{"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-0.35-20260430-143919/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-0.35-20260430-143919/margin_logs/step_0000002.npy"} {"epoch": 0.004188481675392671, "step": 3, "batch_size": 128, "mean": 0.08885043859481812, "std": 0.8816279172897339, "min": -2.45379638671875, "p10": -0.9426879882812499, "median": 0.10031700134277344, "p90": 1.1569900512695308, "max": 3.13323974609375, "pos_frac": 0.5546875, "sample": [0.53857421875, 0.17107009887695312, -0.1837158203125, -0.2694511413574219, 0.2940826416015625, -0.7346649169921875, 1.50946044921875, -0.22735595703125, 0.0, -0.829620361328125, 1.588134765625, -0.918975830078125, -0.551727294921875, 2.294769287109375, -0.42236328125, 0.208099365234375, -1.032806396484375, 1.241851806640625, 0.8358001708984375, 0.636016845703125, -0.6342010498046875, -0.327972412109375, 0.32525634765625, 0.22795867919921875, 0.1031494140625, -0.12957763671875, 1.4718017578125, -2.45379638671875, 0.79443359375, 0.057586669921875, -0.11486434936523438, 0.08338546752929688, 0.745269775390625, 0.912567138671875, 1.27581787109375, -0.24466705322265625, -0.11798095703125, -0.77899169921875, -0.54766845703125, 1.52008056640625, -0.1612548828125, -0.8875732421875, 0.749969482421875, 0.3801422119140625, 0.805389404296875, -1.478729248046875, -0.5308685302734375, 1.4685821533203125, -0.30419921875, -1.147308349609375, -0.45599365234375, 0.22867965698242188, 0.480316162109375, 0.10770988464355469, 0.0178680419921875, 1.872772216796875, 0.44261932373046875, 2.65814208984375, 0.047420501708984375, -0.12381362915039062, 0.0, -1.834075927734375, 3.13323974609375, -0.998016357421875, 0.33798980712890625, -0.1610260009765625, -1.854339599609375, -1.548980712890625, 0.09748458862304688, -0.47742462158203125, -1.185546875, 0.176666259765625, 1.529144287109375, -0.766876220703125, -0.08292388916015625, -0.064453125, 0.56243896484375, 0.111541748046875, -0.00592041015625, -0.417999267578125, 0.1736297607421875, 0.1877593994140625, 0.9146995544433594, 0.19770050048828125, -0.874481201171875, 0.336700439453125, -1.187835693359375, 0.402679443359375, -0.16363525390625, 0.5542449951171875, 0.262420654296875, 0.0735321044921875, -0.6881103515625, -1.65850830078125, 0.9573478698730469, 0.4842376708984375, 0.12740325927734375, 0.37413787841796875, 0.7090339660644531, 0.30637359619140625, -0.5127372741699219, 0.543304443359375, -0.178009033203125, -0.36224365234375, 0.5904388427734375, 0.9637451171875, -0.1779632568359375, 0.5218048095703125, 0.346832275390625, -0.47818756103515625, -1.41534423828125, 0.010284423828125, 1.1206207275390625, 0.229644775390625, 0.55712890625, 0.45404815673828125, 0.3885040283203125, -0.605712890625, -0.751312255859375, 0.17203521728515625, 1.92279052734375, -0.3887176513671875, 0.49600982666015625, -0.1349315643310547, -1.3358993530273438, 0.569976806640625, 0.43817901611328125, -0.136322021484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000003.npy"} {"epoch": 0.0062827225130890054, "step": 4, "batch_size": 128, "mean": 0.05941309034824371, "std": 1.022767424583435, "min": -3.710968017578125, "p10": -1.1701721191406247, "median": 0.10162353515625, "p90": 1.2428192138671874, "max": 2.9666748046875, "pos_frac": 0.5546875, "sample": [-0.15871429443359375, 0.456878662109375, -0.12396812438964844, 2.2205810546875, 0.4649658203125, -0.0136260986328125, 0.226348876953125, -0.23785400390625, 0.7530517578125, 0.82293701171875, -0.6121978759765625, 1.27508544921875, -0.0645904541015625, 0.023162841796875, 0.3500862121582031, 0.0883636474609375, 1.6765899658203125, -3.710968017578125, 0.2060546875, -0.911529541015625, -1.50567626953125, 0.864501953125, 1.3673095703125, 0.467193603515625, -1.567169189453125, 0.3066253662109375, 2.9666748046875, 0.08849716186523438, 0.4691162109375, -0.9968032836914062, -1.6855621337890625, 0.05192756652832031, 0.848297119140625, -1.5905914306640625, 0.3803253173828125, -1.3460693359375, 0.106048583984375, -1.911346435546875, -0.5634613037109375, 0.557220458984375, -0.23996734619140625, 0.523681640625, -0.4709148406982422, -0.6162109375, -1.9794921875, 0.097198486328125, 0.35211181640625, 0.3524322509765625, 1.247711181640625, -0.09661865234375, 1.03729248046875, 0.184906005859375, -0.847076416015625, -0.945587158203125, 0.0843353271484375, -0.4289054870605469, -0.775665283203125, 1.24072265625, 0.8084716796875, -0.060150146484375, -0.3004150390625, -1.09478759765625, 0.5475540161132812, -0.34857177734375, 0.2484130859375, -1.0152587890625, 0.9673919677734375, -0.67852783203125, -0.5224990844726562, -0.827423095703125, -0.1815185546875, 0.44158935546875, 0.7222900390625, -0.22357177734375, 0.07037353515625, -0.14678955078125, -3.13421630859375, -0.2216796875, 0.39959716796875, -0.24983978271484375, 1.1405715942382812, -0.01275634765625, 0.143280029296875, -1.842193603515625, 0.4403533935546875, 1.390777587890625, 0.6965103149414062, 0.99774169921875, 1.6324462890625, -0.224517822265625, -0.20301055908203125, -1.07476806640625, -0.7112884521484375, 0.6149444580078125, 2.870697021484375, 1.0760040283203125, 0.126708984375, -0.18267440795898438, -2.0147705078125, 0.9620361328125, 0.301177978515625, 0.1348419189453125, 0.5315437316894531, 1.213165283203125, 0.7070770263671875, -1.82318115234375, 0.2220001220703125, -0.16632080078125, 1.51348876953125, 0.980712890625, -0.02616119384765625, 0.434051513671875, 0.4456787109375, 1.61712646484375, 0.450439453125, 0.953094482421875, -1.398681640625, 0.9410858154296875, -0.13336181640625, -0.449066162109375, -0.0148468017578125, -0.6605224609375, 1.32708740234375, 0.43378448486328125, -0.54766845703125, 0.547882080078125, 1.3165283203125, -1.0302734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000004.npy"} {"epoch": 0.008376963350785341, "step": 5, "batch_size": 128, "mean": -0.036664366722106934, "std": 0.8168643712997437, "min": -2.8267822265625, "p10": -1.10833740234375, "median": 0.03166675567626953, "p90": 0.9103942871093749, "max": 2.2431640625, "pos_frac": 0.515625, "sample": [0.05667877197265625, -1.0869140625, 0.4953155517578125, -0.0672607421875, 0.9794082641601562, 1.2021026611328125, -0.432403564453125, -0.93353271484375, 0.3717041015625, -0.194580078125, 0.861328125, 0.806793212890625, -0.6183013916015625, -0.8046493530273438, 0.06528472900390625, 1.610382080078125, 0.6420326232910156, -0.15087890625, 1.3707275390625, -2.8267822265625, -1.200103759765625, 0.1422119140625, -0.5105133056640625, -0.5238037109375, -0.802398681640625, 0.7735595703125, -0.1292266845703125, 0.668853759765625, 0.26810455322265625, -0.06362533569335938, 0.99365234375, -0.7055511474609375, -0.300445556640625, 0.03656005859375, 0.3386383056640625, -1.3992462158203125, 0.258514404296875, -0.59820556640625, 0.237152099609375, 1.59857177734375, 0.75091552734375, 1.26129150390625, 1.37091064453125, 0.456085205078125, -0.0218505859375, -0.16578102111816406, 0.146148681640625, 0.428802490234375, 0.03412818908691406, -0.878662109375, 0.77447509765625, -0.574615478515625, -1.1583251953125, -0.05634498596191406, 0.649261474609375, 0.095123291015625, 0.5640869140625, -0.093780517578125, 0.89459228515625, 0.68536376953125, 0.438720703125, -1.414459228515625, 0.457275390625, -0.17087554931640625, 0.25677490234375, -0.628692626953125, -0.20623779296875, -1.40911865234375, -0.0074462890625, -0.5102691650390625, 0.5941162109375, -0.114166259765625, 0.5008392333984375, 0.39333343505859375, 0.055828094482421875, -0.388824462890625, -1.233062744140625, -0.525146484375, 0.4622802734375, 1.1452255249023438, -0.806671142578125, 2.2431640625, -0.6024169921875, -2.080841064453125, 0.47589111328125, -1.245849609375, 0.49090576171875, -0.228118896484375, -0.245819091796875, 0.1846466064453125, -1.199066162109375, 0.07099342346191406, -2.30572509765625, -0.271392822265625, -0.387664794921875, -0.5475616455078125, -1.838287353515625, -0.3292388916015625, -0.3852691650390625, 0.467559814453125, -0.62506103515625, 0.353973388671875, 0.2364501953125, -0.587890625, 0.40555572509765625, -2.064056396484375, -0.8121337890625, 0.10920333862304688, 0.1836395263671875, 0.947265625, 0.25958251953125, -0.49724578857421875, 1.0476303100585938, 1.169036865234375, -0.1384735107421875, 0.19134902954101562, -0.0488739013671875, -0.2730865478515625, 0.780670166015625, 0.34381103515625, 0.029205322265625, 0.3485260009765625, 0.351531982421875, 0.008327484130859375, -0.8946533203125, 0.40362548828125, -0.18151092529296875, -0.485748291015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000005.npy"} {"epoch": 0.010471204188481676, "step": 6, "batch_size": 128, "mean": 0.015265822410583496, "std": 0.9207139611244202, "min": -3.13079833984375, "p10": -1.020806884765625, "median": -0.020893096923828125, "p90": 1.1086181640625, "max": 3.592193603515625, "pos_frac": 0.484375, "sample": [-0.1754913330078125, -0.2113037109375, 0.6547927856445312, -0.2462158203125, -0.799530029296875, 0.9868621826171875, 0.43255615234375, 0.16633987426757812, -0.122467041015625, -0.6749725341796875, -0.746124267578125, 1.978668212890625, -1.0569305419921875, -0.317352294921875, -1.4079132080078125, -0.8327789306640625, 0.19573974609375, -0.040637969970703125, -0.21983718872070312, 0.6234130859375, 0.170166015625, 0.13270187377929688, 0.216217041015625, 0.8211669921875, -0.8401947021484375, -0.1226959228515625, -0.11090087890625, -0.0040283203125, 0.80706787109375, 0.2712726593017578, -0.47017669677734375, -0.2829437255859375, 0.6313323974609375, 0.5471038818359375, -0.4570770263671875, 0.424407958984375, -0.70068359375, -0.66461181640625, -0.1404571533203125, 1.82904052734375, -0.3012847900390625, -1.54901123046875, -0.56256103515625, -1.02777099609375, 0.0924530029296875, -0.85443115234375, 0.06906700134277344, 1.598419189453125, 0.42263031005859375, 0.36517333984375, 0.398895263671875, 1.08880615234375, 1.9693603515625, -0.053924560546875, -0.4892578125, 0.34438323974609375, -0.1534423828125, -0.06097412109375, -0.9791259765625, 0.886962890625, 0.2732505798339844, -0.72711181640625, 0.9790802001953125, 1.0687942504882812, -0.89764404296875, 0.2197265625, 0.20477294921875, 1.648834228515625, 1.25299072265625, 0.2144317626953125, 0.1234283447265625, -0.05059814453125, 1.780731201171875, 0.58642578125, 0.5156402587890625, 0.0561370849609375, 0.39810943603515625, -1.0700225830078125, 0.0, 0.078826904296875, 0.49433135986328125, -0.498870849609375, -0.5047225952148438, -0.31060791015625, -0.2607879638671875, -0.1408233642578125, 2.1806793212890625, 0.8464813232421875, 0.1866455078125, -0.35590362548828125, 0.124847412109375, 0.04998779296875, 0.8857536315917969, -1.313629150390625, 1.15484619140625, -0.1216888427734375, 0.150390625, -0.29572296142578125, -0.736175537109375, -0.42266845703125, 1.00921630859375, -1.42572021484375, -1.1135406494140625, 1.58349609375, -0.6278457641601562, -0.4191436767578125, -1.13543701171875, -2.533935546875, -0.34783935546875, -0.26613616943359375, -0.56024169921875, -0.03775787353515625, 3.592193603515625, 0.077850341796875, 0.17696380615234375, 1.2591552734375, -1.017822265625, -0.099456787109375, -2.173095703125, 0.28607177734375, 0.20711898803710938, -3.13079833984375, -0.5008087158203125, -1.0303955078125, 1.37811279296875, -0.81744384765625, 0.22955322265625, 0.17565536499023438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000006.npy"} {"epoch": 0.012565445026178011, "step": 7, "batch_size": 128, "mean": -0.03379000723361969, "std": 0.965813398361206, "min": -2.73565673828125, "p10": -1.2207626342773437, "median": -0.004405975341796875, "p90": 1.048095703125, "max": 2.82476806640625, "pos_frac": 0.4921875, "sample": [0.8660888671875, 1.7061767578125, -1.568206787109375, -0.721893310546875, 1.0477752685546875, 0.0843048095703125, 0.4464874267578125, -0.9554443359375, -1.556640625, -0.1996917724609375, 0.5574569702148438, 0.372039794921875, -2.02716064453125, -0.21826934814453125, 1.0488433837890625, -0.80181884765625, 0.5485076904296875, 0.6301422119140625, 0.0, -0.028350830078125, 0.095947265625, -1.3461761474609375, 0.03284454345703125, -1.0258331298828125, 0.190887451171875, -0.12042236328125, -0.19818115234375, -0.1794891357421875, -0.407958984375, -1.051605224609375, -0.980499267578125, 0.6754150390625, -0.3559074401855469, 0.4888763427734375, -0.506561279296875, 2.82476806640625, 0.93756103515625, -0.268463134765625, -1.65081787109375, 0.4398651123046875, -0.821044921875, 0.08172607421875, 0.002899169921875, 0.668304443359375, 0.75439453125, 0.5843505859375, 0.575531005859375, 0.22356033325195312, 0.103424072265625, 0.98382568359375, 1.27471923828125, -1.860198974609375, 0.840667724609375, -0.492584228515625, 1.40069580078125, 0.1819610595703125, -0.451629638671875, -0.130645751953125, -1.0563888549804688, 1.18707275390625, -0.49530029296875, 2.68817138671875, 0.5608997344970703, -1.62451171875, -1.3953857421875, 0.985015869140625, -0.76019287109375, 0.0912933349609375, 0.769195556640625, -1.081390380859375, -0.3912353515625, 0.53961181640625, -0.080108642578125, -0.27880859375, 0.35821533203125, -0.807403564453125, -0.35352325439453125, -0.87139892578125, -1.9937744140625, 0.012359619140625, -0.177886962890625, 0.0541839599609375, 1.149322509765625, 0.49211883544921875, 0.30877685546875, 0.8385848999023438, -1.0203857421875, -0.601226806640625, 0.596405029296875, -0.00881195068359375, -2.73565673828125, -0.4225311279296875, 0.030120849609375, -0.3977241516113281, -1.2241058349609375, -0.969207763671875, -0.2751579284667969, 0.09364128112792969, -0.215667724609375, -1.219329833984375, 0.03692626953125, -0.3381500244140625, -0.6474609375, 2.1143798828125, 0.28216552734375, 0.555267333984375, 0.85235595703125, 0.937744140625, -0.119384765625, -1.17840576171875, 0.9312591552734375, -0.6318626403808594, -0.5360107421875, 2.3125152587890625, -0.9650115966796875, -0.6511993408203125, -0.90521240234375, 0.3367919921875, 0.63458251953125, 1.37811279296875, -1.63323974609375, 0.20196533203125, -0.09279632568359375, -0.21559715270996094, 0.55084228515625, -1.7215576171875, 1.780853271484375, 1.362579345703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000007.npy"} {"epoch": 0.014659685863874346, "step": 8, "batch_size": 128, "mean": -0.1919558048248291, "std": 1.1286702156066895, "min": -4.306182861328125, "p10": -1.4724670410156249, "median": -0.1143951416015625, "p90": 1.0879291534423827, "max": 2.97308349609375, "pos_frac": 0.4140625, "sample": [-1.504852294921875, -0.08583831787109375, 0.020965576171875, -1.15240478515625, -0.15966796875, -2.36016845703125, -0.218017578125, -0.2728843688964844, 0.1571216583251953, -2.7926025390625, -0.3854713439941406, -2.652740478515625, -0.093505859375, 2.1561279296875, -0.0289764404296875, -0.3961029052734375, -0.61541748046875, -1.4560546875, -0.0042438507080078125, 0.06304931640625, 0.07635498046875, 2.06298828125, -1.21282958984375, -1.27734375, -0.5486297607421875, -0.117523193359375, -0.458831787109375, 0.195098876953125, -1.252777099609375, -2.077880859375, -1.2928466796875, 2.97308349609375, 0.30352783203125, -1.16387939453125, 0.81524658203125, -1.26239013671875, 0.47754669189453125, -0.0389862060546875, -0.6451416015625, -2.20440673828125, 2.2545166015625, -0.76800537109375, 0.185699462890625, -0.925567626953125, 0.7318878173828125, 0.6374664306640625, 0.32984161376953125, -0.7637481689453125, -1.545196533203125, -0.8475341796875, -0.9607696533203125, -0.6401519775390625, 0.7828369140625, 0.3701152801513672, -0.7188568115234375, -0.5225982666015625, -0.51678466796875, 0.13525390625, -0.878326416015625, -4.306182861328125, -0.853240966796875, -1.35870361328125, -1.6329345703125, 1.952789306640625, 0.0164337158203125, -0.15771484375, -1.7755889892578125, -0.151763916015625, -2.37445068359375, 2.64031982421875, -0.66668701171875, 0.4920654296875, -0.6298980712890625, -0.02071380615234375, 0.479736328125, -0.11126708984375, 0.986419677734375, 0.2904548645019531, 0.1416645050048828, -1.10693359375, 0.18377304077148438, 0.46307373046875, -1.52630615234375, 0.08854293823242188, 0.4963836669921875, 0.2994270324707031, -1.62603759765625, -0.803131103515625, 0.006969451904296875, 1.172698974609375, 0.599700927734375, -1.458587646484375, 1.2169952392578125, 0.39056396484375, -0.21192169189453125, 0.6103439331054688, 0.67852783203125, 1.53692626953125, 1.1050376892089844, -0.079833984375, 0.5254287719726562, 0.14813232421875, -0.323089599609375, 2.796478271484375, -0.3594512939453125, -0.07380294799804688, 0.02923583984375, 0.5882492065429688, -0.7672119140625, -1.1697998046875, -0.12712478637695312, -0.07061767578125, 1.3199462890625, -0.1422119140625, 0.071136474609375, 0.3336334228515625, 1.080596923828125, -1.278350830078125, -0.08095550537109375, -0.21783447265625, -0.5983428955078125, -1.1002044677734375, -0.1575927734375, -0.71563720703125, -0.26686859130859375, 0.31197357177734375, 2.530303955078125, 0.2379150390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000008.npy"} {"epoch": 0.016753926701570682, "step": 9, "batch_size": 128, "mean": 0.11899170279502869, "std": 0.964026153087616, "min": -2.5528564453125, "p10": -1.0387374877929687, "median": 0.10322189331054688, "p90": 1.2296134948730466, "max": 3.359130859375, "pos_frac": 0.5546875, "sample": [-0.67645263671875, 0.4395751953125, -0.5167083740234375, 0.10644912719726562, 0.639312744140625, -0.90289306640625, 0.44757080078125, -0.2566108703613281, 1.59136962890625, 0.21530914306640625, 0.2884674072265625, -1.41021728515625, 0.149444580078125, -1.173797607421875, -2.3419342041015625, 0.2841796875, 0.87841796875, 0.0978240966796875, 1.1907424926757812, -0.2242431640625, 0.6226882934570312, 2.7742919921875, 0.04036712646484375, 0.90972900390625, 0.10823822021484375, -0.2916259765625, -0.7183837890625, 1.4143524169921875, 0.512176513671875, -2.5528564453125, 0.679779052734375, -0.56927490234375, -0.098541259765625, -1.4857940673828125, -0.9605560302734375, -0.17710113525390625, 1.32244873046875, -0.04079437255859375, 0.5942764282226562, 0.4328765869140625, -1.0218048095703125, 0.2720947265625, 0.393157958984375, 0.347686767578125, 1.39984130859375, -0.508331298828125, -1.817535400390625, -2.019989013671875, -0.8079833984375, 0.001983642578125, 0.444305419921875, -0.539947509765625, 0.2020263671875, 0.02142333984375, -0.20953369140625, 0.5272750854492188, -0.64697265625, 1.140472412109375, -0.86260986328125, -0.992401123046875, 0.485198974609375, -0.2095489501953125, -1.0782470703125, -0.015987396240234375, 0.7502212524414062, 1.5355224609375, 0.38440704345703125, 1.02020263671875, -0.24908447265625, 0.8847579956054688, -0.14202880859375, -0.011175155639648438, 0.7247314453125, 0.025737762451171875, 0.68133544921875, -0.019073486328125, -1.128509521484375, 0.2624053955078125, 0.2737083435058594, 0.3838043212890625, 0.959228515625, -0.037384033203125, -0.3434600830078125, 0.488922119140625, 0.2322845458984375, -0.2880706787109375, 0.0, -0.7717742919921875, 0.404693603515625, -1.216339111328125, 1.9801025390625, 0.7883033752441406, 0.3770599365234375, 0.5534515380859375, 0.8583526611328125, 0.40172576904296875, -0.47174072265625, -0.79412841796875, -0.09703826904296875, 1.3203125, 0.3137550354003906, 0.0, 1.95623779296875, 0.42352294921875, -0.22271728515625, -0.08988761901855469, 0.0, 1.1510009765625, -0.22211074829101562, -0.08970260620117188, 1.426361083984375, 0.6092147827148438, -0.699127197265625, -1.3663330078125, -1.754638671875, 0.09999465942382812, -0.6540374755859375, 0.394561767578125, -0.111053466796875, 0.5694351196289062, 0.2566375732421875, 3.359130859375, 0.04062652587890625, 0.477508544921875, -1.4171295166015625, -0.238037109375, 2.563720703125, 2.8858642578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000009.npy"} {"epoch": 0.018848167539267015, "step": 10, "batch_size": 128, "mean": -0.10534200072288513, "std": 0.8764756321907043, "min": -3.13922119140625, "p10": -1.1974609375, "median": -0.0786590576171875, "p90": 0.9298507690429686, "max": 2.81317138671875, "pos_frac": 0.4453125, "sample": [-1.0848388671875, 0.0548248291015625, 0.03302001953125, -0.220184326171875, 0.6813888549804688, -1.512939453125, -0.3487815856933594, 0.063140869140625, 0.06655120849609375, -0.0860443115234375, 0.914886474609375, -0.014522552490234375, -0.2057952880859375, 1.95556640625, -1.0894775390625, -0.505950927734375, 0.13964080810546875, -1.67608642578125, 0.3733062744140625, 0.9647674560546875, -1.2001953125, -0.08824920654296875, -0.1617584228515625, -2.155517578125, 0.3077850341796875, -0.33191680908203125, 1.9039306640625, 0.166961669921875, -1.286651611328125, 0.474090576171875, -0.47568511962890625, 0.1754150390625, -0.009674072265625, 0.9803009033203125, 0.019298553466796875, -0.2860260009765625, 1.138580322265625, -0.6389007568359375, 1.08929443359375, 0.2203826904296875, -0.2965087890625, -0.20972442626953125, -0.23397254943847656, 0.39697265625, -0.503631591796875, 0.25274658203125, -0.3615760803222656, -0.39654541015625, 0.29376220703125, -0.53948974609375, 0.1441822052001953, -1.0931625366210938, 1.84228515625, -1.1962890625, -0.5072021484375, -0.0712432861328125, -1.30078125, 1.7754058837890625, 0.2958564758300781, 0.07720947265625, -0.630889892578125, -0.6304969787597656, 0.6131591796875, -0.16894149780273438, -0.0390625, -0.824249267578125, 0.0753021240234375, -0.36083984375, -0.1333465576171875, 0.0531005859375, -0.082122802734375, 0.0914764404296875, 1.25860595703125, -0.2326507568359375, -0.11824798583984375, 0.41141510009765625, 0.01288604736328125, 0.446929931640625, -0.3424072265625, 2.81317138671875, -0.34406280517578125, -0.4210205078125, -0.4642333984375, 0.510284423828125, 0.19422531127929688, 0.225616455078125, 0.03662109375, -0.13899993896484375, -0.019287109375, -1.3275146484375, 0.080841064453125, 0.03981590270996094, 0.35842132568359375, 0.0001220703125, -1.929046630859375, 0.6299285888671875, -0.2516632080078125, 0.25616455078125, -0.9936370849609375, 0.0044403076171875, -0.2915496826171875, -0.5601539611816406, 0.0199737548828125, -0.0751953125, -0.819091796875, -0.326873779296875, -1.5214004516601562, -1.618255615234375, 1.4578323364257812, 0.18218994140625, -0.950408935546875, -0.26180458068847656, -0.3066558837890625, 0.0, 0.447357177734375, -3.13922119140625, -0.18738174438476562, -1.864990234375, 0.07402801513671875, -0.557342529296875, 1.86932373046875, -0.331817626953125, 2.5634765625, 0.1142120361328125, -0.25592041015625, -1.464324951171875, -1.083984375, 0.03216552734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000010.npy"} {"epoch": 0.020942408376963352, "step": 11, "batch_size": 128, "mean": 0.07062935829162598, "std": 0.9957197308540344, "min": -2.25714111328125, "p10": -0.9902389526367187, "median": 0.00646209716796875, "p90": 1.1907287597656244, "max": 3.722900390625, "pos_frac": 0.5, "sample": [-0.0870361328125, -1.458251953125, -0.13507080078125, 0.8006591796875, -0.564239501953125, -0.07503509521484375, 0.9345703125, 0.434661865234375, -0.7061767578125, 0.422210693359375, -0.38961029052734375, 0.0915679931640625, 1.036529541015625, -0.12506103515625, -0.1105499267578125, -0.26605987548828125, -0.812042236328125, 0.0564727783203125, -1.475341796875, 0.39591217041015625, 0.36944007873535156, 0.8477935791015625, 1.7772216796875, 2.654296875, -0.9663848876953125, 0.09783935546875, 0.8775482177734375, 0.2819366455078125, 1.13409423828125, -0.9034576416015625, -1.212310791015625, -0.7752685546875, -0.25185394287109375, -0.87225341796875, 0.2664337158203125, 2.210052490234375, -1.689117431640625, 0.35926055908203125, 0.0347747802734375, 0.3525352478027344, 0.0780029296875, -1.790985107421875, -2.10675048828125, -1.34014892578125, -0.069610595703125, 1.397735595703125, -0.26619720458984375, 1.8152923583984375, 3.722900390625, 1.3228759765625, 0.4301910400390625, -1.0458984375, -0.0363006591796875, -2.25714111328125, -0.44158935546875, 0.11578369140625, -0.5631866455078125, 0.88531494140625, 1.562286376953125, 0.4473724365234375, -0.25360107421875, 0.1263427734375, -0.957244873046875, -0.007293701171875, 0.061553955078125, -0.2266674041748047, 0.9777984619140625, -0.07320213317871094, 0.076141357421875, -0.2703895568847656, 0.9962158203125, -0.5405807495117188, -0.231781005859375, -0.4464874267578125, 0.24930763244628906, 0.439208984375, -0.1382904052734375, -0.28167724609375, 0.652313232421875, 2.3514404296875, 0.46075439453125, -0.45159912109375, -0.3365325927734375, -0.34229278564453125, -1.5218963623046875, 0.7684898376464844, -0.03148651123046875, 0.174896240234375, 0.7529296875, -0.304595947265625, -0.381256103515625, -0.8062210083007812, -1.846527099609375, 0.6863365173339844, -1.5941619873046875, -0.95184326171875, 0.7130126953125, -0.62518310546875, 0.1519927978515625, -0.2978630065917969, -0.3711662292480469, 0.357208251953125, 3.38916015625, -0.09234619140625, 0.04058837890625, 0.3607177734375, -0.736297607421875, -0.163787841796875, 0.486175537109375, 0.5732803344726562, 0.2881622314453125, -0.4904937744140625, -0.183349609375, -1.4581527709960938, 0.0202178955078125, -0.932281494140625, 0.797393798828125, -0.46075439453125, -0.870361328125, 1.85113525390625, 1.62164306640625, 0.79461669921875, 0.8712615966796875, 0.1219940185546875, 0.3323974609375, 0.1303558349609375, -0.0636749267578125, 1.68621826171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000011.npy"} {"epoch": 0.023036649214659685, "step": 12, "batch_size": 128, "mean": -0.05256234109401703, "std": 0.9185609221458435, "min": -3.2359619140625, "p10": -1.2036224365234374, "median": -0.0621337890625, "p90": 1.0080528259277344, "max": 2.65570068359375, "pos_frac": 0.4765625, "sample": [-1.227783203125, 0.68988037109375, 2.073272705078125, 1.0325927734375, 0.0008106231689453125, -1.0513534545898438, -1.8298797607421875, 0.8196258544921875, -0.22943115234375, -0.2457733154296875, -0.8721923828125, 0.7018890380859375, -1.4195556640625, 0.9975357055664062, 0.49310302734375, -0.057769775390625, 0.2490081787109375, -0.191558837890625, 0.513671875, 0.8278961181640625, -1.487640380859375, -1.1715087890625, 2.65570068359375, -0.7056732177734375, -0.4017333984375, 0.161376953125, -0.3271446228027344, 0.5803108215332031, -0.406158447265625, 0.6506805419921875, -0.886688232421875, 0.30324554443359375, -0.5270538330078125, -2.15045166015625, -0.7613525390625, 1.403472900390625, 0.0212249755859375, -0.4963226318359375, -1.02191162109375, 0.16741943359375, -0.23207855224609375, 0.21533203125, -1.43414306640625, 0.9406280517578125, -0.1798248291015625, -0.1855316162109375, -0.30584716796875, 0.5786361694335938, -0.528594970703125, -0.5830078125, 1.284149169921875, -0.00689697265625, 0.072845458984375, -0.195556640625, -0.34401702880859375, -0.185455322265625, 0.76708984375, -1.193267822265625, -1.6488189697265625, 1.0520172119140625, 0.0107421875, 0.41107177734375, -0.066497802734375, -0.7474365234375, -0.3992118835449219, 0.97515869140625, 0.1934967041015625, 1.39300537109375, -0.07732391357421875, 0.455963134765625, 0.27097320556640625, -1.7423858642578125, 0.8894577026367188, -0.6781005859375, 0.10924530029296875, 0.320281982421875, 0.43875885009765625, -0.3873291015625, -0.585052490234375, -0.220977783203125, 1.60321044921875, 0.2357330322265625, -0.32183837890625, -0.35041046142578125, 0.4140892028808594, 0.36016845703125, 0.06432342529296875, 0.619049072265625, -3.2359619140625, -0.9334716796875, 1.428009033203125, 0.3447914123535156, -0.65496826171875, -0.711822509765625, -0.63336181640625, 0.01812744140625, -0.75909423828125, -0.16374969482421875, -1.06170654296875, 0.124664306640625, -0.820526123046875, -0.485260009765625, -0.582763671875, 0.91876220703125, -0.28790283203125, -1.250823974609375, 0.6912612915039062, 0.460845947265625, 0.006439208984375, 1.43316650390625, 0.1724853515625, -0.04107666015625, -0.17531585693359375, 0.3036651611328125, 2.348175048828125, -0.61474609375, -0.23116683959960938, -0.13018798828125, -1.82806396484375, 0.3612213134765625, 1.92974853515625, -1.37786865234375, 0.674224853515625, -0.999114990234375, 0.703704833984375, -1.279449462890625, 0.420135498046875, 1.24139404296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000012.npy"} {"epoch": 0.025130890052356022, "step": 13, "batch_size": 128, "mean": -0.03749191761016846, "std": 0.8850112557411194, "min": -3.23101806640625, "p10": -1.0580032348632813, "median": 0.014776229858398438, "p90": 0.8680633544921874, "max": 3.31817626953125, "pos_frac": 0.5078125, "sample": [-0.4773712158203125, -0.122161865234375, 0.05615234375, -3.23101806640625, 0.8030548095703125, 0.569793701171875, 0.82470703125, -0.588897705078125, 0.3128814697265625, 0.46944427490234375, 0.3885498046875, -1.867095947265625, -0.455108642578125, -0.10737991333007812, 0.11621284484863281, 0.008951187133789062, 0.255340576171875, 0.4747314453125, 0.57220458984375, 0.3491668701171875, -0.18805313110351562, -1.53643798828125, 1.3687744140625, 0.037841796875, 0.7579345703125, -1.017120361328125, -0.06290054321289062, -0.865570068359375, -0.2123870849609375, -1.103668212890625, 0.716583251953125, -0.32842254638671875, 0.15533447265625, -0.4911651611328125, 0.4283905029296875, 0.15839385986328125, 0.18378639221191406, -0.50311279296875, 0.073638916015625, -0.529022216796875, 0.4126110076904297, 1.963531494140625, 0.552886962890625, -0.020172119140625, -0.0588836669921875, 0.1313629150390625, -0.77740478515625, -1.9254150390625, -0.657806396484375, 0.2127685546875, -0.1094207763671875, -1.574859619140625, -0.19657516479492188, 0.2688446044921875, -0.395751953125, 0.0, -1.0567626953125, 0.163238525390625, 0.075408935546875, -1.050506591796875, -0.21289825439453125, -1.636627197265625, -1.0779571533203125, 0.21610069274902344, 1.425140380859375, 0.2320556640625, -0.288909912109375, -0.022674560546875, 0.7274322509765625, 0.020601272583007812, -1.2259521484375, -0.9044952392578125, 0.27730560302734375, 1.9100341796875, -0.0899658203125, 0.37530517578125, -0.13338470458984375, 3.31817626953125, 0.25661468505859375, -0.2089691162109375, 0.13202285766601562, 0.8942337036132812, -0.07973480224609375, 0.147979736328125, -0.05524444580078125, 1.35821533203125, 0.2760505676269531, -0.331878662109375, -0.3006134033203125, -0.7257461547851562, 0.183807373046875, 0.2462615966796875, 1.6049957275390625, -0.6540298461914062, 0.6169586181640625, -0.9370498657226562, 0.0447845458984375, 0.86285400390625, -1.0608978271484375, -0.4061279296875, -0.716156005859375, -1.0795669555664062, -0.7643966674804688, -0.2292327880859375, 0.6032562255859375, -0.24945068359375, 0.56658935546875, 0.88043212890625, 1.08660888671875, 0.406951904296875, 0.08889579772949219, -0.5968017578125, -1.3430404663085938, -0.3223400115966797, 1.5672607421875, 2.71270751953125, -0.626617431640625, -0.9859619140625, 0.880218505859375, -0.40020751953125, -2.129638671875, 0.360504150390625, 0.4558868408203125, 0.10866546630859375, -0.59033203125, 0.037136077880859375, -0.6690826416015625, 0.0229034423828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000013.npy"} {"epoch": 0.027225130890052355, "step": 14, "batch_size": 128, "mean": -0.006156221032142639, "std": 0.8432223200798035, "min": -2.846435546875, "p10": -1.0118682861328125, "median": -0.013641357421875, "p90": 1.0021194458007812, "max": 2.4102783203125, "pos_frac": 0.484375, "sample": [-0.129425048828125, 0.2260589599609375, -0.8121795654296875, 2.1226806640625, 0.01892852783203125, -0.8312835693359375, -1.033935546875, -1.357666015625, 0.13974761962890625, 1.71807861328125, -0.26946258544921875, -0.7353363037109375, -2.846435546875, -0.3818359375, 0.291778564453125, 0.1410675048828125, -1.195465087890625, -2.117340087890625, -0.2679939270019531, 0.0989837646484375, -0.1937713623046875, -0.03160285949707031, 2.12005615234375, 0.50897216796875, 0.343414306640625, -0.05633544921875, -0.629974365234375, 0.37973785400390625, 0.679840087890625, 0.702423095703125, 1.5467529296875, 0.519775390625, -0.59930419921875, -0.3595733642578125, 0.6503829956054688, -1.3743896484375, -0.631805419921875, -0.5158405303955078, 0.338226318359375, -1.532440185546875, 0.04972648620605469, 1.594818115234375, 0.244384765625, 0.628753662109375, 0.0, 0.10486984252929688, -0.0541839599609375, 0.364410400390625, -0.0392913818359375, -0.092529296875, 0.0971221923828125, 0.039947509765625, 0.08134078979492188, -0.0084228515625, 0.04727935791015625, -0.52850341796875, 0.285614013671875, 1.237548828125, -0.563812255859375, -0.34161376953125, 0.22607421875, -1.3924560546875, 1.51104736328125, -0.113433837890625, -0.7008514404296875, -0.0955047607421875, -0.121490478515625, -0.29632568359375, 0.89910888671875, -0.947601318359375, 0.33705902099609375, -0.6275787353515625, -0.4911308288574219, 1.895721435546875, 1.2215576171875, 0.7866744995117188, 0.26811981201171875, 1.4906005859375, 0.927093505859375, -0.36121368408203125, -0.1426849365234375, -0.8430633544921875, 0.231658935546875, -1.3182830810546875, -0.6369781494140625, -0.23348236083984375, -0.127288818359375, -0.19366455078125, -0.8932418823242188, 0.2222747802734375, -0.990966796875, 0.36573028564453125, 0.17425537109375, -0.262054443359375, -1.40631103515625, 0.22021484375, -0.34815406799316406, -1.1561737060546875, 0.9488525390625, -0.02674102783203125, -0.626220703125, -0.223419189453125, 0.590240478515625, 0.2838134765625, -0.040618896484375, 0.1097564697265625, -0.46486663818359375, 0.0669403076171875, 0.9983062744140625, 0.190643310546875, 0.8312225341796875, 1.011016845703125, -1.004241943359375, 0.03479766845703125, 0.35264015197753906, -1.8642578125, 0.3301239013671875, 0.435821533203125, -0.01885986328125, -0.15038681030273438, -1.0296630859375, 0.3461761474609375, -0.294525146484375, -0.418365478515625, -0.10236358642578125, 0.3388671875, 2.4102783203125, 1.3288116455078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000014.npy"} {"epoch": 0.02931937172774869, "step": 15, "batch_size": 128, "mean": -0.037596121430397034, "std": 0.8116958141326904, "min": -3.51824951171875, "p10": -0.8670074462890625, "median": -0.051097869873046875, "p90": 0.8070152282714842, "max": 2.325347900390625, "pos_frac": 0.453125, "sample": [0.5699844360351562, -1.027923583984375, -0.484344482421875, -0.10546875, -0.0579681396484375, 0.06536102294921875, 1.1321868896484375, -0.640655517578125, -0.2861480712890625, -0.26035308837890625, 0.2892875671386719, -0.92034912109375, -1.077484130859375, 0.4335060119628906, -0.04422760009765625, -1.03271484375, -0.6208877563476562, -0.12483024597167969, -0.04370880126953125, -1.3603515625, -0.5938720703125, -0.491455078125, -0.029876708984375, -0.844146728515625, 0.658447265625, 1.5321197509765625, 0.5620880126953125, -2.455078125, -0.160736083984375, -0.36766815185546875, 1.47747802734375, 1.57318115234375, -1.399169921875, -0.279754638671875, 0.483306884765625, -0.111328125, 1.57891845703125, -0.0675048828125, 0.724822998046875, -0.37030792236328125, -0.3255615234375, -1.2254638671875, 0.752716064453125, -0.308563232421875, 0.206939697265625, 0.116668701171875, 0.12982559204101562, 1.4648895263671875, 0.6180267333984375, -0.018402099609375, -0.45155906677246094, 0.280670166015625, 0.2169189453125, 0.212310791015625, -0.27154541015625, 0.423583984375, -0.75531005859375, -0.312713623046875, -0.16837120056152344, -0.326385498046875, 0.21768951416015625, 0.257293701171875, 0.07330322265625, 0.28243255615234375, -0.53570556640625, -0.7804412841796875, 0.8369598388671875, 0.249786376953125, 0.461700439453125, -0.435791015625, -0.37310028076171875, -0.32917022705078125, 0.4631805419921875, -1.45062255859375, 0.0086669921875, 1.3736419677734375, -0.4824066162109375, 0.41796112060546875, -0.605712890625, 0.65130615234375, 0.891021728515625, 0.0030269622802734375, -0.01751708984375, 0.056640625, -0.57330322265625, -0.41448974609375, -0.79949951171875, 0.77740478515625, 0.2144184112548828, 0.3154296875, 0.20428848266601562, -0.0076904296875, 0.343475341796875, -1.018585205078125, 0.38055419921875, 1.889404296875, -0.4239387512207031, -0.58734130859375, -0.7216796875, -0.633544921875, -1.19189453125, 2.325347900390625, 0.60443115234375, 1.54144287109375, 0.16845703125, 0.904571533203125, -3.51824951171875, -0.72119140625, -0.61517333984375, -0.1913909912109375, 0.6914749145507812, 0.066253662109375, -0.21169281005859375, -0.0695953369140625, -0.1968231201171875, -0.14110946655273438, -0.77960205078125, 0.7941818237304688, 0.6129837036132812, 0.131500244140625, -0.313873291015625, -0.18805694580078125, -0.2027130126953125, -1.9615478515625, 0.588104248046875, 0.7867431640625, -0.5134506225585938, 0.528472900390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000015.npy"} {"epoch": 0.031413612565445025, "step": 16, "batch_size": 128, "mean": -0.002897113561630249, "std": 0.9007660150527954, "min": -2.0302734375, "p10": -1.1622070312499997, "median": -0.0079193115234375, "p90": 1.11326904296875, "max": 3.6927490234375, "pos_frac": 0.4921875, "sample": [-0.0980682373046875, -0.0249786376953125, 1.100250244140625, 0.6436309814453125, 0.449676513671875, 0.021392822265625, 0.807037353515625, 1.072845458984375, 0.20867919921875, 0.83880615234375, 0.5212860107421875, -0.2549285888671875, 0.7207794189453125, -0.98138427734375, 0.2509765625, -0.0980072021484375, -0.3985595703125, 0.6243896484375, -1.0616455078125, -0.1206817626953125, 0.4167633056640625, -0.17957305908203125, -0.5890350341796875, 0.194580078125, 0.25555419921875, 0.13743209838867188, 0.94580078125, 1.667572021484375, -0.84759521484375, -0.255035400390625, 1.143646240234375, -0.60089111328125, -1.6044921875, -1.752105712890625, -1.89508056640625, -0.964141845703125, 3.6927490234375, -0.599456787109375, -0.13850784301757812, -0.19977188110351562, 0.6271209716796875, -0.5075836181640625, 0.358062744140625, 0.6944580078125, -0.741119384765625, -0.14478111267089844, -1.5654296875, -0.8779754638671875, 1.198211669921875, 0.03125, 0.025783538818359375, 1.228729248046875, 0.6453704833984375, -1.102752685546875, -0.63519287109375, -0.504638671875, 0.39881134033203125, 0.07134246826171875, 0.49268341064453125, 0.0560302734375, 0.368804931640625, 0.06943130493164062, 0.1367645263671875, -0.087066650390625, -0.139068603515625, -1.572265625, -0.32967376708984375, -0.0350494384765625, 1.70361328125, -2.0302734375, 0.06705474853515625, 0.044189453125, -0.9883880615234375, 0.324737548828125, 0.4869384765625, 0.286163330078125, -1.5531005859375, 1.483123779296875, -0.1728668212890625, -0.183349609375, -0.297119140625, -0.17740631103515625, -1.4251708984375, 0.4280548095703125, -0.787567138671875, 0.710174560546875, 0.0, 0.08124542236328125, 1.204864501953125, -0.50579833984375, -0.18111419677734375, 1.2921142578125, -0.2141571044921875, -0.64569091796875, 1.84112548828125, -0.19921875, 0.42974090576171875, 0.7041778564453125, 0.4257965087890625, 0.8528575897216797, 0.241455078125, 2.0919189453125, -1.300933837890625, 0.552093505859375, -0.8263931274414062, -0.0565185546875, -0.0472412109375, -0.350189208984375, -1.40625, -0.191162109375, 0.64483642578125, -0.7419967651367188, -1.0675201416015625, -0.524444580078125, -1.35552978515625, 0.236328125, 1.2740631103515625, 0.2305450439453125, -1.67291259765625, 0.7183380126953125, -0.8356552124023438, -0.015838623046875, 0.70703125, -0.770050048828125, 0.8582763671875, -0.586334228515625, 1.2977294921875, -1.68939208984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000016.npy"} {"epoch": 0.033507853403141365, "step": 17, "batch_size": 128, "mean": 0.08988560736179352, "std": 0.7468373775482178, "min": -3.114105224609375, "p10": -0.76263427734375, "median": 0.12321090698242188, "p90": 0.9053466796874999, "max": 2.35211181640625, "pos_frac": 0.59375, "sample": [-0.659454345703125, -0.1310882568359375, -0.40997314453125, 0.22174072265625, 0.0126953125, -0.3936614990234375, -0.001068115234375, 0.29388427734375, -0.2249298095703125, -0.00597381591796875, 0.5966339111328125, 0.025787353515625, 0.052642822265625, 0.669769287109375, 0.12701416015625, 0.7264633178710938, 0.8112945556640625, -0.02410888671875, 0.8946533203125, 0.22491455078125, 0.3328857421875, 1.058074951171875, -0.59197998046875, 0.040618896484375, 0.1826629638671875, 0.778228759765625, -0.989349365234375, -0.43951416015625, -0.75201416015625, -0.98101806640625, 0.22607421875, 1.259765625, 0.722686767578125, 0.744110107421875, -1.452880859375, 0.2569160461425781, 1.316741943359375, -3.114105224609375, -0.0789337158203125, -0.08984375, 1.506622314453125, -0.3938140869140625, 0.04400634765625, 0.07902145385742188, -0.3347320556640625, 0.8603439331054688, 0.1924591064453125, -0.050365447998046875, 0.2339019775390625, 1.28436279296875, 0.57421875, 0.33246612548828125, 0.0421905517578125, 0.9302978515625, -0.45452880859375, 0.1581573486328125, -0.1313629150390625, -0.305511474609375, 0.16295623779296875, 0.26837158203125, 0.7096099853515625, 0.2622833251953125, -0.137359619140625, 0.11940765380859375, -0.959381103515625, 0.2081756591796875, -0.5615234375, -0.0436553955078125, -0.49634552001953125, 0.1939239501953125, 0.058746337890625, -1.186737060546875, -0.95703125, -1.2530364990234375, 0.173126220703125, 2.0933837890625, 0.6143341064453125, 0.23371124267578125, 1.8309478759765625, -0.14827728271484375, 0.161163330078125, 0.69134521484375, -0.5336151123046875, -0.225982666015625, -0.16397857666015625, -0.179290771484375, 0.63995361328125, 1.01422119140625, -1.5711517333984375, 0.819793701171875, 0.16510963439941406, 0.29437255859375, -0.38348388671875, 0.637054443359375, -0.4856414794921875, 0.163848876953125, -1.6826324462890625, 1.323486328125, 0.22719573974609375, 1.515167236328125, -0.07078933715820312, 0.202545166015625, 0.61822509765625, 2.35211181640625, 0.3418922424316406, 0.4945068359375, 0.2521400451660156, 0.1581573486328125, 0.157318115234375, 0.03631591796875, 0.04848480224609375, -0.24721527099609375, -0.78741455078125, -0.31900787353515625, -0.23931884765625, -0.269012451171875, 0.037994384765625, 0.40155029296875, 0.66827392578125, -1.39617919921875, 0.16375732421875, -0.43804931640625, -0.12152099609375, 0.2440948486328125, -0.80096435546875, -0.5752410888671875, -0.1461944580078125, 1.322235107421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000017.npy"} {"epoch": 0.0356020942408377, "step": 18, "batch_size": 128, "mean": -0.09905266761779785, "std": 0.8698012232780457, "min": -2.65032958984375, "p10": -1.2705963134765625, "median": -0.06304931640625, "p90": 0.88941650390625, "max": 2.233062744140625, "pos_frac": 0.453125, "sample": [-1.39984130859375, -1.275787353515625, 0.35369300842285156, 0.4181365966796875, 0.5399742126464844, -1.373046875, -0.09217453002929688, -0.7946548461914062, -0.09644317626953125, -0.006313323974609375, -0.079345703125, 0.1598358154296875, 0.45746612548828125, 0.344207763671875, 1.77239990234375, -0.0555419921875, 1.09271240234375, -1.93121337890625, 0.1497039794921875, -0.5417556762695312, 0.0409393310546875, 1.0091171264648438, 0.3486328125, 0.3601226806640625, 0.021759033203125, -0.16265106201171875, -1.5362396240234375, -0.6473236083984375, -0.4665985107421875, -0.3757972717285156, 0.88897705078125, -0.85748291015625, -0.023406982421875, -0.308807373046875, -0.49365234375, -0.391815185546875, 0.603912353515625, 0.5006980895996094, -0.182373046875, 0.89044189453125, 0.1892547607421875, 1.0189208984375, -1.78875732421875, -0.606231689453125, 0.6522750854492188, -0.1148529052734375, -0.697235107421875, -0.3089752197265625, 0.3006439208984375, 0.05767822265625, 0.2652587890625, -0.656585693359375, 0.8430023193359375, -0.10854339599609375, -1.7454833984375, -0.1395263671875, 0.9220733642578125, 1.2845458984375, 1.50738525390625, -0.745513916015625, 0.708587646484375, -0.15511322021484375, -0.0648193359375, -2.105316162109375, -2.65032958984375, -0.676788330078125, -1.26837158203125, 2.233062744140625, -0.248382568359375, -2.46539306640625, -0.09852790832519531, 0.16698074340820312, 0.3676300048828125, -0.3236083984375, -0.3348846435546875, -0.39107513427734375, -1.58148193359375, -0.24539947509765625, 0.6152267456054688, -0.050628662109375, -0.83880615234375, 0.2925262451171875, -0.061279296875, 1.06036376953125, 0.317779541015625, -0.28813934326171875, -0.6009674072265625, 0.8687744140625, 0.718780517578125, -0.5996246337890625, -0.730804443359375, 0.01090240478515625, 0.3790740966796875, -0.13918304443359375, -1.984375, 0.46527099609375, 0.05962371826171875, 0.05336761474609375, -0.2356109619140625, 0.48542022705078125, -0.7488479614257812, 0.74102783203125, 1.39910888671875, -0.05511474609375, 0.3643035888671875, 0.80194091796875, -0.78228759765625, -1.232086181640625, -1.1025466918945312, 0.5526275634765625, -0.43206787109375, -0.08184051513671875, -0.579986572265625, 1.0604248046875, 0.6840057373046875, -0.783355712890625, 0.3140716552734375, 0.67156982421875, -0.283905029296875, 1.8379669189453125, 0.54473876953125, 0.0870513916015625, 0.34918212890625, -1.2034912109375, -0.33306884765625, 0.2718505859375, -0.1263275146484375, -2.2679443359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000018.npy"} {"epoch": 0.03769633507853403, "step": 19, "batch_size": 128, "mean": -0.14113175868988037, "std": 0.924579381942749, "min": -3.558197021484375, "p10": -1.217762756347656, "median": -0.08076286315917969, "p90": 0.8864257812499998, "max": 2.387725830078125, "pos_frac": 0.4375, "sample": [-0.038280487060546875, -0.2158355712890625, 0.2993621826171875, -1.76141357421875, 0.7685546875, 0.42901611328125, -1.104644775390625, -0.068389892578125, -0.939208984375, -0.40673828125, 0.11394500732421875, -0.09313583374023438, 1.2380828857421875, -2.05615234375, 0.060230255126953125, -0.4694633483886719, 0.1197052001953125, -1.04437255859375, 1.0527420043945312, 0.93017578125, 0.6763534545898438, 0.312774658203125, -0.2917671203613281, -0.502838134765625, 0.56732177734375, -0.019866943359375, -0.013708114624023438, -0.46663665771484375, -1.96282958984375, 0.293701171875, 0.86767578125, 0.7232666015625, -3.558197021484375, -0.62164306640625, 0.37396240234375, -0.4974861145019531, 0.082489013671875, -1.34136962890625, 0.2487030029296875, 0.15160560607910156, -0.1604156494140625, -0.09491157531738281, -2.38934326171875, -2.064117431640625, 1.592529296875, 0.75396728515625, -0.0127410888671875, 0.421966552734375, -1.999603271484375, -0.41152191162109375, -0.27286529541015625, -0.627349853515625, 0.1266326904296875, -1.315185546875, -0.23516845703125, 0.0970458984375, -1.1417236328125, 0.96234130859375, 0.35933685302734375, -0.6576385498046875, -0.3930206298828125, 0.0891265869140625, -0.307586669921875, -0.5084419250488281, -1.621063232421875, -0.9561004638671875, -0.2533111572265625, 1.3258514404296875, -0.15970611572265625, -0.477935791015625, -0.250701904296875, -0.538543701171875, -0.889617919921875, 0.36651611328125, 0.5605087280273438, -0.40997314453125, 1.220458984375, 0.27333831787109375, 0.241058349609375, 0.31988525390625, 0.1976776123046875, -0.54486083984375, 0.22855567932128906, 0.0265655517578125, -1.1760101318359375, -0.7256088256835938, -0.022430419921875, -0.288818359375, 0.31780242919921875, 0.223876953125, 1.29339599609375, -2.06939697265625, 1.509185791015625, -0.8549652099609375, 0.1650390625, 1.55322265625, 0.7069091796875, -0.4109382629394531, 0.03025054931640625, 0.5265989303588867, -0.13587188720703125, -0.3163909912109375, -0.7854766845703125, 1.7967529296875, 0.80206298828125, 0.751800537109375, 0.6605224609375, -1.10546875, -0.1800689697265625, -0.256500244140625, -0.357208251953125, -0.1116485595703125, 2.387725830078125, -0.956451416015625, -1.08349609375, -1.3839874267578125, -1.151702880859375, -0.827972412109375, 2.173004150390625, -0.01546478271484375, 0.32550048828125, -0.122283935546875, 0.36932373046875, -0.224853515625, 0.40781402587890625, -1.8979949951171875, -0.014923095703125, 0.10467529296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000019.npy"} {"epoch": 0.039790575916230364, "step": 20, "batch_size": 128, "mean": -0.0368005633354187, "std": 0.9183447957038879, "min": -3.185516357421875, "p10": -1.1519180297851561, "median": -0.0072154998779296875, "p90": 1.1511138916015622, "max": 2.16241455078125, "pos_frac": 0.484375, "sample": [-0.3290252685546875, 0.858612060546875, 1.249603271484375, 1.87890625, 0.34039306640625, -0.604095458984375, -1.27301025390625, -2.00323486328125, -0.3175048828125, 0.31634521484375, -1.016632080078125, -0.225799560546875, 0.9026641845703125, 0.74005126953125, 0.650115966796875, 0.2393035888671875, 0.190643310546875, -0.37493896484375, 0.4768218994140625, -0.5831413269042969, -0.2402496337890625, 0.437774658203125, 1.923431396484375, -0.8167610168457031, 0.2508583068847656, 0.48685455322265625, -0.0638427734375, 0.27392578125, -0.765380859375, -1.4581298828125, 0.09051513671875, 0.06065559387207031, -0.09011459350585938, -0.34326171875, -0.5170516967773438, -0.983856201171875, 0.02911376953125, -0.047332763671875, 0.193603515625, -0.613037109375, -1.422943115234375, -1.05853271484375, 0.139404296875, -0.6765975952148438, -0.301055908203125, 0.502685546875, 0.266876220703125, 0.0810546875, -0.1859130859375, 0.6186447143554688, 0.4346923828125, -0.17986297607421875, 0.0, 0.9898681640625, 2.027618408203125, 0.09769439697265625, -0.1665973663330078, 0.7178955078125, 1.3305816650390625, -0.6083221435546875, 0.5329971313476562, 0.1123046875, -0.643951416015625, 1.753173828125, -0.17486572265625, -1.579681396484375, 1.1204071044921875, 0.00852203369140625, -0.4150390625, 0.46539306640625, -0.7283935546875, 0.30287933349609375, 0.12255859375, -0.008205413818359375, -0.61474609375, 0.18255615234375, -0.0062255859375, -0.427886962890625, 0.8928756713867188, 1.5550918579101562, 0.39422607421875, -0.01715087890625, -0.6055793762207031, 0.6065483093261719, 0.215179443359375, -0.37349700927734375, -2.4693603515625, -1.2401123046875, 0.00858306884765625, -1.506927490234375, 0.1255035400390625, 2.16241455078125, 1.86651611328125, -1.57806396484375, -1.1141204833984375, 0.8743896484375, -0.671905517578125, -0.8802108764648438, -0.6337890625, -0.3034172058105469, 1.42962646484375, -0.67022705078125, -0.313629150390625, -0.12705230712890625, -0.87646484375, -1.9815826416015625, -0.0772705078125, -1.338775634765625, -0.5896759033203125, -0.16998291015625, 1.371612548828125, -1.0630950927734375, -0.37579345703125, 0.4754638671875, -0.071136474609375, 0.41960906982421875, 1.5089111328125, -0.5045318603515625, -1.715545654296875, 0.17327499389648438, 0.920440673828125, 0.726104736328125, 0.7718505859375, 0.5078125, -0.16185379028320312, 0.14421844482421875, 1.2227630615234375, -3.185516357421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000020.npy"} {"epoch": 0.041884816753926704, "step": 21, "batch_size": 128, "mean": 0.12164326012134552, "std": 0.9039376378059387, "min": -2.8372802734375, "p10": -0.871185302734375, "median": 0.08883094787597656, "p90": 1.059351348876953, "max": 2.976165771484375, "pos_frac": 0.5625, "sample": [0.7451171875, 0.26992034912109375, -0.2705230712890625, -0.3446502685546875, -0.8289947509765625, -0.05950927734375, 0.36328125, 0.048370361328125, 0.06048583984375, -0.1253204345703125, 1.035888671875, 0.260955810546875, -0.172454833984375, -1.776519775390625, -0.580108642578125, -0.685638427734375, 0.32550048828125, 1.2015838623046875, -1.000640869140625, -0.2027759552001953, 2.976165771484375, 0.014829635620117188, -2.8372802734375, 1.871612548828125, 0.7933807373046875, 0.894195556640625, -1.026275634765625, 2.579620361328125, -0.42386627197265625, -0.40914154052734375, 0.20845794677734375, -0.076171875, 0.453125, 0.3050689697265625, 0.4834556579589844, -0.060333251953125, -0.9139785766601562, -0.07965087890625, -2.298492431640625, 1.0564727783203125, 0.0965423583984375, 0.069000244140625, -0.4461669921875, 0.1120758056640625, -0.1514739990234375, 0.8411865234375, 0.09730911254882812, -0.2435302734375, -0.5722122192382812, 0.57733154296875, -0.5909347534179688, 0.9171981811523438, 0.0, -0.8792953491210938, -0.868072509765625, 0.09502410888671875, -2.34967041015625, -0.878448486328125, -0.793975830078125, 0.01230621337890625, -0.58355712890625, -0.02716064453125, 0.13112640380859375, 2.45758056640625, 1.1861572265625, 0.0726776123046875, -0.1252288818359375, 0.45306396484375, -0.0948944091796875, 0.79962158203125, 0.8035888671875, -0.0251617431640625, -0.274444580078125, -1.350189208984375, 0.5072860717773438, 0.640869140625, -1.47161865234375, 1.988922119140625, -1.330902099609375, 0.2530517578125, -0.7851028442382812, 0.11078643798828125, 1.6839599609375, -0.354034423828125, 0.693206787109375, 0.238983154296875, 0.5893516540527344, 0.0007171630859375, 0.4118194580078125, 0.08263778686523438, -1.27191162109375, 1.1181716918945312, 1.7462158203125, 0.2969169616699219, -0.42791748046875, 0.5623054504394531, 0.4213409423828125, -0.313812255859375, 0.8504486083984375, 1.0611724853515625, 0.62176513671875, -0.257080078125, 2.06475830078125, 0.971588134765625, -0.712890625, 0.96490478515625, -0.1157379150390625, 0.3049774169921875, 0.46091461181640625, 0.595001220703125, -0.3799896240234375, -0.4807891845703125, 0.69415283203125, 1.235687255859375, 1.0585708618164062, -0.86328125, -0.3707427978515625, 0.69073486328125, 0.50775146484375, 0.95709228515625, -0.234375, 0.31243896484375, 0.21880149841308594, -0.6846923828125, -0.19793701171875, -0.0819549560546875, 0.14534378051757812, 0.5999259948730469], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000021.npy"} {"epoch": 0.04397905759162304, "step": 22, "batch_size": 128, "mean": 0.21360869705677032, "std": 0.900295078754425, "min": -2.166656494140625, "p10": -0.7543533325195312, "median": 0.0913543701171875, "p90": 1.4445922851562498, "max": 2.2467041015625, "pos_frac": 0.5625, "sample": [2.161712646484375, 0.3565826416015625, -0.2818145751953125, 0.5858154296875, 1.946929931640625, -0.7461090087890625, 0.0329742431640625, 1.0828018188476562, -1.706390380859375, 0.849945068359375, 0.852020263671875, -0.5109481811523438, -0.4397125244140625, -0.011676788330078125, -1.8295745849609375, -1.66558837890625, 0.36871337890625, 0.277008056640625, 0.43310546875, -0.3202972412109375, 1.483642578125, 0.6862945556640625, -0.773590087890625, -0.4007415771484375, -0.210113525390625, 0.047943115234375, 1.05572509765625, -1.33599853515625, 0.1432037353515625, -0.14947891235351562, -1.225982666015625, -0.2746734619140625, 0.0, 0.338470458984375, -1.2853775024414062, -0.585540771484375, -0.2656707763671875, 0.416900634765625, 0.7471237182617188, -0.940185546875, -0.23065185546875, 0.1824798583984375, 0.228424072265625, 2.20672607421875, -0.0212249755859375, 0.2505340576171875, -1.135955810546875, -0.67572021484375, 1.5909423828125, 0.811004638671875, 0.2687492370605469, -0.10926437377929688, -0.021484375, -0.735809326171875, 0.11297607421875, 0.568572998046875, -0.07979202270507812, 0.940155029296875, 1.921478271484375, -0.64117431640625, -0.072235107421875, 0.787017822265625, -0.32904052734375, -1.2609786987304688, 0.156890869140625, 0.13690185546875, 1.0255126953125, -0.3843994140625, -0.39487457275390625, -0.10638427734375, -0.11763763427734375, -0.599212646484375, -0.42746734619140625, 1.25164794921875, 2.017425537109375, 0.04779052734375, 0.6394805908203125, -0.3093109130859375, 2.16790771484375, -0.2308349609375, -0.2183837890625, -0.449920654296875, -0.011810302734375, 0.12266159057617188, 0.1586761474609375, 1.3841552734375, 1.0479736328125, -0.3132476806640625, 1.288818359375, -0.863525390625, 1.279205322265625, -0.4035797119140625, -0.130462646484375, 0.7520904541015625, 0.7013092041015625, -0.33538055419921875, -2.166656494140625, -0.491973876953125, 1.41339111328125, 0.54144287109375, 2.229278564453125, -0.07720184326171875, 1.024169921875, 0.27899169921875, 0.2137603759765625, 0.635528564453125, 0.1986083984375, 0.2681884765625, 0.818817138671875, -0.2618408203125, 0.18321609497070312, 0.069732666015625, 0.21118927001953125, -0.47628021240234375, 0.010120391845703125, 1.4278564453125, 0.633087158203125, 1.823486328125, -1.012725830078125, 2.155517578125, 0.02907562255859375, 1.668060302734375, -0.26947784423828125, 0.0146026611328125, 0.681396484375, 2.2467041015625, 0.015932083129882812, 0.960723876953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000022.npy"} {"epoch": 0.04607329842931937, "step": 23, "batch_size": 128, "mean": -0.05267469584941864, "std": 0.9013878107070923, "min": -2.90533447265625, "p10": -1.1243576049804687, "median": -0.00896453857421875, "p90": 0.943133544921875, "max": 3.78759765625, "pos_frac": 0.4921875, "sample": [-0.4023323059082031, 0.1501007080078125, -0.6104888916015625, -0.763519287109375, 0.814666748046875, 3.78759765625, 1.219146728515625, 0.67169189453125, -0.430877685546875, -1.36029052734375, 1.180419921875, -1.25958251953125, -0.294281005859375, -1.70068359375, -1.343902587890625, -0.3759002685546875, 1.57720947265625, 0.1988372802734375, -0.13137435913085938, -0.9593963623046875, 0.26160430908203125, -2.698486328125, -0.5029296875, -0.4259033203125, 0.25852203369140625, 0.5769271850585938, 0.57733154296875, -0.202362060546875, -0.5116424560546875, -0.8875961303710938, 0.0, 0.307830810546875, -0.08550262451171875, 0.1773681640625, -2.176971435546875, 0.269744873046875, -0.05020904541015625, -1.113311767578125, -1.285614013671875, -2.90533447265625, 0.074920654296875, -1.690399169921875, 0.95001220703125, -0.0179290771484375, 0.540802001953125, 0.123504638671875, 0.24273681640625, -1.0457992553710938, 1.18829345703125, -0.40964508056640625, 0.0425567626953125, 0.1881732940673828, -0.175872802734375, 0.3703460693359375, 0.87744140625, -0.032073974609375, 0.32830810546875, -0.40819549560546875, 0.57763671875, -0.19891357421875, -0.9835205078125, -0.51080322265625, 0.47650146484375, -0.19219970703125, 0.85546875, 0.281036376953125, 0.51409912109375, 1.25140380859375, 0.315673828125, -0.082000732421875, 0.9175949096679688, 0.5352516174316406, -0.662200927734375, -0.17022705078125, 0.775390625, 0.5742340087890625, 0.785186767578125, 0.2322711944580078, 1.7341461181640625, 0.1197052001953125, -0.6305618286132812, -0.5603847503662109, 1.1804161071777344, 0.37561798095703125, -1.089813232421875, -1.4249267578125, -0.681610107421875, 0.1967315673828125, 0.3925933837890625, 0.10333251953125, -1.6181564331054688, -0.12939834594726562, -0.322998046875, 0.6965141296386719, -0.4771575927734375, -0.932098388671875, 0.068206787109375, 0.940185546875, -0.111968994140625, -0.2613525390625, 1.831451416015625, 0.112030029296875, 0.3880462646484375, -0.079132080078125, -1.1501312255859375, -0.638671875, 0.89373779296875, 1.2218017578125, -0.12811279296875, 0.550445556640625, 0.19849395751953125, -1.2572021484375, 0.19117355346679688, -0.46978759765625, -0.3343086242675781, -0.34295654296875, -0.414703369140625, -0.711212158203125, 0.628204345703125, 0.024169921875, 0.7666015625, -0.966583251953125, 0.9832916259765625, 0.01324462890625, 0.9518280029296875, -0.9759063720703125, -0.8696250915527344, -0.7151412963867188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000023.npy"} {"epoch": 0.048167539267015703, "step": 24, "batch_size": 128, "mean": -0.07679517567157745, "std": 0.9854691028594971, "min": -3.085968017578125, "p10": -1.3674041748046875, "median": -0.08144187927246094, "p90": 1.040264892578125, "max": 3.510650634765625, "pos_frac": 0.4453125, "sample": [0.10834503173828125, -0.2664337158203125, -0.5191116333007812, 0.4568939208984375, -0.62371826171875, 1.63250732421875, 1.15045166015625, 1.90911865234375, 0.225372314453125, 0.47235107421875, 0.3761444091796875, 0.026287078857421875, -1.29541015625, -0.0435638427734375, -0.3112983703613281, -0.519287109375, 1.01922607421875, 1.054534912109375, -0.527618408203125, 0.3852214813232422, -0.1746826171875, -0.065032958984375, -0.002105712890625, 0.0, 0.043727874755859375, 0.4217567443847656, -0.38311767578125, -0.1436004638671875, -0.09453582763671875, -0.505462646484375, 0.9244384765625, -0.24039840698242188, -0.846160888671875, 1.77264404296875, 0.16729736328125, -0.642822265625, -1.856658935546875, -2.104522705078125, -0.14397430419921875, 0.127593994140625, -0.1018524169921875, -0.3756561279296875, -0.23167037963867188, -0.745452880859375, 3.510650634765625, -0.7115402221679688, 0.0, 0.218017578125, -0.0968017578125, -0.13053131103515625, -1.468719482421875, 0.39981842041015625, 0.2023162841796875, 0.17284393310546875, 1.05780029296875, -0.9603729248046875, -0.375244140625, 0.10038375854492188, 0.706573486328125, 0.592041015625, -3.085968017578125, -0.216552734375, -0.2249755859375, 0.0919647216796875, -0.78826904296875, 0.9771347045898438, -0.1838836669921875, 0.5161514282226562, -0.4640960693359375, 0.192626953125, 0.135223388671875, 0.267791748046875, -0.72821044921875, -1.41864013671875, -2.1785736083984375, 1.194366455078125, 0.214447021484375, -0.5465545654296875, 1.718048095703125, 0.0828094482421875, -0.712860107421875, -0.553924560546875, -1.711029052734375, 0.4820556640625, -0.6758880615234375, -0.286895751953125, -0.9530029296875, 0.39373779296875, -1.379302978515625, 0.5389556884765625, 0.1439666748046875, 1.53057861328125, -1.340240478515625, -0.7782135009765625, -0.1338653564453125, -0.09000778198242188, -0.0999755859375, 0.4381866455078125, -0.433746337890625, 0.7000732421875, -1.3623046875, -0.2174072265625, 1.507080078125, 1.81201171875, -1.553253173828125, -1.5810546875, -0.0330657958984375, -0.885101318359375, 0.042724609375, -0.0728759765625, 0.96923828125, -0.737091064453125, -0.641815185546875, 0.876922607421875, 0.6040878295898438, 1.034149169921875, 0.42572021484375, -0.37390899658203125, -1.1954345703125, -0.7612152099609375, 3.34375, -1.683502197265625, 0.14299774169921875, -2.1085205078125, 0.59735107421875, 0.0760650634765625, -1.663787841796875, 0.248016357421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000024.npy"} {"epoch": 0.050261780104712044, "step": 25, "batch_size": 128, "mean": 0.029784739017486572, "std": 0.771125316619873, "min": -3.166473388671875, "p10": -0.9443832397460937, "median": 0.0, "p90": 0.9239624023437499, "max": 2.2481689453125, "pos_frac": 0.4921875, "sample": [-1.1071014404296875, -0.78857421875, 1.076751708984375, 0.1557464599609375, -0.7010498046875, -1.25262451171875, -0.232177734375, 0.3714141845703125, 1.602783203125, -0.4576416015625, -0.1410980224609375, -0.3157501220703125, 0.1569671630859375, 0.478912353515625, 0.493560791015625, -1.01141357421875, -0.092864990234375, 0.5224838256835938, -0.344085693359375, 0.339111328125, -0.42047119140625, 0.08050727844238281, -0.2694091796875, 0.412567138671875, -1.38653564453125, -0.5065765380859375, -0.7889556884765625, 0.07293701171875, 0.34665679931640625, -0.6136474609375, -0.22904396057128906, 0.20269775390625, 0.0, -0.7745704650878906, 0.251800537109375, -0.014295578002929688, -0.38417816162109375, -0.3442535400390625, -1.134429931640625, -0.748138427734375, -1.5218353271484375, 0.0, 2.1362762451171875, -0.16119384765625, 0.4897308349609375, -0.045379638671875, -0.1245269775390625, 1.7952880859375, 0.8289337158203125, -0.64764404296875, -0.11922454833984375, -0.532623291015625, -0.00701904296875, 0.08165740966796875, 2.2481689453125, 0.010223388671875, -1.1083908081054688, -0.357940673828125, 1.0481719970703125, -0.097503662109375, 0.810150146484375, 0.137725830078125, -0.73291015625, -0.36700439453125, -0.503021240234375, 0.20968246459960938, 0.505950927734375, -0.2960662841796875, -3.166473388671875, 0.344970703125, 0.90673828125, 0.4333648681640625, -0.03537750244140625, 1.37310791015625, 0.48828125, -0.079986572265625, -0.37120819091796875, 0.5231513977050781, 1.0612335205078125, 1.11602783203125, -0.9985198974609375, 0.7349853515625, 1.222198486328125, 1.39324951171875, 0.779693603515625, 0.941650390625, -0.3038501739501953, 0.2528076171875, 0.6509895324707031, -0.16912841796875, 0.468719482421875, -0.4454345703125, 0.1348876953125, 0.789306640625, 0.9163818359375, 0.4345703125, 0.04683685302734375, 1.297271728515625, 0.871795654296875, 0.310333251953125, 0.01987457275390625, -0.42212677001953125, -0.3732757568359375, 0.0067901611328125, -1.3487548828125, 0.8104248046875, 0.525970458984375, -0.926910400390625, -0.46150970458984375, -0.37054443359375, -0.9851531982421875, 0.7475967407226562, 0.4427490234375, 0.34558868408203125, 0.18206787109375, -0.0534820556640625, -0.038604736328125, -0.16732406616210938, 0.4814910888671875, -0.01779937744140625, 0.24139022827148438, 0.24691200256347656, 0.6365966796875, -1.0415115356445312, -0.761322021484375, -0.81231689453125, -1.011444091796875, -0.19118690490722656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000025.npy"} {"epoch": 0.05235602094240838, "step": 26, "batch_size": 128, "mean": 0.0860128402709961, "std": 0.8689505457878113, "min": -1.9568023681640625, "p10": -0.8482559204101563, "median": 0.027856826782226562, "p90": 1.2226959228515621, "max": 3.46881103515625, "pos_frac": 0.5078125, "sample": [-1.914306640625, -0.13906478881835938, 3.46881103515625, -0.376312255859375, -1.053497314453125, 0.45334625244140625, 0.4685535430908203, 1.0413436889648438, 0.165771484375, 0.06987380981445312, 0.017414093017578125, -0.7806396484375, 0.0, -0.6553192138671875, 1.376861572265625, -0.7762908935546875, 1.17938232421875, 0.19646453857421875, 1.69610595703125, 0.2660369873046875, 0.424774169921875, -1.111907958984375, -0.7079505920410156, 1.323760986328125, 0.038299560546875, 0.10538101196289062, -0.5458526611328125, -0.4452362060546875, 0.054901123046875, -0.2994842529296875, 0.0703887939453125, -1.242156982421875, -0.029571533203125, 1.498779296875, -0.1728515625, 1.803466796875, 0.613189697265625, 0.27520751953125, 0.2048797607421875, -0.8496551513671875, -0.6038818359375, -0.3343467712402344, 0.8275222778320312, 0.88897705078125, 0.914947509765625, -0.815338134765625, 0.23622894287109375, -0.5373382568359375, -0.4847259521484375, -0.1770172119140625, -0.23419189453125, 0.8563232421875, -0.55877685546875, -0.492919921875, 0.4271240234375, -0.17362594604492188, -0.0524444580078125, -0.5272216796875, -0.0823974609375, 1.44873046875, 0.5331497192382812, -1.764068603515625, 1.769195556640625, -0.288482666015625, -0.301177978515625, -0.0010833740234375, -0.003021240234375, 0.9245147705078125, -0.64599609375, 0.5148391723632812, 0.5083465576171875, -0.45055198669433594, 0.181884765625, 0.202423095703125, 0.5024948120117188, -0.001926422119140625, -0.0410919189453125, 0.5481243133544922, 1.96429443359375, 0.25598907470703125, 0.121826171875, 2.018646240234375, -0.2187957763671875, -0.980316162109375, 1.590057373046875, 0.3497581481933594, 0.27581787109375, -1.9568023681640625, -0.060848236083984375, -0.6660003662109375, 0.077880859375, -0.11840057373046875, 0.4574317932128906, -0.614471435546875, 1.049835205078125, -1.576019287109375, 1.077117919921875, -0.12819290161132812, -0.05325508117675781, 0.4726104736328125, 0.072601318359375, -0.3495330810546875, 1.01788330078125, 1.349090576171875, 0.3097076416015625, -1.8402099609375, -0.1960601806640625, 0.0, -1.12054443359375, -0.4611358642578125, -0.1781768798828125, 0.442291259765625, 2.0164794921875, -0.7805252075195312, -0.130340576171875, 0.474761962890625, 0.191192626953125, 1.0738525390625, -0.84765625, 0.2093963623046875, -0.0294189453125, 0.66412353515625, 0.2166595458984375, 0.266265869140625, -0.5650634765625, 0.1232452392578125, -1.5506591796875, -1.132843017578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000026.npy"} {"epoch": 0.05445026178010471, "step": 27, "batch_size": 128, "mean": 0.06736427545547485, "std": 0.8935533761978149, "min": -2.25048828125, "p10": -0.8863861083984375, "median": -0.043633460998535156, "p90": 1.264779663085937, "max": 3.84136962890625, "pos_frac": 0.4765625, "sample": [0.39654541015625, 2.0406494140625, -0.2803153991699219, -0.13480377197265625, 1.0736083984375, -0.38995361328125, -0.545806884765625, -0.429962158203125, 0.0417327880859375, 0.32021331787109375, 0.0, -0.681396484375, 0.83758544921875, -0.4058380126953125, -0.0792236328125, -0.45574951171875, -0.44071197509765625, -0.0666961669921875, 0.1960906982421875, 1.20928955078125, 0.49645233154296875, 0.3087310791015625, 1.80633544921875, 0.3312530517578125, 1.6712646484375, 2.182281494140625, 0.10101318359375, 0.51318359375, -0.3782615661621094, -0.22463226318359375, -2.25048828125, -0.02704620361328125, 1.065399169921875, -0.1322021484375, 0.081268310546875, -0.565948486328125, 0.21233367919921875, -0.10064697265625, -1.262969970703125, 1.8937225341796875, 2.32452392578125, 0.9373779296875, -1.282928466796875, 0.3572540283203125, 0.3484039306640625, 0.64306640625, -0.535614013671875, -1.19036865234375, 1.70184326171875, 1.4058151245117188, -1.0726318359375, 0.4622039794921875, -0.538543701171875, 0.730865478515625, -0.0906982421875, -0.303436279296875, 0.18250656127929688, -0.325836181640625, 0.28510284423828125, 0.884033203125, -0.42718505859375, 0.2921905517578125, -0.529876708984375, -1.262725830078125, 0.11307525634765625, 0.1625213623046875, -0.3130950927734375, -0.3114738464355469, 3.84136962890625, 0.294189453125, -0.889312744140625, -1.347259521484375, -1.701629638671875, -0.74566650390625, 0.230865478515625, 0.9480819702148438, -0.78997802734375, 0.75579833984375, 1.49749755859375, 0.065032958984375, -0.22850799560546875, 1.5186767578125, 0.52838134765625, -0.4746551513671875, -0.735076904296875, -0.039398193359375, 1.890594482421875, -0.2241077423095703, 0.06655120849609375, 1.096954345703125, -0.1774749755859375, -0.49353790283203125, -1.66741943359375, -0.576019287109375, -0.86920166015625, -0.570648193359375, -0.133575439453125, -0.4553184509277344, -0.4242401123046875, 0.364776611328125, 0.438812255859375, 0.22899818420410156, -0.6489181518554688, -0.8851318359375, 0.14017486572265625, -0.053955078125, -0.989166259765625, -0.32254791259765625, -0.36277008056640625, -0.04786872863769531, 0.400390625, 0.23077392578125, 0.05518341064453125, 0.28863525390625, -0.2406463623046875, 0.4387359619140625, 0.6266021728515625, -0.103973388671875, 0.0722808837890625, 0.61553955078125, 1.394256591796875, 0.23596954345703125, -1.33294677734375, -0.823333740234375, -1.321990966796875, -0.105316162109375, -0.2456207275390625, -0.18795204162597656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000027.npy"} {"epoch": 0.05654450261780105, "step": 28, "batch_size": 128, "mean": 0.06961117684841156, "std": 0.724982738494873, "min": -2.2273406982421875, "p10": -0.6878341674804687, "median": -0.020913124084472656, "p90": 1.0359405517578124, "max": 2.954254150390625, "pos_frac": 0.484375, "sample": [0.619232177734375, -0.6112060546875, -0.6449737548828125, 1.13079833984375, 0.02390289306640625, -0.1736297607421875, -0.1450653076171875, 0.20751953125, -0.12142181396484375, -0.4432373046875, 0.1520519256591797, -0.311370849609375, 1.541168212890625, 0.5221099853515625, 0.582916259765625, -0.075347900390625, 0.220245361328125, -0.3609771728515625, 0.6662139892578125, -0.2784423828125, -0.10300636291503906, -0.013887405395507812, -0.572998046875, -0.382293701171875, -0.37471771240234375, 0.3906402587890625, 1.7724609375, 0.10799980163574219, -0.324737548828125, -2.2273406982421875, 0.31402587890625, 0.6416015625, 0.16741943359375, -0.1251983642578125, -0.17987442016601562, -0.9322509765625, 