656, 89.74745178222656, 48.613006591796875, 32.860992431640625, 31.58734130859375, 46.61956787109375, -3.665924072265625, 9.907073974609375, 23.80914306640625, 86.98147583007812, -10.135663986206055, -13.625335693359375, 76.59187316894531, 70.0230712890625, 51.76824188232422], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000428.npy"}
{"epoch": 0.8963350785340314, "step": 429, "batch_size": 128, "mean": 32.93855667114258, "std": 52.67481231689453, "min": -98.05766296386719, "p10": -30.721910858154295, "median": 25.30804443359375, "p90": 102.91313323974609, "max": 127.21807861328125, "pos_frac": 0.7109375, "sample": [73.68247985839844, -84.29302978515625, 5.814117431640625, 67.54362487792969, 89.82099914550781, 86.07763671875, 22.80272674560547, 24.02899169921875, -35.28102111816406, 73.00140380859375, 7.1088409423828125, -67.08712768554688, 31.116844177246094, 85.95391845703125, 87.982177734375, 85.44438171386719, 9.226375579833984, 95.5013427734375, -1.56915283203125, 18.82379150390625, 49.978065490722656, 75.56620025634766, 7.735967636108398, 85.23663330078125, -30.883522033691406, 96.98638916015625, 45.54808044433594, 113.47755432128906, 26.823104858398438, 60.7437744140625, -43.1790771484375, 77.50660705566406, 82.28419494628906, 84.38339233398438, 15.971847534179688, 20.705520629882812, -61.312347412109375, -9.21136474609375, 87.400390625, 126.90103149414062, 77.52392578125, 85.85939025878906, -13.030044555664062, -0.3455352783203125, 36.57069396972656, -2.7843017578125, 121.87625122070312, 40.23028564453125, 5.8351593017578125, -25.07501220703125, 4.962682723999023, 103.82443237304688, 117.12667846679688, 94.42899322509766, 16.35894775390625, 32.255584716796875, 113.86691284179688, -41.133514404296875, -59.84284973144531, 80.3275146484375, -20.9246826171875, 44.20318603515625, 89.49188232421875, 100.42364501953125, -24.9644775390625, 20.70879364013672, 39.56982421875, 10.991241455078125, 11.31805419921875, 52.57124328613281, -15.554290771484375, -22.38943862915039, -17.861419677734375, -29.333023071289062, 82.576171875, -98.05766296386719, 103.25927734375, -25.53862762451172, 26.111572265625, 85.05572509765625, -3.3658103942871094, 0.0, 91.33824157714844, 122.7301025390625, 56.105010986328125, 39.84393310546875, 102.76478576660156, 24.5045166015625, -0.600830078125, 2.7520103454589844, 127.21807861328125, -25.0904541015625, 57.011749267578125, -4.9931793212890625, 2.0980072021484375, 34.85638427734375, 17.526107788085938, 6.768699645996094, -16.690216064453125, 57.72502136230469, -32.795440673828125, -1.52581787109375, -16.453582763671875, 5.1921234130859375, -30.65264892578125, 60.663238525390625, 32.86602783203125, 105.9139404296875, -73.9885482788086, 23.612869262695312, 107.26596069335938, 88.9029541015625, -41.849693298339844, 91.92396545410156, 117.02288818359375, 14.695098876953125, 2.446392059326172, 2.7276687622070312, 77.734619140625, 108.21865844726562, 0.48577880859375, 74.99688720703125, -10.785026550292969, -16.658905029296875, 36.070770263671875, 17.22003173828125, 58.670562744140625, -67.14266967773438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000429.npy"}
{"epoch": 0.8984293193717278, "step": 430, "batch_size": 128, "mean": 28.82379722595215, "std": 46.954620361328125, "min": -65.76193237304688, "p10": -28.669686889648432, "median": 21.330936431884766, "p90": 93.42246246337889, "max": 133.40646362304688, "pos_frac": 0.734375, "sample": [-31.240234375, -4.2982330322265625, -40.12574768066406, 3.069427490234375, 63.524322509765625, 77.65435791015625, -40.67106628417969, 67.8067626953125, -16.485076904296875, 3.433868408203125, 69.0733642578125, 14.910686492919922, 26.695159912109375, -3.79071044921875, 28.982437133789062, -59.710723876953125, 115.47442626953125, 8.968994140625, 3.668712615966797, 75.08364868164062, 55.97607421875, -39.20025634765625, -14.114822387695312, 77.52108764648438, 14.905029296875, 22.13690185546875, -23.05999755859375, 0.0, 103.29458618164062, 44.4493408203125, 52.71477508544922, -19.564453125, 76.047607421875, 35.13304138183594, 12.970596313476562, 0.0, 17.094146728515625, -3.4228134155273438, 4.4132080078125, 59.71287536621094, -50.882049560546875, 82.33743286132812, 3.571533203125, 58.25762939453125, 107.78970336914062, 53.861480712890625, 50.43194580078125, 10.807106018066406, -16.7620849609375, 133.40646362304688, 44.20112609863281, 6.983154296875, 79.41102600097656, 15.45556640625, 95.853759765625, 102.64280700683594, 2.966033935546875, -24.102294921875, 78.5889892578125, -57.54052734375, 82.06550598144531, 92.38047790527344, 96.53268432617188, 91.2083969116211, 4.7187957763671875, 47.31781005859375, 73.72186279296875, 79.18649291992188, 7.027076721191406, 117.62445068359375, 15.3311767578125, 17.287124633789062, 111.59503173828125, 81.515869140625, -27.568023681640625, -19.72600555419922, 16.302490234375, 21.420135498046875, -1.9897174835205078, 28.04913330078125, 78.0843505859375, 22.7894287109375, -24.662811279296875, 9.775566101074219, -9.283454895019531, 23.700653076171875, -14.9312744140625, 127.66494750976562, 0.0, 3.3099746704101562, 24.155181884765625, 13.737747192382812, 44.502777099609375, 15.191375732421875, 104.6702880859375, 48.958580017089844, 24.962646484375, 56.46173095703125, 1.0826759338378906, -61.55885314941406, 108.22981262207031, 105.14398193359375, 47.682472229003906, -26.565643310546875, -39.803466796875, 51.85308837890625, 73.6366195678711, 21.241737365722656, 88.4224853515625, 43.28448486328125, -63.9949951171875, 2.9747314453125, 49.249298095703125, -14.057785034179688, 29.509872436523438, 69.92828369140625, 5.131256103515625, -32.821990966796875, 9.949935913085938, -65.76193237304688, -11.42071533203125, -60.846527099609375, 43.063720703125, 64.43163299560547, 51.90801239013672, 14.7945556640625, 2.262603759765625, 83.09786987304688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000430.npy"}
{"epoch": 0.900523560209424, "step": 431, "batch_size": 128, "mean": 28.72683334350586, "std": 45.781742095947266, "min": -81.32091522216797, "p10": -23.222122955322263, "median": 22.432235717773438, "p90": 89.85065612792968, "max": 141.4337158203125, "pos_frac": 0.703125, "sample": [-8.415817260742188, -67.63104248046875, 42.629913330078125, 72.19993591308594, 64.59182739257812, 47.427459716796875, 6.6544952392578125, 59.33799743652344, 14.425262451171875, -25.651214599609375, 0.54632568359375, 27.908447265625, 13.128814697265625, 25.785247802734375, -36.60321044921875, 70.60147094726562, 1.1597900390625, 71.55108642578125, 44.99125671386719, 12.648006439208984, -7.90728759765625, 47.773651123046875, 28.958343505859375, -39.92310333251953, 39.93792724609375, 79.26332092285156, 45.018829345703125, -30.276397705078125, 87.42094421386719, 124.56048583984375, 64.50141906738281, 47.3375244140625, 16.684844970703125, -41.384910583496094, -30.981277465820312, 78.13778686523438, 52.868019104003906, -10.097076416015625, 0.05680084228515625, 60.403289794921875, -12.222957611083984, -66.399658203125, 11.658205032348633, 4.69403076171875, 75.15780639648438, 58.159942626953125, 104.97789001464844, -12.418601989746094, -13.74671745300293, -81.32091522216797, 6.355817794799805, 26.29119873046875, -5.928916931152344, 77.5552978515625, 0.47418212890625, 73.45599365234375, 33.6900634765625, 88.631103515625, 89.28732299804688, -17.510726928710938, -0.123291015625, 23.459209442138672, 50.06394958496094, 23.177947998046875, 13.067474365234375, 29.21392822265625, 46.5565185546875, 93.3974609375, -15.158638000488281, -15.3446044921875, 94.19769287109375, -22.18108367919922, 0.8660888671875, 108.54595947265625, -3.249725341796875, 141.4337158203125, -3.123870849609375, 104.55343627929688, 37.35408020019531, 38.56526184082031, 106.5059814453125, 136.72860717773438, 22.57342529296875, -29.023529052734375, -1.139373779296875, -8.6397705078125, 56.093719482421875, 25.598342895507812, -31.626129150390625, 19.141342163085938, 122.53485107421875, 71.38931274414062, 91.16510009765625, -20.7259521484375, 1.7652206420898438, 133.7694091796875, -9.274200439453125, 68.5201416015625, 1.8457489013671875, 21.28057861328125, 23.373611450195312, 47.294891357421875, 2.449188232421875, 7.9742431640625, 22.291046142578125, -0.1616668701171875, 81.19108581542969, 41.5318603515625, 64.9427719116211, -5.7555694580078125, 14.146011352539062, -7.2761077880859375, 47.43595886230469, 85.39959716796875, 21.384063720703125, 59.52345275878906, -2.75701904296875, -6.31195068359375, -47.26201248168945, 44.068023681640625, -0.9789047241210938, 3.10101318359375, 9.466598510742188, -4.25799560546875, -42.283203125, 118.93812561035156, 65.004150390625, 20.329742431640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000431.npy"}
{"epoch": 0.9026178010471204, "step": 432, "batch_size": 128, "mean": 33.26490020751953, "std": 46.90750503540039, "min": -83.08810424804688, "p10": -17.190899658203126, "median": 29.317562103271484, "p90": 96.89746551513672, "max": 141.21795654296875, "pos_frac": 0.7109375, "sample": [-12.04107666015625, 91.0009765625, -0.0882415771484375, 79.5846176147461, 128.04534912109375, 96.24945068359375, 48.51725769042969, 50.820281982421875, -40.648681640625, -70.80401611328125, 110.314208984375, -28.577255249023438, 85.3343505859375, 32.01605224609375, 26.895248413085938, 13.238716125488281, 26.25750732421875, 48.29930114746094, -6.1938323974609375, 1.658447265625, 46.86540222167969, -71.56320190429688, 14.472686767578125, 1.3924942016601562, 16.5423583984375, 6.036251068115234, -16.23193359375, 83.19276428222656, 31.59354019165039, -5.379974365234375, 18.51715087890625, 91.59162902832031, 121.07637023925781, 93.67425537109375, -11.936553955078125, -5.4462890625, 46.19175720214844, -28.216644287109375, 101.0908203125, 23.936859130859375, 68.2609634399414, 63.669647216796875, 44.918365478515625, 1.980926513671875, 4.896720886230469, -7.2640228271484375, 35.886474609375, 29.09117889404297, -83.08810424804688, -6.2535400390625, 59.853790283203125, 16.518264770507812, 8.151237487792969, 102.64749145507812, -12.937789916992188, 8.157363891601562, -39.15745544433594, 82.92984771728516, -60.48490524291992, 22.534934997558594, -9.48150634765625, 13.7445068359375, 67.44625854492188, -16.5628662109375, 59.74440002441406, 3.1640472412109375, 107.32778930664062, 70.74943542480469, 50.26289367675781, 67.61404418945312, -19.570335388183594, 30.96337890625, 6.82635498046875, -4.12823486328125, 18.00194549560547, 111.46920776367188, 76.24798583984375, 124.65972900390625, -1.8022003173828125, 83.06283569335938, 38.60894775390625, -20.13127899169922, 141.21795654296875, -10.617652893066406, 39.992462158203125, 46.3109130859375, 98.40950012207031, 71.8405990600586, -23.0313720703125, 73.76254272460938, 6.0194854736328125, -10.016399383544922, 0.0, 70.02545166015625, 93.94589233398438, -16.900421142578125, 7.490814208984375, 11.182373046875, -2.520263671875, 33.8638916015625, 29.5439453125, -5.337955474853516, -15.536588668823242, 72.30572509765625, 56.416282653808594, 41.7352294921875, 119.26119995117188, 82.06423950195312, 125.00927734375, 91.54899597167969, -9.26226806640625, 69.81845092773438, 29.564422607421875, 113.10296630859375, -2.047178268432617, 18.1573486328125, 58.327301025390625, 50.09722900390625, 8.066572189331055, 57.82476806640625, 45.62908935546875, 78.16229248046875, -17.868682861328125, 4.695075988769531, 57.65324401855469, 55.56184387207031, -0.2786979675292969, -23.09002685546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000432.npy"}
{"epoch": 0.9047120418848168, "step": 433, "batch_size": 128, "mean": 28.07382583618164, "std": 45.074363708496094, "min": -107.17791748046875, "p10": -22.91815490722656, "median": 30.441009521484375, "p90": 77.76854629516602, "max": 128.96884155273438, "pos_frac": 0.78125, "sample": [68.42897033691406, -84.44967651367188, -107.17791748046875, 78.99066162109375, 21.451385498046875, 14.500396728515625, 49.90106201171875, -8.8753662109375, -17.093244552612305, 34.2578125, 21.100311279296875, 6.9459228515625, 72.19091033935547, 47.487762451171875, 35.1756591796875, 13.677032470703125, -22.160308837890625, 71.23297882080078, 51.5426025390625, 13.460792541503906, 36.45489501953125, 3.33489990234375, 73.5189208984375, -15.448089599609375, -0.8389816284179688, 16.475051879882812, 77.50802612304688, 6.406261444091797, 112.44903564453125, 53.9432373046875, 80.1028060913086, -8.380332946777344, 69.65811157226562, 38.52825927734375, 50.35137939453125, 118.35528564453125, 70.23741149902344, 71.83126068115234, -63.01631164550781, 73.5010986328125, -8.775993347167969, 33.64886474609375, 23.7142333984375, 62.65495300292969, 17.68829345703125, 82.15084838867188, 43.0714111328125, 3.2133140563964844, 18.6910400390625, 72.00831604003906, 46.61217498779297, 34.20698928833008, 67.90789031982422, -9.4725341796875, 6.374603271484375, -49.13623046875, 27.324554443359375, 74.90049743652344, 13.814212799072266, 4.4188385009765625, -9.673187255859375, 20.043685913085938, -73.26602935791016, 35.70603942871094, 98.52783203125, 39.59124755859375, 54.950782775878906, -89.6640625, 34.628326416015625, 65.97123718261719, -7.6881103515625, 17.311691284179688, 75.38999938964844, -3.7472305297851562, 5.6539306640625, 18.980514526367188, 61.77648162841797, 68.4764633178711, 76.49111938476562, -44.623748779296875, -42.538055419921875, 88.14035034179688, 43.46333312988281, -37.129425048828125, 23.0565185546875, 0.863800048828125, 78.37642669677734, 6.655071258544922, 43.601715087890625, 29.7431640625, 67.21376037597656, 0.37606048583984375, -94.96817016601562, 10.809864044189453, 17.448150634765625, 35.045841217041016, 65.87969970703125, 73.61921691894531, 55.191131591796875, -35.67132568359375, 49.666046142578125, 11.8140869140625, 2.1845169067382812, 128.96884155273438, -14.0611572265625, 37.812469482421875, 87.9154052734375, 60.008575439453125, 78.48286437988281, 1.3283424377441406, 66.229248046875, 23.90008544921875, 55.66114807128906, -17.439437866210938, 28.1568603515625, 50.374420166015625, 30.449462890625, -4.469928741455078, 63.86578369140625, 113.37283325195312, 30.43255615234375, 51.694580078125, -73.90446472167969, 17.272220611572266, -10.420219421386719, -24.68646240234375, 80.81752014160156, 3.4330787658691406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000433.npy"}
{"epoch": 0.9068062827225131, "step": 434, "batch_size": 128, "mean": 30.323331832885742, "std": 46.41093826293945, "min": -99.48983764648438, "p10": -21.682143402099605, "median": 23.957672119140625, "p90": 90.43983764648438, "max": 144.93096923828125, "pos_frac": 0.7265625, "sample": [16.803070068359375, 3.6813201904296875, -6.801513671875, 3.29827880859375, 87.23135375976562, 11.5567626953125, 36.636146545410156, 56.272705078125, -27.704437255859375, 135.48199462890625, -12.387941360473633, 86.25965881347656, -23.3514404296875, 11.551971435546875, -23.34302520751953, 6.752044677734375, -20.9703369140625, 65.28947448730469, 28.069442749023438, -7.706512451171875, 15.69671630859375, 138.8380126953125, -3.9713134765625, 144.93096923828125, 88.25881958007812, -59.08319091796875, 0.0, -15.923408508300781, -5.873924255371094, 61.663818359375, 104.7811279296875, 53.37884521484375, -17.983062744140625, 77.9776611328125, 36.16082763671875, 79.1827392578125, -8.0228271484375, 23.3934326171875, 19.32379150390625, 75.00885009765625, 35.59330749511719, 41.536712646484375, 92.19801330566406, -26.43927001953125, 18.437393188476562, -42.27808380126953, 64.53353881835938, 66.33554077148438, 21.1412353515625, 49.880462646484375, 90.164794921875, 71.7005615234375, 60.90447998046875, 134.23779296875, 13.419723510742188, 37.4580078125, 81.03447723388672, 14.9356689453125, -1.065093994140625, 80.890869140625, 16.64508056640625, 35.4703369140625, 127.3992919921875, -18.53805160522461, 5.9119873046875, 36.77629089355469, 65.36538696289062, 91.08160400390625, 9.5833740234375, 31.681068420410156, 102.52023315429688, 27.433837890625, -13.389144897460938, 99.75054931640625, -45.33984375, 10.22085189819336, -24.8455810546875, 101.79158020019531, 8.9803466796875, 56.9329833984375, 89.76287841796875, -26.650115966796875, 42.74444580078125, 30.958343505859375, 23.254669189453125, 48.75787353515625, 62.12944030761719, -91.64666748046875, 48.78276062011719, -9.093841552734375, 48.74864196777344, 10.485321044921875, -58.40394592285156, 0.0, 49.123992919921875, 20.148948669433594, 10.841705322265625, 98.134521484375, 97.13040161132812, 2.4294166564941406, 2.537261962890625, -0.3082122802734375, -11.570901870727539, 59.457733154296875, -5.8187713623046875, 9.221885681152344, -7.318399429321289, -9.034866333007812, 72.731689453125, 38.59718322753906, 34.451080322265625, 5.33845329284668, 0.0, 67.850341796875, 29.211669921875, 0.891815185546875, 32.72795104980469, 52.94696044921875, 22.580047607421875, 42.2635498046875, -99.48983764648438, 13.918212890625, 24.52191162109375, -7.009429931640625, -56.010772705078125, 88.27185821533203, 38.57781219482422, 79.8022689819336], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000434.npy"}