0.17608642578125, 0.8143463134765625, 0.858795166015625, 0.456695556640625, -0.239410400390625, 0.1397705078125, 0.081268310546875, -0.062286376953125, -0.2850189208984375, -0.8060302734375, -0.063323974609375, 1.0655517578125, 0.33343505859375, 0.23748779296875, 1.074310302734375, -0.09365081787109375, -0.42962646484375, -1.04150390625, 2.954254150390625, -0.8296661376953125, 0.5114479064941406, -0.085052490234375, 0.385894775390625, 1.2239227294921875, -0.650848388671875, 1.2987060546875, 0.145416259765625, -0.4744453430175781, 1.3309326171875, 0.454864501953125, -0.1429901123046875, 0.4356536865234375, 0.675567626953125, 0.311614990234375, 0.54913330078125, 1.04595947265625, -0.0544586181640625, -0.6391983032226562, 0.2937469482421875, 0.93829345703125, -1.86785888671875, -0.11598587036132812, -0.2226715087890625, 1.3067626953125, -0.1351318359375, -0.5480804443359375, 0.1800537109375, -0.21051025390625, -0.0279388427734375, 0.09728240966796875, -0.276092529296875, -0.2806854248046875, -0.7460784912109375, 0.025793075561523438, -0.35748291015625, -0.13787841796875, 0.9315338134765625, -1.2232666015625, -0.12238693237304688, -0.11526870727539062, 0.67803955078125, 0.043853759765625, 0.648406982421875, -0.972442626953125, -0.214874267578125, -0.662872314453125, 0.5003509521484375, 1.031646728515625, -0.2753448486328125, -0.6156005859375, 0.7970962524414062, 0.219512939453125, 0.90582275390625, 0.5389404296875, 0.587738037109375, -0.065155029296875, -0.85870361328125, -0.26782798767089844, 1.072662353515625, 0.15172576904296875, -1.4151153564453125, 0.13073348999023438, -0.117340087890625, 0.456146240234375, -0.5123443603515625, -0.4893798828125, -0.84710693359375, 0.47402191162109375, 1.32208251953125, -0.3791637420654297, 0.0, -1.663360595703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000028.npy"} {"epoch": 0.05863874345549738, "step": 29, "batch_size": 128, "mean": 0.1204247772693634, "std": 0.996165931224823, "min": -2.4778594970703125, "p10": -1.162066650390625, "median": 0.08746337890625, "p90": 1.3104934692382812, "max": 3.9453125, "pos_frac": 0.5546875, "sample": [0.010324478149414062, -0.371185302734375, 1.09796142578125, -2.05450439453125, -0.327423095703125, -1.572998046875, 1.28778076171875, 3.9453125, -1.438720703125, 1.247589111328125, 0.19024658203125, 2.43743896484375, -0.11053466796875, 0.6565399169921875, 0.29535675048828125, 0.0, -1.67279052734375, 0.096038818359375, -0.975982666015625, 0.2530708312988281, 1.556549072265625, 0.16064453125, -0.3801422119140625, -0.20745849609375, 0.6497344970703125, 1.3067626953125, 1.359466552734375, 0.6987762451171875, -0.16934967041015625, 0.22875213623046875, 0.2624053955078125, -2.4778594970703125, 0.369537353515625, 0.60400390625, -1.2481918334960938, 2.01220703125, -1.67340087890625, -0.436309814453125, 0.47418212890625, -0.36359405517578125, -0.20458984375, -0.05882072448730469, 0.999969482421875, 0.59820556640625, 1.616302490234375, 0.31824493408203125, -0.9659347534179688, -0.33404541015625, 0.4250335693359375, -0.2529449462890625, -1.2073974609375, -0.54150390625, -0.070953369140625, -0.339630126953125, 1.0261383056640625, 1.46728515625, 1.90380859375, 0.272186279296875, -0.88861083984375, 0.48681640625, 0.343902587890625, -0.497467041015625, 1.9835205078125, 1.91741943359375, -0.089935302734375, 0.14322662353515625, -0.5205078125, 1.0009765625, 0.0, -0.5633468627929688, 1.0352783203125, -0.010711669921875, -1.4169464111328125, -1.1145477294921875, 0.1864776611328125, -0.44220733642578125, -1.320556640625, 0.05014991760253906, -0.386962890625, -0.01806640625, 0.887420654296875, -0.48006439208984375, -1.525634765625, 0.11315155029296875, 1.2666015625, 0.1495513916015625, 0.9552154541015625, 1.3191986083984375, -0.6732940673828125, 3.406280517578125, 0.48952484130859375, -0.02899169921875, 0.4161224365234375, -0.31549072265625, 0.02337646484375, -0.73638916015625, 0.4116058349609375, 0.5162506103515625, 0.18259429931640625, 0.2025909423828125, 0.42431640625, 0.5962409973144531, -0.18853759765625, -0.8259811401367188, 0.328094482421875, 1.1268768310546875, 0.236541748046875, -1.15618896484375, -0.4365997314453125, -1.518463134765625, 0.85894775390625, 0.2029571533203125, 0.029052734375, 0.3656463623046875, 0.948760986328125, 0.00579833984375, -0.4309234619140625, -1.0360107421875, 0.1229248046875, -0.427947998046875, 0.05206108093261719, 0.078887939453125, 1.379486083984375, -0.7303619384765625, -1.17578125, -0.54718017578125, -0.183258056640625, 0.48590087890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000029.npy"} {"epoch": 0.060732984293193716, "step": 30, "batch_size": 128, "mean": 0.03173212707042694, "std": 0.866070568561554, "min": -3.28619384765625, "p10": -0.858148193359375, "median": -0.01087188720703125, "p90": 1.0212066650390623, "max": 2.37005615234375, "pos_frac": 0.4921875, "sample": [1.1177978515625, 1.6971282958984375, -1.165679931640625, -0.801422119140625, -0.3238868713378906, -0.218109130859375, -0.115386962890625, 0.4076690673828125, 0.75732421875, -0.27402496337890625, 0.154632568359375, 0.21759033203125, 2.112884521484375, 0.0948638916015625, -0.0213775634765625, -0.87408447265625, -1.99090576171875, 1.556732177734375, -0.9923019409179688, -0.4134674072265625, -0.846435546875, 0.55035400390625, -0.4757537841796875, 0.21334457397460938, 0.0409088134765625, -0.402618408203125, 0.915771484375, 0.17266464233398438, -0.11791229248046875, 0.3287353515625, 0.3263397216796875, -0.53765869140625, -0.30242919921875, -0.69549560546875, -0.4271354675292969, 0.75732421875, -0.065460205078125, 0.5408248901367188, -0.193634033203125, -2.28521728515625, -0.947021484375, -0.08028411865234375, 0.61480712890625, 0.96844482421875, -0.47418212890625, 1.00982666015625, -0.42806243896484375, -0.83551025390625, 0.31197357177734375, -0.588897705078125, 1.1512832641601562, 0.9587631225585938, -0.795257568359375, -0.6002655029296875, 0.418212890625, 0.37298583984375, -1.379852294921875, 0.17514801025390625, -0.851318359375, -0.438446044921875, 1.009033203125, 0.306640625, 1.9727935791015625, -0.33099365234375, 0.2741851806640625, -0.6042327880859375, 0.2886199951171875, -0.98870849609375, -0.3464698791503906, 0.303009033203125, 0.020076751708984375, 0.075836181640625, -0.5823211669921875, 0.0561065673828125, 0.78521728515625, 2.37005615234375, -0.4756317138671875, -0.53253173828125, 0.07956695556640625, -0.1786651611328125, 1.2603530883789062, 1.047760009765625, -2.490020751953125, 0.44293212890625, 0.6940994262695312, 0.623138427734375, 0.902618408203125, -0.26629638671875, 1.908203125, -0.715911865234375, -0.371490478515625, 0.8265380859375, 0.6654510498046875, -3.28619384765625, -0.971923828125, -0.0003662109375, 1.050018310546875, 0.002780914306640625, -0.312255859375, 1.8663330078125, -0.1334228515625, -0.4719696044921875, 0.4736785888671875, 0.2324371337890625, -0.04638671875, 0.5045166015625, -0.59552001953125, -0.445587158203125, -0.0325775146484375, -0.2525634765625, -0.6629638671875, -0.1718597412109375, 0.93206787109375, 0.43951416015625, 0.29561805725097656, 0.23520660400390625, 0.43158721923828125, 1.1953582763671875, -0.9980659484863281, -0.460845947265625, -0.955230712890625, -0.03252410888671875, 0.8397064208984375, -0.3779296875, -0.11169815063476562, 0.3499755859375, -0.3290252685546875, 0.8480224609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000030.npy"} {"epoch": 0.06282722513089005, "step": 31, "batch_size": 128, "mean": 0.09210464358329773, "std": 0.9335258603096008, "min": -2.5721435546875, "p10": -0.9912689208984374, "median": 0.01693248748779297, "p90": 1.1325897216796874, "max": 3.17303466796875, "pos_frac": 0.5, "sample": [1.0220870971679688, -0.921051025390625, -1.10186767578125, -0.660919189453125, -0.54296875, 1.4210586547851562, 1.0778350830078125, 1.4930419921875, 1.1003646850585938, -1.05499267578125, 2.322021484375, -0.6783447265625, 0.2698974609375, -1.200042724609375, -0.05804443359375, -0.1102447509765625, 0.6212158203125, 0.5260848999023438, 0.125732421875, -0.061431884765625, -1.40911865234375, -0.49200439453125, -0.01806640625, 1.346282958984375, 1.0859375, 1.58221435546875, -1.3850250244140625, 1.100341796875, -0.504241943359375, 0.4145050048828125, -0.124114990234375, 2.104400634765625, -0.33563995361328125, -0.5619697570800781, -0.963958740234375, 0.89996337890625, -0.082763671875, -0.4184722900390625, -0.8778533935546875, 0.0, -0.7967681884765625, 0.1502685546875, 0.0, 0.586334228515625, 0.6341934204101562, -0.583770751953125, 1.8889617919921875, 0.5749053955078125, -1.758514404296875, -2.0025787353515625, 1.490081787109375, -2.155364990234375, -0.058837890625, 0.055450439453125, 0.42767333984375, -0.069732666015625, -0.394683837890625, 3.17303466796875, -1.166839599609375, -0.7461051940917969, 0.0, -0.5179443359375, -0.66107177734375, -2.5721435546875, -0.893890380859375, -0.4523773193359375, -0.240264892578125, 0.33172607421875, -0.257568359375, -0.3412933349609375, -1.0751953125, -1.333984375, -0.15050125122070312, 1.028839111328125, -0.3814239501953125, 1.175506591796875, -0.4327392578125, -0.18955230712890625, 0.646240234375, -0.84271240234375, 0.18066024780273438, -0.08681488037109375, 0.715118408203125, 0.11585044860839844, 1.2796554565429688, -0.221893310546875, 2.16217041015625, 0.22259521484375, -0.3347625732421875, 0.987335205078125, -0.2197113037109375, 0.20273780822753906, 1.01416015625, 1.418365478515625, -0.326934814453125, -0.45168304443359375, 1.11419677734375, 0.241607666015625, 0.2639617919921875, 0.0, -0.6307373046875, -2.4168701171875, 0.5345611572265625, 0.6684341430664062, -0.2845458984375, 0.2103271484375, -0.627960205078125, 0.03386497497558594, 0.506439208984375, 0.7136688232421875, 0.750579833984375, -0.19504928588867188, 0.29183197021484375, 0.5699462890625, 0.54290771484375, -0.07486724853515625, 0.4186687469482422, 0.529571533203125, 0.9930419921875, 0.82794189453125, 0.76995849609375, 0.440673828125, 0.4679069519042969, 0.87457275390625, 0.2456512451171875, -0.174560546875, 0.4518585205078125, 0.03775787353515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000031.npy"} {"epoch": 0.06492146596858639, "step": 32, "batch_size": 128, "mean": 0.04823726415634155, "std": 0.830391526222229, "min": -2.412445068359375, "p10": -1.042083740234375, "median": 0.09407424926757812, "p90": 1.0423568725585937, "max": 2.0555267333984375, "pos_frac": 0.5390625, "sample": [0.254638671875, -0.251617431640625, -2.412445068359375, -1.284454345703125, -0.4431610107421875, -1.141845703125, -0.053394317626953125, 0.759490966796875, -0.5166778564453125, 0.201995849609375, -0.482086181640625, -0.3370513916015625, 1.0135955810546875, 0.543853759765625, -0.0089874267578125, -0.9898414611816406, -0.100616455078125, 1.10418701171875, 1.2314453125, 0.3687744140625, 0.21399688720703125, -0.83563232421875, -0.06396484375, -1.191192626953125, 1.3362884521484375, 0.490081787109375, 0.766326904296875, -0.36139488220214844, 0.2028656005859375, 0.21945953369140625, 0.18665313720703125, 0.33370208740234375, -0.25373077392578125, 0.64080810546875, 1.188995361328125, 0.831878662109375, -0.493927001953125, -0.529876708984375, 0.1905517578125, 0.4960174560546875, -0.2626495361328125, -2.11627197265625, 1.65557861328125, -0.27559661865234375, 0.4932861328125, 0.008575439453125, 2.0555267333984375, 0.40862274169921875, -1.13079833984375, 0.502777099609375, 0.434173583984375, -0.54290771484375, -0.465606689453125, -1.0386962890625, 1.88134765625, -0.4545745849609375, 1.037811279296875, 0.83319091796875, 0.5125732421875, -1.71234130859375, 0.44622802734375, 0.3017425537109375, -1.04998779296875, -0.2258453369140625, -0.33941650390625, -0.19268798828125, -1.220062255859375, 1.0529632568359375, 0.59625244140625, 0.6493682861328125, 0.857421875, 0.291168212890625, 0.4404106140136719, -0.753631591796875, 0.83502197265625, -0.371307373046875, -0.5143890380859375, -1.58795166015625, 0.04859161376953125, 1.885650634765625, 1.143585205078125, -0.98309326171875, 0.16544723510742188, -0.2475452423095703, 1.3077392578125, 0.8083953857421875, 0.8907318115234375, 0.328826904296875, -0.0953369140625, -0.61407470703125, 0.08516693115234375, -0.2882537841796875, 1.54010009765625, 0.80938720703125, -0.178741455078125, 0.3543701171875, -0.2852783203125, 0.246368408203125, 0.7314834594726562, 0.7303466796875, -0.04950714111328125, 0.4759521484375, -0.2955780029296875, 0.009979248046875, 0.2889404296875, 0.0714874267578125, -0.22751808166503906, 0.3585968017578125, -0.1455230712890625, 0.16817474365234375, -0.624908447265625, 0.1029815673828125, -0.543121337890625, -0.223602294921875, -0.84332275390625, -0.533721923828125, -1.6983642578125, 0.556365966796875, -0.580963134765625, -1.6741180419921875, -0.55816650390625, 0.5403842926025391, 0.819549560546875, 0.421417236328125, 0.3326416015625, -0.757843017578125, -1.35345458984375, 1.890716552734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000032.npy"} {"epoch": 0.06701570680628273, "step": 33, "batch_size": 128, "mean": 0.17485833168029785, "std": 1.0131292343139648, "min": -2.85662841796875, "p10": -1.1452213287353514, "median": 0.17236328125, "p90": 1.389961242675781, "max": 2.947906494140625, "pos_frac": 0.609375, "sample": [0.2512474060058594, 0.521087646484375, 0.1669921875, 2.0659637451171875, -0.473663330078125, 0.6345367431640625, 1.4557037353515625, 0.338165283203125, 1.96221923828125, -0.1587677001953125, 0.513458251953125, -0.275360107421875, 1.183380126953125, -0.62506103515625, 0.0, 0.214691162109375, 0.666412353515625, 0.368438720703125, -0.7565765380859375, -0.64434814453125, -0.623809814453125, -0.171234130859375, 0.06517219543457031, -1.61627197265625, 0.8049468994140625, 0.052276611328125, 0.7111358642578125, -0.3874664306640625, 1.7567138671875, 2.3126220703125, 0.70904541015625, -0.235321044921875, -0.714508056640625, -1.998870849609375, -0.36212730407714844, 2.947906494140625, 0.6228485107421875, -1.451629638671875, -0.671905517578125, -0.395263671875, 0.0829620361328125, -1.7273101806640625, -0.3632316589355469, 0.94525146484375, 0.15423583984375, 0.2496185302734375, 1.55621337890625, -0.2660369873046875, 0.5820693969726562, -0.713958740234375, 1.07110595703125, 0.8816680908203125, 0.010711669921875, 0.0, 0.299163818359375, -0.10870361328125, 0.42052650451660156, 1.191375732421875, 1.7755126953125, 0.9812164306640625, 1.90301513671875, -0.48590087890625, -1.1342430114746094, 0.177734375, 0.07794952392578125, -1.284393310546875, -0.2300090789794922, 0.22992706298828125, -2.3253173828125, 1.27874755859375, -0.9787139892578125, 0.7552490234375, 0.072723388671875, 0.986541748046875, 0.336212158203125, -0.129638671875, -1.3756256103515625, 0.5703125, 0.0276336669921875, -0.20257568359375, 0.7784423828125, 0.816925048828125, 2.4312744140625, 0.625274658203125, 1.6298828125, -0.74542236328125, -0.0829315185546875, -0.49078369140625, 0.3284454345703125, 1.361785888671875, 0.01922607421875, 0.060543060302734375, 1.1147918701171875, 0.356109619140625, 0.0682525634765625, -1.17083740234375, 0.2459869384765625, 1.473175048828125, -2.28265380859375, 0.6268310546875, -0.78125, 0.09428787231445312, 0.6128997802734375, 0.91064453125, 0.6247482299804688, 0.5177078247070312, -0.5575714111328125, 1.3450927734375, 0.8490447998046875, -1.589111328125, 0.91058349609375, 1.0164642333984375, -0.6572265625, 0.00408935546875, 0.67584228515625, -1.0024566650390625, -0.5286407470703125, -0.047389984130859375, 1.07305908203125, 0.8187713623046875, -1.273040771484375, -0.38409423828125, -2.85662841796875, 0.6194000244140625, -1.84014892578125, 0.3261146545410156, -0.014430999755859375, 2.295989990234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000033.npy"} {"epoch": 0.06910994764397906, "step": 34, "batch_size": 128, "mean": 0.10043707489967346, "std": 1.057915449142456, "min": -3.18914794921875, "p10": -1.0942169189453124, "median": 0.1310272216796875, "p90": 1.2621322631835936, "max": 3.86199951171875, "pos_frac": 0.546875, "sample": [0.386749267578125, 1.857696533203125, 0.84454345703125, -0.6011734008789062, 0.2111663818359375, -0.480621337890625, -1.229034423828125, 0.32135772705078125, -0.2257537841796875, 1.7784423828125, 0.83636474609375, -0.64697265625, 1.45361328125, -2.32769775390625, -0.2868194580078125, 1.67047119140625, 1.029052734375, 0.49942779541015625, -0.011112213134765625, -1.6067123413085938, 0.5381431579589844, -0.70562744140625, 0.4947509765625, 0.872100830078125, -0.345489501953125, -0.3712005615234375, 2.7197265625, -0.25974273681640625, 0.16511917114257812, 0.036376953125, -0.323455810546875, 0.45249176025390625, 0.776336669921875, -0.3521881103515625, 0.527587890625, -0.81060791015625, 3.17950439453125, -0.6536865234375, -1.03643798828125, -0.72259521484375, 0.47479248046875, -1.49530029296875, 1.209625244140625, 3.86199951171875, 0.160247802734375, 0.37621307373046875, -0.3860931396484375, -0.320648193359375, 0.441619873046875, -1.391998291015625, 0.5749359130859375, 0.059326171875, -0.0753173828125, -0.88543701171875, -3.18914794921875, 0.059173583984375, 0.6057815551757812, 0.47845458984375, 0.88311767578125, -0.434783935546875, 1.105194091796875, 0.2743644714355469, 0.8893585205078125, -1.01458740234375, 0.366973876953125, 0.85662841796875, 1.2695770263671875, -0.04289436340332031, 0.514129638671875, -0.560760498046875, 0.25286865234375, 0.48089599609375, -1.912353515625, -0.6623992919921875, -0.596099853515625, 0.2384796142578125, 0.4709358215332031, -0.024969100952148438, 0.4703826904296875, -0.4915580749511719, 1.770660400390625, -0.587249755859375, -0.7734375, -0.86126708984375, 0.539031982421875, 0.55499267578125, 1.077423095703125, -0.263946533203125, 0.7556381225585938, -1.52996826171875, -0.74371337890625, -0.760498046875, 1.11236572265625, -1.5902099609375, 0.38043212890625, -2.397735595703125, 0.0806884765625, -1.236083984375, -0.7550048828125, 0.4408721923828125, -0.22454833984375, 0.101806640625, 0.931915283203125, -0.47674560546875, 2.26092529296875, -0.7464981079101562, -1.448699951171875, 0.22878265380859375, -0.3908729553222656, 1.4541015625, 1.333953857421875, -0.22808837890625, 0.91168212890625, -0.4439697265625, 0.834228515625, 0.6762313842773438, -0.952545166015625, 0.8286590576171875, -0.465362548828125, 0.34539794921875, 0.403839111328125, 2.3560791015625, 0.6419219970703125, 1.258941650390625, -1.3692626953125, -0.555328369140625, 0.060302734375, -0.2287139892578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000034.npy"} {"epoch": 0.0712041884816754, "step": 35, "batch_size": 128, "mean": 0.04918116331100464, "std": 0.9458433389663696, "min": -2.917205810546875, "p10": -1.0824447631835938, "median": 0.027191162109375, "p90": 1.3133087158203125, "max": 2.525238037109375, "pos_frac": 0.515625, "sample": [-0.21477508544921875, -1.63043212890625, -1.540435791015625, 0.037841796875, 0.751434326171875, 0.868621826171875, -0.51409912109375, -0.20550537109375, -2.917205810546875, 0.12548828125, 1.4776611328125, 0.35904693603515625, 0.608062744140625, 0.758941650390625, 1.2615203857421875, 0.0761260986328125, -2.07794189453125, -1.17572021484375, -0.547149658203125, -0.11524200439453125, 0.20804595947265625, 0.7520751953125, 1.310516357421875, -0.333038330078125, 0.014617919921875, 0.198883056640625, 1.34368896484375, -0.264129638671875, 0.2059326171875, -0.87274169921875, 0.1670684814453125, -0.62628173828125, 0.3750457763671875, 0.3499603271484375, 1.1080322265625, 1.144622802734375, -0.3385467529296875, -0.78021240234375, 0.31868553161621094, 0.0, -0.701080322265625, 0.12872314453125, 0.01953125, -0.721832275390625, -0.03101348876953125, 0.04949951171875, 2.47003173828125, 0.0772857666015625, -0.166778564453125, -2.281494140625, -1.0828399658203125, -0.17819976806640625, 0.117919921875, -1.082275390625, 1.331512451171875, -0.00188446044921875, 2.525238037109375, 0.51019287109375, 0.5581130981445312, 0.0805511474609375, -0.52655029296875, 1.73828125, 2.32562255859375, 0.23931121826171875, 1.31982421875, 0.163909912109375, 0.695037841796875, 0.2745532989501953, -0.73773193359375, 0.10649681091308594, -0.1263275146484375, 0.946197509765625, -0.4531707763671875, -2.091949462890625, -0.05913543701171875, 0.425445556640625, -0.6484603881835938, 1.688690185546875, 1.48980712890625, 0.03485107421875, -0.036960601806640625, 0.19182586669921875, 0.43548583984375, 0.15996551513671875, 1.6099853515625, -1.2670745849609375, -0.07424545288085938, -0.49576568603515625, 0.81201171875, 0.5708885192871094, -0.66943359375, -0.32668304443359375, -0.14804649353027344, 0.2166900634765625, -0.09688568115234375, -0.396209716796875, -1.145263671875, -0.24609375, 0.72705078125, 2.12164306640625, -1.158416748046875, 1.28802490234375, 0.0, -0.3718414306640625, -0.00665283203125, -0.0674285888671875, -0.6637115478515625, -0.2124481201171875, 2.161468505859375, -0.5200729370117188, 0.168609619140625, -0.72222900390625, -1.142791748046875, 0.3210296630859375, 0.5516014099121094, 0.09033203125, -0.09746551513671875, 0.9589080810546875, -0.8360137939453125, 0.08249664306640625, -0.15643310546875, 1.073516845703125, -0.9391021728515625, -0.497711181640625, 0.3483734130859375, -2.1552734375, -0.53790283203125, -0.700927734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000035.npy"} {"epoch": 0.07329842931937172, "step": 36, "batch_size": 128, "mean": 0.2948346734046936, "std": 1.150039553642273, "min": -3.819183349609375, "p10": -1.0530624389648438, "median": 0.33100128173828125, "p90": 1.92755126953125, "max": 3.662078857421875, "pos_frac": 0.609375, "sample": [-0.53131103515625, -2.164520263671875, 0.886199951171875, -0.01506805419921875, 1.195159912109375, 1.00640869140625, -0.1800537109375, -0.3651275634765625, 0.43487548828125, -0.11091995239257812, 0.8480224609375, 0.4102630615234375, 1.21435546875, -0.82183837890625, 0.0206298828125, 0.7185592651367188, 0.04871368408203125, 0.0772705078125, 0.4117431640625, -0.12337493896484375, 0.7488479614257812, 0.06195068359375, -0.3949546813964844, 0.0, -0.22600555419921875, 0.51446533203125, -0.14447021484375, -1.583740234375, -0.3037872314453125, -0.15330123901367188, 2.23565673828125, 0.772369384765625, 0.408233642578125, 0.513671875, -2.0574951171875, -1.0495147705078125, 0.9020233154296875, -0.385284423828125, 0.150054931640625, 0.3941802978515625, 3.662078857421875, -0.1066741943359375, 0.4346771240234375, 0.4010162353515625, -1.011566162109375, 0.5357666015625, 0.7629547119140625, 2.137939453125, -0.465850830078125, 0.6737060546875, 1.02728271484375, 0.4307403564453125, 1.926544189453125, -0.5786361694335938, -1.0644989013671875, 0.729949951171875, 1.1934051513671875, -0.157958984375, 2.963623046875, 2.1680908203125, 0.0, -1.3463134765625, 1.260223388671875, -0.5460739135742188, -0.328887939453125, 1.40020751953125, -0.0845794677734375, 1.0912551879882812, -0.4001350402832031, 0.3337554931640625, 0.353607177734375, 1.9840087890625, -1.06134033203125, 0.949188232421875, -0.00927734375, 0.3848114013671875, 0.059478759765625, 0.963592529296875, -3.819183349609375, -0.401519775390625, 1.5089111328125, -1.967376708984375, 0.3203582763671875, -0.19574737548828125, -3.00494384765625, -0.4959602355957031, 1.35284423828125, -0.4810638427734375, 0.426055908203125, 0.6740646362304688, 1.929901123046875, 0.156158447265625, 0.9648895263671875, 0.15917205810546875, -2.47772216796875, -0.13232421875, -0.14432334899902344, 2.8990936279296875, -0.7433319091796875, 1.3041763305664062, -0.361083984375, 1.62835693359375, 0.7564697265625, 2.1235504150390625, 0.1683349609375, 0.3282470703125, 0.578857421875, 0.694427490234375, 0.654022216796875, -1.5847320556640625, -0.065185546875, 0.3501434326171875, 1.135101318359375, 2.23687744140625, -0.043853759765625, 0.01529693603515625, 2.176666259765625, 0.85443115234375, 2.074462890625, -1.3570709228515625, 0.154510498046875, 2.5814208984375, -0.3211669921875, 0.31143951416015625, 0.6569976806640625, 0.5726852416992188, 0.9293212890625, -1.370819091796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000036.npy"} {"epoch": 0.07539267015706806, "step": 37, "batch_size": 128, "mean": 0.24796608090400696, "std": 1.0888601541519165, "min": -2.71881103515625, "p10": -1.0443725585937498, "median": 0.155364990234375, "p90": 1.6751495361328126, "max": 3.40814208984375, "pos_frac": 0.578125, "sample": [-0.630096435546875, -0.0007476806640625, 1.128204345703125, 1.753326416015625, 1.5045166015625, 0.026233673095703125, 1.680511474609375, 0.3627815246582031, 0.2197113037109375, 0.238677978515625, 1.42083740234375, 0.05105781555175781, -1.95831298828125, 0.8207015991210938, -0.1781005859375, 1.449951171875, -0.4962310791015625, 0.6142196655273438, 1.3585205078125, -1.5948944091796875, 0.14339447021484375, -0.26544189453125, -0.3322772979736328, -0.760894775390625, -0.57647705078125, 0.45074462890625, -0.12091827392578125, 0.89697265625, 0.25623321533203125, -0.297149658203125, 1.142181396484375, -0.2273712158203125, -0.16199493408203125, -0.787200927734375, -0.400634765625, 2.57830810546875, -2.71881103515625, 1.235626220703125, 1.60748291015625, -0.014556884765625, 0.07509994506835938, -1.6929931640625, 0.280120849609375, 1.486480712890625, 1.5549163818359375, 2.99615478515625, 0.123260498046875, 0.12941741943359375, -0.12076568603515625, 0.603759765625, 1.3756256103515625, -0.481781005859375, -0.388427734375, 1.987762451171875, -1.5733642578125, -0.8250732421875, 1.9223480224609375, 0.5544891357421875, 0.516082763671875, -1.689453125, 1.176422119140625, 0.25823974609375, 1.1221466064453125, -0.642181396484375, -2.0369873046875, -1.205047607421875, 0.3193359375, -0.47369384765625, 0.165191650390625, 1.25592041015625, 0.0299072265625, 0.5233917236328125, 0.75927734375, 1.877838134765625, 0.2500457763671875, 0.7413330078125, -0.248138427734375, -0.25457763671875, 0.91644287109375, -1.183441162109375, 0.5620498657226562, 3.223175048828125, 0.47928619384765625, -1.489166259765625, 0.48492431640625, -0.698760986328125, 0.3284912109375, 0.453857421875, 0.145538330078125, 1.6728515625, -1.454833984375, -1.66851806640625, -0.59539794921875, -0.15673828125, -0.18361663818359375, 0.09001541137695312, 0.41339111328125, -0.3850822448730469, -0.27588653564453125, -0.33050537109375, -0.306243896484375, -0.11923599243164062, -0.73779296875, -0.438812255859375, 0.45163726806640625, 0.21734619140625, 3.40814208984375, 0.660430908203125, 0.3089752197265625, 0.21536636352539062, 0.887176513671875, -0.512115478515625, 0.322479248046875, -0.984771728515625, 0.091766357421875, 2.09771728515625, 1.890716552734375, -0.0257720947265625, -0.1269989013671875, -0.5010528564453125, -0.280487060546875, -1.5074005126953125, 0.890960693359375, 0.2712860107421875, 2.59539794921875, 0.6532135009765625, -0.115081787109375, 2.214569091796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000037.npy"} {"epoch": 0.0774869109947644, "step": 38, "batch_size": 128, "mean": 0.1785726696252823, "std": 0.8876856565475464, "min": -2.234619140625, "p10": -1.0341812133789061, "median": 0.21637344360351562, "p90": 1.2335144042968749, "max": 2.2418975830078125, "pos_frac": 0.625, "sample": [1.16705322265625, 0.21683502197265625, -0.24932289123535156, 0.683685302734375, -0.30165863037109375, -0.30850982666015625, -0.39417266845703125, 0.854034423828125, 1.257781982421875, 0.9945068359375, 0.7000732421875, -0.247314453125, 0.5704574584960938, 1.0704345703125, -0.141448974609375, 0.225616455078125, 0.574462890625, 0.21465301513671875, 0.4482269287109375, 0.5415496826171875, -0.34564208984375, -0.7616729736328125, 0.7935562133789062, 0.40367889404296875, 0.215911865234375, -1.47491455078125, -0.0806732177734375, 1.4794921875, 0.4124603271484375, -0.1891021728515625, -1.25946044921875, 1.77642822265625, -1.41497802734375, -1.80023193359375, 0.1785125732421875, 1.728851318359375, 1.0666961669921875, -0.25346946716308594, 0.95428466796875, 0.11004638671875, -0.3737335205078125, -0.4169921875, 0.742645263671875, -0.118316650390625, -0.870086669921875, 0.7409286499023438, -0.68023681640625, 0.0140380859375, 0.623931884765625, 1.223114013671875, 1.12188720703125, 0.4125213623046875, -1.5699462890625, -0.2393341064453125, 0.41109466552734375, -1.0737762451171875, -0.826171875, 0.049652099609375, -0.3339996337890625, 0.076385498046875, 0.331298828125, 0.3314208984375, -1.0172119140625, -0.5213623046875, 1.2110519409179688, 0.14171981811523438, 0.971893310546875, -2.1410675048828125, 2.2418975830078125, -0.545166015625, 0.4095458984375, -0.202117919921875, 1.4208984375, -0.739410400390625, -0.507904052734375, -1.9365234375, 1.3961334228515625, 0.472686767578125, 0.20037078857421875, -1.11956787109375, 0.820587158203125, -1.6051483154296875, -0.60821533203125, 0.13323974609375, -0.05075836181640625, 2.232208251953125, -0.51141357421875, -0.4079437255859375, 0.781951904296875, 0.34017181396484375, 1.2150192260742188, 0.05352783203125, 1.9423828125, 0.669281005859375, 0.450164794921875, 0.23792266845703125, 0.8376960754394531, 1.1303558349609375, 1.72119140625, -0.526580810546875, 0.0476226806640625, 0.0, 0.408905029296875, 0.01812744140625, 0.1366424560546875, -0.029022216796875, 0.217529296875, 0.4019775390625, -0.064056396484375, 0.052642822265625, 0.110137939453125, -2.234619140625, 0.4544830322265625, 0.39105224609375, 1.734375, 0.2882537841796875, 0.7239990234375, 1.350341796875, 0.8319091796875, 0.553466796875, -1.2593994140625, -0.3719940185546875, -1.469329833984375, 0.7831268310546875, 0.793182373046875, 1.403656005859375, -0.2929973602294922, 0.2927398681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000038.npy"} {"epoch": 0.07958115183246073, "step": 39, "batch_size": 128, "mean": 0.4789374768733978, "std": 1.040976643562317, "min": -1.9818115234375, "p10": -0.8675674438476562, "median": 0.3726997375488281, "p90": 1.8472076416015624, "max": 3.54949951171875, "pos_frac": 0.7109375, "sample": [1.0414581298828125, 0.354461669921875, -1.0430908203125, 0.3519134521484375, 1.510589599609375, 0.0198974609375, 1.0197906494140625, 0.45046234130859375, -1.171661376953125, 0.537353515625, 1.86688232421875, 1.838775634765625, 1.02532958984375, 0.723358154296875, 0.147247314453125, -0.6798095703125, -0.3348846435546875, 3.275299072265625, 2.2996673583984375, -0.011016845703125, 0.086212158203125, 0.640655517578125, 0.6858139038085938, -0.943115234375, 2.474639892578125, 0.37178802490234375, 0.8951339721679688, -0.4516143798828125, 0.3736114501953125, -1.24932861328125, 1.95281982421875, 0.449676513671875, 0.41741943359375, 0.27264404296875, 0.617584228515625, -1.365753173828125, -0.06014251708984375, 0.5505523681640625, 2.064453125, 0.6264801025390625, 0.4083251953125, 0.20058059692382812, 1.575775146484375, 0.3552970886230469, 0.029140472412109375, 1.340301513671875, 1.362396240234375, 0.00930023193359375, 1.06390380859375, 3.24810791015625, -1.9818115234375, -0.26879119873046875, -0.353485107421875, 0.918365478515625, 0.6858673095703125, 0.0409393310546875, -0.03179931640625, 1.65972900390625, 0.23089599609375, -1.717315673828125, -0.2845878601074219, -0.97467041015625, -0.9076004028320312, 0.377044677734375, 1.2807769775390625, 2.0460662841796875, 0.2502593994140625, -0.20362281799316406, 0.2413482666015625, -0.9402618408203125, -0.99542236328125, 0.2490234375, -0.11534881591796875, -1.645660400390625, 0.593048095703125, 1.046417236328125, 0.96026611328125, 0.687408447265625, 1.1284942626953125, -1.60888671875, 0.97381591796875, 0.13821029663085938, 1.065460205078125, 0.16326904296875, 0.5909271240234375, -0.06793212890625, 2.697357177734375, 3.05902099609375, 1.5406951904296875, 0.130767822265625, 0.20549583435058594, 0.052825927734375, -0.16715240478515625, -0.8504104614257812, 0.0, 0.481719970703125, 0.2410125732421875, 0.599456787109375, 1.302764892578125, 0.325469970703125, 1.7304229736328125, -0.35544586181640625, 1.027435302734375, -0.4176788330078125, 2.913909912109375, 1.1622467041015625, -0.841278076171875, 0.0973358154296875, 0.66943359375, -0.26495361328125, 0.13482666015625, 0.754302978515625, 0.4647388458251953, 0.985809326171875, -0.53448486328125, -0.6822242736816406, 0.9044952392578125, 1.922210693359375, 0.30084228515625, 0.14451217651367188, 1.1816253662109375, 0.45664215087890625, -0.07574462890625, 3.54949951171875, 1.315185546875, -0.0401763916015625, -0.32568359375, 1.0520477294921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000039.npy"} {"epoch": 0.08167539267015707, "step": 40, "batch_size": 128, "mean": 0.4076656103134155, "std": 1.1672630310058594, "min": -2.394866943359375, "p10": -0.5982421874999999, "median": 0.183868408203125, "p90": 1.931982421875, "max": 6.02569580078125, "pos_frac": 0.5859375, "sample": [-0.2831153869628906, -1.85443115234375, -0.6210250854492188, 0.22857666015625, -0.4072704315185547, -0.095916748046875, -2.394866943359375, 3.549407958984375, -0.15052032470703125, 1.2872314453125, 1.023651123046875, 0.8543243408203125, 0.13031005859375, 0.6507568359375, -0.0762786865234375, 0.6671867370605469, -0.2828941345214844, 1.350738525390625, -0.2255229949951172, 0.40659332275390625, -0.572540283203125, 3.20574951171875, -0.3736076354980469, -0.2701148986816406, 0.6611862182617188, 1.9716796875, -1.559783935546875, 0.243499755859375, -0.04248046875, 1.0600128173828125, -0.5582313537597656, 0.38336181640625, -0.5884780883789062, -0.13824462890625, 0.49273681640625, -1.289642333984375, -1.1503448486328125, -0.358734130859375, 1.9342041015625, -0.3214111328125, 6.02569580078125, 1.16082763671875, -1.1640625, 0.0155029296875, 0.003662109375, -0.5169219970703125, -0.496124267578125, 1.9310302734375, 0.630279541015625, 1.7464141845703125, -0.1266937255859375, 0.1768798828125, 0.34130859375, -0.012969970703125, 1.509857177734375, 2.61859130859375, 0.4201812744140625, 1.0715866088867188, 2.3939208984375, 0.534942626953125, -0.703460693359375, -0.038936614990234375, -0.51300048828125, -0.055419921875, -0.26580810546875, 0.340789794921875, 0.31109619140625, 1.030487060546875, 0.19085693359375, 0.54339599609375, 0.4803314208984375, -0.4004669189453125, 1.0367279052734375, 2.02069091796875, 0.12109375, -1.1025619506835938, 0.90155029296875, 0.08829116821289062, -0.53173828125, 0.31268310546875, -0.0531158447265625, 1.73876953125, -0.13897705078125, -0.6408615112304688, -0.4311676025390625, 0.7252693176269531, -0.4054374694824219, 4.047760009765625, -0.37195587158203125, -1.35784912109375, 2.16510009765625, 2.328857421875, -0.328857421875, 1.142730712890625, 1.408599853515625, -0.212890625, -0.321319580078125, -0.5140838623046875, 0.550628662109375, 0.38758087158203125, 0.655426025390625, 0.9638671875, 0.059112548828125, 0.070465087890625, 0.12353515625, 3.39874267578125, 0.4275054931640625, -0.010715484619140625, -0.05962371826171875, -0.5693359375, 0.39007568359375, 0.1739501953125, 1.218048095703125, 0.21885299682617188, 0.72125244140625, -0.671173095703125, -0.2435150146484375, -0.2282257080078125, 1.54449462890625, 0.8790283203125, 0.96331787109375, 0.1379566192626953, 0.422607421875, 1.338104248046875, 0.4434814453125, 2.110198974609375, -1.267547607421875, 0.73626708984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000040.npy"} {"epoch": 0.08376963350785341, "step": 41, "batch_size": 128, "mean": 0.22250136733055115, "std": 1.0642849206924438, "min": -2.7086181640625, "p10": -0.8792480468749999, "median": 0.18135881423950195, "p90": 1.4689502716064449, "max": 3.3585205078125, "pos_frac": 0.6015625, "sample": [2.21954345703125, -0.2120819091796875, 0.70977783203125, -0.22544097900390625, -0.09100341796875, -2.377593994140625, 0.580352783203125, 0.3006591796875, 0.3563995361328125, 0.287078857421875, 0.7056312561035156, 0.692230224609375, 0.27716827392578125, 0.2603759765625, -0.591949462890625, 0.3076171875, -0.0930938720703125, -0.7163238525390625, -0.019012451171875, 0.80218505859375, 0.31585693359375, 2.2059478759765625, 0.3814239501953125, 0.934722900390625, -0.908599853515625, -0.603790283203125, -0.421875, -0.28668212890625, 0.80706787109375, 0.2966899871826172, 1.3846588134765625, -1.187408447265625, 0.14182281494140625, 1.585601806640625, -0.4189300537109375, -0.23739242553710938, 1.70037841796875, 2.46820068359375, 0.060699462890625, 0.2846221923828125, -0.3411865234375, 0.27259063720703125, -0.7200164794921875, 0.79949951171875, 0.9262237548828125, -0.323455810546875, 0.77825927734375, -0.24852752685546875, 0.04443359375, -0.267913818359375, -0.2119903564453125, -0.35221099853515625, 0.2342529296875, -0.7919769287109375, 3.3585205078125, -0.5096435546875, 1.1854629516601562, 0.35955810546875, -1.188140869140625, 1.3498611450195312, 0.0269775390625, -0.33770751953125, 0.053619384765625, 0.96343994140625, -0.165008544921875, 0.0561370849609375, -0.5523681640625, 0.7976608276367188, 0.0, 0.786163330078125, -0.1595458984375, -1.37384033203125, 0.4666900634765625, -0.866668701171875, -2.611236572265625, -1.7686767578125, 2.1869888305664062, 0.02994537353515625, 0.064605712890625, 0.218841552734375, 0.3561248779296875, 0.8703689575195312, -0.590911865234375, 1.39312744140625, -0.048370361328125, 1.7408905029296875, 0.926422119140625, 0.49481201171875, 3.1172332763671875, 0.113067626953125, -0.526336669921875, -0.00421142578125, 1.1710281372070312, -0.7833099365234375, -0.494659423828125, 0.543304443359375, -0.980804443359375, -0.2754364013671875, 0.1438760757446289, -0.9200286865234375, 2.3536376953125, 0.35498046875, -0.598480224609375, -0.7863082885742188, -0.215179443359375, 0.7136688232421875, 0.0265655517578125, 0.079315185546875, 0.13543701171875, 0.35669708251953125, 1.4189567565917969, 1.1173095703125, 0.38525390625, -1.753631591796875, 2.74603271484375, 0.4912109375, 1.860595703125, -1.78851318359375, 0.2366485595703125, -0.036895751953125, -2.7086181640625, 0.716461181640625, 2.0299072265625, 0.992889404296875, 1.1821441650390625, 1.1458740234375, 0.5528106689453125, -2.61993408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000041.npy"} {"epoch": 0.08586387434554973, "step": 42, "batch_size": 128, "mean": 0.26422321796417236, "std": 1.047226071357727, "min": -3.65899658203125, "p10": -0.6622177124023436, "median": 0.1628570556640625, "p90": 1.5516265869140624, "max": 3.5047607421875, "pos_frac": 0.609375, "sample": [-0.5940494537353516, -0.26507568359375, 0.5797271728515625, 2.3458251953125, 0.62896728515625, 1.7569580078125, -0.4434051513671875, 0.79180908203125, 1.555023193359375, 0.08393096923828125, 1.29736328125, 2.36126708984375, -0.25152587890625, 0.158416748046875, -1.593231201171875, -0.06570243835449219, -0.7908935546875, 1.905120849609375, -0.30612945556640625, 1.32342529296875, -0.002227783203125, -0.6263885498046875, 0.7205352783203125, 0.22210693359375, 1.003082275390625, -0.2520599365234375, 0.73724365234375, 0.09429168701171875, -0.08807373046875, -1.0027008056640625, -1.30322265625, 1.0255165100097656, 0.455230712890625, 0.32928466796875, 1.86285400390625, 0.50531005859375, 1.290283203125, -0.4170989990234375, -0.7724456787109375, 0.6478271484375, -0.0885009765625, 0.032222747802734375, 0.1875, 0.1699981689453125, 0.602294921875, 0.0138702392578125, 0.860931396484375, 0.799774169921875, 0.128387451171875, 0.13681411743164062, 0.7083892822265625, -0.013147354125976562, 0.92315673828125, 0.7208023071289062, -0.11484146118164062, 0.19110870361328125, 0.16729736328125, -0.343902587890625, 0.122314453125, -0.0406036376953125, -0.273651123046875, -1.69976806640625, 0.381622314453125, 0.44952392578125, -0.3552398681640625, 1.431060791015625, 0.27873992919921875, -0.01016998291015625, -0.20937347412109375, 0.08427238464355469, -0.31170654296875, 0.46689605712890625, -0.745819091796875, -0.065216064453125, -2.976806640625, -0.015729904174804688, -3.12664794921875, -0.43386077880859375, 1.6446533203125, 0.048679351806640625, 0.48551177978515625, 1.10009765625, 1.88739013671875, 0.722900390625, 1.053070068359375, 0.7486572265625, 3.5047607421875, 0.0482177734375, 0.325897216796875, 0.215789794921875, 1.15557861328125, -1.433929443359375, 1.575927734375, -0.03232574462890625, 1.636688232421875, 0.81591796875, -0.154998779296875, 3.30126953125, 0.5483856201171875, -0.24806976318359375, 0.893341064453125, 0.11214447021484375, 0.0, -0.172821044921875, -2.35882568359375, 0.7158546447753906, 1.158660888671875, -0.01799774169921875, 0.1227264404296875, -0.4267425537109375, 1.31683349609375, -0.39923095703125, 0.8120498657226562, 0.2601966857910156, 1.5501708984375, 1.182373046875, 0.623687744140625, -0.44890594482421875, 1.6527786254882812, -0.6045684814453125, -0.405120849609375, 1.2825775146484375, -3.65899658203125, 0.0027923583984375, -1.04229736328125, -0.300689697265625, -0.09405517578125, 0.17540740966796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000042.npy"} {"epoch": 0.08795811518324607, "step": 43, "batch_size": 128, "mean": 0.3445459008216858, "std": 1.1919546127319336, "min": -3.70733642578125, "p10": -0.8638442993164062, "median": 0.3937387466430664, "p90": 1.97275390625, "max": 3.125579833984375, "pos_frac": 0.6328125, "sample": [0.1131591796875, 1.730133056640625, 1.588165283203125, -0.93988037109375, 1.8572845458984375, 1.1499481201171875, 3.125579833984375, 0.21652984619140625, 0.0648345947265625, -0.6007308959960938, 0.5722503662109375, -0.046722412109375, 0.395751953125, 1.97198486328125, 0.0118408203125, -0.318603515625, -0.7544403076171875, -0.029815673828125, 0.7056427001953125, -1.31787109375, 1.74432373046875, 0.7725753784179688, 2.11737060546875, 0.49346923828125, -1.2698822021484375, -0.849365234375, -1.015716552734375, -0.1439971923828125, -0.06795692443847656, 2.143035888671875, -1.3123779296875, 1.7782440185546875, -0.6351318359375, -3.62030029296875, 0.870086669921875, 0.3917255401611328, 0.2872161865234375, 0.482940673828125, 0.19768333435058594, -0.19274139404296875, -0.16521453857421875, 1.2943115234375, -0.40450477600097656, 0.816131591796875, 0.5105209350585938, 0.14781761169433594, 0.3213958740234375, -0.697052001953125, -3.70733642578125, 0.7238922119140625, -0.8976287841796875, 1.1735954284667969, 0.1751270294189453, -0.3991546630859375, 2.43048095703125, -0.197021484375, 2.329833984375, 0.6749420166015625, -0.5134963989257812, 0.2049560546875, 0.196502685546875, -0.4615020751953125, 2.0142822265625, 0.988250732421875, 2.4349365234375, -1.746795654296875, 0.93701171875, -2.0987548828125, 0.475677490234375, 1.5970306396484375, 1.596405029296875, 1.991424560546875, 1.1938018798828125, 0.822845458984375, -0.586273193359375, -0.6724319458007812, -0.2477874755859375, -1.66790771484375, -0.12400054931640625, 2.0439453125, 0.47528076171875, 0.43310546875, 1.16375732421875, 0.4475555419921875, 1.0551300048828125, 0.4542236328125, 0.6870269775390625, 0.03090667724609375, 0.904022216796875, -0.52593994140625, 1.554443359375, 1.97454833984375, -2.38525390625, 1.0945816040039062, 2.5277099609375, -0.35077667236328125, 0.546112060546875, 1.3433837890625, -0.773193359375, 0.725433349609375, 0.4415283203125, 0.9410324096679688, 1.941131591796875, 0.2868537902832031, 0.6507568359375, -0.14044189453125, 2.56414794921875, -0.5096435546875, 1.6308135986328125, 2.30401611328125, -0.31884765625, 0.512298583984375, 0.5418014526367188, -3.1095657348632812, 0.35491943359375, 0.5946502685546875, -0.4808502197265625, 0.7474441528320312, 0.791900634765625, -0.05316162109375, -0.2913665771484375, -0.44677734375, -0.009191513061523438, 0.510498046875, -0.39990234375, -0.6887664794921875, 0.1548614501953125, 0.025178909301757812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000043.npy"} {"epoch": 0.09005235602094241, "step": 44, "batch_size": 128, "mean": 0.5084197521209717, "std": 1.1600193977355957, "min": -3.327880859375, "p10": -1.0022628784179688, "median": 0.4660453796386719, "p90": 2.031292724609375, "max": 3.8736572265625, "pos_frac": 0.6953125, "sample": [0.099761962890625, -0.0336151123046875, 2.4964599609375, -1.0031814575195312, -0.24965667724609375, 2.2030029296875, 0.45188140869140625, -0.032085418701171875, 0.7815399169921875, -0.3175048828125, 0.0175628662109375, 0.3909912109375, 0.3619232177734375, -0.5137557983398438, 1.5858154296875, 1.102691650390625, -1.24053955078125, 4.57763671875e-05, -2.4619140625, 1.9290008544921875, 1.921112060546875, 1.2012939453125, 0.3925933837890625, 0.15960311889648438, 0.7965087890625, 0.7007827758789062, 1.754058837890625, 0.6489982604980469, -2.078125, 0.11871337890625, 1.0557708740234375, 0.724945068359375, -0.396270751953125, -0.59619140625, -0.01239776611328125, 0.253173828125, 0.337158203125, -0.33402442932128906, 1.459564208984375, 2.1284027099609375, 2.60137939453125, -0.2491588592529297, 0.353607177734375, 1.64788818359375, 0.488739013671875, 3.8736572265625, 0.33197021484375, 2.546966552734375, 1.1242523193359375, -0.6132354736328125, 0.25731849670410156, 0.04547119140625, -1.0018692016601562, 0.261474609375, 0.39586639404296875, 1.37030029296875, -0.53619384765625, 0.57373046875, 0.3742942810058594, -0.851043701171875, 1.1819000244140625, 1.6470947265625, 0.344390869140625, 2.3861083984375, 1.58306884765625, 0.8188095092773438, 0.72784423828125, -1.542755126953125, 1.40277099609375, 3.313079833984375, 1.190216064453125, 0.4802093505859375, 1.3778076171875, 0.78387451171875, 2.182647705078125, 1.2175445556640625, 0.85015869140625, -1.20184326171875, 0.5391464233398438, -1.335357666015625, 1.5067138671875, 1.1281280517578125, 2.1964111328125, 0.50177001953125, 0.71917724609375, -0.18369293212890625, 0.4500274658203125, -0.74957275390625, 2.11370849609375, 0.0015716552734375, -0.277008056640625, 1.9959716796875, -0.55572509765625, 1.34478759765625, -0.5406646728515625, 1.18487548828125, 1.068878173828125, -1.522247314453125, -0.4788360595703125, -1.4853515625, -1.165283203125, -0.08466339111328125, -0.4940032958984375, -0.035247802734375, 0.3155975341796875, 1.290771484375, -0.23486328125, 1.64959716796875, 0.05631256103515625, -1.29150390625, 2.45513916015625, 1.546722412109375, 0.14995956420898438, 2.220672607421875, 0.7832260131835938, 1.105133056640625, 0.5800552368164062, 0.43328857421875, -3.327880859375, -0.42618560791015625, -1.2150115966796875, 0.6499214172363281, 0.5103912353515625, 1.1788482666015625, 1.95782470703125, -0.32692718505859375, 0.6343231201171875, 0.996368408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000044.npy"} {"epoch": 0.09214659685863874, "step": 45, "batch_size": 128, "mean": 0.534998893737793, "std": 1.3676496744155884, "min": -4.356689453125, "p10": -0.7119247436523436, "median": 0.3321533203125, "p90": 2.1443115234375, "max": 5.310516357421875, "pos_frac": 0.6796875, "sample": [-0.099395751953125, 0.0386505126953125, 0.651092529296875, -0.3006439208984375, -0.06286048889160156, 0.696136474609375, 0.12053298950195312, 1.015228271484375, 1.1594734191894531, 1.53900146484375, 1.808441162109375, 3.33551025390625, 0.91937255859375, -0.14041900634765625, 1.35784912109375, 1.4526443481445312, 0.030071258544921875, 0.29583740234375, 0.0, -1.8427734375, 0.1099853515625, 0.787628173828125, 0.515960693359375, 0.7704277038574219, -1.132598876953125, 3.97479248046875, -0.3126220703125, -0.1937255859375, 0.9422149658203125, 0.23534393310546875, 0.04803466796875, -0.24744033813476562, -1.1304931640625, -0.30450439453125, -0.34703826904296875, 1.686065673828125, -1.227081298828125, -0.12842178344726562, 2.08221435546875, 1.732940673828125, 4.6988525390625, -0.34906005859375, 0.1858062744140625, 0.280426025390625, 4.07684326171875, 0.4386329650878906, -0.14892578125, 0.214752197265625, -1.14599609375, 0.85595703125, -0.6378173828125, 0.1244049072265625, -1.2518310546875, 1.162872314453125, 0.052349090576171875, 0.3708648681640625, 0.7053604125976562, 0.21636962890625, -0.20552825927734375, -4.356689453125, -0.6553955078125, 2.0147705078125, -0.49634552001953125, 0.461090087890625, -0.49664306640625, 0.9330596923828125, 0.324676513671875, 0.39520263671875, 0.4027099609375, -0.2198467254638672, -0.1373291015625, 2.473388671875, -0.4732513427734375, 0.4109992980957031, 3.406097412109375, 0.36297607421875, 0.555999755859375, 0.151702880859375, 1.866912841796875, 0.09417724609375, 0.8946762084960938, 2.98272705078125, 0.61370849609375, 2.6765594482421875, 0.5923080444335938, -0.8438262939453125, 0.037872314453125, 0.11795806884765625, -0.65032958984375, -3.2078857421875, 1.133880615234375, 2.281158447265625, 0.57159423828125, -1.4814300537109375, 2.95208740234375, 0.7949371337890625, 1.65838623046875, -0.476715087890625, 1.9898681640625, -0.379058837890625, -1.89666748046875, 0.339630126953125, -0.01854705810546875, 0.960693359375, 1.047943115234375, 2.8341064453125, -0.5151214599609375, -1.5937957763671875, 0.42034912109375, 0.11618423461914062, 0.1071319580078125, -0.3174591064453125, 2.136199951171875, 2.01922607421875, 0.306884765625, 0.3056640625, 0.679168701171875, 2.163238525390625, 5.310516357421875, 1.370361328125, 1.3831787109375, 1.024658203125, 0.5678977966308594, -0.1620635986328125, 1.1614990234375, 0.19539642333984375, -1.01763916015625, 0.7947235107421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000045.npy"} {"epoch": 0.09424083769633508, "step": 46, "batch_size": 128, "mean": 0.6320484280586243, "std": 1.2782378196716309, "min": -2.857452392578125, "p10": -0.6043983459472656, "median": 0.44033050537109375, "p90": 2.2183349609374994, "max": 5.69482421875, "pos_frac": 0.65625, "sample": [1.2999267578125, 2.12152099609375, 0.8581695556640625, -0.29862213134765625, -0.03582763671875, 0.86199951171875, 1.46575927734375, -0.62481689453125, -0.05169677734375, -0.6182785034179688, -0.827239990234375, 0.501953125, 3.4884033203125, -0.104461669921875, 1.036285400390625, -0.59844970703125, 0.51336669921875, 0.3490104675292969, 1.831298828125, 1.281829833984375, -0.49725341796875, 0.947021484375, -0.17919921875, 0.668792724609375, -0.48956298828125, -1.471221923828125, -2.189727783203125, -0.113677978515625, 5.69482421875, -0.06594085693359375, -0.0029430389404296875, 0.6519775390625, 0.769256591796875, 1.27301025390625, 1.190216064453125, 1.6099853515625, -0.0933837890625, 0.32222747802734375, -0.4485931396484375, 1.06390380859375, 0.023834228515625, 1.356903076171875, 1.4384231567382812, 0.60235595703125, -0.214111328125, -2.07867431640625, 1.71527099609375, 0.4311065673828125, 0.9267501831054688, 0.6378936767578125, 3.07476806640625, 1.4897613525390625, 1.2415084838867188, 0.0708770751953125, 0.0699615478515625, 0.535614013671875, 0.27960205078125, 3.4344482421875, -0.010528564453125, 3.22174072265625, 0.01172637939453125, 0.4595947265625, -0.5499153137207031, -0.25189208984375, 1.438079833984375, -0.69580078125, 0.04601478576660156, 0.449554443359375, -0.932098388671875, 0.99346923828125, 0.345489501953125, 0.04096221923828125, 2.03643798828125, -0.14996337890625, -0.7565536499023438, 1.099456787109375, 0.2552757263183594, -0.8122100830078125, 0.3262653350830078, 0.65185546875, 0.8638153076171875, 2.466644287109375, 1.737579345703125, -0.13671875, 0.5267333984375, 0.22019195556640625, -2.857452392578125, -0.30698394775390625, 0.6432952880859375, 0.652740478515625, -0.2610321044921875, 0.232147216796875, 2.40350341796875, 0.259307861328125, -0.21985244750976562, 2.13897705078125, 0.06287384033203125, -0.3339080810546875, 0.965576171875, 0.943939208984375, -0.2069377899169922, -0.271484375, 0.8514404296875, -0.1904449462890625, 3.8250732421875, 0.7133865356445312, 1.245361328125, 1.3893585205078125, 0.787322998046875, 0.08154296875, 2.731353759765625, -0.68731689453125, 0.005306243896484375, 0.3064384460449219, -0.17576217651367188, 3.21002197265625, 1.54656982421875, 0.962310791015625, 0.45660400390625, 4.3275146484375, 0.0, 1.52423095703125, -0.20880126953125, -1.2315139770507812, -0.02542877197265625, 3.86090087890625, -0.0303955078125, 2.761077880859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000046.npy"} {"epoch": 0.09633507853403141, "step": 47, "batch_size": 128, "mean": 0.3551369905471802, "std": 1.3988271951675415, "min": -4.004638671875, "p10": -1.3693840026855468, "median": 0.34942626953125, "p90": 2.3213989257812497, "max": 4.23590087890625, "pos_frac": 0.6640625, "sample": [-1.1285400390625, 0.2237110137939453, 1.0031356811523438, -0.63446044921875, 4.23590087890625, -1.427734375, -4.004638671875, 0.2703094482421875, 1.664642333984375, 0.437469482421875, 0.303680419921875, 0.38159942626953125, 0.1425933837890625, -0.39910888671875, -1.750762939453125, -0.2601318359375, 0.360107421875, -1.02056884765625, 2.7086181640625, -0.586151123046875, 1.104705810546875, -2.116363525390625, 0.7391128540039062, -0.0763702392578125, -0.13134765625, 0.188446044921875, -0.3499908447265625, 0.478302001953125, -2.893280029296875, -1.19500732421875, 0.9739227294921875, 0.0, 1.432281494140625, -1.3841934204101562, 0.324462890625, 1.08056640625, 0.75732421875, 1.05938720703125, -0.4986114501953125, 0.5805683135986328, -0.4785614013671875, 0.02787017822265625, 2.182861328125, 3.3526611328125, 1.880401611328125, 1.322662353515625, -1.264190673828125, -1.53143310546875, 1.783966064453125, 0.21671485900878906, 0.4699554443359375, -0.14850234985351562, 3.413665771484375, 0.1018218994140625, -1.618865966796875, 2.5858154296875, -2.59136962890625, 2.34344482421875, 2.7959442138671875, 1.0753631591796875, 0.1941070556640625, 1.118194580078125, -0.6501045227050781, 2.31195068359375, 1.3924102783203125, -0.6916885375976562, 0.032073974609375, 0.7915267944335938, 0.6280117034912109, 2.843780517578125, 0.356201171875, 0.26867103576660156, -0.0411529541015625, -0.279388427734375, 1.3063507080078125, 1.2923507690429688, 0.0, 2.3936920166015625, 2.177459716796875, 0.5650634765625, 1.0121078491210938, 0.40988922119140625, -1.2055206298828125, -1.5272064208984375, 2.627532958984375, 0.3647613525390625, 0.06254959106445312, 0.07720947265625, 0.912322998046875, 0.5698013305664062, 0.7500228881835938, 0.001983642578125, -0.556182861328125, 0.3426513671875, 0.7087326049804688, 0.59930419921875, 1.95452880859375, -0.4400634765625, 0.587646484375, 1.0487594604492188, -0.69903564453125, 1.1451187133789062, -0.1720428466796875, 0.0489501953125, 0.917236328125, -1.29693603515625, 0.43121337890625, -1.363037109375, -1.1199493408203125, 0.17407608032226562, 0.268096923828125, 0.9634246826171875, 3.206085205078125, 0.3591194152832031, 1.24853515625, 2.805908203125, -1.0255203247070312, 0.8666877746582031, 0.39183807373046875, 0.0318603515625, -1.69158935546875, 2.198577880859375, 0.7270126342773438, -0.695465087890625, 2.596923828125, -2.95208740234375, -2.86663818359375, 0.13501739501953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000047.npy"} {"epoch": 0.09842931937172775, "step": 48, "batch_size": 128, "mean": 0.6199018955230713, "std": 1.4153958559036255, "min": -2.372802734375, "p10": -0.8313434600830077, "median": 0.3825111389160156, "p90": 2.3818832397460934, "max": 6.480712890625, "pos_frac": 0.6328125, "sample": [-0.327911376953125, 1.40191650390625, 0.6414031982421875, 0.8954544067382812, 2.364959716796875, 1.360687255859375, 0.1153564453125, 3.57891845703125, 1.58245849609375, 0.041839599609375, -0.2855682373046875, -0.74066162109375, -0.07901382446289062, -0.585052490234375, -0.590484619140625, 2.0947265625, 0.06414794921875, -1.9942474365234375, -1.675689697265625, -0.2692718505859375, 3.146331787109375, 0.6518135070800781, -0.364349365234375, -1.0736083984375, 1.5187835693359375, 0.33617401123046875, -0.9554595947265625, 0.9189682006835938, -1.59381103515625, 1.2350616455078125, 0.4623737335205078, 2.179168701171875, 2.27764892578125, 0.371612548828125, 3.007659912109375, 1.0948944091796875, 0.004726409912109375, 2.3643798828125, 1.601531982421875, 2.6973724365234375, 0.39340972900390625, 2.348541259765625, 0.9049835205078125, 0.2568359375, 0.752471923828125, 0.1516876220703125, -2.372802734375, -0.195587158203125, 0.75347900390625, 0.9117355346679688, 3.2335205078125, 0.790802001953125, 0.1674346923828125, 0.03246307373046875, -0.200347900390625, 0.9044952392578125, 0.7995223999023438, -0.0830230712890625, 0.07491493225097656, 1.645477294921875, 1.36993408203125, 1.107666015625, -0.667144775390625, 1.5039138793945312, -0.363372802734375, 0.270965576171875, 0.93817138671875, -0.681915283203125, 1.0549240112304688, 1.2083740234375, -1.3236083984375, 0.01153564453125, 0.0, -0.8693389892578125, 1.471466064453125, 0.0952911376953125, 2.227783203125, 0.28055572509765625, 0.5465850830078125, 2.7353515625, 0.60162353515625, 3.591583251953125, 0.5365753173828125, 4.08917236328125, -0.716064453125, 3.2034912109375, 0.9195556640625, 0.3481597900390625, -0.2843608856201172, 0.3303070068359375, -1.485198974609375, 1.0009307861328125, -2.0291595458984375, -0.06113433837890625, 1.680999755859375, 3.451904296875, -1.18096923828125, 1.683258056640625, -1.77569580078125, 1.1325149536132812, -0.0573272705078125, 2.4213714599609375, 0.0, 0.98779296875, 1.879913330078125, -0.330841064453125, 6.480712890625, -0.12921905517578125, -0.64227294921875, 1.247222900390625, -0.49517822265625, 0.44828033447265625, 0.5333251953125, 0.6568603515625, -0.4782524108886719, -0.1284332275390625, -0.798614501953125, -0.8150596618652344, -0.03260040283203125, 0.8009681701660156, -0.5888519287109375, -0.120697021484375, -0.7490234375, -0.159637451171875, -1.2817535400390625, -0.068115234375, 4.1351318359375, 1.931854248046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000048.npy"} {"epoch": 0.10052356020942409, "step": 49, "batch_size": 128, "mean": 0.4983615279197693, "std": 1.6273279190063477, "min": -5.22198486328125, "p10": -1.3924240112304687, "median": 0.41829681396484375, "p90": 2.61800537109375, "max": 5.037689208984375, "pos_frac": 0.6640625, "sample": [1.24884033203125, 0.333709716796875, 0.6043434143066406, 0.8881149291992188, -1.1544036865234375, -1.1462860107421875, -0.168701171875, 4.3348388671875, 2.06988525390625, 0.9367008209228516, 0.35196685791015625, 1.2974700927734375, -1.37103271484375, -0.3743133544921875, 0.04803466796875, 3.11358642578125, 1.866668701171875, 0.9505767822265625, -0.102935791015625, 0.83489990234375, -1.64617919921875, -0.23209762573242188, 2.98150634765625, 0.7396240234375, 0.408843994140625, 0.1326904296875, -1.4423370361328125, 5.037689208984375, -0.04425048828125, 0.79608154296875, 0.833831787109375, 1.978302001953125, 1.967926025390625, 0.8532028198242188, 2.6151123046875, 0.3983268737792969, 1.0170745849609375, 0.54144287109375, 0.1182403564453125, -1.35595703125, 0.1544036865234375, 1.4345703125, -1.65142822265625, 1.1844100952148438, 1.27740478515625, 0.113525390625, 2.121490478515625, 2.1113433837890625, 2.348907470703125, -0.225433349609375, -0.01617431640625, -1.859283447265625, -0.029941558837890625, 0.5450592041015625, 0.6582298278808594, 1.68890380859375, 0.4672088623046875, 0.42867279052734375, -0.5989532470703125, 0.8397369384765625, 0.018438339233398438, 3.43365478515625, -0.016845703125, 1.23504638671875, 0.46631622314453125, -0.2207794189453125, 1.058563232421875, -1.6118240356445312, 2.624755859375, 1.467437744140625, 0.13458251953125, 1.7484893798828125, -0.46868133544921875, -0.6421432495117188, -0.1327648162841797, 3.4617919921875, -4.068511962890625, -1.799072265625, 2.107574462890625, -2.644866943359375, 0.29265785217285156, 3.4698486328125, 0.990875244140625, -0.39539337158203125, 0.6165390014648438, 1.576751708984375, -3.83489990234375, 1.2599029541015625, 4.3779296875, 2.72869873046875, -0.74530029296875, -0.01802825927734375, 0.39836883544921875, 0.20452880859375, -0.0379180908203125, 0.4083251953125, -0.661956787109375, 0.261260986328125, 0.5758552551269531, -2.76739501953125, -1.20098876953125, 1.727325439453125, -2.38543701171875, 0.970001220703125, 3.53662109375, -1.179412841796875, 0.129608154296875, 3.08319091796875, 1.60498046875, 0.28363037109375, 3.1050872802734375, 0.09418487548828125, 0.67901611328125, -0.3262672424316406, 0.3946380615234375, 1.2289657592773438, 0.23484420776367188, 2.439178466796875, 0.7212142944335938, -0.00146484375, -5.22198486328125, -0.33428192138671875, -0.8795166015625, 0.80023193359375, 0.4277496337890625, -2.256103515625, -0.5350341796875, 0.5447921752929688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000049.npy"} {"epoch": 0.10261780104712041, "step": 50, "batch_size": 128, "mean": 0.49733060598373413, "std": 1.5684844255447388, "min": -4.664825439453125, "p10": -1.1613372802734374, "median": 0.4150657653808594, "p90": 2.43294677734375, "max": 5.4403076171875, "pos_frac": 0.65625, "sample": [0.2868499755859375, -0.2486572265625, 4.5264892578125, 0.29160499572753906, 0.78240966796875, -0.6981658935546875, 0.982666015625, -0.3591156005859375, 0.119903564453125, 5.4403076171875, 0.7716064453125, -0.24664878845214844, -0.47247314453125, 2.427490234375, 2.59820556640625, 1.52685546875, 1.0911712646484375, -4.664825439453125, 0.19073486328125, 0.46566009521484375, 0.417999267578125, -1.84222412109375, -2.50006103515625, 0.2476654052734375, -0.22662353515625, 1.87969970703125, 1.5421218872070312, -0.177459716796875, 1.255615234375, -1.119842529296875, -0.789886474609375, 4.2021484375, 0.774658203125, 0.172882080078125, 0.0803375244140625, -1.167724609375, -0.48758697509765625, 0.033203125, 0.755950927734375, -0.136016845703125, -0.2657737731933594, -1.025665283203125, 1.6009521484375, 2.4456787109375, -0.22357177734375, 0.79681396484375, -0.3826904296875, 1.67926025390625, -0.31081581115722656, 0.41213226318359375, 2.450439453125, 0.326202392578125, -0.111358642578125, -0.023326873779296875, 0.455474853515625, 0.944122314453125, 3.107391357421875, -2.5143890380859375, 2.300567626953125, 2.416259765625, 1.0498199462890625, 1.620880126953125, -3.00390625, -0.13701248168945312, 0.06976318359375, -1.529541015625, 0.00592041015625, 1.0063934326171875, 1.6195220947265625, -0.2544384002685547, 1.08184814453125, 0.095733642578125, -0.4878273010253906, 0.6309356689453125, 0.26963043212890625, -2.97802734375, 2.64483642578125, 0.8558349609375, 1.7570343017578125, -0.04552268981933594, -0.84893798828125, -0.354034423828125, -3.208648681640625, -0.702972412109375, 0.8503017425537109, 3.027587890625, -2.02923583984375, -2.81201171875, 1.563812255859375, -2.404571533203125, 0.880340576171875, 0.04437255859375, 2.029632568359375, 1.261505126953125, 0.930511474609375, 2.4610595703125, 0.45282745361328125, 0.2434539794921875, 1.194580078125, 0.88909912109375, 2.229156494140625, 3.873870849609375, 0.152374267578125, 1.06488037109375, 0.511260986328125, 0.22661590576171875, 0.4956321716308594, 0.998809814453125, -0.469573974609375, -1.158599853515625, -1.22808837890625, 1.225555419921875, 2.8695068359375, 1.85919189453125, 0.299652099609375, -0.48321533203125, 2.54840087890625, 2.31866455078125, 0.149261474609375, 0.5288639068603516, 2.2479248046875, 0.46532440185546875, 1.789215087890625, -1.028594970703125, 1.40533447265625, -0.81488037109375, -0.3030967712402344, 2.339691162109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000050.npy"} {"epoch": 0.10471204188481675, "step": 51, "batch_size": 128, "mean": 0.6575089693069458, "std": 1.9014719724655151, "min": -5.57635498046875, "p10": -1.400432205200195, "median": 0.485748291015625, "p90": 3.16890869140625, "max": 5.8289794921875, "pos_frac": 0.640625, "sample": [1.43267822265625, 0.5980300903320312, 1.6061153411865234, 1.141510009765625, 1.4995956420898438, -1.2663650512695312, 0.877532958984375, 4.374969482421875, -0.492218017578125, -0.280364990234375, 1.4797821044921875, 3.0553436279296875, 1.702545166015625, 2.185943603515625, -1.2944450378417969, -0.9719085693359375, -0.14794921875, -0.30853271484375, 0.2778472900390625, 1.945465087890625, -5.57635498046875, 0.202667236328125, 0.251312255859375, 0.1545867919921875, 2.7527923583984375, 5.8289794921875, -0.3324317932128906, -0.31884765625, 3.7537078857421875, -0.246612548828125, -2.25396728515625, 3.74774169921875, 3.157958984375, 1.7641448974609375, 1.0959396362304688, 3.0830078125, 1.08905029296875, 1.278228759765625, 0.67108154296875, 0.7239151000976562, 1.1155242919921875, 1.435150146484375, -2.2458648681640625, 2.2535400390625, -0.42270660400390625, 2.911285400390625, -0.1202850341796875, 0.13788604736328125, -1.684326171875, -3.0159912109375, 3.861053466796875, 0.0077667236328125, -3.936676025390625, 1.27960205078125, 0.82220458984375, 2.6044158935546875, 0.4171600341796875, 1.6689910888671875, 0.234130859375, 2.9911651611328125, 0.2279052734375, 0.4913330078125, -0.4597320556640625, 0.9094009399414062, -3.832763671875, -1.833038330078125, -0.09499359130859375, 0.0703582763671875, 0.7719268798828125, 2.8824462890625, -0.43859100341796875, 0.492431640625, 0.62689208984375, 3.755859375, -2.73876953125, 0.0, 3.794891357421875, -0.687713623046875, 1.9285354614257812, 1.58990478515625, 0.4420166015625, -0.6749839782714844, 0.000274658203125, 0.5692596435546875, -1.647735595703125, -0.454437255859375, -0.487823486328125, -0.5811691284179688, -0.647796630859375, 2.25140380859375, -0.550933837890625, -0.718475341796875, -0.1609039306640625, -0.5513763427734375, -0.1645660400390625, 1.7237548828125, -0.007293701171875, 1.1109619140625, 2.5472564697265625, -0.1345977783203125, -3.587158203125, -0.8835906982421875, -0.904876708984375, 5.4718017578125, 4.348388671875, 3.4368896484375, 1.9070663452148438, 3.2484130859375, 0.803497314453125, 1.910003662109375, 3.1944580078125, 0.4252815246582031, 0.233001708984375, 1.4098129272460938, 2.039947509765625, 3.833892822265625, -2.5103912353515625, -3.281402587890625, 1.46429443359375, 1.422576904296875, 0.48016357421875, 0.03082275390625, 0.14959716796875, -0.4727325439453125, 1.10137939453125, 1.0246734619140625, 0.0602874755859375, -0.0705718994140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000051.npy"} {"epoch": 0.1068062827225131, "step": 52, "batch_size": 128, "mean": 0.8414908647537231, "std": 1.9173743724822998, "min": -5.42578125, "p10": -1.203376770019531, "median": 0.4819793701171875, "p90": 3.072167968749999, "max": 5.839752197265625, "pos_frac": 0.6875, "sample": [0.414093017578125, 1.435943603515625, -0.2769775390625, -1.650390625, -0.01082611083984375, -0.2524566650390625, 2.496368408203125, 0.7155380249023438, -0.07545089721679688, 4.18768310546875, -0.300628662109375, 2.646026611328125, 0.984161376953125, -5.42578125, 2.583984375, 4.3702392578125, 0.97625732421875, 1.4205169677734375, -0.22491455078125, -0.45513153076171875, 1.99688720703125, 1.0298118591308594, 0.29437255859375, 0.9550552368164062, 0.8331222534179688, -0.9887924194335938, 0.418975830078125, 2.572845458984375, 1.8856353759765625, -1.7300872802734375, -2.19598388671875, -0.7756271362304688, -0.3102130889892578, 1.1531219482421875, 0.3203125, 0.15381622314453125, -0.0665130615234375, -1.83203125, 0.19830703735351562, 1.68359375, 3.77044677734375, 0.5894279479980469, -1.3058624267578125, 0.51605224609375, 2.0333251953125, 1.0429306030273438, 0.282958984375, 4.485595703125, -0.0697021484375, 3.87451171875, 1.16473388671875, 1.6057586669921875, 0.33502197265625, 0.2723846435546875, -0.12303924560546875, 2.526336669921875, 0.23247528076171875, 1.298095703125, 0.036285400390625, -0.753753662109375, 0.5779647827148438, -0.5361328125, 2.627288818359375, -1.91180419921875, 2.18853759765625, 2.53302001953125, -3.5693359375, 0.08642959594726562, 2.0510482788085938, 0.08919429779052734, -1.1275634765625, 1.05712890625, 2.8684844970703125, 2.6940155029296875, -1.66119384765625, -2.404022216796875, 0.8535575866699219, -0.69091796875, 2.1714935302734375, 1.25848388671875, 3.2733154296875, 5.65667724609375, 0.1550750732421875, 0.09842681884765625, 2.929168701171875, -1.35845947265625, -0.4276008605957031, 2.8958740234375, 5.167327880859375, 2.53363037109375, 1.9058151245117188, -0.6934967041015625, -0.1623382568359375, 4.817657470703125, -1.159454345703125, 0.5186805725097656, -0.927520751953125, 5.590545654296875, 0.447906494140625, 0.26613616943359375, 0.642578125, 1.035736083984375, 0.1547393798828125, -2.22540283203125, 5.4398193359375, 2.027263641357422, -0.01495361328125, -0.379669189453125, 5.839752197265625, 1.915008544921875, 1.45501708984375, 2.9859619140625, -0.6875839233398438, -0.6735000610351562, 1.7333221435546875, 0.05782318115234375, 0.1525115966796875, 0.3557853698730469, 0.41961669921875, 1.37933349609375, 2.539276123046875, 1.4490966796875, -3.669342041015625, -0.4763069152832031, 0.5910491943359375, 0.23082351684570312, 3.721435546875, 0.05974578857421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000052.npy"} {"epoch": 0.10890052356020942, "step": 53, "batch_size": 128, "mean": 0.78252112865448, "std": 2.0612473487854004, "min": -6.02130126953125, "p10": -1.3825851440429686, "median": 0.6467208862304688, "p90": 3.264949035644531, "max": 8.1617431640625, "pos_frac": 0.6484375, "sample": [1.8389892578125, -1.6428375244140625, 0.9050445556640625, -1.5992202758789062, 1.5690765380859375, 0.951446533203125, 3.3328704833984375, -1.5703887939453125, -1.469818115234375, -0.4607391357421875, 3.5577392578125, 0.2706298828125, -0.58843994140625, -3.660369873046875, 1.537872314453125, 2.7957763671875, 1.80084228515625, -0.33536529541015625, 1.0545196533203125, 1.495635986328125, -0.388336181640625, -0.714111328125, 0.633026123046875, 2.341796875, 3.23583984375, 5.308349609375, 2.899200439453125, 5.6119384765625, -0.33502197265625, 4.836029052734375, 2.0196990966796875, 3.99151611328125, -0.2858390808105469, 0.4707794189453125, 0.8834762573242188, -0.31815338134765625, 1.066619873046875, -4.2044677734375, -0.20271873474121094, 1.36181640625, 1.830169677734375, 0.6210594177246094, 0.6367721557617188, -2.889251708984375, -0.2557029724121094, -1.5736083984375, -0.92999267578125, -0.242919921875, 1.56036376953125, 1.19134521484375, 0.341339111328125, 5.498382568359375, 1.7882537841796875, -1.695648193359375, 0.5186882019042969, -0.8347091674804688, 1.837432861328125, 0.2729988098144531, -0.8176345825195312, 2.82806396484375, 3.07257080078125, 0.17399978637695312, 0.033172607421875, 1.62860107421875, 0.7109832763671875, 0.37772369384765625, 2.85858154296875, -0.27008819580078125, 0.6566696166992188, 0.6591606140136719, 0.31203460693359375, 1.63519287109375, 4.9581298828125, 0.9131011962890625, 3.4088134765625, 2.03179931640625, 0.2512969970703125, -0.229217529296875, 2.0350341796875, 1.07501220703125, 2.1859130859375, 0.5309333801269531, 1.21856689453125, 0.273773193359375, -0.37969207763671875, 2.91949462890625, -6.02130126953125, 1.154449462890625, 0.058124542236328125, -1.0718536376953125, 1.3147125244140625, -1.0325164794921875, 3.89105224609375, -1.3451995849609375, 0.58709716796875, 8.1617431640625, -0.243377685546875, -2.583770751953125, -0.5478897094726562, -1.183074951171875, -0.2693367004394531, 1.5529632568359375, -0.28050994873046875, 0.6646728515625, -0.14414215087890625, -1.1495132446289062, 0.590606689453125, 2.979461669921875, 0.6312408447265625, -5.56396484375, -0.6727981567382812, -0.1509246826171875, 3.090545654296875, 2.1279144287109375, 1.3482666015625, 0.6583251953125, 3.83770751953125, 0.0, 1.944091796875, 0.745849609375, 1.44158935546875, -0.0948944091796875, 1.6206932067871094, -2.288604736328125, 1.708648681640625, 0.7081680297851562, -0.64324951171875, 3.9100341796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000053.npy"} {"epoch": 0.11099476439790576, "step": 54, "batch_size": 128, "mean": 0.8322997093200684, "std": 2.2783291339874268, "min": -6.41729736328125, "p10": -1.81824951171875, "median": 0.4670581817626953, "p90": 3.6246948242187496, "max": 8.357666015625, "pos_frac": 0.6484375, "sample": [4.0694580078125, 4.160888671875, 3.007171630859375, 2.666046142578125, 2.18927001953125, 1.285064697265625, -0.35595703125, 3.46466064453125, 0.07874488830566406, -3.391937255859375, 2.927642822265625, -1.8076019287109375, -0.34789276123046875, -2.2124481201171875, 5.01483154296875, -0.1568927764892578, 5.9854736328125, 0.6245498657226562, 0.662628173828125, 0.394927978515625, 2.1310272216796875, 4.281768798828125, -0.0405120849609375, 0.4130859375, -0.722900390625, 0.18975830078125, 3.5732421875, -0.1114654541015625, -1.1885986328125, 0.18295669555664062, 0.563934326171875, 0.1142578125, -1.91064453125, 0.6510696411132812, 0.71185302734375, 2.692535400390625, 5.9244384765625, -3.3453369140625, 2.9393310546875, 0.2978515625, 0.9342041015625, 0.4718589782714844, 3.5037841796875, -0.39044189453125, 5.024749755859375, -0.034515380859375, -4.45806884765625, 0.0850372314453125, -0.6165313720703125, -0.529327392578125, 2.65118408203125, 6.537322998046875, 0.9142303466796875, 0.4817047119140625, -0.07462501525878906, 0.320068359375, -0.4119873046875, 1.6765289306640625, 1.486724853515625, 0.46225738525390625, 2.658374786376953, 0.4937591552734375, -2.272064208984375, -1.397918701171875, 2.301513671875, 1.2296905517578125, -0.8385467529296875, 1.3158798217773438, 2.45758056640625, 2.8380126953125, 0.0277252197265625, 1.4372940063476562, 1.529083251953125, 0.793548583984375, -1.062774658203125, 0.7695159912109375, 3.41522216796875, -1.742706298828125, 0.1087799072265625, -0.1314258575439453, 0.08456039428710938, -0.6723556518554688, 0.8418045043945312, 0.15497589111328125, 1.0, 2.4735107421875, 0.1689453125, -1.8430938720703125, 1.8271064758300781, 0.4130401611328125, 0.44873046875, 3.7447509765625, -3.94451904296875, 2.430633544921875, 0.8370361328125, -0.2770538330078125, -1.39447021484375, 8.357666015625, 6.36236572265625, 2.592132568359375, 0.4136962890625, 3.899749755859375, 3.505615234375, -0.26641845703125, 3.0819091796875, -0.5208625793457031, -2.16082763671875, -0.0972442626953125, -6.41729736328125, 1.220947265625, -0.766632080078125, -2.1301116943359375, -0.43292236328125, -0.781341552734375, 4.3751220703125, -0.727264404296875, -1.3213882446289062, -0.8072509765625, 0.73919677734375, -0.3287353515625, 1.885284423828125, 3.14080810546875, 1.4146060943603516, -2.44329833984375, -2.13629150390625, 0.22711181640625, 1.6187744140625, 1.1726837158203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000054.npy"} {"epoch": 0.1130890052356021, "step": 55, "batch_size": 128, "mean": 0.5179433822631836, "std": 1.7774183750152588, "min": -5.4998779296875, "p10": -1.2817596435546874, "median": 0.28682708740234375, "p90": 2.450527954101562, "max": 5.927520751953125, "pos_frac": 0.609375, "sample": [-0.9971160888671875, 1.8499908447265625, 0.76861572265625, 0.101959228515625, -0.9574127197265625, -0.97698974609375, 0.8300704956054688, 1.10845947265625, 0.443328857421875, -2.081817626953125, 1.3540191650390625, 2.403900146484375, 2.365203857421875, -0.1007232666015625, -0.2397613525390625, 0.4539031982421875, 1.1758880615234375, -0.14208984375, 0.651123046875, 1.2026710510253906, 0.01497650146484375, 0.154052734375, 1.0592041015625, 2.82135009765625, 2.0751495361328125, 0.08151435852050781, 0.6819839477539062, 2.362823486328125, -0.0833740234375, 0.1792755126953125, -1.124542236328125, 0.02118682861328125, -1.38018798828125, 1.686767578125, -0.0265960693359375, 1.6597175598144531, 0.402069091796875, 0.827239990234375, -0.3053741455078125, 3.476776123046875, -4.230255126953125, -0.0808258056640625, -0.6793594360351562, -0.981964111328125, -0.673095703125, 1.1317520141601562, -0.593475341796875, -0.7631683349609375, 0.233978271484375, 4.229766845703125, 0.6901702880859375, -1.0782928466796875, -2.3597412109375, 0.38146209716796875, 0.3333740234375, -1.9855194091796875, -2.549652099609375, 1.3341064453125, -1.2948226928710938, 2.214996337890625, 0.2402801513671875, 0.577728271484375, 5.323760986328125, 0.15555191040039062, 0.801513671875, 0.698516845703125, -0.95428466796875, 2.331390380859375, 1.3309173583984375, -1.8138465881347656, -0.015960693359375, -0.09516143798828125, -0.7960205078125, 0.0093841552734375, -1.2761611938476562, 0.10191726684570312, -0.152130126953125, 2.3947296142578125, -0.47119903564453125, 1.3162994384765625, -1.0364303588867188, 3.9930419921875, -1.46722412109375, 0.6438789367675781, 0.0723419189453125, 0.4900779724121094, -0.8859195709228516, 3.6603851318359375, 3.711517333984375, -0.2164154052734375, -0.1629180908203125, 2.1733474731445312, 0.417205810546875, 0.03753662109375, 5.927520751953125, -0.08723640441894531, 2.95184326171875, -0.871917724609375, 0.712158203125, 3.30853271484375, -1.561431884765625, 1.2062606811523438, 0.5562896728515625, 2.22900390625, 0.9574737548828125, -2.6961212158203125, 2.559326171875, 1.1082000732421875, 2.2253036499023438, -1.0009918212890625, -5.4998779296875, 0.029876708984375, 2.2487030029296875, -0.2302703857421875, -0.5182037353515625, -2.09930419921875, 2.301605224609375, -1.06158447265625, 1.4027786254882812, -0.1234130859375, 1.1526412963867188, -0.12237548828125, -1.1939697265625, 2.021942138671875, 1.355682373046875, 5.05963134765625, 5.1691741943359375, 0.6651840209960938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000055.npy"} {"epoch": 0.11518324607329843, "step": 56, "batch_size": 128, "mean": 0.9814174771308899, "std": 2.3433401584625244, "min": -7.858734130859375, "p10": -1.1612274169921872, "median": 0.6239089965820312, "p90": 3.9653320312499982, "max": 7.865478515625, "pos_frac": 0.703125, "sample": [1.13909912109375, -3.6453704833984375, 2.903411865234375, 4.60418701171875, 0.6395111083984375, 1.5261383056640625, 1.47833251953125, 0.157928466796875, -0.3013763427734375, 6.2392578125, 1.7299652099609375, 1.560028076171875, 0.05587005615234375, -0.57080078125, 0.5272064208984375, 0.46038818359375, 0.18934249877929688, -0.600738525390625, 3.42901611328125, -0.399139404296875, 0.608306884765625, 0.442474365234375, 2.689239501953125, -1.412567138671875, 0.1763916015625, 4.38702392578125, 2.389312744140625, 4.83319091796875, 2.5430908203125, 0.6919403076171875, 2.765625, 3.491668701171875, 1.1148834228515625, -0.9833984375, 7.48504638671875, 2.1449432373046875, -2.255157470703125, 0.957763671875, 0.6909027099609375, -0.24362945556640625, 4.89080810546875, 1.686309814453125, 0.6439476013183594, 0.17159652709960938, 0.03426551818847656, 7.865478515625, 2.806304931640625, -0.20659828186035156, -1.0397186279296875, -0.25409698486328125, -0.8328857421875, -0.933929443359375, 1.140472412109375, -4.882080078125, 0.4824371337890625, -7.858734130859375, 0.15978240966796875, -0.46858978271484375, 1.3209228515625, 0.4201507568359375, 2.7312469482421875, 6.45452880859375, 4.445953369140625, 2.896514892578125, 0.163970947265625, -1.6039581298828125, 2.442626953125, 1.6892242431640625, 3.78460693359375, 0.873199462890625, -0.38237762451171875, -0.3629474639892578, -3.246246337890625, 0.6000461578369141, 0.053119659423828125, -0.2769012451171875, -1.326080322265625, 1.998779296875, -0.1058502197265625, 1.33953857421875, -4.804443359375, 1.9346923828125, 2.5435943603515625, -1.353302001953125, 1.4099502563476562, 6.047607421875, 0.38671875, 3.240142822265625, -0.5468902587890625, 2.2062339782714844, -0.018768310546875, 6.4774169921875, 1.1497802734375, 1.6494216918945312, -0.9952392578125, 5.54229736328125, 1.4072799682617188, -0.436370849609375, 0.15863800048828125, 0.711578369140625, 1.6300048828125, -0.397369384765625, 1.8069610595703125, -1.65594482421875, -2.80670166015625, 4.64630126953125, -0.9996795654296875, 2.6474609375, 0.850128173828125, 1.178985595703125, 1.0414276123046875, -0.384490966796875, 2.074859619140625, -1.4536285400390625, 0.4074077606201172, 0.3072662353515625, -1.090576171875, 0.032772064208984375, 0.0508880615234375, 0.4166259765625, 0.591094970703125, -0.7651596069335938, 1.3846435546875, 1.811126708984375, 0.11285400390625, 3.686767578125, 2.4599609375, 0.37296295166015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000056.npy"} {"epoch": 0.11727748691099477, "step": 57, "batch_size": 128, "mean": 1.3680766820907593, "std": 2.3784008026123047, "min": -3.1953277587890625, "p10": -1.0637542724609375, "median": 0.7627849578857422, "p90": 4.29403076171875, "max": 14.876953125, "pos_frac": 0.7421875, "sample": [-1.020355224609375, 3.35101318359375, -3.1953277587890625, 3.803070068359375, -0.706268310546875, -1.00628662109375, 3.90496826171875, 0.42108917236328125, -2.591552734375, 0.4125823974609375, -1.16650390625, -0.0909423828125, 3.904083251953125, 0.29607391357421875, 2.6655349731445312, 0.18832015991210938, 1.0597686767578125, 0.85113525390625, 2.606689453125, 0.9849853515625, 0.5567646026611328, 3.58526611328125, 0.5538654327392578, 0.1632080078125, 0.3401336669921875, 2.027801513671875, -1.324859619140625, -0.802642822265625, -0.7273025512695312, -0.7741622924804688, 1.2908935546875, 2.492889404296875, 1.408935546875, 0.6038360595703125, 7.0091552734375, 14.876953125, 0.7718505859375, -1.611175537109375, -1.2652816772460938, 3.622467041015625, 3.98638916015625, 0.8868789672851562, -1.341461181640625, 5.5449981689453125, 1.4668502807617188, 0.3777313232421875, -0.01033782958984375, 0.504364013671875, -1.15960693359375, 1.4681396484375, 1.4700927734375, 0.6682662963867188, 0.123077392578125, 6.66094970703125, 1.0071868896484375, 0.10638427734375, 0.9116287231445312, 0.9188385009765625, 2.173126220703125, 0.49957275390625, 1.366729736328125, 0.56268310546875, 0.09167671203613281, 1.849822998046875, 0.6019744873046875, 4.66790771484375, 4.18359375, 2.29217529296875, -1.411895751953125, 1.6851806640625, 4.606353759765625, 0.20758056640625, -2.0637588500976562, -0.6129722595214844, 3.0888671875, 2.74853515625, 0.9857177734375, 3.90216064453125, 2.1484222412109375, 2.516510009765625, 1.160980224609375, 0.965850830078125, 3.751953125, 4.958465576171875, -0.324615478515625, -0.38702392578125, -1.649322509765625, 3.65423583984375, -2.06353759765625, -0.6554412841796875, 2.3121871948242188, 4.241424560546875, 0.50372314453125, 0.7537193298339844, 0.0738525390625, 0.031352996826171875, 3.15863037109375, -0.9796524047851562, 0.36297607421875, -0.943115234375, -2.15972900390625, -1.022674560546875, 5.38201904296875, 2.2252197265625, 0.531097412109375, 5.14385986328125, 0.5297508239746094, 4.416778564453125, 0.6699371337890625, -0.38539886474609375, -0.37640380859375, 0.743621826171875, 5.145751953125, -0.676513671875, 2.3681259155273438, 2.844818115234375, 2.23236083984375, -0.4827251434326172, 0.7195854187011719, 0.2616100311279297, 2.936981201171875, -0.365142822265625, 4.91552734375, 1.841552734375, 2.9240493774414062, 0.04465484619140625, 4.112548828125, 4.5149078369140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000057.npy"} {"epoch": 0.1193717277486911, "step": 58, "batch_size": 128, "mean": 0.973612904548645, "std": 2.0907115936279297, "min": -10.107086181640625, "p10": -1.4001953125, "median": 0.8843727111816406, "p90": 3.6880981445312497, "max": 5.8919677734375, "pos_frac": 0.6640625, "sample": [0.74786376953125, 4.213226318359375, -0.707305908203125, 3.676910400390625, 1.1027679443359375, -0.18958663940429688, 2.1210098266601562, 0.72991943359375, 2.38720703125, 4.15875244140625, 2.1226272583007812, 3.06573486328125, 1.6746826171875, 4.52496337890625, 1.793670654296875, -0.9122467041015625, 4.24169921875, 2.37396240234375, 2.2471771240234375, -1.0849609375, 5.22845458984375, 4.44744873046875, 0.174774169921875, -1.43505859375, 1.29705810546875, 1.7906646728515625, 2.8085174560546875, 1.9969482421875, 0.84454345703125, -1.88140869140625, -0.009677886962890625, 1.3437042236328125, 0.6441650390625, -0.12346076965332031, 2.76708984375, 0.6637802124023438, 1.174530029296875, 1.3703460693359375, -0.3390350341796875, -1.38525390625, -2.38360595703125, 1.368927001953125, 0.9062118530273438, 0.604522705078125, 3.143310546875, -1.2736129760742188, 0.5222949981689453, 3.386260986328125, -0.14609527587890625, -10.107086181640625, -0.1849822998046875, 4.968994140625, -0.5418701171875, -1.8956375122070312, 0.8525390625, 0.4993438720703125, -0.35369110107421875, 1.094552993774414, -0.30743408203125, 4.135498046875, 0.9127044677734375, -0.087310791015625, 2.1046295166015625, -0.1622161865234375, -0.055206298828125, 0.552398681640625, -0.41864013671875, 1.986480712890625, 0.828369140625, 0.819580078125, -1.352783203125, -1.7878990173339844, -0.9620742797851562, 2.596099853515625, 1.048187255859375, 0.347869873046875, 5.852203369140625, 0.8625335693359375, -1.63616943359375, -0.3690948486328125, 2.0846099853515625, -2.55401611328125, 5.8919677734375, -3.406524658203125, 1.187530517578125, 1.3773040771484375, -0.1552886962890625, 0.25110626220703125, 0.18973541259765625, 3.7574462890625, -0.04962921142578125, 2.066986083984375, 0.53863525390625, 1.485260009765625, 2.1468505859375, -0.119842529296875, 3.714202880859375, -1.9117431640625, 0.97607421875, 1.618560791015625, 1.754791259765625, -0.25969505310058594, 0.0109405517578125, -1.718414306640625, 2.5315399169921875, 1.1723480224609375, 0.9404792785644531, 2.181884765625, 3.14276123046875, 2.615142822265625, -1.77288818359375, 3.01593017578125, 0.5922470092773438, 4.202667236328125, 1.80718994140625, 0.607818603515625, -0.4296875, -0.35394287109375, 1.9199295043945312, 2.966644287109375, 2.2213287353515625, 1.322723388671875, -1.02008056640625, 2.148101806640625, -0.14124298095703125, -1.632110595703125, 3.402984619140625, -0.73046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000058.npy"} {"epoch": 0.12146596858638743, "step": 59, "batch_size": 128, "mean": 0.877183198928833, "std": 2.7218892574310303, "min": -7.11505126953125, "p10": -1.7987289428710938, "median": 0.5944366455078125, "p90": 4.461618041992186, "max": 8.607818603515625, "pos_frac": 0.6328125, "sample": [-0.593048095703125, 1.64215087890625, -1.87567138671875, -7.11505126953125, 1.3682785034179688, 1.469024658203125, 0.2386474609375, 1.627471923828125, 1.0534210205078125, -0.85943603515625, 0.3713188171386719, 1.4605560302734375, 1.6613616943359375, 3.164306640625, -3.106719970703125, 0.569549560546875, -0.3683624267578125, -0.3262367248535156, -1.864715576171875, 1.3272857666015625, -0.9222869873046875, 1.2803382873535156, 0.05224418640136719, -0.0148773193359375, 1.8215789794921875, 1.2356491088867188, 1.6644287109375, 2.487091064453125, -3.503082275390625, 0.15179443359375, -0.454132080078125, 0.7665519714355469, -0.12640380859375, 2.142059326171875, 0.0, 7.2908935546875, 0.11925315856933594, 0.8608245849609375, -0.46166419982910156, -3.700042724609375, -0.017658233642578125, 2.57763671875, 6.1993408203125, 7.87554931640625, 1.112640380859375, -5.62347412109375, 1.2179946899414062, 8.607818603515625, 3.2744140625, 8.337890625, 1.3185958862304688, -0.342376708984375, 6.0989990234375, 6.0435791015625, 0.140411376953125, 1.5120010375976562, -0.13079071044921875, 2.344268798828125, -5.4996337890625, 0.0435791015625, -0.819122314453125, 2.2065887451171875, 1.25762939453125, 0.178131103515625, 0.3534698486328125, 4.2999267578125, 4.864837646484375, -1.4166259765625, 3.427490234375, 1.7893447875976562, -1.0522613525390625, 2.4495849609375, 0.61932373046875, 5.99639892578125, -1.9725151062011719, -5.571044921875, 1.0307464599609375, -3.24273681640625, 0.4161834716796875, 0.3288116455078125, 2.96038818359375, 0.9512939453125, 2.883941650390625, 1.1176605224609375, 1.62969970703125, -1.8333892822265625, 1.44061279296875, 0.2715301513671875, -1.6428070068359375, -0.3213043212890625, 2.506805419921875, 2.3141021728515625, -0.164459228515625, -0.5227508544921875, -1.2935562133789062, 5.423187255859375, 2.2701416015625, -0.0810546875, -1.074615478515625, -1.599029541015625, -0.231292724609375, -0.4083251953125, 0.7626419067382812, 1.0853271484375, -0.44921875, 1.3189697265625, -1.6697845458984375, -3.55279541015625, -1.5059280395507812, 1.88238525390625, 2.58935546875, 2.332122802734375, 7.568603515625, -0.173828125, -1.0059928894042969, 0.2849273681640625, 8.1729736328125, 1.417083740234375, 1.430389404296875, 0.0176849365234375, -1.78387451171875, 0.5347137451171875, 4.838897705078125, 1.714111328125, 0.04251861572265625, -0.47486114501953125, 1.9508056640625, -0.41385650634765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000059.npy"} {"epoch": 0.12356020942408377, "step": 60, "batch_size": 128, "mean": 1.1955628395080566, "std": 2.3934524059295654, "min": -6.23626708984375, "p10": -1.0806922912597654, "median": 0.9897689819335938, "p90": 4.155944061279296, "max": 8.77911376953125, "pos_frac": 0.6953125, "sample": [-5.645751953125, 2.811614990234375, 4.705413818359375, -0.664825439453125, 2.68743896484375, 3.4835662841796875, -0.3598060607910156, 3.2703857421875, 1.646087646484375, 1.492431640625, 0.4420318603515625, -0.7126235961914062, -0.23362159729003906, 1.630401611328125, 0.4925079345703125, 0.0606689453125, 0.9729766845703125, -0.229949951171875, 2.123016357421875, 4.69268798828125, -0.054718017578125, -2.025604248046875, 3.6270751953125, -0.5545196533203125, -1.2587203979492188, 0.4341278076171875, 0.8631439208984375, 0.0, -0.5990772247314453, 4.5714111328125, -0.145660400390625, 5.181640625, 0.965179443359375, 1.904876708984375, 0.36358642578125, -0.39532470703125, 4.394195556640625, -3.273834228515625, 2.5936431884765625, -0.021945953369140625, -0.5932998657226562, 1.9278182983398438, -0.690673828125, 2.952911376953125, 1.2234764099121094, -0.08160400390625, -1.5333309173583984, 1.0684814453125, 0.429595947265625, 0.2896575927734375, 0.210662841796875, -0.64849853515625, 0.36431884765625, 3.700408935546875, 0.06240081787109375, -1.36810302734375, 3.7787322998046875, -2.2545166015625, 1.57476806640625, 4.234519958496094, 0.8997116088867188, 3.0650634765625, 3.5058135986328125, 0.43724632263183594, -6.23626708984375, 1.8224029541015625, 2.16650390625, 0.16147613525390625, -0.4011383056640625, -0.5538330078125, -1.9866943359375, 3.8828125, 3.3406295776367188, 4.27020263671875, 2.5390625, 1.247528076171875, -0.41562652587890625, 2.6053619384765625, 1.6649398803710938, 7.1210479736328125, 4.1222686767578125, 0.6170196533203125, -0.4776649475097656, 1.4806060791015625, 4.015106201171875, 0.4409637451171875, 1.601654052734375, -3.447906494140625, 1.3496971130371094, 3.0691986083984375, 2.186798095703125, 0.622222900390625, -4.95196533203125, 0.3635063171386719, 1.046142578125, 3.68756103515625, 0.4747314453125, 5.460205078125, -0.8673477172851562, -0.42266845703125, 1.101837158203125, 5.14599609375, 0.9216766357421875, -2.87152099609375, 4.37542724609375, 3.56817626953125, -0.6796875, 1.9849090576171875, 8.77911376953125, 0.0, 6.56072998046875, -1.00439453125, 1.11444091796875, 0.441070556640625, -0.5708885192871094, 0.5524139404296875, 2.122772216796875, 0.5977859497070312, 2.2534332275390625, -4.36767578125, 2.464202880859375, 3.2490997314453125, 1.8306808471679688, 3.91748046875, 2.789947509765625, 1.006561279296875, 2.420379638671875, 1.939849853515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000060.npy"} {"epoch": 0.1256544502617801, "step": 61, "batch_size": 128, "mean": 1.1737103462219238, "std": 2.4435060024261475, "min": -5.0535888671875, "p10": -1.486757659912109, "median": 0.8888044357299805, "p90": 4.5412445068359375, "max": 7.82220458984375, "pos_frac": 0.7109375, "sample": [-1.28314208984375, -1.1572265625, 1.2534828186035156, 0.13232421875, 5.5985107421875, 4.1219482421875, -0.3542022705078125, 0.33447265625, 0.1378498077392578, 1.225341796875, 1.790496826171875, -3.400421142578125, -2.21063232421875, 1.2401885986328125, -0.2829742431640625, 0.8954906463623047, -4.384552001953125, -1.4062347412109375, -3.7955322265625, -0.6389236450195312, 0.8170166015625, 0.8821182250976562, 6.0205078125, 5.373016357421875, 3.63751220703125, -0.19427490234375, 3.399627685546875, 0.0975341796875, 3.175323486328125, 2.6208038330078125, 0.3195629119873047, 0.7152900695800781, 0.4032173156738281, -0.2822608947753906, 5.269744873046875, 0.79693603515625, -0.28940582275390625, 3.856292724609375, 1.060272216796875, 0.5284996032714844, -0.18674468994140625, 1.020355224609375, 2.926177978515625, -0.4449920654296875, 1.792694091796875, 0.635833740234375, 4.60626220703125, 4.601593017578125, 1.73297119140625, 1.787567138671875, 0.01617431640625, 3.3153076171875, 5.5498046875, 5.8388671875, 3.65887451171875, 1.4412841796875, 1.9657440185546875, 0.7478961944580078, 7.2083740234375, -0.638519287109375, 0.76959228515625, 2.7961769104003906, 1.02880859375, -1.0105667114257812, 0.9145965576171875, 4.03460693359375, 4.974639892578125, 4.515380859375, 3.824432373046875, 0.8059844970703125, 1.69012451171875, 0.542205810546875, 0.9429702758789062, -0.92327880859375, -1.2613410949707031, 1.553863525390625, 0.019794464111328125, 0.486968994140625, 1.3748321533203125, -1.924407958984375, 6.552459716796875, 0.60394287109375, 3.69622802734375, 2.89068603515625, 1.748779296875, 2.2630081176757812, 0.45467376708984375, -1.9798049926757812, -0.7001953125, 1.056610107421875, -0.37200164794921875, -2.132568359375, -1.0810546875, 1.257568359375, 0.20733261108398438, 1.4356765747070312, 0.6944236755371094, 3.77374267578125, -0.1386260986328125, 7.126220703125, -0.3707275390625, 7.82220458984375, 0.982177734375, 0.0, 0.50982666015625, 1.6686172485351562, 3.1095428466796875, 1.02618408203125, 1.4907684326171875, -3.6126708984375, 4.29583740234375, -2.211395263671875, -0.5376129150390625, -0.4683990478515625, -4.736968994140625, 2.6523284912109375, -5.0535888671875, 0.1530303955078125, -1.840301513671875, 1.705596923828125, 1.337890625, 0.5495223999023438, -0.7817306518554688, 1.3846206665039062, 2.026508331298828, 4.292022705078125, 0.4046783447265625, -1.6746444702148438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000061.npy"} {"epoch": 0.12774869109947645, "step": 62, "batch_size": 128, "mean": 1.1402709484100342, "std": 2.701298236846924, "min": -5.1613311767578125, "p10": -2.379899597167969, "median": 0.9509963989257812, "p90": 4.119326782226563, "max": 9.088226318359375, "pos_frac": 0.703125, "sample": [4.118988037109375, 0.5977935791015625, 8.91192626953125, 3.7827911376953125, 4.216033935546875, 0.20402145385742188, 1.376708984375, 2.501434326171875, -3.575927734375, -0.4311981201171875, -2.3454437255859375, -1.0571823120117188, -0.49236297607421875, 3.629852294921875, 2.54022216796875, 1.5691375732421875, 1.567657470703125, 2.2733230590820312, -0.7247161865234375, -1.3060455322265625, 1.8637619018554688, 0.6491928100585938, 3.246368408203125, 0.8831100463867188, -3.953582763671875, 6.46435546875, -5.1613311767578125, 0.550872802734375, 0.7503509521484375, 1.29656982421875, 1.156463623046875, 1.774566650390625, 2.91583251953125, 1.3807220458984375, 0.03179168701171875, 0.14676856994628906, 2.07421875, 0.2601776123046875, 2.124359130859375, 0.3553466796875, 0.9701080322265625, 3.8577880859375, 4.1832122802734375, -0.43675994873046875, -2.844268798828125, 1.243865966796875, 0.4151039123535156, 2.46624755859375, 2.87908935546875, 9.088226318359375, -4.48040771484375, 4.1201171875, -0.7628173828125, 2.534698486328125, 1.66943359375, -3.796539306640625, 7.339111328125, 0.0, 3.74932861328125, 0.29305267333984375, -3.415313720703125, -2.1291046142578125, 5.0792236328125, 0.3156280517578125, 1.7787017822265625, -0.7186508178710938, 0.8772430419921875, -0.18511962890625, 0.248016357421875, 2.78033447265625, 3.83038330078125, 1.6078338623046875, 0.4906730651855469, -3.037567138671875, 2.3844757080078125, 3.748687744140625, 1.038421630859375, -2.53570556640625, 0.11767578125, 2.094453811645508, -0.09769439697265625, -1.790863037109375, 2.10076904296875, 2.42919921875, -1.9361572265625, 3.74176025390625, 7.756500244140625, 1.62603759765625, 0.067291259765625, 2.96002197265625, 0.19853782653808594, -1.169281005859375, 9.0623779296875, 3.3424072265625, 4.968658447265625, 3.08807373046875, -2.71575927734375, 6.3226318359375, -2.647369384765625, -1.5277099609375, 3.8519439697265625, 0.466888427734375, 1.0354080200195312, 2.150238037109375, 5.563262939453125, 1.4411468505859375, 1.353973388671875, 0.38703155517578125, 0.37615203857421875, 3.73040771484375, 1.7459716796875, -1.16143798828125, -0.707366943359375, -2.4793701171875, 0.931884765625, -1.1889572143554688, -1.1722412109375, -0.479766845703125, 2.8462142944335938, 0.26191139221191406, 0.471435546875, -1.947998046875, 2.6002197265625, 1.3993682861328125, -0.9746551513671875, 0.570068359375, -2.460296630859375, -1.4619903564453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000062.npy"} {"epoch": 0.12984293193717278, "step": 63, "batch_size": 128, "mean": 1.3663294315338135, "std": 2.9731767177581787, "min": -7.88165283203125, "p10": -2.2385795593261717, "median": 1.2370796203613281, "p90": 4.6413635253906245, "max": 10.63037109375, "pos_frac": 0.6796875, "sample": [3.56585693359375, 4.6199951171875, 0.3754863739013672, -7.88165283203125, -4.68603515625, -0.3637256622314453, 2.6307373046875, 0.9152984619140625, 0.2318572998046875, 3.217376708984375, -3.874176025390625, -2.7513961791992188, -2.177154541015625, -2.2324295043945312, 2.4366455078125, -0.37115478515625, 1.445037841796875, 4.69122314453125, 6.45751953125, 7.119781494140625, -2.2529296875, -2.686920166015625, 8.70599365234375, -3.288238525390625, 2.8648681640625, 1.837493896484375, 2.96392822265625, -3.1389694213867188, 0.270050048828125, 4.1392364501953125, -0.043933868408203125, 0.43182373046875, -1.1118316650390625, 0.27371978759765625, 1.5342655181884766, 10.63037109375, -3.58306884765625, 1.8487091064453125, -2.04400634765625, 4.268280029296875, -0.490234375, 0.7511520385742188, 5.245147705078125, -4.136688232421875, 1.409942626953125, 2.639495849609375, 8.2684326171875, -6.37615966796875, -0.0943603515625, 5.33843994140625, 0.46587181091308594, 5.598846435546875, 3.3292236328125, 0.0, -0.5920982360839844, 2.3756103515625, 3.89654541015625, -1.6471176147460938, -0.15564727783203125, 1.1462821960449219, 1.1175537109375, 0.121917724609375, 0.82720947265625, 4.600128173828125, -1.442901611328125, 0.11333465576171875, 3.677490234375, 2.3016357421875, -1.5167388916015625, 4.475830078125, 5.70977783203125, 1.6499481201171875, 3.30712890625, -0.4290771484375, 3.81549072265625, -3.11529541015625, 3.1461105346679688, 1.49749755859375, 1.8170928955078125, -0.23150634765625, 0.6060791015625, 9.187469482421875, 0.8932037353515625, -0.5587310791015625, 2.9324798583984375, 0.33350372314453125, 2.348133087158203, 0.0504608154296875, 1.1543807983398438, 2.3095779418945312, 2.0884857177734375, -0.65850830078125, -1.49725341796875, 3.471527099609375, 0.6153087615966797, 3.788665771484375, -0.217193603515625, 3.71624755859375, 2.5477447509765625, 3.3895263671875, 3.67901611328125, 3.2201995849609375, 4.056976318359375, -0.3019561767578125, 2.3299484252929688, 0.7603607177734375, -0.422454833984375, -0.34151458740234375, 3.84429931640625, 1.1233978271484375, 3.083740234375, -0.4328155517578125, 2.7750091552734375, 2.330413818359375, 1.6340255737304688, 4.867919921875, 0.6608123779296875, 1.3197784423828125, 1.94140625, 1.0235366821289062, -3.375518798828125, -1.2808380126953125, 2.28656005859375, 1.831756591796875, -1.333953857421875, -0.15909576416015625, 4.123870849609375, 5.740936279296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000063.npy"} {"epoch": 0.1319371727748691, "step": 64, "batch_size": 128, "mean": 1.4827189445495605, "std": 3.2560784816741943, "min": -7.3162841796875, "p10": -2.140666198730468, "median": 1.0321903228759766, "p90": 5.762124633789062, "max": 11.1207275390625, "pos_frac": 0.7109375, "sample": [4.181243896484375, 0.87567138671875, -3.1875, 8.93927001953125, -2.8256988525390625, 4.31097412109375, 2.2865753173828125, 1.87542724609375, 7.7337646484375, 0.7244415283203125, 2.3996124267578125, -0.35052490234375, 6.08001708984375, 2.9739990234375, 1.487091064453125, -4.5955810546875, 7.18255615234375, 1.9794769287109375, 0.2671356201171875, -4.9866943359375, 1.4357452392578125, 0.9590301513671875, -1.9376983642578125, -2.6142578125, 0.7566909790039062, -4.968292236328125, 2.4821510314941406, 0.5618896484375, 1.1707763671875, 4.3221435546875, 0.65203857421875, 2.004547119140625, 2.574462890625, 0.4263801574707031, 2.23724365234375, -0.10295867919921875, 0.575225830078125, 2.8149185180664062, 2.1718978881835938, 0.545501708984375, 2.632080078125, -0.2588043212890625, -1.26507568359375, 1.945068359375, -0.9084014892578125, 9.226104736328125, 8.459716796875, -1.1068572998046875, 7.55413818359375, 0.8564910888671875, 5.53179931640625, -1.6593475341796875, 0.25074195861816406, 0.72796630859375, 0.8196067810058594, -0.7482452392578125, -3.275238037109375, 2.5589675903320312, 4.9280853271484375, 2.298675537109375, 3.28515625, 3.4686737060546875, -1.059051513671875, 4.9336395263671875, 1.567535400390625, -0.189117431640625, -0.22078704833984375, -5.7744140625, 0.6907958984375, 1.870208740234375, -0.4884033203125, 0.31365966796875, 9.63250732421875, 1.664337158203125, 0.5830955505371094, 1.405029296875, 2.74432373046875, -0.22833251953125, -0.8704681396484375, 0.121307373046875, -1.6288414001464844, 6.489166259765625, -0.9100151062011719, 0.12252044677734375, 5.0767822265625, 5.772186279296875, 0.9491424560546875, 1.4615631103515625, 1.0155982971191406, -3.34918212890625, 1.3631439208984375, 11.1207275390625, 3.685211181640625, 6.31475830078125, 0.29364013671875, 0.94879150390625, -7.3162841796875, 1.0487823486328125, -0.96923828125, -0.70941162109375, 3.750274658203125, 4.6875, 2.964385986328125, 0.0, 0.20759201049804688, 0.32253265380859375, -0.2330322265625, 1.9015655517578125, 4.3524932861328125, -3.206207275390625, 5.7578125, 1.2042808532714844, -0.47725677490234375, 5.322723388671875, 2.3001251220703125, -0.8864555358886719, 0.054351806640625, 7.357635498046875, -4.6583251953125, 1.0703582763671875, 0.0, 2.24627685546875, 5.60772705078125, 2.1884994506835938, 4.4689788818359375, 2.6772308349609375, -5.47210693359375, 0.0641632080078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000064.npy"} {"epoch": 0.13403141361256546, "step": 65, "batch_size": 128, "mean": 1.1228067874908447, "std": 2.8196301460266113, "min": -7.56103515625, "p10": -1.819923400878906, "median": 0.9190139770507812, "p90": 4.964999389648437, "max": 8.854736328125, "pos_frac": 0.6484375, "sample": [-0.09920120239257812, -2.7845458984375, 2.624267578125, 0.015390396118164062, 6.614013671875, 1.3665084838867188, 0.8530120849609375, 2.91046142578125, 2.7095184326171875, -5.2598876953125, -3.524200439453125, -4.58563232421875, -0.70050048828125, 3.672607421875, -1.5909996032714844, 4.9146728515625, 3.579986572265625, -2.5643310546875, 0.9682655334472656, -1.551849365234375, -3.364013671875, 0.3428306579589844, 3.7710113525390625, 5.145172119140625, 2.458738327026367, 2.611083984375, 1.7991943359375, 1.4898834228515625, -3.0714111328125, -1.15545654296875, -0.1468963623046875, -0.9135589599609375, -0.7786026000976562, 2.6021728515625, 8.854736328125, -0.4469146728515625, 0.22119140625, 5.1884765625, -0.36151885986328125, -0.5090484619140625, 3.120147705078125, -3.0131988525390625, 1.69281005859375, 1.5519027709960938, 6.3826904296875, -0.2009124755859375, 5.082427978515625, 2.8274993896484375, 5.844696044921875, 0.52423095703125, 2.8813552856445312, 1.9022064208984375, -1.44195556640625, 2.49713134765625, 2.64288330078125, -0.667327880859375, 5.123443603515625, 0.9274444580078125, 0.3353004455566406, 2.4474868774414062, 0.62554931640625, 0.208831787109375, 0.8216018676757812, -1.57763671875, 1.824462890625, 0.91058349609375, 1.6196136474609375, 0.278045654296875, -1.77655029296875, 0.102569580078125, 3.3618011474609375, 3.968597412109375, -1.33758544921875, 0.27063751220703125, 3.286163330078125, 0.127532958984375, -1.6464080810546875, 0.011709213256835938, -0.460479736328125, -0.7660064697265625, -5.158233642578125, -1.5821533203125, 0.0, 1.77813720703125, 3.693206787109375, 2.32354736328125, -1.1330718994140625, 0.8829498291015625, 3.9553985595703125, 2.9356536865234375, 0.419708251953125, 2.55780029296875, 6.57855224609375, 2.8863677978515625, 4.739471435546875, -1.26605224609375, 4.654205322265625, 2.106201171875, -0.4107627868652344, 7.228515625, 1.760467529296875, 4.50347900390625, -7.56103515625, 1.30096435546875, 2.4929733276367188, 5.3324737548828125, -1.0464019775390625, 6.90594482421875, 2.383026123046875, 6.448486328125, 0.6081085205078125, 0.0833740234375, 1.835540771484375, 1.41552734375, -4.79150390625, -0.8418312072753906, 3.4423370361328125, -2.95465087890625, -0.124237060546875, 0.953460693359375, -1.3406982421875, -1.9211273193359375, 2.183502197265625, -0.612884521484375, 4.126708984375, -0.34497833251953125, 1.0399703979492188, -0.36508941650390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000065.npy"} {"epoch": 0.13612565445026178, "step": 66, "batch_size": 128, "mean": 1.502687931060791, "std": 3.6614935398101807, "min": -11.233001708984375, "p10": -2.5177932739257813, "median": 1.2416915893554688, "p90": 6.1328796386718745, "max": 12.30645751953125, "pos_frac": 0.6953125, "sample": [0.408050537109375, 1.0369873046875, 7.000244140625, -2.48370361328125, -1.4797515869140625, 5.00006103515625, 0.091949462890625, 1.15911865234375, 1.2383575439453125, 0.9557037353515625, -0.08615684509277344, 0.013431549072265625, 1.83880615234375, 1.1435394287109375, 1.6761245727539062, 0.724884033203125, -0.39548683166503906, 0.88739013671875, 8.445465087890625, 1.7862167358398438, 1.8017578125, -6.26513671875, 4.966056823730469, 3.16265869140625, 2.878662109375, 1.538360595703125, -0.9979705810546875, 1.781463623046875, -0.7590599060058594, 6.2724609375, 1.733673095703125, 0.933074951171875, 11.726272583007812, -3.99127197265625, 1.1252975463867188, -0.688140869140625, 1.89251708984375, 11.2730712890625, 2.6220855712890625, -1.111328125, -2.6474761962890625, 2.046947479248047, 6.7966461181640625, 1.9176025390625, 0.074798583984375, 2.89349365234375, -0.83612060546875, 8.388153076171875, 8.06854248046875, -2.5973358154296875, -6.03497314453125, 0.9042205810546875, 1.1048736572265625, 12.30645751953125, 8.8106689453125, 0.3498687744140625, 0.7626953125, 3.6897735595703125, -11.233001708984375, 2.602569580078125, -1.0625457763671875, 1.149810791015625, 9.63323974609375, 1.7541046142578125, 5.030670166015625, -6.9215545654296875, 0.91448974609375, -1.1258087158203125, 2.629913330078125, -1.14501953125, -0.384674072265625, 1.54058837890625, 6.084503173828125, -3.065032958984375, 1.896820068359375, 1.4298019409179688, 3.57415771484375, -0.7865982055664062, 4.58087158203125, -6.018585205078125, 3.4007568359375, 1.8558692932128906, -2.07672119140625, 3.1546630859375, -1.2382736206054688, 2.074005126953125, -0.32866668701171875, 0.48614501953125, 4.16644287109375, 2.7067718505859375, 0.8832855224609375, 6.0050048828125, 1.7191162109375, -6.233154296875, 0.3044281005859375, -1.72576904296875, -2.714691162109375, -3.114776611328125, 3.761474609375, 2.2264404296875, 5.1290283203125, 3.204986572265625, 1.021575927734375, 0.0, 2.5825462341308594, -1.7309494018554688, -0.8469390869140625, 5.7008209228515625, 1.8037948608398438, 4.212371826171875, 0.707366943359375, 2.13873291015625, -0.2817802429199219, -3.40985107421875, 1.245025634765625, 2.812286376953125, -1.177947998046875, 6.245758056640625, 2.663665771484375, 3.5797119140625, -1.655181884765625, -0.1357421875, 2.1513519287109375, -2.062694549560547, 5.190338134765625, 0.7735671997070312, 6.5750732421875, 4.6614990234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000066.npy"} {"epoch": 0.1382198952879581, "step": 67, "batch_size": 128, "mean": 1.2717524766921997, "std": 3.2977724075317383, "min": -7.392578125, "p10": -2.72943115234375, "median": 1.1993865966796875, "p90": 5.455589294433593, "max": 10.486328125, "pos_frac": 0.6953125, "sample": [-0.39360809326171875, 4.3165283203125, -0.17156982421875, 7.273712158203125, 0.4367046356201172, -4.054931640625, 0.7060470581054688, 3.99609375, 4.010833740234375, -0.8025588989257812, 1.667572021484375, 9.159423828125, 0.7473907470703125, 2.21502685546875, 2.18035888671875, 5.95147705078125, 3.8038330078125, 4.6619873046875, -1.012664794921875, -0.9056549072265625, 1.1451416015625, -1.0604400634765625, 3.7886962890625, -0.3162078857421875, 1.6115798950195312, 5.39227294921875, -2.17840576171875, -0.70965576171875, 2.1242904663085938, 0.735107421875, -4.51165771484375, 0.14663314819335938, -1.22320556640625, 1.6893386840820312, 1.439178466796875, -3.732421875, 0.0, 0.64678955078125, 3.9600830078125, 2.3628883361816406, -2.86358642578125, 4.79168701171875, 3.1290130615234375, 6.834625244140625, 1.4456558227539062, -1.157191276550293, 2.5318450927734375, -1.52703857421875, -4.53839111328125, -0.3193397521972656, -4.82879638671875, 4.9093780517578125, 0.00424957275390625, 6.153411865234375, -0.57745361328125, 1.8193359375, 6.42120361328125, -4.733673095703125, 1.3016204833984375, 0.9676303863525391, 3.33001708984375, 2.986156463623047, 4.193603515625, -2.67193603515625, 2.6375656127929688, 0.12732696533203125, 0.24557113647460938, -7.12957763671875, -1.33465576171875, 0.2741680145263672, 5.5838165283203125, 1.253631591796875, 0.27797889709472656, 6.28009033203125, -7.25738525390625, 10.486328125, 9.133880615234375, 4.5909423828125, 0.3762359619140625, 1.9377250671386719, 2.4866485595703125, -1.86114501953125, 2.43609619140625, 0.953125, -4.35162353515625, -0.12982940673828125, 5.707324981689453, 0.5996017456054688, 0.16351318359375, 3.908355712890625, 6.83673095703125, -7.392578125, 5.624176025390625, 3.4063949584960938, -4.71051025390625, 3.0078582763671875, 1.44012451171875, 1.4738922119140625, 3.01220703125, 5.400634765625, 2.998077392578125, 1.7375106811523438, 0.168121337890625, 0.6628265380859375, 1.3853759765625, -0.011260986328125, -0.9635124206542969, -6.69427490234375, -1.309661865234375, 1.5312461853027344, 1.4307098388671875, 3.463104248046875, 0.4087066650390625, -1.0978775024414062, -0.9341850280761719, 4.0848388671875, 4.46136474609375, 0.6723289489746094, 3.021759033203125, 1.0091094970703125, 2.32391357421875, 0.8328628540039062, 3.73162841796875, 0.9025039672851562, 0.388946533203125, -1.2637176513671875, 2.91534423828125, -1.2641487121582031], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000067.npy"} {"epoch": 0.14031413612565444, "step": 68, "batch_size": 128, "mean": 1.6918784379959106, "std": 3.940326690673828, "min": -8.28021240234375, "p10": -3.3285202026367187, "median": 1.0791168212890625, "p90": 6.485028076171873, "max": 15.4058837890625, "pos_frac": 0.7265625, "sample": [7.568016052246094, 0.1809062957763672, 0.798675537109375, -2.320159912109375, -0.06568527221679688, 0.1113128662109375, 0.679168701171875, 3.9839324951171875, -0.39101219177246094, 1.0435943603515625, 1.0235366821289062, -0.348876953125, 1.111541748046875, 4.741241455078125, 5.766571044921875, -5.233978271484375, -0.550811767578125, 0.5379791259765625, 3.2513427734375, 2.1138381958007812, 6.96563720703125, 0.9166793823242188, -1.8160057067871094, 1.7739410400390625, 15.4058837890625, 1.0729217529296875, 1.3885555267333984, 3.2335205078125, 0.22086334228515625, -5.652099609375, 4.2718505859375, 3.3262939453125, 10.3375244140625, 7.193511962890625, 0.20471763610839844, 5.825653076171875, 0.0647125244140625, 1.0764923095703125, -4.5786285400390625, -6.1357421875, -1.64776611328125, -1.03668212890625, 4.235477447509766, 3.24322509765625, 0.6861038208007812, -6.101806640625, 0.5289154052734375, 2.8217086791992188, 3.3514404296875, 6.279052734375, 1.617340087890625, -3.2693634033203125, 3.55438232421875, 4.90960693359375, 11.957656860351562, -0.6365966796875, 2.1358642578125, 8.19744873046875, -0.3846702575683594, 0.9969406127929688, 4.561798095703125, 2.0724334716796875, -0.3702430725097656, 2.4799957275390625, 0.3996410369873047, 0.0427398681640625, -8.138275146484375, 5.503753662109375, 1.574951171875, 10.24462890625, 0.11557769775390625, -3.673583984375, 0.9574737548828125, 0.14926910400390625, -3.466552734375, 1.93939208984375, -0.0335693359375, -3.7111053466796875, 9.312225341796875, 0.032421112060546875, -3.59521484375, 2.4153900146484375, -0.4570465087890625, 4.176177978515625, 1.40673828125, 0.080596923828125, -1.99114990234375, 5.779144287109375, -8.28021240234375, 9.385383605957031, 3.83001708984375, 1.0382156372070312, 0.807708740234375, 1.5211181640625, 2.64306640625, 5.459869384765625, -4.327178955078125, -3.08770751953125, 4.037353515625, 1.7586746215820312, 4.31732177734375, 1.0076026916503906, 1.6051559448242188, 7.775390625, 7.22808837890625, 0.833343505859375, 4.8342132568359375, 6.24786376953125, 2.3404693603515625, 4.391326904296875, 3.88568115234375, -3.6117706298828125, 2.6070556640625, 1.656728744506836, -0.4263458251953125, 1.2243537902832031, 3.163665771484375, 0.06262969970703125, -0.2905731201171875, 1.2756729125976562, -1.1433944702148438, -2.0648651123046875, 0.387481689453125, 11.735565185546875, 5.7252197265625, -1.4573593139648438, -0.9614334106445312, 1.0817413330078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000068.npy"} {"epoch": 0.1424083769633508, "step": 69, "batch_size": 128, "mean": 2.1560559272766113, "std": 4.031048774719238, "min": -10.36065673828125, "p10": -1.7285614013671868, "median": 1.6861419677734375, "p90": 7.7519287109375, "max": 15.21514892578125, "pos_frac": 0.75, "sample": [4.7288055419921875, -1.558135986328125, 2.1849822998046875, 4.241363525390625, 0.471282958984375, -10.36065673828125, 3.4733428955078125, 1.8272628784179688, -2.80291748046875, 0.09940910339355469, 5.086814880371094, 1.8515281677246094, 10.806488037109375, 0.7542724609375, 7.752655029296875, -0.05246734619140625, 7.793182373046875, 0.02878570556640625, 2.0751953125, 0.9525909423828125, -0.7103271484375, 0.4012260437011719, -3.3974380493164062, -5.8406829833984375, 7.11505126953125, 11.204071044921875, 4.965728759765625, 4.254486083984375, 0.14604949951171875, 3.282806396484375, -9.64385986328125, 5.128448486328125, -0.2074127197265625, 1.2124786376953125, 2.8168487548828125, -0.2838897705078125, 0.7534370422363281, 1.0999374389648438, 7.669342041015625, 1.627685546875, 9.80157470703125, -5.422637939453125, 0.1107177734375, 2.442413330078125, 0.9423294067382812, 1.4550247192382812, -2.878936767578125, 8.1185302734375, 3.251617431640625, 0.7765045166015625, 7.02349853515625, -1.18902587890625, 7.16119384765625, 3.51611328125, 5.023895263671875, -3.83837890625, -5.232452392578125, 0.347564697265625, 3.21185302734375, -0.5631084442138672, 0.243499755859375, 6.298431396484375, 0.8657341003417969, 5.244026184082031, -0.4390106201171875, 3.5729522705078125, 1.744598388671875, 5.386566162109375, 3.49884033203125, 4.92108154296875, 9.690185546875, 3.5639801025390625, 1.902496337890625, 0.0482177734375, 2.135223388671875, 0.4012298583984375, 7.20245361328125, 1.1623916625976562, 2.6233978271484375, 1.493927001953125, -1.3963260650634766, -1.0618896484375, 1.8250560760498047, 0.8003349304199219, -1.139739990234375, -0.066619873046875, 3.75823974609375, 1.9074249267578125, 3.8215484619140625, 9.636566162109375, 9.17718505859375, 8.441757202148438, 0.9629898071289062, 4.3043060302734375, -3.2193603515625, -2.38037109375, 0.119781494140625, 1.130126953125, 9.35540771484375, -0.36798858642578125, 3.921783447265625, 2.214202880859375, 15.21514892578125, 1.993377685546875, 4.7549896240234375, 4.2567138671875, -2.126220703125, 1.568817138671875, -0.9697265625, 2.2167739868164062, -1.43865966796875, -1.45379638671875, 0.4613075256347656, 5.085357666015625, 5.83453369140625, -6.356353759765625, -0.4345207214355469, 0.006927490234375, 3.707305908203125, 9.857040405273438, 1.2084197998046875, 3.0434417724609375, 0.477142333984375, -0.26056480407714844, -0.30825042724609375, 2.262451171875, 7.751617431640625, 1.3111839294433594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000069.npy"} {"epoch": 0.14450261780104712, "step": 70, "batch_size": 128, "mean": 2.1732125282287598, "std": 3.7106518745422363, "min": -10.91943359375, "p10": -1.349446105957031, "median": 1.5323433876037598, "p90": 6.8650054931640625, "max": 13.6029052734375, "pos_frac": 0.7421875, "sample": [-0.344635009765625, 3.2255401611328125, 1.0482559204101562, -2.1472320556640625, -1.424652099609375, 0.601593017578125, 4.45855712890625, 3.9087142944335938, -4.7659912109375, 7.505523681640625, -0.8862838745117188, -0.010967254638671875, 0.68646240234375, 4.8134765625, 6.849853515625, 4.092811584472656, 2.807830810546875, 4.37249755859375, 0.6483917236328125, 1.23919677734375, 1.2169723510742188, 6.98858642578125, 0.21240234375, 1.5704669952392578, -4.0322265625, 3.1368560791015625, 0.4453582763671875, 1.9222869873046875, 2.6884689331054688, -1.4925003051757812, 6.261932373046875, 1.0767288208007812, 0.5074920654296875, 0.1312255859375, 2.9237060546875, -0.173797607421875, 7.847686767578125, -3.140655517578125, -3.8330078125, 0.04473876953125, 1.570220947265625, -0.10791778564453125, 10.8406982421875, 1.93170166015625, 1.9085426330566406, -1.624237060546875, 0.465545654296875, 1.0121002197265625, -0.5354385375976562, -0.9412841796875, 0.508453369140625, 0.8010711669921875, -0.6119232177734375, 0.4246788024902344, -1.0173835754394531, 5.156829833984375, 10.169219970703125, 4.74066162109375, 4.92462158203125, 1.4944658279418945, 2.4785842895507812, 3.96063232421875, -0.6475296020507812, 0.6568756103515625, 2.63568115234375, 4.95074462890625, -0.0497894287109375, 5.954254150390625, 1.8327159881591797, 0.7974853515625, -3.81732177734375, 1.2873458862304688, -0.07140350341796875, 1.4690093994140625, 1.6184768676757812, 7.64813232421875, 0.2838134765625, 1.9381484985351562, 9.15936279296875, 2.769012451171875, 11.63873291015625, 6.854522705078125, 3.001922607421875, 2.61614990234375, 5.3870086669921875, 3.6044921875, -1.4936981201171875, 2.83221435546875, 0.12921905517578125, 10.620361328125, 5.106681823730469, 6.193145751953125, 0.0, 0.2511940002441406, 4.09619140625, 8.72442626953125, 5.790283203125, 4.6942138671875, 0.13140869140625, 12.5682373046875, 4.048828125, 6.88946533203125, 1.01348876953125, -2.1923179626464844, 2.6659622192382812, -1.132904052734375, 5.819244384765625, 1.4156036376953125, 5.39093017578125, -1.3172149658203125, -0.2136383056640625, -0.7672119140625, 0.8913040161132812, 3.1181869506835938, -6.83099365234375, -10.91943359375, 2.3903656005859375, 0.48697662353515625, 1.82452392578125, 6.199798583984375, 2.13848876953125, 2.7904434204101562, 2.149749755859375, -0.24719810485839844, 13.6029052734375, 0.0, 0.292205810546875, -1.0295639038085938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000070.npy"} {"epoch": 0.14659685863874344, "step": 71, "batch_size": 128, "mean": 2.0154664516448975, "std": 4.504321575164795, "min": -14.34185791015625, "p10": -3.2779174804687496, "median": 2.07952880859375, "p90": 7.829141235351562, "max": 12.8551025390625, "pos_frac": 0.6875, "sample": [-0.225372314453125, 8.67034912109375, -1.3740768432617188, -5.304656982421875, -1.725799560546875, -5.59246826171875, -0.45558929443359375, 10.63427734375, -7.84344482421875, 3.2984619140625, 1.3116874694824219, 3.6924209594726562, -3.6051025390625, 3.656005859375, 2.1082916259765625, 5.479583740234375, 6.3202056884765625, 0.173675537109375, 3.1037750244140625, 9.82623291015625, 2.569183349609375, -0.884185791015625, 3.10076904296875, 10.817535400390625, 2.56488037109375, -1.1091690063476562, 0.8593368530273438, 7.824188232421875, 3.1742782592773438, 3.80950927734375, -3.8499908447265625, 5.112525939941406, 1.946868896484375, 9.348785400390625, -1.319854736328125, 4.129302978515625, 9.3775634765625, 0.5366945266723633, 7.576385498046875, -4.226318359375, -0.2911376953125, 0.5, -1.0201416015625, 2.8221664428710938, 2.0507659912109375, -0.41573333740234375, 1.314208984375, -13.5169677734375, -3.1376953125, -1.4421424865722656, 3.0100250244140625, 4.87127685546875, -1.2943916320800781, 4.743686676025391, 1.5046501159667969, 5.2373046875, 11.64617919921875, 2.392120361328125, -0.3191795349121094, -0.04730224609375, 3.9893646240234375, -1.656768798828125, 6.689727783203125, 6.60308837890625, 0.11531257629394531, -0.48974609375, 0.2235565185546875, -2.49884033203125, 4.34478759765625, -0.008106231689453125, 1.8837738037109375, 12.03411865234375, 3.551513671875, -4.62921142578125, -4.285747528076172, 4.8199462890625, 7.8406982421875, -6.644775390625, -2.8177642822265625, 0.6181259155273438, -0.75128173828125, -2.34051513671875, 1.924072265625, 1.5971908569335938, 8.212738037109375, 6.857513427734375, 1.1553955078125, 5.488525390625, 6.59881591796875, 3.447540283203125, 1.7122859954833984, 4.004119873046875, 0.253570556640625, 0.6645383834838867, 0.74774169921875, 9.859573364257812, -0.7264175415039062, 8.81689453125, 3.342090606689453, 5.85235595703125, 6.88714599609375, -4.241172790527344, -0.26718902587890625, 4.45294189453125, 7.656097412109375, -0.2928619384765625, 0.62896728515625, 0.8464984893798828, -5.0458984375, 2.15008544921875, 2.7655487060546875, 1.0272769927978516, 3.554229736328125, 2.2543487548828125, 2.192596435546875, 2.6305294036865234, 3.492767333984375, 4.67401123046875, 2.298297882080078, 3.0829620361328125, 3.2152099609375, 1.7706375122070312, 2.655609130859375, -0.21410369873046875, -0.31711578369140625, 12.8551025390625, -14.34185791015625, 3.094818115234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000071.npy"} {"epoch": 0.1486910994764398, "step": 72, "batch_size": 128, "mean": 2.0543031692504883, "std": 4.2037672996521, "min": -10.786102294921875, "p10": -3.0144317626953123, "median": 1.9885978698730469, "p90": 7.6597045898437495, "max": 11.654296875, "pos_frac": 0.6796875, "sample": [3.63775634765625, 3.639923095703125, -2.7165374755859375, 1.779052734375, 11.17236328125, -2.00811767578125, 1.8956756591796875, 5.13427734375, 2.98211669921875, -0.47374725341796875, -0.2603607177734375, 1.531982421875, 2.042144775390625, 3.4521331787109375, 4.17041015625, 8.775421142578125, -2.3416595458984375, 2.27374267578125, -3.10101318359375, 0.332427978515625, 2.8486251831054688, 6.569580078125, 1.391357421875, 10.9627685546875, 4.2019805908203125, 0.6142368316650391, -1.8696670532226562, 3.352447509765625, -5.066764831542969, 1.020416259765625, 3.895660400390625, -10.786102294921875, -1.74310302734375, -5.39447021484375, 2.33050537109375, 1.82403564453125, 0.8551177978515625, -0.5581645965576172, 2.49267578125, 1.1909332275390625, 3.975311279296875, -1.71429443359375, 7.0869140625, 6.60345458984375, 7.640533447265625, 2.879985809326172, 3.3722076416015625, 0.41522216796875, -4.398345947265625, -2.780059814453125, -2.977325439453125, 4.652191162109375, -1.506561279296875, 0.0, 7.471923828125, 7.704437255859375, 10.093856811523438, -0.45305633544921875, 4.29046630859375, 9.846435546875, 0.3001251220703125, -4.7474212646484375, 4.062232971191406, 5.630470275878906, 2.7758026123046875, 6.7621307373046875, 1.8033065795898438, 0.6107635498046875, -0.4954795837402344, 2.7486572265625, -4.1087188720703125, 2.493234634399414, -0.3197174072265625, 2.989471435546875, 8.7470703125, 2.1427574157714844, 0.05797386169433594, 4.77838134765625, 0.107574462890625, 9.520286560058594, 0.86285400390625, 1.6345882415771484, -1.9716720581054688, -0.981781005859375, -2.6956024169921875, -3.64044189453125, 1.525054931640625, 0.8258209228515625, -1.839498519897461, 9.14996337890625, -6.541534423828125, 8.057647705078125, -1.08074951171875, 3.316162109375, 2.1881561279296875, 5.892173767089844, 6.421661376953125, -1.1996612548828125, 1.688385009765625, 5.8541259765625, 2.57476806640625, 7.6035614013671875, 0.0, 5.07098388671875, -3.4756736755371094, 0.8395767211914062, 8.186248779296875, 11.169525146484375, 7.0971221923828125, 6.587944030761719, 3.2633438110351562, -1.36517333984375, -1.5698699951171875, -5.542327880859375, 11.654296875, -0.18011474609375, -2.0418739318847656, 0.0, 1.9350509643554688, 4.2360076904296875, 3.0746726989746094, 2.5283584594726562, -6.418487548828125, 3.85345458984375, 2.338653564453125, -3.868499755859375, 2.705322265625, 7.11602783203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000072.npy"} {"epoch": 0.15078534031413612, "step": 73, "batch_size": 128, "mean": 1.9830081462860107, "std": 4.749943733215332, "min": -17.465667724609375, "p10": -3.1256393432617187, "median": 1.6726064682006836, "p90": 7.873501586914062, "max": 17.413055419921875, "pos_frac": 0.71875, "sample": [1.6199607849121094, 2.0543212890625, 9.628662109375, 4.53045654296875, -1.75543212890625, 1.8316268920898438, -0.3424072265625, -0.371612548828125, 1.1639556884765625, 7.14715576171875, 2.307708740234375, 1.411376953125, 1.5819091796875, 9.704345703125, 0.2832908630371094, 0.39665985107421875, -7.33624267578125, -9.632720947265625, 1.2629013061523438, 0.6679840087890625, 2.3428955078125, 1.0498046875, 0.207122802734375, 4.08203125, 0.7682266235351562, -0.029003143310546875, 1.9795417785644531, 1.4059715270996094, 1.5395660400390625, 4.0259552001953125, -6.272918701171875, -2.8646621704101562, 4.571075439453125, 9.466522216796875, -1.53717041015625, 1.686187744140625, -3.0453338623046875, 13.368408203125, 5.504508972167969, 3.024993896484375, 3.944732666015625, 1.4152565002441406, 4.19122314453125, -3.0905532836914062, 17.413055419921875, -0.9398193359375, -0.8087005615234375, -1.44927978515625, 2.0944671630859375, 11.400146484375, -6.724212646484375, 2.5030517578125, 9.722747802734375, -17.465667724609375, 4.4791259765625, 1.1707572937011719, 2.373687744140625, -4.914207458496094, 5.4831695556640625, -2.7562408447265625, 2.785430908203125, 0.63079833984375, 4.208984375, -1.3695068359375, -1.1791839599609375, 7.10552978515625, 4.257637023925781, -5.559722900390625, 6.587799072265625, -0.03583526611328125, -1.0753822326660156, -3.0816650390625, 4.313140869140625, 5.5071868896484375, 0.371185302734375, 1.6590251922607422, -6.2792510986328125, -4.51971435546875, -2.1442108154296875, 1.58319091796875, 5.742523193359375, 5.52825927734375, 4.8866119384765625, 0.1965484619140625, -2.55657958984375, 3.3517303466796875, 0.5292205810546875, 2.1475372314453125, -0.7551498413085938, 5.81207275390625, 1.7106285095214844, 1.993927001953125, 7.109832763671875, 3.5651168823242188, 7.045928955078125, 6.068359375, 2.036243438720703, 4.457305908203125, -6.4226226806640625, 5.4436187744140625, -0.0381011962890625, 5.66796875, 4.696281433105469, 4.5155029296875, 7.8917236328125, 6.1906585693359375, 0.347412109375, -1.37689208984375, 11.2152099609375, 7.909271240234375, 1.84478759765625, -3.0972900390625, 1.5945587158203125, 0.12441825866699219, 9.751258850097656, -4.0845947265625, 0.4557151794433594, 7.865692138671875, -3.1917877197265625, 0.40822601318359375, 9.36395263671875, 2.23162841796875, 7.896820068359375, 3.7296905517578125, 1.5075531005859375, 0.18511199951171875, 6.81304931640625, -3.719696044921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000073.npy"} {"epoch": 0.15287958115183245, "step": 74, "batch_size": 128, "mean": 1.8528268337249756, "std": 4.708539009094238, "min": -11.263427734375, "p10": -4.302780151367187, "median": 1.807830810546875, "p90": 7.782110595703124, "max": 15.77166748046875, "pos_frac": 0.6796875, "sample": [2.688232421875, 9.46856689453125, 3.733551025390625, 1.28363037109375, 9.2734375, -1.8475341796875, -4.105438232421875, 2.026123046875, 6.3948822021484375, 9.78662109375, -2.47747802734375, 9.310775756835938, 0.0, 1.47979736328125, 1.194427490234375, 6.998291015625, 5.74688720703125, 11.193328857421875, -1.7536163330078125, 0.9670867919921875, 2.797515869140625, 3.6990966796875, 1.71881103515625, 2.1658782958984375, 8.9080810546875, -5.069091796875, -6.92877197265625, -0.9058837890625, -1.1881103515625, -3.6063995361328125, -1.4765625, -1.634521484375, -4.76324462890625, -1.1290054321289062, 4.4519500732421875, 1.5529327392578125, -0.10533905029296875, 7.231212615966797, 0.11182022094726562, 6.6839599609375, -0.4784393310546875, -1.48272705078125, -10.461273193359375, -6.600616455078125, 1.0629119873046875, 3.298126220703125, 1.142333984375, -2.17901611328125, -1.836944580078125, 4.128662109375, -9.974029541015625, -7.3515625, 0.93072509765625, 2.349029541015625, 1.086181640625, -1.2941551208496094, 1.3702888488769531, 1.949554443359375, -8.26995849609375, 5.509521484375, 3.8429718017578125, 3.5950164794921875, 0.7646484375, 7.6219482421875, 6.2789306640625, 2.86865234375, 8.15582275390625, 5.103546142578125, 1.9418487548828125, 15.77166748046875, 3.352203369140625, 1.377197265625, -1.6318588256835938, 5.366668701171875, -0.24561309814453125, 2.492706298828125, 2.1951141357421875, 3.482177734375, -0.2236480712890625, 5.0053863525390625, 3.49615478515625, -0.16066741943359375, 1.782470703125, -11.263427734375, 12.307373046875, 2.0463485717773438, 1.44927978515625, 1.2603759765625, 1.5411033630371094, -0.7375411987304688, -0.37137603759765625, -4.0220184326171875, -0.533203125, 5.722450256347656, 4.0492706298828125, 4.49560546875, 3.534423828125, -8.04779052734375, -6.101104736328125, 6.24981689453125, 4.7881011962890625, -4.9667510986328125, 2.6661834716796875, 6.31884765625, 1.83319091796875, 5.92437744140625, 4.383880615234375, -5.10675048828125, 5.74261474609375, -2.41204833984375, -0.6088180541992188, 1.6632461547851562, 4.62274169921875, 3.144683837890625, 11.28125, -0.3076286315917969, 1.4164657592773438, 3.422027587890625, 3.997100830078125, 0.432952880859375, 3.286529541015625, 2.4525222778320312, 1.2573394775390625, 8.935218811035156, 9.770751953125, 4.12554931640625, 0.28460693359375, 8.2261962890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000074.npy"} {"epoch": 0.1549738219895288, "step": 75, "batch_size": 128, "mean": 2.796572685241699, "std": 4.509194850921631, "min": -9.182098388671875, "p10": -2.8426315307617185, "median": 2.299285888671875, "p90": 8.551837158203124, "max": 14.87506103515625, "pos_frac": 0.75, "sample": [2.2123565673828125, 3.196807861328125, 2.0210418701171875, 1.2211456298828125, -1.82989501953125, 8.55487060546875, 5.8533172607421875, -3.16693115234375, 4.76263427734375, 2.5385208129882812, 0.6855716705322266, -3.867034912109375, 2.342254638671875, -0.8599452972412109, 8.747802734375, -0.052669525146484375, 14.87506103515625, 3.74560546875, 7.35992431640625, 4.4664459228515625, 7.2279510498046875, 6.6911468505859375, 7.1392822265625, -3.2556610107421875, 0.7365570068359375, 13.29461669921875, -2.10888671875, 7.0690460205078125, -0.525787353515625, 1.7300262451171875, 3.9080810546875, 6.9040679931640625, 4.592041015625, 2.871826171875, 0.971710205078125, -1.7674331665039062, 3.606964111328125, 3.0443115234375, 3.7603836059570312, 1.1533203125, 10.713043212890625, 5.854190826416016, 0.0233154296875, -9.182098388671875, 0.9235992431640625, 1.3504638671875, 0.795989990234375, -0.187164306640625, 8.89031982421875, 0.985595703125, 4.825347900390625, 3.703704833984375, 2.642669677734375, -0.1011962890625, 4.3636474609375, 5.3474884033203125, -3.181182861328125, 9.248382568359375, 0.8793182373046875, 7.046630859375, 4.0966796875, 1.9829483032226562, 0.9337234497070312, 7.9518585205078125, 1.0578536987304688, 1.8407440185546875, 8.281463623046875, -1.6861495971679688, 1.9211235046386719, 1.487030029296875, 1.819305419921875, -2.391693115234375, 1.8761978149414062, 12.485107421875, 8.550537109375, 0.016675949096679688, -6.301910400390625, -7.384185791015625, -1.17999267578125, 3.4383163452148438, 10.060455322265625, 5.0466461181640625, 5.76678466796875, -4.592445373535156, -0.4268646240234375, 9.267959594726562, 0.8745536804199219, -0.826324462890625, 3.947601318359375, 6.4430084228515625, 5.39813232421875, -1.6828765869140625, 7.0234375, 2.77178955078125, 7.658905029296875, 13.467864990234375, 0.5661773681640625, 5.41387939453125, 0.0506591796875, 6.9488372802734375, -8.5557861328125, -0.026947021484375, 5.392669677734375, 10.41943359375, 6.6094970703125, 2.880218505859375, -0.2001495361328125, 5.5305633544921875, 2.0683670043945312, -5.052276611328125, 6.037628173828125, -0.18929290771484375, -0.351318359375, 5.688682556152344, 4.793218612670898, 7.8682861328125, 1.74615478515625, 1.5710716247558594, 3.3735198974609375, -3.049835205078125, 2.256317138671875, 0.958709716796875, 5.065059661865234, 9.2642822265625, -2.7538299560546875, -3.187469482421875, -7.501838684082031, 0.5400543212890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000075.npy"} {"epoch": 0.15706806282722513, "step": 76, "batch_size": 128, "mean": 2.9753010272979736, "std": 4.966036796569824, "min": -9.52252197265625, "p10": -1.8819801330566401, "median": 2.0750465393066406, "p90": 9.6177734375, "max": 22.938217163085938, "pos_frac": 0.734375, "sample": [6.781890869140625, -2.3330841064453125, 1.6514663696289062, 1.4604339599609375, 2.996337890625, 3.514556884765625, 1.901397705078125, 1.2990741729736328, 2.668691635131836, 3.9807281494140625, 8.762939453125, 0.6124401092529297, -0.3910064697265625, -0.870208740234375, 15.230224609375, 5.32977294921875, -0.6950607299804688, -2.6075897216796875, 2.0365447998046875, -0.41666412353515625, 11.47015380859375, -0.3652191162109375, 6.6201171875, 2.82293701171875, 3.2199554443359375, 12.46160888671875, 2.221649169921875, 3.0550994873046875, 1.7157402038574219, 1.411468505859375, 2.305999755859375, 0.04123115539550781, 5.45184326171875, 8.213302612304688, -9.52252197265625, 15.69317626953125, 3.7017822265625, 4.639923095703125, 1.98809814453125, 5.38031005859375, 9.679595947265625, 5.93212890625, 6.165283203125, 6.134246826171875, -6.50286865234375, 1.4380226135253906, 3.4992599487304688, 4.6008148193359375, 0.7261505126953125, 9.591278076171875, 0.0, 4.2514495849609375, -0.1904144287109375, 0.5523300170898438, 7.710823059082031, -2.2965087890625, 10.47052001953125, -6.762725830078125, -0.862762451171875, -5.30462646484375, 5.9081573486328125, 11.75140380859375, -2.20867919921875, 2.02410888671875, 3.181060791015625, 0.9612846374511719, 1.9813690185546875, 4.21380615234375, 5.6275482177734375, -1.7388458251953125, -1.2882843017578125, 9.2374267578125, -2.51177978515625, 0.25707244873046875, -3.49273681640625, 0.0, 10.679702758789062, 4.232170104980469, 3.86151123046875, 13.899139404296875, 2.463775634765625, 0.09663200378417969, 2.144439697265625, 2.0997543334960938, -0.26099395751953125, 2.446533203125, 1.9037551879882812, 7.74676513671875, 2.92919921875, 4.0652923583984375, 4.2432708740234375, 2.150970458984375, -0.6508560180664062, 1.7749786376953125, -0.1360454559326172, -0.88922119140625, -3.9868850708007812, -0.456329345703125, 1.8182220458984375, -0.5667037963867188, 3.114501953125, 1.4177017211914062, 7.4981689453125, 13.5555419921875, 8.868499755859375, 16.310577392578125, 10.439956665039062, 1.178131103515625, 1.4302978515625, -3.79119873046875, 0.8062896728515625, -1.7419662475585938, 22.938217163085938, -1.440673828125, -7.7529296875, 0.0189971923828125, 2.779541015625, 2.0503387451171875, 7.03961181640625, 1.5368881225585938, 2.3309097290039062, 7.65972900390625, 2.2386398315429688, 2.0167465209960938, -0.6981201171875, 3.132843017578125, -0.8051300048828125, 0.9228973388671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000076.npy"} {"epoch": 0.15916230366492146, "step": 77, "batch_size": 128, "mean": 2.4289917945861816, "std": 5.285545349121094, "min": -13.15606689453125, "p10": -3.2662719726562495, "median": 1.8930244445800781, "p90": 9.1076904296875, "max": 18.9609375, "pos_frac": 0.6796875, "sample": [-0.9161262512207031, 3.320220947265625, 1.5787353515625, 3.1217041015625, 2.0606689453125, 0.7393798828125, 9.1353759765625, 5.0410308837890625, 0.22129249572753906, 4.370124816894531, -4.231292724609375, -0.023136138916015625, 9.081100463867188, 6.382537841796875, 11.708694458007812, 1.5527915954589844, 0.4322509765625, 2.9551162719726562, -13.15606689453125, 6.2578125, -1.2708053588867188, -0.43701171875, -2.33660888671875, -2.966644287109375, 6.632080078125, 8.109848022460938, 5.6619110107421875, 1.8878402709960938, 3.9874114990234375, 4.9833984375, -2.23089599609375, -1.642852783203125, 1.086944580078125, 8.373504638671875, -0.99005126953125, 1.3103485107421875, 1.874481201171875, -0.0198974609375, 7.1123046875, -1.2783737182617188, 0.2330780029296875, 1.0065536499023438, 7.25970458984375, 2.509246826171875, 4.013786315917969, 3.4144287109375, 0.9198684692382812, 18.87738037109375, -1.9691314697265625, 0.342864990234375, 5.7340087890625, -1.3841552734375, 3.3094482421875, 10.640716552734375, 8.475662231445312, 5.60235595703125, 4.639892578125, -2.687744140625, 2.656402587890625, 0.2367706298828125, 0.5178985595703125, -3.61627197265625, 8.86358642578125, 18.9609375, -3.11627197265625, 5.06170654296875, 3.7619094848632812, -1.23883056640625, 11.3502197265625, 2.08148193359375, 9.76995849609375, 2.3331680297851562, -8.704681396484375, -4.490081787109375, -2.7657546997070312, -1.599822998046875, 5.215972900390625, 1.6528396606445312, 1.9612579345703125, 2.24090576171875, -1.765350341796875, 10.146484375, -2.7246170043945312, -1.99981689453125, 1.0923309326171875, 5.181793212890625, 9.0958251953125, -8.682403564453125, -4.328460693359375, 3.9712982177734375, 2.596435546875, 1.9556732177734375, -1.9911346435546875, 3.37359619140625, 5.4140777587890625, -5.11285400390625, -1.42816162109375, -7.6226806640625, 9.192779541015625, 4.97369384765625, -3.62335205078125, 5.3896636962890625, -0.7785663604736328, 1.583404541015625, 6.4587554931640625, 1.03912353515625, 0.59429931640625, 11.926025390625, 12.32733154296875, -0.01873779296875, 1.7700309753417969, -4.989898681640625, -5.0686492919921875, 16.787384033203125, -0.4497947692871094, 4.658050537109375, -4.5246734619140625, 2.508331298828125, -2.086088180541992, 0.496337890625, 7.567047119140625, 1.8982086181640625, 4.3693084716796875, 5.5433349609375, 5.397064208984375, 13.98199462890625, -2.9384765625, 0.20456695556640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000077.npy"} {"epoch": 0.1612565445026178, "step": 78, "batch_size": 128, "mean": 2.956638813018799, "std": 5.386080741882324, "min": -12.682968139648438, "p10": -3.283417510986328, "median": 2.356466293334961, "p90": 9.842164611816406, "max": 23.59429931640625, "pos_frac": 0.734375, "sample": [-1.656280517578125, 4.88714599609375, -4.15576171875, 5.4678497314453125, 8.922943115234375, 1.8904266357421875, 3.1922760009765625, 3.142852783203125, -3.9164505004882812, 5.4087982177734375, 0.6675186157226562, 1.6980056762695312, 4.496543884277344, 3.23443603515625, -1.2329177856445312, 2.853363037109375, 3.884368896484375, 5.8727264404296875, 10.514266967773438, -0.0270538330078125, 0.0, 11.646575927734375, -4.6598968505859375, 2.988983154296875, 0.7867279052734375, 1.1382255554199219, 7.0360107421875, 2.8725032806396484, -3.2126083374023438, 6.022789001464844, -2.450225830078125, 4.793243408203125, 9.838058471679688, 1.23919677734375, 7.742218017578125, -2.26666259765625, 1.191070556640625, 15.40557861328125, 0.8336181640625, -4.8981475830078125, 4.647590637207031, 6.03662109375, -0.649383544921875, 23.59429931640625, 12.6837158203125, -3.448638916015625, 4.3136138916015625, 2.7188720703125, 0.660308837890625, 4.2865447998046875, 5.94830322265625, -12.682968139648438, 1.0725555419921875, 2.37127685546875, -0.12786865234375, -0.3719043731689453, 6.0800018310546875, 4.7609710693359375, 3.5740814208984375, -0.80596923828125, 9.85174560546875, -0.555816650390625, 5.070587158203125, 7.0747833251953125, 2.347637176513672, -3.4646987915039062, 1.431427001953125, 3.217132568359375, 4.68048095703125, -10.45770263671875, 12.860260009765625, 2.1269912719726562, 1.60699462890625, 2.0983734130859375, 10.303955078125, 7.10919189453125, -0.082489013671875, 2.4285507202148438, -1.1107215881347656, 1.57415771484375, -4.2559661865234375, -0.9285717010498047, 9.465362548828125, 15.0301513671875, 1.210479736328125, 2.717376708984375, -0.7677040100097656, 3.095947265625, -6.634307861328125, 6.825042724609375, 3.2623367309570312, -2.38653564453125, 8.35137939453125, -0.638214111328125, -8.235809326171875, 2.2439651489257812, 2.1573486328125, -4.5572509765625, 12.16339111328125, 0.0044403076171875, 0.8045654296875, 0.9036026000976562, -2.3845443725585938, 8.9986572265625, 6.331268310546875, 2.4746246337890625, 5.9428863525390625, 0.710174560546875, 1.9200515747070312, 2.5092697143554688, 13.178558349609375, 0.095245361328125, 2.36529541015625, 5.8880462646484375, 2.784912109375, 2.0021286010742188, 6.836090087890625, 13.43682861328125, 9.836822509765625, -4.5024566650390625, -2.178466796875, 0.5049095153808594, 16.922760009765625, 0.8215293884277344, 1.653289794921875, 2.273773193359375, -1.6546478271484375, 7.886566162109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000078.npy"} {"epoch": 0.16335078534031414, "step": 79, "batch_size": 128, "mean": 3.0043623447418213, "std": 6.942393779754639, "min": -42.5029296875, "p10": -3.2623748779296875, "median": 2.2585220336914062, "p90": 10.420620727539061, "max": 23.1326904296875, "pos_frac": 0.6484375, "sample": [-0.26923370361328125, 6.505119323730469, 10.1038818359375, -3.486663818359375, 5.891510009765625, -0.768463134765625, 1.978057861328125, -0.6221084594726562, 3.6798553466796875, -3.48077392578125, 12.06890869140625, -1.302520751953125, 0.0, 7.1500244140625, -0.13074874877929688, -0.9530105590820312, 0.344940185546875, -2.0900802612304688, -6.509712219238281, -1.005584716796875, 2.44073486328125, -2.021270751953125, -3.23138427734375, -1.9311027526855469, -3.19647216796875, 4.384986877441406, 3.4253082275390625, 0.0, 10.222442626953125, -0.06427192687988281, -6.43634033203125, 10.65478515625, 5.35015869140625, 5.887290954589844, 4.390907287597656, 1.6241378784179688, 3.910186767578125, 1.4206085205078125, 3.7715301513671875, -0.6690673828125, 6.731842041015625, -8.79693603515625, 10.5687255859375, 0.4698600769042969, 0.406768798828125, 5.5098114013671875, 6.462335586547852, 13.805267333984375, 2.0763092041015625, 3.052520751953125, -1.176239013671875, -3.5200347900390625, -42.5029296875, -5.607696533203125, 3.498046875, 8.32891845703125, -1.0997314453125, 15.557159423828125, 8.868026733398438, 9.84716796875, -0.5468368530273438, 1.92779541015625, 3.0398406982421875, 7.1906890869140625, 11.548721313476562, 3.5045166015625, 3.2372589111328125, 2.531423568725586, -2.0224761962890625, 2.68316650390625, -0.661163330078125, 5.199005126953125, 0.4844512939453125, -0.70123291015625, -2.7432861328125, -3.334686279296875, 8.7979736328125, -1.6480712890625, 9.240570068359375, 1.6094818115234375, -3.9950714111328125, 16.5333251953125, 8.20062255859375, 7.00860595703125, 3.582489013671875, 7.700439453125, 4.247627258300781, 0.9586753845214844, 0.3707122802734375, 1.0513114929199219, 0.4905681610107422, 6.176334381103516, -0.29370880126953125, -0.3740081787109375, 1.9582710266113281, 23.1326904296875, -0.1880207061767578, 4.716400146484375, -4.4493255615234375, 9.5155029296875, 14.111053466796875, 10.176025390625, 4.41082763671875, 