{"epoch": 0.9089005235602095, "step": 435, "batch_size": 128, "mean": 28.572433471679688, "std": 43.86581802368164, "min": -94.15744018554688, "p10": -22.56511840820312, "median": 26.473461151123047, "p90": 85.65127258300781, "max": 125.73284912109375, "pos_frac": 0.7578125, "sample": [59.697845458984375, 7.55523681640625, 7.040103912353516, 25.25501251220703, 30.293258666992188, 102.66896057128906, 81.04780578613281, -60.09498596191406, 69.92478942871094, -2.004180908203125, -3.8011207580566406, -11.743331909179688, -94.15744018554688, 33.649070739746094, 5.75593376159668, -9.396942138671875, 49.98455810546875, 54.029998779296875, 85.89401245117188, 46.815216064453125, 10.1875, 10.958595275878906, 75.35527038574219, 0.0, 3.332212448120117, -5.601837158203125, 34.2265625, 14.364017486572266, 69.41921997070312, 32.69313049316406, 42.30442810058594, 82.085693359375, 51.88429260253906, -11.316631317138672, -11.886970520019531, 115.61306762695312, 41.318328857421875, 57.56512451171875, 30.937255859375, 10.664459228515625, 94.5633544921875, 97.82943725585938, 94.6314697265625, 27.691909790039062, 74.15960693359375, 16.667755126953125, 84.6846923828125, -21.14764404296875, 72.64329528808594, 7.34429931640625, 85.5472412109375, 51.51531982421875, 80.51922607421875, 69.8973388671875, 13.723638534545898, 3.8726043701171875, 52.792633056640625, 20.134658813476562, 40.875457763671875, -44.66704559326172, 9.282669067382812, 103.30645751953125, 6.285526275634766, -84.58879089355469, 7.6730804443359375, 13.26739501953125, 32.07659912109375, 106.667724609375, -59.37482452392578, -35.31195831298828, 17.38909912109375, -0.4862060546875, 4.3936767578125, 100.15280151367188, 102.45683288574219, 11.441925048828125, 73.81259155273438, 83.45379638671875, -41.901649475097656, 43.10923767089844, -6.325286865234375, 70.86264038085938, 36.75653076171875, 58.17927551269531, -57.7196044921875, -2.174468994140625, 58.2467041015625, 3.735065460205078, 0.9830322265625, -5.122344970703125, -9.501180648803711, 73.11627197265625, 54.771484375, 125.73284912109375, 58.57664489746094, 89.92208862304688, 39.522613525390625, 50.987030029296875, -35.948936462402344, 44.47607421875, -4.252349853515625, 34.5284423828125, 30.08819580078125, 5.422075271606445, 25.17144775390625, 14.460418701171875, 23.901519775390625, 9.273880004882812, 55.307403564453125, 5.57086181640625, -6.681365966796875, -26.632055282592773, -56.177764892578125, -6.158439636230469, -2.993621826171875, 114.9403076171875, 5.459320068359375, 9.703033447265625, 33.744598388671875, -56.959991455078125, 39.623870849609375, 59.166046142578125, 45.93098449707031, 0.0400543212890625, -25.87255859375, 64.37965393066406, 21.91180419921875, 46.396514892578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000435.npy"}
{"epoch": 0.9109947643979057, "step": 436, "batch_size": 128, "mean": 22.02836799621582, "std": 44.709449768066406, "min": -84.29501342773438, "p10": -30.128727722167966, "median": 13.106868743896484, "p90": 86.11027221679687, "max": 141.70028686523438, "pos_frac": 0.6640625, "sample": [44.84393310546875, -22.435791015625, 102.9429931640625, 70.89141845703125, 3.613372802734375, 23.289260864257812, 83.97466278076172, 20.507232666015625, -29.413330078125, 82.0885009765625, -19.518829345703125, 38.070709228515625, -43.494659423828125, 33.05792236328125, 30.23333740234375, 102.15008544921875, 25.51483154296875, -7.8235931396484375, -35.584747314453125, -62.8175048828125, -0.7204132080078125, 57.1357421875, 44.639007568359375, 68.4559326171875, 103.43695068359375, -28.47137451171875, 97.18655395507812, -0.70611572265625, 70.80184173583984, 24.982666015625, 34.68232727050781, 12.206787109375, 0.5741100311279297, 18.50360107421875, 78.76481628417969, 29.50220489501953, -1.5557785034179688, 12.644462585449219, 13.56927490234375, -19.711807250976562, 59.420799255371094, 5.26495361328125, 13.648841857910156, 49.735992431640625, 93.61407470703125, 26.6749267578125, -2.486358642578125, 6.591644287109375, 11.504425048828125, 90.28414916992188, 38.847076416015625, -17.44635009765625, 4.99212646484375, -11.97998046875, 54.484405517578125, -12.816856384277344, 94.40951538085938, -68.96380615234375, -2.009063720703125, 88.42120361328125, 80.8428955078125, 19.891891479492188, 10.198745727539062, 28.078243255615234, 38.219635009765625, 76.66024780273438, -17.0478515625, -66.88571166992188, -5.729248046875, 68.80270385742188, -2.2443389892578125, 141.70028686523438, 59.943695068359375, -54.46173095703125, 21.014373779296875, 17.0994873046875, 11.450979232788086, 56.797080993652344, 12.29248046875, -7.65264892578125, 53.740997314453125, -6.63623046875, 15.29376220703125, -4.66326904296875, 40.13702392578125, 85.119873046875, 3.47772216796875, -31.797988891601562, 18.93921661376953, -39.42071533203125, 94.61819458007812, -3.3956375122070312, -2.5690479278564453, -37.09185791015625, 12.47674560546875, 35.177734375, 128.681884765625, 4.762298583984375, 4.935800552368164, -20.70458984375, 0.493927001953125, -84.29501342773438, 9.330131530761719, 44.925048828125, 29.265289306640625, 10.6309814453125, -40.473846435546875, 66.42924499511719, 130.72213745117188, -13.933807373046875, -16.786109924316406, 20.417373657226562, -36.30474853515625, 73.8056640625, -2.654052734375, 70.14994812011719, -1.375335693359375, 88.65155029296875, -15.466766357421875, 10.37762451171875, -14.06195068359375, 34.841278076171875, -57.59623718261719, 4.858001708984375, -22.15545654296875, 76.06817626953125, 29.594207763671875, 5.9183349609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000436.npy"}
{"epoch": 0.9130890052356021, "step": 437, "batch_size": 128, "mean": 35.15369415283203, "std": 46.05380630493164, "min": -75.90396118164062, "p10": -19.282989501953125, "median": 31.95429229736328, "p90": 97.72495727539062, "max": 138.94000244140625, "pos_frac": 0.7265625, "sample": [24.45166015625, 102.73464965820312, 52.9527587890625, 69.2640380859375, -48.817108154296875, 59.86572265625, 5.615386962890625, 8.757461547851562, 63.704345703125, 6.649658203125, 92.17434692382812, 13.504913330078125, 56.20982360839844, 54.107330322265625, -4.3873443603515625, 49.576904296875, -19.880706787109375, -2.3093109130859375, 32.14117431640625, 77.97833251953125, -3.0315399169921875, 97.50625610351562, 108.51229858398438, 90.71987915039062, 0.609344482421875, 18.506103515625, -4.03533935546875, 82.37301635742188, 78.23100280761719, 32.3782958984375, -14.81243896484375, 16.75579833984375, 5.74151611328125, 50.84617614746094, -0.50531005859375, 9.221572875976562, -50.346710205078125, -2.2840728759765625, 7.1184539794921875, 24.27423095703125, 75.23680114746094, 40.2235107421875, 95.00181579589844, 89.42727661132812, -29.00701904296875, -3.17333984375, 23.875732421875, 17.4769287109375, 98.98561096191406, 47.70648193359375, 66.20506286621094, 113.46337890625, 60.69544219970703, 87.57490539550781, 11.210868835449219, 14.151275634765625, -19.026824951171875, -12.12295913696289, 111.07723999023438, 99.09735107421875, 71.0433578491211, 0.48828125, 73.62825012207031, 86.5130615234375, 123.82183837890625, 32.9261474609375, 14.629791259765625, 43.1851806640625, -21.051040649414062, -8.513969421386719, 49.121299743652344, 8.37457275390625, -22.466552734375, 104.21630859375, -75.90396118164062, 91.62774658203125, 9.589947700500488, 55.54798889160156, -44.925994873046875, 0.0, 82.87348937988281, -32.72163391113281, -3.2158660888671875, -1.3161125183105469, 103.004150390625, 67.51483154296875, -35.696136474609375, 27.864013671875, -42.070281982421875, 44.88847351074219, 113.47711181640625, 31.767410278320312, -4.821666717529297, 69.662109375, 10.822998046875, -3.373779296875, 55.78985595703125, 18.355735778808594, 64.52273559570312, 49.06669235229492, 98.23526000976562, 105.72854614257812, 64.51025390625, 86.82373046875, 30.677734375, -12.202682495117188, 34.371429443359375, 50.782081604003906, 93.03396606445312, -56.3870849609375, 1.4823455810546875, -1.816009521484375, 73.34657287597656, -61.99853515625, -9.453033447265625, 87.05669403076172, 8.680816650390625, 84.1842041015625, 62.649749755859375, -0.745880126953125, 39.800811767578125, -17.417144775390625, 21.909759521484375, 62.86236572265625, 0.0, 4.257720947265625, 71.96036529541016, 138.94000244140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000437.npy"}
{"epoch": 0.9151832460732985, "step": 438, "batch_size": 128, "mean": 18.540283203125, "std": 44.936824798583984, "min": -81.7149658203125, "p10": -36.99493675231933, "median": 12.260946273803711, "p90": 82.16986694335937, "max": 132.5709228515625, "pos_frac": 0.671875, "sample": [6.781608581542969, -71.18527221679688, 54.917388916015625, 3.3153076171875, -4.655574798583984, 54.959716796875, 32.8760986328125, 20.751434326171875, 0.0, 12.031742095947266, 102.87144470214844, -56.122528076171875, 16.352996826171875, 71.99885559082031, 34.2801513671875, 6.869678497314453, 17.54736328125, 80.73838806152344, 49.88750457763672, -12.040664672851562, -27.1005859375, 13.45660400390625, 43.519378662109375, 2.7124710083007812, -20.195404052734375, -0.6585159301757812, -18.4365234375, 60.058494567871094, 9.6422119140625, 24.88653564453125, 13.423540115356445, 64.6697998046875, 42.430206298828125, 8.269775390625, 22.441650390625, 90.3505859375, -68.43977355957031, -18.37694549560547, 17.183334350585938, -73.44427490234375, -79.66482543945312, -38.37590026855469, 37.5633544921875, 51.66063690185547, -10.946334838867188, 28.918807983398438, 0.8765106201171875, 20.69097900390625, -51.645233154296875, -6.555442810058594, 3.90618896484375, -64.19969177246094, -17.94188690185547, 15.495895385742188, 6.56427001953125, 14.603302001953125, 46.72381591796875, 15.364990234375, 50.678497314453125, 60.639068603515625, 2.0398941040039062, -18.704347610473633, -3.072751998901367, -68.94698333740234, 27.24200439453125, -2.205312728881836, 0.184173583984375, 20.859298706054688, 70.73931884765625, -57.50312805175781, 85.56399536132812, -36.40309524536133, 86.8499755859375, -5.808982849121094, 29.12249755859375, 53.52215576171875, 58.597503662109375, -38.81398010253906, 76.10861206054688, 81.88931274414062, -34.035400390625, 82.82449340820312, -29.870529174804688, 1.448455810546875, 110.62225341796875, -29.760635375976562, 68.2093505859375, 77.20205688476562, -81.7149658203125, 12.490150451660156, 48.011131286621094, 7.463966369628906, 4.42164421081543, 30.7196044921875, 89.83694458007812, 76.56396484375, 0.5131759643554688, 86.77223205566406, 126.5277099609375, -12.59130859375, -16.054275512695312, -19.126983642578125, 22.875, 57.43145751953125, 52.47675323486328, -11.34967041015625, 11.622634887695312, 8.06524658203125, -12.6278076171875, 80.32687377929688, 92.86312866210938, 10.487707138061523, -18.883941650390625, 28.272212982177734, 0.0, -1.8146209716796875, 32.07017517089844, 9.030990600585938, -2.5208511352539062, 64.67839050292969, 88.51446533203125, -40.71539306640625, 6.64600944519043, 84.168701171875, 132.5709228515625, -6.421073913574219, 5.431182861328125, 25.299591064453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000438.npy"}
{"epoch": 0.9172774869109948, "step": 439, "batch_size": 128, "mean": 30.14676284790039, "std": 44.91813278198242, "min": -79.32809448242188, "p10": -21.459359741210935, "median": 21.887463569641113, "p90": 88.01937561035156, "max": 121.6890869140625, "pos_frac": 0.7109375, "sample": [-20.42584228515625, -38.192535400390625, -70.22610473632812, 114.37844848632812, -46.0894775390625, 88.34542846679688, 14.645782470703125, 18.224708557128906, 25.217010498046875, 53.306671142578125, -2.6230545043945312, 3.433074951171875, 21.92669105529785, 0.314849853515625, 64.97918701171875, -46.256927490234375, 6.660652160644531, 57.21559143066406, -2.205902099609375, 21.848236083984375, 106.74774169921875, 81.26956939697266, 8.047576904296875, 40.197113037109375, 0.686431884765625, 69.13855743408203, 19.161361694335938, 78.48270416259766, 61.45062255859375, 17.90593910217285, -6.810508728027344, 46.366119384765625, 3.6814918518066406, 112.44500732421875, -22.11456298828125, 57.675933837890625, 121.6890869140625, 40.749359130859375, 121.2154541015625, 15.101119995117188, -19.710906982421875, 13.215900421142578, 79.39358520507812, 80.44316101074219, 8.68841552734375, 76.329833984375, 74.96199035644531, 14.781097412109375, 3.8996429443359375, -12.781768798828125, 3.5257568359375, 12.70233154296875, -32.259246826171875, 12.18560791015625, 63.25830078125, 90.542724609375, 66.92544555664062, -17.62964630126953, 48.65467834472656, 0.0, 14.15081787109375, 0.0, -16.09674072265625, 20.944114685058594, -15.31927490234375, 83.220947265625, 21.9697265625, -15.40362548828125, 10.843942642211914, -27.45745849609375, -2.4696044921875, 87.879638671875, 17.029006958007812, -16.359161376953125, -10.852771759033203, 31.245437622070312, -17.59334945678711, 49.43682098388672, -31.8966064453125, 50.110137939453125, -24.909103393554688, 55.70423889160156, 53.70343017578125, 22.656219482421875, 50.281982421875, 51.059844970703125, -21.178558349609375, -4.3062591552734375, -26.988868713378906, -11.9884033203125, 82.70613098144531, -73.69760131835938, 0.0, 92.78854370117188, 66.09991455078125, 15.510948181152344, 112.317626953125, 78.60848999023438, 29.1470947265625, -8.1746826171875, 36.9683837890625, -24.27440643310547, 53.45066833496094, 31.429733276367188, 39.631072998046875, 5.91021728515625, 59.55499267578125, 45.212852478027344, 72.54185485839844, -7.9549560546875, 86.11459350585938, 58.49945068359375, 33.9095344543457, 86.445068359375, 79.43728637695312, 0.6160545349121094, -79.32809448242188, 101.98341369628906, 67.13836669921875, 101.09375, 84.41317749023438, 74.10594177246094, 7.5972747802734375, -13.827301025390625, 100.76579284667969, 0.0, 98.02481079101562, 81.91363525390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000439.npy"}
{"epoch": 0.9193717277486911, "step": 440, "batch_size": 128, "mean": 24.70290184020996, "std": 47.4720573425293, "min": -103.10107421875, "p10": -33.513866043090815, "median": 22.401676177978516, "p90": 89.02503128051757, "max": 158.472412109375, "pos_frac": 0.6953125, "sample": [-23.24749755859375, 0.0, 26.0362548828125, -70.03924560546875, -13.640531539916992, 41.94317626953125, -52.76995849609375, 61.79844665527344, 19.788467407226562, 39.81267166137695, 25.58135986328125, 3.9267578125, -46.47894287109375, 0.54791259765625, 33.8533935546875, -30.2503662109375, 38.712364196777344, 67.49253845214844, 73.59037017822266, 57.0526123046875, 42.08934020996094, 107.2542724609375, -1.5576629638671875, 9.331634521484375, -30.1754150390625, -3.03155517578125, -103.10107421875, 63.59336853027344, 38.59100341796875, 63.696746826171875, 106.12225341796875, 115.7283935546875, 10.085275650024414, -17.4560546875, -17.37872314453125, 3.42462158203125, 68.875, -79.04400634765625, 0.0, 39.627197265625, -93.8953857421875, 11.314605712890625, 1.0882568359375, 29.152503967285156, 91.93923950195312, 18.97370147705078, -48.5386962890625, -5.408111572265625, 12.045509338378906, 71.55918884277344, 119.11346435546875, -1.3283462524414062, 67.76689147949219, -0.03996849060058594, 27.714263916015625, 103.885986328125, -1.7617931365966797, -61.71577453613281, -36.76055908203125, -0.042022705078125, -15.862701416015625, 2.131103515625, 10.466278076171875, -30.4227294921875, 158.472412109375, 37.32530212402344, 107.14334106445312, 72.7707290649414, 64.03021240234375, 42.1324462890625, -2.086894989013672, 15.803115844726562, 5.9139556884765625, 67.08378601074219, 80.10174560546875, 44.15275573730469, 1.5701141357421875, -35.33218002319336, 26.40850830078125, 11.489669799804688, 13.656723022460938, 29.375778198242188, 69.26085662841797, 28.82379150390625, 92.48133850097656, -48.3233642578125, 14.718353271484375, 97.8787841796875, 39.220123291015625, -25.73853874206543, 108.339111328125, -24.22833251953125, 87.77608489990234, -4.269203186035156, 66.0302734375, -0.008087158203125, 50.07600402832031, 30.645000457763672, 71.72520446777344, 27.947998046875, 82.62091064453125, -1.5667343139648438, 21.233009338378906, 12.823143005371094, 33.156982421875, 55.656005859375, 5.7546539306640625, 95.30255126953125, 33.51939392089844, 47.05157470703125, 23.570343017578125, 34.13151550292969, 15.41375732421875, 78.3929672241211, -30.736358642578125, 79.90365600585938, 44.69435119628906, 15.466705322265625, 13.894927978515625, 44.906494140625, -32.734588623046875, 107.21536254882812, 26.246139526367188, -17.049034118652344, -36.311798095703125, 7.864288330078125, -42.408180236816406, 79.8331298828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000440.npy"}