5.5527496337890625, -4.962371826171875, 8.861244201660156, 8.779296875, -0.3041343688964844, 9.01605224609375, -0.38986968994140625, -3.977386474609375, 16.78863525390625, 0.0438385009765625, -0.6727638244628906, 4.998987197875977, 6.8081512451171875, 6.5172271728515625, 4.65325927734375, 0.5782394409179688, 5.230194091796875, -0.47183799743652344, 10.357147216796875, 10.8438720703125, 19.24481201171875, 0.2662200927734375, -3.0038909912109375, 1.5868606567382812, 14.319610595703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000079.npy"} {"epoch": 0.16544502617801046, "step": 80, "batch_size": 128, "mean": 2.6710267066955566, "std": 6.1410441398620605, "min": -20.939361572265625, "p10": -3.186279296875, "median": 2.1039352416992188, "p90": 10.63662109375, "max": 23.49761962890625, "pos_frac": 0.6875, "sample": [1.788787841796875, 4.430633544921875, -1.6825408935546875, 1.7593193054199219, 2.372833251953125, -2.2018890380859375, 1.09417724609375, -5.952484130859375, 6.1998291015625, -3.0602264404296875, 0.00824737548828125, -1.5828857421875, 5.920013427734375, 4.73101806640625, 5.2181243896484375, -2.6483688354492188, -0.621490478515625, 9.76812744140625, 3.5318527221679688, 4.281646728515625, 17.004425048828125, -0.54022216796875, 2.0261993408203125, 4.508270263671875, 0.865692138671875, 10.223663330078125, 7.48974609375, -3.1610107421875, 3.414764404296875, 10.148468017578125, 2.850189208984375, 0.7368984222412109, 3.74078369140625, -2.3129329681396484, -12.69000244140625, 1.1314620971679688, -20.939361572265625, 6.2991943359375, 0.0, 12.9449462890625, -1.004058837890625, 8.5172119140625, -3.3120880126953125, -2.638092041015625, 3.45916748046875, -9.6112060546875, -1.940673828125, 5.9837799072265625, -2.212005615234375, -2.5175247192382812, 1.4260940551757812, 2.120758056640625, 2.0882568359375, 6.0091400146484375, 9.32110595703125, 0.17133712768554688, 0.47467613220214844, 0.4897308349609375, -1.9150543212890625, 13.139358520507812, 3.7383270263671875, 2.1196136474609375, 5.155567169189453, 5.2557373046875, -3.9053955078125, 1.3369293212890625, 5.1390838623046875, 0.35665130615234375, 7.101966857910156, 3.0507965087890625, -1.97705078125, 4.5111541748046875, 3.11557674407959, -4.579856872558594, 18.267913818359375, -1.17877197265625, -0.1520233154296875, 0.091461181640625, 6.551300048828125, -7.122283935546875, 11.361968994140625, 3.61834716796875, -1.077362060546875, 10.92791748046875, 5.5191650390625, 3.2262229919433594, -6.60296630859375, 1.84014892578125, -1.7199630737304688, 15.993804931640625, 6.1174774169921875, 2.96441650390625, 12.422012329101562, 5.740966796875, 4.272577285766602, 12.889739990234375, 8.231597900390625, 1.1366729736328125, -9.877883911132812, 0.7410507202148438, -7.03753662109375, 11.11175537109375, -2.8178863525390625, 4.84222412109375, 3.62872314453125, 10.51177978515625, 1.1633758544921875, -1.1674041748046875, 3.85955810546875, 8.5325927734375, -0.511138916015625, -0.000762939453125, 0.15325927734375, -0.527008056640625, 1.343841552734375, -5.148231506347656, 3.694263458251953, -1.2024688720703125, 0.9730930328369141, -3.2452392578125, 6.809600830078125, 11.0867919921875, 9.53155517578125, 4.1610565185546875, 23.49761962890625, 0.26393699645996094, 3.3454818725585938, 15.2901611328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000080.npy"} {"epoch": 0.16753926701570682, "step": 81, "batch_size": 128, "mean": 2.8696107864379883, "std": 6.6621785163879395, "min": -18.657470703125, "p10": -3.3299354553222655, "median": 2.2968673706054688, "p90": 12.184716796875, "max": 16.014678955078125, "pos_frac": 0.671875, "sample": [-0.844268798828125, -0.12308311462402344, 8.881706237792969, -11.047348022460938, 4.693084716796875, -0.6650390625, -8.832122802734375, 3.0442123413085938, 16.014678955078125, 7.031524658203125, 15.303359985351562, 11.56439208984375, -2.8032150268554688, -2.661773681640625, 8.2010498046875, 0.4864673614501953, 5.155670166015625, 8.510452270507812, 5.488128662109375, 4.915283203125, 0.9587211608886719, 12.20770263671875, 0.4928741455078125, -1.9393768310546875, 4.253929138183594, 3.484375, 2.28240966796875, -0.8007049560546875, 10.38714599609375, 8.18017578125, 14.9815673828125, 3.2846832275390625, 8.334190368652344, 3.467193603515625, -2.69915771484375, 3.7286605834960938, -5.824951171875, 14.7137451171875, 13.75830078125, 2.674896240234375, -4.9040374755859375, 1.302947998046875, 11.27398681640625, -0.65478515625, -1.146209716796875, 1.599761962890625, -14.896942138671875, -2.32586669921875, 7.74957275390625, 11.2384033203125, 0.69464111328125, 3.0817108154296875, -6.291717529296875, 9.081146240234375, -4.5464630126953125, -0.05535888671875, 5.82940673828125, 4.2542266845703125, 1.8021240234375, -16.07275390625, -4.91937255859375, 1.9451370239257812, 5.04620361328125, 1.829071044921875, -0.04486656188964844, 3.1149063110351562, 1.7708740234375, 0.27698516845703125, 4.2353973388671875, 6.2711334228515625, -0.621917724609375, -2.489593505859375, -0.4404296875, 12.68890380859375, 14.59765625, 9.106292724609375, -2.1201324462890625, 5.5728607177734375, 2.116558074951172, 4.924896240234375, -13.7041015625, 5.7262725830078125, 8.700408935546875, -3.2669677734375, 1.49700927734375, 2.0469131469726562, -1.4957923889160156, -2.506591796875, -0.813232421875, 2.3798294067382812, -18.657470703125, -2.4566192626953125, -14.182220458984375, 0.475189208984375, 2.9759445190429688, 1.6038055419921875, -0.546905517578125, -0.29882049560546875, 13.56396484375, 13.1470947265625, 12.936248779296875, 12.17486572265625, 5.34320068359375, -2.69384765625, 9.12237548828125, 2.3617095947265625, -1.6656494140625, 1.1525802612304688, 0.02719879150390625, 6.73138427734375, -3.4768600463867188, 5.724639892578125, 0.10994529724121094, -1.188201904296875, 2.3113250732421875, 10.562530517578125, 8.841598510742188, -1.4710578918457031, 9.8763427734375, 6.965789794921875, 6.26898193359375, 12.999267578125, 13.783233642578125, 9.96405029296875, 2.060302734375, 0.5327911376953125, 6.202827453613281, -0.548980712890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000081.npy"} {"epoch": 0.16963350785340314, "step": 82, "batch_size": 128, "mean": 3.070629119873047, "std": 5.92679500579834, "min": -14.1072998046875, "p10": -3.85950927734375, "median": 3.1250457763671875, "p90": 10.824951171875, "max": 23.78326416015625, "pos_frac": 0.734375, "sample": [2.6940155029296875, 5.090911865234375, 0.4518585205078125, -6.8071441650390625, -0.96453857421875, 3.32769775390625, 0.50775146484375, 11.75274658203125, 7.6178741455078125, -0.3762626647949219, 1.94000244140625, 6.534515380859375, -4.989631652832031, -2.3122406005859375, -2.644287109375, 7.5612640380859375, 4.34576416015625, 3.1485137939453125, 12.836044311523438, 5.800994873046875, 1.3701705932617188, -3.4644775390625, 7.098731994628906, -1.1226654052734375, 0.07910919189453125, 10.996490478515625, -8.796661376953125, 3.856658935546875, 3.1729736328125, 4.96331787109375, 18.00909423828125, 2.6237335205078125, 10.174240112304688, 5.091766357421875, 9.677505493164062, 12.51226806640625, -0.840423583984375, 5.31512451171875, 5.18621826171875, -14.1072998046875, -4.32025146484375, -0.4673614501953125, 3.335968017578125, 3.4892654418945312, -2.6619110107421875, 9.92547607421875, -2.101348876953125, 8.0972900390625, 17.29974365234375, -0.92626953125, -1.925628662109375, 18.238662719726562, 4.7089996337890625, 5.014739990234375, 2.69415283203125, 0.5794277191162109, 12.354522705078125, -4.21368408203125, 0.43511962890625, 2.6334228515625, 3.8691253662109375, -5.06298828125, -12.643798828125, 3.418149948120117, -3.7947998046875, 0.9098052978515625, -4.2662353515625, 1.648345947265625, 10.751434326171875, 3.891937255859375, -0.6046981811523438, 7.3492279052734375, 0.942596435546875, 4.60107421875, 0.249542236328125, 2.3077545166015625, -1.509246826171875, 1.3257865905761719, 4.548622131347656, 1.5271377563476562, 1.9477996826171875, 23.78326416015625, -3.4389801025390625, 6.1649169921875, -0.574127197265625, 3.1567001342773438, 5.494171142578125, 10.285903930664062, 2.6926116943359375, 2.5396270751953125, 3.7076263427734375, -2.0206451416015625, 2.722545623779297, 5.39599609375, 6.711122512817383, 6.15081787109375, -4.21978759765625, 2.04443359375, 4.5969390869140625, 12.601806640625, 5.326080322265625, 0.16834449768066406, 0.7340469360351562, 5.959991455078125, 6.55682373046875, 3.9722118377685547, 0.2920417785644531, -1.875030517578125, -4.36639404296875, 1.829010009765625, -3.0355262756347656, 3.883453369140625, 6.566474914550781, 7.482761383056641, -4.010498046875, 12.9820556640625, 13.580657958984375, 3.1015777587890625, 4.413047790527344, 4.8907012939453125, 4.4691314697265625, 5.318267822265625, 12.000732421875, 6.84173583984375, 0.9160900115966797, 1.1308860778808594, -11.22198486328125, -3.5697174072265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000082.npy"} {"epoch": 0.17172774869109947, "step": 83, "batch_size": 128, "mean": 3.5388307571411133, "std": 6.943986415863037, "min": -16.035491943359375, "p10": -2.941253662109375, "median": 2.736520767211914, "p90": 12.155758666992186, "max": 28.639312744140625, "pos_frac": 0.703125, "sample": [6.132659912109375, 0.0, 10.23065185546875, 4.67919921875, 0.514251708984375, 0.03540802001953125, -0.0420989990234375, 9.38348388671875, 2.359619140625, 0.18311309814453125, -1.43194580078125, -16.035491943359375, -0.345428466796875, -11.3846435546875, 0.2525482177734375, 1.13726806640625, 4.9887542724609375, 10.32708740234375, 8.277168273925781, 2.776142120361328, 9.35546875, 6.476802825927734, -2.115814208984375, 4.764190673828125, 9.7164306640625, -4.0142822265625, -2.2019500732421875, -0.5998458862304688, -2.69427490234375, 12.23272705078125, 7.229347229003906, 0.86590576171875, 4.52020263671875, 0.9387741088867188, -6.81341552734375, 15.513427734375, 4.1050262451171875, 2.6968994140625, 19.625320434570312, 2.0231475830078125, 4.466487884521484, -1.3896942138671875, 2.6644363403320312, 3.1683502197265625, -3.0039520263671875, 11.81396484375, 5.208770751953125, 5.264312744140625, -2.862823486328125, -2.8539199829101562, 6.9678192138671875, 5.95184326171875, 19.10333251953125, 10.3221435546875, -11.41864013671875, 1.0352630615234375, 10.6002197265625, 4.13482666015625, 28.639312744140625, 2.328033447265625, -2.810821533203125, 11.713577270507812, 4.830413818359375, 7.526458740234375, 2.0899581909179688, -1.1336669921875, 0.80316162109375, 0.36163330078125, -2.9143829345703125, 8.687332153320312, -7.9205780029296875, 6.2864837646484375, 0.150146484375, -2.299591064453125, -2.39215087890625, 3.51800537109375, 16.05902099609375, -3.80560302734375, 3.54644775390625, 8.788558959960938, -2.913818359375, 13.975738525390625, 4.912933349609375, 4.1749267578125, -14.66650390625, -4.4102783203125, 6.411155700683594, 14.0364990234375, -9.11199951171875, 12.122772216796875, -0.13820838928222656, -2.4953231811523438, 0.343719482421875, -4.853302001953125, 5.027629852294922, 15.83612060546875, 14.67626953125, 2.542327880859375, 2.3415985107421875, 1.072021484375, 14.430023193359375, -1.7769775390625, 5.05242919921875, 2.237518310546875, -1.1805877685546875, 0.3597831726074219, 8.365585327148438, 4.987152099609375, 2.376129150390625, -0.74896240234375, 8.929054260253906, -3.927154541015625, 9.801239013671875, 8.815933227539062, 2.78460693359375, 4.267578125, 2.4610595703125, 16.64569091796875, 20.691650390625, -2.8921966552734375, 1.6068172454833984, -0.55035400390625, 9.412857055664062, -0.15424346923828125, 3.9504852294921875, 5.899925231933594, 3.9481201171875, 3.406585693359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000083.npy"} {"epoch": 0.17382198952879582, "step": 84, "batch_size": 128, "mean": 2.8044614791870117, "std": 7.091738700866699, "min": -21.56805419921875, "p10": -3.937847900390625, "median": 3.17962646484375, "p90": 11.147512817382813, "max": 18.5784912109375, "pos_frac": 0.671875, "sample": [-1.3811798095703125, 6.986541748046875, 16.64703369140625, -0.185699462890625, -3.9544677734375, -4.2226715087890625, 4.214111328125, 4.4676513671875, 4.673271179199219, -3.554840087890625, 1.08978271484375, 9.45343017578125, 1.0977935791015625, -2.7447509765625, 3.5372314453125, 4.6701507568359375, 6.528236389160156, 1.1915740966796875, -0.3466606140136719, 9.352203369140625, -3.2774887084960938, 6.537689208984375, 1.59326171875, 5.2799072265625, 7.23974609375, 0.3609619140625, -8.200382232666016, 7.463226318359375, 7.89208984375, 4.1971282958984375, -2.68048095703125, -20.60467529296875, -21.56805419921875, 13.57342529296875, 3.108856201171875, -1.2441558837890625, 8.002777099609375, 2.6503829956054688, 0.9349212646484375, -0.14484405517578125, 5.28118896484375, 3.47064208984375, 12.368988037109375, -3.2205047607421875, 10.414794921875, -1.7614898681640625, 1.6342544555664062, -3.3699264526367188, 12.726165771484375, 4.8285064697265625, 18.284942626953125, -17.4862060546875, 9.571807861328125, 3.1064453125, 4.040740966796875, -3.8596038818359375, 3.2963638305664062, -5.63836669921875, 7.1699981689453125, 8.467884063720703, 5.85791015625, 4.1254425048828125, -0.090850830078125, 4.8236236572265625, 11.093414306640625, 1.7184906005859375, 3.70135498046875, -0.10381317138671875, -0.8729248046875, 3.199981689453125, -3.93072509765625, 5.122711181640625, 2.788909912109375, -0.53033447265625, 2.4579925537109375, 11.27374267578125, 7.707328796386719, 8.891731262207031, -11.974700927734375, 3.681182861328125, 3.159271240234375, 1.710845947265625, -3.5944900512695312, 2.371929168701172, -4.7557373046875, -0.28101158142089844, -0.553009033203125, 9.62677001953125, 18.019287109375, -0.362457275390625, 12.692535400390625, 4.081382751464844, -6.471405029296875, 6.082672119140625, -0.4835662841796875, -0.915191650390625, 3.5665130615234375, 3.148256301879883, 3.3422088623046875, 3.6930084228515625, 5.415557861328125, -0.660797119140625, 2.32061767578125, 2.7826309204101562, 10.005233764648438, 5.809226989746094, -3.540069580078125, 18.5784912109375, -12.1170654296875, 10.371437072753906, 16.42608642578125, 1.5318145751953125, 6.569225311279297, 4.807373046875, 17.46417236328125, 9.52569580078125, -7.1661376953125, -15.852890014648438, 4.523712158203125, -0.88214111328125, 0.32387733459472656, 12.902801513671875, 14.977249145507812, -2.9845504760742188, 3.614532470703125, 5.7705841064453125, 1.2123565673828125, -3.7358856201171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000084.npy"} {"epoch": 0.17591623036649215, "step": 85, "batch_size": 128, "mean": 2.0265555381774902, "std": 5.863761901855469, "min": -16.3883056640625, "p10": -5.919511413574218, "median": 2.313833236694336, "p90": 9.11131591796875, "max": 17.077972412109375, "pos_frac": 0.65625, "sample": [3.8651123046875, -3.686737060546875, -1.8643684387207031, 17.077972412109375, 6.830463409423828, 5.6349945068359375, -1.9351959228515625, -0.816986083984375, 5.720977783203125, 1.9422607421875, 2.4386138916015625, -3.0347061157226562, 1.8844490051269531, 9.0390625, 6.94720458984375, 3.033203125, 8.459991455078125, 5.8879852294921875, 5.30517578125, 2.3998336791992188, -0.75811767578125, -0.4861297607421875, 9.8035888671875, 13.298248291015625, 1.7384300231933594, -2.484039306640625, -0.230133056640625, -5.336112976074219, -6.505035400390625, 5.027313232421875, 6.64434814453125, 5.1491241455078125, -0.8085784912109375, 4.90924072265625, 9.2799072265625, 10.843917846679688, -3.13104248046875, 4.90631103515625, 0.46350860595703125, -1.750457763671875, 9.97821044921875, 5.977142333984375, -0.9593505859375, 5.1398162841796875, 11.278839111328125, -16.3883056640625, 2.3187103271484375, 3.2357940673828125, 3.5241756439208984, 14.326141357421875, -0.768768310546875, -6.428955078125, 5.07989501953125, 0.30694580078125, 2.580656051635742, -3.8685531616210938, 1.931732177734375, 1.5060958862304688, 5.096611022949219, 1.9884567260742188, -4.740203857421875, 5.953125, 3.7000350952148438, -2.4535064697265625, 8.668891906738281, 7.206756591796875, 4.72021484375, 4.77191162109375, -5.452766418457031, -9.564895629882812, -3.674957275390625, 6.559242248535156, -8.021575927734375, -3.535400390625, -0.16634368896484375, -8.33441162109375, 11.73309326171875, 7.6910400390625, 0.5424156188964844, 5.392578125, 3.118865966796875, 10.766510009765625, 0.835784912109375, 5.796600341796875, 14.3011474609375, -12.89227294921875, -0.0697021484375, 8.993988037109375, 4.247091293334961, -5.7967071533203125, -10.133064270019531, 1.5001792907714844, 2.79095458984375, 2.7447032928466797, 10.52685546875, 2.3089561462402344, -1.53338623046875, 4.0561065673828125, -7.3847503662109375, 4.945075988769531, -0.4676246643066406, 4.42547607421875, 1.0625362396240234, -0.763275146484375, 0.01824951171875, 0.988677978515625, -6.9303741455078125, 0.276092529296875, 7.409088134765625, 3.8102798461914062, -6.2060546875, -6.4183349609375, 0.2209320068359375, 6.747955322265625, -3.1312713623046875, 1.5546875, 6.5947265625, 1.0733489990234375, 16.410064697265625, -7.740165710449219, -2.889190673828125, 3.885101318359375, -4.147216796875, 1.94427490234375, 3.4251708984375, -2.452800750732422, 3.6529998779296875, 5.368679046630859], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000085.npy"} {"epoch": 0.17801047120418848, "step": 86, "batch_size": 128, "mean": 3.7104103565216064, "std": 7.79500150680542, "min": -16.67987060546875, "p10": -4.16065673828125, "median": 3.2807998657226562, "p90": 12.946478271484374, "max": 25.188629150390625, "pos_frac": 0.6953125, "sample": [4.0485992431640625, -10.50982666015625, 2.209136962890625, -8.728668212890625, 4.308433532714844, 2.945526123046875, 4.11322021484375, 3.117095947265625, 6.467529296875, 6.33416748046875, 25.188629150390625, -2.73431396484375, 1.3898086547851562, 1.8735504150390625, 6.608634948730469, -10.814613342285156, -12.17767333984375, -3.2943878173828125, 0.8161373138427734, -0.8931503295898438, 4.221923828125, 6.139524459838867, -2.291414260864258, -1.3687286376953125, 15.4281005859375, -0.9228363037109375, -13.57672119140625, 0.25384521484375, 20.1051025390625, 12.559112548828125, -8.379638671875, 11.694580078125, -0.10723876953125, 13.486083984375, 20.7801513671875, 10.178680419921875, 12.687896728515625, -4.07122802734375, 3.51922607421875, 13.4189453125, 3.9048233032226562, -3.6793060302734375, 9.861740112304688, -2.1014862060546875, 3.1018753051757812, 1.26141357421875, 8.300834655761719, 12.857177734375, 10.84661865234375, 5.156707763671875, -16.41455078125, 11.25836181640625, 1.072601318359375, 7.864532470703125, 4.486083984375, -1.9912567138671875, 12.80010986328125, -16.67987060546875, 1.504150390625, 10.257415771484375, 21.153594970703125, 3.300079345703125, 2.66668701171875, 13.41265869140625, 8.5162353515625, -3.935546875, 0.0, -2.18182373046875, 3.2615203857421875, 7.516357421875, 5.085479736328125, 13.0946044921875, -11.53070068359375, 5.0842132568359375, 2.3924331665039062, 1.12176513671875, -0.6900291442871094, 0.615631103515625, 11.993637084960938, 10.7962646484375, -2.7000274658203125, 6.794342041015625, 11.51995849609375, -0.36824798583984375, 6.384757995605469, -8.496978759765625, -11.469390869140625, 7.0443572998046875, -3.087646484375, 1.2803497314453125, -0.906494140625, 5.657501220703125, 2.47113037109375, 2.1404876708984375, 1.93890380859375, -5.08551025390625, 4.130950927734375, 6.1453857421875, 3.24127197265625, 3.5668106079101562, 0.402496337890625, -4.36932373046875, -3.941864013671875, 12.49713134765625, 15.62152099609375, 8.963729858398438, -1.2746944427490234, 24.280914306640625, 2.1446456909179688, 3.65240478515625, -2.417552947998047, 5.32073974609375, 13.979080200195312, 4.096832275390625, 12.2906494140625, -1.3846359252929688, 7.5721588134765625, 15.421356201171875, -0.06204414367675781, 5.369720458984375, 6.62939453125, 10.104835510253906, 8.363204956054688, 12.88299560546875, -2.248046875, -3.446929931640625, 1.7493896484375, 3.1682586669921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000086.npy"} {"epoch": 0.18010471204188483, "step": 87, "batch_size": 128, "mean": 4.11115837097168, "std": 6.644282817840576, "min": -16.745819091796875, "p10": -2.831385803222656, "median": 3.0015029907226562, "p90": 14.21392364501953, "max": 23.52301025390625, "pos_frac": 0.71875, "sample": [4.55340576171875, 9.290512084960938, 6.548919677734375, 8.703460693359375, 18.0062255859375, 0.0, -0.1713733673095703, -2.8657684326171875, 8.948436737060547, 15.512939453125, -4.914794921875, 6.982170104980469, -2.88592529296875, 9.774789810180664, 23.52301025390625, -0.20858001708984375, 1.0373611450195312, 8.55401611328125, 15.990142822265625, 4.362884521484375, -2.816650390625, -6.712249755859375, 1.596160888671875, 4.72674560546875, 1.750885009765625, -2.6550979614257812, -0.1652069091796875, 5.8035888671875, 16.48297119140625, 0.6213207244873047, 10.09771728515625, -10.11602783203125, 6.511016845703125, 17.555419921875, 5.3945465087890625, 4.387718200683594, 4.6468353271484375, 3.8291015625, 5.710601806640625, 3.7794189453125, -0.8910198211669922, -0.224365234375, 7.482177734375, -1.864105224609375, -2.2309188842773438, -1.0782585144042969, 9.383804321289062, -3.73016357421875, 10.070526123046875, 2.3736572265625, 7.3904266357421875, 0.16143798828125, 14.6634521484375, 16.66156005859375, 10.670501708984375, 7.386474609375, 5.295654296875, 9.14190673828125, 10.411376953125, 16.776397705078125, 1.43731689453125, -0.1212310791015625, 2.402599334716797, 10.516128540039062, 3.0389404296875, 21.2005615234375, -0.3585205078125, 0.0, 3.38262939453125, 4.618988037109375, 4.38177490234375, 5.923492431640625, 14.427001953125, 4.8954315185546875, 14.122604370117188, 2.667205810546875, 3.958587646484375, -2.0574111938476562, 9.66766357421875, -0.3269157409667969, -7.7705078125, -3.7771453857421875, -0.9771957397460938, 1.3197021484375, -0.9394073486328125, 10.522186279296875, -3.335296630859375, 6.3863372802734375, 3.6067123413085938, 6.7427978515625, 3.9474334716796875, 0.9979019165039062, 2.9558486938476562, -0.813323974609375, 0.96435546875, -8.3621826171875, 1.7185897827148438, 1.2585945129394531, -7.144073486328125, 14.038543701171875, -0.271240234375, 2.9640655517578125, 0.12029266357421875, 0.72796630859375, 12.49273681640625, -16.745819091796875, 4.80462646484375, -5.2985992431640625, 0.003643035888671875, 2.739776611328125, 13.29730224609375, 15.9100341796875, -0.1104736328125, 7.8499908447265625, -1.811981201171875, 0.4978179931640625, 10.732025146484375, 2.4010543823242188, 2.478302001953125, 0.8184890747070312, 3.922637939453125, 0.4866790771484375, -1.9552230834960938, 2.7527732849121094, 6.580787658691406, 0.98016357421875, 14.781326293945312, 0.9412040710449219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000087.npy"} {"epoch": 0.18219895287958116, "step": 88, "batch_size": 128, "mean": 3.22519588470459, "std": 7.246361255645752, "min": -27.55804443359375, "p10": -4.0638671875, "median": 2.2784414291381836, "p90": 12.737503051757812, "max": 25.614349365234375, "pos_frac": 0.6640625, "sample": [8.913143157958984, 14.73193359375, 12.79656982421875, 0.6453666687011719, 16.9537353515625, 2.9202346801757812, -4.214385986328125, -0.22461700439453125, 8.2052001953125, 2.3521728515625, 2.2199535369873047, -3.114593505859375, 2.6863861083984375, 20.25067138671875, -0.24380874633789062, 0.55023193359375, -0.20443153381347656, 9.133026123046875, 1.3474273681640625, -2.29071044921875, 12.18853759765625, -4.514556884765625, 10.598602294921875, -3.999359130859375, -0.96783447265625, 6.9052734375, -5.435020446777344, 1.0075759887695312, 15.7945556640625, -5.1267547607421875, -5.93798828125, 10.651954650878906, 4.588775634765625, 5.09698486328125, 2.3369293212890625, -0.9948387145996094, 20.48150634765625, 0.0, 0.10033988952636719, 2.122344970703125, 6.96783447265625, 4.383392333984375, 10.014602661132812, -1.460968017578125, 14.941879272460938, 13.616424560546875, 0.791259765625, -6.6616058349609375, 2.84698486328125, 4.2944488525390625, 0.0, 2.407958984375, 3.510498046875, 4.58612060546875, 12.712188720703125, 0.8385009765625, -0.9493865966796875, 1.0843887329101562, -0.2886505126953125, 4.71014404296875, 9.796539306640625, -0.6596279144287109, 0.6966323852539062, -5.875251770019531, 5.435802459716797, 3.03125, -3.8226394653320312, 10.4881591796875, 0.27532196044921875, 10.4591064453125, 3.465259552001953, -4.222198486328125, 10.046630859375, 0.05708503723144531, -1.49383544921875, -19.18157958984375, 2.8832244873046875, 1.2498664855957031, -0.132080078125, -2.3183746337890625, 7.5693817138671875, 7.5488433837890625, 8.029571533203125, -1.9309730529785156, -0.912445068359375, -4.22393798828125, -2.9687271118164062, -5.2330322265625, 6.9548797607421875, 7.612884521484375, 0.4886932373046875, 5.914886474609375, 2.7081146240234375, 2.62371826171875, 5.4336090087890625, 15.829315185546875, 1.28369140625, -2.70721435546875, -2.7926177978515625, 17.90020751953125, 2.47747802734375, 2.999664306640625, -1.6094722747802734, 9.514923095703125, -1.72833251953125, 25.614349365234375, 13.4212646484375, -0.5275096893310547, 8.75885009765625, -7.3128662109375, 9.56500244140625, 0.13234710693359375, 1.599456787109375, 2.725616455078125, 21.03204345703125, 3.9167613983154297, -1.5005989074707031, 0.008928298950195312, 1.13446044921875, -0.5579833984375, -27.55804443359375, 6.113739013671875, -2.901611328125, 7.42413330078125, 2.1702804565429688, 8.40362548828125, 4.4356689453125, -0.8917999267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000088.npy"} {"epoch": 0.18429319371727748, "step": 89, "batch_size": 128, "mean": 3.878573417663574, "std": 6.7189812660217285, "min": -17.63433837890625, "p10": -2.8394676208496095, "median": 2.950305938720703, "p90": 11.305084228515625, "max": 27.7852783203125, "pos_frac": 0.796875, "sample": [0.3512306213378906, -10.622695922851562, 15.109710693359375, 1.5998077392578125, 5.078277587890625, 16.79815673828125, -1.3060016632080078, 1.6348648071289062, 2.21636962890625, 8.106979370117188, 4.11431884765625, 2.97943115234375, 2.0697479248046875, 11.198333740234375, 2.12579345703125, 10.548431396484375, 0.8970966339111328, -4.294097900390625, 5.526451110839844, 2.3318328857421875, 4.58306884765625, 13.049102783203125, -2.9233627319335938, 0.7418022155761719, 1.3197860717773438, 1.8056106567382812, 5.538604736328125, -0.159942626953125, -1.2750701904296875, 2.9211807250976562, 2.0867767333984375, 0.759124755859375, -0.1337890625, 14.642578125, 12.499481201171875, 27.7852783203125, 11.468505859375, -7.1451416015625, 1.5474090576171875, 1.835601806640625, -17.63433837890625, -0.457183837890625, 2.00732421875, -7.4884033203125, 1.22137451171875, 4.995086669921875, 7.634613037109375, -11.539352416992188, 0.99853515625, -4.03192138671875, -0.20711898803710938, 7.7087249755859375, 4.004730224609375, 0.1060943603515625, 9.020401000976562, 8.485198974609375, 6.0391845703125, -3.0753555297851562, 7.276481628417969, 10.69024658203125, 2.8249549865722656, 16.3658447265625, 8.138763427734375, 7.918830871582031, 7.725830078125, 0.98333740234375, -7.1232452392578125, 1.3331298828125, 0.7852249145507812, -2.8035125732421875, 4.739677429199219, 5.650115966796875, 4.644966125488281, 6.3403472900390625, -4.9701690673828125, 0.42340850830078125, 3.6886215209960938, 8.15350341796875, 22.361328125, -5.170806884765625, 0.732879638671875, 7.20068359375, 2.5048065185546875, 8.798126220703125, 1.799652099609375, 11.23504638671875, 2.8460922241210938, 13.88299560546875, 9.64471435546875, 20.619110107421875, 5.605064392089844, -1.13818359375, 12.278099060058594, 2.153076171875, 3.3681488037109375, 3.2995452880859375, 25.5185546875, -2.0353546142578125, 1.7177200317382812, 4.031532287597656, 7.57733154296875, 1.7945785522460938, 3.0552215576171875, -1.422027587890625, 3.8754711151123047, 5.414817810058594, -1.8746185302734375, -1.201904296875, 1.7384033203125, 4.6709136962890625, 4.1592559814453125, 3.2957115173339844, 9.5230712890625, 11.1373291015625, 5.2198486328125, 4.08258056640625, 4.828834533691406, 3.24468994140625, 8.020278930664062, -13.376937866210938, 1.6057682037353516, 10.45245361328125, 5.0101318359375, 2.4051284790039062, 2.182432174682617, 2.80419921875, 1.7863006591796875, -2.789276123046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000089.npy"} {"epoch": 0.18638743455497384, "step": 90, "batch_size": 128, "mean": 3.4592151641845703, "std": 7.2304368019104, "min": -18.22052001953125, "p10": -4.448576354980468, "median": 2.6097564697265625, "p90": 12.807598876953124, "max": 23.939361572265625, "pos_frac": 0.703125, "sample": [8.705718994140625, 3.28851318359375, 1.6028366088867188, 15.86712646484375, -10.110092163085938, -0.36968994140625, -3.045654296875, 0.8441848754882812, -1.99041748046875, 11.559295654296875, 1.9607067108154297, 2.8697357177734375, -7.0270538330078125, -0.3197975158691406, -0.10662841796875, 6.90936279296875, 21.545944213867188, 16.655303955078125, 14.6754150390625, 1.1392059326171875, 5.8066864013671875, 0.01947021484375, 2.5443878173828125, 3.101654052734375, -12.751785278320312, -4.5531768798828125, 11.15517807006836, 7.511932373046875, 6.016265869140625, 2.62188720703125, -0.78021240234375, 6.196113586425781, 8.4522705078125, 3.587249755859375, 8.103057861328125, 2.2330780029296875, -4.2230224609375, -4.40374755859375, 0.64111328125, 6.52789306640625, -4.886165618896484, 2.0740203857421875, -1.6277389526367188, -9.5487060546875, 6.430694580078125, 12.291015625, 5.335296630859375, 8.470596313476562, 5.371192932128906, 2.597625732421875, 9.657485961914062, -2.8798599243164062, 0.532073974609375, 1.9319629669189453, 1.28143310546875, 1.3462677001953125, -12.67034912109375, -0.36443328857421875, -0.17429351806640625, 6.0011138916015625, 0.59222412109375, 0.0, 12.87164306640625, 7.1572723388671875, 10.018386840820312, -0.270416259765625, 6.979766845703125, 3.3121261596679688, 0.38470458984375, 12.7801513671875, 1.9039764404296875, 1.690399169921875, 6.0035400390625, 18.67938232421875, 11.3724365234375, 1.763824462890625, 0.18944263458251953, 3.8893661499023438, 2.8057174682617188, 9.67803955078125, 10.4544677734375, -18.22052001953125, 9.2701416015625, -11.58416748046875, 0.00453948974609375, 3.672027587890625, 4.846900939941406, 2.4958229064941406, -0.4911537170410156, -2.961090087890625, -0.568756103515625, -3.803955078125, 11.22021484375, 12.072998046875, -3.0243988037109375, 3.119873046875, -5.3478851318359375, 5.2920379638671875, 9.254837036132812, 10.511077880859375, -1.90478515625, 10.91684341430664, 8.328704833984375, 0.48188018798828125, -1.81390380859375, 23.939361572265625, -2.9943161010742188, 4.089508056640625, -7.245330810546875, -13.07806396484375, 4.6790313720703125, 1.8303985595703125, -5.65997314453125, 13.2435302734375, 2.85101318359375, -0.4622821807861328, 13.175048828125, 2.485321044921875, 14.557815551757812, 3.04241943359375, -1.4474716186523438, -2.3592987060546875, 20.95611572265625, 13.74017333984375, 3.5791053771972656, 12.183151245117188, 1.119384765625, 12.901596069335938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000090.npy"} {"epoch": 0.18848167539267016, "step": 91, "batch_size": 128, "mean": 3.4569621086120605, "std": 7.640954971313477, "min": -13.009414672851562, "p10": -6.1342010498046875, "median": 3.09197998046875, "p90": 13.632415771484375, "max": 24.762176513671875, "pos_frac": 0.6640625, "sample": [6.071136474609375, 1.5850372314453125, -1.2905807495117188, 16.970108032226562, 8.915374755859375, 13.78729248046875, -0.07367897033691406, 3.753448486328125, 4.9974365234375, -3.8571395874023438, 0.9765777587890625, 5.4088592529296875, 7.171806335449219, 6.708244323730469, 6.301225662231445, 8.20928955078125, 1.5027313232421875, -13.009414672851562, -3.59356689453125, 0.3811321258544922, 8.170333862304688, 3.7483978271484375, 5.2283782958984375, 7.0345611572265625, -1.5671730041503906, 6.345245361328125, 0.11803054809570312, -0.71099853515625, -0.5460987091064453, -8.341140747070312, 13.6103515625, -0.23516845703125, -0.793731689453125, -6.6219482421875, -3.1045761108398438, 10.644058227539062, -5.5387725830078125, 6.644622802734375, 5.307777404785156, 6.628448486328125, -7.867828369140625, -0.5188369750976562, 3.482696533203125, -2.395538330078125, 2.169464111328125, 18.9815673828125, -6.11895751953125, -4.7506561279296875, 0.0, -11.346054077148438, -5.8365325927734375, -3.5531005859375, -10.929168701171875, 2.22808837890625, 6.521697998046875, 6.302703857421875, 0.4322357177734375, -5.63848876953125, 13.516754150390625, 5.97906494140625, 9.334228515625, -8.72802734375, 3.9857330322265625, -0.10112380981445312, -2.610136032104492, -1.505462646484375, 23.178558349609375, 1.7847900390625, 0.2616119384765625, 10.993392944335938, -7.951202392578125, -4.1131591796875, 24.762176513671875, 3.831026077270508, 15.853851318359375, 1.1026973724365234, 4.3847503662109375, 0.7114658355712891, 3.405609130859375, 11.583587646484375, -8.963943481445312, -6.169769287109375, 22.57080078125, 10.632369995117188, 13.68389892578125, 0.0, 8.836372375488281, 3.6431884765625, 8.84130859375, 11.2218017578125, 4.0054931640625, 9.286651611328125, 4.763648986816406, -9.146484375, 0.81622314453125, 13.69000244140625, -8.358932495117188, 2.0942115783691406, 4.4874114990234375, 19.58587646484375, 8.618560791015625, 8.9473876953125, 7.78692626953125, 5.038787841796875, 5.213165283203125, -9.143447875976562, -3.728363037109375, 11.756645202636719, -0.875213623046875, 18.2945556640625, -3.5417938232421875, 17.72216796875, 20.142486572265625, -0.29456520080566406, 0.332977294921875, 10.261810302734375, 2.3248748779296875, -4.29815673828125, 8.15179443359375, 1.3497390747070312, 9.023239135742188, -3.3240966796875, 1.919219970703125, 1.7646942138671875, 4.94384765625, 2.778350830078125, 6.98272705078125, 1.063323974609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000091.npy"} {"epoch": 0.1905759162303665, "step": 92, "batch_size": 128, "mean": 3.780871868133545, "std": 7.97656774520874, "min": -27.07965087890625, "p10": -5.178341674804687, "median": 3.9127674102783203, "p90": 14.720816040039063, "max": 21.78656005859375, "pos_frac": 0.7421875, "sample": [-5.10076904296875, 0.15703582763671875, -3.37725830078125, 4.701072692871094, 2.7436370849609375, 6.150909423828125, 4.1996612548828125, 5.39288330078125, 14.75860595703125, 7.969573974609375, 1.2143402099609375, 2.3260669708251953, 4.667633056640625, -1.833892822265625, 21.428619384765625, 17.359619140625, 21.78656005859375, -7.4722900390625, 8.08209228515625, 1.2206573486328125, -6.919586181640625, 7.202423095703125, 15.332443237304688, 12.481353759765625, 1.36859130859375, -3.1262969970703125, 3.5216522216796875, 14.36639404296875, 2.907806396484375, -1.818389892578125, 4.87738037109375, 0.161773681640625, 13.19256591796875, 7.6276092529296875, 1.5469512939453125, 0.831939697265625, 0.553619384765625, 4.3365020751953125, 14.1357421875, 7.96728515625, 1.6750030517578125, 1.898193359375, 0.08825492858886719, -27.07965087890625, 11.314620971679688, 3.146331787109375, 4.320587158203125, 2.714111328125, 1.182373046875, 1.3851089477539062, -5.359344482421875, -8.9564208984375, -7.661346435546875, -1.233795166015625, 15.80657958984375, -5.9557342529296875, -10.809326171875, 0.07320404052734375, -6.95263671875, 7.4638671875, 5.294502258300781, 5.6370391845703125, 6.86676025390625, 4.962890625, 6.32568359375, 4.048332214355469, 10.240753173828125, 0.0, -2.50836181640625, 11.283203125, -2.467803955078125, 5.64288330078125, 1.95574951171875, -1.030792236328125, -2.2084808349609375, 8.244041442871094, -0.35407257080078125, 1.6894607543945312, 10.154678344726562, -0.9453744888305664, -8.259552001953125, 14.9752197265625, 5.39532470703125, -22.219207763671875, -1.8068313598632812, 0.301483154296875, 5.962493896484375, 18.38751220703125, 19.297637939453125, 4.9575653076171875, 8.00601577758789, 6.480224609375, 0.21120452880859375, -0.9081268310546875, 0.47878074645996094, 21.122711181640625, -1.289581298828125, -4.5367431640625, 19.61920166015625, 5.5012664794921875, 10.06195068359375, 10.576309204101562, -13.626953125, 14.704620361328125, 8.926971435546875, 4.239776611328125, 3.054473876953125, 4.008819580078125, 2.475341796875, 4.300811767578125, 8.7086181640625, 4.368598937988281, 8.995513916015625, -11.7840576171875, 6.6112823486328125, 7.76995849609375, 0.3274688720703125, 0.3555145263671875, -1.07635498046875, 9.101531982421875, -3.97515869140625, 8.859445571899414, 3.8167152404785156, 2.2789688110351562, 15.9403076171875, 18.3338623046875, 9.63201904296875, -1.5249481201171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000092.npy"} {"epoch": 0.19267015706806281, "step": 93, "batch_size": 128, "mean": 3.6341428756713867, "std": 7.513704299926758, "min": -18.925994873046875, "p10": -4.7602073669433596, "median": 2.502216339111328, "p90": 14.977032470703126, "max": 24.1461181640625, "pos_frac": 0.6953125, "sample": [0.25441741943359375, 14.9598388671875, 1.17034912109375, 22.11474609375, -4.96087646484375, 8.112770080566406, -2.24542236328125, 12.1927490234375, 4.034423828125, 5.38578987121582, 3.8819580078125, -8.466278076171875, 1.57666015625, 3.739410400390625, 4.82403564453125, 12.229248046875, 2.152130126953125, 16.877288818359375, -2.2581863403320312, 4.5821380615234375, 6.1966094970703125, 4.916259765625, 3.1891231536865234, 4.1346893310546875, 16.286720275878906, 0.92132568359375, -0.29392242431640625, 3.5398406982421875, 12.929718017578125, 10.399017333984375, 3.1547088623046875, -7.466461181640625, 0.5582733154296875, -1.8745269775390625, 2.7233123779296875, 6.15240478515625, 4.2585906982421875, 4.36932373046875, -4.4979400634765625, 14.451904296875, 3.232574462890625, 6.0809326171875, 3.5089874267578125, -4.7769927978515625, 0.138153076171875, 4.303058624267578, 0.214599609375, 15.01715087890625, 1.190582275390625, -1.0226058959960938, 1.4188308715820312, 6.7384796142578125, 17.970489501953125, -6.819122314453125, 8.294189453125, 1.557952880859375, -1.6034469604492188, -0.7209892272949219, 8.753509521484375, 1.0978279113769531, 7.6884613037109375, 5.60089111328125, 16.28619384765625, -0.7075653076171875, -1.199493408203125, -6.50177001953125, 11.022308349609375, 16.64752197265625, -6.3388214111328125, 2.15716552734375, 3.418752670288086, 2.992349624633789, -4.114837646484375, -1.313873291015625, 6.8745269775390625, -3.201263427734375, 13.4857177734375, -0.932861328125, 19.0120849609375, -18.925994873046875, 21.71832275390625, 7.0802001953125, 15.445465087890625, -6.45489501953125, 7.1629638671875, 1.9816970825195312, 0.0, -9.499481201171875, 2.195028305053711, 4.828254699707031, 16.416412353515625, 12.87091064453125, -4.17919921875, 2.2433319091796875, 5.429210662841797, -12.72607421875, 6.5373992919921875, -0.9303550720214844, -0.970916748046875, -6.637060165405273, 10.38140869140625, 8.088455200195312, 2.073822021484375, 1.939849853515625, 3.8516845703125, 0.0, 1.4841194152832031, -1.3967208862304688, -4.753013610839844, -0.787384033203125, -1.5181427001953125, 0.22769546508789062, 5.9098663330078125, 2.2811203002929688, -12.297943115234375, 23.1971435546875, 24.1461181640625, -2.0717201232910156, -1.1727523803710938, -1.3340682983398438, 4.917205810546875, 3.25836181640625, 1.3270263671875, 14.67633056640625, 0.2503528594970703, 2.1886138916015625, 0.7954177856445312, 6.266387939453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000093.npy"} {"epoch": 0.19476439790575917, "step": 94, "batch_size": 128, "mean": 4.109147071838379, "std": 9.126523971557617, "min": -19.314453125, "p10": -7.035502624511718, "median": 2.9252662658691406, "p90": 15.878195190429686, "max": 29.6976318359375, "pos_frac": 0.71875, "sample": [-0.628753662109375, 11.205650329589844, -6.115509033203125, 16.46875, 5.246368408203125, 9.98138427734375, 18.92657470703125, 2.586824417114258, 5.02642822265625, 9.309295654296875, 3.27191162109375, -5.152992248535156, -13.359695434570312, 0.5381851196289062, 11.922119140625, -0.46881103515625, 13.758697509765625, 0.8743515014648438, -0.67791748046875, 0.7235565185546875, 13.353790283203125, 13.508071899414062, 2.9796142578125, -1.6077957153320312, 18.948165893554688, 16.304351806640625, 2.0754013061523438, 5.4623565673828125, -11.5576171875, -1.772979736328125, 12.214447021484375, 21.0594482421875, -11.037750244140625, 24.4080810546875, 7.033969879150391, 6.218780517578125, 9.34295654296875, 2.157665252685547, -7.5952911376953125, 22.29205322265625, 4.819480895996094, 29.6976318359375, 14.483642578125, -0.28498077392578125, 15.695556640625, 3.9771881103515625, 15.39031982421875, -5.60089111328125, 7.814971923828125, -2.466114044189453, 1.359527587890625, 0.27059173583984375, -0.6209335327148438, 7.752410888671875, 3.3237380981445312, -6.2230987548828125, 7.63836669921875, -3.45556640625, 8.73883056640625, -8.1776123046875, -0.091552734375, -3.24652099609375, 2.4703521728515625, -3.9668426513671875, 1.942535400390625, 14.49566650390625, 14.434478759765625, 0.7411727905273438, -6.194328308105469, 13.166046142578125, 1.02313232421875, 0.6301078796386719, 6.61102294921875, 12.521865844726562, 20.0281982421875, -6.79559326171875, 3.57464599609375, -2.0505638122558594, 2.7342529296875, 3.5311279296875, 3.848846435546875, 1.2112884521484375, -6.0464324951171875, -6.553741455078125, -11.570365905761719, 20.20013427734375, 13.2962646484375, -8.40618896484375, 17.044830322265625, -12.666595458984375, -19.314453125, 4.6593780517578125, 2.725688934326172, 9.174812316894531, 5.355560302734375, 13.584449768066406, 6.9100341796875, 8.333450317382812, 17.65887451171875, -8.0130615234375, 6.8477935791015625, -8.4539794921875, 11.555023193359375, 2.0561599731445312, 3.7575225830078125, 7.4554290771484375, 2.8709182739257812, 9.454559326171875, 0.712310791015625, 1.8943557739257812, 4.2171478271484375, 4.949005126953125, 1.2476654052734375, -17.80084228515625, 0.9871826171875, 0.071624755859375, 8.106201171875, 1.27996826171875, 0.4970550537109375, -6.662353515625, 9.700729370117188, 2.60711669921875, 28.55560302734375, 12.640632629394531, -11.12774658203125, 0.46759033203125, 0.11989498138427734, -0.384918212890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000094.npy"} {"epoch": 0.1968586387434555, "step": 95, "batch_size": 128, "mean": 4.5721259117126465, "std": 9.134565353393555, "min": -24.954559326171875, "p10": -6.688613891601562, "median": 4.0120697021484375, "p90": 16.731369018554684, "max": 27.1103515625, "pos_frac": 0.71875, "sample": [-2.8577003479003906, -0.3663482666015625, 3.041015625, 24.728271484375, 7.56817626953125, 2.860992431640625, 2.2928314208984375, 6.588470458984375, 24.03875732421875, 9.171630859375, -0.943328857421875, 0.494171142578125, 2.27569580078125, 2.6271095275878906, -8.009201049804688, 6.958740234375, 3.929901123046875, 6.909904479980469, 11.642288208007812, 2.674896240234375, 13.52032470703125, 6.4836578369140625, -9.0994873046875, 8.561187744140625, -2.423828125, 2.4547882080078125, 26.770172119140625, -5.382904052734375, 16.240692138671875, -24.954559326171875, 23.36370849609375, -1.2471923828125, 9.603412628173828, -6.3690032958984375, 25.39190673828125, 11.37091064453125, 2.80615234375, 7.194976806640625, 3.3325042724609375, 10.024658203125, 3.0032958984375, 7.16192626953125, -1.625030517578125, -0.26972198486328125, -1.4021377563476562, 7.819305419921875, 4.707633972167969, 10.538070678710938, 20.42327880859375, 2.618621826171875, -0.1290130615234375, 8.499954223632812, 5.981109619140625, 0.28664207458496094, 6.649139404296875, 6.667510986328125, 4.794700622558594, 4.684654235839844, 5.3018951416015625, 10.113311767578125, 7.3525390625, -1.340606689453125, 3.3529052734375, -0.13470458984375, 2.6070480346679688, 6.975860595703125, -5.5030364990234375, 11.042510986328125, 6.0937957763671875, 3.685770034790039, -5.619895935058594, 20.1571044921875, 8.71112060546875, 17.87628173828125, 0.652008056640625, -1.0774993896484375, 0.20460891723632812, 15.805496215820312, 3.2560348510742188, 5.8028411865234375, -8.951950073242188, 10.227130889892578, 4.9376678466796875, 1.2626018524169922, 5.561065673828125, 24.30615234375, 6.6274566650390625, 8.116287231445312, -1.6391448974609375, -14.81573486328125, 4.4653472900390625, 6.768402099609375, -3.983795166015625, 1.942108154296875, -13.1080322265625, 2.919219970703125, 8.93536376953125, 6.56500244140625, -7.489013671875, 4.09423828125, -4.3541717529296875, -2.5508193969726562, -7.4343719482421875, 1.8110275268554688, 11.306324005126953, -9.748466491699219, 5.480670928955078, -1.6244049072265625, 24.677886962890625, -9.855316162109375, 3.7809906005859375, 3.9093017578125, -2.3101654052734375, 8.117919921875, 3.0196533203125, 15.03106689453125, 1.2220458984375, 10.011283874511719, 7.413604736328125, -13.8135986328125, -8.33575439453125, -0.08557891845703125, 5.9985809326171875, 20.795013427734375, 4.471405029296875, 26.322479248046875, -10.868896484375, 27.1103515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000095.npy"} {"epoch": 0.19895287958115182, "step": 96, "batch_size": 128, "mean": 4.245677947998047, "std": 8.52260684967041, "min": -20.3970947265625, "p10": -4.974618530273437, "median": 3.667142868041992, "p90": 15.919281005859371, "max": 29.5897216796875, "pos_frac": 0.7109375, "sample": [-2.39202880859375, -10.7779541015625, -9.131378173828125, 1.7194900512695312, 17.6099853515625, 8.981399536132812, -13.639837265014648, -4.666748046875, 2.257781982421875, 1.6434783935546875, 3.6013450622558594, -5.4027099609375, -5.370391845703125, -0.8001708984375, 8.682247161865234, 1.7955799102783203, 4.116912841796875, -5.3161163330078125, 1.651824951171875, 3.8151092529296875, 9.85968017578125, 2.55694580078125, 7.478851318359375, 7.3784942626953125, -4.3579559326171875, 3.8637237548828125, 17.890670776367188, 14.2032470703125, 2.1907272338867188, 4.633331298828125, 26.636688232421875, 6.1186676025390625, 10.68878173828125, 11.703140258789062, 7.62103271484375, 3.8750381469726562, 4.141502380371094, 3.4717178344726562, 2.2776336669921875, 1.1981582641601562, 6.3880767822265625, 4.24102783203125, -3.0645751953125, 19.5899658203125, -5.124176025390625, 5.403511047363281, -8.585716247558594, -0.4617118835449219, 4.0988922119140625, -2.116729736328125, 4.800323486328125, -2.499420166015625, 8.688064575195312, -4.533447265625, 4.414306640625, -3.0397796630859375, 5.188804626464844, 14.38348388671875, 3.323760986328125, 15.134521484375, -1.1306419372558594, 3.8196640014648438, 12.430877685546875, 9.358673095703125, 6.2935943603515625, 22.124267578125, 29.5897216796875, 3.17108154296875, -1.685546875, 15.585113525390625, 4.314208984375, 15.480903625488281, 2.1845016479492188, -11.586502075195312, 24.76507568359375, 5.278167724609375, 9.345001220703125, -2.4341278076171875, 17.947021484375, 0.1912841796875, 8.013259887695312, 1.0028533935546875, 4.816810607910156, -7.62408447265625, 1.5937957763671875, -1.4388465881347656, 0.2621307373046875, 11.618179321289062, 16.835357666015625, 0.1339569091796875, -1.0621814727783203, -0.46451568603515625, 4.620729446411133, 16.699005126953125, 18.79632568359375, -6.42547607421875, 3.732940673828125, 7.304389953613281, 10.2713623046875, 13.16961669921875, -14.768798828125, 21.093719482421875, 2.0605010986328125, -1.858175277709961, 0.2113800048828125, 11.436538696289062, -0.493408203125, 27.12353515625, 0.7475433349609375, 2.903076171875, 9.533203125, 0.0, 1.4366912841796875, 2.733264923095703, 0.6194896697998047, 3.8746299743652344, -4.9105224609375, 7.149658203125, 4.561767578125, 10.56060791015625, -3.804342269897461, 7.193159103393555, -3.6482810974121094, -3.0638961791992188, 2.482666015625, 8.42010498046875, -2.6852569580078125, -20.3970947265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000096.npy"} {"epoch": 0.20104712041884817, "step": 97, "batch_size": 128, "mean": 4.597841262817383, "std": 9.981614112854004, "min": -18.570648193359375, "p10": -7.317329406738281, "median": 3.9876327514648438, "p90": 17.527389526367188, "max": 32.3145751953125, "pos_frac": 0.6953125, "sample": [5.576129913330078, 6.96429443359375, 6.6177978515625, -7.131866455078125, 6.688911437988281, 3.0518569946289062, -1.1150627136230469, 5.501258850097656, 8.0537109375, 8.283538818359375, -1.6228370666503906, 32.3145751953125, 10.226959228515625, 16.523040771484375, 0.5642318725585938, 3.5159912109375, 17.341949462890625, -0.88055419921875, -5.26580810546875, 11.83038330078125, 3.8771209716796875, 2.2548370361328125, -2.966064453125, -0.652923583984375, 8.5186767578125, -8.382537841796875, 4.09814453125, -6.3722686767578125, 8.428997039794922, 8.398292541503906, -7.4450225830078125, 0.45288848876953125, -3.9990997314453125, -6.777801513671875, 7.28472900390625, 25.879135131835938, 10.501911163330078, -4.8646240234375, -6.75933837890625, -9.656494140625, 2.2552947998046875, 2.23675537109375, 2.6903533935546875, 22.1356201171875, -8.882354736328125, 6.5179443359375, 5.737762451171875, 8.96258544921875, 1.2952880859375, 8.764976501464844, 12.63079833984375, -7.262603759765625, 24.82012939453125, 1.826171875, 8.317108154296875, 12.176101684570312, -6.81011962890625, 8.44158935546875, 1.9984130859375, -0.6624755859375, 10.073844909667969, 6.7098388671875, -1.7960052490234375, 7.773384094238281, 4.6329498291015625, 0.0, 27.29730224609375, 8.37060546875, 9.527099609375, 30.238922119140625, 1.2251949310302734, 11.037338256835938, 8.64666748046875, 5.388275146484375, -0.7231960296630859, 27.81060791015625, 1.807373046875, -15.400421142578125, -7.869850158691406, 19.2476806640625, 7.5948486328125, 1.1840324401855469, -5.683349609375, 4.26483154296875, 13.119659423828125, 5.629329681396484, 0.06353950500488281, 28.477203369140625, 5.2261962890625, -2.6280670166015625, 7.875461578369141, -11.55865478515625, 7.711151123046875, 1.8663406372070312, 8.26593017578125, -18.570648193359375, 0.0, 7.56451416015625, 12.747268676757812, 1.067169189453125, 2.001312255859375, 28.7158203125, -10.055831909179688, 2.7070083618164062, 3.3022384643554688, 2.41729736328125, -1.4620208740234375, -6.482574462890625, 25.6790771484375, 7.9176025390625, -4.409423828125, 25.046722412109375, -8.71408462524414, -12.254302978515625, 0.032745361328125, 12.43206787109375, 3.7083892822265625, 8.307037353515625, -0.34229278564453125, -12.632049560546875, 9.714630126953125, -4.418220520019531, 1.2194061279296875, 12.9619140625, 17.9600830078125, -13.983001708984375, 6.06402587890625, 8.829345703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000097.npy"} {"epoch": 0.2031413612565445, "step": 98, "batch_size": 128, "mean": 4.718062877655029, "std": 10.269811630249023, "min": -27.5106201171875, "p10": -5.615592384338379, "median": 3.5158538818359375, "p90": 17.705661010742187, "max": 33.441619873046875, "pos_frac": 0.6640625, "sample": [11.609149932861328, 11.6326904296875, 23.144775390625, 3.113117218017578, 30.52716064453125, 18.048004150390625, 15.277023315429688, -5.6095733642578125, 3.635009765625, -1.0410537719726562, 15.658660888671875, -13.189849853515625, -1.5811767578125, 5.5316619873046875, 8.81121826171875, 8.168853759765625, 1.2480621337890625, -12.32220458984375, -6.7372894287109375, 1.7521133422851562, 11.431915283203125, -4.68060302734375, -5.844085693359375, 4.606109619140625, 2.8237533569335938, 4.060401916503906, -0.005218505859375, -5.4620208740234375, -5.52947998046875, -17.873809814453125, 1.9660472869873047, -27.5106201171875, -5.12939453125, -8.76617431640625, 3.300579071044922, -1.220367431640625, -5.629636764526367, 15.26959228515625, 3.4898223876953125, 3.0941390991210938, 2.83233642578125, 8.72869873046875, 0.57110595703125, 6.3292388916015625, -0.0229949951171875, 17.84423828125, 1.04638671875, 17.56402587890625, -5.8198394775390625, 14.221355438232422, 32.6292724609375, 13.530120849609375, -0.8401870727539062, 3.662017822265625, -0.2894477844238281, -1.914703369140625, 30.60357666015625, 13.7308349609375, -3.07989501953125, 21.462982177734375, -15.690139770507812, -3.9471282958984375, -1.1667633056640625, -5.260467529296875, 23.788909912109375, -1.3813629150390625, 10.465087890625, 5.138885498046875, 6.376821517944336, -2.8487091064453125, 9.019790649414062, 16.28424072265625, -13.429107666015625, 3.5418853759765625, 20.705001831054688, 14.521575927734375, 3.436981201171875, 12.26416015625, 1.01287841796875, 17.646270751953125, 5.8302459716796875, 17.967254638671875, -9.423553466796875, -1.0351524353027344, 16.148681640625, -4.223793029785156, -0.6525039672851562, 1.7617969512939453, 3.4412994384765625, -0.634033203125, 21.77252197265625, 7.700164794921875, 5.9802703857421875, -1.324127197265625, 7.3789825439453125, 7.9288330078125, 0.6927032470703125, 7.2544708251953125, 2.5373992919921875, 5.4915771484375, 1.008392333984375, 15.470733642578125, 33.441619873046875, 4.672882080078125, 7.0469970703125, -4.981365203857422, -3.6748809814453125, 6.4901580810546875, 5.0724945068359375, 5.2857208251953125, -0.75201416015625, 3.046966552734375, 4.306060791015625, 7.234355926513672, -1.0241165161132812, 8.492767333984375, 2.38055419921875, -2.366455078125, 7.2482757568359375, 11.811279296875, 12.164863586425781, -5.14886474609375, 15.587646484375, 7.7557830810546875, 20.164520263671875, 3.4167938232421875, -20.00799560546875, 11.840606689453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000098.npy"} {"epoch": 0.20523560209424083, "step": 99, "batch_size": 128, "mean": 5.145061492919922, "std": 10.554427146911621, "min": -24.86822509765625, "p10": -7.682640075683593, "median": 4.897870063781738, "p90": 18.967984008789063, "max": 26.994232177734375, "pos_frac": 0.7265625, "sample": [-19.985626220703125, 0.64617919921875, -10.44061279296875, 26.994232177734375, 4.5685882568359375, 12.140777587890625, -0.93878173828125, 16.71588134765625, 0.45263671875, -4.341644287109375, 6.39202880859375, -11.125732421875, -0.113616943359375, 6.431892395019531, -7.4713592529296875, 16.383636474609375, 16.0223388671875, 8.636360168457031, 7.854248046875, -2.5720977783203125, 5.124706268310547, 3.3817520141601562, 10.86224365234375, -17.702392578125, 25.16973876953125, 8.30413818359375, 4.86433219909668, -10.901336669921875, 6.82427978515625, 15.9573974609375, 14.469818115234375, -4.9560546875, 19.17889404296875, 2.4201583862304688, 6.2845458984375, 0.9019775390625, 5.3052978515625, 8.702239990234375, 24.741241455078125, 4.7782135009765625, 10.163894653320312, 19.51690673828125, 6.955902099609375, 17.743408203125, 6.8856658935546875, -4.8519134521484375, 12.478950500488281, 2.042938232421875, -0.28546142578125, 0.946807861328125, -13.1895751953125, -20.61822509765625, -20.19873046875, 11.230659484863281, 11.611602783203125, -3.3943099975585938, 14.550287246704102, -18.393692016601562, 15.707275390625, 10.32049560546875, 5.09930419921875, 0.641693115234375, 20.01348876953125, 1.0372390747070312, -3.427154541015625, 15.48809814453125, -24.86822509765625, 9.119720458984375, 3.2316741943359375, 18.877593994140625, 1.9981689453125, 17.354339599609375, -2.1545562744140625, 0.6899051666259766, -1.21771240234375, 7.943634033203125, 4.931407928466797, 20.240936279296875, 5.468605041503906, 3.683441162109375, 8.409576416015625, 1.6957244873046875, -1.757568359375, 4.434791564941406, 16.9405517578125, 3.8486595153808594, 13.4962158203125, -2.31390380859375, 7.67742919921875, 4.586212158203125, -8.175628662109375, 2.9493408203125, 2.1619873046875, 7.649009704589844, 19.335693359375, 0.639434814453125, 16.6024169921875, -11.250335693359375, 1.702362060546875, -4.807464599609375, 19.2841796875, -0.7886962890625, 0.0, 5.209571838378906, -2.40374755859375, 18.055023193359375, 3.5351181030273438, -3.181671142578125, 16.631927490234375, 22.975250244140625, 22.304046630859375, 12.84515380859375, -22.080718994140625, 14.55181884765625, 2.4200706481933594, 7.343070983886719, 11.4969482421875, 21.415771484375, -1.387939453125, 2.053009033203125, 3.4287796020507812, -2.65032958984375, 15.44573974609375, -0.777130126953125, 13.052459716796875, 7.819923400878906, 21.526275634765625, 3.284454345703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000099.npy"} {"epoch": 0.20732984293193718, "step": 100, "batch_size": 128, "mean": 4.396217346191406, "std": 9.802384376525879, "min": -18.52777099609375, "p10": -8.538117980957031, "median": 3.231168746948242, "p90": 18.940263366699217, "max": 28.98388671875, "pos_frac": 0.6875, "sample": [8.047760009765625, 2.91143798828125, 12.273574829101562, 3.1834869384765625, 15.88726806640625, 4.28961181640625, -10.351715087890625, 18.846527099609375, 0.552825927734375, 9.368927001953125, -3.8243408203125, 1.140899658203125, 0.8499755859375, 4.219970703125, 12.978790283203125, 28.98388671875, 0.9536209106445312, 1.0740203857421875, -9.577667236328125, 8.29730224609375, 0.359161376953125, 6.167388916015625, -8.47308349609375, -2.0528621673583984, 1.7397308349609375, 11.11962890625, 2.268646240234375, -9.321319580078125, -0.10361099243164062, 17.022705078125, -12.991973876953125, 0.146575927734375, -1.226715087890625, 4.612174987792969, 4.89703369140625, 4.30810546875, -1.621673583984375, 19.158981323242188, 2.6581649780273438, 14.75445556640625, -3.424285888671875, 12.195846557617188, 6.31365966796875, 2.016357421875, -1.7908554077148438, 5.293701171875, 7.741203308105469, 22.7757568359375, 24.77001953125, 13.929351806640625, -1.6234207153320312, -8.689865112304688, -5.637790679931641, 5.63093376159668, 3.3722057342529297, 10.298095703125, -15.16107177734375, -2.966064453125, -18.52777099609375, 2.5839157104492188, 0.6131973266601562, 27.035491943359375, -5.584716796875, 4.8138580322265625, -0.474090576171875, 5.622802734375, -4.9146575927734375, -13.609771728515625, 3.803466796875, 5.61016845703125, 27.714874267578125, 8.98431396484375, 4.9454193115234375, 5.995464324951172, -1.0641403198242188, -8.730224609375, -5.80914306640625, 6.408660888671875, 0.7987823486328125, 11.495410919189453, -12.337577819824219, 5.232566833496094, 4.848419189453125, 2.055450439453125, 6.36676025390625, 3.278850555419922, -4.064140319824219, -1.0101089477539062, 15.58294677734375, 25.95977783203125, 0.81890869140625, 3.5762481689453125, -3.453857421875, 1.9591064453125, 6.5623779296875, 1.5888671875, 4.013633728027344, 2.752992630004883, -0.2801246643066406, 11.782867431640625, 22.47393798828125, -0.664093017578125, 24.239852905273438, 15.15673828125, 5.2053985595703125, -3.69635009765625, 16.31454086303711, 24.940109252929688, 0.85821533203125, 2.09228515625, 24.597076416015625, 5.658203125, -0.4234466552734375, -3.9747695922851562, -10.451507568359375, 16.085189819335938, -0.11209487915039062, 3.838958740234375, 21.374359130859375, -10.261749267578125, 0.8460578918457031, 9.649261474609375, 15.16632080078125, -3.969226837158203, 8.37094497680664, -0.6592254638671875, 21.485321044921875, -12.941238403320312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000100.npy"} {"epoch": 0.2094240837696335, "step": 101, "batch_size": 128, "mean": 4.451606273651123, "std": 10.341442108154297, "min": -23.12005615234375, "p10": -6.478089904785156, "median": 4.484272003173828, "p90": 15.52752685546875, "max": 35.16748046875, "pos_frac": 0.7265625, "sample": [6.2146148681640625, -23.12005615234375, 4.752166748046875, 15.621307373046875, 4.216377258300781, -0.157623291015625, 3.314300537109375, 5.7213134765625, -4.863525390625, 0.976837158203125, 17.92694091796875, 11.599945068359375, -5.676483154296875, 20.12237548828125, 32.5687255859375, 26.6126708984375, 2.7467823028564453, 5.141639709472656, -1.4270610809326172, 10.550521850585938, 8.117584228515625, 4.9884033203125, 11.506317138671875, 9.375045776367188, 1.289175033569336, 20.267471313476562, 6.2186737060546875, 1.3515472412109375, 0.035736083984375, 29.24114990234375, -1.2482681274414062, 6.17974853515625, 4.092685699462891, 5.086578369140625, 15.46710205078125, 8.919448852539062, 22.103958129882812, 15.487335205078125, 12.93820571899414, 1.376220703125, -9.375274658203125, 4.021614074707031, 9.55712890625, 10.863739013671875, 19.371009826660156, 15.337020874023438, 7.57135009765625, -2.9189910888671875, -0.3758697509765625, -12.2227783203125, -2.3335418701171875, 6.787109375, -17.09515380859375, 11.202987670898438, 20.894866943359375, 7.1605377197265625, 6.8842315673828125, -17.299041748046875, -1.268747329711914, 3.9338912963867188, -4.21978759765625, 6.844146728515625, 3.95068359375, 11.160102844238281, 0.18963623046875, -4.110357284545898, 6.066261291503906, 9.855819702148438, -0.7711334228515625, 15.273513793945312, -0.010467529296875, 3.418966293334961, -0.2113037109375, 5.47271728515625, 0.9622859954833984, 2.4239540100097656, 20.2628173828125, 1.2938613891601562, 3.5075836181640625, -21.039581298828125, -21.900634765625, 6.80755615234375, -9.33156967163086, 12.964141845703125, 3.77716064453125, 2.449859619140625, 10.841278076171875, 14.504497528076172, 1.2589950561523438, -2.5049591064453125, 10.20309066772461, 6.004550933837891, 7.478797912597656, 5.159088134765625, 6.916412353515625, 2.002960205078125, -3.54339599609375, 29.611236572265625, 14.4971923828125, 5.6086578369140625, -16.79888916015625, -20.566253662109375, 9.911741256713867, 0.9138641357421875, 5.7880096435546875, -5.181640625, 8.23052978515625, 4.17474365234375, 35.16748046875, -6.9940643310546875, 9.7994384765625, 0.15288543701171875, 11.50408935546875, -3.51153564453125, 0.37451171875, 6.295387268066406, -1.6668701171875, -5.027374267578125, -6.2569580078125, 6.7747039794921875, -9.15313720703125, -0.9908447265625, 2.011932373046875, 0.5241680145263672, -13.648445129394531, 1.2186012268066406, 6.56939697265625, 10.73349380493164], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000101.npy"} {"epoch": 0.21151832460732983, "step": 102, "batch_size": 128, "mean": 6.402046203613281, "std": 9.570538520812988, "min": -18.341888427734375, "p10": -5.594486999511719, "median": 5.897279739379883, "p90": 18.24680938720703, "max": 33.395904541015625, "pos_frac": 0.734375, "sample": [-5.6373291015625, -1.021820068359375, 13.408103942871094, 6.235908508300781, 2.1728668212890625, 18.00262451171875, 8.970367431640625, -0.07732391357421875, 3.261962890625, 13.867706298828125, 5.8050994873046875, 5.989459991455078, 11.707489013671875, 8.790115356445312, -0.5787811279296875, 16.159637451171875, 9.98162841796875, -10.358123779296875, -2.5652618408203125, -0.31026458740234375, -3.6935272216796875, 11.912567138671875, 12.851226806640625, 24.05804443359375, 16.775711059570312, -3.277130126953125, 6.032688140869141, -8.32672119140625, 5.169731140136719, -8.550048828125, -7.009246826171875, 6.5159759521484375, 3.433380126953125, 7.5216522216796875, -5.5638427734375, 21.52301025390625, 15.808258056640625, -1.1636428833007812, 0.933746337890625, 0.6938838958740234, -1.44610595703125, 0.5962600708007812, -2.059934616088867, 16.701400756835938, 6.947052001953125, 5.657440185546875, 3.2943878173828125, -6.109161376953125, -9.15478515625, -6.36871337890625, 17.6495361328125, -5.0108795166015625, 5.120475769042969, 6.9912109375, 5.154876708984375, 16.376983642578125, 7.2354888916015625, 33.395904541015625, 10.00830078125, 16.07354736328125, 1.4888687133789062, -0.836669921875, 17.455413818359375, 10.558746337890625, 9.01654052734375, -5.5761260986328125, -9.823028564453125, 1.488128662109375, -3.945690155029297, 5.385337829589844, 11.47299575805664, 16.68890380859375, 22.347618103027344, 15.91925048828125, 5.702388763427734, 18.795608520507812, 18.327808380126953, 4.815460205078125, 8.33575439453125, -14.61944580078125, 0.5534267425537109, 9.80548095703125, 0.0, 6.063629150390625, -1.2982330322265625, 18.49798583984375, 0.0, 6.83905029296875, 14.6842041015625, 0.19618606567382812, 12.8731689453125, 12.9205322265625, -1.1838836669921875, 3.523345947265625, 16.114898681640625, 5.766265869140625, 12.915817260742188, 1.3124008178710938, 10.4171142578125, 15.166221618652344, 18.3204345703125, 5.484291076660156, 2.409038543701172, 25.0880126953125, 6.66094970703125, 9.641708374023438, 2.5678882598876953, 18.608116149902344, 0.7996292114257812, 18.215255737304688, -18.341888427734375, 10.34771728515625, -2.2099761962890625, 3.1216812133789062, 12.611343383789062, -14.45428466796875, 15.728950500488281, 2.611419677734375, 31.44598388671875, 2.7259979248046875, 18.495208740234375, 4.0264892578125, 13.079635620117188, 9.674957275390625, 10.511886596679688, 27.1029052734375, -1.183380126953125, -12.296600341796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000102.npy"} {"epoch": 0.2136125654450262, "step": 103, "batch_size": 128, "mean": 4.79279899597168, "std": 10.719864845275879, "min": -23.0523681640625, "p10": -6.471695709228513, "median": 2.4814987182617188, "p90": 19.000753784179686, "max": 46.87242126464844, "pos_frac": 0.7109375, "sample": [-20.78582763671875, -8.547027587890625, 8.107818603515625, 9.50299072265625, 18.2734375, 15.50925064086914, -0.30681610107421875, -2.821502685546875, 23.2197265625, 1.0806884765625, 0.30841064453125, 16.31097412109375, 19.921295166015625, 3.265655517578125, 0.0, -1.653045654296875, -12.717315673828125, -11.239608764648438, -3.001251220703125, 8.637786865234375, -0.6570329666137695, 1.7245864868164062, 10.08929443359375, -23.0523681640625, 2.1691970825195312, 3.49462890625, 1.019287109375, 1.3478546142578125, 6.2415924072265625, 0.8784561157226562, 11.445465087890625, 9.930580139160156, 0.19199371337890625, 5.713897705078125, -4.51593017578125, 7.108173370361328, -2.253662109375, 16.293373107910156, 8.704803466796875, 7.785621643066406, -1.7087154388427734, 14.767547607421875, 1.15478515625, 31.3055419921875, -1.5393524169921875, 12.354949951171875, 2.7665328979492188, 10.09832763671875, 13.20599365234375, 46.87242126464844, 18.788818359375, 13.00701904296875, 9.050628662109375, -13.08123779296875, 2.3368682861328125, 0.146270751953125, -10.544967651367188, 8.716033935546875, 9.24664306640625, 10.1641845703125, 16.772186279296875, 7.372283935546875, 0.8259658813476562, 11.61761474609375, 4.489448547363281, 20.55438232421875, 10.822052001953125, 11.236831665039062, 3.7920608520507812, 7.2919464111328125, 1.6346511840820312, 0.920806884765625, -0.031341552734375, 2.2626953125, -0.4661865234375, 0.6847171783447266, -8.908132553100586, -17.17755126953125, 2.626129150390625, 2.6295623779296875, 22.813140869140625, 13.839317321777344, 0.45355796813964844, 13.29437255859375, -1.102783203125, 8.553009033203125, 5.838478088378906, 1.5309677124023438, 27.988861083984375, -1.26727294921875, 1.9161605834960938, 1.46783447265625, 15.979248046875, 6.2526092529296875, -3.1686325073242188, 1.0867881774902344, 23.137847900390625, -3.1171875, -4.7595062255859375, -13.22540283203125, 0.0, 3.3904876708984375, 3.6190757751464844, 17.646148681640625, -2.0047607421875, 20.9970703125, -11.60931396484375, 2.111572265625, 1.481658935546875, 12.083545684814453, 12.812118530273438, -4.635948181152344, -3.5293350219726562, 19.495269775390625, 0.7443695068359375, 0.09654808044433594, 21.5194091796875, 24.42181396484375, 11.298736572265625, 19.59698486328125, -2.386932373046875, 6.596244812011719, -1.6884765625, -15.031005859375, 1.070404052734375, -19.896331787109375, -5.582267761230469, 0.5659332275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000103.npy"} {"epoch": 0.2157068062827225, "step": 104, "batch_size": 128, "mean": 4.355037689208984, "std": 11.487180709838867, "min": -26.039398193359375, "p10": -8.385761260986328, "median": 3.1405277252197266, "p90": 19.755438232421874, "max": 48.24786376953125, "pos_frac": 0.6640625, "sample": [1.71209716796875, -4.06256103515625, 7.9064178466796875, 15.369384765625, -10.61224365234375, 4.827156066894531, 4.1814422607421875, -0.48587989807128906, 19.70001220703125, -8.364166259765625, 16.556427001953125, -7.682262420654297, 3.319366455078125, -17.222686767578125, 8.392181396484375, 7.909122467041016, 27.573867797851562, 21.654464721679688, -1.90899658203125, -3.5401077270507812, 2.62615966796875, -2.664102554321289, 8.274337768554688, -22.15570068359375, 25.136566162109375, -3.522796630859375, 3.341827392578125, -1.9486980438232422, 2.694742202758789, 24.9482421875, 6.919242858886719, 15.32318115234375, -12.28863525390625, -0.06085205078125, 2.757843017578125, -8.436149597167969, 0.6985435485839844, -8.019660949707031, 23.327911376953125, -3.215118408203125, 12.399505615234375, 0.16357421875, 1.0768661499023438, -11.677337646484375, -10.2413330078125, 11.218719482421875, -0.1204833984375, 7.9360198974609375, -1.749176025390625, -3.4954833984375, -4.470367431640625, 9.012344360351562, 9.920028686523438, 3.155670166015625, 24.718505859375, 5.632938385009766, 19.884765625, -5.8319244384765625, -2.623809814453125, 2.2732791900634766, 5.231767654418945, 9.977531433105469, 5.15069580078125, 6.844629287719727, 6.433624267578125, 3.125385284423828, -1.5148086547851562, 14.452667236328125, 4.630622863769531, -14.628005981445312, 10.05926513671875, 18.131561279296875, 0.3816986083984375, -26.039398193359375, 48.24786376953125, 5.404747009277344, 23.88677978515625, 9.637664794921875, 11.617874145507812, 22.666534423828125, 6.2279205322265625, -25.11492919921875, -10.53125, 11.907882690429688, 2.61041259765625, 0.32561492919921875, 0.9866600036621094, 7.145477294921875, -6.55157470703125, 18.643035888671875, 23.82171630859375, 0.7997283935546875, 4.211181640625, 8.50341796875, 1.1138763427734375, -0.0080413818359375, -6.83880615234375, 1.0314960479736328, 25.099273681640625, 11.89129638671875, 14.524017333984375, 28.723114013671875, -5.4847259521484375, 15.669830322265625, -18.474365234375, 1.019195556640625, -6.203369140625, 2.6531448364257812, 4.488258361816406, -13.8419189453125, 0.0, 10.179519653320312, -0.7333908081054688, 13.443229675292969, 3.5601959228515625, 6.4661865234375, 11.329559326171875, -6.426460266113281, 8.201644897460938, 0.43731689453125, -0.84185791015625, 16.968719482421875, 2.8623294830322266, 6.6456451416015625, 17.77203369140625, -2.443206787109375, -0.28460693359375, 2.1194610595703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000104.npy"} {"epoch": 0.21780104712041884, "step": 105, "batch_size": 128, "mean": 8.132970809936523, "std": 11.18293285369873, "min": -23.019989013671875, "p10": -5.014015197753905, "median": 7.240753173828125, "p90": 22.58988037109375, "max": 35.3359375, "pos_frac": 0.796875, "sample": [5.66314697265625, 3.45855712890625, 14.804061889648438, 1.3554096221923828, -7.95068359375, 17.700531005859375, 9.446868896484375, 15.2822265625, 11.268524169921875, 2.380889892578125, 16.737030029296875, 4.72198486328125, 21.20654296875, 19.2191162109375, -10.939910888671875, 22.993927001953125, 6.3804931640625, -1.10943603515625, 27.236679077148438, 8.006134033203125, -0.9013099670410156, 0.506011962890625, 7.9615325927734375, 7.291908264160156, 23.565093994140625, 29.21368408203125, -20.530181884765625, 0.6982269287109375, -3.563690185546875, 30.81353759765625, 5.6372833251953125, 14.451278686523438, 0.5644454956054688, 17.9075927734375, -1.24090576171875, 19.9202880859375, 22.984893798828125, 4.7569122314453125, 20.37200927734375, -11.262481689453125, 2.9200897216796875, 22.493896484375, 16.307373046875, 17.496673583984375, 2.5178375244140625, 17.175018310546875, 8.883392333984375, -2.251861572265625, 21.208251953125, 22.195404052734375, 3.911956787109375, 13.223907470703125, -0.157196044921875, 14.183502197265625, -1.8609771728515625, 12.60430908203125, 10.487457275390625, 22.8232421875, 8.430671691894531, -12.94677734375, 2.559417724609375, -10.51910400390625, 0.069091796875, 3.6169586181640625, 6.698760986328125, 8.369415283203125, 14.547882080078125, -1.9805908203125, 2.4475231170654297, 2.546295166015625, 19.90606689453125, -19.80126953125, 16.81926727294922, 21.0384521484375, 7.19024658203125, -4.875152587890625, 12.84228515625, 17.65008544921875, 25.84405517578125, 2.4225616455078125, 7.291259765625, 2.4690399169921875, 7.643341064453125, 0.8661041259765625, 5.44775390625, 19.84600830078125, -12.3687744140625, 22.8138427734375, 7.103790283203125, -23.019989013671875, 3.107128143310547, 10.54278564453125, 25.275650024414062, -10.8265380859375, 6.555015563964844, 2.5750579833984375, 9.460508346557617, -5.7604217529296875, 11.94537353515625, 0.0, 4.578142166137695, 8.266937255859375, 0.5450210571289062, -0.8504638671875, 14.830787658691406, -2.7121429443359375, 33.284912109375, 26.7445068359375, 6.4775390625, 5.06146240234375, 7.4132232666015625, 21.540618896484375, 0.2329559326171875, 4.712066650390625, 18.501922607421875, 13.25103759765625, -5.3380279541015625, 11.19671630859375, 2.9286346435546875, -7.694244384765625, 8.162818908691406, 20.9449462890625, 4.664886474609375, 5.081584930419922, -0.2252197265625, 35.3359375, 19.259132385253906, 9.8109130859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000105.npy"} {"epoch": 0.2198952879581152, "step": 106, "batch_size": 128, "mean": 5.464231491088867, "std": 11.705928802490234, "min": -21.980300903320312, "p10": -8.539123535156248, "median": 4.635364532470703, "p90": 20.955355834960937, "max": 36.761077880859375, "pos_frac": 0.671875, "sample": [-6.317634582519531, -15.54058837890625, 15.5123291015625, -3.101104736328125, -3.598358154296875, 22.630859375, -6.883209228515625, 7.333953857421875, -14.39373779296875, 15.490478515625, -13.591583251953125, 0.6594009399414062, 4.6684722900390625, 19.373321533203125, 13.19287109375, 31.755615234375, 26.605987548828125, -8.01654052734375, -4.2133636474609375, -2.0211334228515625, 13.675247192382812, 2.995452880859375, 3.1367034912109375, 4.328178405761719, 12.919464111328125, 1.158447265625, 3.4895286560058594, 7.062202453613281, 12.187454223632812, 24.374191284179688, 1.191009521484375, 9.376144409179688, 17.643524169921875, -0.3118133544921875, 0.3404064178466797, -6.284332275390625, 11.47406005859375, 2.73095703125, 7.0017852783203125, 7.484527587890625, 20.780914306640625, 1.1031341552734375, 7.978607177734375, -16.463348388671875, -2.4434051513671875, 14.96466064453125, -7.7025604248046875, -0.129241943359375, 12.679611206054688, 5.0038604736328125, 6.011474609375, 8.478973388671875, -4.928323745727539, 14.784332275390625, -17.64703369140625, -21.980300903320312, 11.4859619140625, 30.5262451171875, 28.743194580078125, 20.806549072265625, 5.49591064453125, 8.7742919921875, -9.75848388671875, -6.946807861328125, 9.93499755859375, -0.2293853759765625, -5.695831298828125, -4.5885009765625, 23.778564453125, -13.345138549804688, 13.220687866210938, 6.3270263671875, -0.364837646484375, 5.759033203125, -12.03717041015625, 1.3676815032958984, -3.3405914306640625, 2.4624481201171875, 17.82208251953125, 2.2397613525390625, -0.092864990234375, 36.761077880859375, 7.924224853515625, 1.9508743286132812, -4.810211181640625, 2.2586669921875, 20.73907470703125, -0.10136795043945312, 14.759613037109375, -10.154144287109375, 10.237167358398438, 3.7630462646484375, 17.4429931640625, -1.2345428466796875, 1.18743896484375, 4.602256774902344, 13.767364501953125, 19.642059326171875, 5.738128662109375, 8.554718017578125, 28.85009765625, -17.065521240234375, 1.390167236328125, -2.5790863037109375, 6.7748565673828125, 12.85189437866211, 14.352066040039062, -12.387771606445312, 3.59259033203125, -5.849517822265625, -10.087677001953125, -4.7508544921875, 32.06597900390625, 7.782199859619141, -7.497161865234375, 5.8712005615234375, 12.173812866210938, -5.781898498535156, 20.843643188476562, 2.880859375, 21.6192626953125, 7.7256011962890625, 9.473739624023438, 23.479400634765625, 21.216018676757812, -0.1077880859375, 9.977813720703125, 1.223876953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000106.npy"} {"epoch": 0.22198952879581152, "step": 107, "batch_size": 128, "mean": 7.293955326080322, "std": 12.344758033752441, "min": -56.04638671875, "p10": -4.936019134521484, "median": 5.770164489746094, "p90": 23.622132873535154, "max": 33.523529052734375, "pos_frac": 0.734375, "sample": [0.9927825927734375, -4.00927734375, 4.5070953369140625, 5.1334381103515625, 9.996795654296875, -3.91387939453125, 14.659637451171875, 25.737335205078125, 6.212158203125, 5.75189208984375, -4.808685302734375, 4.6872406005859375, 1.225738525390625, 12.725982666015625, 12.799541473388672, -3.41534423828125, 13.90155029296875, -6.214874267578125, 11.44256591796875, 19.19134521484375, -2.59454345703125, 22.679733276367188, 7.992820739746094, 14.30255126953125, -11.7447509765625, 27.99090576171875, 7.81109619140625, -13.214874267578125, 18.4266357421875, -6.765724182128906, 18.7630615234375, 1.1742630004882812, 1.43701171875, 17.718055725097656, 20.391998291015625, 23.285369873046875, 11.31536865234375, 6.214637756347656, 31.8529052734375, 0.6365966796875, 27.339614868164062, 8.073486328125, 6.0208740234375, 12.428657531738281, 2.9522247314453125, 1.1416091918945312, -3.315155029296875, 5.7884368896484375, 13.927799224853516, 6.833343505859375, 8.048809051513672, -1.0703964233398438, 3.2533340454101562, 3.6267852783203125, 22.356689453125, -5.233131408691406, 4.1380615234375, 0.8484840393066406, 22.02099609375, 10.401763916015625, 5.074676513671875, 3.7625579833984375, -3.6375732421875, 33.2550048828125, -0.483184814453125, -6.661865234375, 4.371543884277344, 31.183319091796875, 22.822479248046875, 17.49981689453125, -12.952392578125, 24.407913208007812, -4.669158935546875, 14.6751708984375, 20.6649169921875, 1.9029731750488281, 2.196685791015625, 19.261932373046875, 12.631256103515625, 12.564952850341797, 28.839385986328125, 15.077972412109375, -3.00048828125, -12.305831909179688, -0.7291259765625, 4.1048126220703125, 6.689781188964844, -1.7754592895507812, 33.523529052734375, -7.8096923828125, 28.919342041015625, 12.055145263671875, -1.115509033203125, 5.79376220703125, 5.96893310546875, 1.6118831634521484, 16.202789306640625, 2.38677978515625, -0.4365997314453125, -0.008640289306640625, -56.04638671875, -10.126861572265625, 9.03009033203125, 26.855438232421875, 18.474693298339844, -2.9888839721679688, 25.82550048828125, -0.357574462890625, 7.9028472900390625, 1.1275386810302734, 19.752761840820312, -0.9730224609375, 10.339035034179688, 3.49554443359375, 5.60540771484375, 17.465576171875, -3.4029083251953125, -2.21685791015625, -6.70947265625, 0.08282470703125, 2.05267333984375, 8.677490234375, -5.900146484375, 7.054229736328125, 0.08991622924804688, 31.04638671875, 5.142360687255859, 16.6058349609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000107.npy"} {"epoch": 0.22408376963350785, "step": 108, "batch_size": 128, "mean": 8.16340160369873, "std": 10.975932121276855, "min": -19.15081787109375, "p10": -5.26512451171875, "median": 7.201179504394531, "p90": 24.67862548828125, "max": 33.863372802734375, "pos_frac": 0.765625, "sample": [24.738494873046875, 15.790985107421875, -1.9487457275390625, 7.0885772705078125, 13.216964721679688, -7.15118408203125, 18.615997314453125, 8.92593002319336, 3.4745407104492188, 13.240081787109375, 1.862762451171875, -4.93414306640625, -5.1998291015625, 10.938827514648438, 4.809272766113281, -7.717796325683594, 27.743682861328125, 5.121551513671875, -2.145343780517578, 9.2020263671875, 13.3837890625, 5.9620208740234375, 10.682449340820312, -6.7362060546875, -0.464447021484375, 9.691131591796875, 26.5035400390625, -2.6241455078125, 0.0, -19.15081787109375, 14.237136840820312, 3.031585693359375, 9.749847412109375, 5.199928283691406, 16.6417236328125, 1.975555419921875, 26.3829345703125, 7.31378173828125, 17.373260498046875, -0.3977699279785156, 3.5619373321533203, 10.347320556640625, 3.122772216796875, -5.638824462890625, 6.3985595703125, 28.007827758789062, 11.325439453125, -15.103515625, 4.179931640625, 24.72613525390625, 14.888286590576172, 4.9613037109375, 8.48052978515625, 33.863372802734375, 5.682674407958984, 1.6105499267578125, 0.0, -5.41748046875, 1.5073070526123047, 0.13679122924804688, 2.3014888763427734, 22.50628662109375, 14.500152587890625, 9.861404418945312, -18.484375, 9.5499267578125, 21.128173828125, 11.956615447998047, -1.025665283203125, 10.48974609375, 9.333465576171875, 4.9606170654296875, -1.250579833984375, 1.98486328125, -4.407890319824219, 24.65826416015625, 19.77288818359375, 1.8411483764648438, 12.71075439453125, 17.772735595703125, 11.61212158203125, -6.1277313232421875, 28.910888671875, 12.073577880859375, 4.49542236328125, -0.0761566162109375, 22.8446044921875, 10.776313781738281, 9.174270629882812, 29.419036865234375, 6.923553466796875, 19.822265625, 3.0237884521484375, 10.946456909179688, 8.884361267089844, -2.23590087890625, 33.0487060546875, 15.697906494140625, 27.41656494140625, 20.3592529296875, 15.91879653930664, 5.209991455078125, 7.952911376953125, 0.0, -0.6257171630859375, 8.920631408691406, 5.946083068847656, 32.642181396484375, 4.888057708740234, 24.27642822265625, 7.757598876953125, -9.50372314453125, 4.664497375488281, 0.6029739379882812, 16.564437866210938, 28.42236328125, 22.36981201171875, 1.858612060546875, 2.8649063110351562, 10.877815246582031, -11.202423095703125, -11.127830505371094, -10.620391845703125, 6.9611968994140625, -0.026275634765625, 9.1900634765625, 4.204925537109375, 17.70928955078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000108.npy"} {"epoch": 0.2261780104712042, "step": 109, "batch_size": 128, "mean": 8.47254467010498, "std": 13.762791633605957, "min": -29.88543701171875, "p10": -5.919546508789062, "median": 6.705966949462891, "p90": 23.919512939453124, "max": 60.02783203125, "pos_frac": 0.734375, "sample": [3.726318359375, 12.0447998046875, -10.307563781738281, 23.44268798828125, 5.4772796630859375, 8.131202697753906, 14.731979370117188, 17.540008544921875, -9.944778442382812, 6.589111328125, 4.6783447265625, 19.88232421875, -0.015735626220703125, -3.28057861328125, -7.0507659912109375, -4.1650390625, -1.1781463623046875, 30.98785400390625, 41.23602294921875, 1.7138595581054688, 0.9609603881835938, -0.1181640625, -4.1543426513671875, -14.56976318359375, -7.167573928833008, 11.739608764648438, 0.296356201171875, -0.6447792053222656, -0.3105888366699219, 9.968505859375, 6.9119720458984375, -0.6204986572265625, 16.099334716796875, 35.970367431640625, 26.08587646484375, 10.220989227294922, 12.72772216796875, 19.532684326171875, 1.5959548950195312, 22.439178466796875, 12.5191650390625, 18.771331787109375, 5.5639801025390625, -0.3983783721923828, 15.15460205078125, 33.032684326171875, 0.136627197265625, 30.81585693359375, 6.822822570800781, 6.465538024902344, 13.015472412109375, 6.318756103515625, 6.170654296875, 4.1334228515625, -21.137298583984375, 7.33544921875, 7.914520263671875, 7.175821304321289, 12.783363342285156, 7.1820068359375, 16.95855712890625, 3.6808700561523438, -0.0328369140625, 21.394561767578125, 8.0545654296875, 10.940376281738281, -5.64013671875, 11.092742919921875, 9.8642578125, 12.040245056152344, 16.869049072265625, 3.349334716796875, -7.4620819091796875, 18.256591796875, 10.688472747802734, 22.9002685546875, 48.677978515625, -2.1956787109375, 1.1328659057617188, 29.24835205078125, 4.07916259765625, 5.5121002197265625, 18.993019104003906, 13.867347717285156, 0.424468994140625, 0.7567214965820312, 13.199310302734375, 0.6581077575683594, -9.262115478515625, 43.58642578125, 23.71533203125, 18.995834350585938, 6.449504852294922, 60.02783203125, 8.934883117675781, -2.14630126953125, 1.6314010620117188, 37.375732421875, -10.980239868164062, 4.4662017822265625, -5.70404052734375, 18.639205932617188, 0.0, 2.1511764526367188, 30.602447509765625, 9.266311645507812, 14.082077026367188, 9.065750122070312, -1.2641677856445312, 10.036895751953125, -1.51556396484375, 6.260566711425781, 1.6979522705078125, -2.1939544677734375, 21.068267822265625, -1.84576416015625, 12.862937927246094, 22.863037109375, -10.70916748046875, -29.88543701171875, -26.28728485107422, 24.39593505859375, 5.35125732421875, 17.088417053222656, 4.511962890625, -6.422393798828125, -5.460693359375, 14.747512817382812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000109.npy"} {"epoch": 0.22827225130890053, "step": 110, "batch_size": 128, "mean": 7.956274032592773, "std": 13.483417510986328, "min": -38.228485107421875, "p10": -6.348573303222656, "median": 5.215660095214844, "p90": 26.12867431640625, "max": 42.8319091796875, "pos_frac": 0.7109375, "sample": [-0.5397872924804688, -38.228485107421875, -11.82601547241211, 29.385986328125, -2.1804733276367188, 1.533376693725586, -2.8269424438476562, 2.44140625, -0.66143798828125, 27.680877685546875, 3.858642578125, 4.391357421875, 3.252765655517578, 22.986907958984375, -5.103061676025391, -0.43743133544921875, 12.654052734375, 4.502235412597656, 16.418670654296875, 2.9805908203125, -9.028419494628906, 18.552093505859375, 2.3318939208984375, 23.716522216796875, 8.11846923828125, 3.589691162109375, -6.161109924316406, 13.579730987548828, 0.815673828125, -4.541778564453125, 28.87908935546875, 13.393096923828125, 27.513671875, 2.3653697967529297, 35.28826904296875, -3.07208251953125, 17.339096069335938, 17.881103515625, 10.46514892578125, -2.868743896484375, 5.698638916015625, 5.4117431640625, 11.564071655273438, 26.1163330078125, -10.6439208984375, -0.00099945068359375, 15.688201904296875, -6.9627685546875, 10.460639953613281, -0.1002349853515625, 21.480674743652344, -21.663589477539062, 11.92950439453125, 1.18511962890625, 32.784942626953125, 12.348602294921875, 4.139984130859375, -30.650955200195312, 16.1361083984375, 19.340576171875, 3.9151840209960938, 6.7842254638671875, 2.8092422485351562, -0.783538818359375, 1.878509521484375, 27.00689697265625, -6.785987854003906, 25.07861328125, 18.86553955078125, -14.4637451171875, -0.013515472412109375, 18.58740234375, 5.0195770263671875, 2.4558639526367188, 12.043540954589844, 7.10308837890625, 0.21809005737304688, -3.0470352172851562, -1.0660858154296875, 11.230850219726562, 0.0, 5.006103515625, 23.525390625, 14.039474487304688, -10.517776489257812, 14.673553466796875, 0.4642353057861328, -7.1905364990234375, 6.749176025390625, 26.2821044921875, 15.012252807617188, 38.641387939453125, 14.927520751953125, 24.557876586914062, 4.856407165527344, 19.87603759765625, 24.83978271484375, -2.660888671875, -2.426982879638672, 22.04541015625, 26.157470703125, 7.0211944580078125, 1.2839279174804688, 23.580474853515625, -16.860130310058594, 8.491455078125, 4.196968078613281, -5.1627197265625, 24.82110595703125, 20.179641723632812, -1.5250167846679688, 29.147964477539062, -2.9070510864257812, 8.570831298828125, 12.0245361328125, 28.23089599609375, 42.8319091796875, -3.1342086791992188, 19.060272216796875, 22.083770751953125, 1.08843994140625, 19.03155517578125, 5.0128936767578125, 7.587726593017578, -13.609619140625, 2.733062744140625, -0.43438720703125, 6.66015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000110.npy"} {"epoch": 0.23036649214659685, "step": 111, "batch_size": 128, "mean": 8.718975067138672, "std": 13.70277214050293, "min": -22.695068359375, "p10": -7.663871002197265, "median": 7.7945404052734375, "p90": 25.94262237548828, "max": 48.19390869140625, "pos_frac": 0.75, "sample": [4.3208465576171875, -11.43328857421875, 12.6552734375, -22.695068359375, 48.19390869140625, -7.402137756347656, 11.375701904296875, 26.7413330078125, 0.5687103271484375, 0.3716888427734375, 25.66448974609375, 5.82135009765625, 6.8603057861328125, 5.899749755859375, 5.319488525390625, 15.049140930175781, 7.798248291015625, 34.357391357421875, -17.65155029296875, -13.401458740234375, 11.024093627929688, 5.8572845458984375, 14.0206298828125, 4.119873046875, 15.871429443359375, 1.613189697265625, 4.71173095703125, 25.940658569335938, 1.2780227661132812, 11.37347412109375, 16.59217071533203, 21.7113037109375, -3.0457115173339844, 34.8841552734375, -4.35296630859375, 6.3571014404296875, -5.62066650390625, 17.515869140625, 0.1814117431640625, 11.420211791992188, 1.5555877685546875, 10.734664916992188, -8.145072937011719, 21.609954833984375, 13.451873779296875, 1.43414306640625, 17.302764892578125, -0.983306884765625, 31.127197265625, 7.904548645019531, 2.3270263671875, 47.3836669921875, -12.011322021484375, -2.933685302734375, -0.75067138671875, 17.9632568359375, 12.93963623046875, 22.31536865234375, 4.6818084716796875, 31.04254150390625, 4.96136474609375, 11.504196166992188, 5.329450607299805, 10.276336669921875, 2.6053314208984375, -7.131134033203125, 3.2674102783203125, 13.610641479492188, 40.427978515625, 11.16015625, 6.497161865234375, -1.36151123046875, 8.40802001953125, -1.0240631103515625, 5.2017822265625, -0.1706981658935547, -1.18115234375, 25.94720458984375, 10.444259643554688, -17.1168212890625, 1.4264907836914062, 10.0911865234375, -7.4576416015625, 3.059173583984375, -2.764984130859375, -5.99896240234375, 0.2533683776855469, 1.88067626953125, 17.812828063964844, 27.64306640625, 16.366119384765625, 32.08807373046875, 17.8426513671875, -1.9869861602783203, 15.486862182617188, 10.4444580078125, 16.210987091064453, -3.172138214111328, 43.790252685546875, 3.7774734497070312, 20.40380096435547, 9.720664978027344, -16.419036865234375, -9.9486083984375, -1.614349365234375, 25.293975830078125, -12.0826416015625, 20.193603515625, -1.880584716796875, 20.218246459960938, 15.93255615234375, 8.532073974609375, 15.7899169921875, 37.34112548828125, 21.976531982421875, 17.61016845703125, 24.18756103515625, -16.711456298828125, 13.798980712890625, 14.411304473876953, 8.1341552734375, -10.732818603515625, 13.864776611328125, 7.79083251953125, 19.60986328125, 2.5196609497070312, -11.164077758789062, 1.986419677734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000111.npy"} {"epoch": 0.2324607329842932, "step": 112, "batch_size": 128, "mean": 4.334019660949707, "std": 15.298911094665527, "min": -49.02935791015625, "p10": -12.773721313476562, "median": 3.518341064453125, "p90": 25.167794799804685, "max": 46.97589111328125, "pos_frac": 0.609375, "sample": [15.228073120117188, 3.27972412109375, -4.0508270263671875, 10.7125244140625, 0.6826705932617188, 8.991546630859375, -0.6410903930664062, -30.3167724609375, -4.216670989990234, 3.21734619140625, -1.6844463348388672, 6.706451416015625, -10.98455810546875, 19.148483276367188, -1.85247802734375, 3.6450347900390625, 15.38079833984375, 15.074600219726562, 6.111961364746094, 24.88067626953125, -8.340179443359375, 20.73004150390625, 12.789901733398438, 13.9268798828125, -49.02935791015625, -14.32843017578125, -9.825004577636719, 13.095870971679688, 9.093231201171875, 8.527183532714844, -9.792320251464844, 4.844551086425781, -26.477294921875, 6.015228271484375, 4.480949401855469, -18.9716796875, 4.6265869140625, 5.661495208740234, -0.2561492919921875, 3.9605865478515625, 16.478515625, -3.037109375, 6.3626708984375, 14.309341430664062, 12.922805786132812, -1.57598876953125, 14.936279296875, -0.0566253662109375, 16.570327758789062, -6.648223876953125, 9.562515258789062, 5.1157684326171875, 0.9139480590820312, 15.1700439453125, -17.18450927734375, -3.173412322998047, -14.337371826171875, -10.0645751953125, -35.87762451171875, -8.928375244140625, -1.8940811157226562, -0.7006454467773438, 17.802520751953125, 36.723907470703125, -4.97021484375, 3.3916473388671875, 30.505828857421875, 15.381378173828125, 0.661346435546875, 7.271324157714844, 7.469390869140625, -5.6643524169921875, 15.849151611328125, 15.02044677734375, 8.410781860351562, 37.102203369140625, -4.326042175292969, 0.071929931640625, 29.7012939453125, -17.85979461669922, -2.38946533203125, 10.309326171875, 0.63287353515625, -1.41094970703125, 20.981689453125, 24.5384521484375, 11.228408813476562, -1.693511962890625, -5.51727294921875, 1.93450927734375, 25.837738037109375, -12.46966552734375, 29.67022705078125, 4.404792785644531, -5.246795654296875, 16.958206176757812, -1.9534492492675781, 0.7476863861083984, -37.085357666015625, -6.858154296875, -8.53509521484375, 27.36334228515625, -2.553985595703125, -0.60772705078125, 27.256149291992188, 16.3612060546875, -13.483184814453125, 27.092376708984375, 5.645538330078125, 17.1485595703125, 34.396240234375, -9.9537353515625, 2.7591552734375, 27.193084716796875, 5.578575134277344, 1.7357940673828125, -17.99951171875, -2.2096405029296875, 0.1601104736328125, 1.351409912109375, 8.6470947265625, -4.212249755859375, 24.79730224609375, 27.666641235351562, 46.97589111328125, -17.765472412109375, -4.070426940917969, 9.946258544921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000112.npy"} {"epoch": 0.23455497382198953, "step": 113, "batch_size": 128, "mean": 10.092288970947266, "std": 15.011159896850586, "min": -28.6304931640625, "p10": -7.426852416992187, "median": 8.656940460205078, "p90": 31.548735046386717, "max": 57.0379638671875, "pos_frac": 0.7421875, "sample": [5.1936492919921875, 16.728492736816406, 9.7562255859375, 0.402679443359375, 14.8330078125, 1.0520248413085938, -7.0972900390625, 15.023773193359375, 25.987701416015625, 5.622772216796875, 6.9373321533203125, 7.3965301513671875, 19.7376708984375, -11.330703735351562, -1.1132965087890625, 3.241628646850586, 3.734102249145508, 1.8059616088867188, 5.340057373046875, 12.835588455200195, 57.0379638671875, -3.70611572265625, -2.7239303588867188, 41.524200439453125, 12.96173095703125, -28.6304931640625, 4.790191650390625, 23.87005615234375, 12.4564208984375, -2.7613983154296875, -2.235015869140625, -11.98675537109375, 23.72662353515625, 32.01953125, 41.25189208984375, -2.019775390625, 15.16534423828125, -6.0114593505859375, 28.495574951171875, -11.4049072265625, 13.2978515625, 31.584548950195312, -0.194488525390625, 14.04608154296875, 10.1046142578125, 4.494762420654297, -12.02703857421875, 0.7901458740234375, 0.4847602844238281, 31.53338623046875, 4.3480377197265625, 17.948585510253906, -26.538665771484375, 26.6192626953125, 32.55963134765625, 24.637130737304688, 11.446739196777344, 32.38543701171875, 5.877471923828125, -11.705917358398438, 8.780342102050781, 0.9302825927734375, -4.660530090332031, 24.8720703125, 0.039257049560546875, 8.533538818359375, 6.140270233154297, 6.331207275390625, -15.9918212890625, 17.667999267578125, 6.13909912109375, 31.80419921875, -1.7229690551757812, 11.734512329101562, -9.134857177734375, 4.490333557128906, 13.337066650390625, 31.031219482421875, 15.297027587890625, 15.650222778320312, 14.835365295410156, -11.298797607421875, 17.580902099609375, 28.03118896484375, 36.58270263671875, 13.405677795410156, 13.356246948242188, -3.8755569458007812, -3.73175048828125, 6.591461181640625, 35.97210693359375, 25.88372039794922, 20.553741455078125, 15.911895751953125, -6.5902099609375, 25.3487548828125, -0.4801483154296875, 13.33489990234375, 15.519439697265625, 7.792236328125, 5.144828796386719, -0.3924560546875, 10.982940673828125, 3.5821762084960938, 23.8370361328125, 39.1466064453125, 23.860504150390625, 39.496612548828125, -9.92767333984375, 3.30670166015625, 22.0841064453125, 17.46826171875, 12.80279541015625, 42.80560302734375, -5.3288421630859375, -8.195831298828125, 2.753662109375, 3.3293685913085938, 20.293548583984375, 16.991012573242188, -0.4056396484375, 16.546661376953125, -17.839248657226562, 10.741943359375, 3.0676422119140625, -3.6655120849609375, 19.845489501953125, -6.111606597900391], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000113.npy"} {"epoch": 0.23664921465968586, "step": 114, "batch_size": 128, "mean": 5.136881351470947, "std": 13.791298866271973, "min": -26.916259765625, "p10": -12.616398620605468, "median": 4.438137054443359, "p90": 23.717074584960937, "max": 42.72991943359375, "pos_frac": 0.6875, "sample": [3.9625587463378906, 21.497894287109375, 3.8646087646484375, 26.289714813232422, 0.20085906982421875, 6.13262939453125, 2.1465492248535156, 5.263759613037109, -0.5921859741210938, -5.90679931640625, -3.3261260986328125, 22.74676513671875, -0.85382080078125, 28.252227783203125, 15.940505981445312, 37.70806884765625, -11.893726348876953, 0.75732421875, -19.91693115234375, 1.9620208740234375, 1.411346435546875, -15.774871826171875, 5.98822021484375, 14.842323303222656, 2.0492782592773438, 15.612152099609375, 9.72027587890625, 10.7960205078125, 11.807159423828125, 15.2947998046875, 3.02783203125, 15.840805053710938, 4.455841064453125, 7.9555511474609375, 15.89385986328125, -12.41351318359375, 10.855735778808594, -26.647735595703125, 27.0706787109375, 4.9073486328125, -4.811180114746094, -4.724365234375, 2.7931461334228516, -26.916259765625, 5.05426025390625, 2.1550827026367188, 0.063995361328125, 0.989410400390625, 7.187385559082031, -9.102508544921875, -0.03656768798828125, -13.089797973632812, 10.409591674804688, -8.313980102539062, -17.13519287109375, -19.067947387695312, 11.7449951171875, 42.72991943359375, 23.544708251953125, -9.551559448242188, 27.211624145507812, -24.6077880859375, 18.598663330078125, -7.131317138671875, -3.45269775390625, 19.244384765625, 37.9171142578125, -8.860015869140625, 0.41400909423828125, 13.242919921875, -23.820220947265625, 0.7379074096679688, -15.33245849609375, 9.954940795898438, 3.07220458984375, 33.73541259765625, 1.3872146606445312, 0.258331298828125, -2.8027191162109375, -16.215728759765625, 2.9605026245117188, 9.675880432128906, 7.121490478515625, -18.629302978515625, 0.9463348388671875, -1.4583759307861328, -18.479034423828125, -4.181396484375, 0.0, 24.1192626953125, 10.652374267578125, -1.6694564819335938, 13.13018798828125, 6.097930908203125, -1.146697998046875, 24.2022705078125, -2.5418701171875, 18.150360107421875, 2.4866943359375, 5.925468444824219, 9.923362731933594, 32.63157653808594, 13.062393188476562, 9.70489501953125, 37.52801513671875, 15.734565734863281, 5.809844970703125, 4.420433044433594, -8.561614990234375, 0.0, 33.1470947265625, -2.2385101318359375, -6.485115051269531, 11.297149658203125, 8.070770263671875, 20.762481689453125, 6.6653900146484375, 4.78900146484375, 6.203639984130859, 6.50909423828125, 2.9531707763671875, 8.16827392578125, 6.565155029296875, -1.869842529296875, 15.310333251953125, 7.581611633300781, 2.862762451171875, 15.208343505859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000114.npy"} {"epoch": 0.2387434554973822, "step": 115, "batch_size": 128, "mean": 5.767663955688477, "std": 15.29000186920166, "min": -30.443511962890625, "p10": -11.959991836547852, "median": 5.283025741577148, "p90": 28.34342041015625, "max": 57.9896240234375, "pos_frac": 0.6796875, "sample": [31.368988037109375, -15.213043212890625, 13.525115966796875, -11.412933349609375, -1.9285888671875, -13.549087524414062, -10.73858642578125, 2.63818359375, 2.7221450805664062, 20.330772399902344, -5.157344818115234, 10.957572937011719, 7.2635650634765625, 9.211688995361328, 2.9910507202148438, -9.771461486816406, 9.282936096191406, -4.745025634765625, 15.288055419921875, 1.221343994140625, -12.246318817138672, 2.37994384765625, 7.323577880859375, 7.47979736328125, 5.40349006652832, -3.1207122802734375, 1.73583984375, 13.554931640625, 12.502777099609375, 15.4044189453125, -4.109649658203125, 12.26654052734375, 11.15606689453125, 3.2777786254882812, 3.8708248138427734, 8.654632568359375, -9.621063232421875, -20.56689453125, -1.323455810546875, 31.65618896484375, 6.549644470214844, 28.127593994140625, 38.9061279296875, 6.9954833984375, -20.314208984375, -18.5433349609375, 24.532142639160156, -9.911689758300781, 27.205001831054688, 43.008544921875, 5.54632568359375, 57.9896240234375, -10.767669677734375, -2.018829345703125, -12.911651611328125, 15.073043823242188, 28.847015380859375, 23.123260498046875, 12.057968139648438, -23.732513427734375, -17.35357666015625, -7.0521240234375, 28.910491943359375, 1.993896484375, -4.3058624267578125, 11.85479736328125, 6.272026062011719, 1.8563232421875, 5.846397399902344, 31.56640625, 0.8403911590576172, 8.49627685546875, 14.231258392333984, 9.44366455078125, -3.1480178833007812, 6.095300674438477, 44.8006591796875, 0.867584228515625, -19.26934814453125, -7.615074157714844, 2.71844482421875, 0.145233154296875, -12.6031494140625, 36.2205810546875, 21.01490020751953, 4.40679931640625, -27.260589599609375, 9.705879211425781, 2.4121856689453125, 30.264678955078125, -4.561122894287109, 3.528656005859375, -4.7685546875, 33.726715087890625, 17.4263916015625, -5.2873687744140625, 2.53973388671875, 6.140869140625, -8.717941284179688, 6.83056640625, 10.964996337890625, 5.177272796630859, 9.468460083007812, 8.521041870117188, 9.047882080078125, 3.2393341064453125, 42.6298828125, 5.3887786865234375, -7.78851318359375, -11.8372802734375, 11.983245849609375, 4.883108139038086, 16.543121337890625, 6.9367218017578125, 12.746097564697266, -11.0919189453125, -5.579490661621094, 13.983654022216797, -0.9941329956054688, 20.86444091796875, -30.443511962890625, -3.1516494750976562, 11.793960571289062, 0.4564056396484375, 14.016281127929688, 3.698760986328125, -2.42999267578125, 11.293792724609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000115.npy"} {"epoch": 0.24083769633507854, "step": 116, "batch_size": 128, "mean": 8.059527397155762, "std": 14.646001815795898, "min": -40.7652587890625, "p10": -7.723500823974609, "median": 6.510387420654297, "p90": 24.885698699951167, "max": 56.1361083984375, "pos_frac": 0.71875, "sample": [3.606475830078125, -4.233852386474609, 5.7256011962890625, 3.25, 6.2634124755859375, 12.25494384765625, 8.71002197265625, 26.021484375, 42.8284912109375, 7.778289794921875, 19.368988037109375, 8.916015625, 20.2933349609375, 3.4192657470703125, 17.394805908203125, 11.400527954101562, -6.7624359130859375, 12.606155395507812, 4.122804641723633, -9.951301574707031, -21.24041748046875, -6.283075332641602, 22.394241333007812, 22.859039306640625, 0.2931995391845703, -9.64385986328125, 11.360260009765625, -0.6257553100585938, 5.5458984375, -0.09130859375, 2.2569026947021484, 24.00225830078125, -10.34246826171875, 16.46734619140625, -5.0792388916015625, 1.8422088623046875, 22.502029418945312, 13.81793212890625, -7.784698486328125, -6.38958740234375, -2.5936279296875, 19.473388671875, 16.088897705078125, -9.811538696289062, 7.2189483642578125, 4.813117980957031, -2.44964599609375, 4.9930419921875, 22.10736083984375, 34.3211669921875, -40.7652587890625, 2.887451171875, 13.684158325195312, -17.4107666015625, -1.1301193237304688, 26.1551513671875, 9.982025146484375, 24.39893341064453, -4.04718017578125, 0.5061149597167969, 10.028533935546875, 11.988174438476562, 6.12261962890625, 4.600128173828125, 9.48775863647461, 0.0, 7.4024658203125, 35.4998779296875, -2.4864425659179688, -9.168594360351562, 17.696197509765625, 34.47314453125, -1.316986083984375, -5.6600494384765625, -7.697273254394531, 2.3058242797851562, 2.1704940795898438, 1.6156272888183594, 5.951499938964844, 23.028106689453125, 3.2669639587402344, 37.4107666015625, 18.2769775390625, 0.0, 12.75320816040039, 6.962196350097656, 19.873748779296875, 5.4512786865234375, 35.181884765625, -1.27685546875, -1.8804855346679688, 56.1361083984375, 13.98370361328125, -36.21368408203125, 6.55487060546875, 31.16619873046875, 19.222183227539062, 4.9895172119140625, 10.087539672851562, 8.140716552734375, 11.366172790527344, -0.363311767578125, 8.277328491210938, 27.904708862304688, 20.15216064453125, -21.368446350097656, 7.4403076171875, 12.610763549804688, 1.9634246826171875, -11.892967224121094, 11.078094482421875, 7.939208984375, 44.785430908203125, -0.9681167602539062, 24.039077758789062, 16.891464233398438, 5.5169677734375, 16.11029052734375, 22.635406494140625, 13.182746887207031, -7.2265777587890625, 6.465904235839844, 0.1543121337890625, 30.0184326171875, -12.042144775390625, 6.271274566650391, -1.702789306640625, 18.9569091796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000116.npy"} {"epoch": 0.24293193717277486, "step": 117, "batch_size": 128, "mean": 8.39564323425293, "std": 13.80836009979248, "min": -25.100616455078125, "p10": -6.472610473632812, "median": 6.898429870605469, "p90": 28.211776733398438, "max": 43.168853759765625, "pos_frac": 0.7265625, "sample": [2.2043228149414062, 8.072460174560547, 11.81781005859375, 16.620803833007812, 5.414360046386719, 13.794052124023438, 1.6001739501953125, -1.60101318359375, 11.1944580078125, -3.460052490234375, 13.20562744140625, -7.57879638671875, -1.787139892578125, 13.858917236328125, 18.155193328857422, 3.49969482421875, 9.313034057617188, 9.281005859375, 25.3050537109375, 4.85772705078125, 19.278839111328125, -8.804473876953125, 6.8823394775390625, 2.1799564361572266, 14.470355987548828, 27.780517578125, 1.5755462646484375, -18.59551239013672, 11.5191650390625, 43.168853759765625, 15.525957107543945, 31.01324462890625, 34.197723388671875, 14.16622543334961, 9.462610244750977, -11.507110595703125, 24.122711181640625, 0.7489166259765625, 1.270486831665039, -8.469482421875, 34.9031982421875, 6.8854827880859375, 5.050994873046875, 28.0328369140625, 0.0, 26.368988037109375, -5.455047607421875, 6.0660400390625, 39.08587646484375, 23.737274169921875, 8.696121215820312, 34.8094482421875, -23.318466186523438, 17.786285400390625, 24.131103515625, 21.495452880859375, -5.4959716796875, 3.1931705474853516, 4.125, 6.9419097900390625, 17.47607421875, 14.9326171875, 13.05865478515625, 12.394084930419922, 6.29498291015625, 12.371734619140625, 13.165420532226562, -0.2054901123046875, -6.251007080078125, -6.98968505859375, 7.985633850097656, -21.60223388671875, -0.014377593994140625, 9.71146011352539, 28.629302978515625, -13.20574951171875, 11.044921875, 13.0850830078125, 20.553466796875, 2.3136634826660156, -2.5720748901367188, -2.1707763671875, 7.521759033203125, 33.25335693359375, 4.5276947021484375, -4.99542236328125, -5.7729339599609375, -14.86572265625, 0.315582275390625, 4.33134651184082, -25.100616455078125, 17.16290283203125, 29.374282836914062, 28.765289306640625, 8.755069732666016, 3.581939697265625, -9.837158203125, -2.4728546142578125, -1.0420150756835938, 29.17816162109375, 3.513519287109375, -4.355499267578125, 2.358030319213867, -4.628662109375, -4.6752777099609375, 6.911376953125, 27.710678100585938, 6.223365783691406, 26.5172119140625, -3.9306793212890625, 15.425464630126953, 16.87188720703125, 22.0162353515625, 0.226806640625, -0.0037078857421875, -5.4974822998046875, 4.18455696105957, 0.6733283996582031, 3.8571624755859375, -3.5781288146972656, -18.887420654296875, 38.315216064453125, 11.53643798828125, 14.14084243774414, 2.3095016479492188, 17.043922424316406, 37.25909423828125, 9.6239013671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000117.npy"} {"epoch": 0.2450261780104712, "step": 118, "batch_size": 128, "mean": 9.039901733398438, "std": 15.18082046508789, "min": -24.602691650390625, "p10": -9.338301086425782, "median": 7.6294708251953125, "p90": 26.47886962890625, "max": 52.440582275390625, "pos_frac": 0.75, "sample": [7.541816711425781, 9.567779541015625, 20.201751708984375, 3.92236328125, 29.342483520507812, 26.26654052734375, -21.827362060546875, -14.01266860961914, 19.370208740234375, 27.3699951171875, 13.709075927734375, 6.1978607177734375, 6.878177642822266, 6.12628173828125, -14.84613037109375, 33.784271240234375, 17.0679931640625, 19.7916259765625, 21.97203826904297, 6.835906982421875, 9.857452392578125, 18.6646728515625, 32.734153747558594, 0.602783203125, -24.03338623046875, -24.602691650390625, 16.379486083984375, 2.555877685546875, -20.651229858398438, 23.322479248046875, 18.489044189453125, 25.17339324951172, 44.16265869140625, 15.78173828125, 3.8318328857421875, 6.4930877685546875, 2.4465789794921875, 5.643913269042969, 18.607376098632812, 21.552108764648438, -3.505828857421875, 22.822509765625, 31.686492919921875, -2.8142642974853516, 2.2754058837890625, -5.3266448974609375, 19.60302734375, 16.40850830078125, 12.0211181640625, 0.064117431640625, 0.6038665771484375, 2.208740234375, 25.3131103515625, 17.32379150390625, 26.97430419921875, 40.471923828125, -7.997611999511719, 10.134307861328125, 0.000492095947265625, 22.438339233398438, -6.289295196533203, 0.0, -9.22100830078125, 9.685089111328125, -10.832122802734375, -6.841098785400391, -19.92364501953125, 6.5564422607421875, 15.290878295898438, -11.141036987304688, -0.467437744140625, 31.139129638671875, 0.551666259765625, 16.010589599609375, 6.412078857421875, 24.32611083984375, 24.798492431640625, 15.895614624023438, -9.611984252929688, 25.196014404296875, 5.003837585449219, 22.774795532226562, 8.7269287109375, -5.2247772216796875, 13.757850646972656, 25.463653564453125, -19.143524169921875, 1.964111328125, -9.068267822265625, -8.283218383789062, 49.568084716796875, -4.45263671875, 25.243209838867188, 8.459228515625, -2.3484420776367188, 10.278861999511719, 1.497467041015625, -14.594993591308594, 10.971435546875, -7.6399993896484375, 4.983057022094727, 36.1558837890625, 7.62078857421875, 13.142486572265625, -0.0400238037109375, 11.592849731445312, 5.16925048828125, 5.325347900390625, 10.504493713378906, 32.85621643066406, -6.119632720947266, -6.81146240234375, 17.01645278930664, 13.49566650390625, 4.9212646484375, 2.6346206665039062, -20.925506591796875, 10.49249267578125, 7.638153076171875, 20.70135498046875, 2.305450439453125, 15.6124267578125, 4.4232330322265625, 23.661895751953125, 0.0, 4.1868896484375, 16.657989501953125, 52.440582275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000118.npy"} {"epoch": 0.24712041884816754, "step": 119, "batch_size": 128, "mean": 7.528423309326172, "std": 18.212087631225586, "min": -40.283905029296875, "p10": -14.944094848632812, "median": 4.725155830383301, "p90": 30.1638671875, "max": 67.86239624023438, "pos_frac": 0.671875, "sample": [-40.283905029296875, -18.408103942871094, 24.28704833984375, 3.066802978515625, 2.5576400756835938, 4.721672058105469, -8.126579284667969, -7.371429443359375, 9.430908203125, -3.6722259521484375, 21.890655517578125, -10.275344848632812, -12.59942626953125, 31.323974609375, -15.3197021484375, 9.882720947265625, -13.96527099609375, 23.3382568359375, 27.410751342773438, -3.567901611328125, 7.59075927734375, 18.666946411132812, -0.8691349029541016, -8.881080627441406, 6.221466064453125, 15.075736999511719, 20.883499145507812, 7.59368896484375, 33.24749755859375, 4.673591613769531, 4.728639602661133, 2.2876129150390625, -3.953826904296875, 1.4361724853515625, 11.527130126953125, 9.999557495117188, 11.459625244140625, 3.548564910888672, 8.561691284179688, 2.8925399780273438, 23.061981201171875, -10.23626708984375, -0.835662841796875, 0.10921478271484375, 30.336944580078125, 0.7360668182373047, 4.5334014892578125, -33.8360595703125, -14.791961669921875, -20.924835205078125, 51.254364013671875, 11.73553466796875, 3.2667999267578125, 29.772674560546875, -4.818328857421875, -1.4287872314453125, 15.585906982421875, 2.144115447998047, 16.174636840820312, 23.26519012451172, 4.3603515625, -1.3511962890625, 24.402938842773438, 15.53631591796875, 18.98040771484375, -8.91583251953125, 39.736480712890625, 12.241065979003906, 33.62725830078125, 19.096466064453125, 0.3696155548095703, 7.856170654296875, 13.909637451171875, 0.0, 37.14715576171875, -2.283538818359375, -1.9185714721679688, -25.541229248046875, 20.6192626953125, -23.441558837890625, 22.125450134277344, 12.325469970703125, 4.469696044921875, -15.299072265625, 12.459426879882812, 0.0990753173828125, 31.20147705078125, -7.651527404785156, 16.005111694335938, 7.1568450927734375, 22.0032958984375, -15.99969482421875, -2.8803558349609375, 59.22113037109375, -3.9633636474609375, 67.86239624023438, 7.3349609375, -0.818572998046875, 7.706024169921875, -27.0880126953125, 22.93255615234375, 30.089691162109375, 2.04852294921875, 18.754364013671875, -7.579925537109375, 22.792236328125, -7.237823486328125, 4.30126953125, 5.216880798339844, -15.63763427734375, 2.1301116943359375, -22.4271240234375, 49.206817626953125, 14.447975158691406, 56.50433349609375, 6.7581939697265625, -2.9071502685546875, 16.184173583984375, -0.9487762451171875, 18.61920166015625, 6.91265869140625, 3.96405029296875, 26.76470947265625, -5.924383163452148, 39.199798583984375, 4.4877777099609375, 6.17401123046875, -18.00543212890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000119.npy"} {"epoch": 0.24921465968586387, "step": 120, "batch_size": 128, "mean": 9.0379638671875, "std": 17.22722625732422, "min": -33.32958984375, "p10": -12.511999511718745, "median": 9.250785827636719, "p90": 31.184835815429683, "max": 58.78094482421875, "pos_frac": 0.6640625, "sample": [-11.3701171875, 25.158599853515625, -5.22943115234375, 10.933685302734375, -4.049072265625, 12.07330322265625, 24.162261962890625, -10.519882202148438, 6.1060791015625, 17.37567138671875, -2.7386245727539062, -3.98883056640625, 21.05706024169922, 5.52288818359375, -17.435577392578125, -2.02020263671875, 32.535125732421875, -3.5144824981689453, 9.052688598632812, 17.283517837524414, -6.59197998046875, 32.798370361328125, 29.1954345703125, 25.68963623046875, 27.428619384765625, 12.366073608398438, 5.5689697265625, -1.574554443359375, -5.468395233154297, 30.60614013671875, 26.789306640625, -0.320343017578125, 4.834442138671875, 3.634552001953125, -24.854644775390625, 7.8593597412109375, -1.0030517578125, 14.97015380859375, 21.0859375, 22.18841552734375, 1.4953956604003906, -1.7308311462402344, 12.98297119140625, -26.526763916015625, 0.0, -2.2314682006835938, -21.963592529296875, -2.74310302734375, -6.630607604980469, 5.50048828125, 12.000564575195312, 12.300636291503906, 50.8880615234375, 20.416717529296875, 13.342498779296875, 9.370025634765625, -15.242080688476562, 14.795074462890625, 8.78387451171875, 3.2429885864257812, -3.3158187866210938, 0.0, 10.981689453125, 19.815963745117188, 58.78094482421875, 14.84576416015625, 16.010528564453125, 38.887054443359375, -15.1763916015625, 20.446372985839844, 22.1824951171875, 9.135848999023438, 4.240806579589844, 16.3165283203125, 14.440322875976562, -7.013336181640625, 9.2783203125, 33.75950622558594, 20.650238037109375, 18.83612060546875, 11.936859130859375, -16.089691162109375, -19.02655029296875, -8.550796508789062, -11.214508056640625, 38.678985595703125, 9.223251342773438, 36.55255126953125, 43.83447265625, 16.01842498779297, -0.5920333862304688, 26.307891845703125, -29.054443359375, 33.823333740234375, 36.861717224121094, 4.9238128662109375, 0.3105316162109375, 16.998550415039062, 12.157333374023438, 18.787643432617188, -22.57379913330078, 2.231769561767578, 0.25278472900390625, 19.255661010742188, 28.586318969726562, -1.2957725524902344, 11.108795166015625, -1.1940841674804688, 40.9879150390625, -33.32958984375, 22.23284912109375, 22.301666259765625, 20.20690155029297, -16.835906982421875, -23.37823486328125, 16.43072509765625, 14.046882629394531, 19.293609619140625, 6.3578948974609375, -1.620391845703125, -4.318416595458984, 15.114028930664062, -3.398956298828125, 50.026458740234375, 3.3366928100585938, 2.62298583984375, -3.113616943359375, 14.88604736328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000120.npy"} {"epoch": 0.2513089005235602, "step": 121, "batch_size": 128, "mean": 5.375214576721191, "std": 15.204877853393555, "min": -34.456268310546875, "p10": -11.842683410644527, "median": 4.455902099609375, "p90": 24.58706665039062, "max": 46.54815673828125, "pos_frac": 0.640625, "sample": [-1.6168746948242188, -0.271484375, 2.4969940185546875, 7.143089294433594, 0.8728485107421875, 4.4395751953125, -1.88970947265625, 6.356719970703125, 7.92889404296875, 27.66448974609375, 3.3602867126464844, -1.7923431396484375, 33.73857116699219, 5.233846664428711, 17.671077728271484, -0.4273223876953125, -3.549591064453125, 3.4954986572265625, 0.5299911499023438, 24.084716796875, 18.464187622070312, -16.675933837890625, 19.88385009765625, 16.76885986328125, -26.09710693359375, 0.0, 9.339103698730469, 5.0809326171875, 0.8337554931640625, 18.737823486328125, 9.333648681640625, 9.015766143798828, -32.46246337890625, 25.75921630859375, 5.933841705322266, -34.456268310546875, -3.8978424072265625, 6.1655426025390625, 21.649810791015625, 12.618667602539062, -2.9488677978515625, -8.645538330078125, 7.285308837890625, 1.9332399368286133, 6.2578887939453125, 1.3732070922851562, 16.650894165039062, -10.9234619140625, -9.140594482421875, 0.0, 8.64279556274414, -4.208620071411133, -0.6656646728515625, -3.48486328125, 15.473281860351562, -7.843086242675781, 0.910247802734375, 5.299896240234375, 18.507781982421875, -13.987533569335938, -2.241424560546875, -18.431411743164062, -2.7942276000976562, 46.103759765625, 13.166946411132812, 9.046199798583984, 42.330657958984375, -5.62939453125, -4.502197265625, -1.934112548828125, -5.139137268066406, 15.71551513671875, -4.88507080078125, 14.391998291015625, -9.079917907714844, 2.14404296875, 4.223941802978516, 8.358055114746094, -1.3881683349609375, 14.978557586669922, 13.727890014648438, -1.08477783203125, -7.8688507080078125, 28.738143920898438, 45.5267333984375, 5.29840087890625, -25.431114196777344, -1.64190673828125, -21.249267578125, 46.54815673828125, 16.515762329101562, 19.905563354492188, 17.7325439453125, 2.615081787109375, 33.30879211425781, 3.0960540771484375, -21.852874755859375, 4.4810028076171875, 12.341300964355469, 1.0024185180664062, 2.6989593505859375, 8.436538696289062, 14.114799499511719, 8.757598876953125, 4.28717041015625, 14.5946044921875, -1.34112548828125, 4.680107116699219, -3.7417678833007812, -3.0938873291015625, 4.817413330078125, 36.0399169921875, 8.810760498046875, 4.47222900390625, 8.268157958984375, 9.256599426269531, 27.651687622070312, -25.148300170898438, 28.428314208984375, 13.428619384765625, -1.1195297241210938, -14.7999267578125, 40.52830505371094, 13.956527709960938, 1.037820816040039, -16.8778076171875, -22.981781005859375, 14.770751953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000121.npy"} {"epoch": 0.2534031413612565, "step": 122, "batch_size": 128, "mean": 7.439023971557617, "std": 18.26180076599121, "min": -37.487518310546875, "p10": -15.670248413085936, "median": 6.9904022216796875, "p90": 30.61583557128906, "max": 60.11151123046875, "pos_frac": 0.625, "sample": [21.25177001953125, -15.07659912109375, 26.183425903320312, -15.38983154296875, 14.331893920898438, 22.326644897460938, 10.317596435546875, 4.4087371826171875, -33.8751220703125, 29.419448852539062, 25.37109375, 0.0, 8.905792236328125, -24.70379638671875, -0.4706153869628906, -37.487518310546875, 12.078033447265625, 17.36419677734375, -12.465499877929688, 13.5311279296875, -4.32489013671875, 22.52447509765625, -22.213958740234375, -30.87371826171875, 6.488037109375, 33.13714599609375, -13.687469482421875, 10.191621780395508, 46.206329345703125, 2.867156982421875, 9.247528076171875, -3.112335205078125, 8.749786376953125, 8.428672790527344, 16.977935791015625, -5.797904968261719, 4.128044128417969, -22.235870361328125, -0.037109375, 9.827842712402344, -29.221450805664062, 12.61883544921875, 7.724365234375, 45.03570556640625, -1.6758346557617188, -23.23468017578125, -10.828903198242188, 8.411745071411133, 39.640289306640625, -6.979339599609375, -6.554588317871094, 10.267669677734375, 13.07379150390625, -21.629318237304688, 12.619354248046875, 13.865280151367188, -0.3433094024658203, 23.748512268066406, -29.32037353515625, 10.557373046875, 14.690032958984375, 20.672943115234375, 19.9176025390625, -1.30926513671875, -8.340034484863281, 34.1478271484375, 14.985420227050781, 25.06903076171875, 20.84454345703125, 30.3140869140625, 46.876312255859375, 13.396806716918945, 1.1579437255859375, 37.72210693359375, 60.11151123046875, 7.4614715576171875, 5.05548095703125, -2.437835693359375, 31.5562744140625, 23.7276611328125, -3.736083984375, -16.324554443359375, 3.18658447265625, -9.959136962890625, -5.2236328125, -2.341888427734375, -0.6147842407226562, 12.367256164550781, 6.777435302734375, -5.3045501708984375, 17.361083984375, -0.4360809326171875, 2.5885353088378906, -17.878890991210938, -3.7220382690429688, -19.305084228515625, -1.6172027587890625, 27.874404907226562, 22.765594482421875, -0.830352783203125, 5.7835845947265625, -5.3502349853515625, 22.5390625, 18.368621826171875, 2.7423973083496094, 23.859832763671875, 21.97784423828125, 28.57611846923828, 4.465339660644531, 2.2010231018066406, -0.9259185791015625, -1.41015625, -6.9975128173828125, 15.798843383789062, 50.3670654296875, 0.659027099609375, 33.0372314453125, -3.7465362548828125, 2.7239227294921875, 26.078689575195312, 11.743682861328125, 20.74871826171875, -4.139751434326172, 36.73724365234375, -5.943248748779297, 2.243194580078125, 31.319915771484375, 7.203369140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000122.npy"} {"epoch": 0.2554973821989529, "step": 123, "batch_size": 128, "mean": 10.552115440368652, "std": 18.66358757019043, "min": -49.49884033203125, "p10": -11.341659545898438, "median": 9.475677490234375, "p90": 34.113156127929685, "max": 68.37789916992188, "pos_frac": 0.7265625, "sample": [25.192367553710938, -10.496612548828125, 30.071090698242188, -11.1268310546875, 7.2764739990234375, 6.52703857421875, 9.930450439453125, 3.96142578125, 11.55584716796875, 10.86956787109375, -0.6259040832519531, 2.853271484375, -2.3640594482421875, 2.152679443359375, 7.298131942749023, -9.431915283203125, -11.842926025390625, 17.352569580078125, 45.828277587890625, 8.95733642578125, -21.12994384765625, -22.605621337890625, 9.602615356445312, 16.474361419677734, 46.7525634765625, 3.2847137451171875, 19.90277099609375, 29.385498046875, 29.37310791015625, 1.2956466674804688, 28.516387939453125, 23.568084716796875, 19.57806396484375, 24.572174072265625, 9.348739624023438, 23.446990966796875, 23.10626220703125, -24.9541015625, 2.52020263671875, 30.761016845703125, 19.188735961914062, 3.6649017333984375, 0.36737060546875, 33.790771484375, 9.945037841796875, 57.107696533203125, 1.6106834411621094, 3.098529815673828, 19.72430419921875, 28.60894775390625, 3.164844512939453, 18.399948120117188, 20.609527587890625, 68.37789916992188, 47.216033935546875, 12.489418029785156, -8.278141021728516, -1.587646484375, 3.4971923828125, 46.202423095703125, -3.0494232177734375, 6.018119812011719, -49.49884033203125, 33.15796661376953, 8.848625183105469, -17.880462646484375, 23.87939453125, 5.832489013671875, 34.865386962890625, 7.59172248840332, 12.40423583984375, -1.1357421875, 8.7513427734375, -18.647056579589844, 10.939559936523438, 27.391693115234375, 22.9566650390625, -4.415679931640625, -17.8448486328125, 44.31890869140625, -4.8373260498046875, 11.743499755859375, 20.19906997680664, -5.34051513671875, 8.231330871582031, 19.529510498046875, -20.218597412109375, 17.036575317382812, 18.289581298828125, -6.60845947265625, 0.0, -3.8992538452148438, 37.507659912109375, 11.898971557617188, 10.7293701171875, 0.0, 12.859268188476562, 9.25384521484375, 0.7819442749023438, 20.586685180664062, -5.735874176025391, -14.112548828125, 5.5359954833984375, 26.611602783203125, -1.23175048828125, 11.912788391113281, 10.2913818359375, 35.331817626953125, 16.347259521484375, -16.250701904296875, -2.206573486328125, -2.02484130859375, 29.713043212890625, 8.918167114257812, -30.379623413085938, 49.6298828125, 14.71636962890625, -7.516571044921875, 13.79315185546875, -2.9754562377929688, 0.019256591796875, 9.882781982421875, 6.496673583984375, 34.962799072265625, 44.54969787597656, -19.72747802734375, 16.925811767578125, 21.030181884765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000123.npy"} {"epoch": 0.25759162303664923, "step": 124, "batch_size": 128, "mean": 9.558961868286133, "std": 17.343643188476562, "min": -50.43658447265625, "p10": -7.979665565490722, "median": 7.302577972412109, "p90": 33.75798797607422, "max": 57.224212646484375, "pos_frac": 0.765625, "sample": [-0.52813720703125, 29.05902862548828, -50.43658447265625, 17.410552978515625, -3.5806331634521484, 2.3124542236328125, 26.045379638671875, 7.977783203125, 34.299041748046875, 46.416259765625, 4.385467529296875, -28.9681396484375, 6.493011474609375, -7.7294158935546875, 21.659210205078125, 4.718879699707031, 5.246612548828125, 20.712615966796875, -6.652301788330078, 12.4317626953125, 5.6352386474609375, 5.6170654296875, -0.10955810546875, 9.211669921875, 20.50634765625, 2.591033935546875, -1.080810546875, -5.358127593994141, 3.6152725219726562, 8.566184997558594, 29.422637939453125, -15.006256103515625, 4.05364990234375, 24.131866455078125, 45.610015869140625, 38.00065612792969, 19.815467834472656, 13.216262817382812, 5.91986083984375, 23.03436279296875, 21.674652099609375, 18.21703338623047, 39.753082275390625, 9.9796142578125, 3.623636245727539, -2.313507080078125, 0.051849365234375, 13.999603271484375, 46.644744873046875, -8.563581466674805, 9.839263916015625, -17.932384490966797, 6.439300537109375, 0.0, 2.3925018310546875, -2.6885986328125, 25.424163818359375, 5.452056884765625, 3.646636962890625, -11.510246276855469, 33.52610778808594, -37.117156982421875, -26.65130615234375, 40.04803466796875, 13.736175537109375, 24.57098388671875, 8.6263427734375, 32.65460205078125, 25.13827133178711, -2.6361541748046875, 11.85223388671875, 5.1314697265625, -20.8409423828125, -26.739532470703125, 21.30138397216797, 2.466644287109375, 4.266120910644531, 13.302963256835938, 34.647552490234375, 22.23480224609375, 11.004302978515625, 29.43853759765625, 7.9073486328125, -17.196319580078125, -4.96942138671875, 6.4246673583984375, 3.2718963623046875, 14.667953491210938, 34.830291748046875, 16.553577423095703, 2.635101318359375, 13.778640747070312, -2.8605194091796875, 12.500152587890625, 9.927871704101562, 7.620283126831055, 0.883392333984375, 57.224212646484375, 12.696874618530273, 21.49427032470703, -3.617340087890625, 1.5006103515625, 14.220306396484375, 2.3983078002929688, 1.2203369140625, 50.57012939453125, -9.719985961914062, -0.55474853515625, 7.286231994628906, 18.34423828125, -3.798065185546875, 1.39263916015625, 7.180755615234375, 12.37615966796875, 7.3189239501953125, 20.977365493774414, 9.013595581054688, 34.554901123046875, 0.3320884704589844, -2.3616790771484375, 4.270172119140625, 15.735992431640625, 23.680023193359375, 44.580413818359375, 4.5460205078125, 11.830718994140625, 3.7027511596679688, -13.580886840820312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000124.npy"} {"epoch": 0.25968586387434556, "step": 125, "batch_size": 128, "mean": 9.315194129943848, "std": 19.017906188964844, "min": -33.16729736328125, "p10": -11.962199783325195, "median": 6.563804626464844, "p90": 35.349047851562496, "max": 64.07733154296875, "pos_frac": 0.640625, "sample": [4.697509765625, -8.082063674926758, -33.16729736328125, 38.9952392578125, 1.6627864837646484, 13.743301391601562, 16.044830322265625, -11.826171875, 3.4042749404907227, -16.194305419921875, 3.76287841796875, -2.7816696166992188, 32.3814697265625, -1.53814697265625, 12.9783935546875, 24.732269287109375, -27.897857666015625, 13.556259155273438, 13.901168823242188, 14.386154174804688, -6.99456787109375, -4.018779754638672, 19.1134033203125, -5.774650573730469, 16.6590576171875, 13.693077087402344, 11.785293579101562, 14.79608154296875, -3.5546722412109375, 45.9161376953125, 5.1702880859375, -6.6471099853515625, 11.821395874023438, -1.5898246765136719, 7.115943908691406, 32.2647705078125, 21.3060302734375, 37.0535888671875, 15.520614624023438, -4.033729553222656, 1.529022216796875, -22.623504638671875, -11.510589599609375, -6.937999725341797, -6.441986083984375, 42.6156005859375, 0.23968505859375, -7.641021728515625, -32.78517150878906, 15.081275939941406, 18.421630859375, -16.023574829101562, 24.6329345703125, 64.07733154296875, 7.3442230224609375, 13.402885437011719, 8.997886657714844, 20.2774658203125, 19.848236083984375, 26.937255859375, -4.43792724609375, 36.2264404296875, 14.092819213867188, -0.9724903106689453, 0.061065673828125, 14.26416015625, 4.209686279296875, 0.4423408508300781, 6.24261474609375, 5.35357666015625, 55.979400634765625, 11.555099487304688, 29.510337829589844, 43.35394287109375, -12.279598236083984, -18.905120849609375, 6.5673828125, 24.204360961914062, 34.9730224609375, 24.692066192626953, 31.933868408203125, 4.5609283447265625, -2.308319091796875, 42.57354736328125, -0.077423095703125, 12.996963500976562, -14.09381103515625, -7.300537109375, 20.833242416381836, 31.692840576171875, 28.031021118164062, 4.074432373046875, -6.3837432861328125, 13.11480712890625, -9.2044677734375, -1.1472702026367188, -9.70318603515625, 28.02386474609375, 27.194671630859375, -0.3437042236328125, -6.7633819580078125, 0.0, 3.307497024536133, -3.4248809814453125, 40.146575927734375, -8.856353759765625, -0.02215576171875, 4.576934814453125, 6.5602264404296875, 47.907318115234375, -6.8619842529296875, -13.09442138671875, 19.325103759765625, 30.904861450195312, 0.6902332305908203, 26.342742919921875, 50.59027099609375, 11.31170654296875, -15.200469970703125, 17.01837158203125, 27.638946533203125, -32.00172424316406, -7.810966491699219, 47.77056884765625, 12.981819152832031, 18.274017333984375, -3.1904144287109375, -17.17950439453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000125.npy"} {"epoch": 0.2617801047120419, "step": 126, "batch_size": 128, "mean": 10.590585708618164, "std": 19.14838409423828, "min": -43.432525634765625, "p10": -11.032455444335938, "median": 9.29534912109375, "p90": 33.647561645507814, "max": 52.04412841796875, "pos_frac": 0.6953125, "sample": [22.31866455078125, 51.593292236328125, 12.947341918945312, 17.5743408203125, 28.6427001953125, 42.70147705078125, 12.674186706542969, 40.37469482421875, 4.285617828369141, -1.8537406921386719, -43.432525634765625, 24.655792236328125, 2.86871337890625, 18.780059814453125, 27.500682830810547, 3.798126220703125, 27.7257080078125, -4.6998291015625, -21.70953369140625, -7.235198974609375, -14.2684326171875, 29.493179321289062, -10.59112548828125, -7.54713249206543, 5.1317138671875, -9.250686645507812, -3.219940185546875, -3.05706787109375, 2.356842041015625, -6.46783447265625, 9.2452392578125, -2.5891876220703125, 16.94731903076172, 14.102981567382812, 32.84234619140625, -11.019866943359375, -8.491630554199219, 0.9091796875, -0.575927734375, 5.592987060546875, 8.379402160644531, 23.020751953125, 33.8466796875, -1.0487747192382812, 13.019744873046875, 24.4669189453125, 8.606250762939453, 14.9781494140625, -14.89349365234375, -4.370880126953125, 33.562225341796875, -0.30545806884765625, 12.400634765625, 49.69403076171875, 3.2287139892578125, 36.36412811279297, 16.73340606689453, 6.3499908447265625, 4.148750305175781, 21.193334579467773, 10.93758773803711, 29.853759765625, 20.0015869140625, 23.919769287109375, -14.699859619140625, 17.697509765625, 5.651557922363281, 25.298019409179688, 21.784103393554688, 47.60986328125, 39.4779052734375, -5.03106689453125, 27.165679931640625, 28.524375915527344, -3.7362594604492188, 6.4041290283203125, 10.30126953125, -7.4090728759765625, -0.06614875793457031, -25.53607177734375, 28.140029907226562, 10.538818359375, -2.9838027954101562, 45.19500732421875, 22.133056640625, 2.945301055908203, -29.317230224609375, 5.888763427734375, 20.6365966796875, -37.12054443359375, 13.72833251953125, -9.545860290527344, 25.5810546875, 32.881622314453125, -42.999237060546875, -10.344100952148438, 4.232576370239258, 9.345458984375, 7.081939697265625, -15.904510498046875, 24.274070739746094, 26.539138793945312, 33.405029296875, 43.360626220703125, 6.38665771484375, 20.059249877929688, -7.1849365234375, 45.892364501953125, 21.645263671875, 37.376312255859375, -11.06182861328125, 20.350341796875, 17.963356018066406, 33.19294738769531, 20.608509063720703, 4.16717529296875, 4.219940185546875, -2.280364990234375, 10.6739501953125, -0.9133758544921875, 14.6553955078125, -11.144500732421875, 2.42083740234375, 4.272735595703125, 30.15606689453125, 52.04412841796875, 4.3128814697265625, -16.486907958984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000126.npy"} {"epoch": 0.2638743455497382, "step": 127, "batch_size": 128, "mean": 12.38076400756836, "std": 19.267282485961914, "min": -41.784271240234375, "p10": -10.151010131835935, "median": 9.104278564453125, "p90": 37.592987060546875, "max": 56.8076171875, "pos_frac": 0.734375, "sample": [-0.8459625244140625, -0.90582275390625, -9.65277099609375, 26.10357666015625, 10.64849853515625, 3.513824462890625, 19.502349853515625, 12.434417724609375, -11.313568115234375, -41.784271240234375, 28.087982177734375, 3.847949981689453, 20.0426025390625, 17.114233016967773, 9.26312255859375, -33.08038330078125, -8.180206298828125, 13.271316528320312, -17.989593505859375, 7.78375244140625, 8.54827880859375, 7.3294677734375, 21.216552734375, 3.8451690673828125, 19.599960327148438, 49.0535888671875, 20.97821044921875, 36.287139892578125, 12.395217895507812, 24.878448486328125, 37.117645263671875, 14.448783874511719, 29.91180419921875, 4.176197052001953, 9.34027099609375, 46.47784423828125, -9.517425537109375, -20.353607177734375, -1.6559066772460938, 9.264335632324219, 47.65008544921875, 7.4785003662109375, 6.8516693115234375, 7.602294921875, -15.317169189453125, -24.254898071289062, 36.78678894042969, 37.255043029785156, 11.587677001953125, -20.53662109375, 17.406707763671875, 34.788970947265625, 20.074859619140625, -6.9893951416015625, 30.203033447265625, 39.624664306640625, 24.5650634765625, 14.712631225585938, -3.82427978515625, 2.7016754150390625, 4.33782958984375, 5.105316162109375, 38.1080322265625, 23.91778564453125, 8.388351440429688, 33.422119140625, -7.237335205078125, 6.572021484375, 31.058090209960938, 37.37225341796875, 25.72650146484375, -2.4567108154296875, -11.386459350585938, -2.4525279998779297, 8.9454345703125, -11.375457763671875, -1.6759033203125, -15.46978759765625, 7.3189849853515625, -3.8306617736816406, 5.51715087890625, 22.348464965820312, 56.8076171875, -1.7983207702636719, 41.08331298828125, 19.159072875976562, 4.6344146728515625, 55.4481201171875, 31.10894775390625, 21.098388671875, 46.436004638671875, -8.19830322265625, 7.76495361328125, 10.4410400390625, 3.9728660583496094, -2.0422420501708984, -20.70452308654785, 6.71978759765625, 9.376317977905273, 30.537109375, -5.2800750732421875, 8.075531005859375, -9.478897094726562, 30.757415771484375, 23.48155975341797, 4.298301696777344, 39.789825439453125, 48.96728515625, 2.094329833984375, -3.438232421875, 6.1059417724609375, 3.836883544921875, 24.080795288085938, 36.016265869140625, -17.385421752929688, 4.232749938964844, 37.04632568359375, 39.20086669921875, -2.304851531982422, 10.168121337890625, 28.323883056640625, 22.29461669921875, 1.17791748046875, 30.523590087890625, 3.0054931640625, -6.278076171875, 22.782379150390625, 48.97491455078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000127.npy"} {"epoch": 0.26596858638743454, "step": 128, "batch_size": 128, "mean": 13.56689167022705, "std": 20.874980926513672, "min": -45.77911376953125, "p10": -11.06713409423828, "median": 12.88832950592041, "p90": 38.501862335205075, "max": 73.70220947265625, "pos_frac": 0.7109375, "sample": [27.955078125, 4.5749664306640625, 19.855926513671875, 24.4656982421875, -35.081085205078125, -33.5859375, -19.5260009765625, 8.117538452148438, 24.305084228515625, 29.578811645507812, 24.629371643066406, 2.6848297119140625, 35.76239776611328, -0.0354766845703125, 39.29402160644531, 0.8547935485839844, 3.5998802185058594, 23.636917114257812, 27.0543212890625, 23.713211059570312, -3.5502777099609375, -1.3005905151367188, 35.83568572998047, 66.92913818359375, 3.034149169921875, 7.9801483154296875, -3.64398193359375, 4.78631591796875, 29.192230224609375, 21.59088134765625, 9.35845947265625, 11.927003860473633, -19.32391357421875, 34.79571533203125, -7.763162612915039, 31.55323028564453, -16.313583374023438, 7.571632385253906, -1.18353271484375, -21.3643798828125, 24.279273986816406, 3.7210693359375, -3.3918380737304688, 17.3878173828125, 33.109649658203125, -3.1179656982421875, 0.49930572509765625, 24.964508056640625, 19.94830322265625, 38.27605438232422, 35.2332763671875, 26.12359619140625, -3.722686767578125, 35.1776123046875, -1.8146820068359375, -5.7175445556640625, 2.9190673828125, 24.944862365722656, 18.407241821289062, 41.158782958984375, 27.723979949951172, 6.496612548828125, 14.19575309753418, -0.49469757080078125, 1.0794830322265625, 7.3050079345703125, 39.02874755859375, 29.7811279296875, 30.381683349609375, 13.849655151367188, 15.25665283203125, 28.820404052734375, 49.579925537109375, 17.5731201171875, 47.790557861328125, 34.99957275390625, 11.421981811523438, -5.7468414306640625, -2.41070556640625, -1.1839332580566406, 40.34657287597656, 32.992431640625, -14.301544189453125, 17.952133178710938, 30.06695556640625, -45.77911376953125, 20.85931396484375, -21.599090576171875, 18.864578247070312, 73.70220947265625, 9.560516357421875, -11.52032470703125, 34.518402099609375, 17.079559326171875, -7.85186767578125, 1.066009521484375, -15.7327880859375, -6.973358154296875, 7.2388916015625, 11.053337097167969, 36.00897216796875, -5.6278839111328125, 4.3798828125, 45.27808380126953, 22.97003936767578, 35.49755859375, -7.354625701904297, -5.2253570556640625, 56.339874267578125, 1.0664482116699219, 7.43402099609375, 0.0, -21.185989379882812, -7.285430908203125, 26.5633544921875, -11.529205322265625, -10.872909545898438, 14.79217529296875, 24.151535034179688, 42.20953369140625, 4.23358154296875, 20.18218994140625, 11.542007446289062, 39.822601318359375, 60.965057373046875, 25.82122802734375, 19.57867431640625, -0.5353927612304688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000128.npy"} {"epoch": 0.2680628272251309, "step": 129, "batch_size": 128, "mean": 12.691354751586914, "std": 21.606300354003906, "min": -46.919708251953125, "p10": -11.68356246948242, "median": 11.464057922363281, "p90": 41.914495849609374, "max": 80.28433227539062, "pos_frac": 0.7734375, "sample": [-46.919708251953125, 28.7918701171875, -23.487899780273438, -13.16943359375, 11.377853393554688, 44.7461051940918, 27.8399658203125, 37.33576202392578, 30.410446166992188, 2.6888427734375, 32.40667724609375, 2.3485889434814453, 13.42645263671875, 0.77508544921875, -6.588203430175781, -10.666458129882812, 27.50750732421875, 36.594573974609375, 13.298126220703125, -11.046760559082031, 12.247852325439453, 12.378089904785156, 1.4697494506835938, 20.660125732421875, 42.959625244140625, -6.2520751953125, 16.736236572265625, -2.4028472900390625, 80.28433227539062, 60.3670654296875, 5.374237060546875, 21.62835693359375, 52.766693115234375, 16.700347900390625, 1.1180801391601562, -42.4383544921875, 33.6182861328125, 6.201202392578125, 61.572265625, 11.597911834716797, -4.8597869873046875, 14.66778564453125, 42.36639404296875, 44.07476806640625, 12.003585815429688, 43.093475341796875, -25.366943359375, 10.07728385925293, 1.0156211853027344, 4.344184875488281, 28.284912109375, 17.82586669921875, 8.586654663085938, 25.84307861328125, -10.561767578125, 8.176155090332031, 6.7855682373046875, 41.7208251953125, 2.58258056640625, 29.10387420654297, 1.5439338684082031, 2.7437744140625, 41.6119384765625, 22.16338348388672, 61.155303955078125, -6.822509765625, 14.602447509765625, 17.203399658203125, 27.006103515625, 3.5027923583984375, 4.723480224609375, 6.450782775878906, 2.7361297607421875, -4.265174865722656, 4.4305419921875, -14.510040283203125, 29.319564819335938, 21.59162139892578, -2.9172935485839844, 60.6439208984375, 27.242507934570312, 9.822036743164062, 20.94476318359375, 4.127464294433594, 2.284503936767578, 1.470733642578125, 30.435653686523438, 4.6063079833984375, 19.301666259765625, 39.696380615234375, -10.124542236328125, -27.2764892578125, -2.776702880859375, 0.544097900390625, -6.1701507568359375, 13.8602294921875, -10.832000732421875, -33.192291259765625, 15.699203491210938, 12.1552734375, 46.0599365234375, -15.1380615234375, 11.550262451171875, -3.6540794372558594, -20.566085815429688, -23.835662841796875, 12.06365966796875, 3.60845947265625, 16.35101318359375, 24.505035400390625, 29.292404174804688, 6.201751708984375, 18.784912109375, 11.09808349609375, 9.884605407714844, 45.906951904296875, 9.895225524902344, 30.48089599609375, -20.7540283203125, 18.844314575195312, 10.632179260253906, -0.49423980712890625, 3.6945877075195312, 31.308609008789062, 14.514060974121094, -14.248283386230469, 27.0101318359375, 32.74336624145508], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000129.npy"} {"epoch": 0.27015706806282724, "step": 130, "batch_size": 128, "mean": 11.693221092224121, "std": 19.642427444458008, "min": -38.328765869140625, "p10": -11.353901672363278, "median": 9.145927429199219, "p90": 37.90469970703125, "max": 56.135040283203125, "pos_frac": 0.734375, "sample": [7.8927001953125, -2.852813720703125, 30.600128173828125, 4.779449462890625, 5.54541015625, 41.42242431640625, 8.882080078125, 29.210693359375, 40.391502380371094, -9.227569580078125, -3.1896438598632812, 8.86968994140625, 23.387527465820312, 39.88677978515625, 0.93878173828125, 6.891845703125, -3.67889404296875, -16.521636962890625, 7.8832855224609375, 20.10455322265625, 17.488128662109375, 38.637939453125, -1.849822998046875, 9.232864379882812, -2.460878372192383, 56.135040283203125, 0.946319580078125, 22.80914306640625, 48.543243408203125, 50.0986328125, -31.3505859375, 0.595001220703125, 0.0, -10.716598510742188, 25.227996826171875, 11.883209228515625, 24.39849853515625, -36.869384765625, 8.395206451416016, 22.642181396484375, 25.786659240722656, 14.925262451171875, 47.504547119140625, -10.208267211914062, -37.9395751953125, -38.328765869140625, -3.255767822265625, 34.31196594238281, 4.770111083984375, -0.7388229370117188, 35.5966796875, 1.2185630798339844, 10.54541015625, 33.54588317871094, -16.236083984375, -0.7200164794921875, 12.985607147216797, 18.879791259765625, 37.5904541015625, -1.97265625, 9.465591430664062, 0.9955101013183594, 23.9532470703125, 16.60784912109375, 5.986328125, 43.59434509277344, 28.0228271484375, -17.08538818359375, -12.8409423828125, -1.4882049560546875, 39.2982177734375, 8.484458923339844, 30.71026611328125, -25.51678466796875, 5.2632904052734375, 17.59100341796875, -15.604827880859375, 29.617828369140625, 9.173843383789062, 15.543636322021484, 29.125991821289062, 18.509979248046875, 30.82073974609375, 16.4013671875, 32.612884521484375, 0.5616817474365234, 33.138946533203125, 9.118011474609375, 27.608123779296875, -3.466510772705078, 11.460693359375, 7.0655517578125, 4.78546142578125, 3.3036651611328125, 22.206619262695312, -28.95062255859375, -5.9920654296875, 5.6977386474609375, 2.772369384765625, 48.5028076171875, 44.0020751953125, 35.52978515625, 20.277664184570312, 1.0631790161132812, -25.23944091796875, 25.91998291015625, 4.569671630859375, 5.403961181640625, 18.909515380859375, -5.0779266357421875, 34.78263854980469, 7.069202423095703, 28.4979248046875, -5.345726013183594, 19.92877197265625, 19.054489135742188, 38.844146728515625, 21.0032958984375, -1.3426628112792969, 