{"epoch": 0.9214659685863874, "step": 441, "batch_size": 128, "mean": 35.387176513671875, "std": 41.448402404785156, "min": -64.84475708007812, "p10": -12.049920654296875, "median": 35.284515380859375, "p90": 92.06560058593749, "max": 126.53021240234375, "pos_frac": 0.7578125, "sample": [-11.89422607421875, 30.4345703125, -64.84475708007812, 66.04289245605469, -0.47129058837890625, -8.275115966796875, 109.71673583984375, -5.52716064453125, 5.50421142578125, -0.76025390625, -40.80247497558594, 126.53021240234375, 39.51073455810547, 86.2684326171875, 21.002395629882812, 72.36480712890625, 61.89984130859375, 83.63458251953125, 52.100914001464844, 62.58888244628906, 44.36968994140625, 68.1539535522461, 13.631086349487305, 14.7469482421875, 67.59149169921875, 27.270523071289062, 52.72412109375, 97.22941589355469, 57.00933837890625, -5.2745361328125, 23.69805908203125, 58.7061767578125, 43.31611633300781, -7.821468353271484, 75.16350555419922, 73.12835693359375, 14.8302001953125, 96.67828369140625, 66.76296997070312, -22.07721710205078, 82.72969055175781, 60.91712951660156, -12.958526611328125, -5.68408203125, 58.13226318359375, -1.2221088409423828, 34.140960693359375, 27.031417846679688, 43.814422607421875, 81.249267578125, 91.00141906738281, 8.2816162109375, 35.51824951171875, 7.614358901977539, 13.91537094116211, 9.4207763671875, 2.354583740234375, 74.45065307617188, 4.9701385498046875, -21.199615478515625, 6.5830078125, 19.612991333007812, 2.7971229553222656, 80.51791381835938, -2.448944091796875, 0.0, 4.6275787353515625, 41.070648193359375, 45.65458679199219, 81.954345703125, -23.37255859375, -2.523387908935547, -5.230566024780273, 31.938064575195312, 0.0, 61.827667236328125, 42.64593505859375, 38.384246826171875, 71.698486328125, 36.574981689453125, 5.564292907714844, 90.26495361328125, 31.017471313476562, -47.22503662109375, 35.05078125, 55.65333557128906, 73.16688537597656, 44.17449951171875, -4.5256500244140625, 15.84674072265625, 1.393341064453125, 19.5987548828125, 35.009368896484375, 50.222747802734375, 52.47600555419922, 10.36773681640625, 102.72508239746094, -35.47595977783203, 106.08056640625, 37.908935546875, 100.15533447265625, -42.18800354003906, 51.735015869140625, 9.975677490234375, -53.195281982421875, -12.4132080078125, 103.78512573242188, -55.06021499633789, 72.46659851074219, -0.51171875, 18.238235473632812, 97.142578125, 95.72836303710938, 94.54869079589844, 110.44027709960938, 65.89556884765625, 46.22804260253906, -4.651458740234375, -20.284637451171875, 7.0709228515625, 98.45257568359375, -2.1639785766601562, 69.12703704833984, 79.58877563476562, 81.7364501953125, 65.17707824707031, 70.74224090576172, 30.84637451171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000441.npy"}
{"epoch": 0.9235602094240838, "step": 442, "batch_size": 128, "mean": 27.289581298828125, "std": 50.58068084716797, "min": -106.31951904296875, "p10": -36.524697875976564, "median": 20.91680145263672, "p90": 90.3112968444824, "max": 159.88360595703125, "pos_frac": 0.7265625, "sample": [76.028076171875, 12.566619873046875, 16.380416870117188, 16.947650909423828, 126.0765380859375, 87.21041870117188, 3.710512161254883, 22.970504760742188, 97.45928955078125, 3.433135986328125, 46.16658020019531, 16.433578491210938, -25.4930419921875, -5.3160552978515625, 58.1593017578125, -76.65463256835938, 16.308670043945312, 17.5738525390625, 13.28082275390625, -52.15923309326172, 13.472015380859375, 35.93705749511719, -20.598220825195312, -3.534421920776367, 63.57135009765625, -12.137542724609375, 51.614173889160156, -15.070281982421875, 111.56376647949219, 102.92866516113281, -68.45309448242188, -1.370147705078125, -33.75372314453125, 82.36724853515625, 40.522308349609375, 78.18359375, 25.065933227539062, 0.9251708984375, -57.94317626953125, 3.7470550537109375, 13.045425415039062, -35.32475280761719, 38.45204162597656, 3.4796371459960938, -29.20965576171875, -65.00552368164062, 20.309951782226562, 32.058319091796875, 4.905357360839844, 5.311767578125, 14.332015991210938, -1.998016357421875, 35.37434387207031, 96.98651123046875, 7.213375091552734, 21.523651123046875, 71.60261535644531, 0.0, 38.016632080078125, -31.0543212890625, -33.915435791015625, -1.8062667846679688, -79.0986557006836, 46.072662353515625, -80.80374145507812, 112.49602508544922, 56.0577392578125, -16.195709228515625, 43.021697998046875, 73.4555892944336, 1.5276203155517578, -106.31951904296875, 102.99327087402344, -3.3380584716796875, 63.287567138671875, 51.210540771484375, 117.36724853515625, -15.7666015625, 71.53302001953125, 14.273529052734375, 40.459930419921875, 50.18780517578125, -5.868740081787109, 106.39326477050781, 84.09909057617188, 85.77745056152344, -36.5994873046875, -61.0546875, -8.499237060546875, -36.492645263671875, 30.74542236328125, 48.49420928955078, 9.689537048339844, 73.466064453125, 3.73773193359375, 76.71623992919922, 57.89863586425781, 79.1294174194336, 68.44745635986328, 76.9549331665039, 106.3399658203125, 79.353271484375, 88.55725860595703, 94.404052734375, 12.307281494140625, 159.88360595703125, 17.3538818359375, -12.046745300292969, -37.051788330078125, 105.17776489257812, 76.4365234375, 39.279258728027344, 24.713592529296875, 70.36912536621094, 5.3740081787109375, 17.312530517578125, 7.3922119140625, 16.27874755859375, 68.98539733886719, -45.0712890625, 35.53953552246094, 71.84911346435547, 84.0113296508789, 29.062347412109375, -70.68270874023438, 53.663330078125, 66.96741485595703, 59.43247985839844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000442.npy"}
{"epoch": 0.9256544502617801, "step": 443, "batch_size": 128, "mean": 28.574121475219727, "std": 44.45750045776367, "min": -96.17595672607422, "p10": -17.364417839050294, "median": 23.472877502441406, "p90": 88.62537002563477, "max": 137.22613525390625, "pos_frac": 0.765625, "sample": [18.143692016601562, 12.043914794921875, 69.0015869140625, 10.05059814453125, -22.47418212890625, 45.1834716796875, 112.10368347167969, 0.5154571533203125, 98.93707275390625, -5.555809020996094, 17.795948028564453, 51.906494140625, 4.139671325683594, 66.218994140625, -9.10675048828125, -20.421165466308594, -12.0211181640625, 28.701126098632812, 84.74324035644531, -1.3269691467285156, 26.584365844726562, 72.56341552734375, 43.41337585449219, 20.265411376953125, -15.3504638671875, 28.087440490722656, 52.2059326171875, 83.17713928222656, 25.376846313476562, 2.1719284057617188, -8.211669921875, 39.9619140625, -64.34902954101562, 6.248424530029297, -10.42333984375, 29.126983642578125, 67.66632080078125, 26.37326431274414, 87.97757720947266, 11.41439437866211, 110.20904541015625, -15.383758544921875, -6.049468994140625, 3.091156005859375, 13.333999633789062, 26.7998046875, 137.22613525390625, 15.253890991210938, 28.071762084960938, -25.120391845703125, 49.60009765625, -96.17595672607422, 2.169158935546875, -2.906005859375, 52.33953857421875, 66.50841522216797, 58.127410888671875, 125.96633911132812, 12.057235717773438, 94.97982788085938, 69.14895629882812, 29.203369140625, -2.4581680297851562, 30.7313232421875, 48.512725830078125, 11.145668029785156, -2.1812591552734375, 36.800933837890625, 68.67826843261719, 85.16976928710938, -6.9994354248046875, 21.56890869140625, 93.054931640625, 9.231731414794922, 86.30320739746094, 5.703765869140625, 3.0505447387695312, 74.39544677734375, -50.797943115234375, 8.782745361328125, -17.271827697753906, 91.03228759765625, 76.473876953125, 46.427581787109375, 32.25701904296875, -44.71045684814453, 16.455101013183594, -17.580461502075195, 68.37503051757812, 129.989013671875, 94.695068359375, 64.18310546875, 13.618377685546875, 42.81364440917969, -20.301773071289062, 69.09378051757812, 9.688796997070312, 116.17202758789062, 72.7537841796875, 51.568634033203125, 45.268310546875, -1.7964324951171875, -87.21627807617188, 66.92344665527344, 29.978240966796875, 3.5532989501953125, 27.497146606445312, 38.980926513671875, 45.406646728515625, 8.51385498046875, 33.640586853027344, 4.928680419921875, -25.119232177734375, 1.3228073120117188, -8.958755493164062, -6.961906433105469, 14.0814208984375, 97.84733581542969, 14.875518798828125, -82.97967529296875, 70.18984985351562, 68.53207397460938, 90.13688659667969, 13.742240905761719, 0.62860107421875, -57.30998229980469, 5.857231140136719, 8.18914794921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000443.npy"}
{"epoch": 0.9277486910994764, "step": 444, "batch_size": 128, "mean": 29.338577270507812, "std": 47.92215347290039, "min": -117.426513671875, "p10": -21.49110107421875, "median": 19.47113609313965, "p90": 94.58628845214844, "max": 153.37521362304688, "pos_frac": 0.6953125, "sample": [-1.0000419616699219, 11.187820434570312, 101.34255981445312, 16.20965576171875, 69.45004272460938, 53.92596435546875, -26.798736572265625, 1.14208984375, 16.624534606933594, 10.942138671875, -8.944412231445312, 17.877655029296875, -21.484527587890625, 25.441925048828125, 67.89104461669922, 31.770263671875, 69.97917175292969, 30.990333557128906, -18.78424072265625, 44.122520446777344, 60.76454162597656, 23.046875, 70.4984130859375, -27.7679443359375, 16.703399658203125, -22.186614990234375, 81.04498291015625, 7.152900695800781, 5.0671234130859375, 12.451995849609375, 11.841567993164062, 81.00272369384766, 64.82710266113281, -5.2056427001953125, 96.44853210449219, -15.213714599609375, -14.158615112304688, 40.83489990234375, 97.15557861328125, 72.39132690429688, 42.354644775390625, 85.13763427734375, -10.9664306640625, -3.445903778076172, 22.877716064453125, 65.93988037109375, 13.21734619140625, -60.548362731933594, -55.100494384765625, -13.2762451171875, 140.105224609375, 44.964263916015625, -27.16912841796875, 126.0675048828125, -3.351419448852539, 57.6224365234375, 153.37521362304688, 99.38436889648438, -6.37445068359375, -3.876495361328125, 1.331197738647461, 107.0960693359375, 79.7148208618164, -16.371597290039062, 40.77882385253906, -21.506439208984375, 99.6583251953125, 59.84294128417969, -25.162994384765625, 117.49267578125, 57.76918029785156, 35.897003173828125, -6.1260986328125, 75.3865966796875, -14.405975341796875, 38.004608154296875, 82.8583984375, 98.43975830078125, 81.48068237304688, 5.163177490234375, 93.83779907226562, -10.425016403198242, 2.862131118774414, 91.7491455078125, -6.567798614501953, 88.17507934570312, -3.5796661376953125, 76.83528137207031, 52.98229217529297, -70.03820037841797, 0.500213623046875, 99.7716064453125, -10.62038803100586, 40.10706329345703, 23.864967346191406, 18.706085205078125, 29.953155517578125, -14.861801147460938, 2.0129852294921875, -33.47784423828125, -7.288116455078125, 5.8120880126953125, 81.73745727539062, 5.592220306396484, -16.566490173339844, 38.5830078125, -5.233316421508789, 77.73858642578125, -24.932785034179688, 76.68757629394531, 3.504291534423828, 30.014991760253906, 6.126434326171875, 58.65851974487305, 43.78593444824219, 16.05553436279297, 96.332763671875, 76.6392822265625, 68.0405044555664, 69.6412124633789, -2.843759536743164, 20.236186981201172, -15.1644287109375, 0.9626731872558594, -100.42727661132812, 17.855712890625, 66.56497192382812, -117.426513671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000444.npy"}
{"epoch": 0.9298429319371728, "step": 445, "batch_size": 128, "mean": 25.653188705444336, "std": 46.604122161865234, "min": -88.19560241699219, "p10": -28.610110473632812, "median": 17.93547821044922, "p90": 92.23882446289062, "max": 138.85397338867188, "pos_frac": 0.71875, "sample": [17.430931091308594, -88.19560241699219, 103.46585083007812, 42.67918395996094, 34.86573791503906, 93.17782592773438, -2.1315689086914062, 16.724647521972656, 100.4063949584961, -39.9132080078125, 5.9370880126953125, 53.17655944824219, 82.42800903320312, -45.58782958984375, -23.9566650390625, 5.073699951171875, 33.053314208984375, 17.463272094726562, 108.78509521484375, 31.991195678710938, 99.05679321289062, 2.00128173828125, 16.16881561279297, -46.096012115478516, 19.922943115234375, 54.69505310058594, 13.878692626953125, -1.4264183044433594, 47.83940124511719, 17.90221405029297, 110.15594482421875, 60.610931396484375, 30.70751953125, 2.3993148803710938, -11.49786376953125, -6.9605560302734375, 18.36174774169922, 14.481170654296875, 62.711326599121094, 54.47247314453125, 7.0799560546875, 94.44595336914062, 100.12155151367188, -0.572906494140625, 67.38253021240234, 4.161806106567383, 8.261001586914062, 26.556884765625, 34.50920867919922, -66.61611938476562, -27.331260681152344, 25.694854736328125, 69.198974609375, 6.149810791015625, 70.0219497680664, -0.155487060546875, 33.677093505859375, 83.28915405273438, 9.510223388671875, 80.69386291503906, 37.95091247558594, 23.063140869140625, 138.85397338867188, 35.415008544921875, 66.57394409179688, 59.456878662109375, 57.396331787109375, -72.08924865722656, 33.74114990234375, 116.5426025390625, 44.62554931640625, 65.011962890625, -16.67694091796875, 77.13461303710938, 67.02365112304688, -1.8008270263671875, -6.059173583984375, -71.81381225585938, 88.94253540039062, 35.11125946044922, -3.799346923828125, -13.49298095703125, -22.09185791015625, -61.590087890625, 7.116697311401367, 29.208751678466797, 16.222381591796875, -43.78802490234375, 40.07861328125, 15.819725036621094, 21.533538818359375, 57.21990966796875, 119.19534301757812, 79.29776000976562, 37.700164794921875, 42.11431884765625, -2.5725231170654297, -75.69743347167969, 105.27745819091797, -5.95416259765625, 5.043876647949219, 31.347579956054688, 88.31344604492188, 17.96874237060547, -7.5814208984375, 14.915237426757812, 8.622734069824219, 14.544837951660156, 98.3052978515625, 8.690216064453125, -27.93072509765625, 6.8616790771484375, 0.0, 6.660163879394531, 8.679222106933594, -28.121246337890625, 2.7851409912109375, -4.45318603515625, 32.90342712402344, 91.83639526367188, -31.47784423828125, -29.75079345703125, -78.066162109375, -2.444110870361328, 60.68988037109375, 74.52534484863281, -19.381431579589844, 87.58204650878906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000445.npy"}
{"epoch": 0.9319371727748691, "step": 446, "batch_size": 128, "mean": 33.65616226196289, "std": 47.280269622802734, "min": -106.12322998046875, "p10": -14.615724182128904, "median": 32.989036560058594, "p90": 93.58300857543945, "max": 140.19140625, "pos_frac": 0.765625, "sample": [74.93231201171875, 92.29840087890625, 32.209739685058594, 7.027587890625, 1.06982421875, 30.952354431152344, 88.38967895507812, 24.18944549560547, 53.64924621582031, -5.1098175048828125, 31.277496337890625, -40.63130187988281, 35.17431640625, 28.94463348388672, 13.2677001953125, 34.25393295288086, 116.55850219726562, 19.01091766357422, -12.309791564941406, 2.6458282470703125, 12.169525146484375, 72.507080078125, 93.88665771484375, 54.57037353515625, -5.733015060424805, -53.2354736328125, 78.6802978515625, 6.1458740234375, 3.2011871337890625, -20.919357299804688, 9.195404052734375, 11.712738037109375, -23.98553466796875, 21.061546325683594, -38.716094970703125, 13.252784729003906, 93.45287322998047, -2.0376663208007812, 17.75018310546875, 46.174232482910156, 59.946380615234375, 7.097694396972656, 50.176780700683594, 36.02716064453125, 36.55377197265625, 68.19256591796875, 99.80364990234375, 80.45816040039062, 82.68389129638672, -106.12322998046875, 12.720062255859375, 32.15673828125, 122.45474243164062, 1.4870758056640625, 67.13496398925781, 75.81647491455078, 37.66552734375, 37.63395690917969, -97.01483154296875, 36.84437561035156, 94.4676284790039, -1.9243736267089844, 9.963836669921875, 0.0, 28.135528564453125, 81.9783935546875, 39.01837158203125, 54.35967254638672, 59.763427734375, 83.21240997314453, 3.561553955078125, 74.78047180175781, 62.619659423828125, 55.72674560546875, 55.836395263671875, 17.938446044921875, 91.674072265625, 73.64117431640625, -8.39385986328125, 0.0, 59.73565673828125, -3.413848876953125, 113.08212280273438, -6.046051025390625, 12.357681274414062, -26.593505859375, 54.745208740234375, 114.55343627929688, 40.6121826171875, 39.07525634765625, -17.645187377929688, 140.19140625, 126.28488159179688, 87.41546630859375, 61.544769287109375, 45.85052490234375, 37.82122802734375, 64.07611083984375, -1.6286125183105469, 93.1358642578125, 12.327804565429688, -11.628082275390625, 114.45135498046875, 104.18817138671875, -97.47604370117188, -71.92216491699219, -10.046646118164062, 17.06927490234375, -7.7175445556640625, 4.602294921875, 38.20383834838867, 0.0, -15.989395141601562, 5.797515869140625, 78.23023986816406, -37.030303955078125, 114.40084838867188, 12.388458251953125, 120.52395629882812, 15.756210327148438, 6.416595458984375, 41.7855224609375, -14.027008056640625, -11.320465087890625, 71.833740234375, 33.768333435058594, 73.04766082763672, 50.189697265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000446.npy"}