17.3599853515625, 15.372039794921875, -4.483551025390625, -1.578094482421875, 2.5211334228515625, 30.84326171875, -23.39166259765625, 4.7891387939453125, 27.130615234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000130.npy"} {"epoch": 0.27225130890052357, "step": 131, "batch_size": 128, "mean": 9.59052848815918, "std": 22.08632469177246, "min": -46.46379089355469, "p10": -17.669525146484375, "median": 8.229389190673828, "p90": 39.691604614257805, "max": 70.37777709960938, "pos_frac": 0.6328125, "sample": [30.785934448242188, 9.615701675415039, 37.724151611328125, -21.502166748046875, 12.602035522460938, 3.1414337158203125, 2.450977325439453, 23.204071044921875, -17.163101196289062, 41.56573486328125, 31.234153747558594, 31.348297119140625, 42.91387939453125, 17.066375732421875, -16.6632080078125, 1.42864990234375, 2.07635498046875, -16.82867431640625, 9.140724182128906, 47.056243896484375, 34.169158935546875, 8.1754150390625, -6.737274169921875, -5.683828353881836, 30.106117248535156, -5.094879150390625, -11.79925537109375, 42.53460693359375, 11.223007202148438, 13.4285888671875, -1.4188766479492188, 58.73187255859375, -6.8467254638671875, 0.002288818359375, 19.3131103515625, -3.2847156524658203, -22.895591735839844, 42.146484375, -2.31671142578125, 30.389801025390625, 9.04083251953125, 3.7775115966796875, -2.976703643798828, -16.23333740234375, 24.982421875, 7.180488586425781, 57.375518798828125, -17.20819091796875, 18.511932373046875, 25.499267578125, -1.6576309204101562, -3.828857421875, -17.140899658203125, 14.77618408203125, 14.093994140625, 40.622039794921875, -23.96636962890625, -5.919525146484375, 25.736347198486328, 35.29981994628906, -8.420150756835938, -21.17059326171875, -18.7459716796875, 31.190208435058594, 7.682769775390625, 6.1110687255859375, 16.223236083984375, 29.4007568359375, 70.37777709960938, 8.083648681640625, -46.46379089355469, 3.7730941772460938, 11.309722900390625, -10.174549102783203, -3.7890243530273438, 18.755126953125, -27.961502075195312, 62.2777099609375, -6.7308197021484375, 35.18714904785156, 10.904541015625, -21.77618408203125, 2.1089706420898438, -13.783454895019531, 11.783843994140625, 19.708465576171875, -14.71844482421875, 38.06292724609375, 42.8475341796875, -24.932891845703125, -0.72833251953125, -3.6156082153320312, 44.196014404296875, -21.927806854248047, 27.5267333984375, 5.966621398925781, 1.3457965850830078, 8.283363342285156, -8.172477722167969, 16.769859313964844, 11.339370727539062, -19.451812744140625, 28.89215087890625, -4.263191223144531, 10.049224853515625, 34.40830993652344, -2.091217041015625, -33.802978515625, 53.6934814453125, 7.814239501953125, 32.8492431640625, 39.2928466796875, 15.632171630859375, 19.21527099609375, -14.212646484375, -2.20904541015625, 21.076904296875, -9.8394775390625, 27.815433502197266, 9.950836181640625, 19.777435302734375, 13.651626586914062, 23.14892578125, 17.808944702148438, -5.2821044921875, -20.769493103027344, -8.894073486328125, 1.9168586730957031], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000131.npy"} {"epoch": 0.2743455497382199, "step": 132, "batch_size": 128, "mean": 12.922745704650879, "std": 24.91107177734375, "min": -46.173736572265625, "p10": -17.007156372070312, "median": 10.429153442382812, "p90": 44.22190704345703, "max": 96.35284423828125, "pos_frac": 0.6953125, "sample": [-9.8516845703125, 26.016357421875, 37.706329345703125, 12.947494506835938, -0.530426025390625, -9.6427001953125, 27.70722198486328, 11.0771484375, -3.704559326171875, 37.1480712890625, 3.8778610229492188, -10.723800659179688, -20.33624267578125, 36.07330322265625, 22.022171020507812, 9.102127075195312, 0.865936279296875, 45.09587097167969, 41.83094787597656, 61.323577880859375, 34.648162841796875, 30.731216430664062, 73.97369384765625, 48.0859375, 14.586265563964844, 1.3315773010253906, -0.8800163269042969, -32.64067077636719, 14.76507568359375, 19.955718994140625, -2.211151123046875, 46.975494384765625, 2.5446929931640625, 20.15447998046875, -21.4259033203125, -16.33960723876953, -16.712684631347656, 9.269561767578125, 4.4048919677734375, 0.907867431640625, 42.622650146484375, 16.290237426757812, -13.011871337890625, -15.380599975585938, -22.710163116455078, 0.1868133544921875, 5.8160400390625, -17.206787109375, 12.3271484375, -11.649555206298828, 16.594512939453125, -10.214296340942383, 16.751968383789062, 30.42047119140625, 3.720977783203125, 0.975830078125, 6.191650390625, 16.925811767578125, 96.35284423828125, -0.2580413818359375, 27.18902587890625, 30.114959716796875, 22.306976318359375, -25.39642333984375, 36.30353546142578, -2.33197021484375, -2.5533599853515625, 11.631317138671875, -19.64508056640625, 0.8089447021484375, 17.248748779296875, 4.069190979003906, 12.688812255859375, 37.58042907714844, -5.8838958740234375, -1.95074462890625, 0.0, 34.408294677734375, 13.021469116210938, -20.503173828125, 9.1883544921875, 38.202301025390625, 50.63594055175781, -33.25921630859375, -46.173736572265625, 54.906494140625, 13.847808837890625, 18.372833251953125, 2.81146240234375, 10.28143310546875, 23.3048095703125, 13.025680541992188, 31.78363037109375, 24.996490478515625, 52.90083312988281, 92.03814697265625, -15.0157470703125, 27.957015991210938, 1.0914764404296875, 15.24212646484375, 24.50946044921875, 38.772674560546875, 2.063568115234375, 0.7949047088623047, 10.576873779296875, 51.244598388671875, 6.267608642578125, 7.4397125244140625, 0.0, -18.36114501953125, 32.00440979003906, 39.293212890625, -10.382652282714844, 1.6082592010498047, -7.818756103515625, 9.484039306640625, -3.591583251953125, 55.794891357421875, -27.125457763671875, 43.84735107421875, 33.79750061035156, 27.581634521484375, -25.016342163085938, 15.081809997558594, -16.921600341796875, 17.827896118164062, -0.0650634765625, 65.28323364257812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000132.npy"} {"epoch": 0.2764397905759162, "step": 133, "batch_size": 128, "mean": 13.507908821105957, "std": 22.27564811706543, "min": -67.5667724609375, "p10": -11.912147521972651, "median": 12.050041198730469, "p90": 41.26946411132812, "max": 71.16314697265625, "pos_frac": 0.7734375, "sample": [20.882110595703125, 1.654937744140625, -34.02839660644531, 0.0, 1.4097137451171875, 33.097412109375, -1.0825996398925781, 13.982721328735352, 4.2725372314453125, 27.865280151367188, 18.2218017578125, 17.88970947265625, 5.32183837890625, 25.08489990234375, 38.893829345703125, 50.3946533203125, -10.124374389648438, -2.324951171875, -1.4589996337890625, -14.998092651367188, 33.06422424316406, 49.6158447265625, 71.16314697265625, 4.90020751953125, 25.291961669921875, 43.78802490234375, 22.01691436767578, 53.077728271484375, 0.0, 0.651580810546875, 21.31640625, 21.241226196289062, -3.3642654418945312, 2.932220458984375, 28.0078125, 32.37251281738281, 22.27313232421875, 25.651580810546875, -26.869964599609375, 32.63201904296875, 40.8326416015625, 6.25213623046875, -3.24755859375, -6.944915771484375, 10.1826171875, 1.278900146484375, -16.40949249267578, 8.6939697265625, 3.425994873046875, 8.24224853515625, 15.9620361328125, 40.94647216796875, 3.20587158203125, 30.550857543945312, 51.1141357421875, 8.237777709960938, 6.0323333740234375, 40.86468505859375, -4.643768310546875, 30.7476806640625, -22.62335205078125, 12.102783203125, 34.606536865234375, 2.73516845703125, 18.026575088500977, 8.261295318603516, 25.422454833984375, 42.950439453125, 31.537689208984375, 29.812576293945312, -35.94537353515625, 0.580535888671875, 11.997299194335938, 10.009475708007812, 49.9559326171875, 8.750110626220703, -4.702522277832031, -21.18255615234375, 3.797323226928711, 42.62603759765625, -6.923057556152344, 1.92388916015625, 9.381988525390625, -10.589599609375, 21.1455078125, -10.192237854003906, 7.864593505859375, 26.445960998535156, 24.57684326171875, 41.96343994140625, -5.3304595947265625, 46.234283447265625, 28.929443359375, 12.932525634765625, 14.291824340820312, 5.23284912109375, 7.89984130859375, 40.9720458984375, 24.801971435546875, 44.50056457519531, 9.686408996582031, 29.29266357421875, 22.47821807861328, 25.377086639404297, -27.87213134765625, 20.748321533203125, 19.49884033203125, 29.265121459960938, -25.704010009765625, 49.27703857421875, 21.7130126953125, 7.992698669433594, 6.5982666015625, -16.4129638671875, -49.291412353515625, 17.417221069335938, -67.5667724609375, 4.8589324951171875, 3.0087661743164062, 38.0625, 5.160675048828125, 24.78570556640625, 36.112937927246094, -5.4761199951171875, -22.954925537109375, 2.018707275390625, 26.041656494140625, 40.074249267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000133.npy"} {"epoch": 0.27853403141361255, "step": 134, "batch_size": 128, "mean": 9.359418869018555, "std": 21.22076988220215, "min": -42.63262939453125, "p10": -15.986801147460938, "median": 7.7456464767456055, "p90": 33.606607818603514, "max": 93.84054565429688, "pos_frac": 0.65625, "sample": [39.28022766113281, 26.300125122070312, -16.17767333984375, -5.078948974609375, -25.721298217773438, -12.9114990234375, 8.66632080078125, -16.976715087890625, -4.90771484375, 25.6309814453125, -40.6522216796875, -6.309173583984375, 11.3424072265625, 17.29302978515625, 27.591781616210938, -1.0474185943603516, -5.156332015991211, 9.532752990722656, 21.734619140625, 5.884967803955078, 11.51959228515625, 60.368438720703125, 0.0, 20.16461181640625, 24.53704833984375, 21.241058349609375, 31.47869873046875, -3.5220108032226562, 5.32232666015625, 46.722900390625, -29.27240562438965, -5.64300537109375, 16.98956298828125, 2.6067123413085938, 29.11574935913086, 2.8231964111328125, 26.492218017578125, 15.136672973632812, 15.746124267578125, 5.575286865234375, -10.756744384765625, 24.959068298339844, 13.383377075195312, 26.97979736328125, 93.84054565429688, 35.253570556640625, 0.378875732421875, -12.0325927734375, 69.37088012695312, 28.147003173828125, -23.466888427734375, 34.25762939453125, 10.6158447265625, -13.870155334472656, -0.024169921875, 1.7704391479492188, 30.516326904296875, 21.781356811523438, 39.86952209472656, 8.790550231933594, -10.879638671875, -2.4655914306640625, 9.370941162109375, 6.67047119140625, -26.82086181640625, 40.68995666503906, -2.6977996826171875, 15.686553955078125, 25.2769775390625, 10.508743286132812, 2.3578052520751953, 16.98529052734375, -17.78199005126953, -4.521820068359375, -0.212738037109375, 41.95042419433594, 11.03887939453125, 26.39202880859375, 8.285608291625977, -33.53788757324219, 17.173828125, 32.240753173828125, 28.129669189453125, -10.9266357421875, -0.6195068359375, -2.429229736328125, 10.448554992675781, 14.317907333374023, 44.349395751953125, 1.3018779754638672, -1.2472686767578125, -5.28216552734375, 9.854480743408203, -0.192047119140625, -42.63262939453125, 27.47858428955078, 21.144561767578125, 13.722412109375, 6.190696716308594, -8.9869384765625, 5.988658905029297, 12.389434814453125, -15.904998779296875, -38.389007568359375, -17.056320190429688, 20.315673828125, -16.437744140625, 0.431304931640625, 35.285064697265625, 1.0477294921875, 7.205684661865234, 32.241065979003906, 0.6194000244140625, 20.0623779296875, 40.1146240234375, 26.009449005126953, 33.327598571777344, 4.911895751953125, 3.709136962890625, -7.86907958984375, -2.13385009765625, 0.235870361328125, -3.9916458129882812, -6.69403076171875, 33.011016845703125, 19.386411666870117, 4.373039245605469, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000134.npy"} {"epoch": 0.2806282722513089, "step": 135, "batch_size": 128, "mean": 13.265047073364258, "std": 20.877321243286133, "min": -49.906494140625, "p10": -9.597443771362304, "median": 10.263824462890625, "p90": 44.23398132324218, "max": 60.38287353515625, "pos_frac": 0.78125, "sample": [0.3579425811767578, -26.575103759765625, 11.215408325195312, 23.587738037109375, -9.208560943603516, 42.683197021484375, 0.7852954864501953, 15.1514892578125, 30.880630493164062, 3.1056060791015625, 52.464508056640625, 17.100982666015625, -7.02191162109375, -3.2023773193359375, 10.35565185546875, 6.3651123046875, 8.7711181640625, 25.4068603515625, 25.365829467773438, 3.1634902954101562, 1.7090911865234375, 6.964977264404297, -0.8754425048828125, 2.88482666015625, -29.2366943359375, -0.0338897705078125, 32.738494873046875, 26.20105743408203, 18.43267059326172, 47.65252685546875, 33.93353271484375, 0.3664817810058594, 31.850860595703125, 4.463829040527344, -17.89875030517578, 37.744903564453125, 3.68212890625, 15.77630615234375, 28.522064208984375, 7.62969970703125, 4.33270263671875, 12.398773193359375, 13.14825439453125, 36.851287841796875, 59.18951416015625, 18.88886260986328, 44.695556640625, 10.1719970703125, 1.3826904296875, -1.93548583984375, 18.410751342773438, 20.006820678710938, 2.2433853149414062, -10.504837036132812, -18.187423706054688, -7.840301513671875, 44.036163330078125, 6.182403564453125, 40.853485107421875, 18.19635009765625, 23.521324157714844, 17.337554931640625, 32.3216552734375, -5.493499755859375, -3.5730209350585938, 11.135391235351562, -2.49365234375, 20.60784912109375, 18.379425048828125, 3.402374267578125, 16.974624633789062, 6.9457244873046875, 50.379974365234375, 16.484375, 34.162147521972656, 31.33301544189453, 16.284881591796875, 5.74493408203125, -5.78009033203125, 50.422760009765625, 18.871536254882812, -18.033950805664062, -0.6434135437011719, 1.27587890625, 7.816070556640625, 50.42198181152344, 1.029998779296875, 52.2564697265625, 37.62005615234375, 49.26678466796875, 20.81610107421875, 39.3585205078125, 2.953826904296875, -39.90948486328125, 45.96026611328125, 6.7201080322265625, 6.480873107910156, 7.502939224243164, 1.54205322265625, 24.325042724609375, 8.609375, 13.02825927734375, 31.212486267089844, 60.38287353515625, 2.8076820373535156, 0.113494873046875, 4.6505126953125, 14.219917297363281, 47.454193115234375, 59.428466796875, -16.8182373046875, -15.013504028320312, 5.2298583984375, 10.905914306640625, 0.0, 20.957855224609375, -18.9632568359375, -49.906494140625, -0.8360977172851562, 10.620231628417969, -5.78619384765625, 20.038833618164062, 5.5238494873046875, 43.925201416015625, 28.20379638671875, -14.041915893554688, -17.830810546875, 0.297821044921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000135.npy"} {"epoch": 0.28272251308900526, "step": 136, "batch_size": 128, "mean": 12.510882377624512, "std": 23.127172470092773, "min": -56.410736083984375, "p10": -9.488161468505858, "median": 10.55545425415039, "p90": 40.661914062499996, "max": 82.30023193359375, "pos_frac": 0.765625, "sample": [44.5870361328125, 18.40264129638672, 1.6455535888671875, 14.206283569335938, 34.36638641357422, 39.8446044921875, 45.970123291015625, 11.49664306640625, 34.672637939453125, 61.008148193359375, 6.277442932128906, 3.5686378479003906, 42.5689697265625, 6.529991149902344, 25.39340591430664, 38.38188171386719, 13.799514770507812, 39.35939025878906, 10.749755859375, 0.451446533203125, -43.78285217285156, 35.812583923339844, 6.82586669921875, 24.39781951904297, -18.9385986328125, 3.67608642578125, -6.833465576171875, -41.490692138671875, 7.6741485595703125, 12.602813720703125, 4.794647216796875, -0.36676025390625, -5.966552734375, 8.415435791015625, 25.093528747558594, -27.9737548828125, -8.8065185546875, 28.235748291015625, 13.752365112304688, -5.754364013671875, 27.024398803710938, -20.801681518554688, 59.19293212890625, -56.410736083984375, 11.9395751953125, -6.546669006347656, 3.3487548828125, 4.0978851318359375, 15.897705078125, -5.042449951171875, 19.84686279296875, -35.71026611328125, 43.50236511230469, 16.80743408203125, 32.31488037109375, 71.33642578125, -8.955322265625, 7.3828887939453125, 7.95526123046875, 32.22596740722656, 42.6728515625, 21.15447998046875, 36.350128173828125, -12.726299285888672, 10.688362121582031, -0.573394775390625, -2.4835662841796875, 9.2203369140625, 34.01947021484375, -2.878082275390625, 2.6534881591796875, 14.143798828125, 7.605621337890625, 8.712982177734375, 4.2057342529296875, 53.51385498046875, -1.5702323913574219, 13.679718017578125, 11.25006103515625, -8.302337646484375, 1.23065185546875, -26.513748168945312, 24.52154541015625, 57.638031005859375, 36.774658203125, 24.7369384765625, 1.6794700622558594, 17.482044219970703, -6.915307998657227, 46.57330322265625, 18.352752685546875, 10.42254638671875, -10.731452941894531, 12.75, 5.31207275390625, 1.5017852783203125, 56.58154296875, 14.285438537597656, 28.728843688964844, 22.723434448242188, -0.028181076049804688, 5.536357879638672, 34.89166259765625, 1.0614013671875, 31.28839111328125, -12.859664916992188, -8.242263793945312, 2.9161834716796875, 34.756103515625, 38.971771240234375, 36.21697998046875, 10.753412246704102, -21.6688232421875, -6.115570068359375, 4.2581787109375, 8.83734130859375, 82.30023193359375, 4.499900817871094, 2.01708984375, 9.976348876953125, 12.981201171875, 14.9442138671875, 11.778129577636719, 2.9040679931640625, 13.961799621582031, -51.174285888671875, 30.239898681640625, 3.8653640747070312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000136.npy"} {"epoch": 0.2848167539267016, "step": 137, "batch_size": 128, "mean": 11.119251251220703, "std": 21.839317321777344, "min": -48.39044189453125, "p10": -15.53371238708496, "median": 8.007904052734375, "p90": 36.64623260498047, "max": 75.01052856445312, "pos_frac": 0.75, "sample": [-30.1236572265625, 2.4051513671875, 35.76275634765625, -5.9700775146484375, 11.93634033203125, 20.17156982421875, 39.08399963378906, 12.095794677734375, 3.681610107421875, 2.3783721923828125, 18.858192443847656, 15.144943237304688, 25.9691162109375, 55.888153076171875, 17.862869262695312, 5.149688720703125, -0.1938629150390625, 31.58795166015625, 1.68701171875, -14.110992431640625, 29.449295043945312, 19.220733642578125, 6.442352294921875, 11.257720947265625, -43.39453125, 21.565139770507812, 2.5628814697265625, -16.5809326171875, 29.60869598388672, 7.575019836425781, 48.018829345703125, 15.684036254882812, -2.9956207275390625, 38.5, -17.93133544921875, -1.1639404296875, 2.545989990234375, 7.2831268310546875, 0.865966796875, -9.484954833984375, 8.4755859375, -4.607585906982422, -19.879669189453125, 0.737579345703125, -48.39044189453125, 5.664276123046875, 7.306022644042969, 14.788681030273438, -45.48454284667969, 16.302001953125, 7.76031494140625, 45.014007568359375, 7.535797119140625, 11.9735107421875, 15.629425048828125, 34.62608337402344, -17.58056640625, 38.25152587890625, -16.79547119140625, 2.0966796875, 6.555206298828125, -0.8266067504882812, 26.74224853515625, -1.0240631103515625, 14.1558837890625, 57.75030517578125, -28.057098388671875, 25.91717529296875, 7.723968505859375, -2.383697509765625, 1.452880859375, 23.62548828125, 30.4217529296875, 12.278282165527344, 34.23081588745117, 1.7789154052734375, 1.0632266998291016, 75.01052856445312, 28.844940185546875, 16.454238891601562, 6.8729400634765625, 58.27197265625, -1.1118850708007812, 36.24676513671875, 16.666473388671875, 20.7999267578125, 53.70458984375, 17.940689086914062, -4.421195983886719, 36.698944091796875, 1.1119461059570312, -11.910606384277344, -1.2255325317382812, 6.537109375, 1.7171401977539062, 63.831329345703125, 24.12481689453125, 3.131000518798828, 17.78985595703125, 16.001205444335938, -36.86041259765625, -30.6158447265625, 7.73089599609375, -32.59748840332031, -9.275352478027344, 51.38291931152344, -15.084903717041016, 29.89202880859375, 36.62364196777344, 33.151458740234375, 24.750732421875, 32.29203796386719, -1.2835693359375, 22.275970458984375, 14.371337890625, 23.742961883544922, -3.1455230712890625, 5.987449645996094, 0.88348388671875, 6.7748870849609375, 8.2554931640625, 20.2603759765625, 5.159210205078125, 12.962905883789062, 14.76873779296875, 19.95947265625, 29.476959228515625, -4.784263610839844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000137.npy"} {"epoch": 0.2869109947643979, "step": 138, "batch_size": 128, "mean": 11.132911682128906, "std": 24.576322555541992, "min": -46.723358154296875, "p10": -17.768994140625, "median": 8.491329193115234, "p90": 41.56491088867187, "max": 81.96670532226562, "pos_frac": 0.6796875, "sample": [69.69036865234375, 41.1751708984375, 43.12890625, 17.0908203125, -11.736946105957031, 5.751190185546875, -6.4056396484375, 68.73239135742188, 6.924072265625, 23.270538330078125, 10.259567260742188, 10.84307861328125, 2.5061492919921875, 12.231826782226562, 12.430274963378906, -31.0308837890625, -17.638641357421875, 28.9625244140625, -11.562408447265625, -28.310821533203125, 20.959548950195312, 4.78369140625, -12.29876708984375, -7.54876708984375, -46.723358154296875, 25.423782348632812, -3.8262176513671875, 35.12071228027344, 2.4048385620117188, 81.96670532226562, -8.0594482421875, 7.0887451171875, 3.9493408203125, 42.47430419921875, 1.1975631713867188, 26.435958862304688, -26.696792602539062, 10.4671630859375, 40.5054931640625, -1.4508132934570312, 11.086456298828125, -2.291227340698242, 30.585845947265625, 35.76336669921875, 16.614990234375, 32.4769287109375, 6.343719482421875, 3.9791259765625, 14.647918701171875, -14.9644775390625, 12.632369995117188, 1.3809890747070312, 38.341766357421875, -17.223037719726562, 44.307403564453125, 11.803497314453125, 38.746917724609375, -0.311737060546875, 20.712627410888672, -18.719970703125, -18.073150634765625, 2.203277587890625, -1.66961669921875, -0.9354400634765625, 20.639862060546875, -29.037582397460938, 67.2052001953125, 31.349655151367188, 27.881103515625, -2.1851806640625, -29.15869140625, 16.952880859375, -5.27703857421875, 54.576568603515625, 15.871994018554688, -9.924118041992188, -3.9936370849609375, 21.667694091796875, 12.74114990234375, 1.7589874267578125, 14.254470825195312, 33.8863525390625, 27.867721557617188, 33.61810302734375, -39.33770751953125, -11.93304443359375, 12.971466064453125, 0.7325057983398438, -14.449996948242188, -18.223419189453125, 27.597488403320312, 56.307891845703125, 44.61802673339844, 2.38623046875, -2.559926986694336, 26.70806884765625, 26.886886596679688, 25.011016845703125, 0.940582275390625, -4.777923583984375, 28.759368896484375, -15.253875732421875, 72.9984130859375, -5.410594940185547, 8.692649841308594, 16.94439697265625, -7.357208251953125, -16.76605224609375, 4.7395172119140625, 39.683937072753906, -19.43817138671875, 3.844451904296875, 1.1497726440429688, 18.610458374023438, 27.262054443359375, 2.614269256591797, 57.480682373046875, -33.734344482421875, 33.7987060546875, -38.764068603515625, 4.456043243408203, 21.759063720703125, 3.947967529296875, 19.804244995117188, 49.45538330078125, 26.970504760742188, 8.290008544921875, -13.022201538085938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000138.npy"} {"epoch": 0.28900523560209423, "step": 139, "batch_size": 128, "mean": 8.212000846862793, "std": 24.184280395507812, "min": -45.949371337890625, "p10": -22.19134063720703, "median": 5.958518981933594, "p90": 41.178506469726564, "max": 68.52813720703125, "pos_frac": 0.625, "sample": [-36.74394226074219, -3.8332977294921875, -29.746917724609375, 7.3375396728515625, -30.377197265625, 14.4964599609375, 31.605506896972656, 21.3529052734375, -2.662322998046875, -11.30710220336914, 51.20599365234375, 3.2867889404296875, 21.04669189453125, 11.321746826171875, -26.967193603515625, 45.32928466796875, 53.49102783203125, 10.866058349609375, -14.584068298339844, 10.003662109375, 5.417083740234375, 23.965667724609375, 30.646041870117188, -15.750732421875, 12.3955078125, -12.857864379882812, 5.888206481933594, 35.51148986816406, -6.948646545410156, -45.949371337890625, 0.6190528869628906, -7.151885986328125, 37.22709655761719, 40.959869384765625, 25.42828369140625, 0.1540393829345703, 11.627143859863281, -9.247314453125, 2.84600830078125, 29.105133056640625, -2.5320816040039062, -20.395339965820312, 1.381591796875, 21.066680908203125, 13.030357360839844, -6.906982421875, -39.41851806640625, 0.07257080078125, -11.836471557617188, 39.7769775390625, -10.2213134765625, -0.4742431640625, -22.81201171875, -21.925338745117188, 7.814781188964844, 42.04132080078125, -1.222198486328125, -0.0733795166015625, -25.04736328125, -19.01068115234375, 41.68865966796875, -13.249073028564453, 43.79197692871094, 24.35008430480957, 27.386260986328125, 15.64508056640625, 0.5274524688720703, 1.1005096435546875, 39.203369140625, 22.547409057617188, -16.036094665527344, 19.417713165283203, 13.30535888671875, 42.58522033691406, -3.8451766967773438, -10.872650146484375, 5.675628662109375, 68.52813720703125, -11.133544921875, 40.368804931640625, 24.15607452392578, 22.5299072265625, 14.856246948242188, 51.70759582519531, 4.8372802734375, -14.946342468261719, 53.624755859375, -38.5142822265625, 19.216873168945312, -4.783580780029297, 28.695281982421875, 46.68853759765625, 20.10186767578125, 14.2860107421875, -18.681732177734375, 14.13739013671875, -0.295379638671875, -40.16188049316406, -5.9618377685546875, 2.1574344635009766, 24.6947021484375, 7.218963623046875, -8.452590942382812, 3.49560546875, 13.088958740234375, -10.622711181640625, 27.327667236328125, 67.61337280273438, 10.295310974121094, 58.585601806640625, 40.163299560546875, 18.533447265625, 3.718353271484375, -34.318634033203125, 7.2851104736328125, -33.23829650878906, 0.0, 5.7316436767578125, 14.04534912109375, 33.21236801147461, -7.924583435058594, -31.415771484375, 19.754539489746094, -5.0423583984375, 20.75872802734375, -13.3009033203125, 36.949981689453125, 6.028831481933594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000139.npy"} {"epoch": 0.29109947643979056, "step": 140, "batch_size": 128, "mean": 14.967662811279297, "std": 22.448715209960938, "min": -43.5142822265625, "p10": -10.204063415527342, "median": 10.383228302001953, "p90": 46.34557952880859, "max": 76.697509765625, "pos_frac": 0.7109375, "sample": [-6.889617919921875, 24.674957275390625, 32.532806396484375, -9.929725646972656, -10.844184875488281, 49.40570068359375, -5.710563659667969, 40.60591125488281, -3.300018310546875, -0.647918701171875, -15.801475524902344, 11.500625610351562, 30.22076416015625, 21.27983856201172, 5.7982177734375, 10.105438232421875, 2.34783935546875, 19.782928466796875, 46.15765380859375, 10.3623046875, 7.649620056152344, -17.735641479492188, 28.47130584716797, -27.345947265625, 21.238082885742188, 39.2669677734375, -4.139556884765625, 7.3107147216796875, 43.75709533691406, 41.665504455566406, 29.20648193359375, 31.850875854492188, 65.18389892578125, 38.415283203125, -9.51202392578125, 55.898712158203125, -2.27777099609375, 4.697235107421875, 48.63200378417969, 50.37034606933594, -1.5981063842773438, -25.545028686523438, 10.9537353515625, 0.3007087707519531, -2.087066650390625, 37.9766845703125, 51.16912841796875, 9.871734619140625, -8.334014892578125, 76.697509765625, 9.650459289550781, 23.99261474609375, 31.377044677734375, 11.44818115234375, 2.6899261474609375, 22.079299926757812, 22.102569580078125, 5.737323760986328, -8.25018310546875, 22.782638549804688, 27.378265380859375, 0.9275894165039062, 4.130928039550781, 21.34466552734375, 5.7468414306640625, 0.0334014892578125, -3.5395736694335938, -11.04248046875, 0.0, -43.5142822265625, -0.9806976318359375, -26.923095703125, 66.35665893554688, 0.9813995361328125, 4.1796417236328125, 14.734916687011719, 24.036041259765625, 32.8333740234375, 3.5325241088867188, 46.78407287597656, 10.404151916503906, 3.596588134765625, 0.07527351379394531, 42.77630615234375, 50.664825439453125, 35.55600357055664, 30.14666748046875, -2.7647228240966797, 34.64418029785156, 37.79710388183594, 5.983884811401367, 7.576301574707031, 17.77660369873047, -3.3166046142578125, 29.24634552001953, 31.011489868164062, 0.0, -6.181947708129883, 18.6475830078125, -1.0398788452148438, -7.5283203125, 49.2423095703125, -16.09136962890625, 46.12255859375, 9.676193237304688, -18.260818481445312, 42.0660400390625, 18.889022827148438, 22.956802368164062, 23.34185791015625, 44.338104248046875, 22.304779052734375, 51.12860107421875, 20.658206939697266, 27.492996215820312, -8.410552978515625, -2.3212127685546875, -14.27117919921875, 7.897918701171875, -24.43950653076172, 12.405364990234375, 55.224365234375, 9.238311767578125, -2.98907470703125, 6.181304931640625, -14.97711181640625, 27.83880615234375, 21.276336669921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000140.npy"} {"epoch": 0.2931937172774869, "step": 141, "batch_size": 128, "mean": 14.336434364318848, "std": 25.244476318359375, "min": -75.11483764648438, "p10": -10.9705810546875, "median": 12.62304973602295, "p90": 46.53515014648437, "max": 88.59771728515625, "pos_frac": 0.75, "sample": [40.634735107421875, 16.96295166015625, -24.910839080810547, -9.733917236328125, 6.8660888671875, 13.176673889160156, -7.620758056640625, 21.827980041503906, -48.5067138671875, 88.59771728515625, 5.384033203125, 11.17535400390625, 30.68602752685547, 32.700775146484375, 16.59173583984375, 44.284820556640625, 26.532272338867188, 18.575576782226562, 18.223081588745117, 23.604156494140625, 56.40362548828125, -4.90301513671875, 1.6270523071289062, 43.26875305175781, 0.12411880493164062, -13.679445266723633, 38.95614242553711, 12.272560119628906, 0.8867034912109375, 49.41363525390625, 32.94158935546875, -18.016143798828125, -10.739959716796875, 69.923095703125, 3.71929931640625, 19.18756103515625, -25.159629821777344, -9.439085006713867, 10.012039184570312, -19.429779052734375, 46.02880859375, 36.47039794921875, 12.973539352416992, 18.623809814453125, -6.5405426025390625, 9.60150146484375, 1.828125, -7.592124938964844, 58.679656982421875, 8.782684326171875, 13.137893676757812, 25.327301025390625, 3.0957679748535156, -9.196929931640625, 29.73760986328125, 22.055419921875, 1.2541275024414062, 1.5091476440429688, 24.95904541015625, 0.983001708984375, 35.4691162109375, 34.12151336669922, -14.84063720703125, -44.18626403808594, -5.99462890625, 68.64154052734375, -2.233530044555664, 25.917007446289062, -5.65472412109375, 37.48527526855469, 4.2635650634765625, 12.126922607421875, 6.694049835205078, 23.430458068847656, 29.2681884765625, 35.304893493652344, 32.430419921875, -7.552978515625, 0.0, -14.817436218261719, 40.74528503417969, -1.700653076171875, -0.293548583984375, 58.8944091796875, 5.87994384765625, 54.0860595703125, -75.11483764648438, 15.8433837890625, 19.400054931640625, -1.3123626708984375, 25.24090576171875, 18.590423583984375, 34.150978088378906, -44.625091552734375, -6.834869384765625, 17.044845581054688, 0.8115005493164062, 56.1083984375, 0.286529541015625, 32.367210388183594, -6.5057525634765625, 11.20196533203125, 19.277130126953125, 39.5704345703125, 15.8743896484375, 4.035131454467773, 23.9024658203125, 47.71661376953125, 41.775482177734375, -11.508697509765625, 7.835971832275391, 13.47552490234375, 57.54290771484375, -37.67303466796875, 1.8458099365234375, 9.668197631835938, 10.921661376953125, 11.855499267578125, 18.85919189453125, 20.629112243652344, -7.635993957519531, 47.95562744140625, 43.3819580078125, 7.98870849609375, 11.632377624511719, 58.02728271484375, 15.83502197265625, 3.9982528686523438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000141.npy"} {"epoch": 0.29528795811518327, "step": 142, "batch_size": 128, "mean": 14.066268920898438, "std": 25.35465431213379, "min": -65.97012329101562, "p10": -14.238179016113277, "median": 9.491111755371094, "p90": 48.638798522949216, "max": 79.83291625976562, "pos_frac": 0.6875, "sample": [-2.4783477783203125, -2.58428955078125, 9.318798065185547, 27.4912109375, 2.231138229370117, -17.665771484375, 11.26544189453125, -1.26470947265625, 73.10836791992188, 38.99577331542969, 43.224639892578125, -3.582794189453125, -6.538421630859375, 35.329193115234375, 17.763824462890625, 28.990325927734375, 0.00649261474609375, -4.275360107421875, 54.01300048828125, 26.513473510742188, 0.600738525390625, 25.463973999023438, 0.0, 20.292381286621094, 21.14605712890625, 49.165496826171875, 10.67474365234375, 3.6125030517578125, -2.8062820434570312, 0.7581787109375, -2.72039794921875, 28.550811767578125, -2.703411102294922, -23.294647216796875, 3.0905914306640625, 3.0801773071289062, 23.7803955078125, 14.316871643066406, -25.840896606445312, 26.875213623046875, 19.37127685546875, 50.59454345703125, 44.448486328125, 17.602989196777344, 16.98004150390625, -11.006240844726562, -1.1527423858642578, 45.14631652832031, 20.096023559570312, 69.62457275390625, 13.83404541015625, -6.251373291015625, 4.749786376953125, 13.6678466796875, -12.769210815429688, 3.014049530029297, -0.711578369140625, 34.81047058105469, -18.4827880859375, -19.91070556640625, 20.28363037109375, 31.90765380859375, -10.364410400390625, -7.924407958984375, 48.41307067871094, 37.48478698730469, 18.22625732421875, 0.18638229370117188, 45.62084197998047, -3.7765045166015625, 7.884346008300781, 20.970077514648438, 1.794097900390625, 6.175220489501953, -23.387283325195312, 31.036712646484375, -39.23896789550781, 28.436058044433594, 42.3475341796875, -25.072509765625, -5.46929931640625, 69.18093872070312, 79.83291625976562, -1.1729736328125, -2.2674331665039062, 13.424560546875, 9.617889404296875, 51.39178466796875, 9.364334106445312, 5.847114562988281, -2.9791908264160156, -3.6436805725097656, 33.331756591796875, 41.78211975097656, 56.19940185546875, 5.74322509765625, 36.547149658203125, 26.233123779296875, 21.69244384765625, 54.526947021484375, -65.97012329101562, 53.13218688964844, -30.8096923828125, 1.1243553161621094, -0.8080883026123047, -27.339080810546875, 35.82477569580078, 18.429824829101562, 32.00201416015625, 0.23357391357421875, 36.220703125, -36.360198974609375, 3.9795761108398438, 5.18536376953125, 21.979652404785156, -8.3709716796875, 29.630950927734375, -1.449747085571289, 4.731964111328125, 2.761016845703125, -20.859527587890625, 62.62164306640625, 0.25452613830566406, -2.003662109375, 30.8909912109375, 29.788009643554688, 66.75086975097656, 41.16551971435547], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000142.npy"} {"epoch": 0.2973821989528796, "step": 143, "batch_size": 128, "mean": 16.066205978393555, "std": 27.57000160217285, "min": -80.81378173828125, "p10": -10.854431152343748, "median": 13.204254150390625, "p90": 47.89609069824218, "max": 100.1934814453125, "pos_frac": 0.7578125, "sample": [26.198139190673828, 48.262176513671875, 55.445648193359375, 31.09918212890625, 40.936134338378906, 15.862060546875, 29.426239013671875, 9.254226684570312, -18.343948364257812, -7.905790328979492, 29.409423828125, 0.18280029296875, 21.4600830078125, 12.851943969726562, 2.568897247314453, -11.51409912109375, 47.73919677734375, 4.0760498046875, -19.677078247070312, 100.1934814453125, -20.097412109375, -7.3609619140625, 44.64283752441406, 46.119537353515625, 22.46489715576172, 44.05021667480469, 35.382080078125, 62.4505615234375, 87.94281005859375, 0.762664794921875, -2.43902587890625, 42.54624938964844, 28.24322509765625, 32.3690071105957, 25.94611358642578, -7.73345947265625, -80.81378173828125, -1.1855621337890625, 28.409194946289062, 9.722808837890625, 9.335563659667969, -30.84832000732422, 10.158004760742188, 39.806671142578125, 5.235069274902344, 6.17376708984375, 9.686151504516602, 21.966751098632812, 15.97341537475586, -66.04449462890625, 6.6546630859375, 28.010009765625, 3.8487091064453125, 38.78875732421875, 18.4788818359375, 22.5828857421875, -1.69500732421875, 21.964996337890625, -13.972152709960938, -0.38687896728515625, -9.8182373046875, 0.8670921325683594, 68.49880981445312, 19.452239990234375, 5.383258819580078, -1.819580078125, 59.19708251953125, 32.19590759277344, 38.199951171875, 2.701507568359375, 30.332855224609375, 35.171295166015625, -9.021919250488281, 3.8531665802001953, 33.1915283203125, -10.57171630859375, -39.46026611328125, 1.3274765014648438, 12.043548583984375, 13.625732421875, 3.83819580078125, -1.675537109375, 8.241043090820312, 17.497299194335938, 11.4927978515625, -3.337738037109375, -17.010162353515625, 2.549468994140625, 0.9047164916992188, 13.556564331054688, 34.8095588684082, 15.733856201171875, -2.3386688232421875, 23.230850219726562, 19.59979248046875, 24.383895874023438, 0.7504692077636719, 16.259490966796875, -4.585689544677734, 2.1060791015625, 81.7479248046875, 30.91009521484375, -4.3242340087890625, 11.912925720214844, -34.360076904296875, 75.3529052734375, 33.73619842529297, 3.2806854248046875, 2.837656021118164, -2.160247802734375, 27.858978271484375, 56.072509765625, 28.303939819335938, 67.74868774414062, 27.84906005859375, 62.08367919921875, 43.035614013671875, 43.677947998046875, -39.429931640625, 25.543148040771484, -16.5238037109375, 51.1217041015625, -6.76629638671875, 9.47589111328125, 46.020416259765625, 8.607177734375, 15.74859619140625, 1.093027114868164], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000143.npy"} {"epoch": 0.2994764397905759, "step": 144, "batch_size": 128, "mean": 13.333430290222168, "std": 25.159290313720703, "min": -67.29981994628906, "p10": -15.722161865234375, "median": 6.671337127685547, "p90": 45.548089599609376, "max": 90.77133178710938, "pos_frac": 0.7265625, "sample": [-1.8190231323242188, 18.46393585205078, 43.805145263671875, 5.175392150878906, 5.330047607421875, 10.23834228515625, -21.038375854492188, -9.806961059570312, 1.4949684143066406, 90.77133178710938, 1.72265625, -15.71075439453125, 6.729515075683594, 58.13433837890625, 4.7027587890625, 33.429962158203125, 2.1585617065429688, 45.28532409667969, 9.283935546875, -0.363433837890625, 2.61322021484375, 22.06683349609375, -15.748779296875, 3.77752685546875, 1.041259765625, -4.99859619140625, -21.359527587890625, -2.9025917053222656, 26.929641723632812, 5.1608123779296875, -38.3946533203125, 6.096546173095703, -0.23382568359375, 25.02099609375, 50.801666259765625, 39.11958312988281, 31.433563232421875, 68.5137939453125, 8.356361389160156, 59.216766357421875, -1.0441398620605469, -9.060821533203125, 5.875175476074219, 21.27728271484375, -0.7527542114257812, -18.076583862304688, 7.75286865234375, 33.970428466796875, -18.05615234375, 21.827056884765625, 13.354423522949219, 56.564971923828125, 37.62786865234375, -17.46466064453125, -67.29981994628906, -27.59381103515625, 16.00897216796875, -11.472923278808594, -9.760528564453125, 2.919036865234375, 34.759857177734375, 0.320953369140625, -8.216217041015625, 5.562408447265625, 28.670310974121094, 13.5323486328125, 68.32904052734375, 63.520263671875, 38.58172607421875, 39.2371826171875, 1.404632568359375, 19.039718627929688, 45.9508056640625, -27.350341796875, 22.51458740234375, 27.45452880859375, 4.692718505859375, 10.118301391601562, 45.52459716796875, 42.973388671875, 45.20343017578125, 37.64728927612305, 5.9210052490234375, -8.01123046875, 14.368865966796875, 25.844436645507812, -2.58428955078125, 7.562042236328125, 2.2841644287109375, 32.38330078125, 70.03619384765625, 32.18658447265625, 12.56072998046875, 0.6239089965820312, 11.787284851074219, 46.6170654296875, 2.8597412109375, -2.16778564453125, 6.282379150390625, 35.8807373046875, 7.223876953125, 0.09409141540527344, 0.0, -0.6922149658203125, -13.067855834960938, -12.073867797851562, 4.68572998046875, -34.224029541015625, 25.8101806640625, 24.701091766357422, 18.832275390625, 0.32110595703125, 42.727996826171875, -7.2694091796875, 27.625885009765625, 1.5230712890625, 19.089630126953125, 6.527473449707031, 16.87042236328125, -25.832366943359375, 13.35491943359375, 36.176666259765625, 6.6131591796875, 45.6029052734375, -18.337310791015625, -0.76458740234375, 72.7098388671875, 3.4496002197265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000144.npy"} {"epoch": 0.30157068062827225, "step": 145, "batch_size": 128, "mean": 17.32861328125, "std": 26.165180206298828, "min": -52.64202880859375, "p10": -14.031126403808594, "median": 15.290306091308594, "p90": 52.78699951171874, "max": 83.11041259765625, "pos_frac": 0.7109375, "sample": [-14.220428466796875, 38.23566436767578, -41.7132568359375, 36.92816162109375, 13.480249404907227, -28.17352294921875, 41.77978515625, 19.17730712890625, 80.29312133789062, 25.37982177734375, 19.638275146484375, 12.0435791015625, 23.44189453125, 33.36308288574219, 29.645957946777344, 63.240966796875, 4.40484619140625, 30.854949951171875, 8.63525390625, 5.8435821533203125, 4.3999176025390625, -7.173431396484375, 49.2901611328125, -9.367233276367188, -8.283531188964844, 67.16485595703125, -27.939666748046875, 51.98931884765625, 0.125, 21.140167236328125, 1.76727294921875, 31.15923309326172, 9.195991516113281, 35.85888671875, 54.64825439453125, 22.370697021484375, 76.32394409179688, -52.64202880859375, 8.648040771484375, -0.19268798828125, 19.12408447265625, 18.6822509765625, -15.550544738769531, 65.38662719726562, 69.63327026367188, 15.929168701171875, 36.303314208984375, 33.2034912109375, 10.837127685546875, 21.49237060546875, 17.55445098876953, -6.8417510986328125, -5.249488830566406, 33.17121124267578, 46.20454406738281, 44.01307678222656, 0.716400146484375, -4.395408630371094, -14.629730224609375, 3.7125396728515625, 14.377044677734375, 6.83807373046875, -7.20880126953125, 42.481201171875, 64.85736083984375, -5.0111541748046875, 57.32867431640625, -10.773406982421875, 55.097808837890625, 31.611204147338867, 54.902984619140625, -10.41168212890625, 5.395698547363281, 34.486724853515625, 47.0528564453125, 10.181549072265625, 1.1944999694824219, 11.331024169921875, 26.700912475585938, -19.918182373046875, -20.8287353515625, 5.027740478515625, 9.870758056640625, 29.94830322265625, 83.11041259765625, 22.674705505371094, 5.7996673583984375, 25.121490478515625, -23.23894500732422, -12.928779602050781, -2.7099533081054688, 0.0, 6.9249114990234375, 19.79754638671875, 9.640739440917969, 3.5531387329101562, 41.42024230957031, 70.622314453125, 41.19221496582031, 24.313217163085938, -2.0676422119140625, -9.235580444335938, 26.736236572265625, 24.40392303466797, 14.651443481445312, -14.567718505859375, 25.576553344726562, 29.19488525390625, -2.568187713623047, -1.0302066802978516, -15.84576416015625, 49.436981201171875, 26.60639190673828, 42.11767578125, -2.521759033203125, -1.8570556640625, 4.261039733886719, 30.26314926147461, 43.777008056640625, 48.598175048828125, 33.886077880859375, -13.949996948242188, -1.8022956848144531, -4.096626281738281, 27.19464111328125, -21.0484619140625, -8.162673950195312, 20.227386474609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000145.npy"} {"epoch": 0.3036649214659686, "step": 146, "batch_size": 128, "mean": 13.794811248779297, "std": 25.24655532836914, "min": -61.914642333984375, "p10": -16.057464599609375, "median": 8.424051284790039, "p90": 44.98786315917968, "max": 79.3291015625, "pos_frac": 0.734375, "sample": [5.117793083190918, 2.7245559692382812, 18.401294708251953, -33.70263671875, 12.441802978515625, 39.8074951171875, -26.731369018554688, 11.815673828125, 33.750694274902344, -25.512969970703125, 79.3291015625, 75.43527221679688, 11.0245361328125, -8.501983642578125, 17.592620849609375, 2.4806900024414062, 41.47564697265625, 23.1922607421875, 3.08734130859375, 2.90338134765625, 28.40643310546875, 2.3621673583984375, 8.252826690673828, 12.630760192871094, -17.927703857421875, -8.79052734375, 4.1749114990234375, 16.668350219726562, 16.172409057617188, 20.809326171875, 7.9034881591796875, 0.6590480804443359, 37.02838134765625, 0.720367431640625, 4.653228759765625, 19.257675170898438, 16.235855102539062, -23.789535522460938, 15.198234558105469, 3.449047088623047, 7.8058624267578125, 47.64067077636719, 0.8421897888183594, 35.819366455078125, 0.0, 45.99116516113281, -5.44189453125, -1.317605972290039, 27.195480346679688, 25.7901611328125, 11.773681640625, -15.97918701171875, 9.103372573852539, -1.927215576171875, 53.734893798828125, 60.811309814453125, -59.195953369140625, 30.39362335205078, 4.641242980957031, 34.836456298828125, 35.83319091796875, 5.98846435546875, 4.600807189941406, 47.91058349609375, 35.9127197265625, -14.000686645507812, -8.278228759765625, -12.88140869140625, 6.390663146972656, -19.140975952148438, 51.360137939453125, -21.003067016601562, 3.044342041015625, 41.1878662109375, -20.900985717773438, 61.104461669921875, 8.59527587890625, 9.056610107421875, -9.71221923828125, 70.94024658203125, 8.20458984375, 33.188201904296875, 2.4926071166992188, -11.9173583984375, 36.43133544921875, 0.68328857421875, -6.2659149169921875, 3.4840545654296875, 0.0, 37.998077392578125, -2.8105926513671875, -4.636299133300781, 26.033401489257812, -22.120361328125, 2.1895751953125, 57.55842590332031, 21.3919677734375, 39.28849792480469, 7.291099548339844, 51.7568359375, -22.282241821289062, -16.2401123046875, 5.6571197509765625, 13.037322998046875, 40.891357421875, -8.569198608398438, 33.2506103515625, 6.126220703125, 52.68865966796875, 44.55787658691406, 43.919281005859375, -61.914642333984375, 30.510677337646484, 43.863502502441406, 15.05218505859375, 36.576171875, -9.359786987304688, 43.04706573486328, 2.9686050415039062, 39.382568359375, -0.457916259765625, 6.904003143310547, 23.360809326171875, 42.83741760253906, 30.086959838867188, -9.90469741821289, 29.47442626953125, -8.701080322265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000146.npy"} {"epoch": 0.3057591623036649, "step": 147, "batch_size": 128, "mean": 11.693140983581543, "std": 28.326513290405273, "min": -76.30413818359375, "p10": -18.906491088867188, "median": 9.181427001953125, "p90": 48.929747009277335, "max": 88.34823608398438, "pos_frac": 0.6875, "sample": [0.3644866943359375, -3.756988525390625, 7.5131072998046875, 29.446075439453125, -9.455223083496094, -2.7382049560546875, 84.44235229492188, -14.324859619140625, 0.0, 70.15679931640625, -16.1827392578125, 20.903884887695312, 14.68408203125, 6.2126617431640625, 25.247316360473633, 4.70294189453125, -19.423736572265625, 43.8846435546875, 51.8079833984375, 3.2199020385742188, -4.17645263671875, 17.61077880859375, 0.12665557861328125, 35.70909118652344, -1.7727584838867188, 50.677642822265625, 0.5936393737792969, 43.97528076171875, 9.179931640625, 1.6865234375, 67.55252075195312, 35.4342041015625, 18.376537322998047, 23.12884521484375, 11.35601806640625, -28.30780029296875, -42.253623962402344, -5.530426025390625, -16.73162841796875, 17.06500244140625, 10.72027587890625, 34.324039459228516, 31.97054672241211, 6.3131561279296875, -31.250843048095703, 29.383758544921875, -28.460372924804688, -4.060968399047852, -0.3067150115966797, 1.1874923706054688, -3.1013336181640625, 50.977081298828125, 10.729232788085938, 13.343185424804688, 27.544090270996094, 29.04589080810547, 80.1871337890625, 12.246414184570312, 4.265907287597656, -4.836029052734375, 20.88623046875, 22.887298583984375, 57.93278503417969, 8.41226577758789, 53.871337890625, 2.037872314453125, 1.45013427734375, 59.481842041015625, 39.528282165527344, 2.524993896484375, -43.673553466796875, 12.1209716796875, 2.303009033203125, 0.893951416015625, 32.25738525390625, 5.52056884765625, 13.431707382202148, 3.7814102172851562, 0.15601348876953125, -0.6005783081054688, 27.961410522460938, -9.25262451171875, 0.4903411865234375, 35.79583740234375, 17.13225555419922, 37.3216552734375, 9.18292236328125, 31.3572998046875, -11.2384033203125, 32.71886444091797, 14.0430908203125, 42.9793701171875, -25.053466796875, -76.30413818359375, -38.40709686279297, -1.6727523803710938, 9.908554077148438, 29.41900634765625, -18.365631103515625, -40.93360900878906, -9.555984497070312, 4.081153869628906, 52.981353759765625, 15.01593017578125, -18.684814453125, -8.035446166992188, 45.14691162109375, -3.2399444580078125, -40.74543762207031, -15.174835205078125, -65.34403228759766, 10.609405517578125, 7.7775421142578125, 38.74629211425781, 48.18064880371094, 36.54150390625, 10.697425842285156, 53.615020751953125, -4.8704376220703125, -7.290008544921875, 16.05840301513672, 42.33397674560547, 44.02714538574219, 14.3651123046875, 88.34823608398438, 12.037277221679688, -4.705415725708008, -29.108154296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000147.npy"} {"epoch": 0.3078534031413613, "step": 148, "batch_size": 128, "mean": 12.354327201843262, "std": 26.86064338684082, "min": -58.21788024902344, "p10": -24.382000732421876, "median": 10.81943130493164, "p90": 48.662498474121094, "max": 92.66525268554688, "pos_frac": 0.6484375, "sample": [12.1385498046875, 12.319808959960938, 18.34283447265625, -14.71124267578125, -1.5567054748535156, 67.962646484375, -0.04575347900390625, 19.671295166015625, 72.33892822265625, 25.689170837402344, 5.695400238037109, 20.427841186523438, -3.10162353515625, 55.27947998046875, 12.397491455078125, 37.26206970214844, 11.4337158203125, -18.12542724609375, 15.248626708984375, 26.414691925048828, -20.842254638671875, 42.05963134765625, -46.70172882080078, 0.4056205749511719, 9.634857177734375, -24.12567138671875, 14.357177734375, -5.4575653076171875, 15.207427978515625, -8.753190994262695, 20.93377685546875, 15.018234252929688, -8.842941284179688, -58.21788024902344, -8.416351318359375, -15.132354736328125, 60.372589111328125, 8.25048828125, -27.32763671875, 20.77630615234375, 57.8477783203125, 1.481130599975586, -15.51373291015625, -32.47198486328125, 1.0104827880859375, 5.894596099853516, 1.955780029296875, -15.6671142578125, 22.875732421875, 29.597747802734375, 13.383598327636719, 8.117752075195312, 7.126300811767578, 18.84613037109375, -30.28570556640625, 69.16375732421875, 8.854347229003906, 37.06666564941406, 10.205146789550781, 48.34149169921875, 22.34320068359375, -7.1767120361328125, -0.2992668151855469, -1.0630645751953125, 0.0, 8.575469970703125, 55.5623779296875, 25.278366088867188, 36.00885009765625, 39.29119873046875, -3.990407943725586, -29.52674102783203, -4.9739837646484375, -36.4432373046875, 38.64982604980469, 8.3997802734375, -27.85077667236328, 31.457611083984375, -1.6755218505859375, 31.091018676757812, -0.6039638519287109, 19.549530029296875, -24.9801025390625, 30.806549072265625, 58.09233093261719, 18.855709075927734, 2.094024658203125, 16.793243408203125, -9.506362915039062, -16.32000732421875, 27.23796844482422, 20.197490692138672, -0.7765998840332031, 28.027618408203125, 25.7633056640625, 48.08287048339844, -27.26068115234375, 6.103118896484375, -11.528213500976562, 63.06269836425781, 3.0342254638671875, 26.806884765625, 92.66525268554688, 36.297119140625, 2.2550506591796875, -8.215391159057617, 31.974445343017578, -28.700042724609375, 32.02618408203125, 23.681434631347656, 49.41151428222656, 43.74066162109375, 16.593505859375, 28.96240234375, -27.249549865722656, -2.22930908203125, -27.83697509765625, 16.026931762695312, 61.334625244140625, 0.0, -8.8182373046875, 13.200775146484375, 34.40020751953125, -0.7100753784179688, 7.9796600341796875, 43.58198547363281, 61.36930847167969, -3.6873779296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000148.npy"} {"epoch": 0.3099476439790576, "step": 149, "batch_size": 128, "mean": 15.413253784179688, "std": 26.159282684326172, "min": -41.745208740234375, "p10": -19.665377807617183, "median": 11.685348510742188, "p90": 47.55890808105468, "max": 95.892822265625, "pos_frac": 0.7578125, "sample": [48.907318115234375, -25.202056884765625, 23.68048095703125, 34.34197235107422, 42.633636474609375, 46.98101806640625, 22.854171752929688, 5.944187164306641, 8.5625, 5.945587158203125, 18.7535400390625, 2.85504150390625, 43.572174072265625, 16.3779296875, 28.64910888671875, -2.3292388916015625, 43.926513671875, 9.839740753173828, -1.94183349609375, -7.772705078125, 16.38591766357422, 4.902244567871094, 19.07568359375, 4.25640869140625, 25.78360366821289, 11.487457275390625, -40.636741638183594, 39.016448974609375, 49.1148681640625, 15.37115478515625, 7.648406982421875, 17.115936279296875, 33.11669921875, 11.990188598632812, 1.9121942520141602, 0.15478515625, 51.28132629394531, 2.076244354248047, -18.280990600585938, 0.6789627075195312, 34.9267578125, 68.57278442382812, -11.742881774902344, 2.17156982421875, 77.29116821289062, 45.6927490234375, 10.950714111328125, -22.895614624023438, 37.527557373046875, 50.182861328125, 21.211204528808594, -3.6995086669921875, 9.74755859375, 25.247360229492188, 5.0206756591796875, 76.5794677734375, -29.94952392578125, 16.73394775390625, 16.690231323242188, 11.88323974609375, -7.1619873046875, 24.129127502441406, -25.57232666015625, 39.24897766113281, -25.6343994140625, 2.9873046875, 7.7442626953125, -6.42681884765625, 23.666702270507812, 22.37726402282715, 39.63428497314453, 19.726425170898438, 32.30600357055664, 12.606842041015625, 40.618804931640625, 14.398246765136719, 50.707916259765625, 5.8076019287109375, 56.766082763671875, -0.04547119140625, 40.932647705078125, 0.6759109497070312, 32.79845428466797, 28.194580078125, 11.355560302734375, -10.0352783203125, 55.49822998046875, 42.123321533203125, -39.857696533203125, 2.6784439086914062, 11.149429321289062, 33.59722900390625, -5.36041259765625, -26.4100341796875, -2.884246826171875, -5.069854736328125, 65.77310180664062, 6.8538818359375, 43.08232116699219, 6.423309326171875, -7.1168365478515625, 3.6667327880859375, 32.87750244140625, 1.2449951171875, 28.572952270507812, -0.998077392578125, -28.146697998046875, -23.6524658203125, 0.392333984375, 95.892822265625, -5.5281982421875, -4.65435791015625, 7.86322021484375, 29.591522216796875, 5.218109130859375, -11.094329833984375, 42.125633239746094, 27.941604614257812, -41.675270080566406, -41.745208740234375, 19.999649047851562, 37.70458984375, -25.332839965820312, 73.98599243164062, 39.602561950683594, 24.014266967773438, 3.332672119140625, 8.237518310546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000149.npy"} {"epoch": 0.31204188481675393, "step": 150, "batch_size": 128, "mean": 14.469941139221191, "std": 24.96263885498047, "min": -51.11865234375, "p10": -14.397694396972657, "median": 12.59688949584961, "p90": 45.47791290283203, "max": 82.21463012695312, "pos_frac": 0.7734375, "sample": [20.697021484375, 32.41295623779297, -3.8473663330078125, 33.421112060546875, 10.077789306640625, -27.117584228515625, -3.0479583740234375, 17.177413940429688, -33.332244873046875, 31.30413818359375, 25.850357055664062, 22.520233154296875, 44.45068359375, 17.1312255859375, -14.48577880859375, 3.302490234375, 8.534622192382812, 28.80579376220703, 6.9854736328125, 38.194244384765625, 24.418563842773438, -14.384918212890625, -2.056915283203125, 3.5836944580078125, -5.493152618408203, 15.75128173828125, 8.675628662109375, 6.79669189453125, 65.04893493652344, 34.25153350830078, 66.9808349609375, 0.7565574645996094, 19.68310546875, 8.51262092590332, 17.801742553710938, 26.719009399414062, 21.212921142578125, -38.65447998046875, 38.68022155761719, 9.977340698242188, 25.339691162109375, 42.963287353515625, 16.641845703125, 40.08587646484375, -8.668670654296875, 1.2996673583984375, 0.771453857421875, 5.5804443359375, 13.698394775390625, 63.6602783203125, 16.357852935791016, 74.86270141601562, 1.419891357421875, 31.507843017578125, 45.102020263671875, 62.085296630859375, 51.9185791015625, 6.14122200012207, -0.652862548828125, 0.7459945678710938, 0.0944976806640625, -0.7659053802490234, -32.761566162109375, 46.371803283691406, 3.80010986328125, 2.5873870849609375, 5.8860626220703125, 1.8321533203125, 43.004188537597656, 43.58721923828125, 20.133880615234375, 37.9039306640625, 2.9054107666015625, 46.35499572753906, 20.992610931396484, -9.637161254882812, -41.059783935546875, 8.353057861328125, 3.0471324920654297, 23.672775268554688, 13.316154479980469, -51.11865234375, -2.33984375, -19.778244018554688, -5.791130065917969, 15.808837890625, -22.27947998046875, 82.21463012695312, 34.232177734375, 11.87762451171875, 14.15057373046875, -10.496429443359375, 6.0641326904296875, 25.453453063964844, -48.440185546875, 9.129096984863281, 40.34138488769531, -2.268798828125, -24.699859619140625, 1.65814208984375, 16.65245819091797, 31.347938537597656, 9.783050537109375, 52.132354736328125, 10.132047653198242, 18.647857666015625, 36.5816650390625, 10.116836547851562, 41.433013916015625, 30.050018310546875, 2.0742950439453125, 15.431396484375, 4.807119369506836, 51.098602294921875, 7.7447357177734375, -0.248626708984375, 15.258201599121094, -10.498779296875, 50.76470947265625, -14.427505493164062, 0.0, 27.03594970703125, 74.1739501953125, 1.563629150390625, 14.28531265258789, 17.027267456054688, 18.708091735839844, -31.016128540039062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000150.npy"} {"epoch": 0.31413612565445026, "step": 151, "batch_size": 128, "mean": 22.23900032043457, "std": 28.61642837524414, "min": -54.74229431152344, "p10": -16.037470245361327, "median": 25.017807006835938, "p90": 56.300991821289045, "max": 83.828125, "pos_frac": 0.7578125, "sample": [39.36564636230469, 20.901611328125, 0.7934150695800781, -20.639976501464844, 0.20452880859375, 81.46368408203125, 51.714630126953125, -24.43462371826172, 1.25030517578125, 40.511749267578125, -0.865997314453125, -5.50347900390625, 23.305831909179688, 36.576416015625, -7.380912780761719, -15.586296081542969, -3.6877288818359375, 40.95829772949219, 37.714202880859375, 22.633499145507812, -54.74229431152344, 4.3177490234375, -11.295562744140625, 24.76263427734375, 5.788818359375, 41.67315673828125, 4.8980560302734375, -3.2542495727539062, 26.894927978515625, 42.230377197265625, 47.822265625, 4.42645263671875, -4.1968231201171875, 22.63524627685547, 38.46995544433594, 14.700607299804688, 18.133575439453125, 17.039886474609375, -5.2749786376953125, 78.89456176757812, 29.090530395507812, 32.28462219238281, 13.473678588867188, 19.872467041015625, 44.81500244140625, -20.962905883789062, 25.012298583984375, 7.669708251953125, 45.664794921875, 44.47692108154297, 43.583702087402344, -8.77056884765625, 4.089332580566406, -0.287353515625, 36.77333068847656, 48.26985168457031, 15.44677734375, 25.50244140625, -28.456298828125, 21.147689819335938, -5.426116943359375, 83.828125, 49.209716796875, -10.113040924072266, 30.06158447265625, 43.133697509765625, 29.00115966796875, -22.416595458984375, 41.184661865234375, 60.315673828125, -36.97674560546875, -29.13824462890625, 45.39532470703125, 33.9652099609375, 32.96906280517578, 25.0233154296875, 54.580413818359375, 39.11444091796875, 74.8348388671875, -17.8133544921875, -23.726318359375, 44.509002685546875, 70.13967895507812, 47.531585693359375, 77.85302734375, 8.105995178222656, 37.236114501953125, -0.596435546875, 0.0, 12.715179443359375, 25.88262939453125, 40.8204345703125, 33.22447204589844, -10.097061157226562, 34.72235107421875, 33.12811279296875, 50.66204833984375, 69.49420166015625, -13.625450134277344, 73.4700927734375, -32.988006591796875, 36.51373291015625, 2.642425537109375, 35.752166748046875, 62.203704833984375, 47.332275390625, 16.10009765625, -17.0902099609375, 30.609329223632812, 33.209808349609375, 17.4033203125, 67.0888671875, 14.502487182617188, -15.552204132080078, 75.23336791992188, 19.182586669921875, 51.167694091796875, 37.747161865234375, 67.20889282226562, 25.18084716796875, 41.66780090332031, 9.736083984375, 11.662200927734375, 31.8675537109375, 38.165122985839844, 1.4427490234375, 8.51617431640625, -34.012001037597656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000151.npy"} {"epoch": 0.3162303664921466, "step": 152, "batch_size": 128, "mean": 16.827287673950195, "std": 28.631921768188477, "min": -56.084564208984375, "p10": -14.29251480102539, "median": 15.245189666748047, "p90": 56.467620849609375, "max": 97.24777221679688, "pos_frac": 0.703125, "sample": [34.510650634765625, 48.99591064453125, -4.0562896728515625, 15.977638244628906, 6.194549560546875, 31.51397705078125, 12.2139892578125, 32.73149108886719, 57.12242126464844, 30.190032958984375, 17.48593521118164, 38.958038330078125, 68.59782409667969, 7.000579833984375, 38.8619384765625, -12.535003662109375, 0.5623779296875, 3.0626983642578125, 1.894256591796875, -22.82733154296875, -55.93177795410156, 29.990631103515625, -56.084564208984375, 21.877960205078125, -9.386058807373047, 37.133087158203125, 15.502281188964844, -0.27504730224609375, 52.948028564453125, -1.045989990234375, -0.8460369110107422, 46.99871826171875, 56.552215576171875, 29.903472900390625, 27.961090087890625, 34.49876403808594, 50.280029296875, 42.71708679199219, 61.10009765625, 30.097549438476562, -41.29064178466797, 97.24777221679688, -27.29290771484375, 7.202056884765625, 32.42042541503906, -1.7872562408447266, -10.122283935546875, 22.6497802734375, -8.841653823852539, 1.17889404296875, 5.734771728515625, -7.4315643310546875, 14.98809814453125, -6.0946044921875, -6.741397857666016, 26.630279541015625, 10.795745849609375, -23.0030517578125, -7.5736236572265625, -1.3406982421875, 62.4361572265625, 17.379852294921875, -17.381378173828125, 19.0047607421875, 67.12109375, -21.45440673828125, 16.85699462890625, -14.839469909667969, -6.402366638183594, 41.601806640625, 7.565629959106445, -10.657318115234375, -6.18670654296875, 9.950286865234375, 1.9903335571289062, 65.473388671875, 39.916229248046875, 19.543121337890625, 71.8570556640625, 32.413448333740234, 56.431365966796875, -28.793853759765625, 1.805398941040039, -6.754974365234375, 1.41717529296875, 3.5042343139648438, 37.779388427734375, 35.205780029296875, 69.73406982421875, 13.39213752746582, 7.424552917480469, -50.876678466796875, -9.340164184570312, 64.35096740722656, 35.930580139160156, 30.00072479248047, -15.87213134765625, 5.516326904296875, 23.828872680664062, -14.05810546875, 20.1080322265625, 46.228759765625, 29.078826904296875, 24.87811279296875, -11.13653564453125, 59.5364990234375, 4.49224853515625, 10.862884521484375, -7.4091796875, -6.753997802734375, 42.779422760009766, 9.834320068359375, -44.834228515625, 27.891876220703125, -1.1783790588378906, 11.688690185546875, 31.49018096923828, 1.1778182983398438, 45.652862548828125, 60.70452880859375, 46.178253173828125, 51.92962646484375, 46.95109558105469, 34.809173583984375, 5.2683258056640625, 25.09893798828125, -9.106781005859375, 43.081825256347656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000152.npy"} {"epoch": 0.3183246073298429, "step": 153, "batch_size": 128, "mean": 24.108661651611328, "std": 27.00099754333496, "min": -56.314605712890625, "p10": -2.883473205566406, "median": 19.7017822265625, "p90": 62.75394287109375, "max": 90.9688720703125, "pos_frac": 0.859375, "sample": [14.477447509765625, 0.0223846435546875, 73.37939453125, 15.55194091796875, 59.692169189453125, 46.795616149902344, 46.306671142578125, 1.7970428466796875, 90.9688720703125, 6.0208740234375, 46.712005615234375, 4.1431121826171875, -12.475616455078125, 24.167495727539062, 35.16371154785156, 4.678337097167969, -12.984085083007812, 2.64349365234375, 28.934165954589844, 68.09837341308594, 62.37542724609375, 72.19429016113281, 35.29682159423828, 4.0086517333984375, 6.522491455078125, -4.255058288574219, 35.256431579589844, -1.8935813903808594, 1.5707130432128906, 3.3985595703125, -16.603729248046875, 5.986961364746094, 47.21173095703125, 4.2178955078125, 4.4830322265625, 15.137405395507812, 54.05157470703125, 44.78681945800781, 29.378782272338867, 0.0, 61.88067626953125, 39.67802429199219, 57.713134765625, 21.266555786132812, 48.28778076171875, 39.09263610839844, 43.36517333984375, -3.0062103271484375, 19.59991455078125, 18.455474853515625, 16.529327392578125, 48.81756591796875, -56.314605712890625, 4.0847625732421875, -2.83087158203125, 35.55401611328125, 50.00390625, 37.689697265625, 1.0380096435546875, 68.56756591796875, 16.445831298828125, 6.5999298095703125, 21.096969604492188, 64.3753662109375, -7.647392272949219, 50.956024169921875, 10.2679443359375, 11.423736572265625, 19.109817504882812, 19.80364990234375, -6.9229583740234375, 21.2568359375, 5.1472930908203125, 39.44996643066406, 0.9510498046875, -0.4551239013671875, 10.0318603515625, 33.1400146484375, 16.799850463867188, 33.863525390625, 36.268463134765625, -11.676666259765625, 50.114288330078125, 69.96365356445312, 23.713958740234375, 21.51226806640625, 5.639808654785156, 13.962203979492188, 40.362388610839844, -0.8719482421875, 3.7962188720703125, 19.277191162109375, 64.53587341308594, 45.142547607421875, 55.43421936035156, 11.51230239868164, 5.801523208618164, 18.001861572265625, 9.63323974609375, 9.392120361328125, 48.28839111328125, 47.417327880859375, 40.126708984375, 8.555313110351562, 2.8775634765625, 3.168365478515625, 28.495819091796875, 78.26364135742188, 28.544971466064453, 72.96759033203125, 63.63714599609375, 23.582046508789062, 5.13104248046875, 64.0068359375, -20.626882553100586, 27.7584228515625, 11.512588500976562, 61.77215576171875, 29.967910766601562, 6.0802001953125, -18.2503662109375, 69.71717834472656, 58.8408203125, 13.191741943359375, -47.726661682128906, -38.80462646484375, 32.9326171875, 30.579986572265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000153.npy"} {"epoch": 0.3204188481675393, "step": 154, "batch_size": 128, "mean": 17.610185623168945, "std": 31.83588218688965, "min": -77.18527221679688, "p10": -23.19345092773437, "median": 17.99991226196289, "p90": 60.089683532714844, "max": 85.40765380859375, "pos_frac": 0.7265625, "sample": [25.561676025390625, 50.328369140625, -26.070770263671875, 61.83367919921875, 33.82752990722656, 7.288536071777344, -6.151012420654297, 59.22538757324219, 7.274993896484375, 16.429168701171875, -49.82191467285156, 38.52092742919922, 18.521774291992188, 14.827545166015625, 34.03205871582031, -77.18527221679688, 36.59013366699219, -26.264297485351562, 35.84686279296875, 2.530609130859375, 12.721160888671875, 0.752655029296875, 7.242919921875, -26.164779663085938, -3.508270263671875, 4.650108337402344, 36.3338623046875, 7.7273101806640625, 7.0308837890625, 72.94668579101562, 12.080780029296875, -6.99737548828125, 58.570404052734375, 68.123046875, 4.09210205078125, 49.931724548339844, 30.3421630859375, 71.46759033203125, 39.752227783203125, 64.15032958984375, 4.4649658203125, 59.3515625, 65.63252258300781, 30.026161193847656, -53.2491455078125, 18.439956665039062, 79.0897216796875, -2.0710296630859375, -7.34161376953125, -26.175567626953125, -4.762138366699219, -7.7409820556640625, 18.904953002929688, -25.07440185546875, 28.36077880859375, 41.48048400878906, 30.699462890625, -6.05743408203125, 3.3856964111328125, -21.85406494140625, -2.250621795654297, 16.763076782226562, 47.151947021484375, 26.086631774902344, 26.829559326171875, -49.935943603515625, -17.0638484954834, 0.153961181640625, -2.7598495483398438, -9.22015380859375, -8.783809661865234, -48.30523681640625, 53.90925598144531, 67.32717895507812, -4.981117248535156, -9.721565246582031, 11.909568786621094, -6.4666748046875, 34.371551513671875, 9.064071655273438, -37.04075622558594, 11.473468780517578, -1.1042633056640625, 24.470306396484375, 34.79449462890625, 10.235107421875, 54.180938720703125, 33.321319580078125, -7.573333740234375, 33.22838592529297, 20.94384765625, 20.9359130859375, 59.98451232910156, 48.39177703857422, 36.87689208984375, -17.139434814453125, 22.1533203125, 6.614013671875, 85.40765380859375, 14.966217041015625, 20.7296142578125, 49.470733642578125, -22.3873291015625, -64.81608581542969, 41.31806945800781, 29.180206298828125, 36.503265380859375, 31.629364013671875, 17.340728759765625, 24.099578857421875, 68.77940368652344, 34.15470886230469, 33.77032470703125, 57.18232727050781, 2.13946533203125, 5.511285781860352, 80.8258056640625, 17.55986785888672, 17.08934783935547, 25.263824462890625, -42.761810302734375, -12.890552520751953, 60.3350830078125, 30.949493408203125, 4.087715148925781, 24.041015625, 37.98966979980469, 63.908958435058594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000154.npy"} {"epoch": 0.3225130890052356, "step": 155, "batch_size": 128, "mean": 12.791027069091797, "std": 28.905765533447266, "min": -61.8182373046875, "p10": -22.22660446166992, "median": 10.574287414550781, "p90": 49.852877044677726, "max": 83.46142578125, "pos_frac": 0.6796875, "sample": [67.13082885742188, 25.717239379882812, 15.21832275390625, -6.04132080078125, 35.28007507324219, -0.42589759826660156, 10.245101928710938, -1.0523681640625, 8.0438232421875, 48.75238800048828, 43.724430084228516, 4.272190093994141, 29.7611083984375, -23.828109741210938, 24.722991943359375, 29.29736328125, -15.87677001953125, 73.60543823242188, 83.46142578125, -6.523834228515625, 10.903472900390625, 27.642608642578125, 46.155364990234375, 81.82269287109375, 5.0995635986328125, 42.30853271484375, -41.81727600097656, 0.0, -28.889892578125, -7.325653076171875, -21.489723205566406, 56.069091796875, 21.276535034179688, 47.168663024902344, -0.42022705078125, -14.888229370117188, -21.540245056152344, 39.479896545410156, -9.1602783203125, 1.233978271484375, 8.92071533203125, 27.475250244140625, 17.9759521484375, 1.46875, 36.90728759765625, 57.41912841796875, 31.624267578125, 12.31939697265625, 7.0543975830078125, 35.95056915283203, -11.085700988769531, -53.447509765625, 8.360078811645508, 8.23651123046875, 3.5217533111572266, -36.964080810546875, -1.5479736328125, 36.689453125, 73.999267578125, 52.420684814453125, 2.7039451599121094, 4.302001953125, 18.73870849609375, -9.277503967285156, -43.74614715576172, -9.475875854492188, 45.15318298339844, 57.63201904296875, 13.634052276611328, 18.319290161132812, -2.07171630859375, -29.960716247558594, 39.993438720703125, 44.024169921875, -14.01568603515625, 33.103851318359375, 8.7027587890625, -6.474479675292969, 36.84918212890625, -28.57208251953125, 34.750457763671875, 54.04679870605469, -25.108489990234375, -14.272811889648438, 3.1875457763671875, 7.990753173828125, 38.39691162109375, -39.320068359375, 0.416412353515625, 42.613616943359375, 18.87664794921875, 27.967514038085938, 22.6162109375, -1.74420166015625, -41.08612060546875, 14.769569396972656, 9.076019287109375, 64.41474914550781, 12.838287353515625, -15.95306396484375, -16.291961669921875, -13.619949340820312, -5.0460357666015625, 11.181087493896484, 20.03533935546875, 40.45506286621094, 4.8176116943359375, -5.693977355957031, -61.8182373046875, 1.5487136840820312, -41.693603515625, 12.449569702148438, 37.76742172241211, -12.37109375, 53.4234619140625, 31.16473388671875, 36.355140686035156, 26.356704711914062, 15.15777587890625, 60.568145751953125, -21.15386962890625, 0.2735748291015625, 0.34275054931640625, 16.181442260742188, 17.59197998046875, 24.009307861328125, 13.88543701171875, 0.92413330078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000155.npy"} {"epoch": 0.32460732984293195, "step": 156, "batch_size": 128, "mean": 21.715551376342773, "std": 33.17417907714844, "min": -77.7095947265625, "p10": -22.924919128417965, "median": 26.214599609375, "p90": 61.11421356201172, "max": 94.76690673828125, "pos_frac": 0.78125, "sample": [36.80029296875, 42.056121826171875, 30.07695770263672, 20.30328369140625, -0.960723876953125, 4.135429382324219, 47.02099609375, 47.02923583984375, 32.125755310058594, 28.84638786315918, 44.18196105957031, 27.751602172851562, 49.7119140625, 32.99224853515625, 3.7296371459960938, -22.2864990234375, -6.063484191894531, 34.4713134765625, 5.921966552734375, 42.501625061035156, 2.80255126953125, 45.253173828125, -12.249237060546875, -29.769195556640625, 30.834213256835938, 9.486465454101562, -3.4488754272460938, 21.749679565429688, 0.69622802734375, 44.516845703125, 68.40283203125, -4.9368438720703125, 71.52728271484375, -5.6586151123046875, 11.87966537475586, -25.983413696289062, -38.247039794921875, 5.9542694091796875, 49.073028564453125, 2.8060760498046875, 28.25164794921875, 60.663360595703125, 30.906494140625, 7.413238525390625, 4.3166046142578125, 2.8844757080078125, 43.9212646484375, -35.29364013671875, 46.69883346557617, 31.513031005859375, 2.8388671875, 36.220848083496094, 22.080047607421875, 25.45123291015625, 85.72760009765625, -25.1724853515625, -37.99082946777344, 86.45550537109375, 0.013824462890625, 8.923385620117188, -7.730682373046875, 83.6693115234375, 34.171844482421875, 50.57981872558594, 31.142566680908203, 58.68963623046875, 51.84716796875, 10.99151611328125, 61.25079345703125, 6.39862060546875, 21.83935546875, 0.39453125, -4.509037017822266, 66.330078125, 36.30250549316406, 34.648590087890625, 94.76690673828125, 18.30328369140625, 0.4406852722167969, -3.9164295196533203, 83.40335083007812, 35.06135559082031, 71.659423828125, 2.058441162109375, 9.549163818359375, 30.552101135253906, 2.6490306854248047, 34.08612060546875, 31.44849395751953, -31.302474975585938, 45.096923828125, 52.1109619140625, 37.68144226074219, 45.51416015625, 27.58331298828125, 11.37744140625, -21.0146484375, 61.05567932128906, 26.97796630859375, -24.414566040039062, 61.71136474609375, 15.46234130859375, 18.59613037109375, -27.1646728515625, -77.7095947265625, 53.623138427734375, -1.7187442779541016, 86.45867919921875, -2.8301544189453125, -39.88133239746094, 4.636810302734375, 33.013519287109375, 87.5672607421875, 11.8211669921875, -76.77899169921875, -37.55819320678711, 0.69903564453125, 46.364463806152344, 53.7520751953125, 57.0738525390625, 0.3715248107910156, 27.078887939453125, 42.453941345214844, 58.862274169921875, -19.147003173828125, 58.2138671875, -7.7351531982421875, 4.75115966796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000156.npy"} {"epoch": 0.3267015706806283, "step": 157, "batch_size": 128, "mean": 15.066287994384766, "std": 29.641189575195312, "min": -62.98963928222656, "p10": -21.393421936035157, "median": 12.458808898925781, "p90": 52.2958625793457, "max": 97.90533447265625, "pos_frac": 0.703125, "sample": [33.96429443359375, -0.7185516357421875, 24.340057373046875, 63.10438537597656, -0.325347900390625, -21.50323486328125, 12.6915283203125, 20.857574462890625, 52.89251708984375, -7.57769775390625, 27.302825927734375, 0.17450714111328125, -44.317596435546875, 28.740272521972656, 0.6254196166992188, 35.804931640625, 9.801990509033203, -0.6595077514648438, 3.0791587829589844, -13.028640747070312, -20.162246704101562, 67.51666259765625, 31.402206420898438, -11.052703857421875, 48.23233413696289, -1.082937240600586, -34.72615051269531, -1.614166259765625, 4.8195343017578125, 16.9473876953125, 10.641098022460938, -6.3824462890625, 15.608688354492188, -48.352783203125, 35.21788787841797, 51.32661437988281, 21.568695068359375, 31.315628051757812, -44.28533935546875, -62.98963928222656, 81.2110595703125, 15.978225708007812, 36.21821594238281, 20.04425048828125, 81.94869995117188, 39.20924377441406, 12.354660034179688, -4.2812347412109375, 15.582489013671875, 0.0, 18.152908325195312, 49.26434326171875, 10.476333618164062, -6.904571533203125, 66.3377685546875, 12.562957763671875, 40.770721435546875, -6.724395751953125, -4.03125, 1.2266769409179688, 48.998291015625, 23.639678955078125, 54.33367156982422, 11.995773315429688, 31.043060302734375, -2.7691650390625, 50.883880615234375, -4.847747802734375, -49.0601806640625, 37.80377197265625, 0.3472900390625, -15.460601806640625, -43.224578857421875, 8.9923095703125, -12.225494384765625, 16.898284912109375, -33.49822998046875, 41.125732421875, 6.3955230712890625, 55.082183837890625, 20.85552978515625, 42.063507080078125, 0.11156463623046875, -7.4900970458984375, 2.3926162719726562, 13.005828857421875, 7.6006317138671875, 13.814376831054688, 27.32291030883789, 0.0, -29.106842041015625, 60.51141357421875, 23.683425903320312, 16.400909423828125, 5.474334716796875, -3.450164794921875, 28.05873680114746, 55.51457977294922, 97.90533447265625, 52.04015350341797, 10.351837158203125, -21.346359252929688, 35.906341552734375, 6.493890762329102, 3.98138427734375, 2.9540328979492188, 35.06486511230469, 35.44474792480469, 26.3092041015625, -8.16033935546875, -27.58966064453125, 43.909881591796875, 3.2786712646484375, 40.12689208984375, 33.02864074707031, 31.5948486328125, -5.0672607421875, 1.5098190307617188, -25.7174072265625, 47.34764099121094, -26.680938720703125, 7.521751403808594, 6.140954971313477, 29.119447708129883, 5.9046783447265625, 18.47844696044922, 96.63577270507812, 60.18638610839844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000157.npy"} {"epoch": 0.3287958115183246, "step": 158, "batch_size": 128, "mean": 16.788619995117188, "std": 29.56882667541504, "min": -50.680908203125, "p10": -23.10926208496094, "median": 16.286216735839844, "p90": 53.95505981445312, "max": 83.17779541015625, "pos_frac": 0.71875, "sample": [4.877729415893555, -1.9621353149414062, 24.52435302734375, 54.813720703125, -34.6004638671875, 5.3550262451171875, -3.3197021484375, 26.119522094726562, -11.364151000976562, 54.805023193359375, -16.83837890625, 0.40753173828125, -41.94818115234375, -39.375640869140625, 23.102752685546875, 19.3009033203125, 33.855712890625, 29.89319610595703, 8.68328857421875, 29.783584594726562, 16.622451782226562, 62.3082275390625, 40.81170654296875, 3.9698944091796875, 16.66014862060547, 43.05619812011719, 12.159828186035156, -48.08251953125, 42.2218017578125, 22.679946899414062, -10.095855712890625, 37.28684997558594, -14.842803955078125, 54.223541259765625, 19.63653564453125, 7.279655456542969, 40.4158935546875, -35.881160736083984, 16.97156524658203, -41.8035888671875, 41.356658935546875, 7.2208251953125, 11.440509796142578, -26.250308990478516, 12.510986328125, -2.7708663940429688, -7.3008575439453125, 17.76702880859375, 37.87391662597656, 21.556835174560547, 65.20831298828125, -3.9875011444091797, 1.5579586029052734, 10.303611755371094, 35.04357147216797, 47.722991943359375, -36.599388122558594, -23.0316162109375, 62.914825439453125, 19.837448120117188, 62.51776123046875, 7.558067321777344, -5.60211181640625, 10.7099609375, 35.78399658203125, 13.427337646484375, -21.696624755859375, 0.0, 31.308822631835938, 14.9481201171875, 69.6116943359375, -6.648345947265625, -1.768096923828125, 51.181884765625, 25.252471923828125, 53.839996337890625, 4.1691436767578125, -22.858840942382812, -9.684707641601562, -2.4113216400146484, 50.96144104003906, 5.89544677734375, -45.452301025390625, 15.789779663085938, 31.936477661132812, 38.86883544921875, 37.428985595703125, -0.27329254150390625, 44.11289978027344, 40.63726806640625, 47.002532958984375, 29.473541259765625, 13.640419006347656, 18.685516357421875, 68.03555297851562, 45.45477294921875, 82.28646850585938, -6.300075531005859, -24.57318115234375, -23.290435791015625, 15.949981689453125, 83.17779541015625, -17.119415283203125, 38.64787292480469, 49.44404602050781, 46.57635498046875, 54.243743896484375, 36.81261444091797, 34.720916748046875, 0.4492950439453125, 15.71533203125, 33.5504150390625, 37.442039489746094, 5.3393707275390625, 28.360366821289062, -14.84564208984375, 53.455596923828125, 3.571979522705078, -32.56587219238281, -7.3955078125, 29.46923828125, 45.842987060546875, -50.680908203125, 8.9376220703125, 81.99615478515625, 14.827392578125, 27.887657165527344, 1.0850753784179688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000158.npy"} {"epoch": 0.3308900523560209, "step": 159, "batch_size": 128, "mean": 13.043231010437012, "std": 34.51919937133789, "min": -90.4290771484375, "p10": -31.009339141845704, "median": 9.463394165039062, "p90": 60.50778808593749, "max": 98.65470886230469, "pos_frac": 0.609375, "sample": [-19.573516845703125, 57.72389221191406, 25.22735595703125, 31.196533203125, -1.830291748046875, -14.18841552734375, -37.6187744140625, 48.80694580078125, -2.534271240234375, 63.207977294921875, -52.81477355957031, 6.132987976074219, -90.4290771484375, -17.834327697753906, 6.049556732177734, 40.26630401611328, 12.486717224121094, -5.228729248046875, -34.924041748046875, 45.20893859863281, 27.9320068359375, -0.7415771484375, -13.230545043945312, -28.61090087890625, 22.384063720703125, 59.841339111328125, 22.958541870117188, 59.63761901855469, 31.92559814453125, -7.2518768310546875, 22.566038131713867, 2.3819580078125, 38.820281982421875, 16.42957305908203, 20.903121948242188, -19.1375732421875, 1.997955322265625, -25.579833984375, -4.004753112792969, -33.707855224609375, 48.452232360839844, 16.156982421875, 23.025390625, 42.233795166015625, -9.529571533203125, 20.008544921875, -30.799224853515625, 3.8457393646240234, -44.70367431640625, -16.83771514892578, 1.0049285888671875, -7.270172119140625, 50.38829040527344, 9.81256103515625, 18.53045654296875, 50.66932678222656, 2.7670249938964844, 49.604583740234375, -10.846237182617188, -3.223876953125, 98.65470886230469, 9.297836303710938, -30.97863006591797, 11.945541381835938, 76.21136474609375, 12.906036376953125, 53.135986328125, -0.999053955078125, -10.147491455078125, 11.672233581542969, 0.6302337646484375, 35.43682861328125, -7.963123321533203, 47.938140869140625, 37.343231201171875, -33.065948486328125, 2.4336090087890625, 63.221649169921875, -17.118240356445312, 18.9268798828125, -20.767120361328125, -38.11572265625, 33.25444793701172, 71.87319946289062, 79.76361083984375, -10.730133056640625, -10.2315673828125, -31.08099365234375, -9.532928466796875, 62.320892333984375, 27.29254913330078, 16.326568603515625, -16.62847900390625, 25.512420654296875, 11.290435791015625, 31.547500610351562, -34.04986572265625, 65.0120849609375, -15.547225952148438, 39.4652099609375, 33.10614013671875, -28.041444778442383, -36.79022216796875, -0.5146942138671875, 85.84320068359375, 77.863525390625, 22.556640625, -53.520050048828125, 50.691795349121094, 7.064754486083984, -3.9238433837890625, -14.8021240234375, 62.062835693359375, -46.219512939453125, 76.27166748046875, -8.726318359375, 50.6295166015625, 35.668212890625, 54.746337890625, 45.22760009765625, 9.5283203125, 8.755496978759766, -3.6766357421875, 71.64666748046875, 5.900299072265625, 9.398468017578125, -9.80224609375, 43.99895095825195], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000159.npy"} {"epoch": 0.33298429319371725, "step": 160, "batch_size": 128, "mean": 20.157310485839844, "std": 30.483272552490234, "min": -46.980194091796875, "p10": -14.232567024230956, "median": 14.852951049804688, "p90": 59.08974456787109, "max": 108.82171630859375, "pos_frac": 0.7578125, "sample": [-8.034088134765625, 45.846588134765625, 0.0, 87.09426879882812, 25.66950225830078, 27.2584228515625, 29.76824951171875, 14.9534912109375, 80.0029296875, -0.070831298828125, 54.31925964355469, 68.60305786132812, 26.4693546295166, 74.64202880859375, 17.131332397460938, -4.21539306640625, 71.8009033203125, 26.001388549804688, 8.155937194824219, 15.506088256835938, 10.832122802734375, 5.5152587890625, 33.53160095214844, 35.26005554199219, 1.6974639892578125, -7.56170654296875, -7.399200439453125, 81.48728942871094, 1.8858661651611328, -46.980194091796875, 14.752410888671875, 106.00811767578125, -40.802711486816406, 55.804893493652344, 18.11480712890625, 3.1641883850097656, 11.578826904296875, 3.576221466064453, 29.302688598632812, 9.497499465942383, 35.81060028076172, 51.08355712890625, 8.91552734375, 0.0, 36.3070068359375, -13.84368896484375, 34.903289794921875, -24.200714111328125, -15.967910766601562, 6.4976654052734375, 27.399261474609375, 36.0023193359375, 108.82171630859375, 72.45138549804688, -23.07427978515625, 18.279754638671875, 26.423583984375, -14.27734375, 9.685501098632812, -13.088752746582031, -11.1466064453125, 53.36750793457031, 7.87152099609375, -38.844482421875, -12.1842041015625, 57.47015380859375, 60.06060791015625, 56.30461120605469, 4.466104507446289, 14.271484375, 35.32599639892578, 1.1580810546875, -0.9097671508789062, 11.575927734375, 32.10394287109375, 32.322052001953125, 68.46170043945312, 50.63191223144531, 35.133148193359375, 19.05926513671875, -25.55206298828125, -6.005889892578125, 42.72008514404297, -31.194011688232422, -12.5538330078125, 58.67366027832031, 29.12957000732422, 12.9669189453125, 43.943695068359375, -14.787353515625, 8.529281616210938, 8.980415344238281, 4.5556640625, -28.4422607421875, 33.885162353515625, -46.948211669921875, -1.5238037109375, 28.525115966796875, 62.841400146484375, 0.53387451171875, 14.18704605102539, 12.150840759277344, -14.213376998901367, 25.857986450195312, 17.7105712890625, 7.296417236328125, 6.784460067749023, -9.05859375, 38.6982421875, 32.27958679199219, 55.661834716796875, 2.0778045654296875, 13.1640625, 44.10499572753906, -6.566192626953125, 26.94709014892578, 55.2353515625, 12.458221435546875, 34.29499816894531, 54.94190979003906, 60.284912109375, 31.19775390625, -34.156341552734375, 33.411376953125, 12.465713500976562, 51.152130126953125, 3.9665985107421875, 4.727407455444336], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000160.npy"} {"epoch": 0.33507853403141363, "step": 161, "batch_size": 128, "mean": 21.274681091308594, "std": 32.994651794433594, "min": -69.96920776367188, "p10": -15.386656951904296, "median": 17.662017822265625, "p90": 69.88068237304687, "max": 87.11712646484375, "pos_frac": 0.7109375, "sample": [-3.4572982788085938, 1.3060798645019531, -4.335906982421875, -68.1171875, 1.274160385131836, 23.079376220703125, 2.5887908935546875, 7.2552490234375, 6.46868896484375, 3.07421875, 4.308048248291016, 78.404296875, 30.333282470703125, -32.3251953125, 5.299285888671875, 2.7605133056640625, 75.13095092773438, 21.27850341796875, 34.589534759521484, -6.362030029296875, 45.365875244140625, 12.4307861328125, 58.853851318359375, 78.174560546875, -21.437320709228516, 3.47222900390625, 6.2656097412109375, 76.9842529296875, 7.1536712646484375, 34.29930114746094, 6.6924896240234375, 27.635225296020508, -6.361785888671875, -15.143218994140625, 24.97644805908203, 23.843032836914062, 35.12583923339844, 52.560791015625, -1.7921867370605469, 13.266746520996094, -5.85040283203125, -44.039398193359375, -28.0760498046875, -11.790336608886719, 12.622795104980469, 46.179046630859375, -14.539154052734375, -4.035888671875, 0.0, 40.944000244140625, 17.43316650390625, 37.70257568359375, 8.588623046875, 0.31555747985839844, 11.944091796875, -3.16680908203125, 53.68681335449219, 54.66267395019531, 42.54181671142578, 26.638221740722656, 57.449462890625, 65.407470703125, -20.20111083984375, 17.890869140625, -12.993408203125, 66.5819091796875, 56.838623046875, 74.12750244140625, -17.811767578125, 14.79547119140625, 79.5369873046875, 27.28829574584961, 38.978424072265625, 70.2381591796875, -29.263092041015625, 71.21856689453125, 34.37671661376953, -2.2767333984375, 3.2491455078125, 40.17634582519531, -15.469825744628906, -0.36474609375, 87.11712646484375, 69.25274658203125, 30.675201416015625, 73.9403076171875, 2.5805816650390625, 53.15106201171875, 83.49557495117188, -1.5147457122802734, -1.31756591796875, 35.196319580078125, -15.35101318359375, 41.106109619140625, -4.70379638671875, 29.52288818359375, 68.2705078125, -4.667625427246094, 36.147705078125, -6.00518798828125, 23.72527313232422, 61.232147216796875, 22.68316650390625, -16.85882568359375, 10.341804504394531, 43.4627685546875, -15.757766723632812, 29.467132568359375, 40.44379425048828, 51.18585205078125, 60.180389404296875, 69.72747802734375, -14.391281127929688, 39.9603271484375, 27.974218368530273, -53.67413330078125, 71.87471008300781, 8.663345336914062, 13.018043518066406, 0.0, -3.86138916015625, 70.79092407226562, 60.125091552734375, -69.96920776367188, 33.058319091796875, 19.97027587890625, 2.6628570556640625, 53.7734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000161.npy"} {"epoch": 0.33717277486910996, "step": 162, "batch_size": 128, "mean": 19.14374351501465, "std": 31.588211059570312, "min": -75.93800354003906, "p10": -21.62876510620117, "median": 16.943992614746094, "p90": 60.82508239746094, "max": 89.64834594726562, "pos_frac": 0.7578125, "sample": [-7.916534423828125, 7.450248718261719, 67.54071044921875, 42.894996643066406, 4.0249786376953125, 15.743408203125, -23.065032958984375, 26.628524780273438, 5.993217468261719, 27.887802124023438, 53.13555145263672, 74.9825439453125, 16.28356170654297, 27.16226577758789, -5.3984375, -7.489044189453125, 26.28619384765625, 28.5419921875, 1.6481399536132812, 4.173206329345703, -29.297149658203125, 38.16021728515625, -30.42431640625, 31.284454345703125, 17.326553344726562, 32.13495635986328, 61.101806640625, 12.92218017578125, -42.40242004394531, 6.279731750488281, -0.9875411987304688, 17.0408935546875, 42.990997314453125, -29.69440460205078, 68.474365234375, 8.344833374023438, 17.31475830078125, -62.630828857421875, -9.720369338989258, 37.769012451171875, 3.41668701171875, 54.886993408203125, 3.351133346557617, 65.84014892578125, 64.05242919921875, -75.93800354003906, 60.048095703125, 28.15386962890625, 28.48919677734375, 64.62884521484375, 1.4726219177246094, 39.7144775390625, 43.34307861328125, 6.98492431640625, 30.739151000976562, -3.516956329345703, -21.013221740722656, 43.68644714355469, 71.0166015625, 27.478271484375, 68.98809814453125, 88.07382202148438, -1.88555908203125, -0.1156158447265625, 41.4461669921875, 55.91716003417969, 57.92979431152344, 70.56072235107422, -18.813018798828125, 51.108184814453125, -0.0022640228271484375, 8.73624038696289, -28.202301025390625, 18.7733154296875, 89.64834594726562, -27.053382873535156, -28.177536010742188, 1.544647216796875, 16.847091674804688, 54.587921142578125, -6.450584411621094, 18.824462890625, 44.16071701049805, 27.076507568359375, 60.84027099609375, 58.10736083984375, 7.286174774169922, -1.22979736328125, 28.413482666015625, 28.077163696289062, 10.829261779785156, 12.6815185546875, 8.161077499389648, -58.37432861328125, 0.631103515625, 25.641555786132812, 57.566551208496094, 37.42791748046875, 16.64947509765625, 60.818572998046875, 43.28483581542969, -38.6873779296875, 47.482666015625, 8.660659790039062, 1.7734451293945312, 60.48248291015625, -13.649932861328125, 4.5152587890625, -3.6918601989746094, 48.535972595214844, -26.4105224609375, 5.306419372558594, -9.715972900390625, 41.23388671875, 28.311447143554688, 38.08453369140625, 12.257740020751953, -3.3926258087158203, 28.0555419921875, 19.5238037109375, 7.374048233032227, -20.166473388671875, 0.950592041015625, 0.896636962890625, 5.613609313964844, 60.80348205566406, 27.0447998046875, 9.54083251953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000162.npy"} {"epoch": 0.3392670157068063, "step": 163, "batch_size": 128, "mean": 22.31351089477539, "std": 27.983539581298828, "min": -46.55133056640625, "p10": -12.475260162353516, "median": 22.714614868164062, "p90": 60.42236175537109, "max": 85.71917724609375, "pos_frac": 0.7578125, "sample": [22.363494873046875, -1.2688369750976562, 42.19916534423828, -14.373321533203125, 27.579742431640625, -12.475639343261719, 22.787979125976562, 24.026092529296875, 34.98884582519531, 9.081451416015625, 2.2055206298828125, 64.03671264648438, 42.49894714355469, 13.763458251953125, 71.58929443359375, 16.215606689453125, 11.715072631835938, -12.47509765625, 7.5458526611328125, -7.8569183349609375, -14.778472900390625, 37.52412796020508, 26.48638916015625, -1.4603042602539062, 14.61236572265625, 50.32218933105469, -5.47361946105957, 15.496902465820312, 47.7052001953125, -2.4305419921875, -6.0315704345703125, 60.025787353515625, -1.50634765625, 44.80694580078125, 28.977294921875, 25.992584228515625, 12.292266845703125, 48.428619384765625, -7.1641845703125, -4.9622650146484375, 38.506805419921875, 33.28961181640625, -27.619186401367188, 11.213981628417969, 8.81304931640625, 23.098175048828125, -46.55133056640625, 62.699920654296875, 46.38856506347656, -31.758544921875, 36.597713470458984, 5.9898681640625, 42.898101806640625, 39.858917236328125, -2.4913101196289062, 57.92286682128906, 19.34625244140625, 61.875244140625, 64.9671630859375, 38.842681884765625, 3.367584228515625, -3.2347412109375, -11.356689453125, 34.76721954345703, -8.503433227539062, -12.210906982421875, 9.22138786315918, 48.51365661621094, 6.825164794921875, 36.54307556152344, -45.204833984375, 61.34770202636719, 24.19709014892578, 24.171768188476562, 12.3673095703125, -16.4912109375, 76.2974853515625, -15.932342529296875, 64.49966430664062, 63.82818603515625, 33.68079376220703, 45.66368103027344, 20.14190673828125, 18.058391571044922, 42.41815948486328, 40.481849670410156, 36.25830078125, 1.449493408203125, 62.26570129394531, 49.26835632324219, 6.058950424194336, 5.9605712890625, 22.379600524902344, 85.71917724609375, 36.548095703125, -31.096832275390625, -26.181724548339844, 26.591867446899414, 76.07418823242188, 31.024261474609375, 53.051422119140625, 48.942718505859375, -18.327926635742188, 56.0843505859375, -5.7514801025390625, 18.71728515625, 59.5899658203125, 18.342819213867188, 22.641250610351562, 32.03577423095703, 57.651123046875, 36.10340881347656, 16.79522705078125, 2.2412071228027344, 11.640762329101562, 14.937957763671875, 32.81317138671875, 56.65577697753906, 29.181427001953125, 65.52957916259766, -3.6294918060302734, 4.0551605224609375, -38.716766357421875, 56.608551025390625, 48.55952453613281, 23.600830078125, 8.78996467590332, -4.6912994384765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000163.npy"} {"epoch": 0.3413612565445026, "step": 164, "batch_size": 128, "mean": 19.32340431213379, "std": 30.270366668701172, "min": -52.88468933105469, "p10": -12.225566101074218, "median": 15.660584449768066, "p90": 60.92125091552734, "max": 97.37332153320312, "pos_frac": 0.75, "sample": [-10.247833251953125, 20.759681701660156, -2.739593505859375, 53.708221435546875, 0.22495269775390625, -0.7986907958984375, -6.100727081298828, -4.796173095703125, 28.478347778320312, 32.98773193359375, 32.699951171875, -12.393783569335938, 43.41725158691406, 43.538604736328125, 22.364959716796875, -45.70233154296875, -1.732818603515625, 52.574432373046875, 13.909265518188477, 66.24874877929688, 0.939239501953125, -44.726165771484375, 48.129791259765625, 13.73651123046875, 11.669380187988281, -23.950515747070312, 27.94635009765625, 5.5047149658203125, -12.153472900390625, 6.1751556396484375, 37.2294921875, 0.47137451171875, 5.00054931640625, 59.44847106933594, -5.2100372314453125, 49.201385498046875, -38.36909484863281, 3.1513214111328125, 1.4592437744140625, -41.690399169921875, -9.31365966796875, 68.99127197265625, 44.505096435546875, 29.1383056640625, 67.18624877929688, 7.6236724853515625, 53.453857421875, 0.336669921875, 0.291839599609375, 2.3314208984375, -27.844772338867188, 13.324378967285156, 19.31787109375, 31.66404151916504, -14.270065307617188, 62.7657470703125, -1.7476654052734375, 41.925933837890625, 32.85528564453125, 30.0052490234375, 0.8224639892578125, 8.168792724609375, 62.10498046875, 64.94082641601562, -11.940811157226562, 65.70164489746094, 22.92284393310547, -0.550201416015625, 45.15380859375, 13.508636474609375, 82.41824340820312, 30.240020751953125, 67.95068359375, 1.027008056640625, -8.352439880371094, 0.42071533203125, 21.825469970703125, 3.5787525177001953, -46.983978271484375, 21.88690185546875, 26.94366455078125, 53.97052001953125, -12.919788360595703, 2.4904251098632812, -0.7602386474609375, 35.26971435546875, 54.32916259765625, 60.456817626953125, 39.519500732421875, 25.685699462890625, -5.721736907958984, 47.278053283691406, 0.5320816040039062, 44.175689697265625, 9.696044921875, 49.709930419921875, 48.88665771484375, 62.08990478515625, -11.42047119140625, 1.2081146240234375, -33.12933349609375, 29.397506713867188, 0.003936767578125, 28.883262634277344, 8.471900939941406, -6.853118896484375, 56.81060791015625, -11.41552734375, 45.59593200683594, 9.420211791992188, 30.836288452148438, 44.566162109375, 9.589668273925781, 18.06597900390625, -52.88468933105469, -20.31365966796875, 8.061332702636719, -7.79681396484375, 97.2447509765625, 18.59326171875, 28.25357437133789, 45.875701904296875, 14.0159912109375, 32.36468505859375, 43.88641357421875, 62.00492858886719, 97.37332153320312, 17.305177688598633], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000164.npy"} {"epoch": 0.34345549738219894, "step": 165, "batch_size": 128, "mean": 16.120891571044922, "std": 28.859432220458984, "min": -61.71372985839844, "p10": -15.875078582763672, "median": 9.139314651489258, "p90": 54.0265396118164, "max": 88.64114379882812, "pos_frac": 0.6640625, "sample": [5.9185943603515625, 53.8397216796875, -5.135780334472656, 43.04986572265625, -17.96282958984375, 3.2999343872070312, 31.48626708984375, 3.432952880859375, 47.851318359375, 45.219146728515625, 65.945556640625, 3.937957763671875, -58.4901123046875, 43.288543701171875, 61.1861572265625, 15.729095458984375, 2.72637939453125, 65.94992065429688, -7.6019439697265625, 45.0538330078125, 30.135406494140625, -45.61590576171875, 54.03178405761719, 40.357933044433594, 8.097091674804688, -25.301555633544922, -22.6593017578125, -9.610244750976562, 18.162506103515625, 35.29669189453125, 3.97576904296875, -8.92181396484375, -3.9119873046875, 1.1708526611328125, 0.9632034301757812, 13.141708374023438, 28.529632568359375, -4.4358673095703125, -6.658483505249023, -10.417129516601562, 16.8385009765625, 15.4964599609375, 53.367279052734375, -20.944976806640625, 14.508544921875, 54.0242919921875, -0.7411251068115234, 0.1604461669921875, -16.969802856445312, 9.7557373046875, 85.29022216796875, -0.5326080322265625, 4.11199951171875, -1.7628326416015625, 60.09588623046875, 27.950729370117188, -8.43499755859375, 5.8547210693359375, -4.774509429931641, 13.074249267578125, -3.405548095703125, 66.72943115234375, -8.789314270019531, 46.87933349609375, 26.380706787109375, -15.7718505859375, 24.034774780273438, -3.3140411376953125, -5.217376708984375, -7.1552886962890625, 0.6197509765625, -1.8521499633789062, 18.16888427734375, 18.835153579711914, 29.514801025390625, 4.843906402587891, 39.027557373046875, -25.96575927734375, 42.15313720703125, 15.05780029296875, -25.5928955078125, 29.724281311035156, 67.80960083007812, -15.68231201171875, 8.522891998291016, 39.006072998046875, -1.2687759399414062, 63.3809814453125, 50.665283203125, 20.0616455078125, 12.63140869140625, -0.332489013671875, 38.2467041015625, 43.807193756103516, 50.46974182128906, -16.115943908691406, -8.043464660644531, 24.182373046875, 44.37712860107422, 5.586189270019531, 88.64114379882812, -1.7762832641601562, 34.35493469238281, 41.86412048339844, 58.54450988769531, -30.13257598876953, -7.676704406738281, 15.691116333007812, 53.46754455566406, 2.41619873046875, 1.1376113891601562, 5.8749542236328125, 60.234619140625, 32.30682373046875, 50.92274475097656, 46.59269714355469, 5.4613037109375, 14.233268737792969, 8.401046752929688, 47.417266845703125, -1.9072647094726562, 45.084808349609375, -61.71372985839844, -27.837432861328125, -3.4687271118164062, 55.93675231933594, -8.445556640625, -5.783721923828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000165.npy"} {"epoch": 0.34554973821989526, "step": 166, "batch_size": 128, "mean": 14.69461727142334, "std": 35.251739501953125, "min": -77.21087646484375, "p10": -25.195880126953124, "median": 12.222084045410156, "p90": 61.8302391052246, "max": 104.21173095703125, "pos_frac": 0.6640625, "sample": [14.60675048828125, 3.2984619140625, -28.9571533203125, 14.956146240234375, -3.21636962890625, 12.365989685058594, 64.63973999023438, 58.55706787109375, -3.281961441040039, 3.0629520416259766, 9.104873657226562, -7.156196594238281, 104.21173095703125, -16.598388671875, 2.8112030029296875, -6.9127197265625, 23.382659912109375, 28.177230834960938, 3.7559890747070312, 0.9497528076171875, -1.6521949768066406, -31.768585205078125, 23.926803588867188, 66.95497131347656, -57.546142578125, -25.555877685546875, 1.7887773513793945, 31.218353271484375, -56.04612731933594, -19.11749267578125, -9.10772705078125, 18.648193359375, 28.805145263671875, 71.1734619140625, 60.417877197265625, 77.77880859375, 3.192169189453125, -8.94976806640625, -15.685882568359375, -73.47262573242188, 6.294912338256836, -28.287094116210938, 4.878089904785156, -2.3259201049804688, -16.916107177734375, 64.96603393554688, 25.483257293701172, 19.645553588867188, -18.486175537109375, 36.08746337890625, -2.4976348876953125, 88.56097412109375, -46.381072998046875, 69.1083984375, 94.72796630859375, -51.85774230957031, -35.0438232421875, 93.54559326171875, 30.868072509765625, 41.55970001220703, -23.957054138183594, -11.30718994140625, -19.3984375, 35.45237731933594, 49.92457580566406, -25.041595458984375, 53.14129638671875, 37.6102294921875, 7.459503173828125, -15.153305053710938, 20.867095947265625, 23.0118408203125, 72.08671569824219, -8.543172836303711, -14.382125854492188, 21.93614959716797, 13.691932678222656, -1.0386123657226562, 13.656410217285156, 13.44476318359375, -5.479194641113281, 46.349029541015625, 18.191543579101562, 10.770431518554688, 12.77484130859375, 74.06866455078125, 2.4952392578125, 37.762603759765625, 17.288360595703125, -40.371337890625, 4.6741943359375, -6.684013366699219, 12.319076538085938, 52.837158203125, 51.623138427734375, 27.486358642578125, 16.15557861328125, 50.28362274169922, 29.50250244140625, 6.42626953125, 22.58447265625, 0.8492488861083984, -2.2384185791015625, 44.63079833984375, 1.7811412811279297, 30.193084716796875, 53.41352844238281, 47.293487548828125, -10.177665710449219, 60.62616729736328, -44.6239013671875, -77.21087646484375, 12.125091552734375, 42.50431823730469, 56.481964111328125, -20.896812438964844, 13.706146240234375, -0.6876220703125, 19.74530029296875, -1.9539642333984375, 59.925506591796875, -1.1111297607421875, 103.3079833984375, 13.670089721679688, 1.6410675048828125, 34.3468017578125, 10.943145751953125, 11.416213989257812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000166.npy"} {"epoch": 0.34764397905759165, "step": 167, "batch_size": 128, "mean": 17.45690155029297, "std": 31.923097610473633, "min": -61.90531921386719, "p10": -18.411463928222652, "median": 11.898628234863281, "p90": 61.74599151611328, "max": 96.82943725585938, "pos_frac": 0.7265625, "sample": [46.06865692138672, 5.6993560791015625, 15.04510498046875, 44.073089599609375, 10.549972534179688, 52.517333984375, 44.25677490234375, 71.92453002929688, 33.71142578125, 15.299072265625, 58.221771240234375, -42.751312255859375, 66.26382446289062, 1.2030181884765625, 10.70550537109375, 76.86416625976562, 16.628082275390625, 68.88449096679688, 0.132659912109375, 12.080047607421875, 2.3264312744140625, 37.75946044921875, -54.6617431640625, -61.90531921386719, -2.7904586791992188, 0.26676177978515625, -27.497146606445312, 53.10310363769531, 34.36517333984375, 55.02641296386719, 28.22674560546875, 61.48258972167969, 15.08563232421875, 62.360595703125, 36.69731140136719, 7.015716552734375, 15.185279846191406, -33.435791015625, 19.1041259765625, -9.027961730957031, 68.75071716308594, -4.377494812011719, 7.525177001953125, 2.999359130859375, 0.0466156005859375, -60.04045104980469, 0.0, 1.132904052734375, 1.456085205078125, 33.190338134765625, 14.066665649414062, 72.96923828125, -31.697959899902344, 63.603485107421875, -15.96551513671875, 15.809173583984375, 45.324913024902344, 9.955474853515625, 32.61045837402344, 1.04998779296875, -32.64349365234375, 3.20068359375, 81.44461059570312, 0.2978782653808594, 24.5928955078125, 11.717208862304688, 17.193252563476562, -16.015090942382812, 36.46336364746094, 4.104318618774414, -1.466400146484375, 1.2511940002441406, 38.21752166748047, -36.93743896484375, 47.54265594482422, 40.998451232910156, 56.74806213378906, -1.6534423828125, 25.7685546875, 2.86981201171875, -1.6669235229492188, 45.60890197753906, 36.68536376953125, -25.988525390625, 54.979949951171875, -5.621307373046875, 17.048980712890625, -6.442842483520508, 36.0230712890625, 38.43704605102539, 35.254486083984375, 6.314918518066406, 42.17118835449219, 19.111541748046875, 36.69941711425781, 79.832763671875, -5.653766632080078, -17.616012573242188, -20.26751708984375, -9.660247802734375, -4.603309631347656, 8.25480842590332, -10.49871826171875, 67.75125122070312, -9.424545288085938, 9.770965576171875, 96.82943725585938, 21.357391357421875, 44.637481689453125, -12.35516357421875, 7.7784423828125, 42.0400390625, -30.79534912109375, -2.938446044921875, -3.42645263671875, 3.48236083984375, 41.43414306640625, 0.176788330078125, -3.62469482421875, 51.487449645996094, 40.12156677246094, 23.979026794433594, 10.172348022460938, -1.8223419189453125, -53.9063720703125, 51.09844970703125, 7.0361328125, 77.05183410644531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000167.npy"} {"epoch": 0.34973821989528797, "step": 168, "batch_size": 128, "mean": 23.155166625976562, "std": 31.993560791015625, "min": -67.62136840820312, "p10": -11.33554229736328, "median": 18.821395874023438, "p90": 63.61311340332031, "max": 94.46319580078125, "pos_frac": 0.7421875, "sample": [32.61405944824219, 17.085845947265625, 75.81436157226562, 64.3890380859375, 44.800933837890625, 30.680885314941406, 46.32533264160156, 23.03076934814453, 50.37249755859375, -8.135536193847656, 51.170654296875, -10.976760864257812, 4.9571533203125, 6.2652435302734375, -0.15558624267578125, 3.354583740234375, 27.955001831054688, 31.559112548828125, -6.082393646240234, 51.11824035644531, 16.49634552001953, 58.860687255859375, 17.972274780273438, 23.48162841796875, 58.36767578125, -2.296337127685547, 23.829788208007812, 48.5769157409668, -1.5537796020507812, 2.0805130004882812, 0.0, 13.960006713867188, 75.29022216796875, 16.091140747070312, 62.9678955078125, 28.953811645507812, 17.271453857421875, 6.717193603515625, 32.11016845703125, 4.529296875, 65.99896240234375, 17.4730224609375, -1.58624267578125, 42.620758056640625, -6.13006591796875, 94.4124755859375, 18.345306396484375, 38.575927734375, 78.90353393554688, -14.55535888671875, 49.637939453125, -1.789154052734375, 28.918392181396484, 13.931396484375, -46.21112060546875, 35.46269989013672, 50.05541229248047, 55.226776123046875, 47.11280059814453, 6.8531494140625, -20.245025634765625, -22.76879119873047, 3.19281005859375, 0.9744720458984375, 45.60357666015625, -0.43625450134277344, 58.264404296875, 55.96234130859375, 94.46319580078125, 26.588287353515625, 72.3341064453125, 13.744991302490234, -67.62136840820312, 42.70054626464844, 64.39602661132812, 0.3200855255126953, 9.793563842773438, 48.65705871582031, 5.6143646240234375, -2.2156982421875, -5.207794189453125, 6.783843994140625, 41.94807434082031, 19.12176513671875, 64.28121948242188, 49.49761962890625, 62.49372863769531, 40.837158203125, 76.53472900390625, 53.648529052734375, -28.014148712158203, -4.45672607421875, 63.3267822265625, -12.172698974609375, 7.46807861328125, 45.48309326171875, -13.442054748535156, 16.76251220703125, 31.189743041992188, -55.46989440917969, 18.521026611328125, 9.488510131835938, 52.49408721923828, 33.87109375, 27.856048583984375, 75.28839111328125, -0.7767715454101562, -3.60528564453125, 4.280937194824219, -0.755706787109375, 6.889434814453125, -1.544097900390625, -28.3267822265625, -36.64457702636719, 52.64558410644531, 21.470916748046875, 44.8125, 62.25505065917969, 61.41754150390625, 46.22564697265625, -0.462158203125, 35.00433349609375, 8.126861572265625, -7.39508056640625, -34.933685302734375, 11.6748046875, 65.40396118164062, -62.464569091796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000168.npy"} {"epoch": 0.3518324607329843, "step": 169, "batch_size": 128, "mean": 26.455947875976562, "std": 33.31447982788086, "min": -55.11091613769531, "p10": -7.596064758300781, "median": 17.673072814941406, "p90": 72.16372375488281, "max": 106.0230712890625, "pos_frac": 0.78125, "sample": [-2.0755996704101562, 44.21778106689453, 88.31527709960938, 37.11083984375, -19.144073486328125, 13.60538101196289, 4.028175354003906, 6.934173583984375, 2.53857421875, 15.625015258789062, 19.389434814453125, 55.73402404785156, 56.932132720947266, 59.16357421875, -37.189178466796875, -2.3762435913085938, 53.358612060546875, 11.50384521484375, 5.14277458190918, -3.02056884765625, -3.135223388671875, 16.70722198486328, 49.71410369873047, 15.577194213867188, 50.61442565917969, 65.35003662109375, 28.994384765625, -20.807891845703125, 37.965423583984375, 46.980072021484375, 60.75299072265625, 64.06936645507812, 51.24053955078125, -7.3046875, 4.2469482421875, 9.210182189941406, -14.030763626098633, -55.11091613769531, 27.998123168945312, -2.80474853515625, -7.574371337890625, 65.33279418945312, -10.418281555175781, 14.813697814941406, 0.0, -40.847084045410156, 9.188148498535156, 33.75946044921875, 19.764007568359375, 25.190948486328125, 1.93719482421875, 9.41168212890625, 106.0230712890625, 61.31695556640625, 40.620086669921875, 50.726959228515625, -49.24047088623047, 51.68724060058594, 0.3080158233642578, -0.8385238647460938, 72.0802001953125, 47.947113037109375, 26.634979248046875, -10.019134521484375, 39.98162841796875, 72.35861206054688, 62.406463623046875, 82.54336547851562, 17.778717041015625, 67.68797302246094, 1.141204833984375, 87.43798828125, 7.649620056152344, 1.030029296875, 40.31951904296875, -33.348297119140625, 12.940071105957031, 76.69882202148438, 47.11375427246094, -7.527069091796875, 46.68536376953125, 9.655204772949219, 16.574304580688477, 48.157257080078125, 26.966758728027344, 51.23225402832031, 62.1787109375, -7.6466827392578125, 39.13323974609375, 34.4696044921875, 0.5245285034179688, 37.37969970703125, 5.3741912841796875, 99.96041870117188, 31.51202392578125, 9.686222076416016, 14.28436279296875, 3.864898681640625, 82.82864379882812, 38.9315185546875, 6.138826370239258, 50.155670166015625, 91.09344482421875, 16.7603759765625, -29.267181396484375, -28.715492248535156, 7.79913330078125, -1.8763885498046875, 91.25946044921875, 17.567428588867188, 46.248687744140625, -0.12601280212402344, 69.0369873046875, 53.124267578125, 49.958648681640625, 0.0, 15.045196533203125, 3.394683837890625, 23.2664794921875, 42.450008392333984, 14.214759826660156, -0.21103286743164062, 12.4678955078125, 13.72824478149414, 85.81463623046875, 85.81451416015625, 74.6700439453125, -5.242271423339844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000169.npy"} {"epoch": 0.3539267015706806, "step": 170, "batch_size": 128, "mean": 22.573013305664062, "std": 31.463598251342773, "min": -55.0826416015625, "p10": -10.397607421874998, "median": 20.144119262695312, "p90": 64.07347869873047, "max": 109.31402587890625, "pos_frac": 0.7578125, "sample": [5.727020263671875, 13.587554931640625, 51.00346374511719, -7.801239013671875, 47.831268310546875, 10.43109130859375, 0.07122039794921875, 101.044677734375, 8.118568420410156, 9.782135009765625, 31.102569580078125, 87.01309204101562, 41.85771179199219, 26.016048431396484, -38.267333984375, 98.85519409179688, -24.46038818359375, 31.283645629882812, -0.9810256958007812, 14.898460388183594, -32.37664794921875, 5.09747314453125, 25.702049255371094, 32.342010498046875, 7.7913818359375, -31.566261291503906, -19.991008758544922, 3.6458139419555664, 64.8250732421875, -7.92510986328125, 26.98089599609375, 27.92827606201172, 40.62013244628906, 6.957731246948242, 15.875732421875, -2.5252532958984375, 0.492462158203125, 3.8533477783203125, 46.27203369140625, 58.555694580078125, 2.0077056884765625, 35.34906005859375, 1.5609893798828125, 63.649200439453125, 44.14009094238281, 35.446441650390625, 34.09796142578125, -2.986541748046875, -21.7493896484375, 64.93328857421875, 64.68363952636719, 51.476051330566406, 48.3951416015625, -45.238807678222656, 23.949188232421875, -5.3663330078125, 67.49688720703125, -12.310302734375, 62.04136657714844, 44.75159454345703, 16.15570068359375, 73.39468383789062, 56.60066223144531, -22.351470947265625, 43.3682975769043, 45.378684997558594, -2.69158935546875, 15.4918212890625, 59.31221008300781, 3.604400634765625, 31.201766967773438, 53.509063720703125, 8.079809188842773, 19.9764404296875, 45.384315490722656, 41.220001220703125, 0.0, 68.78363037109375, 20.311798095703125, 49.18797302246094, -30.003448486328125, 45.08135986328125, 5.71173095703125, 44.2764892578125, 56.76409912109375, 34.94877624511719, -29.97662353515625, 11.8526611328125, 37.56982421875, -55.0826416015625, 26.413604736328125, 23.198974609375, -5.19970703125, 40.536285400390625, 0.8014450073242188, -3.907247543334961, 70.98110961914062, -1.3879547119140625, 29.556640625, 109.31402587890625, -0.356231689453125, -8.789901733398438, 22.61949920654297, -3.5975341796875, -7.439788818359375, 7.451690673828125, 25.50439453125, 11.362075805664062, 1.6221847534179688, 18.678436279296875, -2.2328643798828125, 3.1738128662109375, 83.83837890625, 15.83770751953125, -27.851951599121094, 21.260452270507812, 6.801666259765625, 56.803955078125, 26.972564697265625, 3.164308547973633, -0.9003238677978516, 36.882904052734375, 70.70829772949219, 58.60643768310547, 6.203544616699219, 35.42353057861328, -9.577880859375, 63.811981201171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000170.npy"} {"epoch": 0.35602094240837695, "step": 171, "batch_size": 128, "mean": 23.315813064575195, "std": 31.948843002319336, "min": -34.74065399169922, "p10": -18.11961975097656, "median": 20.377487182617188, "p90": 70.45777587890625, "max": 93.25582885742188, "pos_frac": 0.7421875, "sample": [87.58688354492188, 17.2637939453125, 29.147964477539062, 17.12432861328125, 61.702880859375, 63.74226379394531, 61.56407928466797, -24.11126708984375, -11.620851516723633, 16.141220092773438, -19.977340698242188, 0.038330078125, 74.90830993652344, -19.480712890625, 1.3286590576171875, -2.5298805236816406, -14.095088958740234, 26.689790725708008, 30.70269775390625, 75.58502197265625, 52.30360412597656, -3.5497894287109375, 18.854644775390625, 79.47586059570312, 61.64802551269531, 25.28692626953125, -23.502349853515625, 10.792793273925781, -24.26849365234375, 21.19891357421875, 8.630050659179688, 10.700546264648438, -10.800506591796875, 43.51280212402344, 50.513671875, 72.495849609375, 54.14009094238281, 2.75958251953125, 26.177715301513672, -4.767677307128906, 13.001251220703125, -11.83251953125, -6.3450927734375, -7.2884521484375, -14.830162048339844, 78.30743408203125, 75.51873779296875, 12.897174835205078, -34.483734130859375, 5.248809814453125, 65.74484252929688, 28.9984130859375, 8.993194580078125, 62.0582275390625, 30.06634521484375, 25.44287109375, 38.400146484375, 71.70635986328125, 78.1005859375, -19.28021240234375, 52.146484375, 27.335784912109375, -6.2848968505859375, 55.393028259277344, 30.431522369384766, 60.34803771972656, 29.642059326171875, 13.109319686889648, 53.120513916015625, 40.76100158691406, 27.198936462402344, 74.44873046875, -3.6672821044921875, 3.1150054931640625, -31.1964111328125, -14.7139892578125, 71.91520690917969, 4.056156158447266, -28.27850341796875, 4.6138916015625, -0.2150135040283203, 2.5142288208007812, -13.730133056640625, -25.397994995117188, 19.556060791015625, -30.941268920898438, 47.074275970458984, 44.790069580078125, -4.196319580078125, 23.861297607421875, 47.34195327758789, 91.87290954589844, -34.74065399169922, 12.222602844238281, 28.809356689453125, 56.83616638183594, 52.04638671875, 2.4828014373779297, 66.12393188476562, 26.108154296875, 19.251571655273438, 44.492889404296875, -26.351394653320312, 11.486846923828125, 22.941680908203125, 16.4046630859375, 31.939224243164062, 58.71929931640625, 24.245933532714844, 2.268768310546875, 19.50836181640625, -17.622222900390625, 59.08122253417969, 54.117431640625, 48.88775634765625, 15.046173095703125, 23.55987548828125, 5.12554931640625, 22.131210327148438, 11.283279418945312, 93.25582885742188, 48.46922302246094, -8.767681121826172, 0.7572555541992188, -3.2912139892578125, 69.92266845703125, 25.516571044921875, -7.607513427734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000171.npy"} {"epoch": 0.3581151832460733, "step": 172, "batch_size": 128, "mean": 15.538994789123535, "std": 33.26405715942383, "min": -83.81842041015625, "p10": -17.219871520996094, "median": 14.240211486816406, "p90": 61.38524627685546, "max": 82.9017333984375, "pos_frac": 0.6796875, "sample": [14.515228271484375, -9.67193603515625, 7.604766845703125, 19.148681640625, 31.238540649414062, 11.755516052246094, -2.932861328125, -8.272666931152344, 6.010887145996094, -0.69940185546875, -2.1612014770507812, -0.2710113525390625, -61.7371826171875, 5.486144065856934, 44.22340393066406, 12.083703994750977, 33.060890197753906, 71.5230712890625, 38.42724609375, 12.798431396484375, -8.1326904296875, 8.74090576171875, 25.02777099609375, 74.08718872070312, 4.99847412109375, 43.129119873046875, 35.77919006347656, -7.031890869140625, -13.596748352050781, 56.48211669921875, 4.5699462890625, 26.984466552734375, -64.3419189453125, 6.045684814453125, 16.3050537109375, 46.17756652832031, -16.961578369140625, 69.41021728515625, -17.822555541992188, -5.1205291748046875, -11.119669914245605, 15.597381591796875, -2.652923583984375, -1.4447803497314453, 73.85635375976562, -83.81842041015625, -5.7569122314453125, 34.98091125488281, 71.05926513671875, 54.3319091796875, -6.275604248046875, -24.526947021484375, 20.773284912109375, 0.5333404541015625, 30.27911376953125, 14.473609924316406, 60.67010498046875, 66.16172790527344, -14.74273681640625, -4.212974548339844, -2.76910400390625, 19.58892822265625, 23.297195434570312, -2.9359588623046875, 56.037872314453125, 44.27862548828125, -6.35845947265625, 7.3857574462890625, 20.135162353515625, 9.270042419433594, 81.72906494140625, -59.54095458984375, 2.6959762573242188, 39.963134765625, 69.0140380859375, 55.934722900390625, 1.1511688232421875, 2.4550704956054688, 17.50323486328125, -65.62557983398438, 26.897430419921875, 10.136482238769531, 26.715118408203125, -55.61383056640625, 16.572059631347656, 47.389373779296875, 10.557785034179688, 64.03125, 15.35858154296875, -42.72491455078125, 66.14312744140625, -8.20121955871582, 63.5975341796875, -33.1409912109375, -36.87784957885742, 20.639663696289062, 23.81500244140625, 17.338104248046875, -20.09543800354004, 14.006813049316406, -5.00323486328125, 82.9017333984375, 45.96759033203125, 41.6323356628418, 24.036636352539062, 25.280059814453125, -9.452346801757812, 19.939926147460938, 36.844234466552734, 7.995269775390625, 15.29156494140625, 50.72187042236328, 32.69233703613281, 59.07476806640625, 57.554656982421875, 39.86183166503906, -6.458770751953125, 9.9239501953125, 7.942964553833008, 53.9140625, -72.84515380859375, 63.05390930175781, -1.3536834716796875, 8.70343017578125, 0.0, 39.08428955078125, 33.77903747558594, -6.87188720703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000172.npy"} {"epoch": 0.36020942408376966, "step": 173, "batch_size": 128, "mean": 19.203231811523438, "std": 32.040828704833984, "min": -66.53952026367188, "p10": -16.604188919067376, "median": 19.375442504882812, "p90": 59.65892639160156, "max": 114.81353759765625, "pos_frac": 0.7109375, "sample": [-1.48443603515625, -66.53952026367188, -44.560455322265625, 26.081314086914062, 27.48828125, -4.092742919921875, 47.923004150390625, -38.766876220703125, 46.227294921875, 104.78204345703125, -9.572284698486328, 24.540374755859375, -14.707714080810547, 35.094482421875, 45.923797607421875, 59.00054931640625, 9.12908935546875, 4.2453460693359375, 20.698211669921875, 76.16378784179688, 62.47222900390625, 12.845870971679688, 6.8353729248046875, 34.634521484375, 22.08538055419922, 14.79119873046875, -4.07794189453125, 7.1890869140625, 34.435821533203125, 48.44535827636719, 3.09039306640625, 43.05418395996094, -3.50982666015625, 6.055057525634766, 7.06414794921875, 31.909576416015625, -33.086029052734375, 25.58349609375, 20.754459381103516, 40.624786376953125, -12.78021240234375, 24.617919921875, 37.127838134765625, 40.781494140625, 50.08882141113281, 36.63128662109375, -2.363006591796875, -5.8953857421875, 3.0167007446289062, 9.825927734375, -9.94342041015625, -39.47591018676758, -25.839462280273438, 114.81353759765625, 13.2122802734375, -28.206329345703125, 6.985939025878906, -21.029296875, 30.390548706054688, 94.01986694335938, 58.42833709716797, 61.26214599609375, 2.0277442932128906, 52.62518310546875, 55.251556396484375, 59.189910888671875, -46.3807373046875, 29.64117431640625, 22.979736328125, 18.760955810546875, 26.001548767089844, 1.7782669067382812, 65.96856689453125, 46.77748107910156, 3.7030029296875, 45.21044921875, 13.475128173828125, 66.26873779296875, -3.516742706298828, 13.519966125488281, 24.29241943359375, -28.13336181640625, 3.0770606994628906, 4.583343505859375, 54.16009521484375, 6.211090087890625, 36.43650817871094, 60.7532958984375, -4.5257568359375, 7.0126953125, 22.20098876953125, -5.535736083984375, 50.168426513671875, 25.214218139648438, 64.99533081054688, 25.481414794921875, 53.39863586425781, 32.975555419921875, 40.641204833984375, 56.104248046875, 66.23527526855469, 10.065582275390625, -3.5305557250976562, -11.790817260742188, -6.94633674621582, 58.571632385253906, -31.677520751953125, 50.44264221191406, 35.42491912841797, 19.98992919921875, -4.93963623046875, 2.968017578125, -11.77960205078125, 6.970855712890625, 0.9593067169189453, -9.337005615234375, -4.672119140625, -32.77278137207031, 73.77725219726562, -4.777107238769531, 63.43011474609375, -9.96612548828125, 0.0, 23.235946655273438, 28.939453125, -35.6258544921875, -6.425079345703125, 24.00933837890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000173.npy"} {"epoch": 0.362303664921466, "step": 174, "batch_size": 128, "mean": 24.349727630615234, "std": 32.06266403198242, "min": -55.243255615234375, "p10": -12.365817260742187, "median": 19.098899841308594, "p90": 69.92042541503906, "max": 113.11111450195312, "pos_frac": 0.765625, "sample": [51.372047424316406, 1.8498916625976562, 5.00959587097168, 1.2324676513671875, 84.4766845703125, 60.93572998046875, 71.41653442382812, -36.516387939453125, 10.337753295898438, 0.1121673583984375, 63.479736328125, 51.410621643066406, 2.30816650390625, 48.824432373046875, 11.7891845703125, 27.11444091796875, 64.14718627929688, 97.410888671875, 11.958724975585938, 12.319313049316406, -3.01104736328125, 69.27923583984375, 1.61376953125, 76.02841186523438, 75.7706298828125, -5.2950439453125, 43.21159362792969, 8.977157592773438, 76.47613525390625, 48.491943359375, -3.1771163940429688, 51.140960693359375, 113.11111450195312, -17.555755615234375, -17.571250915527344, -11.686904907226562, 56.412567138671875, 71.48458862304688, 84.69912719726562, 2.635040283203125, 15.840227127075195, 34.61244201660156, 35.235191345214844, 30.84729766845703, -2.5028114318847656, 74.66690063476562, 46.850341796875, 36.36785888671875, -6.7701416015625, 99.97384643554688, -10.227783203125, -28.210006713867188, 55.595298767089844, 38.44921875, -10.737075805664062, 3.19683837890625, 7.324775695800781, -1.4921112060546875, 24.13616943359375, 22.355712890625, 21.1104736328125, 6.09881591796875, 73.37100219726562, 64.31533813476562, 18.680633544921875, -12.0155029296875, -3.583038330078125, 39.5516357421875, 10.763282775878906, 33.52210998535156, 23.087059020996094, 47.61474609375, 57.80574035644531, 56.72869873046875, 44.33766174316406, 29.730789184570312, 6.0838165283203125, 13.072723388671875, 58.474700927734375, -12.457183837890625, -35.633880615234375, 38.44709014892578, 0.8385238647460938, 38.5826416015625, 19.517166137695312, 2.3785629272460938, 62.87785339355469, 8.924863815307617, -14.114044189453125, 10.126785278320312, 8.0264892578125, 16.81402587890625, 41.526336669921875, 38.78089904785156, 42.45556640625, 44.350616455078125, -0.054412841796875, 82.94464111328125, -2.08489990234375, 18.641677856445312, 52.462158203125, -17.08251953125, 43.274658203125, 19.999725341796875, -12.32666015625, 13.367996215820312, -12.013420104980469, 0.3797187805175781, -3.04833984375, -29.887939453125, 43.05708312988281, -1.29254150390625, 3.03887939453125, 35.45535659790039, 32.292755126953125, -16.14251708984375, 66.14523315429688, -14.5396728515625, 8.455810546875, -55.243255615234375, 13.905593872070312, 15.502212524414062, 35.776763916015625, 20.668121337890625, 38.6756591796875, 1.3035507202148438, 20.7117919921875, -13.30767822265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000174.npy"} {"epoch": 0.3643979057591623, "step": 175, "batch_size": 128, "mean": 19.940677642822266, "std": 34.09808349609375, "min": -93.3785400390625, "p10": -17.240288543701165, "median": 17.492958068847656, "p90": 60.434933471679685, "max": 116.109130859375, "pos_frac": 0.765625, "sample": [52.584686279296875, 25.419189453125, 38.778961181640625, -2.354280471801758, 1.0470046997070312, -43.44468688964844, 24.225128173828125, 2.5674285888671875, 87.16473388671875, -42.90521240234375, 14.636154174804688, 10.942359924316406, 29.186279296875, 27.246376037597656, 47.31265640258789, 116.109130859375, 4.598066329956055, 45.36956787109375, 40.35203552246094, 70.12867736816406, -32.5389404296875, 55.705078125, 55.52569580078125, 48.292724609375, 42.3212890625, 1.507843017578125, -2.5114288330078125, -30.3358154296875, 22.667770385742188, -30.938873291015625, -4.2412109375, -0.005828857421875, 21.6009521484375, 50.21881103515625, -4.1275177001953125, 50.73382568359375, 10.269927978515625, -3.4615936279296875, 70.42459106445312, 41.46923828125, 4.87359619140625, 10.80853271484375, 13.547515869140625, 18.6405029296875, 9.6976318359375, 67.62905883789062, 50.862213134765625, 51.72479248046875, -13.19708251953125, 80.39926147460938, 35.447113037109375, 10.49951171875, 4.22857666015625, 10.881500244140625, -12.063232421875, -20.462615966796875, 17.621795654296875, -0.49974822998046875, -1.3546676635742188, 69.56845092773438, 20.222806930541992, 57.429107666015625, 7.883026123046875, 38.664344787597656, 50.131195068359375, 7.59417724609375, 2.8244247436523438, 4.375019073486328, 2.427642822265625, 6.101043701171875, 57.316619873046875, -2.5523300170898438, 10.094802856445312, 60.1077880859375, 47.80101013183594, -41.41139221191406, 1.169219970703125, 61.198272705078125, 0.0, -93.3785400390625, 17.364120483398438, 42.22100830078125, 75.56649780273438, 6.6759033203125, 24.8746337890625, -0.3321533203125, 24.1507568359375, -15.859291076660156, -32.9605712890625, 79.33895874023438, 40.404296875, 4.629917144775391, 50.22796630859375, 50.59576416015625, 47.51019287109375, 33.338226318359375, -10.69647216796875, 10.426610946655273, -23.467239379882812, 8.10797119140625, 48.7606201171875, 50.770286560058594, 9.586349487304688, 24.727447509765625, 1.6556396484375, -73.50811767578125, 19.421722412109375, 18.864578247070312, -8.434295654296875, 24.895111083984375, 31.853328704833984, 22.912033081054688, 27.656967163085938, 33.24980163574219, 74.41970825195312, 4.45025634765625, 28.378555297851562, 80.882080078125, 54.54283142089844, 16.939788818359375, 12.047607421875, 5.371368408203125, 21.29693603515625, -68.44064331054688, -48.865203857421875, 0.0, 3.5621795654296875, 84.90081787109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000175.npy"} {"epoch": 0.36649214659685864, "step": 176, "batch_size": 128, "mean": 14.506064414978027, "std": 35.88334274291992, "min": -107.20114135742188, "p10": -23.34557189941406, "median": 9.753379821777344, "p90": 58.27714462280273, "max": 100.82150268554688, "pos_frac": 0.71875, "sample": [-18.813720703125, 12.019691467285156, 3.2952880859375, -17.251068115234375, 21.21685791015625, 41.00786590576172, -54.41218566894531, -52.3182373046875, 14.748336791992188, 15.917510986328125, -77.5966796875, 3.37066650390625, 31.994327545166016, 52.921905517578125, 9.60992431640625, 34.96171569824219, 43.215728759765625, 4.2129058837890625, 2.796600341796875, -8.956298828125, -39.764617919921875, 9.15399169921875, 6.8457183837890625, -70.52032470703125, 12.03619384765625, 73.72760009765625, 30.049339294433594, 95.07896423339844, 33.601409912109375, -0.5765838623046875, 88.2467041015625, -30.6102294921875, -3.358673095703125, 17.944671630859375, 13.891510009765625, 2.491241455078125, -0.5595703125, 24.15008544921875, 14.79815673828125, 8.728923797607422, -47.10523986816406, 11.850446701049805, 46.039825439453125, -3.847747802734375, 7.1636810302734375, 58.502906799316406, 5.8801727294921875, 3.3291854858398438, 33.88093566894531, 75.5767822265625, -38.30394744873047, 50.927032470703125, 45.31208801269531, -19.76214599609375, 13.844573974609375, 50.190032958984375, -0.583953857421875, 43.72210693359375, 41.18524169921875, 3.975025177001953, 4.7686767578125, 1.593353271484375, -4.796356201171875, -16.33135986328125, 44.953819274902344, 58.180389404296875, 33.792755126953125, -12.88885498046875, 66.14263916015625, 3.65277099609375, -13.5030517578125, 70.76765441894531, 18.995010375976562, 3.2415771484375, 100.82150268554688, 48.529510498046875, 1.7181396484375, -15.092681884765625, 14.615325927734375, 9.02117919921875, 35.33863067626953, 34.99415588378906, 50.027408599853516, 42.33104705810547, 41.076507568359375, 54.405540466308594, -6.45050048828125, 9.896835327148438, 2.138946533203125, 80.0166015625, 35.90855407714844, 0.4723663330078125, -30.08154296875, 29.317047119140625, -13.794189453125, 4.544109344482422, 48.38037109375, -68.110595703125, -22.96270751953125, 28.570709228515625, -12.8394775390625, 52.22041320800781, 3.8703155517578125, -15.241790771484375, 7.1596527099609375, -65.54718017578125, 22.766693115234375, 13.578399658203125, -16.361907958984375, 3.941295623779297, 63.044944763183594, -6.849601745605469, 78.74722290039062, 47.802215576171875, 81.44557189941406, 39.63719177246094, 3.048583984375, 19.579620361328125, -24.238922119140625, -107.20114135742188, 60.41424560546875, -1.251495361328125, 0.7132167816162109, 30.70135498046875, 33.36126708984375, 8.84521484375, 35.77252960205078, -3.6161346435546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000176.npy"} {"epoch": 0.36858638743455496, "step": 177, "batch_size": 128, "mean": 14.593791961669922, "std": 38.68341064453125, "min": -104.52943420410156, "p10": -30.933144378662103, "median": 11.13990592956543, "p90": 64.33578796386719, "max": 100.38372802734375, "pos_frac": 0.6796875, "sample": [49.10374450683594, -1.1286582946777344, 1.1128311157226562, 78.65093994140625, -10.486068725585938, 64.82791137695312, 11.79446792602539, -25.4498291015625, 28.49542236328125, 5.786407470703125, 60.441802978515625, -43.532196044921875, -82.56143188476562, -11.622650146484375, 3.2577056884765625, -16.147491455078125, 42.50813674926758, 4.599578857421875, 47.815643310546875, 5.86004638671875, 41.5892333984375, 34.40284729003906, 46.788726806640625, 64.1248779296875, -22.929168701171875, 33.4755859375, -38.368499755859375, 80.77523803710938, 50.585601806640625, 87.0108642578125, 9.429351806640625, 30.08807373046875, -58.432159423828125, 86.04251098632812, -7.0445098876953125, 29.448150634765625, 65.78897094726562, 2.895376205444336, -9.10792350769043, 6.537872314453125, 38.929107666015625, -22.405029296875, -3.6174182891845703, 2.9792633056640625, -7.27203369140625, -29.551368713378906, 4.708766937255859, 13.823776245117188, 47.872962951660156, 44.154693603515625, -10.6676025390625, 0.000274658203125, 86.33209228515625, 26.025344848632812, -5.547943115234375, 59.8690185546875, 21.500701904296875, -5.597869873046875, 74.88148498535156, 3.5848388671875, -45.446075439453125, 75.93035888671875, 1.2115478515625, -20.80743408203125, 62.366729736328125, -2.9409942626953125, -43.26756286621094, -8.32065200805664, 5.952671051025391, 63.80902099609375, -15.4517822265625, 92.01873779296875, -6.6712799072265625, 7.164167404174805, -4.885612487792969, 49.48942565917969, 47.199981689453125, -83.85664367675781, 21.910247802734375, 22.941070556640625, 21.9564208984375, -47.205322265625, -51.49370574951172, 19.5953369140625, -67.2449951171875, 12.0760498046875, 34.09450149536133, 1.8865966796875, -8.84332275390625, 9.911231994628906, 24.74041748046875, 57.20794677734375, 40.01803207397461, 35.51873779296875, -11.89801025390625, -18.07830810546875, 13.650848388671875, 14.255126953125, 55.238922119140625, 10.485343933105469, -23.256378173828125, -3.99566650390625, 12.944305419921875, 42.632080078125, 8.026483535766602, -77.02261352539062, 7.5164337158203125, 0.8552093505859375, -34.15728759765625, -104.52943420410156, 26.992469787597656, 40.090118408203125, 67.70578002929688, 22.8741455078125, 14.351394653320312, 66.6959228515625, 36.610595703125, 38.989349365234375, 100.38372802734375, 12.179534912109375, 54.61328125, 1.5415420532226562, -2.493865966796875, -3.6428985595703125, 18.980484008789062, 4.45220947265625, 30.62603759765625, 57.394081115722656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000177.npy"} {"epoch": 0.3706806282722513, "step": 178, "batch_size": 128, "mean": 23.23711395263672, "std": 36.54656982421875, "min": -69.3055419921875, "p10": -14.450084686279295, "median": 19.945480346679688, "p90": 74.94254608154297, "max": 138.09979248046875, "pos_frac": 0.734375, "sample": [57.561004638671875, 15.621109008789062, 18.461219787597656, -5.8370361328125, -6.8577117919921875, 7.190311431884766, 41.027923583984375, 46.603599548339844, 29.927688598632812, 37.76036071777344, 138.09979248046875, 53.808013916015625, 30.3935546875, 19.743804931640625, -9.294921875, 81.79327392578125, 0.0, 57.871307373046875, 56.057586669921875, 29.286972045898438, 12.980712890625, 86.03683471679688, 7.920614242553711, 73.65045166015625, 52.787567138671875, -15.252799987792969, -7.890350341796875, 94.88552856445312, -19.369186401367188, 32.693084716796875, 4.905948638916016, -5.092430114746094, -1.30303955078125, 48.17389678955078, -63.43595886230469, 11.0994873046875, 27.853836059570312, 67.22242736816406, -28.9012451171875, 52.998779296875, 74.6187744140625, 36.292572021484375, 27.889495849609375, 38.0474853515625, 73.18621826171875, 57.682403564453125, 10.340194702148438, -46.78849792480469, -4.0733642578125, 87.69564819335938, 58.3555908203125, 86.55978393554688, -47.8343505859375, -69.3055419921875, -5.92718505859375, 11.806442260742188, 13.956787109375, 38.02379608154297, -33.57746887207031, 24.460205078125, 0.7267341613769531, 97.45254516601562, 8.977447509765625, 38.90802001953125, 62.249267578125, 27.866397857666016, 24.193756103515625, 29.38507080078125, 37.59405517578125, 8.309928894042969, 87.77142333984375, 21.37201690673828, -12.43984603881836, -2.564788818359375, 12.535795211791992, 79.26193237304688, -45.18304443359375, 13.993820190429688, -6.6894378662109375, 2.47198486328125, 43.564727783203125, 24.661346435546875, 38.15858459472656, -12.35467529296875, -50.7523193359375, -3.9856796264648438, 81.2198486328125, -4.1864013671875, 20.683944702148438, 30.5218505859375, 13.642890930175781, 3.921609878540039, 0.89794921875, 4.367218017578125, 12.647308349609375, 75.69801330566406, 8.105194091796875, 8.834953308105469, 1.0352783203125, 64.47799682617188, 9.390419006347656, -12.248296737670898, -19.42548370361328, 91.324951171875, 23.472152709960938, -13.625534057617188, 15.163398742675781, 63.183746337890625, 63.960205078125, -14.106063842773438, -3.979856491088867, 95.97857666015625, -29.80511474609375, 5.72760009765625, 40.07061767578125, 20.14715576171875, -23.91720962524414, 15.002883911132812, 42.10041809082031, 4.82470703125, 28.035064697265625, 27.905517578125, 25.7982177734375, -3.97039794921875, -5.981407165527344, 28.033538818359375, 51.85943603515625, 39.48797607421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000178.npy"} {"epoch": 0.37277486910994767, "step": 179, "batch_size": 128, "mean": 16.484657287597656, "std": 35.39377212524414, "min": -63.1531982421875, "p10": -34.28028945922851, "median": 12.22491455078125, "p90": 67.54102249145508, "max": 94.2215576171875, "pos_frac": 0.6953125, "sample": [38.132598876953125, 0.5768394470214844, 62.5986328125, -9.5205078125, 11.779922485351562, 63.13856506347656, 7.984092712402344, 10.509231567382812, -52.206390380859375, 2.0302276611328125, 12.51483154296875, -0.62689208984375, 22.9302978515625, 4.91650390625, 68.873779296875, -37.785240173339844, 43.96588134765625, 25.840087890625, 6.76007080078125, -4.878496170043945, 22.37738037109375, 42.331298828125, 4.3497161865234375, 28.708602905273438, 19.31829833984375, -38.71086120605469, -40.77186584472656, -5.473920822143555, -5.479362487792969, -51.357879638671875, 47.15807342529297, 6.47119140625, -45.058319091796875, 5.932830810546875, -8.810333251953125, 28.749969482421875, 45.803314208984375, 63.554595947265625, 21.52532958984375, -6.766212463378906, -32.778167724609375, -49.219085693359375, -4.813720703125, 23.684356689453125, -20.824188232421875, 30.921951293945312, 5.908660888671875, -9.722055435180664, 63.114776611328125, -54.306121826171875, -26.969894409179688, -27.0750732421875, 17.013816833496094, 26.323684692382812, 5.7846832275390625, 74.21871948242188, 67.52144622802734, 59.45794677734375, 20.673828125, -3.9027099609375, -63.1531982421875, -10.156219482421875, 89.53311157226562, 13.01885986328125, -6.648502349853516, 33.008575439453125, 67.58670043945312, -10.732742309570312, -0.3927001953125, 11.93499755859375, 7.670436859130859, 41.61029052734375, 24.27716064453125, 32.178253173828125, 1.6798858642578125, 6.196542739868164, 75.85209655761719, 11.347076416015625, 6.718040466308594, 43.404075622558594, 6.163795471191406, 5.1681060791015625, 90.1590576171875, -10.336280822753906, 13.456878662109375, -39.50994873046875, 78.63269805908203, 71.405517578125, 1.2777290344238281, -40.30714416503906, 47.118186950683594, 3.94317626953125, 76.9755859375, 31.9364013671875, 31.5032958984375, 26.517669677734375, 36.18647766113281, 40.96504211425781, 16.5701904296875, 39.931976318359375, 81.58694458007812, 43.079132080078125, 65.12973022460938, 13.796501159667969, -39.71002197265625, 27.298248291015625, -49.818939208984375, 6.5159759521484375, 23.292251586914062, 61.84882354736328, 12.570648193359375, 39.023712158203125, -28.210357666015625, 94.2215576171875, -15.87066650390625, -15.395866394042969, 42.60781478881836, 48.10211181640625, 0.12890625, -10.4578857421875, 76.11407470703125, 50.986785888671875, -0.8003082275390625, 11.720382690429688, -8.561309814453125, -18.720367431640625, 68.79228210449219, 51.67615509033203], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000179.npy"} {"epoch": 0.374869109947644, "step": 180, "batch_size": 128, "mean": 22.276973724365234, "std": 34.781463623046875, "min": -62.693817138671875, "p10": -19.395359420776366, "median": 16.402755737304688, "p90": 71.2047348022461, "max": 95.22320556640625, "pos_frac": 0.75, "sample": [-0.16901016235351562, -4.330034255981445, -36.04942321777344, 5.1771240234375, 3.919525146484375, 42.281494140625, 66.06170654296875, 6.52874755859375, -5.097900390625, 70.63639831542969, -2.64453125, 53.43194580078125, 93.55679321289062, -58.42720031738281, 7.453216552734375, 64.524169921875, -9.980010986328125, 11.9315185546875, 17.177379608154297, -62.693817138671875, 45.91748046875, 89.33944702148438, 39.105438232421875, 34.99365234375, 22.428192138671875, 48.64293670654297, 18.376556396484375, 61.157684326171875, 93.59756469726562, -5.821798324584961, -60.92073059082031, 8.731704711914062, 72.53085327148438, 95.22320556640625, 73.9576416015625, 10.12274169921875, 14.614349365234375, 8.451583862304688, 10.52215576171875, 14.790695190429688, 65.08892822265625, -4.891693115234375, 47.46540832519531, 31.06843376159668, 85.67813110351562, 17.0130615234375, -27.909713745117188, -2.560028076171875, 4.34405517578125, 0.364776611328125, -24.364479064941406, -5.4455718994140625, 4.71954345703125, 26.0081787109375, 76.51837158203125, 1.49371337890625, 17.975418090820312, 49.46240997314453, 11.260879516601562, 20.767234802246094, 88.95318603515625, 43.263206481933594, -10.20452880859375, 91.29803466796875, 50.82494354248047, -21.014862060546875, 33.464515686035156, 33.417205810546875, -34.661865234375, 15.792449951171875, 38.7611083984375, 4.775810241699219, 13.279884338378906, 59.578643798828125, 68.73014831542969, 69.11282348632812, -0.166015625, 0.3325042724609375, 55.9815673828125, 47.655975341796875, 94.70175170898438, 18.261154174804688, -10.615020751953125, 53.54345703125, 51.97707748413086, 44.13706970214844, 25.925025939941406, -38.94031524658203, 3.8828125, 50.498077392578125, -1.9337005615234375, 17.13427734375, 2.9430313110351562, 29.96478271484375, 52.57621765136719, 12.0892333984375, 24.015411376953125, 28.28509521484375, 5.5506591796875, -17.648590087890625, 4.09326171875, 19.167572021484375, 64.96969604492188, 13.482093811035156, 5.8735198974609375, -19.32968521118164, 4.660923004150391, -20.07611083984375, 74.00375366210938, 4.79595947265625, -12.679962158203125, 42.435428619384766, -4.198291778564453, -21.53961181640625, 47.77751922607422, 31.16748046875, 42.14280700683594, -19.803817749023438, -12.23016357421875, 50.53570556640625, 1.3365039825439453, 19.904876708984375, 3.0009078979492188, 4.9880523681640625, -5.7099151611328125, 85.44216918945312, -19.548599243164062, 18.1617431640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000180.npy"} {"epoch": 0.3769633507853403, "step": 181, "batch_size": 128, "mean": 21.182498931884766, "std": 36.466617584228516, "min": -73.89422607421875, "p10": -25.795977783203124, "median": 19.258445739746094, "p90": 65.18542175292968, "max": 118.31954956054688, "pos_frac": 0.734375, "sample": [0.7764358520507812, 5.248420715332031, 1.8423995971679688, 14.915283203125, 0.0, 43.54265594482422, 45.85906982421875, 31.6673583984375, 8.800590515136719, 34.72277069091797, 62.2591552734375, -10.556095123291016, 48.630821228027344, -7.1081695556640625, 85.4405517578125, 21.058868408203125, 0.0, 66.77983093261719, 45.41484069824219, -18.273361206054688, 57.53801727294922, 8.833251953125, 55.07666015625, 9.356781005859375, -69.58137512207031, 88.73149108886719, -17.0203857421875, 25.124095916748047, 5.473243713378906, 28.338134765625, 61.82754898071289, 2.0483551025390625, 14.703248977661133, 59.295013427734375, 25.044677734375, 41.831077575683594, 64.22119140625, 20.315872192382812, 19.490966796875, 31.9483642578125, 19.025924682617188, 27.816925048828125, -25.55242919921875, -5.284391403198242, 11.479034423828125, 14.037879943847656, 10.431503295898438, 12.31793212890625, 90.97562408447266, -44.32191467285156, 29.5042724609375, -0.11164093017578125, 50.742637634277344, 60.12249755859375, -15.268280029296875, -63.201873779296875, -1.0148506164550781, 15.38916015625, -2.5753173828125, 21.482452392578125, 37.03155517578125, 5.471771240234375, -23.06854248046875, 74.29571533203125, 40.08323669433594, 36.36309051513672, 17.352447509765625, 61.790008544921875, 1.493896484375, 28.36090087890625, 20.459716796875, 42.93699645996094, 27.04515838623047, 59.50880432128906, 100.81005859375, 118.31954956054688, -46.3876953125, 12.319671630859375, -26.3642578125, 63.897674560546875, -30.746994018554688, 29.09124755859375, -5.669134140014648, -64.62139892578125, -8.287490844726562, 65.02590942382812, 14.33331298828125, -9.460166931152344, 15.053802490234375, 4.556488037109375, -29.086715698242188, 33.70556640625, 71.76092529296875, -18.326034545898438, 10.388175964355469, 22.278663635253906, 10.45684814453125, 34.68927764892578, 20.197494506835938, 82.5438232421875, 77.7696533203125, -35.22222900390625, 31.18780517578125, 11.482040405273438, -27.7362060546875, -11.151817321777344, 17.57049560546875, 84.59521484375, -3.98858642578125, -73.89422607421875, 57.426788330078125, 25.59375, 65.5576171875, -5.0936279296875, 