{"epoch": 0.9340314136125655, "step": 447, "batch_size": 128, "mean": 34.747581481933594, "std": 41.20296096801758, "min": -70.59152221679688, "p10": -12.134544277191162, "median": 31.56694984436035, "p90": 88.81240158081054, "max": 134.58697509765625, "pos_frac": 0.8046875, "sample": [-25.282974243164062, 6.621936798095703, 77.9239501953125, 65.64382934570312, 17.357025146484375, 12.212017059326172, 107.79385375976562, 4.9067535400390625, -27.6578369140625, 75.0673828125, 44.67268371582031, 45.26170349121094, 33.22406005859375, 87.91024780273438, 65.42587280273438, 42.06977462768555, 0.0, 34.49169921875, 97.89083862304688, 9.642627716064453, 0.2349853515625, 34.970794677734375, 77.52529907226562, 14.411407470703125, 24.045623779296875, -2.150634765625, 1.069427490234375, 44.41424560546875, -70.59152221679688, 70.679931640625, -34.48515319824219, 20.56903076171875, 39.8143310546875, 36.030555725097656, -6.481685638427734, 20.630088806152344, 50.879241943359375, 78.8211669921875, 13.765144348144531, 75.34474182128906, 4.155914306640625, 49.113983154296875, 93.16668701171875, 100.43829345703125, 10.146774291992188, 3.3131103515625, 64.73330688476562, 77.74038696289062, 86.04080200195312, 0.0, 100.22088623046875, 2.9603500366210938, 38.4150390625, 76.50450897216797, -13.571121215820312, -30.7333984375, -8.888132095336914, 4.82574462890625, -18.417770385742188, 95.23997497558594, 4.1151885986328125, 22.108192443847656, 14.6588134765625, 8.100616455078125, -37.20625305175781, 97.39312744140625, 45.28813171386719, -30.2646484375, -9.18572998046875, 10.76608657836914, 29.854827880859375, -12.17156982421875, -13.554229736328125, 9.6561279296875, 39.756591796875, 0.4372596740722656, 31.171817779541016, -1.9322967529296875, 36.7816162109375, 81.15927124023438, 127.9625244140625, 2.447265625, -2.2667179107666016, 49.146820068359375, -69.9676513671875, 0.0, 9.578887939453125, 89.72989654541016, 11.765396118164062, 62.007568359375, 41.91162109375, 26.170501708984375, 28.37994384765625, 63.10997009277344, 129.44998168945312, 57.7899169921875, 84.56546783447266, 134.58697509765625, 0.0, 88.419189453125, 45.4737548828125, 83.47467803955078, 22.46221923828125, 27.447174072265625, 29.61376953125, 59.12506103515625, 19.90472412109375, -34.110076904296875, 85.92747497558594, 73.63162231445312, 2.0091552734375, 65.62950134277344, 48.7857666015625, 56.35577392578125, 80.6651611328125, 35.328857421875, 1.31280517578125, 67.38841247558594, -6.8884735107421875, -12.11867618560791, 47.306304931640625, 101.894287109375, 70.67483520507812, 31.962081909179688, 29.522369384765625, 102.65518188476562, 12.134918212890625, 38.32354736328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000447.npy"}
{"epoch": 0.9361256544502617, "step": 448, "batch_size": 128, "mean": 30.352706909179688, "std": 43.77871322631836, "min": -77.761962890625, "p10": -17.626060485839844, "median": 25.464771270751953, "p90": 91.61127090454102, "max": 144.790283203125, "pos_frac": 0.78125, "sample": [4.0116424560546875, 6.6572265625, -28.480056762695312, 11.922607421875, 54.1810302734375, 106.2703857421875, -4.933135986328125, -64.78558349609375, 44.02757263183594, 85.780517578125, -5.381378173828125, -23.3175048828125, 25.461273193359375, 29.053070068359375, 3.237060546875, 28.208755493164062, 45.430755615234375, 78.2293701171875, 39.041290283203125, -6.5510101318359375, 38.81861114501953, 99.4849853515625, 66.06611633300781, 5.061977386474609, 24.271446228027344, 16.125823974609375, 53.03450012207031, -13.175724029541016, -19.172607421875, 1.881988525390625, -10.349128723144531, 35.6845703125, 69.29238891601562, -29.742401123046875, 67.3995361328125, -17.32061767578125, 87.11875915527344, 100.48419189453125, 28.802520751953125, 25.84234619140625, 53.475555419921875, 25.46826934814453, 109.95916748046875, 4.2060089111328125, 30.641204833984375, 13.15338134765625, 19.528181076049805, 49.068504333496094, -3.1181640625, 131.058837890625, -2.259796142578125, 3.512176513671875, -5.00750732421875, 61.283538818359375, 5.243080139160156, -13.194061279296875, 36.64764404296875, 44.000885009765625, -32.7369384765625, 80.28659057617188, 73.36531066894531, 47.29376220703125, 19.86474609375, 2.878582000732422, 97.47042083740234, 91.14102172851562, 16.979190826416016, 7.1224212646484375, 81.08584594726562, 110.98681640625, 84.01817321777344, 69.47596740722656, 61.188140869140625, 36.020904541015625, 9.218185424804688, 38.058837890625, 104.61611938476562, 81.85005187988281, 78.20623016357422, 29.672454833984375, -18.338760375976562, 97.94184112548828, -8.046772003173828, 93.26881408691406, 4.0608673095703125, -63.104461669921875, -59.543365478515625, -77.761962890625, 71.92132568359375, 37.1756591796875, -6.5157470703125, -5.0291748046875, 88.05551147460938, 87.80616760253906, 21.301620483398438, -65.8692626953125, 15.974365234375, -2.956104278564453, 28.3843994140625, 60.86531066894531, 80.80702209472656, 11.400583267211914, 98.65138244628906, 3.764759063720703, 40.61993408203125, -21.21624755859375, -43.33234405517578, 50.62568664550781, 26.40728759765625, 9.197067260742188, 87.28315734863281, 2.441619873046875, 9.6033935546875, 12.6959228515625, -11.030534744262695, 15.4554443359375, 6.67584228515625, 4.753814697265625, 7.2123565673828125, 0.23074722290039062, 51.861412048339844, 5.3455810546875, 7.1759490966796875, 92.7085189819336, 144.790283203125, 32.946990966796875, 39.77288818359375, 9.304885864257812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000448.npy"}
{"epoch": 0.9382198952879581, "step": 449, "batch_size": 128, "mean": 32.66149139404297, "std": 44.205589294433594, "min": -101.6800537109375, "p10": -17.467943572998042, "median": 30.68659210205078, "p90": 94.33443908691406, "max": 124.20648193359375, "pos_frac": 0.78125, "sample": [35.286468505859375, 26.027008056640625, -47.66047668457031, 76.91292572021484, 44.17730712890625, -6.029052734375, 2.944183349609375, 93.6922607421875, -4.3951263427734375, 6.747173309326172, 17.451309204101562, 15.0064697265625, 30.419326782226562, 37.575523376464844, 1.918731689453125, 75.83914184570312, 105.56646728515625, 97.28912353515625, 13.12115478515625, -7.3227081298828125, -34.501190185546875, 67.17562866210938, 76.52032470703125, 56.87342071533203, 39.72601318359375, 47.2115478515625, -46.186004638671875, 51.23699951171875, 2.7208776473999023, 99.3587646484375, 121.731201171875, 97.66693115234375, 88.45147705078125, 101.0467529296875, 26.4798583984375, 48.711883544921875, 18.070953369140625, 46.968162536621094, 50.5174560546875, 45.136505126953125, 95.83285522460938, 30.953857421875, 72.1121826171875, 9.465476989746094, 119.1429443359375, 115.90652465820312, 4.4190673828125, 2.4447708129882812, 101.1995849609375, 81.93594360351562, -15.789253234863281, 41.0499153137207, 18.20672607421875, 27.092620849609375, 10.430356979370117, 61.6458740234375, 15.374679565429688, 28.318710327148438, 2.920654296875, 45.30625915527344, -11.771080017089844, 33.00445556640625, 74.05308532714844, 14.9197998046875, 41.15380096435547, 6.407135009765625, 23.888824462890625, 38.347076416015625, 3.486328125, 59.928070068359375, -23.989349365234375, 36.75944519042969, 68.41891479492188, 76.25601196289062, -101.6800537109375, -1.3359012603759766, 5.064689636230469, 78.51504516601562, 27.781997680664062, 91.84405517578125, 78.3668212890625, 59.43217468261719, 124.20648193359375, 14.84820556640625, 58.962188720703125, 34.37828063964844, 90.9540786743164, 31.185882568359375, 63.08355712890625, -32.328983306884766, -7.658973693847656, 26.109283447265625, -12.852630615234375, 110.2938232421875, 45.320556640625, 96.8695068359375, -28.720733642578125, -12.773712158203125, 38.422607421875, 49.01690673828125, -26.5999755859375, 70.45158386230469, -24.947303771972656, -13.444534301757812, 30.210769653320312, -3.51263427734375, -29.18707275390625, -7.238609313964844, 70.18548583984375, 13.009292602539062, 6.543327331542969, 62.24818420410156, 0.8923263549804688, -2.7562217712402344, 23.80224609375, -21.3848876953125, 63.9766845703125, 22.631927490234375, 5.865020751953125, 83.7171630859375, 29.113739013671875, -88.60005187988281, 73.38070678710938, 41.96715545654297, 41.40057373046875, -81.03231811523438, -12.9229736328125, -2.72052001953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000449.npy"}
{"epoch": 0.9403141361256544, "step": 450, "batch_size": 128, "mean": 28.359092712402344, "std": 43.44092559814453, "min": -91.80712890625, "p10": -25.99569396972656, "median": 21.919946670532227, "p90": 88.45911178588865, "max": 131.35076904296875, "pos_frac": 0.71875, "sample": [0.0, -27.134490966796875, 66.63304138183594, -16.769935607910156, 39.4302978515625, -2.7353363037109375, 76.57410430908203, -5.67767333984375, 100.8475341796875, 42.650390625, 61.209228515625, 104.54737854003906, -0.7333526611328125, 22.23199462890625, 75.31005859375, 7.454071044921875, -2.193267822265625, 55.86151123046875, 6.649871826171875, 14.300113677978516, -15.909416198730469, 21.330429077148438, 111.68878173828125, 58.85020446777344, 55.097572326660156, 59.79833984375, -13.486968994140625, 16.28411865234375, 40.55192565917969, 15.337890625, 0.7482185363769531, -5.900001525878906, 47.923004150390625, 5.8827362060546875, 39.780616760253906, 9.4322509765625, 131.35076904296875, -61.34576416015625, 25.341293334960938, -8.644744873046875, -3.034944534301758, 54.460662841796875, 42.44833755493164, 68.19378662109375, 19.143661499023438, 2.6767654418945312, -22.627023696899414, 52.09991455078125, 106.98089599609375, 52.47107696533203, 8.957122802734375, 84.9959716796875, 12.929706573486328, 7.620536804199219, 53.90168762207031, -28.195098876953125, 24.599502563476562, 103.55357360839844, 14.119476318359375, -52.305023193359375, 40.564453125, -91.80712890625, 71.03042602539062, 38.54071044921875, 19.3388671875, 100.06805419921875, 81.75053405761719, 2.537689208984375, 42.26153564453125, 40.699676513671875, -3.53533935546875, 68.33804321289062, -1.404205322265625, 84.83990478515625, 2.17926025390625, 21.607898712158203, 71.21231079101562, 96.02996826171875, -49.97782897949219, 15.174942016601562, 58.504638671875, 36.38462829589844, -36.91145324707031, 8.714263916015625, 105.5655517578125, 24.25396728515625, -1.4099502563476562, 11.89898681640625, 17.716636657714844, -7.186187744140625, 53.09185791015625, 38.81558609008789, 93.16717529296875, 118.15628051757812, 23.514907836914062, 70.76348876953125, -26.710540771484375, -47.9188232421875, -28.880050659179688, -25.6893310546875, 13.421466827392578, 44.98307800292969, 84.81732177734375, 56.676055908203125, 100.12252807617188, 32.3748779296875, -58.42601013183594, 26.289581298828125, 5.9302978515625, 92.90663146972656, 78.30485534667969, 15.443252563476562, 20.952590942382812, -13.55859375, 86.55303192138672, -40.88365173339844, 4.68402099609375, -5.5303955078125, 32.0361328125, 55.692413330078125, 0.0, -8.053306579589844, 45.708160400390625, 55.482261657714844, -9.325775146484375, 75.18399047851562, -18.1231689453125, -34.546653747558594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000450.npy"}
{"epoch": 0.9424083769633508, "step": 451, "batch_size": 128, "mean": 31.553586959838867, "std": 44.76492691040039, "min": -90.28654479980469, "p10": -19.01431884765625, "median": 22.881563186645508, "p90": 94.48581237792969, "max": 130.9842529296875, "pos_frac": 0.7734375, "sample": [62.29205322265625, 29.673542022705078, -11.291351318359375, 6.1047821044921875, 114.85017395019531, 84.98703002929688, 28.806076049804688, 2.731058120727539, -85.79736328125, -19.00750732421875, 61.738800048828125, 3.7230796813964844, 100.56373596191406, 18.480010986328125, 40.268798828125, 50.5980224609375, 64.62994384765625, 8.698089599609375, 105.38986206054688, -11.874969482421875, 77.51785278320312, 3.823638916015625, 17.315719604492188, 10.258514404296875, 115.6148681640625, 1.633880615234375, 128.76968383789062, -13.721328735351562, 4.711082458496094, 28.8211669921875, -19.631561279296875, 21.360023498535156, -90.28654479980469, 92.26187133789062, 14.82464599609375, 23.230857849121094, 19.05596160888672, 40.356475830078125, 32.37933349609375, 74.8362808227539, 22.532268524169922, 73.9708251953125, 70.255859375, 77.81832885742188, 41.72357940673828, 92.19459533691406, -3.52593994140625, 27.846755981445312, 112.98907470703125, -2.88330078125, 5.5082855224609375, 105.72235107421875, 7.777381896972656, 119.56137084960938, 64.04489135742188, 75.55313110351562, -12.217788696289062, 43.22119140625, -57.208160400390625, 68.54879760742188, 46.08085632324219, 72.90499877929688, 12.478530883789062, 94.12960815429688, 1.5666961669921875, 20.853790283203125, -49.318443298339844, 0.8164291381835938, 43.51873779296875, 60.59027099609375, 16.215057373046875, 2.736114501953125, -19.01300048828125, 66.73580932617188, 8.79541015625, -35.844696044921875, 35.42097473144531, 4.90472412109375, 70.45565795898438, 23.85882568359375, 6.11138916015625, -3.3289718627929688, -4.058990478515625, 49.60459899902344, 77.33395385742188, -19.430099487304688, 9.515884399414062, 57.56146240234375, 17.30804443359375, 15.581520080566406, 49.28987121582031, 46.72389221191406, -14.4305419921875, 96.3699951171875, 13.650497436523438, -43.044769287109375, 37.775047302246094, -19.09417724609375, -9.655715942382812, 101.31884765625, 56.444366455078125, 95.31695556640625, 54.789581298828125, 90.72079467773438, -19.01739501953125, -24.05718994140625, -45.77978515625, -10.701835632324219, 55.875885009765625, 42.49165344238281, 53.941253662109375, 8.708847045898438, 97.9593505859375, 80.4074935913086, -1.9933547973632812, 16.960756301879883, 69.60125732421875, 3.6263427734375, 20.134506225585938, 130.9842529296875, 18.515480041503906, -4.5538177490234375, -17.858627319335938, 4.348442077636719, 68.26583862304688, 34.667449951171875, -30.32135009765625, 74.26420593261719], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000451.npy"}
{"epoch": 0.9445026178010472, "step": 452, "batch_size": 128, "mean": 31.27404022216797, "std": 43.927669525146484, "min": -81.89712524414062, "p10": -18.245053100585938, "median": 19.139007568359375, "p90": 93.31939544677734, "max": 133.5463104248047, "pos_frac": 0.796875, "sample": [-35.072021484375, 71.54119873046875, 29.74017333984375, 122.336669921875, -20.865310668945312, 8.236763000488281, 9.363914489746094, 92.74951171875, -9.96087646484375, -25.221946716308594, 91.8216552734375, 5.0958099365234375, 9.740310668945312, 20.278541564941406, -1.1812667846679688, 8.494140625, 53.813812255859375, 15.363311767578125, 2.26312255859375, 10.12109375, 0.0866851806640625, -52.17616271972656, 38.060081481933594, 79.0089111328125, 56.1402587890625, 49.865234375, 26.04852294921875, 16.301742553710938, 1.1635360717773438, -33.873931884765625, -24.46905517578125, 97.11233520507812, 0.71234130859375, 111.60855102539062, 46.037506103515625, -15.356292724609375, 99.93875122070312, 102.33364868164062, -18.106231689453125, 73.60504150390625, 43.400489807128906, 10.9759521484375, 7.943462371826172, 8.58990478515625, 71.09661865234375, 8.804056167602539, 25.020103454589844, 70.2260513305664, 41.3831787109375, 16.612403869628906, 109.05859375, 30.7679443359375, 45.85552978515625, -2.748016357421875, 12.310909271240234, -26.91546630859375, 10.438888549804688, 13.45977783203125, 13.65667724609375, 17.135711669921875, 8.3375244140625, 32.288787841796875, -6.026092529296875, 49.55517578125, -42.90435791015625, 81.54623413085938, 58.47172546386719, 84.31390380859375, 50.616668701171875, 10.99542236328125, 11.987380981445312, 80.39239501953125, 40.41986083984375, -15.899810791015625, 59.91874694824219, 131.2568359375, 80.16841125488281, -15.399765014648438, 22.51104736328125, 65.93412780761719, -72.52635955810547, 69.848876953125, 100.26223754882812, 0.49365997314453125, 95.4830322265625, 76.60354614257812, -3.9484176635742188, 75.9027099609375, 84.00592041015625, 28.822113037109375, 42.39430236816406, 31.325927734375, 0.970611572265625, 0.3728179931640625, 17.999473571777344, -32.51678466796875, 80.90701293945312, 33.42347717285156, 47.065216064453125, -4.78521728515625, 129.31039428710938, 37.944976806640625, 10.757110595703125, 75.68341064453125, 33.8917236328125, 5.106325149536133, -18.5689697265625, 12.115501403808594, 87.40325927734375, 94.64912414550781, 133.5463104248047, 98.27688598632812, 4.558685302734375, 29.3553466796875, 86.94924926757812, 0.0, 34.54277038574219, -1.6845169067382812, -34.107696533203125, -16.709671020507812, 17.712249755859375, 42.97314453125, -81.89712524414062, 17.4403076171875, 2.651336669921875, 89.54850769042969, 9.5670166015625, 1.700296401977539], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000452.npy"}
{"epoch": 0.9465968586387434, "step": 453, "batch_size": 128, "mean": 32.23313522338867, "std": 43.181209564208984, "min": -56.224609375, "p10": -28.629656982421874, "median": 26.5689697265625, "p90": 87.13372497558593, "max": 154.4739990234375, "pos_frac": 0.796875, "sample": [14.784072875976562, 75.58695220947266, -33.092132568359375, 34.786224365234375, 85.200439453125, 4.65478515625, 15.323043823242188, 0.8800125122070312, 77.641845703125, 48.46788024902344, 18.70539093017578, 44.60621643066406, 26.339874267578125, 33.1416015625, -41.43682861328125, -24.191558837890625, 39.843048095703125, -53.802978515625, 5.14593505859375, 72.06533813476562, -19.27435302734375, 76.4373779296875, 77.83969116210938, 73.81336975097656, 35.626220703125, 11.187744140625, -36.387786865234375, 87.24658203125, 40.06907653808594, 86.03089904785156, 16.4295654296875, 70.06315612792969, 31.042617797851562, 68.71029663085938, 17.863922119140625, 31.77862548828125, 26.6585693359375, -6.537689208984375, 92.32652282714844, 103.44961547851562, 5.098175048828125, 94.89044189453125, 25.4759521484375, 20.66094970703125, 55.683982849121094, -47.174041748046875, 14.090606689453125, 21.008621215820312, 13.20745849609375, -30.088104248046875, 12.187713623046875, 46.340301513671875, 6.45404052734375, 3.6760635375976562, -56.224609375, 24.03875732421875, 1.615966796875, -32.495025634765625, 44.88029479980469, 26.4793701171875, 62.127471923828125, -14.669288635253906, 0.0, 62.66455078125, 53.66822814941406, 25.020370483398438, 59.496002197265625, 7.217437744140625, 70.07637023925781, 18.562408447265625, 47.964385986328125, 34.18316650390625, 36.343231201171875, -50.837646484375, 10.95458984375, 67.34908294677734, 107.17529296875, 86.93768310546875, -1.578054428100586, 79.82986450195312, -20.795822143554688, 101.8885498046875, 65.65548706054688, -2.932861328125, 22.159515380859375, 14.76849365234375, 111.96234130859375, -46.33709716796875, -21.8331298828125, 105.27685546875, 30.494110107421875, 71.18218994140625, -36.889129638671875, -28.004608154296875, 87.08535766601562, 11.2532958984375, -51.714202880859375, 82.49522399902344, -3.83056640625, 154.4739990234375, 47.88446807861328, 41.853607177734375, 20.153915405273438, 43.11116027832031, 6.772918701171875, 39.25439453125, 2.6957321166992188, 1.9152679443359375, 89.84030151367188, 18.67298126220703, 83.8796615600586, 72.58735656738281, 18.1671142578125, 73.3060302734375, 34.984344482421875, 66.70555114746094, 89.23839569091797, 102.40994262695312, -7.071563720703125, 8.305252075195312, 38.409515380859375, 8.77764892578125, 62.00255584716797, -7.02923583984375, 67.850830078125, -31.77069091796875, 18.511436462402344, 126.74896240234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000453.npy"}
{"epoch": 0.9486910994764398, "step": 454, "batch_size": 128, "mean": 30.42313003540039, "std": 44.26074981689453, "min": -104.50863647460938, "p10": -18.151478576660153, "median": 20.604537963867188, "p90": 93.13351440429688, "max": 142.26483154296875, "pos_frac": 0.796875, "sample": [38.20196533203125, 86.85873413085938, -68.8548583984375, -7.6074981689453125, -2.534515380859375, 10.75091552734375, 107.5291748046875, 68.12884521484375, 9.056732177734375, 30.522323608398438, 27.03723907470703, 35.27526092529297, -20.176790237426758, 66.43229675292969, 27.232986450195312, 61.453857421875, -45.12090301513672, 5.638097763061523, 6.9378204345703125, -2.854402542114258, 26.40185546875, 10.032341003417969, 7.64996337890625, 3.9236297607421875, 130.61041259765625, 142.26483154296875, -5.48468017578125, 98.957763671875, -104.50863647460938, 2.704132080078125, 17.705116271972656, 77.54689025878906, 24.57672882080078, 68.88359069824219, 73.63517761230469, 16.911392211914062, 79.28155517578125, -16.023727416992188, 13.5621337890625, 20.309921264648438, 37.721923828125, 10.359390258789062, 99.04266357421875, 58.693328857421875, 31.220680236816406, 79.838134765625, 71.83758544921875, -4.1363525390625, 100.998291015625, -10.341522216796875, 20.098922729492188, 26.0811767578125, 21.15069580078125, 3.0306396484375, 42.194915771484375, -11.108642578125, 38.139190673828125, 7.5226287841796875, 87.12274169921875, 108.41290283203125, 67.81163024902344, 77.09233093261719, 99.16351318359375, 53.80754089355469, 94.2142333984375, 90.02859497070312, 71.564453125, 11.05963134765625, 22.28369140625, 9.0350341796875, 88.21224975585938, 74.134765625, 20.3212890625, -36.73801803588867, 7.657012939453125, 20.887786865234375, -31.897804260253906, 44.368927001953125, 13.502532958984375, -63.49359130859375, -39.56340026855469, 0.376129150390625, 97.973876953125, -34.53955078125, -11.370941162109375, 29.94952392578125, 79.44420623779297, 42.8218994140625, -17.8004150390625, 92.67034912109375, 0.3137664794921875, 8.353378295898438, 17.7332763671875, 102.71310424804688, 115.85910034179688, 1.3011474609375, 13.606925964355469, 6.54656982421875, 89.70909118652344, 15.415863037109375, -18.970626831054688, 8.7403564453125, 97.13845825195312, 17.22462272644043, 33.95697021484375, 15.815826416015625, 68.72967529296875, 24.113853454589844, -44.9119873046875, 69.78575134277344, 5.511360168457031, 38.92964172363281, 54.25762939453125, -2.8317413330078125, 15.513214111328125, 77.36775970458984, 26.518798828125, 9.227706909179688, -40.14776611328125, 31.521499633789062, 13.916702270507812, -2.1641387939453125, -1.808685302734375, 87.15554809570312, 30.42352294921875, 7.379150390625, 7.69683837890625, -21.18566131591797], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000454.npy"}
{"epoch": 0.9507853403141361, "step": 455, "batch_size": 128, "mean": 25.77764129638672, "std": 44.0270881652832, "min": -95.85443115234375, "p10": -22.430432128906247, "median": 22.552886962890625, "p90": 89.11443786621093, "max": 133.03573608398438, "pos_frac": 0.703125, "sample": [66.26040649414062, 45.35063934326172, 22.9693603515625, -11.42193603515625, 48.998016357421875, 13.9827880859375, 58.42070007324219, 98.59956359863281, 16.092849731445312, -28.82001495361328, 4.3659820556640625, 68.5867919921875, -10.506622314453125, 43.11077880859375, 28.14828872680664, 90.00669860839844, 97.18226623535156, 58.433658599853516, 123.56619262695312, -27.7967529296875, 5.20281982421875, 33.3856201171875, 11.752273559570312, 94.46975708007812, 133.03573608398438, 83.12428283691406, 85.62091827392578, 19.15716552734375, 23.717632293701172, 0.2755126953125, -9.330062866210938, -95.85443115234375, -0.9159603118896484, 10.2618408203125, 14.279983520507812, 30.031768798828125, -10.292915344238281, -25.384765625, -38.27161407470703, 20.98272705078125, 14.446670532226562, -10.123931884765625, 61.13377380371094, 125.0826416015625, 5.83447265625, 64.586181640625, 65.79891204833984, -11.371490478515625, 93.48405456542969, 27.95672607421875, 3.93829345703125, 9.875946044921875, 54.19386291503906, -3.9542312622070312, -6.7105865478515625, -10.41265869140625, 11.372222900390625, -43.642425537109375, -29.0240478515625, -0.6724815368652344, 10.418838500976562, -5.5043792724609375, 7.074834823608398, -23.20648193359375, -22.09783935546875, -13.029390335083008, 43.726287841796875, 39.947784423828125, 71.444091796875, 35.29182434082031, -86.38034057617188, 54.78416442871094, 60.260528564453125, 84.28775024414062, 37.493896484375, -6.462615966796875, -11.3746337890625, 92.63507080078125, 80.01211547851562, 50.75239562988281, 16.382080078125, -5.224269866943359, 7.410133361816406, -48.99272155761719, 88.73204040527344, -19.4708251953125, -73.93583679199219, 28.622055053710938, -6.19024658203125, 17.577136993408203, -19.827423095703125, 48.8240966796875, 18.74969482421875, 23.715614318847656, 40.407470703125, 29.61920166015625, 109.03253173828125, 122.64630126953125, 22.13641357421875, -18.757675170898438, -8.64532470703125, 4.500213623046875, 26.14642333984375, 39.721893310546875, 8.717208862304688, -1.2693748474121094, 1.9901657104492188, 52.82598876953125, 20.5770263671875, 45.17424774169922, -8.527923583984375, 56.650787353515625, 31.433624267578125, 25.414291381835938, 76.51373291015625, -33.23098373413086, -82.14352416992188, 49.621826171875, -4.0079345703125, 49.02601623535156, 72.43267822265625, 43.77783203125, 91.62753295898438, 39.30584716796875, 50.943756103515625, 41.64154052734375, 90.094482421875, 25.126251220703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000455.npy"}
{"epoch": 0.9528795811518325, "step": 456, "batch_size": 128, "mean": 36.60923385620117, "std": 51.32378005981445, "min": -79.93099975585938, "p10": -23.924765014648433, "median": 26.079345703125, "p90": 101.84947967529297, "max": 173.63348388671875, "pos_frac": 0.7265625, "sample": [21.70806884765625, -26.549957275390625, -54.19342041015625, 72.82470703125, 22.480369567871094, -5.469573974609375, 84.78919982910156, -43.642547607421875, 111.800048828125, 58.285308837890625, 85.91766357421875, -79.93099975585938, 17.380783081054688, -0.9840087890625, -9.1148681640625, 101.5142822265625, 61.14384460449219, -29.704360961914062, 2.5184326171875, 13.661865234375, 98.02030944824219, 25.27056884765625, -9.8057861328125, 91.85570526123047, 1.9474620819091797, -16.802978515625, 45.595703125, -5.55450439453125, 76.28350830078125, -7.114959716796875, 68.39132690429688, -43.27581787109375, 75.18028259277344, -22.7996826171875, 29.41607666015625, 100.00299072265625, 67.18939208984375, 75.6922836303711, 49.99688720703125, 60.899505615234375, 137.04443359375, -14.71661376953125, 101.0950927734375, 19.09015655517578, 102.892822265625, 73.87372589111328, 117.52865600585938, -1.22894287109375, 9.363525390625, 86.4549560546875, 31.752685546875, 33.05827331542969, -1.366668701171875, 14.863000869750977, -38.012054443359375, 31.815780639648438, 8.710418701171875, 13.43621826171875, 9.4847412109375, 123.3070068359375, -0.9326171875, -6.240806579589844, 13.634651184082031, -56.29338836669922, 12.161911010742188, 82.330322265625, 83.80180358886719, -70.12324523925781, 26.99713134765625, 17.755462646484375, 9.889726638793945, 61.552093505859375, 0.6203575134277344, 72.86883544921875, 20.952157974243164, 59.728118896484375, 125.60714721679688, 69.84259033203125, -15.114776611328125, 42.206939697265625, 77.25106811523438, 52.4193115234375, -28.899749755859375, 103.093505859375, -10.402923583984375, -12.7474365234375, 21.064865112304688, 91.82675170898438, 92.08856201171875, 106.60972595214844, 77.40765380859375, 114.38827514648438, -4.7191162109375, 97.63015747070312, 0.0, 5.179201126098633, 68.96217346191406, 99.63616943359375, 58.993934631347656, 68.75366973876953, 17.3052978515625, 11.239044189453125, 26.88812255859375, 2.1799468994140625, 26.954845428466797, -31.820587158203125, 78.04379272460938, 2.0385284423828125, 8.74395751953125, -6.8753662109375, 4.209197998046875, 31.99085235595703, 24.880279541015625, 67.98284912109375, 102.63160705566406, -1.1545886993408203, 168.4400634765625, 10.443405151367188, 173.63348388671875, 117.44647216796875, 89.16107177734375, 29.325164794921875, -29.317779541015625, -8.058868408203125, 95.29890441894531, -14.753509521484375, -59.90899658203125, 95.98406982421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000456.npy"}
{"epoch": 0.9549738219895288, "step": 457, "batch_size": 128, "mean": 34.0906867980957, "std": 46.48457336425781, "min": -99.52755737304688, "p10": -25.47540893554687, "median": 31.12439727783203, "p90": 94.61869812011719, "max": 118.38374328613281, "pos_frac": 0.765625, "sample": [95.4002685546875, 19.35174560546875, 4.8492584228515625, 13.333419799804688, 71.69564819335938, 26.271240234375, 83.63218688964844, 82.44198608398438, 92.60958862304688, 9.214401245117188, 89.97579193115234, -0.8857383728027344, 96.67510986328125, 56.252960205078125, 6.059326171875, 97.47366333007812, 12.219757080078125, 19.15521240234375, -59.967063903808594, 105.91763305664062, 59.28228759765625, 86.35104370117188, 39.782745361328125, 118.38374328613281, 76.66360473632812, 67.08580017089844, 95.15029907226562, 3.0838470458984375, 89.1938705444336, 11.9683837890625, 53.33453369140625, -8.323455810546875, -2.560394287109375, -18.63580322265625, 75.36126708984375, 81.1888427734375, -2.59368896484375, 31.031295776367188, -51.81809997558594, 57.3387451171875, 2.56243896484375, 92.31156158447266, -11.427749633789062, 7.94647216796875, 2.662822723388672, -99.52755737304688, 109.08438110351562, 29.161376953125, 111.05797576904297, -9.043792724609375, -28.1929931640625, 47.13250732421875, 50.64646911621094, -39.07763671875, 33.732177734375, 5.50214958190918, -43.40106201171875, 63.1702880859375, 27.336700439453125, 22.6156005859375, -21.4111328125, -9.731742858886719, 7.538665771484375, -21.085411071777344, 39.46076965332031, 64.37937927246094, 70.22758483886719, 13.54559326171875, 106.56147766113281, -4.1920166015625, 2.96832275390625, 76.7999267578125, 75.42630004882812, 29.20849609375, 21.247634887695312, -64.18949890136719, -30.28857421875, 31.217498779296875, 13.132911682128906, 50.82763671875, 18.3546142578125, 86.74432373046875, 67.34823608398438, 53.410186767578125, 32.4005126953125, -5.066741943359375, 51.223899841308594, 8.098747253417969, 47.881805419921875, 106.18035888671875, 7.728803634643555, -24.31072998046875, 94.390869140625, -31.51495361328125, -33.00738525390625, 74.70488739013672, -39.75653076171875, 72.92657470703125, -12.343536376953125, 89.77593994140625, 64.63745880126953, 28.71466064453125, 19.792510986328125, 38.7308349609375, -68.43505859375, 86.34674072265625, 31.60186767578125, 65.65996551513672, 4.103004455566406, 66.29228210449219, 96.77685546875, 57.629676818847656, 13.46380615234375, 88.25962829589844, 58.90675354003906, 75.67912292480469, 5.116912841796875, -0.5712718963623047, -44.467437744140625, 81.94865417480469, 110.55853271484375, 1.1223773956298828, 0.0, 97.060791015625, -7.149341583251953, 72.82923889160156, 44.13212585449219, 0.8563613891601562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000457.npy"}
{"epoch": 0.9570680628272251, "step": 458, "batch_size": 128, "mean": 31.186250686645508, "std": 47.244083404541016, "min": -95.9197998046875, "p10": -18.541262054443358, "median": 17.304153442382812, "p90": 99.51987304687499, "max": 132.82618713378906, "pos_frac": 0.7578125, "sample": [-0.158599853515625, -36.16351318359375, 11.73541259765625, -89.36228942871094, 114.87632751464844, 13.253837585449219, -31.584228515625, 100.06918334960938, 1.4853057861328125, -22.040985107421875, 16.670745849609375, -28.332408905029297, -38.3968505859375, 6.62530517578125, 1.2223405838012695, 65.31072998046875, 23.874267578125, 28.1417236328125, -19.62353515625, 39.876861572265625, 99.28445434570312, 1.4050731658935547, 70.14251708984375, -0.8305130004882812, 19.090789794921875, 4.081022262573242, 59.7457275390625, 101.37625122070312, -10.2047119140625, 60.9093017578125, 46.39373779296875, -18.077430725097656, 64.27017211914062, 129.00039672851562, 69.93670654296875, 11.68939208984375, 13.442108154296875, 22.3861083984375, 88.62503051757812, -23.706710815429688, 71.0880126953125, -1.9212779998779297, -48.75732421875, 29.82867431640625, 7.372222900390625, 6.0238800048828125, 15.4044189453125, 11.934917449951172, 14.62430191040039, 3.6354598999023438, 3.595684051513672, -14.750747680664062, -9.882858276367188, 93.20086669921875, 86.04071044921875, -7.4989013671875, -10.0621337890625, -1.7038192749023438, 58.99903106689453, 124.43890380859375, 25.4677734375, 2.220672607421875, 105.20263671875, 60.4766845703125, 45.969783782958984, 2.97430419921875, 5.11541748046875, -55.863365173339844, -19.916534423828125, 114.86700439453125, 78.15020751953125, 68.3592529296875, -2.5189971923828125, 17.93756103515625, -8.14306640625, 85.36475372314453, 40.47126770019531, -6.744110107421875, 4.054531097412109, 67.43836975097656, 80.12079620361328, -12.595855712890625, -3.5610733032226562, 11.24053955078125, 123.11624145507812, 21.16754150390625, 132.82618713378906, 2.8228759765625, 34.995758056640625, 82.3175048828125, 13.89007568359375, 115.51654052734375, 77.26597595214844, 70.07630920410156, 115.33625793457031, -1.4140090942382812, 68.85958862304688, -0.421875, 7.337673187255859, 60.44444274902344, 38.706329345703125, 10.5057373046875, 5.555023193359375, 88.39053344726562, 55.606658935546875, 16.666648864746094, 5.957664489746094, 35.366424560546875, 90.99382019042969, 83.58030700683594, 24.620193481445312, 112.06227111816406, 86.20652770996094, 44.46038818359375, 71.6116943359375, 24.999755859375, 66.83004760742188, 2.7418975830078125, -95.9197998046875, -76.12188720703125, 66.82417297363281, 107.37789916992188, 15.267097473144531, 4.073638916015625, 59.46049499511719, 25.779144287109375, 5.374359130859375, -13.411842346191406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000458.npy"}
{"epoch": 0.9591623036649215, "step": 459, "batch_size": 128, "mean": 24.349231719970703, "std": 47.392356872558594, "min": -84.3846435546875, "p10": -32.5879150390625, "median": 16.67938995361328, "p90": 87.8427749633789, "max": 126.31716918945312, "pos_frac": 0.6796875, "sample": [4.26666259765625, -9.571334838867188, 6.608728408813477, 72.578857421875, 7.70562744140625, 56.0135498046875, -70.81236267089844, 32.767578125, 43.03814697265625, 24.768218994140625, 64.06466674804688, 55.12826919555664, 20.538158416748047, -11.08148193359375, -0.02983856201171875, 7.381763458251953, 109.63641357421875, 80.49118041992188, -12.521505355834961, 19.24322509765625, 50.60198974609375, 11.6646728515625, -0.00213623046875, 22.071380615234375, -5.3236083984375, -4.322296142578125, 57.57231140136719, 11.740158081054688, 75.83671569824219, -24.783790588378906, -16.133275985717773, 0.0, 11.526947021484375, -50.40875244140625, 62.978118896484375, -31.050018310546875, 27.39661407470703, -19.82415008544922, -8.26397705078125, 35.906890869140625, 81.67924499511719, -32.968719482421875, 14.9217529296875, 53.37689208984375, -70.77505493164062, 58.70531463623047, -9.457550048828125, 19.14752197265625, -14.89841079711914, 106.86001586914062, 6.213287353515625, 2.6791439056396484, -2.626842498779297, 126.31716918945312, 19.788772583007812, 9.254829406738281, 52.21446228027344, -76.24404907226562, 86.91838073730469, 126.05560302734375, 53.202056884765625, -14.923370361328125, 91.189208984375, 48.475677490234375, 77.34467315673828, 91.281494140625, 51.27537536621094, 62.35575866699219, -43.551055908203125, -48.954345703125, 1.8704833984375, 41.49359130859375, 59.73057556152344, 89.99969482421875, -10.37200927734375, -84.3846435546875, 103.88883972167969, 32.617164611816406, 7.8822021484375, -41.78240966796875, -80.83612060546875, -10.17071533203125, 12.8399658203125, -65.52783203125, 90.1976318359375, 65.12844848632812, -11.8560791015625, 78.3095703125, 0.0, 13.812263488769531, 84.93738555908203, 5.162109375, 18.21453857421875, 113.207763671875, -44.4224853515625, 77.94244384765625, -3.314971923828125, 83.26553344726562, 15.144241333007812, 44.853492736816406, -32.424713134765625, 41.41314697265625, 13.5128173828125, -31.171310424804688, 0.8615570068359375, 10.986305236816406, 67.25048828125, 38.1932373046875, 26.59497833251953, -21.292755126953125, -35.1590576171875, -6.3606719970703125, 46.13232421875, -29.4930419921875, 85.60403442382812, 108.7513427734375, 2.6414794921875, 12.13382339477539, 82.07586669921875, 25.09149169921875, -5.13702392578125, 112.24586486816406, 114.10870361328125, 42.29973602294922, 24.28570556640625, 5.6387939453125, 32.98638916015625, 64.84602355957031], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000459.npy"}
{"epoch": 0.9612565445026178, "step": 460, "batch_size": 128, "mean": 24.85137176513672, "std": 49.639129638671875, "min": -113.273681640625, "p10": -40.67239074707031, "median": 18.355072021484375, "p90": 92.16649475097655, "max": 137.1727294921875, "pos_frac": 0.7265625, "sample": [39.39299011230469, 44.238983154296875, -67.55429077148438, 71.98458862304688, 27.90789794921875, 85.34036254882812, -42.832611083984375, 71.2080078125, -39.74658203125, 97.8648681640625, -59.97206115722656, 7.025665283203125, 17.4930419921875, 7.350372314453125, 35.62652587890625, -5.36773681640625, -14.814056396484375, 8.0909423828125, 86.89920043945312, 95.04385375976562, 31.515960693359375, 22.04095458984375, 14.53643798828125, 33.82891845703125, 38.49359130859375, 7.5221099853515625, 43.251617431640625, -3.18768310546875, 123.61325073242188, -16.553802490234375, -5.4256591796875, 43.96693420410156, 19.44927978515625, 135.2022705078125, 68.21884155273438, -33.233970642089844, 63.602027893066406, -69.9429931640625, 118.27688598632812, -13.000228881835938, -99.83360290527344, 95.68069458007812, -54.47283935546875, 11.50604248046875, 41.276336669921875, 100.9322509765625, -25.665130615234375, 78.58927917480469, 14.814849853515625, -26.411964416503906, 55.260101318359375, 2.376018524169922, 3.153156280517578, 2.6422348022460938, -21.099266052246094, 10.035873413085938, 8.267097473144531, -8.23236083984375, -8.289413452148438, 4.1336212158203125, 100.45748901367188, 40.62371826171875, 67.7906723022461, 15.383705139160156, 22.716201782226562, -43.19305419921875, 6.052032470703125, 30.724578857421875, 78.33468627929688, -0.3617286682128906, -10.913482666015625, 31.445831298828125, 19.21710205078125, 11.65118408203125, 76.87136840820312, 52.434486389160156, 76.49896240234375, 76.655029296875, 38.20343017578125, 16.43545150756836, 94.49072265625, -1.0207061767578125, -38.2958984375, 137.1727294921875, 38.62261962890625, 121.98745727539062, 63.82733154296875, 77.54995727539062, -49.895660400390625, 91.17974853515625, 54.62120056152344, 46.396270751953125, -43.800567626953125, 6.374824523925781, -24.796249389648438, 16.683319091796875, 88.30924987792969, -113.273681640625, 65.52658081054688, -21.65240478515625, 29.67223358154297, -59.395477294921875, 89.35699462890625, 7.72332763671875, 10.855072021484375, -47.06599426269531, -9.272315979003906, -22.672088623046875, 86.61167907714844, 30.36700439453125, 22.38665771484375, 61.92999267578125, 30.2498779296875, 94.46890258789062, -62.07562255859375, 0.4778022766113281, 4.599994659423828, 68.13014221191406, 33.29327392578125, 112.21369934082031, 67.88592529296875, 11.868797302246094, 24.1380615234375, -21.37902069091797, 3.4987945556640625, 3.776857376098633, 13.551620483398438, 6.7272186279296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000460.npy"}
{"epoch": 0.9633507853403142, "step": 461, "batch_size": 128, "mean": 30.120145797729492, "std": 47.89936065673828, "min": -87.970458984375, "p10": -27.2209732055664, "median": 28.127395629882812, "p90": 97.9922637939453, "max": 118.1363525390625, "pos_frac": 0.7109375, "sample": [16.6561279296875, -64.34249877929688, 111.07028198242188, 13.775733947753906, 13.716964721679688, 53.21942138671875, 82.5442123413086, 31.552490234375, -17.127979278564453, 110.92630004882812, 43.6380615234375, 0.0, 7.257476806640625, 118.1363525390625, 37.5443115234375, 61.00847625732422, 67.64690399169922, 28.9085693359375, -8.48773193359375, 10.217529296875, 38.198211669921875, 96.13623046875, 82.23522186279297, -37.35972595214844, 65.59042358398438, -1.865386962890625, 14.85333251953125, 107.82217407226562, -25.138839721679688, 83.61922454833984, -4.713739395141602, 114.34506225585938, 43.1541748046875, 83.37796020507812, -5.5577545166015625, 86.73674011230469, 104.82424926757812, 12.91680908203125, -47.910400390625, -36.057403564453125, 103.46310424804688, -59.93415832519531, 7.720916748046875, 22.156845092773438, 46.595184326171875, 4.619853973388672, 65.81192016601562, 43.776641845703125, 103.13407897949219, 11.65289306640625, 51.21147155761719, -19.979751586914062, 29.51776123046875, -2.431121826171875, 10.502933502197266, -18.968978881835938, 75.34796142578125, -87.970458984375, -43.872161865234375, 33.21405029296875, 52.0576286315918, 3.822723388671875, -7.003662109375, 39.33226013183594, 84.41683959960938, 20.003921508789062, -71.52389526367188, 28.387420654296875, 96.73382568359375, 3.9071807861328125, 18.13644027709961, 90.0491943359375, -10.75259780883789, 92.90969848632812, -6.386627197265625, 103.33251953125, 63.819580078125, 60.6424560546875, -38.76690673828125, -24.11639404296875, 91.57071685791016, 3.41705322265625, 73.44949340820312, 54.4033203125, 92.00650024414062, 1.5543746948242188, -14.59124755859375, -46.50408935546875, -9.525911331176758, 45.14031982421875, 42.44049072265625, 54.77781677246094, 7.031951904296875, 77.65960693359375, 12.3131103515625, 101.92918395996094, -4.323905944824219, 27.86737060546875, 85.544921875, -67.14225006103516, 46.77123260498047, 1.4738178253173828, 104.82698059082031, 14.50421142578125, -18.9449462890625, -6.9868316650390625, 52.07862854003906, 86.94960021972656, 87.5539779663086, -2.0282211303710938, -14.773624420166016, 35.50782775878906, -5.182758331298828, 50.26936340332031, 9.002120971679688, 1.9793014526367188, 54.218772888183594, 52.46116638183594, -58.46051788330078, -13.154937744140625, 31.373687744140625, 10.487510681152344, -32.07928466796875, 104.3665771484375, -12.82526969909668, 18.648651123046875, 59.7559814453125, 100.92861938476562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000461.npy"}
{"epoch": 0.9654450261780104, "step": 462, "batch_size": 128, "mean": 37.065879821777344, "std": 50.133365631103516, "min": -88.9591064453125, "p10": -29.46343231201172, "median": 38.883941650390625, "p90": 97.29525756835938, "max": 141.56005859375, "pos_frac": 0.7578125, "sample": [73.62890625, 47.39744567871094, -3.6588592529296875, 61.26171875, 91.49319458007812, 31.18695068359375, 73.34590148925781, 71.84417724609375, 36.689353942871094, 91.15798950195312, -71.52340698242188, 70.82098388671875, 105.577392578125, 42.69779968261719, 71.39407348632812, 9.772201538085938, 35.35858154296875, 59.738525390625, 38.982391357421875, 1.1807861328125, 30.903045654296875, 69.6514892578125, 77.77774047851562, 35.69207763671875, 44.70723342895508, -0.1500244140625, 73.59715270996094, -1.3494911193847656, 85.231689453125, -44.25459289550781, -61.08905029296875, 45.519927978515625, -1.691741943359375, 42.589111328125, -14.965240478515625, 90.47457885742188, 93.95149230957031, -29.102508544921875, -35.763763427734375, 16.94818115234375, 38.785491943359375, 92.38391876220703, 75.74981689453125, 4.8960418701171875, 136.4442138671875, 49.454872131347656, 40.43617248535156, 93.48779296875, 87.676513671875, 1.631195068359375, 100.26817321777344, 8.508583068847656, 50.74192810058594, -54.364227294921875, 14.328659057617188, -6.1305694580078125, 97.13059997558594, 39.96681213378906, 12.420341491699219, 12.97406005859375, 71.05776977539062, 127.29547119140625, 6.301727294921875, 97.67945861816406, -7.101470947265625, 38.565696716308594, -52.76190948486328, 38.0430908203125, 133.8804931640625, -2.1295623779296875, 83.75650024414062, 96.3855209350586, 86.60914611816406, 82.22518920898438, 51.730316162109375, 45.36170959472656, -16.94659423828125, 15.45855712890625, 4.121417999267578, -44.85430908203125, 10.842235565185547, -20.0281982421875, -8.374603271484375, 11.955795288085938, -68.6375732421875, 23.77496337890625, 68.5579833984375, 136.215087890625, -47.73976135253906, 57.5782470703125, -30.305587768554688, 113.17422485351562, 41.905052185058594, 35.081451416015625, 26.141876220703125, 141.56005859375, -40.86762237548828, 32.154571533203125, 67.25382995605469, 90.92938232421875, 51.18238830566406, -2.333099365234375, 137.6209716796875, 21.646392822265625, -11.476287841796875, 13.861083984375, 61.211647033691406, 31.063446044921875, 45.76092529296875, 7.7601318359375, 49.60955810546875, 6.966888427734375, 46.40252685546875, 99.65109252929688, -65.11508178710938, 120.13851928710938, -19.8232421875, 56.75782012939453, 67.65338897705078, -17.855972290039062, 106.23213195800781, 26.5810546875, -5.1488037109375, -21.602073669433594, 37.4871826171875, 73.31318664550781, 66.18215942382812, -88.9591064453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000462.npy"}
{"epoch": 0.9675392670157068, "step": 463, "batch_size": 128, "mean": 28.50543975830078, "std": 44.492794036865234, "min": -84.13265991210938, "p10": -18.749798583984372, "median": 22.015892028808594, "p90": 88.25756225585937, "max": 125.660888671875, "pos_frac": 0.71875, "sample": [78.114990234375, 6.0984039306640625, 3.7962646484375, 9.611778259277344, 6.130859375, -2.2365798950195312, 113.90264892578125, 6.736083984375, -84.13265991210938, 40.893035888671875, 86.08631134033203, 3.4031982421875, -69.85356140136719, -30.50830841064453, 16.627899169921875, 85.25341796875, 22.931793212890625, 44.592926025390625, 13.7698974609375, 7.21287727355957, 15.376029968261719, -60.55999755859375, -9.883453369140625, 8.085350036621094, 45.58258056640625, 4.0272369384765625, -32.2213134765625, 20.730255126953125, 77.0037841796875, 96.88214111328125, -78.11082458496094, 60.60835266113281, 13.16607666015625, -30.52471923828125, -3.2320556640625, 102.0482177734375, 88.08355712890625, 83.47096252441406, 125.660888671875, -12.048805236816406, 74.4801025390625, -8.769073486328125, 45.76539611816406, 113.376953125, 21.837142944335938, 18.529876708984375, 106.59326171875, -12.420570373535156, 22.19464111328125, -0.9738540649414062, 21.43236541748047, 22.6524658203125, 103.5693588256836, 32.9205322265625, 13.821006774902344, 23.32781982421875, 91.21273803710938, 0.514923095703125, 81.37557983398438, 69.85440826416016, 44.1229248046875, 79.08709716796875, -1.009918212890625, -37.01783752441406, -17.635543823242188, 8.292831420898438, 83.1805419921875, 18.876296997070312, 98.94512939453125, 27.171173095703125, 81.8251953125, -22.9244384765625, -3.52410888671875, 44.888397216796875, 38.540557861328125, 4.48260498046875, 60.441864013671875, -5.2907562255859375, -10.695274353027344, 24.526519775390625, 79.17076110839844, 15.499267578125, 52.79833984375, 72.03173828125, 60.5621337890625, 5.064859390258789, 34.3048095703125, 74.42778778076172, 44.5745849609375, -8.239997863769531, -14.54705810546875, 72.20294189453125, -18.090179443359375, 0.0, -1.266845703125, 4.229213714599609, 6.376630783081055, 111.53262329101562, 23.1746826171875, -10.888694763183594, 46.27616882324219, 63.05137634277344, -14.030136108398438, -2.0652999877929688, 29.29926300048828, -25.868759155273438, 19.720794677734375, 23.42550277709961, 38.717254638671875, -1.5084381103515625, -2.315216064453125, -29.4588623046875, -4.0426025390625, 38.94639587402344, 4.0185394287109375, 79.82989501953125, 90.60166931152344, 41.44449996948242, 74.0225830078125, 52.7568359375, 109.0904541015625, 28.2227783203125, 66.25082397460938, 38.626922607421875, 65.65303039550781, 88.66357421875, -83.4150390625, -20.288909912109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000463.npy"}
{"epoch": 0.9696335078534032, "step": 464, "batch_size": 128, "mean": 29.005157470703125, "std": 45.12446975708008, "min": -124.51473999023438, "p10": -23.778137207031246, "median": 23.392181396484375, "p90": 85.08554306030274, "max": 121.092041015625, "pos_frac": 0.765625, "sample": [0.0, 58.62800598144531, 9.509956359863281, 53.205902099609375, -8.469329833984375, 71.8707046508789, 46.048583984375, -26.070526123046875, 70.16018676757812, -5.072383880615234, 51.142547607421875, 22.461471557617188, 84.84315490722656, 16.8009033203125, 53.11390686035156, 7.041204452514648, 79.56298828125, 3.320953369140625, 85.00609588623047, 40.962158203125, 36.34356689453125, 84.578125, 12.988725662231445, 6.0926513671875, 38.506500244140625, 59.684478759765625, 16.59636688232422, -84.32292938232422, 121.092041015625, 17.281982421875, 12.9039306640625, -6.261199951171875, 15.453628540039062, 95.58465576171875, 81.62823486328125, 94.89619445800781, 24.886444091796875, -5.178352355957031, -8.078948974609375, 50.60209655761719, 9.54412841796875, 25.61627197265625, 71.44189453125, 60.66270446777344, 61.22381591796875, -2.661346435546875, -8.590713500976562, 15.704498291015625, 1.8190155029296875, 59.409881591796875, 19.110671997070312, 10.097625732421875, 3.349884033203125, 74.14886474609375, 11.00497817993164, 40.091705322265625, 71.75808715820312, 66.81158447265625, -74.56878662109375, 71.41354370117188, -2.675140380859375, 68.9114990234375, -5.7761077880859375, 85.27091979980469, 103.35546875, -16.35888671875, 21.523582458496094, 20.290786743164062, 24.71600341796875, 58.189453125, -41.58551025390625, -124.51473999023438, -40.74964904785156, 1.3668441772460938, 16.771820068359375, 24.322891235351562, -19.193603515625, -34.77459716796875, 78.42108154296875, 18.63542938232422, 102.29495239257812, 105.11885070800781, 80.8265380859375, 89.99594116210938, -26.0841064453125, 83.238525390625, 1.3682823181152344, -26.376678466796875, 11.0272216796875, 4.997062683105469, 29.369033813476562, -58.573638916015625, 90.47744750976562, -5.423187255859375, -40.4603271484375, 9.157299041748047, -0.8257980346679688, 76.33786010742188, 8.078399658203125, 8.980690002441406, 60.75751495361328, 97.95930480957031, 26.469993591308594, 93.63396453857422, 73.02435302734375, -22.198089599609375, 48.4393310546875, 12.358047485351562, 78.28396606445312, -34.08058166503906, 13.217864990234375, 8.802825927734375, 1.7738037109375, 87.24008178710938, 69.19525146484375, 61.48757553100586, 47.7296142578125, 86.34937286376953, 73.6412582397461, -101.0364990234375, 30.938934326171875, 19.06298828125, -22.795684814453125, 53.145751953125, 24.389842987060547, -14.783645629882812, 50.064727783203125, 43.183349609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000464.npy"}
{"epoch": 0.9717277486910995, "step": 465, "batch_size": 128, "mean": 36.1943473815918, "std": 47.771705627441406, "min": -93.77597045898438, "p10": -19.051615905761718, "median": 28.17395782470703, "p90": 100.30400085449219, "max": 130.35415649414062, "pos_frac": 0.7578125, "sample": [-0.7877960205078125, -4.312957763671875, 6.78350830078125, 6.07275390625, 93.842041015625, -4.8445587158203125, -18.170318603515625, 113.36318969726562, 11.49493408203125, 16.75274658203125, 4.3542938232421875, 20.740753173828125, 4.022552490234375, 8.390228271484375, -58.31878662109375, -28.42449951171875, 105.00439453125, 88.126708984375, 89.32894897460938, 98.72454833984375, 88.95196533203125, -42.74560546875, 33.41722106933594, 81.28744506835938, 23.7637939453125, 4.837852478027344, 4.55010986328125, 94.74960327148438, 91.98291015625, 65.80714416503906, -18.454055786132812, 78.00799560546875, 24.58740234375, 55.513214111328125, 12.057037353515625, 2.1976318359375, -41.95159912109375, 113.390869140625, 66.16036987304688, 130.35415649414062, 5.0488739013671875, 43.19677734375, -0.541748046875, -23.997528076171875, -10.938468933105469, 31.0245361328125, -0.9652786254882812, 5.399923324584961, -15.736602783203125, 64.43161010742188, 71.87066650390625, 61.2540283203125, 52.961181640625, 72.92462158203125, 46.10707092285156, -33.830116271972656, 88.19718933105469, 29.00592041015625, -7.7364501953125, 25.675918579101562, 47.557281494140625, 27.341995239257812, 36.32379150390625, 50.87574768066406, 101.81442260742188, 21.343490600585938, 102.88223266601562, 104.12078857421875, -3.01702880859375, 21.337846755981445, 26.72580909729004, 90.12906646728516, 71.50283813476562, 58.92585754394531, 52.530723571777344, -54.5933837890625, -8.71588134765625, 33.846832275390625, 34.21504211425781, 46.097740173339844, 19.0078125, 4.874664306640625, 3.471038818359375, 24.0823974609375, 4.117179870605469, 69.48369598388672, 116.7586669921875, -21.921707153320312, 17.85204315185547, -12.409942626953125, 103.719482421875, 108.38272094726562, 125.17535400390625, 14.809913635253906, 37.8050537109375, 20.07421875, 117.77569580078125, 99.65667724609375, 93.75762939453125, 3.8904342651367188, 69.00454711914062, 86.19303894042969, 105.95230102539062, 84.71099853515625, 8.468994140625, 77.15746307373047, -16.44854736328125, 73.405517578125, -39.313201904296875, 30.456390380859375, -38.000885009765625, 7.7295684814453125, 99.13145446777344, 83.20217895507812, 86.08692932128906, 92.02760314941406, -53.361907958984375, 82.15157318115234, -1.083953857421875, 8.00532341003418, -20.4459228515625, 61.703155517578125, -7.95794677734375, -13.762611389160156, 89.4461669921875, -93.77597045898438, -6.69219970703125, 43.318023681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000465.npy"}
{"epoch": 0.9738219895287958, "step": 466, "batch_size": 128, "mean": 38.32670593261719, "std": 41.41657638549805, "min": -74.4180908203125, "p10": -7.947463989257811, "median": 35.8747673034668, "p90": 93.57886657714842, "max": 137.81546020507812, "pos_frac": 0.828125, "sample": [-6.350898742675781, 48.579898834228516, 49.01170349121094, 80.1561279296875, 45.048614501953125, 101.42214965820312, 2.020477294921875, 0.0, 52.40478515625, 35.561607360839844, 79.04214477539062, -17.715106964111328, 62.98243713378906, 54.594329833984375, 15.490234375, 106.74859619140625, 23.558609008789062, 0.2201080322265625, 60.609405517578125, 40.92022705078125, 69.8935546875, 26.28852081298828, 61.82627868652344, -22.627044677734375, 13.16326904296875, 36.18792724609375, 43.87648010253906, 21.788543701171875, 74.70661163330078, 32.81660461425781, -0.681121826171875, 82.03744506835938, 1.9272613525390625, 69.81500244140625, 70.13055419921875, 8.006916046142578, 67.61734008789062, -7.433807373046875, 87.17465209960938, 49.47730255126953, 127.09950256347656, 8.672805786132812, 104.84475708007812, 41.77410888671875, 79.28668212890625, 108.70660400390625, 31.17711639404297, -23.71319580078125, -55.65486145019531, 13.532073974609375, 0.967864990234375, 79.46856689453125, 1.282562255859375, 79.98229217529297, 66.64501953125, 78.3938980102539, -10.800750732421875, -53.95751953125, 19.3963623046875, 91.12515258789062, 53.56013488769531, 77.91941833496094, 56.91145324707031, 54.32673645019531, 12.194976806640625, 63.301307678222656, 74.76708984375, 67.17398071289062, 99.63241577148438, 77.48597717285156, 0.0, 78.85719299316406, 103.80963134765625, -20.1956787109375, -6.417564392089844, 17.322853088378906, 67.70528411865234, -14.152999877929688, 40.110443115234375, 137.81546020507812, 4.807472229003906, 7.61737060546875, 124.29901123046875, 17.86285400390625, 18.51885986328125, -47.99107360839844, 20.2349853515625, 40.429351806640625, 20.50738525390625, 60.72667694091797, 70.7655029296875, 17.417739868164062, 14.513969421386719, -13.143806457519531, 114.939697265625, 1.1575355529785156, 58.5919189453125, 59.2119140625, -24.75322723388672, -74.4180908203125, 38.51255798339844, 23.18695068359375, 0.2736053466796875, 101.41911315917969, 19.13159942626953, 0.0, 99.30419921875, 13.27728271484375, -0.04028511047363281, 55.76649475097656, 74.4163818359375, 52.505462646484375, 0.7952194213867188, 80.07777404785156, 30.86610984802246, 34.467864990234375, 19.403778076171875, 17.321807861328125, 23.583099365234375, 10.4285888671875, 22.485870361328125, 14.846282958984375, 121.51226806640625, 55.43096923828125, -5.256072998046875, -9.14599609375, 44.31005859375, 24.986404418945312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000466.npy"}
{"epoch": 0.9759162303664921, "step": 467, "batch_size": 128, "mean": 28.077857971191406, "std": 49.341732025146484, "min": -94.784912109375, "p10": -31.246383666992184, "median": 18.593765258789062, "p90": 98.23937072753905, "max": 133.0081329345703, "pos_frac": 0.6796875, "sample": [99.90711212158203, 69.15176391601562, 24.588104248046875, 37.53777313232422, 44.175514221191406, -5.4976043701171875, 3.00421142578125, 33.08812713623047, 65.414794921875, 10.4345703125, -94.784912109375, -74.4617919921875, -9.82177734375, -3.1591949462890625, 15.95843505859375, 70.86776733398438, -52.411346435546875, 101.50323486328125, 0.52471923828125, 49.34941101074219, -13.714599609375, 60.8184814453125, -11.163040161132812, 48.81764221191406, 98.07037353515625, 36.23834228515625, 12.315780639648438, 70.89089965820312, 7.028541564941406, 11.767364501953125, 20.1256103515625, 86.5416259765625, 60.6983642578125, 84.47271728515625, 101.33876037597656, 46.10554504394531, 5.0659027099609375, 109.28305053710938, 32.30876159667969, -25.10699462890625, 98.80856323242188, 0.48138427734375, 0.21329879760742188, 84.267578125, 41.076446533203125, 79.4769287109375, -7.119489669799805, -30.720550537109375, 94.18994140625, -60.853614807128906, 9.474288940429688, -57.78440856933594, -18.817787170410156, 84.27070617675781, -32.47332763671875, 74.2200927734375, -13.839786529541016, -19.647918701171875, 83.31778717041016, 109.09768676757812, 3.57708740234375, 8.039255142211914, -1.4843578338623047, 9.787109375, 79.75238037109375, 11.27056884765625, -0.8199691772460938, 31.95221519470215, 1.874847412109375, -27.376117706298828, 10.469329833984375, -72.67256927490234, 76.91860961914062, 0.0, -3.10797119140625, -1.3309783935546875, 48.06988525390625, 110.22254943847656, 3.6512451171875, -0.42078399658203125, 0.0, 15.000656127929688, 56.461273193359375, -5.09552001953125, 81.5916748046875, 46.64964294433594, 78.67469024658203, -28.784652709960938, 133.0081329345703, -56.852577209472656, -45.686614990234375, 110.64666748046875, 64.81134033203125, 44.6815185546875, 95.895263671875, -1.5608978271484375, -12.646636962890625, 23.279815673828125, -2.247467041015625, 34.4932861328125, -25.46453857421875, 63.44639587402344, 87.5638427734375, 27.91705322265625, -23.642471313476562, 17.061920166015625, -38.17012023925781, -39.02581787109375, 8.849249839782715, 91.837646484375, 64.41351318359375, 59.95075988769531, -62.1702880859375, 72.2440185546875, 10.656517028808594, -5.2935791015625, -12.928325653076172, 26.709762573242188, 132.43499755859375, 38.62327194213867, 99.9683837890625, 25.960189819335938, 37.146392822265625, 102.44931030273438, 88.98831176757812, 98.63369750976562, -35.843414306640625, 16.047454833984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000467.npy"}
{"epoch": 0.9780104712041885, "step": 468, "batch_size": 128, "mean": 26.490917205810547, "std": 43.41545486450195, "min": -87.008544921875, "p10": -22.24852752685547, "median": 16.047992706298828, "p90": 88.94185791015624, "max": 144.05340576171875, "pos_frac": 0.71875, "sample": [48.373046875, 116.843505859375, 69.95570373535156, -23.448486328125, 21.08197021484375, 8.95587158203125, -9.091583251953125, 55.065155029296875, 31.784597396850586, 16.85388946533203, 23.15814208984375, -12.43017578125, 20.17755126953125, 97.61090087890625, -29.0091552734375, -11.301750183105469, -7.368865966796875, 94.21664428710938, -38.17189025878906, -9.1275634765625, 41.88555908203125, 11.005964279174805, -29.77999496459961, 28.79705810546875, 48.933563232421875, 113.35711669921875, 42.82025146484375, 41.61384582519531, 20.909683227539062, -21.871685028076172, 1.09088134765625, 8.981399536132812, 107.25845336914062, -10.73924446105957, 29.138214111328125, -13.345170974731445, -3.483856201171875, 91.36386108398438, -87.008544921875, 9.313018798828125, 8.824729919433594, 15.242095947265625, 1.055755615234375, 44.45726013183594, 10.495109558105469, 12.548965454101562, 48.768646240234375, -12.487785339355469, 34.69964599609375, 3.027374267578125, 4.1971588134765625, 1.6385326385498047, 10.044891357421875, 2.11322021484375, 6.85577392578125, 16.94610595703125, 8.765045166015625, 48.215240478515625, 71.11249542236328, 78.17356872558594, 97.15106201171875, -13.105194091796875, -67.63955688476562, 12.738571166992188, 144.05340576171875, 9.98797607421875, 34.160804748535156, 75.87924194335938, -10.081390380859375, 28.643646240234375, 36.387237548828125, 20.23956298828125, -22.914764404296875, 58.89776611328125, 78.45201873779297, 76.71209716796875, 7.974830627441406, 12.411590576171875, 6.732185363769531, 45.77519226074219, -12.776241302490234, 110.46978759765625, 23.02011489868164, 102.03884887695312, 14.5693359375, 55.62578582763672, 88.4727783203125, -11.61478042602539, 75.52693176269531, 52.915687561035156, -3.39569091796875, 5.235525131225586, 1.2669677734375, 111.11395263671875, -1.28485107421875, 28.85247802734375, 64.98577117919922, 3.8708724975585938, 23.08355712890625, -6.836578369140625, 90.036376953125, 86.62865447998047, 51.839019775390625, 5.88653564453125, -45.808837890625, -21.962997436523438, -40.83026123046875, 55.358612060546875, 44.70703125, 85.514892578125, -9.190872192382812, -3.66748046875, 72.51751708984375, -34.82359313964844, -28.297149658203125, 13.5096435546875, 84.23539733886719, -0.00936126708984375, -5.91741943359375, -46.261436462402344, 52.03309631347656, 75.3211669921875, 37.572540283203125, 83.46783447265625, -34.76261901855469, 92.98519897460938, -20.49993896484375, 84.59765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000468.npy"}
{"epoch": 0.9801047120418848, "step": 469, "batch_size": 128, "mean": 34.88108825683594, "std": 48.65401077270508, "min": -106.18585205078125, "p10": -19.622296142578126, "median": 31.981239318847656, "p90": 97.9147850036621, "max": 128.47369384765625, "pos_frac": 0.7578125, "sample": [11.42071533203125, 93.04791259765625, 0.0, 9.313434600830078, 23.12451171875, 40.47308349609375, -10.042556762695312, 23.64154052734375, -9.864334106445312, 54.8199462890625, 57.809417724609375, 28.013916015625, 81.34466552734375, -101.76959228515625, 20.957794189453125, 6.6043243408203125, 103.68801879882812, 72.81297302246094, 14.6702880859375, 73.17086029052734, -28.831192016601562, 53.875213623046875, 6.144275665283203, -0.19172096252441406, -69.39714050292969, 95.03231811523438, 6.032073974609375, 87.29058837890625, 95.66543579101562, 25.501953125, 71.86189270019531, 65.72930908203125, 121.384033203125, 76.42236328125, 1.2187423706054688, 46.90849304199219, 43.70184326171875, 56.43865966796875, 97.7546157836914, 36.504638671875, -1.5675277709960938, -30.5718994140625, 16.6278076171875, 5.39599609375, 76.3363037109375, -6.763580322265625, 16.888015747070312, 20.378753662109375, 31.455032348632812, 27.7061767578125, 50.513153076171875, 103.21539306640625, 93.24615478515625, 99.96282958984375, -48.177703857421875, 39.81255340576172, 58.8673095703125, 120.63174438476562, 77.89532470703125, -35.774566650390625, 6.240724563598633, -4.45574951171875, 110.15399169921875, -43.447265625, -51.90960693359375, 49.48974609375, 20.861221313476562, -2.998046875, -19.5692138671875, 91.44755554199219, 15.249099731445312, 79.37126159667969, 71.40472412109375, 93.03373718261719, -82.425537109375, 3.374227523803711, 33.45042419433594, 80.84964752197266, -10.244125366210938, 97.43557739257812, 90.27362060546875, 45.917938232421875, 29.28070068359375, 1.18756103515625, 99.0216064453125, -3.29071044921875, -106.18585205078125, 4.450105667114258, 35.869781494140625, 19.393882751464844, 42.6182861328125, 99.4532470703125, -1.31524658203125, 90.93392944335938, -19.74615478515625, -8.358123779296875, 2.2507247924804688, 0.6609878540039062, -10.455230712890625, 78.70985412597656, -11.207901000976562, 2.7113265991210938, -6.71868896484375, 128.47369384765625, 18.654821395874023, -33.59344482421875, 96.25442504882812, 45.89391326904297, 65.38958740234375, 105.356201171875, 60.422908782958984, 8.977615356445312, 93.19200134277344, 32.5074462890625, -8.990608215332031, 51.0882568359375, 7.601469039916992, 86.80970764160156, 70.37542724609375, 48.19530487060547, 101.51803588867188, 98.28851318359375, 49.43040466308594, -12.48651123046875, 42.64564514160156, 18.2779541015625, 108.4552001953125, -29.087646484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000469.npy"}
{"epoch": 0.9821989528795811, "step": 470, "batch_size": 128, "mean": 36.574974060058594, "std": 48.97336959838867, "min": -75.88561248779297, "p10": -19.27794494628906, "median": 30.54287338256836, "p90": 102.1395751953125, "max": 128.601318359375, "pos_frac": 0.796875, "sample": [-19.03021240234375, 17.30706787109375, 90.68212890625, 67.44668579101562, -9.559890747070312, 35.044647216796875, 2.2129669189453125, 82.74646759033203, 96.95623779296875, 14.631591796875, -27.6282958984375, 4.0648956298828125, -4.7752838134765625, 76.29281616210938, 58.63494873046875, 85.07907104492188, 96.42240905761719, 1.3956642150878906, -9.082740783691406, 32.746826171875, 59.592960357666016, -45.67388916015625, -50.29058837890625, -54.645904541015625, 128.601318359375, 79.11685180664062, 29.52948760986328, 16.8311767578125, 89.4453125, 73.66061401367188, 101.7547607421875, 63.67285919189453, 2.697967529296875, 1.357666015625, 68.81670379638672, -66.56112670898438, 80.16448974609375, 49.060943603515625, 23.42626953125, -49.04193115234375, 7.7236328125, 103.0374755859375, 61.95855712890625, 21.505584716796875, 12.04779052734375, 82.98519897460938, 110.43296813964844, -4.36334228515625, 26.813514709472656, 55.94281005859375, 1.6049652099609375, 57.26263427734375, 20.7633056640625, 15.65849494934082, -14.843780517578125, 58.96160888671875, 38.99833679199219, 0.16680145263671875, 82.56623840332031, 15.57440185546875, 43.63304138183594, 3.3902511596679688, 90.66380310058594, 9.5440673828125, 23.745506286621094, 12.175548553466797, -5.223747253417969, -61.478759765625, 10.687652587890625, 43.6912841796875, -25.0186767578125, 95.7087631225586, 74.14395141601562, -10.140647888183594, -58.9810791015625, -16.741592407226562, -1.500640869140625, 61.903053283691406, 104.17927551269531, 84.09225463867188, 27.14398193359375, 31.556259155273438, 77.85537719726562, 123.05186462402344, 17.429336547851562, 5.209037780761719, 9.188644409179688, 105.73626708984375, 118.81082153320312, 100.98025512695312, -15.695533752441406, 82.6895751953125, 66.84978485107422, 76.67544555664062, -1.2566986083984375, 26.708038330078125, 17.336891174316406, 71.90846252441406, 75.20492553710938, 3.81781005859375, 91.69319915771484, 94.133056640625, 53.80770492553711, 120.12013244628906, 0.3535003662109375, 50.48297119140625, 3.2512855529785156, 7.0709228515625, 2.158111572265625, -75.88561248779297, 5.917816162109375, -58.81614685058594, 120.977294921875, -7.7906646728515625, 58.58911895751953, 120.11083984375, 0.99261474609375, 41.530181884765625, 55.07122802734375, 42.441436767578125, 119.01861572265625, -49.13478088378906, 43.1326904296875, 17.317596435546875, 119.5540771484375, 74.26738739013672, 103.51278686523438, -19.855987548828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000470.npy"}
{"epoch": 0.9842931937172775, "step": 471, "batch_size": 128, "mean": 25.81607437133789, "std": 44.27755355834961, "min": -94.93695068359375, "p10": -26.131904220581053, "median": 23.009872436523438, "p90": 88.27138061523436, "max": 132.1220703125, "pos_frac": 0.703125, "sample": [-26.643386840820312, 53.75044250488281, 24.902130126953125, 30.9544677734375, 74.44429016113281, 123.0531005859375, 50.80400085449219, -60.4700813293457, -7.5637969970703125, 75.73214721679688, -26.619186401367188, -17.333099365234375, 4.331573486328125, 70.19982147216797, 0.6196136474609375, 24.11676025390625, 17.679351806640625, 21.520782470703125, -17.01543426513672, -9.313323974609375, 26.2685546875, 19.791351318359375, 16.222503662109375, 25.3367919921875, -40.245811462402344, -2.7123489379882812, 117.3050537109375, -3.47222900390625, -9.11962890625, 24.557235717773438, 70.58395385742188, 55.94410705566406, 58.20263671875, 10.98565673828125, 0.00433349609375, 32.443603515625, 37.43370056152344, 49.076637268066406, 21.902984619140625, 12.349853515625, 2.3941802978515625, -4.0469970703125, 87.0877685546875, 92.8489990234375, -11.235027313232422, 32.79949951171875, 26.353439331054688, 107.35185241699219, 71.72021484375, -4.94512939453125, 1.2574005126953125, 14.57269287109375, 95.7581787109375, 35.03875732421875, -31.367507934570312, 65.42218017578125, 57.72910690307617, 4.101291656494141, -10.4815673828125, 16.784889221191406, 129.69207763671875, 132.1220703125, 11.239810943603516, -7.19091796875, -25.92306900024414, -7.947509765625, -10.575355529785156, 2.4988250732421875, -94.93695068359375, 25.693359375, 42.917724609375, 8.35040283203125, 75.37774658203125, 32.7935791015625, -71.29804992675781, 30.861328125, 0.75347900390625, 30.76746368408203, 35.56040954589844, -33.554443359375, 25.024688720703125, -16.363189697265625, -2.8380126953125, -5.5098724365234375, 45.267333984375, 60.79991912841797, 10.348419189453125, 28.18828582763672, 77.71371459960938, -4.45751953125, 51.6126708984375, 89.39764404296875, -1.6621742248535156, 42.508392333984375, 9.97967529296875, 89.54740905761719, 98.493896484375, -21.431842803955078, -3.86602783203125, -17.44579315185547, 26.19070053100586, 12.122062683105469, 17.550933837890625, 35.558837890625, 37.68193054199219, 87.7886962890625, 34.00750732421875, 120.77554321289062, 36.043243408203125, 10.466415405273438, 38.4425048828125, 72.34097290039062, 74.77178955078125, -55.32262420654297, 83.7657470703125, 10.710647583007812, 60.82073974609375, 0.0, 28.766204833984375, -32.768310546875, 113.2607421875, 71.91238403320312, 111.4188232421875, -13.70855712890625, 8.17913818359375, -26.9990234375, -45.460941314697266, -55.52142333984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000471.npy"}
{"epoch": 0.9863874345549738, "step": 472, "batch_size": 128, "mean": 33.89308547973633, "std": 46.12812423706055, "min": -65.58770751953125, "p10": -16.957137870788575, "median": 24.649616241455078, "p90": 91.58238220214844, "max": 141.4107666015625, "pos_frac": 0.75, "sample": [136.4825439453125, -6.064178466796875, -16.946481704711914, -37.277992248535156, 3.642213821411133, 12.577178955078125, 4.877750396728516, 76.2733154296875, -1.34783935546875, 11.082809448242188, 28.152236938476562, 22.405426025390625, 88.97677612304688, 22.190719604492188, -49.624778747558594, 86.0842056274414, -11.647506713867188, 7.585853576660156, -22.37359619140625, 63.37577819824219, 1.593475341796875, 21.241455078125, 61.2713623046875, 31.401123046875, -16.98200225830078, 70.33209228515625, 58.94190216064453, -40.89012145996094, 3.0938682556152344, 82.91681671142578, 32.556549072265625, -6.873405456542969, 13.586212158203125, 127.20819091796875, 3.183441162109375, -22.852951049804688, 27.963836669921875, 92.46945190429688, -2.675048828125, 3.8878860473632812, 13.9083251953125, 68.78775024414062, 46.29132080078125, 51.80108642578125, -9.919532775878906, 16.406394958496094, 6.93988037109375, 6.354522705078125, 32.39300537109375, 13.946746826171875, 79.39041137695312, 34.75419616699219, 91.20220947265625, -62.684814453125, 53.54216003417969, 70.82914733886719, 91.14138793945312, 71.65130615234375, 13.269607543945312, 25.44812774658203, -9.8536376953125, 58.434478759765625, 11.757589340209961, -1.1972198486328125, 51.78033447265625, 84.86627197265625, 48.81245422363281, 50.506103515625, 46.431846618652344, -3.1647491455078125, 25.993358612060547, 23.851104736328125, 22.014617919921875, 83.81895446777344, 58.60151672363281, 115.64944458007812, 105.43997192382812, -7.053565979003906, 90.4736328125, 16.168930053710938, -5.3681640625, 140.21151733398438, 90.14149475097656, 111.29364013671875, 18.452880859375, 129.69744873046875, 110.7281494140625, 96.88043212890625, 78.48098754882812, -65.58770751953125, -11.898300170898438, -64.97500610351562, 46.56272888183594, 32.33929443359375, 8.90008544921875, 66.51568603515625, 89.9271240234375, 10.784896850585938, -25.542572021484375, 62.34527587890625, 84.88421630859375, -8.773574829101562, 76.72206115722656, 74.79974365234375, 15.665000915527344, 47.710670471191406, -2.38787841796875, 11.202667236328125, -2.085786819458008, 6.4368896484375, 96.00115966796875, 100.58052062988281, 3.4112415313720703, 5.240875244140625, 27.7216796875, -51.58744812011719, -1.8420562744140625, 141.4107666015625, -11.311492919921875, 20.85943603515625, 89.02374267578125, 42.06866455078125, -26.811492919921875, 62.396759033203125, 52.14337158203125, 52.76689147949219, -23.134010314941406, -13.270673751831055], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000472.npy"}
{"epoch": 0.9884816753926702, "step": 473, "batch_size": 128, "mean": 34.353790283203125, "std": 47.17509841918945, "min": -100.08790588378906, "p10": -19.14289093017578, "median": 30.878376007080078, "p90": 96.19466857910156, "max": 134.65380859375, "pos_frac": 0.7421875, "sample": [6.417266845703125, 95.08425903320312, 26.125900268554688, -5.738275527954102, 46.31988525390625, -3.3826904296875, 5.286834716796875, 12.99981689453125, 25.07177734375, -4.29400634765625, 37.99174499511719, 26.538116455078125, -55.3931884765625, 9.457324981689453, -33.61444854736328, 56.59262466430664, 21.935943603515625, 8.433807373046875, 39.665191650390625, 51.68743896484375, 44.10302734375, 78.55706024169922, 105.68206787109375, 112.97149658203125, 45.20806884765625, -54.6781005859375, -7.8882904052734375, 73.00128173828125, 44.77790069580078, 25.134063720703125, 104.14181518554688, 97.37713623046875, 76.22369384765625, 30.6363525390625, -9.359550476074219, -39.51637268066406, 2.131988525390625, -28.587135314941406, 114.9908447265625, -6.3369598388671875, 22.86731719970703, 54.5787353515625, 33.383148193359375, -10.795249938964844, -6.0338592529296875, -47.32355499267578, -41.015960693359375, 101.692626953125, 41.02520751953125, 35.6962890625, 40.8331298828125, 19.256607055664062, -0.11417961120605469, 110.93954467773438, -14.89190673828125, 100.2293701171875, -11.474090576171875, 52.0732421875, 109.7681884765625, -71.64749145507812, 6.274810791015625, 100.62623596191406, 88.65921020507812, 3.5902175903320312, 14.50738525390625, -13.115997314453125, 123.77056884765625, 81.56880950927734, 61.019561767578125, 118.215576171875, 62.1182861328125, 54.80536651611328, 54.16180419921875, 10.975128173828125, 3.958446502685547, 80.71442413330078, 79.19953918457031, 6.424369812011719, -30.774078369140625, 87.14691162109375, -6.3674468994140625, -6.259971618652344, 90.44863891601562, 64.53025817871094, 53.5115966796875, 84.33462524414062, -100.08790588378906, -54.831031799316406, 79.57553100585938, -5.6052093505859375, 37.52752685546875, 86.8922119140625, 2.266571044921875, 29.463912963867188, -8.62786865234375, -14.920013427734375, 22.427623748779297, 9.68560791015625, 24.93408966064453, -19.950119018554688, 95.68789672851562, 63.298805236816406, 75.91278076171875, 0.0, 67.67596435546875, 33.57366943359375, -39.69659423828125, 79.40908813476562, 18.188095092773438, 63.80162048339844, 81.35304260253906, 10.324966430664062, 2.502124786376953, 94.48432922363281, 43.691375732421875, -7.2239990234375, 3.785308837890625, -18.79693603515625, 27.088882446289062, 49.55982971191406, 85.67054748535156, 86.69393920898438, 91.58682250976562, 31.120399475097656, 134.65380859375, 75.63899230957031, 67.59593200683594, 22.106475830078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000473.npy"}
{"epoch": 0.9905759162303664, "step": 474, "batch_size": 128, "mean": 28.181812286376953, "std": 41.38920974731445, "min": -69.71014404296875, "p10": -18.52196350097656, "median": 26.602142333984375, "p90": 79.33238220214842, "max": 133.6929931640625, "pos_frac": 0.7578125, "sample": [-2.3273468017578125, 3.56640625, 123.55914306640625, -2.6583023071289062, 2.6005115509033203, -18.604827880859375, 53.73399353027344, 107.154052734375, 41.446311950683594, 76.10227966308594, 8.22406005859375, 38.17596435546875, -17.188720703125, -64.0604248046875, 5.1654205322265625, -47.68904113769531, -4.7419281005859375, -0.0762939453125, 37.523231506347656, 23.23223114013672, 32.179420471191406, 105.75868225097656, 5.87091064453125, -32.04443359375, 52.8558349609375, -20.79387664794922, -40.4012451171875, 43.865081787109375, -13.536827087402344, 32.579559326171875, 64.012451171875, 34.5596923828125, 7.4485015869140625, 70.68496704101562, 33.07344055175781, 56.92388916015625, 116.59368896484375, 13.889350891113281, 7.277351379394531, 100.05059814453125, 48.21875, 70.1712646484375, 109.96273803710938, 99.61857604980469, 46.2679443359375, 7.084468841552734, 0.0, 6.0564117431640625, 11.007598876953125, 6.0013275146484375, -0.49410247802734375, -14.8258056640625, -19.447921752929688, 45.74110412597656, 20.062328338623047, 60.341278076171875, 32.58277893066406, -18.47936248779297, 90.70140075683594, 73.83978271484375, 11.82080078125, -6.803863525390625, 3.6535186767578125, 50.241546630859375, 65.35376739501953, 26.09393310546875, 88.0041275024414, 40.661956787109375, -30.02996826171875, 7.582061767578125, 78.17318725585938, 32.02627944946289, -62.521751403808594, 38.88547134399414, 45.203765869140625, -15.314544677734375, 95.64395141601562, 66.13201904296875, 73.98701477050781, -4.41168212890625, -3.360198974609375, 19.549362182617188, 0.9289741516113281, 1.8208904266357422, 2.8171157836914062, -69.71014404296875, -18.4864501953125, 52.00337219238281, -54.073150634765625, 82.03717041015625, 52.630706787109375, -67.09580993652344, -11.644989013671875, -5.682373046875, 54.84883117675781, 7.107505798339844, 45.44346618652344, -10.0211181640625, 25.720367431640625, 72.86602783203125, 53.8997802734375, 60.80487060546875, 54.51898193359375, 46.1241455078125, 133.6929931640625, 68.39193725585938, 71.42153930664062, 17.885589599609375, 23.923858642578125, 16.28472900390625, 40.096893310546875, 49.77288818359375, 50.347503662109375, 33.01580810546875, 26.23248291015625, 29.220794677734375, 49.1941032409668, 54.134765625, 4.3387603759765625, 45.507171630859375, 36.30956268310547, 94.919189453125, 17.142852783203125, -41.694732666015625, 20.574249267578125, 8.487457275390625, 25.27655029296875, 26.9718017578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000474.npy"}
{"epoch": 0.9926701570680628, "step": 475, "batch_size": 128, "mean": 27.629199981689453, "std": 44.565921783447266, "min": -105.06887817382812, "p10": -21.170970153808593, "median": 20.349761962890625, "p90": 85.00491638183594, "max": 138.76852416992188, "pos_frac": 0.71875, "sample": [108.341064453125, 71.845703125, -17.967361450195312, 84.81348419189453, 69.92484283447266, 7.970245361328125, -105.06887817382812, 111.74887084960938, -21.001724243164062, -37.56695556640625, -6.5107574462890625, -21.075714111328125, -58.48445129394531, -39.098960876464844, 16.3929443359375, 45.183807373046875, 53.05715560913086, 85.45159149169922, 89.11715698242188, -9.242637634277344, 104.90838623046875, -18.50738525390625, 7.357597351074219, 19.477802276611328, 101.0377197265625, 83.56008911132812, 77.6934814453125, -4.257904052734375, 124.6644287109375, -0.025209426879882812, 11.843116760253906, 22.030502319335938, -38.57940673828125, 78.97102355957031, 37.930824279785156, 16.24169158935547, -42.522377014160156, 23.897811889648438, 0.2185497283935547, 89.74349975585938, 65.35830688476562, 28.062164306640625, -62.05970764160156, 34.15696716308594, 61.03889465332031, -14.58587646484375, 52.01685333251953, -12.11279296875, 70.50715637207031, -21.393234252929688, 10.387763977050781, -17.345382690429688, 75.96270751953125, 83.71768188476562, -29.237869262695312, 81.38555908203125, 102.49200439453125, -5.638467788696289, 15.377752304077148, 18.4088134765625, 60.97389221191406, 10.520393371582031, 7.343177795410156, -16.66541290283203, 24.910400390625, 22.625083923339844, 20.949554443359375, 36.947723388671875, 14.343978881835938, 27.697509765625, 10.46380615234375, 80.04254150390625, 32.962371826171875, -10.18975830078125, 46.770751953125, -19.774383544921875, 3.197071075439453, 138.76852416992188, 29.0386962890625, -3.5703506469726562, 43.37834167480469, 98.00701904296875, 66.42132568359375, -27.187530517578125, 78.40808868408203, -18.030521392822266, 41.755001068115234, 6.884765625, 43.390777587890625, 17.02936553955078, 3.8951759338378906, 17.94570541381836, 75.31912231445312, 69.8768310546875, 0.248931884765625, 83.03123474121094, 79.33509826660156, 89.43171691894531, 34.64881896972656, 48.9111328125, 11.444557189941406, -15.12261962890625, 22.654647827148438, -0.3318634033203125, -8.989334106445312, -16.275909423828125, -38.06272888183594, 19.749969482421875, 45.51582336425781, 53.849388122558594, 64.83154296875, 21.878494262695312, 10.689018249511719, 15.328765869140625, 69.1058349609375, -20.93975830078125, 55.11482238769531, -23.155059814453125, -44.4063720703125, 30.76702880859375, 84.236328125, 23.2381591796875, 14.490055084228516, 109.31002807617188, 15.782901763916016, 10.4658203125, 0.09772682189941406, -4.798553466796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000475.npy"}
{"epoch": 0.9947643979057592, "step": 476, "batch_size": 128, "mean": 32.41803741455078, "std": 45.22163772583008, "min": -69.18380737304688, "p10": -13.806586456298827, "median": 24.315135955810547, "p90": 95.06796722412109, "max": 131.77359008789062, "pos_frac": 0.734375, "sample": [-16.174163818359375, -33.6981201171875, -7.6118316650390625, -27.343536376953125, 2.722442626953125, 53.264495849609375, -4.560874938964844, -3.0914669036865234, 17.233856201171875, 28.6619873046875, 65.83056640625, 23.9986572265625, 88.88133239746094, 102.53961181640625, 40.20269775390625, 122.1580810546875, -8.11767578125, 98.74458312988281, 81.63235473632812, -0.9413375854492188, -5.121255874633789, 29.76654052734375, 8.020416259765625, 25.312744140625, 63.38348388671875, -13.716606140136719, 91.78807830810547, -4.310037612915039, -69.18380737304688, 107.12669372558594, 15.3157958984375, 23.2596435546875, -3.6666107177734375, 103.22569274902344, 85.49990844726562, 12.09271240234375, 97.0089111328125, -48.645530700683594, 24.73699951171875, -11.61474609375, 0.226806640625, -62.676177978515625, 68.19046020507812, 15.695419311523438, 39.65412902832031, 27.641693115234375, 73.13375091552734, 115.080810546875, 24.631614685058594, -48.605224609375, 35.393524169921875, 91.5699462890625, 8.8189697265625, 12.334617614746094, 0.3829689025878906, 111.46295166015625, 8.056610107421875, -12.63360595703125, 81.0367431640625, 20.7071533203125, 65.61929321289062, 63.222633361816406, -10.172332763671875, 97.31185913085938, 21.29827880859375, 87.95201110839844, 90.70602416992188, 12.5921630859375, 70.77055358886719, 50.651123046875, 20.17138671875, -8.430191040039062, 113.87986755371094, 32.42132568359375, 30.636878967285156, 2.316568374633789, 66.69448852539062, -59.69807434082031, 77.4093017578125, 39.25262451171875, 85.13578796386719, 72.90623474121094, 53.164337158203125, -2.359893798828125, 6.362701416015625, 5.091583251953125, -3.251708984375, 71.77748107910156, -67.69590759277344, 47.7098388671875, -7.7727508544921875, 35.373046875, 111.57196044921875, 3.2519760131835938, 19.128807067871094, -2.3193359375, -8.73663330078125, 49.32157897949219, 1.9383544921875, 69.24713134765625, -15.82080078125, 31.1697998046875, -13.0452880859375, 131.77359008789062, 16.438987731933594, -8.263565063476562, -8.024383544921875, 9.997337341308594, 52.623130798339844, 61.456085205078125, 27.669036865234375, 83.01211547851562, 90.33053588867188, 7.86846923828125, 8.552825927734375, -14.304656982421875, 75.7923583984375, 6.897026062011719, 2.2920379638671875, 12.826919555664062, 94.94088745117188, 78.70246887207031, -30.10186767578125, 94.38218688964844, -14.01654052734375, 95.36448669433594, 60.50975036621094, 41.32175827026367], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000476.npy"}
{"epoch": 0.9968586387434555, "step": 477, "batch_size": 128, "mean": 33.78434753417969, "std": 49.391597747802734, "min": -105.47885131835938, "p10": -28.23416290283203, "median": 31.14937973022461, "p90": 98.5984878540039, "max": 136.12652587890625, "pos_frac": 0.7109375, "sample": [-39.532867431640625, 14.871978759765625, 76.86512756347656, -44.997314453125, 83.78173828125, 30.636871337890625, -33.41766357421875, 64.06169891357422, -19.845596313476562, 44.50871276855469, -7.668039321899414, 97.95462036132812, 2.5916595458984375, 62.376930236816406, 47.44903564453125, -12.28485107421875, 107.71548461914062, 57.223114013671875, -105.47885131835938, -3.768798828125, 14.635406494140625, -29.43548583984375, 49.11268615722656, 106.62802124023438, 103.52717590332031, 77.5986328125, 61.5216064453125, -3.9804344177246094, 2.975494384765625, 29.0723876953125, 83.93128967285156, 87.11318969726562, 8.011093139648438, 124.19241333007812, 21.433135986328125, 13.455078125, 106.9547119140625, 23.460922241210938, -7.4179840087890625, 0.0, 12.491561889648438, 69.521240234375, 5.13824462890625, -22.355010986328125, 51.17499542236328, -2.6742095947265625, 20.93316650390625, 80.77885437011719, 85.910888671875, -27.797882080078125, -17.014259338378906, -2.3076324462890625, -18.556777954101562, 75.76251220703125, 14.362152099609375, 18.377166748046875, 77.1181640625, 23.047088623046875, 77.1558837890625, 33.36924743652344, 71.332763671875, 106.03421020507812, 131.85302734375, 1.604644775390625, 42.61383056640625, 64.74520111083984, 102.32855224609375, -27.321990966796875, 39.79082489013672, -37.19657897949219, 31.030113220214844, 78.13006591796875, 53.51853942871094, 67.9091796875, 96.34539794921875, 100.10084533691406, -29.252151489257812, -1.7805328369140625, 0.8646278381347656, 77.94741821289062, -15.5517578125, 96.81521606445312, 36.99395751953125, -12.529182434082031, -30.590133666992188, -76.4232406616211, 82.95915222167969, 58.22298049926758, 30.439048767089844, 59.475494384765625, -4.959956169128418, 11.083099365234375, -5.341747283935547, 104.07211303710938, -10.65142822265625, 77.79074096679688, -48.22875213623047, 72.13137817382812, 36.747467041015625, 92.22139739990234, -26.740219116210938, 66.3829345703125, -36.1710205078125, -89.42617797851562, 10.91461181640625, -6.45367431640625, -29.625457763671875, -4.88031005859375, 113.81499481201172, 47.47761535644531, 31.268646240234375, 136.12652587890625, 71.77587890625, 107.05300903320312, 10.290573120117188, 80.49443817138672, 65.00146484375, -25.91937255859375, 94.67003631591797, 17.246002197265625, 46.935546875, 70.33795166015625, 83.78410339355469, 16.424270629882812, 0.6270332336425781, 55.07837677001953, 36.89031982421875, 5.4770050048828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.6-20260430-165125/margin_logs/step_0000477.npy"}