30.340301513671875, 79.83474731445312, -45.850921630859375, 17.37542724609375, 16.735809326171875, -6.32989501953125, 14.776336669921875, 37.90519714355469, 59.366485595703125, -34.755767822265625, 63.73992919921875, 43.4354248046875, 44.618263244628906, 56.855621337890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000181.npy"} {"epoch": 0.37905759162303665, "step": 182, "batch_size": 128, "mean": 26.108627319335938, "std": 31.927921295166016, "min": -65.85614013671875, "p10": -8.518179321289063, "median": 27.256935119628906, "p90": 69.1306671142578, "max": 92.64442443847656, "pos_frac": 0.8046875, "sample": [42.481658935546875, 58.65626525878906, 3.7913818359375, -1.6277027130126953, 23.6070556640625, 8.873176574707031, 65.62211608886719, 29.563385009765625, 21.867691040039062, 4.9880218505859375, 61.79241943359375, 26.0125732421875, 25.3006591796875, 47.44427490234375, 34.34625244140625, 0.164459228515625, 39.738372802734375, 51.126434326171875, 73.15452575683594, 46.42884826660156, 46.555213928222656, 31.040775299072266, 46.00074005126953, 81.05914306640625, -8.74395751953125, 72.40478515625, -6.890022277832031, 44.465850830078125, 89.77615356445312, 31.8436279296875, 53.193023681640625, 28.1805419921875, 32.67327880859375, 32.5760498046875, 13.670646667480469, -4.448179244995117, -12.714096069335938, 11.08941650390625, 8.669357299804688, 9.214462280273438, -49.48309326171875, 22.737289428710938, -65.85614013671875, 35.052955627441406, 67.50469970703125, -8.421417236328125, 81.6756591796875, -63.18223571777344, 60.687408447265625, 84.07168579101562, 29.533599853515625, 59.81781005859375, 60.1685791015625, 71.66342163085938, 42.387298583984375, -6.054351806640625, 13.13306999206543, 6.974449157714844, 8.77130126953125, 39.150421142578125, 31.749908447265625, 66.12834167480469, 36.89111328125, 21.2369384765625, 32.352569580078125, 9.534385681152344, 39.39393615722656, 32.473846435546875, 92.64442443847656, 13.539897918701172, -20.83045196533203, 9.145244598388672, 42.299652099609375, 14.5086669921875, 1.02984619140625, 17.248931884765625, 48.20306396484375, 46.26011657714844, 6.7352142333984375, 7.633087158203125, 7.7284698486328125, -5.603607177734375, 2.65240478515625, 48.1064453125, -3.05126953125, -0.8577957153320312, 46.48359680175781, 4.7572479248046875, 15.417724609375, -2.0015869140625, 75.44473266601562, 52.621826171875, 78.04183959960938, 69.93173217773438, -39.92994689941406, -10.177753448486328, 76.42706298828125, 51.278343200683594, 22.71875, 38.318756103515625, -28.366226196289062, -2.09228515625, 9.79998779296875, 11.138694763183594, -4.904998779296875, 43.993167877197266, 41.70391845703125, 26.333328247070312, 68.787353515625, 8.67279052734375, -15.06451416015625, 54.05328369140625, 65.36041259765625, 2.169414520263672, 1.1127243041992188, 5.3294677734375, 8.78499984741211, 61.80943298339844, 31.71353530883789, -29.424400329589844, -19.482513427734375, 64.42643737792969, 32.30078125, 1.543548583984375, 74.58187866210938, -3.7660293579101562, 29.705780029296875, -28.082504272460938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000182.npy"} {"epoch": 0.381151832460733, "step": 183, "batch_size": 128, "mean": 21.538780212402344, "std": 35.44655227661133, "min": -63.30287170410156, "p10": -21.00976867675781, "median": 19.647354125976562, "p90": 67.21926422119141, "max": 102.93292236328125, "pos_frac": 0.734375, "sample": [-25.064117431640625, 37.996124267578125, -24.226333618164062, 55.647438049316406, -42.516510009765625, 70.19306945800781, -19.631240844726562, 11.058517456054688, 61.37568664550781, 35.87054443359375, -11.93548583984375, 18.636367797851562, 31.824310302734375, 0.72314453125, 19.52081298828125, -56.11676025390625, 2.397420883178711, 50.0169677734375, 16.668670654296875, 2.224550247192383, 25.767501831054688, 6.528343200683594, 50.89723205566406, -17.463409423828125, 11.51031494140625, -47.858428955078125, 64.49375915527344, 31.832672119140625, 11.127883911132812, 11.2303466796875, -8.0875244140625, -32.837646484375, 14.28179931640625, 15.13482666015625, 1.0953598022460938, 60.678741455078125, 10.364486694335938, 70.99673461914062, 28.44415283203125, -16.70501708984375, 81.55328369140625, -5.232883453369141, 39.05267333984375, -17.892578125, 66.2664794921875, 36.77497100830078, 56.67622375488281, -19.239974975585938, 80.2718505859375, 44.64664840698242, 6.65472412109375, 49.592926025390625, 56.06636047363281, -43.1771240234375, 0.0, 0.7840919494628906, -5.9822998046875, 12.410934448242188, -18.046123504638672, 45.17840576171875, 13.255729675292969, 77.12879943847656, 0.2185230255126953, 84.04840087890625, 62.938934326171875, -36.89154052734375, 19.944984436035156, 85.00225830078125, 52.88175964355469, -17.428298950195312, 41.17124938964844, 92.03912353515625, 102.93292236328125, 50.948265075683594, 1.3253326416015625, 13.921745300292969, -49.79066467285156, 25.538543701171875, -2.62933349609375, 39.08143615722656, 31.71221923828125, 24.54730224609375, 12.67596435546875, 25.605730056762695, -7.5155487060546875, 7.680450439453125, 26.530609130859375, 39.936431884765625, 12.15679931640625, 13.55609130859375, -5.327339172363281, 28.576202392578125, 3.7701950073242188, 3.8411178588867188, -6.62493896484375, 67.26272583007812, 56.57279968261719, 48.15802001953125, 34.08172607421875, -0.34979248046875, -63.30287170410156, 4.4095306396484375, -29.277664184570312, 35.94352722167969, -0.3381080627441406, 53.290802001953125, 0.9001197814941406, 67.20063781738281, -4.907173156738281, 84.48373413085938, 31.984893798828125, 84.1585693359375, 37.214599609375, 56.41706848144531, 33.162139892578125, -49.49690246582031, 19.773895263671875, 41.97222900390625, -38.204803466796875, 96.80203247070312, -6.71173095703125, 51.66979217529297, 39.212890625, 33.84521484375, -6.785003662109375, 28.754493713378906, 51.24967956542969, 38.60321044921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000183.npy"} {"epoch": 0.3832460732984293, "step": 184, "batch_size": 128, "mean": 31.213024139404297, "std": 39.675689697265625, "min": -70.55178833007812, "p10": -17.972491455078124, "median": 26.84258270263672, "p90": 80.55586242675781, "max": 112.45187377929688, "pos_frac": 0.796875, "sample": [48.0689697265625, 112.45187377929688, 77.08480834960938, 12.970535278320312, 20.214569091796875, -12.285888671875, 20.4068603515625, -7.164518356323242, 78.23782348632812, 63.414337158203125, 24.371490478515625, 11.146331787109375, -46.420166015625, 14.274765014648438, -6.4085235595703125, 6.940086364746094, 96.95401000976562, 63.516265869140625, 23.742637634277344, 51.574424743652344, 59.473236083984375, 82.2523193359375, -70.55178833007812, 77.03720092773438, -7.137664794921875, 81.39187622070312, 44.835662841796875, -8.633804321289062, -39.09864044189453, 91.4112548828125, 11.41412353515625, 85.74212646484375, 27.007980346679688, -17.668792724609375, 50.88037109375, 80.19757080078125, 5.32391357421875, 63.241180419921875, 54.657318115234375, 93.72340393066406, 25.358856201171875, 92.446533203125, -54.146087646484375, 47.0804557800293, 37.26666259765625, 103.78240966796875, -43.468231201171875, 64.38690185546875, -33.39654541015625, 43.865966796875, 19.26592254638672, -11.90924072265625, 34.540252685546875, 26.67718505859375, -18.681121826171875, 32.722320556640625, 56.2694091796875, 32.59819030761719, -32.43316650390625, 99.89157104492188, 7.829925537109375, 6.356317520141602, 37.32281494140625, -52.08772277832031, 51.075408935546875, 42.617462158203125, 64.71539306640625, 73.14576721191406, 23.862518310546875, 46.34759521484375, 11.8817138671875, 15.265167236328125, 67.4710922241211, 14.31329345703125, 45.21185302734375, 8.981948852539062, 26.463165283203125, 66.61614990234375, 36.22447204589844, 52.10693359375, 73.673095703125, 44.155670166015625, 0.016204833984375, 71.50833129882812, 9.336868286132812, -39.9788818359375, 7.5628662109375, 78.69050598144531, 5.3487701416015625, 23.913482666015625, 7.05889892578125, 68.93667602539062, 18.932876586914062, 15.296821594238281, 59.024017333984375, -10.958183288574219, -27.255861282348633, 21.057205200195312, 1.7784461975097656, 77.92816162109375, 52.52018737792969, 53.41400146484375, 29.8021240234375, 73.52854919433594, 0.8867263793945312, 2.2260074615478516, 64.26669311523438, 19.315231323242188, -10.790565490722656, 89.10240173339844, 26.462265014648438, -35.3614501953125, 66.357421875, 16.325790405273438, 66.61885833740234, -6.7809600830078125, 86.40139770507812, 4.7039031982421875, -16.138870239257812, 49.82476043701172, -16.822021484375, -27.150482177734375, 79.11270141601562, 16.90179443359375, 98.68038940429688, 78.48538208007812, -7.48870849609375, 48.408660888671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000184.npy"} {"epoch": 0.38534031413612563, "step": 185, "batch_size": 128, "mean": 22.347719192504883, "std": 38.56614303588867, "min": -67.43435668945312, "p10": -22.285028839111327, "median": 19.30148696899414, "p90": 75.55305938720703, "max": 106.93051147460938, "pos_frac": 0.6640625, "sample": [51.461997985839844, -36.261810302734375, 57.053436279296875, -16.821044921875, -4.691162109375, 78.34942626953125, 56.20886993408203, 75.54472351074219, 51.97198486328125, -7.5236663818359375, 39.132080078125, 16.205642700195312, 63.54522705078125, 51.44343948364258, 12.691299438476562, 95.64651489257812, 42.85357666015625, -13.69696044921875, 9.248146057128906, 67.77377319335938, 37.4979248046875, 6.74639892578125, 54.56300735473633, 52.02149963378906, 27.918701171875, 15.218505859375, 54.72784423828125, 7.873779296875, 1.7665958404541016, 82.02978515625, -67.43435668945312, 54.510887145996094, 1.4454917907714844, -6.419677734375, 25.732139587402344, 59.05059814453125, 25.03125, 39.275733947753906, 31.82342529296875, 87.95590209960938, -11.16912841796875, 21.09515380859375, -15.403305053710938, 42.6220703125, 2.742401123046875, 78.64216613769531, -10.5513916015625, 44.76063537597656, 22.361083984375, 62.13330078125, -42.45863342285156, -25.126144409179688, -46.195281982421875, -16.521392822265625, 8.150390625, 0.19288063049316406, 11.48773193359375, 0.0, 8.449714660644531, 55.85935974121094, -22.06035614013672, -21.989501953125, 11.593429565429688, -5.666896820068359, 33.451080322265625, 31.210006713867188, 32.187522888183594, 18.4737548828125, 46.6866455078125, -3.852231979370117, -5.18243408203125, 61.04351806640625, 52.2169189453125, 68.65921020507812, -10.907485961914062, -2.138641357421875, -2.360687255859375, 16.01416015625, 24.23328971862793, 64.36331939697266, 106.93051147460938, -18.301376342773438, 11.578155517578125, 0.0, -7.2988433837890625, 59.06792449951172, 75.572509765625, -21.369720458984375, 91.99700927734375, -32.119598388671875, 51.183265686035156, 48.88128662109375, -12.862602233886719, -3.2413368225097656, 43.60205078125, 89.27139282226562, -43.26109313964844, 51.958404541015625, 27.59485626220703, 10.05364990234375, -2.4068984985351562, 32.41499328613281, 46.84638977050781, -22.80926513671875, 91.09323120117188, -52.39239501953125, -16.31494140625, 14.818214416503906, 106.83673095703125, 106.8433837890625, 58.7698974609375, 13.564430236816406, 10.76654052734375, -5.993133544921875, 54.464080810546875, -58.85321044921875, 44.90338134765625, 87.74575805664062, 31.703811645507812, -25.3028564453125, 36.388214111328125, -13.795303344726562, -17.973007202148438, -15.075698852539062, -31.081323623657227, -45.428794860839844, 20.12921905517578, 22.892913818359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000185.npy"} {"epoch": 0.387434554973822, "step": 186, "batch_size": 128, "mean": 21.658172607421875, "std": 34.847076416015625, "min": -71.655029296875, "p10": -11.693555450439453, "median": 18.565704345703125, "p90": 68.56238555908203, "max": 101.09970092773438, "pos_frac": 0.703125, "sample": [23.639602661132812, 0.09523773193359375, 14.87799072265625, 31.32818603515625, 20.724273681640625, 81.62844848632812, 26.4613037109375, -1.76641845703125, 13.808586120605469, 12.427513122558594, 3.8803634643554688, 18.840423583984375, 7.7730865478515625, 55.97605895996094, -18.55694580078125, 35.04486083984375, 49.028350830078125, 2.4323253631591797, 69.02279663085938, 37.46897888183594, 11.082916259765625, 18.89688491821289, -6.738677978515625, 3.61474609375, 2.7840118408203125, 56.422637939453125, 27.1558837890625, 26.828521728515625, -3.1741943359375, -1.4462890625, 8.214702606201172, 37.146484375, 90.11602783203125, 1.4596099853515625, -7.153472900390625, -50.460304260253906, 29.527313232421875, 82.801025390625, 27.50469970703125, 28.700469970703125, 4.806083679199219, 65.26971435546875, 56.29234313964844, 49.83707046508789, 52.28306198120117, 57.290283203125, -44.33892822265625, 5.021730422973633, 49.91503143310547, 37.7154541015625, 15.511199951171875, 18.290985107421875, 54.3751220703125, -0.294586181640625, -12.3165283203125, -21.641616821289062, 42.02606201171875, 22.0289306640625, 0.95416259765625, -4.33612060546875, -6.655693054199219, -10.352447509765625, 78.58685302734375, -35.44952392578125, 68.36506652832031, -10.268310546875, 32.524810791015625, 4.52325439453125, -3.076385498046875, 73.87205505371094, -5.157493591308594, 16.461639404296875, 95.924072265625, 32.918426513671875, -9.014434814453125, 5.097198486328125, 57.365379333496094, -69.13717651367188, -0.4240550994873047, 78.4259033203125, 44.62696838378906, -14.8546142578125, -1.351318359375, 96.82037353515625, 56.1885986328125, -54.25736999511719, 33.33087158203125, -11.195270538330078, 9.001846313476562, -37.551025390625, 20.951278686523438, 20.005889892578125, 38.36803436279297, -3.2916030883789062, 91.187744140625, 86.28793334960938, 27.47412109375, 14.57830810546875, -71.655029296875, 20.538063049316406, 40.63254165649414, 66.22978210449219, 7.979209899902344, -6.48724365234375, 57.86309814453125, 61.00689697265625, 22.94512939453125, -1.335296630859375, -0.59356689453125, 45.420406341552734, 47.42979431152344, -11.426567077636719, 34.474761962890625, -3.0178680419921875, 101.09970092773438, 0.0, 96.28860473632812, 52.72403335571289, -17.336273193359375, 35.1561279296875, -20.067302703857422, -10.7261962890625, 1.437744140625, 2.7514419555664062, 35.75238037109375, -8.149009704589844, 53.119384765625, 13.23406982421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000186.npy"} {"epoch": 0.38952879581151834, "step": 187, "batch_size": 128, "mean": 22.10552215576172, "std": 41.63774108886719, "min": -69.84957885742188, "p10": -33.843801116943354, "median": 21.512725830078125, "p90": 75.09210205078125, "max": 105.5018310546875, "pos_frac": 0.6875, "sample": [56.44110107421875, 71.38427734375, 41.5067138671875, 39.903564453125, -54.7679443359375, 20.898971557617188, 86.25482177734375, 2.3394336700439453, -35.54022979736328, 105.5018310546875, 76.12240600585938, 1.262176513671875, -21.617645263671875, -6.1763916015625, 72.96730041503906, 74.65054321289062, 46.94538879394531, 29.101097106933594, 73.17153930664062, 40.0009765625, 29.10137939453125, -28.224586486816406, 45.127227783203125, -4.3086090087890625, -23.245437622070312, -42.2525634765625, 61.03935241699219, -33.11676025390625, -36.60601806640625, -54.79681396484375, -7.1290283203125, 74.51333618164062, 25.3487548828125, 22.126480102539062, -21.651718139648438, 45.045684814453125, -14.982513427734375, -7.950653076171875, 49.42730712890625, 38.95233154296875, 25.09539794921875, -7.379302978515625, 94.68927001953125, -1.51812744140625, -11.619384765625, 48.078704833984375, 84.77761840820312, 4.64166259765625, -18.939666748046875, 41.999267578125, 0.762939453125, 42.689979553222656, 25.616722106933594, 59.712615966796875, -69.84957885742188, -20.8218994140625, 53.8707275390625, -22.9468994140625, 54.96950912475586, 78.82740783691406, 52.269378662109375, -7.687599182128906, 50.23956298828125, -54.416534423828125, 104.68536376953125, 68.47666931152344, -56.75978088378906, -65.15118408203125, 7.213470458984375, 42.907257080078125, 84.6978759765625, -50.60495376586914, -4.980827331542969, -20.1400146484375, -9.381927490234375, -37.774627685546875, 65.2354736328125, 11.188545227050781, 74.41595458984375, 12.738555908203125, 0.25848388671875, 19.483413696289062, -44.365760803222656, 5.6057891845703125, 9.437728881835938, -1.477447509765625, 19.19183349609375, 6.670562744140625, -55.9542236328125, 42.692352294921875, 46.58648681640625, 71.24798583984375, 67.89320373535156, 45.575653076171875, 43.112327575683594, 87.48320007324219, 0.2889556884765625, 59.603851318359375, 58.56591796875, 44.57317352294922, -21.123748779296875, 43.329620361328125, 99.90283203125, 12.1259765625, 0.1056060791015625, 67.388671875, 2.053070068359375, 22.732192993164062, -23.689666748046875, 89.56134033203125, 49.2689208984375, 28.82244873046875, 10.675003051757812, 18.499778747558594, 9.096979141235352, 68.58724975585938, 85.15086364746094, 0.9881591796875, 4.009767532348633, 7.04803466796875, -10.78485107421875, 49.156036376953125, 35.259559631347656, -6.8749847412109375, 0.0, 79.27178955078125, -12.303131103515625, 28.181381225585938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000187.npy"} {"epoch": 0.39162303664921466, "step": 188, "batch_size": 128, "mean": 23.630451202392578, "std": 40.04135513305664, "min": -87.5208740234375, "p10": -22.800723266601562, "median": 22.197429656982422, "p90": 74.70627288818359, "max": 100.891357421875, "pos_frac": 0.7265625, "sample": [5.067792892456055, 40.657012939453125, 47.571075439453125, 0.0, 21.52202606201172, 4.874237060546875, 5.2100830078125, 84.21255493164062, -6.272640228271484, 48.10729217529297, 58.899993896484375, 0.6363296508789062, 92.21047973632812, -9.75985336303711, 25.67919921875, -34.23291778564453, 58.288909912109375, 25.74932861328125, 13.425033569335938, 100.891357421875, 83.91485595703125, 57.857025146484375, -16.855987548828125, 1.1922874450683594, -12.742767333984375, -22.12518310546875, -26.1119384765625, 11.278121948242188, -24.376983642578125, 58.348838806152344, 71.420166015625, -14.726455688476562, 57.563629150390625, 45.60444641113281, 23.007553100585938, 45.53485107421875, 44.8037109375, 71.94509887695312, 64.83575439453125, 17.676910400390625, 14.58367919921875, 64.37167358398438, 7.739585876464844, 65.91909790039062, 92.21148681640625, -72.73382568359375, 19.371307373046875, 53.534423828125, 47.95123291015625, 40.98423767089844, 11.306976318359375, 60.104827880859375, -21.02752685546875, -4.438140869140625, 32.755859375, 84.49688720703125, 97.91839599609375, 8.71976089477539, 57.033294677734375, 88.6175537109375, 41.87261962890625, 20.83071517944336, 48.19533920288086, 71.24696350097656, 4.794231414794922, 1.094757080078125, 11.452316284179688, 74.32325744628906, 30.676864624023438, 49.71217346191406, 0.570098876953125, 58.3779296875, 100.6397705078125, -1.4818763732910156, -8.076019287109375, 4.6831207275390625, 64.79490661621094, 81.81222534179688, -10.332168579101562, 10.735572814941406, -56.08013916015625, 55.745391845703125, 27.354049682617188, 70.35125732421875, 0.7884674072265625, -6.97303581237793, 37.694580078125, 62.0670166015625, 27.39849853515625, -1.478759765625, -87.5208740234375, 28.175018310546875, -51.108001708984375, -31.09564208984375, 34.69297790527344, -11.8756103515625, -18.072242736816406, -32.7265625, 53.565765380859375, 33.81079864501953, 22.872833251953125, 1.6131057739257812, -48.1031494140625, -85.67337036132812, 48.35595703125, 30.518157958984375, 51.141700744628906, 68.31632995605469, 86.62139892578125, -8.40774154663086, -45.421630859375, 1.3055572509765625, 40.218772888183594, 11.629379272460938, 75.5999755859375, -14.02618408203125, 84.74542236328125, -5.463657379150391, 7.2930908203125, -15.97332763671875, -4.55126953125, 16.40435791015625, 0.0, 8.364410400390625, 36.696685791015625, 35.55422592163086, 4.191139221191406, -39.96209716796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000188.npy"} {"epoch": 0.393717277486911, "step": 189, "batch_size": 128, "mean": 27.567546844482422, "std": 33.81915283203125, "min": -51.468231201171875, "p10": -11.758106994628905, "median": 21.408737182617188, "p90": 75.70638122558593, "max": 103.77587890625, "pos_frac": 0.7734375, "sample": [41.677032470703125, 19.02508544921875, -48.58050537109375, 11.327896118164062, 9.86907958984375, 20.64068603515625, 50.8067626953125, 94.7496337890625, 49.21961212158203, 24.534942626953125, 35.05706024169922, 54.26776123046875, 89.7188720703125, 62.0306396484375, 61.264984130859375, 38.136474609375, 69.42893981933594, -1.6749267578125, -9.559555053710938, 18.95556640625, 76.17312622070312, 80.6976318359375, -14.94952392578125, 88.976318359375, 34.87139892578125, 9.222183227539062, 36.325103759765625, 34.901275634765625, 6.654151916503906, -5.1454315185546875, -25.542236328125, 4.769966125488281, 22.03516387939453, 45.44593048095703, 0.76153564453125, 26.92894744873047, 20.944068908691406, 3.0880584716796875, 14.288330078125, 21.930648803710938, -0.3616447448730469, 80.73626708984375, 64.8248291015625, 57.3438720703125, 57.45831298828125, 46.402740478515625, 21.3240966796875, -15.95916748046875, 70.40936279296875, -42.75267028808594, -31.3402099609375, -6.76708984375, -23.38861083984375, 58.64556884765625, 5.015472412109375, 20.98333740234375, 30.022186279296875, 4.5235443115234375, 13.856658935546875, 15.063201904296875, 69.41494750976562, 33.48393249511719, 47.15423583984375, -3.0827789306640625, 15.06390380859375, 13.145008087158203, 8.460027694702148, 6.514900207519531, 4.671150207519531, 9.732994079589844, 31.218032836914062, 5.305976867675781, -25.699378967285156, 60.36869812011719, -0.26352691650390625, 1.5666351318359375, 13.926849365234375, 5.398349761962891, -11.88671875, 20.20025634765625, 37.59547424316406, 29.77655029296875, 22.286956787109375, -2.28802490234375, 44.00865936279297, 53.7359619140625, 98.33554077148438, -21.018783569335938, -1.9197139739990234, -0.5934352874755859, 14.162078857421875, 79.61404418945312, 26.0948486328125, -51.468231201171875, 1.5349884033203125, 0.0, 103.77587890625, 20.604522705078125, 45.13020324707031, 83.47940063476562, 63.352294921875, 56.81498718261719, 74.12165069580078, 20.911666870117188, 37.39678955078125, 0.0, 27.488189697265625, 91.8778076171875, -13.049072265625, 75.50634765625, 52.870849609375, 13.784942626953125, 52.146453857421875, 10.589788436889648, 21.493377685546875, -17.0584716796875, 53.95240783691406, 73.97998046875, 59.57173156738281, 44.500762939453125, 29.7667236328125, 92.00869750976562, -11.702987670898438, 95.40948486328125, 51.6693115234375, -2.048858642578125, -3.5623931884765625, -9.999561309814453], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000189.npy"} {"epoch": 0.3958115183246073, "step": 190, "batch_size": 128, "mean": 26.735673904418945, "std": 40.11851501464844, "min": -78.96041870117188, "p10": -18.684558868408203, "median": 25.609352111816406, "p90": 76.72267761230468, "max": 122.01019287109375, "pos_frac": 0.765625, "sample": [-32.343772888183594, 13.22625732421875, 93.69183349609375, 47.26490783691406, -14.726387023925781, 33.062339782714844, -58.017059326171875, -3.715818405151367, -22.4429931640625, 5.84381103515625, 48.509613037109375, -59.878639221191406, -19.010520935058594, 75.7178955078125, 66.40264892578125, 1.0207443237304688, 7.877685546875, -15.769943237304688, 19.972312927246094, -76.85760498046875, -12.601846694946289, -19.857824325561523, -37.011085510253906, 66.70137023925781, 65.87639617919922, 7.9620208740234375, 36.34968566894531, -1.024566650390625, 45.17033386230469, 83.68161010742188, 113.0537109375, -16.651840209960938, 16.00444793701172, 23.04920196533203, 63.54949951171875, -4.082366943359375, 73.15557861328125, 13.788188934326172, 54.208168029785156, 42.70979309082031, 72.97225952148438, -76.94528198242188, 6.515869140625, 96.41731262207031, 62.55961608886719, 11.437591552734375, 58.60334777832031, 29.166011810302734, 79.766357421875, 96.62774658203125, 16.910263061523438, 12.017608642578125, 51.843223571777344, -1.1104736328125, 42.90220642089844, 44.7789306640625, 5.485084533691406, -17.128982543945312, 8.5374755859375, -8.815483093261719, -2.17169189453125, 67.49612426757812, 32.5848388671875, 27.377792358398438, 47.80662536621094, 29.88897705078125, 25.339126586914062, -18.54486083984375, 23.449462890625, 57.983184814453125, 54.53722381591797, 30.41208267211914, 52.59611511230469, 0.0, 70.37564086914062, 74.78489685058594, 40.157169342041016, 46.19861602783203, 90.9754638671875, 32.995235443115234, 3.21258544921875, 3.87091064453125, 32.88343811035156, 46.70402526855469, 74.24493408203125, 24.05999755859375, 13.655036926269531, 49.293304443359375, 56.26507568359375, 9.288421630859375, 17.209732055664062, 122.01019287109375, 47.576576232910156, 86.49630737304688, -1.0217971801757812, -78.96041870117188, 13.326065063476562, -17.651031494140625, 72.34838104248047, -26.762435913085938, 48.5286865234375, 19.31072998046875, 34.72747802734375, 79.06716918945312, 50.11370849609375, 40.746124267578125, 13.610061645507812, 5.09368896484375, 2.32635498046875, 47.152740478515625, 4.846691131591797, 105.38775634765625, 20.71407127380371, 51.326141357421875, 2.8172760009765625, 39.10638427734375, -10.176177978515625, 50.550357818603516, -37.52130126953125, 25.87957763671875, 24.391448974609375, 0.7758197784423828, 6.880807876586914, 50.839542388916016, -64.91600036621094, 81.273193359375, -0.953948974609375, 81.57826232910156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000190.npy"} {"epoch": 0.39790575916230364, "step": 191, "batch_size": 128, "mean": 23.361881256103516, "std": 41.68080520629883, "min": -104.04855346679688, "p10": -28.61663055419921, "median": 15.359302520751953, "p90": 78.73802261352539, "max": 109.60824584960938, "pos_frac": 0.7578125, "sample": [86.79641723632812, 57.096038818359375, -13.83111572265625, 0.55743408203125, 35.81523132324219, 92.7789306640625, 11.637603759765625, 15.895427703857422, 7.460807800292969, -34.30462646484375, 15.406700134277344, 50.798282623291016, -2.613698959350586, 7.1017913818359375, -0.5401611328125, -10.197540283203125, 52.006256103515625, -1.89434814453125, 1.5503158569335938, 21.44781494140625, 41.27783203125, 74.8671875, 9.633445739746094, 1.66619873046875, 0.4249420166015625, 16.593582153320312, -7.8959808349609375, -72.10676574707031, 64.09710693359375, 49.86274719238281, 8.401100158691406, 6.230398178100586, -15.511383056640625, 52.2685546875, 7.987548828125, -60.73284912109375, 46.29962158203125, -5.954399108886719, -104.04855346679688, -32.447906494140625, 36.98100280761719, 75.78433227539062, 100.08749389648438, 15.311904907226562, 11.668018341064453, 99.15498352050781, 12.9013671875, -12.974716186523438, 5.701740264892578, 9.190948486328125, 8.371841430664062, 66.56442260742188, 101.35052490234375, 80.39513397216797, 10.49652099609375, 6.0175933837890625, 48.7889518737793, 81.35551452636719, 18.93475341796875, 16.768951416015625, 7.245401382446289, -1.91986083984375, 42.11265563964844, -16.066734313964844, 71.37387084960938, 6.230051040649414, 100.87774658203125, 50.01282501220703, 21.880584716796875, 3.820768356323242, 2.857879638671875, 54.833709716796875, 9.088645935058594, -33.1273193359375, 9.511299133300781, -59.91609191894531, 24.785858154296875, 22.4127197265625, 61.23779296875, -25.622406005859375, 1.979156494140625, 72.91067504882812, 1.7145843505859375, 55.321990966796875, -46.89097595214844, 86.7591552734375, 78.02783203125, -38.8355712890625, 63.481048583984375, 85.59857177734375, 61.4522705078125, 67.86923217773438, 14.604736328125, 2.0859012603759766, -1.9647750854492188, -13.10650634765625, 58.733062744140625, 29.66058349609375, 99.46633911132812, 32.68426513671875, 27.0330810546875, 9.062484741210938, 67.32691955566406, 44.467803955078125, 109.60824584960938, 68.87876892089844, 3.8429107666015625, 0.6988677978515625, -56.93717956542969, 15.780731201171875, -1.88140869140625, -61.20123291015625, 53.5077018737793, 47.35888671875, 56.858428955078125, 54.0576171875, -4.4171142578125, 5.408153533935547, -26.974655151367188, 75.75518798828125, 41.40777587890625, 23.661911010742188, 26.647247314453125, -41.983642578125, 0.0, 70.3310546875, -42.02139663696289, 94.13150024414062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000191.npy"} {"epoch": 0.4, "step": 192, "batch_size": 128, "mean": 20.9859619140625, "std": 42.403343200683594, "min": -105.95297241210938, "p10": -29.609930419921874, "median": 16.556045532226562, "p90": 76.1156509399414, "max": 102.83470153808594, "pos_frac": 0.703125, "sample": [11.98126220703125, 38.632110595703125, 79.18983459472656, -38.27716064453125, 74.4989013671875, 19.232452392578125, -25.627410888671875, 16.003280639648438, -31.340923309326172, 72.4110107421875, 39.551849365234375, 19.201873779296875, 59.618133544921875, 65.8043212890625, 79.15731811523438, -6.86517333984375, -49.80205535888672, 52.1683349609375, -105.95297241210938, 102.83470153808594, -75.882568359375, 94.66085815429688, -10.66412353515625, -59.909305572509766, 41.307891845703125, 58.611663818359375, 8.831886291503906, 9.419929504394531, 13.511665344238281, 62.07566833496094, -88.04620361328125, 33.36046600341797, 53.87516784667969, 57.39666748046875, 1.558441162109375, 3.1359806060791016, -26.046173095703125, 61.130088806152344, 12.941085815429688, 16.430755615234375, 10.554306030273438, -13.425537109375, -12.166046142578125, 61.592498779296875, 77.72979736328125, -30.525390625, 31.8133544921875, -56.75, 0.0, 0.0, 1.9686355590820312, -2.483306884765625, -51.271202087402344, 33.69287109375, -27.180160522460938, 60.734779357910156, 4.5355987548828125, 48.767730712890625, -6.894401550292969, -16.308502197265625, 47.42353057861328, 40.348541259765625, -19.862762451171875, -23.581954956054688, 69.04144287109375, 72.01312255859375, 65.75393676757812, 75.19725036621094, 0.019985198974609375, 67.8275146484375, -29.21759033203125, 5.1986083984375, 29.110595703125, -7.422210693359375, 36.66204071044922, 50.805320739746094, 42.53741455078125, 83.39773559570312, 82.98452758789062, 11.738243103027344, 11.396575927734375, 75.42387390136719, -1.394683837890625, -34.599395751953125, -9.94000244140625, 18.24029541015625, 5.488475799560547, -15.39117431640625, 59.923797607421875, 42.97233581542969, 0.32781982421875, 18.0338134765625, -62.718597412109375, 100.87664794921875, 21.19569206237793, 28.073837280273438, 3.5811080932617188, -8.530155181884766, 85.58157348632812, 53.01678466796875, 11.230926513671875, 16.68133544921875, 8.224273681640625, -23.947189331054688, 6.681598663330078, 28.9541015625, 50.75398254394531, 6.3656158447265625, 97.067138671875, 45.810272216796875, 45.27307891845703, -9.1541748046875, 62.649192810058594, 14.851730346679688, 37.45416259765625, 13.259841918945312, 67.99655151367188, -4.509136199951172, -64.3941650390625, -1.676025390625, 4.593452453613281, 85.05377197265625, 47.68408203125, 19.520462036132812, 16.825241088867188, 83.75428771972656, -10.366615295410156, 83.526611328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000192.npy"} {"epoch": 0.40209424083769635, "step": 193, "batch_size": 128, "mean": 21.96724510192871, "std": 39.96500015258789, "min": -63.21864318847656, "p10": -29.92340393066406, "median": 19.061824798583984, "p90": 77.6846435546875, "max": 100.39181518554688, "pos_frac": 0.6796875, "sample": [72.80745697021484, -4.018333435058594, -4.709136962890625, -6.4108123779296875, -24.814682006835938, 53.42974853515625, 23.24597930908203, 67.24542236328125, 42.49580001831055, 30.61431884765625, 25.146759033203125, 2.9351329803466797, 10.028289794921875, 73.32820129394531, -1.35455322265625, 63.03181457519531, -19.9615478515625, 39.969451904296875, 0.6539306640625, 3.5896549224853516, 65.93951416015625, -19.3260498046875, 60.40618896484375, 7.359130859375, 27.929519653320312, 32.808013916015625, -1.2347259521484375, -52.8533935546875, 60.21549987792969, -2.947113037109375, -39.783668518066406, 2.9803924560546875, 34.5234375, 22.812225341796875, 54.982547760009766, -16.027740478515625, 80.36773681640625, -27.009033203125, 1.6879920959472656, 31.80548858642578, -56.19879150390625, -19.85748291015625, 5.8775482177734375, 38.7664794921875, 9.37677001953125, 37.938323974609375, 78.01080322265625, 33.607635498046875, 77.54486083984375, 84.14942932128906, -2.0863494873046875, 11.127471923828125, 34.37999725341797, 8.469097137451172, 60.70951843261719, 2.3195343017578125, 92.46475982666016, 25.009597778320312, -10.29327392578125, 34.6031494140625, -8.288970947265625, 55.762367248535156, 61.87152099609375, -27.0054931640625, 93.91763305664062, -22.728118896484375, 51.75110626220703, 42.264312744140625, 21.0726318359375, -7.68914794921875, -48.77655029296875, -43.73387145996094, 24.370452880859375, 96.71966552734375, -43.58843231201172, 55.7977294921875, 70.7767333984375, 83.60862731933594, 17.613861083984375, 88.11328125, 68.57904052734375, 24.46875, 67.04547119140625, -34.116302490234375, 90.82705688476562, 85.84259033203125, 9.914772033691406, 2.0931472778320312, -3.8603363037109375, 19.62200927734375, 48.66259765625, -4.5383148193359375, 100.39181518554688, 66.70782470703125, 15.430435180664062, -5.024169921875, 37.82711410522461, 5.121971130371094, 26.11932373046875, 64.58975219726562, 53.74945068359375, 79.08401489257812, 28.2153263092041, -38.57136535644531, 46.916656494140625, -32.903167724609375, 93.41513061523438, -25.884429931640625, -63.21864318847656, -3.6650390625, -3.3713531494140625, 4.733673095703125, -43.1444091796875, 43.30401611328125, 15.691650390625, 18.50164031982422, 71.87289428710938, -28.6463623046875, 13.4466552734375, -7.707122802734375, -3.960906982421875, -2.425445556640625, -61.72088623046875, 8.639923095703125, 58.892608642578125, -40.56690979003906, 4.199676513671875, 61.58811950683594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000193.npy"} {"epoch": 0.4041884816753927, "step": 194, "batch_size": 128, "mean": 16.499408721923828, "std": 38.42276382446289, "min": -80.13717651367188, "p10": -22.568685913085936, "median": 8.640338897705078, "p90": 69.79923706054686, "max": 108.61111450195312, "pos_frac": 0.640625, "sample": [108.61111450195312, 11.473129272460938, 30.451499938964844, -39.67450714111328, 20.130340576171875, 97.87429809570312, 38.30030822753906, -16.650604248046875, 2.457916259765625, -0.3503608703613281, 67.5218734741211, -22.311431884765625, -3.3899478912353516, 5.24066162109375, 10.783538818359375, 2.1012630462646484, -10.694496154785156, 48.30583953857422, 79.94306945800781, -3.6335906982421875, 18.645172119140625, -5.977550506591797, 31.72311019897461, 3.9666290283203125, 12.765411376953125, -23.1689453125, 3.9684906005859375, 26.496826171875, -7.2599334716796875, -38.82208251953125, -13.373687744140625, 23.350425720214844, -10.171546936035156, 12.754119873046875, -2.8443450927734375, 18.908004760742188, 108.10382080078125, 53.33277130126953, 12.951019287109375, 60.13180160522461, -14.14813232421875, 104.2698974609375, -7.713165283203125, -7.9990234375, -12.633163452148438, 60.266998291015625, -27.34624481201172, 54.59198760986328, 68.75765991210938, 7.3790130615234375, -10.506362915039062, 28.68505859375, 28.6717529296875, 34.07806396484375, 105.1417236328125, 31.412796020507812, 17.51678466796875, 27.37591552734375, 20.888473510742188, -11.040985107421875, 15.8487548828125, 10.3172607421875, -5.47393798828125, 58.979339599609375, -47.10809326171875, -15.4267578125, 5.6218414306640625, -80.13717651367188, 26.96661376953125, -16.7581787109375, 99.25042724609375, -9.409454345703125, -28.684234619140625, 82.55099487304688, 0.2464752197265625, 77.11578369140625, -0.09113693237304688, -22.19544219970703, 34.80889892578125, 61.30323791503906, -24.149261474609375, 8.337646484375, 63.14202880859375, 74.1012954711914, -6.129003524780273, -57.9373779296875, 51.173805236816406, -0.29496002197265625, 59.483428955078125, 72.22958374023438, 34.4334716796875, -57.340850830078125, 17.9833984375, -12.292774200439453, 5.838588714599609, 1.8890380859375, 6.824737548828125, -70.97526550292969, 103.3643798828125, 8.50567626953125, -3.876617431640625, 26.4466552734375, 2.3808631896972656, 8.775001525878906, -56.52607727050781, 16.647262573242188, -11.618194580078125, 36.3546142578125, -2.184722900390625, -14.837158203125, 31.61712646484375, 86.79263305664062, 55.278076171875, 8.247886657714844, 32.79302978515625, 26.024627685546875, 52.333885192871094, 2.5222320556640625, -39.118492126464844, 3.93890380859375, 10.807861328125, -5.158050537109375, 56.79058837890625, -16.618026733398438, 3.6125030517578125, -16.846206665039062, 44.146453857421875, 37.66632080078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000194.npy"} {"epoch": 0.406282722513089, "step": 195, "batch_size": 128, "mean": 16.304454803466797, "std": 37.84490966796875, "min": -80.10540771484375, "p10": -27.08980560302734, "median": 9.41360092163086, "p90": 66.73294982910156, "max": 112.8392333984375, "pos_frac": 0.6640625, "sample": [-8.821395874023438, -17.07842254638672, -1.107147216796875, 63.9654541015625, 68.45210266113281, 62.51953125, -29.382781982421875, 0.03271484375, 55.486114501953125, 8.58425521850586, 34.3416748046875, 9.854976654052734, 2.5999183654785156, -9.107959747314453, 2.6444244384765625, 19.597930908203125, -10.001800537109375, -9.882568359375, 16.5858154296875, 45.12371826171875, -12.268001556396484, 12.458816528320312, 65.03628540039062, -4.398490905761719, -16.0130615234375, -3.0529022216796875, 0.6803665161132812, 0.4080810546875, -29.78607177734375, 10.71014404296875, -11.1680908203125, 60.385498046875, -5.7085723876953125, -2.942413330078125, 7.06756591796875, 4.33172607421875, 81.84669494628906, 69.31173706054688, -5.5870361328125, 1.177093505859375, -26.107101440429688, -9.53997802734375, 12.821479797363281, 6.96917724609375, 43.3607177734375, 6.819122314453125, 79.2467041015625, -47.409698486328125, -18.4879150390625, 17.668731689453125, -37.690277099609375, 26.8807373046875, 32.68748474121094, 36.938018798828125, -76.4593505859375, 20.626220703125, -6.148468017578125, -67.38102722167969, 72.1973876953125, -5.382331848144531, 24.586349487304688, -80.10540771484375, 3.5793609619140625, 52.77967834472656, -14.378822326660156, -41.5455322265625, -41.51014709472656, -13.667327880859375, -2.528167724609375, 42.03289794921875, 88.11495971679688, 34.103759765625, 59.6710205078125, 0.620574951171875, 31.531707763671875, 83.88690185546875, 3.36956787109375, 1.720458984375, 26.77448272705078, 15.695465087890625, 45.01957702636719, 64.31893920898438, 30.1317138671875, 41.274810791015625, 13.6512451171875, -2.2297592163085938, 1.2935638427734375, -47.85504150390625, -1.5962438583374023, 11.39678955078125, -1.61279296875, 112.8392333984375, 61.711181640625, 1.2898712158203125, 62.35022735595703, 44.586639404296875, 99.41482543945312, 0.1385498046875, 32.27076721191406, 88.35598754882812, -47.433570861816406, 48.442291259765625, 10.25823974609375, 42.71833801269531, 5.952789306640625, 9.560173034667969, -10.5972900390625, 65.99617004394531, 47.00836181640625, 70.03390502929688, 99.37359619140625, 17.1279296875, -46.035888671875, 13.850494384765625, -25.4942626953125, 6.6099853515625, 55.9649658203125, -58.33770751953125, 9.26702880859375, 59.15510559082031, 10.45306396484375, -6.6650543212890625, -1.2009124755859375, 47.16981506347656, 71.09561920166016, 44.36460876464844, 38.70703125, -10.362152099609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000195.npy"} {"epoch": 0.4083769633507853, "step": 196, "batch_size": 128, "mean": 26.296886444091797, "std": 37.46731185913086, "min": -62.55645751953125, "p10": -18.71807403564453, "median": 20.419089317321777, "p90": 72.9837432861328, "max": 116.86868286132812, "pos_frac": 0.734375, "sample": [2.4271488189697266, 3.51336669921875, 34.94316864013672, 1.20196533203125, 29.124113082885742, 10.150146484375, -20.133209228515625, 20.239797592163086, 60.244110107421875, 0.67010498046875, 3.7513275146484375, 79.13575744628906, 35.960296630859375, 45.61729431152344, 95.4764404296875, -40.124237060546875, 2.065826416015625, 3.2550048828125, 16.523391723632812, 19.014434814453125, 48.95893859863281, 5.772068023681641, 48.05548095703125, 71.17803955078125, 0.9669895172119141, 38.97288513183594, 0.9707145690917969, -1.2694988250732422, -50.86326599121094, 61.23516845703125, 99.84902954101562, 83.14827728271484, 54.526336669921875, 47.04498291015625, -2.78302001953125, 51.430511474609375, 6.624305725097656, -36.07470703125, 49.468597412109375, 41.99237823486328, -22.473556518554688, 88.71429443359375, 36.59271240234375, -9.108154296875, -6.813495635986328, 15.377548217773438, 20.8575439453125, 91.8677978515625, 18.4593505859375, 31.12158203125, -25.465835571289062, -0.034473419189453125, -0.033538818359375, 5.369724273681641, -4.8190155029296875, 75.3824462890625, 70.83059692382812, 3.7496471405029297, 5.39373779296875, 22.28729248046875, 54.64533996582031, 10.210113525390625, 69.8231201171875, -2.238555908203125, 22.318939208984375, 78.49458312988281, -43.54220962524414, 20.79583740234375, -1.018890380859375, -15.44573974609375, 15.044914245605469, 19.3607177734375, 1.3271293640136719, 58.92457580566406, 40.65069580078125, 67.13764953613281, -0.598052978515625, 17.4744873046875, 58.160667419433594, 60.0762939453125, 35.273162841796875, -43.88270568847656, 54.38883972167969, -2.0768165588378906, -12.21380615234375, 58.247802734375, 37.61273193359375, 70.98817443847656, 30.8460693359375, 2.1155242919921875, 68.73272705078125, -62.55645751953125, 38.11956787109375, 99.58755493164062, 42.2506103515625, -8.061553955078125, -18.111587524414062, -12.474700927734375, 69.328857421875, 71.30238342285156, 52.82011413574219, -34.545257568359375, 7.961761474609375, 116.86868286132812, 83.59164428710938, 17.149253845214844, -29.343921661376953, 63.031951904296875, 73.87200927734375, 72.60305786132812, 62.02717590332031, 97.21112060546875, 69.74728393554688, 70.82583618164062, -6.01580810546875, -6.3262176513671875, -4.0206756591796875, 57.57057189941406, 51.255043029785156, 37.497314453125, -0.677398681640625, -22.57257652282715, -35.256927490234375, 53.41145324707031, 20.59838104248047, 8.95709228515625, -6.97430419921875, 4.2001495361328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000196.npy"} {"epoch": 0.41047120418848165, "step": 197, "batch_size": 128, "mean": 33.191864013671875, "std": 39.3669319152832, "min": -72.6707763671875, "p10": -11.942297363281241, "median": 35.24604797363281, "p90": 82.32809753417968, "max": 132.40350341796875, "pos_frac": 0.8046875, "sample": [-61.30729675292969, 6.151336669921875, 3.584716796875, 55.021087646484375, 52.134368896484375, -9.023056030273438, 61.552764892578125, 22.862319946289062, 6.8858642578125, 81.46063232421875, 60.1329345703125, 56.27900695800781, -9.830642700195312, 35.352691650390625, 57.21104431152344, -31.008176803588867, 117.59793090820312, 92.9642105102539, 39.1571044921875, 95.12045288085938, -0.025205612182617188, 66.4691390991211, 85.21681213378906, 19.315582275390625, 39.300048828125, 31.431304931640625, 59.968841552734375, 84.24957275390625, 76.14657592773438, 78.0133056640625, 67.44268798828125, 46.99761199951172, -47.82215881347656, 52.30494689941406, 25.89569091796875, 52.051673889160156, 64.67970275878906, -3.5295944213867188, 33.41753387451172, 88.5120849609375, 36.39496612548828, 51.1943359375, -72.6707763671875, 81.50460815429688, -45.35321044921875, 24.9910888671875, 92.6749267578125, -3.6273651123046875, 9.8505859375, -9.841033935546875, 49.19096374511719, 35.139404296875, 18.666900634765625, -16.845245361328125, 7.973869323730469, -18.945026397705078, 66.02827453613281, 11.433151245117188, 79.77554321289062, 61.74247741699219, 59.22026824951172, 132.40350341796875, 4.64617919921875, 85.8079833984375, -31.9527587890625, -1.8769264221191406, 58.291259765625, 71.26802825927734, 15.060476303100586, 13.930267333984375, 13.058731079101562, 36.28173828125, -1.783477783203125, 67.88031005859375, 1.045440673828125, 53.92585754394531, 18.4019775390625, 36.364105224609375, -23.54351806640625, 3.09063720703125, -4.2196044921875, 12.020919799804688, 37.427093505859375, 49.789825439453125, -24.111083984375, 64.70736694335938, 1.61956787109375, 22.8599853515625, 59.6903076171875, 8.672500610351562, 58.302215576171875, 0.6983642578125, 79.9765625, 39.3262939453125, 57.23883056640625, 13.1114501953125, 26.8931884765625, -38.96221923828125, 16.899757385253906, -49.439727783203125, 11.460540771484375, 42.91612243652344, -4.979480743408203, 65.31982421875, 73.32635498046875, 49.73828125, 31.7928466796875, 28.335250854492188, 80.14776611328125, 39.74383544921875, 103.72210693359375, 97.08953857421875, 8.7838134765625, 19.944747924804688, 31.622161865234375, 17.56591796875, 89.66830444335938, 1.9216594696044922, 62.65254592895508, 15.423843383789062, 0.0, 112.98724365234375, 24.721343994140625, 39.045166015625, 41.31348419189453, -38.438690185546875, 51.203826904296875, -4.105567932128906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000197.npy"} {"epoch": 0.41256544502617803, "step": 198, "batch_size": 128, "mean": 24.81391716003418, "std": 45.64817428588867, "min": -101.22137451171875, "p10": -32.00371475219727, "median": 19.26699161529541, "p90": 86.26334228515624, "max": 118.4759521484375, "pos_frac": 0.7421875, "sample": [69.49822998046875, 19.4510498046875, 8.5361328125, -8.5438232421875, 56.087554931640625, -0.443511962890625, 60.0245361328125, 83.273193359375, 32.85479736328125, 29.27032470703125, -13.788284301757812, -8.188217163085938, 41.22625732421875, 32.900421142578125, 10.826248168945312, -10.332229614257812, 94.78085327148438, 45.77801513671875, 37.05816650390625, 10.1734619140625, 85.60223388671875, -82.6900634765625, -47.662567138671875, 87.165771484375, 108.85662841796875, 110.47576904296875, -73.5648193359375, 11.700927734375, 18.609710693359375, 19.8314208984375, 85.8765869140625, 67.897216796875, 70.59040832519531, 63.349761962890625, 0.68890380859375, 26.36919403076172, -32.547454833984375, 118.4759521484375, -45.83644104003906, 109.28610229492188, -3.793487548828125, 1.4327392578125, 47.7232666015625, 63.60940170288086, -43.333221435546875, 0.2536029815673828, 101.58721923828125, -32.263389587402344, -31.892425537109375, 5.174297332763672, 40.97834777832031, 40.00587463378906, -3.367786407470703, 35.89131164550781, 18.519317626953125, 69.70498657226562, 91.8306884765625, 2.6812362670898438, 24.634521484375, 54.7879638671875, 14.022552490234375, 20.77105712890625, -20.814674377441406, 75.10984802246094, 50.5482177734375, 35.57220458984375, 2.461223602294922, 94.43014526367188, 18.08367919921875, 26.879188537597656, 16.11913299560547, -69.31307983398438, -17.9140625, 25.82281494140625, -101.22137451171875, -18.59637451171875, 18.03839111328125, 73.80160522460938, -17.363182067871094, 65.025390625, 9.593032836914062, 62.82008361816406, 60.81721496582031, -14.85699462890625, 67.51160430908203, 0.01392364501953125, 15.025848388671875, 84.68183898925781, -48.92689514160156, 12.9530029296875, 0.0, 66.14984130859375, 10.725105285644531, -15.932861328125, -48.34657287597656, 10.912734985351562, 19.08293342590332, 54.60820770263672, 93.2738037109375, 25.517303466796875, 93.931884765625, 1.4122581481933594, 71.90481567382812, 47.04231262207031, 27.18389892578125, 58.24540710449219, 37.906402587890625, 3.8360252380371094, 2.288543701171875, -97.37335205078125, 100.56329345703125, -24.7042236328125, 2.2807159423828125, -12.613189697265625, -6.846935272216797, -8.15692138671875, 98.16998291015625, 27.163238525390625, 68.71881103515625, 10.98931884765625, 23.78179931640625, -25.327728271484375, 7.766448974609375, -33.77471923828125, 18.214601516723633, 78.2049560546875, 70.18280029296875, 1.022216796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000198.npy"} {"epoch": 0.41465968586387436, "step": 199, "batch_size": 128, "mean": 25.2368106842041, "std": 41.17595291137695, "min": -74.49343872070312, "p10": -29.45914154052734, "median": 21.355648040771484, "p90": 82.15376663208008, "max": 115.78704833984375, "pos_frac": 0.7421875, "sample": [17.46820068359375, 36.22296142578125, 70.52706909179688, 65.53816223144531, 88.72587585449219, 62.69648742675781, 62.141754150390625, 1.5992164611816406, 81.62680053710938, 14.03564453125, 110.8729248046875, 39.038719177246094, 54.108612060546875, 55.802490234375, 101.00021362304688, -48.262786865234375, 54.75201416015625, 55.63777160644531, 7.561740875244141, -2.1465110778808594, -1.4778518676757812, 1.9704742431640625, -2.9853744506835938, 4.111042022705078, 3.46331787109375, 39.0233154296875, 44.619171142578125, 2.2575149536132812, 12.5804443359375, -2.9676513671875, 15.566558837890625, 44.992218017578125, 56.650672912597656, 9.62054443359375, 41.96723175048828, 19.982025146484375, -56.24243927001953, -8.930679321289062, 68.52772521972656, 11.247802734375, 10.980072021484375, 67.61328125, 12.736480712890625, 30.417095184326172, 10.856483459472656, 70.31903076171875, -1.59014892578125, -2.6958694458007812, -2.9987030029296875, -56.5230712890625, 61.70018768310547, 84.46585083007812, 10.10089111328125, 88.59707641601562, -10.426284790039062, 24.515426635742188, 12.8638916015625, 42.74102783203125, 85.15203857421875, -38.5894775390625, -70.27752685546875, 69.80470275878906, 73.23675537109375, 26.68389892578125, -34.57019805908203, -45.338748931884766, 50.309608459472656, 12.041748046875, -18.481765747070312, 29.428558349609375, 98.72860717773438, 0.93572998046875, 3.117734909057617, 101.55795288085938, 34.997894287109375, -1.9854736328125, 45.6641845703125, 22.246856689453125, -0.8568077087402344, 30.474685668945312, 88.02072143554688, 34.97080993652344, 75.18661499023438, -10.92559814453125, 70.42996215820312, 83.82186889648438, -61.03154754638672, 60.697998046875, 71.28952026367188, 34.376441955566406, -29.141403198242188, -15.7127685546875, 52.57765197753906, 87.38824462890625, -10.25921630859375, 22.35369873046875, 0.042758941650390625, 8.2835693359375, 22.687896728515625, 35.795806884765625, 21.429595947265625, -36.678890228271484, 16.6622314453125, 9.863250732421875, 58.046112060546875, -74.49343872070312, -57.482749938964844, 0.8102054595947266, 27.652957916259766, 28.748870849609375, 21.281700134277344, 11.43890380859375, 69.20144653320312, -3.1288681030273438, 63.93915557861328, 14.962615966796875, 83.38335418701172, 1.8326416015625, -4.983406066894531, 57.107444763183594, 6.718048095703125, -49.47137451171875, -3.16278076171875, -18.743736267089844, 115.78704833984375, 73.07852172851562, 44.983245849609375, -30.200531005859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000199.npy"} {"epoch": 0.4167539267015707, "step": 200, "batch_size": 128, "mean": 20.741317749023438, "std": 48.62843704223633, "min": -96.79008483886719, "p10": -44.66881561279297, "median": 14.977542877197266, "p90": 83.02436523437498, "max": 145.91192626953125, "pos_frac": 0.65625, "sample": [65.61407470703125, -60.935943603515625, 109.04074096679688, 35.850067138671875, -47.039886474609375, 12.669120788574219, -44.63652038574219, 58.411865234375, -5.862762451171875, 7.2524871826171875, -2.5677032470703125, 73.82786560058594, 23.588134765625, -3.008148193359375, -83.12606811523438, 75.09002685546875, 56.754302978515625, -13.640945434570312, 20.7196044921875, 108.8433837890625, -65.43099975585938, -39.006805419921875, 98.51751708984375, 39.65656280517578, 95.571044921875, 66.69610595703125, -22.93596649169922, -6.46502685546875, 62.915130615234375, -6.97442626953125, -34.489501953125, -83.2841796875, 14.258481979370117, -53.88726806640625, 13.852645874023438, 34.18438720703125, -50.39983367919922, 61.367584228515625, 46.906036376953125, -8.357696533203125, 0.0, 51.487464904785156, 56.715301513671875, 21.34943389892578, -10.579696655273438, -75.79721069335938, 78.48236083984375, 5.15118408203125, 113.367919921875, -5.5227813720703125, 17.755126953125, 13.030426025390625, 49.306396484375, 5.242958068847656, -33.739837646484375, -16.293701171875, 26.996597290039062, -70.52030944824219, 92.86700439453125, -1.3802375793457031, 52.169517517089844, 8.18292236328125, 0.31873321533203125, 55.121856689453125, 74.82498168945312, 47.00648880004883, 14.310585021972656, 88.72271728515625, -44.744171142578125, 57.143890380859375, 95.43321228027344, 48.30242919921875, -31.320037841796875, 52.865692138671875, 45.86848449707031, 22.07958984375, -3.739471435546875, -23.214385986328125, 8.182586669921875, 61.75469970703125, -69.07455444335938, 94.867431640625, 145.91192626953125, -0.5391998291015625, 23.112091064453125, 45.313262939453125, -6.155242919921875, 41.05167770385742, -96.79008483886719, 76.77178955078125, 11.316879272460938, 111.5966796875, -43.82313537597656, 80.58221435546875, 51.63629150390625, 80.27239990234375, 6.069633483886719, 94.78875732421875, -26.67645263671875, -41.57337951660156, 14.307586669921875, 40.267059326171875, 68.53765869140625, 36.87883758544922, -15.1767578125, 72.87489318847656, 15.826606750488281, -52.12977600097656, -12.84002685546875, 1.94989013671875, 9.558395385742188, 63.56361389160156, 54.79888916015625, 103.22601318359375, 28.413284301757812, 37.68121337890625, -6.083030700683594, -22.769020080566406, 57.65570068359375, 4.765048980712891, -2.515899658203125, -1.7759904861450195, 16.25194549560547, 4.787628173828125, 2.7776336669921875, 15.644500732421875, 40.579681396484375, 2.4478836059570312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000200.npy"} {"epoch": 0.418848167539267, "step": 201, "batch_size": 128, "mean": 27.594532012939453, "std": 40.86176681518555, "min": -73.29977416992188, "p10": -26.49628295898437, "median": 23.662281036376953, "p90": 81.18174743652344, "max": 122.58084106445312, "pos_frac": 0.7265625, "sample": [-33.170745849609375, -13.203094482421875, 50.260711669921875, 69.57489013671875, 71.06707763671875, 44.7508544921875, 35.407623291015625, 19.235809326171875, 25.078948974609375, 62.01414489746094, 74.25492858886719, -0.4910573959350586, -5.131072998046875, 23.485519409179688, -30.982208251953125, -36.982330322265625, 47.08738708496094, 86.801513671875, 60.44517517089844, 38.62150573730469, 15.852325439453125, 53.31761169433594, 79.99239349365234, 0.0, 59.70454406738281, 33.525390625, 68.96383666992188, 80.15237426757812, 34.88841247558594, 81.0635986328125, 88.87582397460938, 48.37628173828125, 44.366241455078125, -4.5658416748046875, 0.0, -61.21295166015625, 7.5615234375, -1.981292724609375, 8.85861587524414, 5.707244873046875, 36.52691650390625, 37.81846618652344, 68.43809509277344, 68.05848693847656, -40.39337158203125, 13.004608154296875, 43.693443298339844, 56.06153106689453, -14.624061584472656, 45.739715576171875, 14.549259185791016, 23.83904266357422, 21.9501953125, 37.7872314453125, -0.8920078277587891, 18.684471130371094, 17.036231994628906, 79.56430053710938, 64.50123596191406, 65.30197143554688, -7.513603210449219, 44.516326904296875, 93.16848754882812, 11.96661376953125, 84.4503173828125, 60.08921813964844, 31.43505859375, 0.0, 10.656028747558594, -73.29977416992188, 18.034759521484375, 8.588409423828125, -2.661163330078125, 109.1373291015625, 40.36279296875, 14.797225952148438, 4.497211456298828, 8.572952270507812, -8.61224365234375, 30.185333251953125, 25.228485107421875, 53.55369567871094, -23.064491271972656, -14.457389831542969, -0.083984375, 86.87046813964844, -6.079948425292969, 79.40579223632812, -25.186859130859375, 122.58084106445312, 73.1466064453125, 54.3314208984375, 53.164642333984375, -6.644927978515625, -29.551605224609375, -15.691986083984375, 9.913124084472656, 82.59878540039062, 22.569583892822266, 85.0760498046875, 69.97119140625, 64.93705749511719, -56.77911376953125, 64.93255615234375, 16.2532958984375, -6.0053863525390625, -34.675384521484375, 50.96466064453125, 81.45742797851562, -54.61180114746094, 19.143203735351562, 8.072509765625, 14.01947021484375, 4.780120849609375, -2.7014617919921875, 99.0963134765625, 13.181167602539062, 50.884117126464844, 44.94010925292969, 82.46621704101562, 12.608356475830078, -51.251373291015625, 12.447723388671875, 88.19108581542969, 9.785697937011719, 55.29632568359375, -56.33674621582031, -63.234283447265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000201.npy"} {"epoch": 0.42094240837696334, "step": 202, "batch_size": 128, "mean": 25.047733306884766, "std": 46.40884780883789, "min": -97.07757568359375, "p10": -30.54534530639648, "median": 17.480026245117188, "p90": 89.29119720458984, "max": 104.516845703125, "pos_frac": 0.6953125, "sample": [97.84796142578125, 43.253990173339844, 10.57965087890625, 15.38983154296875, 85.66571044921875, 2.8074264526367188, -17.803985595703125, 88.01873016357422, -65.2369384765625, -14.31109619140625, -5.4697265625, 93.2623291015625, 34.295989990234375, 61.981201171875, 64.51746368408203, -1.2011852264404297, -21.311614990234375, 66.22187805175781, 4.876197814941406, 101.10406494140625, 37.076595306396484, 70.674072265625, -11.24908447265625, 17.661834716796875, -10.027786254882812, 10.508758544921875, 99.2205810546875, 1.015742301940918, 53.72120666503906, 8.183723449707031, -48.665489196777344, 27.9974365234375, -29.986038208007812, 92.3558349609375, 11.73291015625, 71.06224060058594, 62.3526611328125, 100.762939453125, 15.114273071289062, -61.877655029296875, 0.19225311279296875, 98.84466552734375, 60.0230712890625, 62.819915771484375, 18.958908081054688, 60.49658203125, 55.83673095703125, 33.60297393798828, 59.8046875, -3.2233123779296875, -75.33491516113281, 53.53004455566406, 80.91612243652344, -35.281898498535156, -69.89935302734375, 51.239837646484375, 44.44610595703125, -13.07403564453125, 23.606002807617188, -31.85039520263672, 26.804885864257812, 58.10322570800781, 91.21682739257812, -23.727874755859375, -5.7083740234375, -17.072052001953125, 99.66290283203125, -1.8257293701171875, 26.194366455078125, 5.7931060791015625, 1.78375244140625, 78.3731689453125, 64.5718994140625, 5.52020263671875, 54.435821533203125, 83.09286499023438, 68.78826904296875, 38.93402099609375, 84.14369201660156, 65.07858276367188, 0.0, 104.516845703125, 0.489898681640625, 12.311981201171875, 15.42138671875, -3.5032806396484375, 1.0066184997558594, -15.28857421875, 4.7417449951171875, -26.921356201171875, 56.956817626953125, 64.91134643554688, 5.463897705078125, 17.2982177734375, 90.54690551757812, 50.890830993652344, -40.33433532714844, -33.023895263671875, -11.3074951171875, -28.84173583984375, -97.07757568359375, 88.75303649902344, 52.07685852050781, 57.78712463378906, 34.210205078125, -11.883026123046875, -24.436248779296875, 17.833465576171875, 45.18220520019531, 1.438730239868164, 56.37005615234375, 71.53617858886719, -38.83137512207031, 12.942108154296875, 56.618743896484375, -94.95437622070312, 12.59808349609375, 61.24337387084961, -48.083892822265625, -7.7391357421875, -25.500341415405273, 80.32597351074219, 4.2210693359375, 9.607025146484375, -0.6020469665527344, 93.45626831054688, -10.537986755371094, 102.27957153320312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000202.npy"} {"epoch": 0.42303664921465967, "step": 203, "batch_size": 128, "mean": 21.156368255615234, "std": 35.19729232788086, "min": -89.92071533203125, "p10": -16.945997619628898, "median": 16.838584899902344, "p90": 68.43513946533203, "max": 114.26515197753906, "pos_frac": 0.71875, "sample": [1.8023662567138672, 5.9796295166015625, 40.315277099609375, 27.86005401611328, 61.711212158203125, 19.819671630859375, 15.983436584472656, 16.338272094726562, 48.0037841796875, 65.01722717285156, 67.12786865234375, 25.131423950195312, 3.551481246948242, 36.393524169921875, 22.561843872070312, 34.9932861328125, -7.270111083984375, -45.90930938720703, 18.021682739257812, 10.203323364257812, 15.027191162109375, 0.0, -38.59423828125, 55.125633239746094, -24.514053344726562, 16.029930114746094, -24.49481201171875, 36.55059814453125, -0.1055908203125, -2.6663970947265625, 25.169464111328125, 41.16192626953125, 29.949554443359375, 74.28076171875, 52.815277099609375, -5.8858642578125, 28.2955322265625, 104.489990234375, 17.338897705078125, -13.272918701171875, 59.00288391113281, 81.54255676269531, 25.33837890625, 61.531394958496094, 1.7212905883789062, -4.3740081787109375, -9.151992797851562, 57.26905822753906, 3.6899871826171875, 57.596031188964844, 1.5587234497070312, 30.282493591308594, 22.657623291015625, 43.03376007080078, 114.26515197753906, -35.009918212890625, 47.057159423828125, -2.3144569396972656, 70.04159545898438, 84.31292724609375, 5.345893859863281, 30.946868896484375, 61.19036865234375, 1.88677978515625, -39.66339111328125, 80.23271179199219, -41.3919677734375, 74.5919189453125, 14.759075164794922, 0.2652435302734375, 22.933807373046875, 3.1318130493164062, -89.92071533203125, 62.276123046875, -5.666656494140625, -2.5880889892578125, -21.41961669921875, 0.884246826171875, -27.3385009765625, 27.7423095703125, 29.970191955566406, 11.020462036132812, -45.314849853515625, 48.05937194824219, 67.74665832519531, 15.562759399414062, 55.956787109375, 19.81329345703125, 89.55743408203125, 66.569091796875, 12.0960693359375, -7.1092529296875, -6.254638671875, 60.96160888671875, 5.39930534362793, -4.95037841796875, 34.2752685546875, -7.5800323486328125, 70.29324340820312, 10.402847290039062, 63.00144958496094, 3.0787200927734375, -4.88494873046875, 11.90045166015625, 18.111989974975586, -53.096832275390625, -8.101654052734375, -4.8018035888671875, 15.4627685546875, 87.28640747070312, 0.0, 22.646575927734375, 29.532875061035156, 25.979583740234375, -1.3590831756591797, -4.49261474609375, 25.659759521484375, 76.40994262695312, 8.126068115234375, -2.404296875, 32.078094482421875, -37.11810302734375, 7.3222198486328125, 77.94158935546875, 48.141448974609375, -15.028732299804688, 12.802501678466797, 24.759811401367188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000203.npy"} {"epoch": 0.42513089005235605, "step": 204, "batch_size": 128, "mean": 29.922130584716797, "std": 41.06574249267578, "min": -79.56085205078125, "p10": -18.38241577148437, "median": 25.67047119140625, "p90": 82.15828552246093, "max": 132.55264282226562, "pos_frac": 0.7578125, "sample": [103.080810546875, 65.30720520019531, 99.947265625, 14.89630126953125, 12.83251953125, 83.02536010742188, 5.040996551513672, 43.04267120361328, 4.827587127685547, 96.56680297851562, 12.250732421875, 7.946929931640625, 66.8076171875, -20.390472412109375, -33.428375244140625, -79.56085205078125, 44.49913024902344, -30.56768798828125, 7.244056701660156, 23.5247802734375, 79.12420654296875, 5.903038024902344, 31.501495361328125, 67.50715637207031, 24.1309814453125, 3.78900146484375, 20.38385772705078, -40.61042785644531, 17.173675537109375, 28.0806884765625, 31.15625, 132.55264282226562, 7.0331268310546875, -27.339187622070312, 42.1038818359375, -14.18670654296875, 27.037813186645508, 113.46343994140625, -3.86737060546875, 73.5927734375, 73.06111145019531, 14.605117797851562, 46.46734619140625, -13.29177474975586, 66.85763549804688, 70.48863220214844, 11.337600708007812, 102.3848876953125, 81.78668212890625, 100.60317993164062, 67.62921905517578, 52.856292724609375, -0.90216064453125, 60.631202697753906, 32.12939453125, -4.3646240234375, 65.34030151367188, 40.09545135498047, 44.546836853027344, -26.55181884765625, 29.51531982421875, 67.18753051757812, 26.529022216796875, -15.913223266601562, -7.412261962890625, 96.6053466796875, 25.292510986328125, 72.40338134765625, -66.2059326171875, 76.26019287109375, -6.68815803527832, 71.7120361328125, -0.9788055419921875, -10.657798767089844, 7.521698951721191, 23.333084106445312, 31.93518829345703, 79.03900909423828, -4.105705261230469, 20.40545654296875, 7.4281463623046875, -6.611846923828125, -57.48772430419922, 12.62603759765625, 52.15716552734375, -3.717132568359375, 47.750701904296875, 11.1158447265625, 72.9945068359375, 69.15994262695312, 86.80926513671875, 21.903366088867188, -36.6064453125, 72.49493408203125, -36.42963409423828, 5.14654541015625, 74.57308959960938, -34.35516357421875, 31.102401733398438, 77.91162109375, 9.22802734375, -0.220672607421875, 7.5419921875, 40.63330078125, 60.126625061035156, 94.2869873046875, 26.048431396484375, 15.571891784667969, 46.4796142578125, -0.44012451171875, 62.2520751953125, -0.5088348388671875, 50.057281494140625, 74.51480102539062, 13.855865478515625, 26.138137817382812, 46.38221740722656, 3.7381744384765625, 8.180477142333984, 20.863067626953125, -17.521820068359375, 28.38214111328125, 85.42103576660156, 34.14337158203125, -35.087066650390625, 109.55352783203125, -0.25524139404296875, 5.7899169921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000204.npy"} {"epoch": 0.4272251308900524, "step": 205, "batch_size": 128, "mean": 26.87209701538086, "std": 43.33024215698242, "min": -72.4041748046875, "p10": -28.713420104980468, "median": 24.83022117614746, "p90": 79.48435058593749, "max": 127.5194091796875, "pos_frac": 0.75, "sample": [27.239227294921875, 0.0, -1.5131072998046875, 3.672567367553711, 24.12509536743164, 72.52447509765625, -24.21392822265625, -17.721771240234375, -40.41798400878906, 0.41727256774902344, 127.5194091796875, 62.05177307128906, 5.949405670166016, 8.736625671386719, -23.13323974609375, 18.23876190185547, 102.44775390625, 93.619140625, 69.30218505859375, 10.211883544921875, 8.426437377929688, 34.886863708496094, -14.2894287109375, 60.343994140625, 31.08123779296875, -22.30047607421875, 68.21719360351562, 14.080070495605469, 95.29415893554688, 45.42578125, 38.11717224121094, 43.30559539794922, 1.0757904052734375, -9.66748046875, 21.536529541015625, 6.791677474975586, 53.693748474121094, 12.79046630859375, 38.701881408691406, 12.345947265625, 30.47149658203125, 1.3395233154296875, -46.14532470703125, 60.263458251953125, 12.35162353515625, 41.1077880859375, 0.0, 30.64788818359375, -62.74805450439453, 29.682647705078125, 3.966571807861328, 25.289913177490234, 84.50628662109375, 85.147705078125, -52.14239501953125, 7.44818115234375, 76.55316162109375, 55.92958068847656, 3.9786376953125, 53.463043212890625, -62.777122497558594, -1.5423583984375, 6.104034423828125, 47.2554931640625, -30.114990234375, 7.47064208984375, 82.42282104492188, -38.30975341796875, -59.65045166015625, 118.0318603515625, 72.85113525390625, 63.6605224609375, -49.813385009765625, 44.42340087890625, -22.25664520263672, 65.68806457519531, 3.637115478515625, 62.76356506347656, 63.160892486572266, -6.925537109375, 24.370529174804688, 100.10336303710938, 59.81103515625, 78.22500610351562, 31.30670166015625, 24.332834243774414, -69.01119995117188, 1.5946044921875, 60.74199676513672, 69.32621765136719, 102.33346557617188, 75.17471313476562, 106.50079345703125, -20.225433349609375, 31.69329833984375, 90.82391357421875, 74.35404968261719, 12.267032623291016, 12.13679313659668, 2.1754608154296875, 101.52294921875, 68.70635986328125, -4.433990478515625, -39.418670654296875, 42.548431396484375, 52.088134765625, -0.667144775390625, 62.713623046875, 60.52442169189453, 68.302978515625, 20.74737548828125, 9.6513671875, -72.4041748046875, 17.222627639770508, 37.130615234375, -28.112747192382812, -51.81431579589844, 51.94633483886719, -12.152984619140625, 65.16326904296875, 44.35107421875, 55.405517578125, -0.2435760498046875, 49.680755615234375, 3.3126678466796875, 76.613525390625, 0.0, 27.10515594482422], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000205.npy"} {"epoch": 0.4293193717277487, "step": 206, "batch_size": 128, "mean": 25.229339599609375, "std": 42.5936164855957, "min": -104.32659912109375, "p10": -26.867335510253906, "median": 18.11547088623047, "p90": 81.7059341430664, "max": 103.29168701171875, "pos_frac": 0.7265625, "sample": [63.778289794921875, -7.969470977783203, -72.20980834960938, 65.451904296875, 103.29168701171875, -4.640045166015625, -19.361328125, 28.610137939453125, 13.074951171875, 75.07412719726562, 39.919769287109375, 81.69688415527344, 72.52267456054688, 2.5010528564453125, -28.261322021484375, 29.671875, 35.26904296875, -7.383064270019531, -4.4803009033203125, 81.45135498046875, 30.325773239135742, 72.38662719726562, -3.2258834838867188, -24.147216796875, -2.612579345703125, 9.99658203125, -38.345977783203125, 51.30975341796875, 21.927223205566406, 18.072250366210938, 2.46820068359375, -10.6356201171875, -65.72491455078125, 67.21351623535156, 69.72613525390625, -21.270671844482422, 99.612548828125, 81.72705078125, 94.164306640625, 45.25787353515625, 9.994873046875, 26.66809844970703, -7.8331298828125, 28.56793212890625, 44.463165283203125, 48.915985107421875, -28.624847412109375, 84.40744018554688, 78.25540161132812, 75.54476928710938, 27.507781982421875, 58.55352020263672, 60.10150146484375, 43.3165283203125, 8.00224494934082, 40.8182373046875, 43.44581604003906, 99.88607788085938, -104.32659912109375, 17.856964111328125, 37.413421630859375, -42.868690490722656, -2.17376708984375, 66.25263977050781, 67.09373474121094, 9.08544921875, 10.19342041015625, -2.58306884765625, -4.502832412719727, 40.93756103515625, 10.7572021484375, 68.49507141113281, 41.53240966796875, -64.78173828125, 84.63714599609375, 3.2967376708984375, -11.347915649414062, 30.43402862548828, 10.90533447265625, 0.9250869750976562, -26.269912719726562, -28.981231689453125, 89.4385986328125, 59.913818359375, 66.369873046875, 9.16656494140625, 49.013092041015625, -69.55411529541016, 9.84419059753418, 0.7543144226074219, 59.149322509765625, 67.04814147949219, 61.44342041015625, 6.170135498046875, -21.6209716796875, 45.0189208984375, 18.15869140625, -14.081924438476562, 0.35223388671875, -9.12991714477539, 70.44827270507812, -0.091827392578125, 92.57186889648438, 83.976318359375, -6.610321044921875, 69.27389526367188, -37.310455322265625, -33.99162292480469, -46.37713623046875, 14.355865478515625, 5.3227081298828125, 99.780517578125, 68.72016906738281, 7.5435791015625, 1.8507881164550781, 86.491455078125, 61.69269561767578, 5.858207702636719, 82.02249145507812, 36.01568603515625, 7.452392578125, 12.9544677734375, 0.097686767578125, 70.20957946777344, 48.343414306640625, 13.245834350585938, -16.1251220703125, 3.9742889404296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000206.npy"} {"epoch": 0.431413612565445, "step": 207, "batch_size": 128, "mean": 23.381336212158203, "std": 48.70920181274414, "min": -94.3760986328125, "p10": -30.222594451904296, "median": 15.454200744628906, "p90": 88.17651062011718, "max": 133.87237548828125, "pos_frac": 0.7109375, "sample": [66.2877197265625, 42.84010314941406, 15.097137451171875, -8.75851821899414, 16.1123046875, 0.39009857177734375, 16.25201416015625, 114.3145751953125, 31.46710205078125, -2.8615188598632812, 25.846174240112305, 108.84515380859375, -51.1619873046875, 3.82763671875, -3.47430419921875, 81.89693450927734, 8.362716674804688, 48.36138916015625, 121.26191711425781, -10.5406494140625, 81.34600067138672, -10.871231079101562, -94.3760986328125, -30.43628692626953, 66.49789428710938, 28.913406372070312, -24.44256591796875, 18.509368896484375, -73.86749267578125, 59.252593994140625, 71.96493530273438, 15.739730834960938, 11.2734375, -29.7230224609375, 68.61001586914062, 70.5132064819336, 13.095794677734375, -1.1559181213378906, 20.844974517822266, 91.75839233398438, -10.716934204101562, 10.267364501953125, 34.722198486328125, 7.2181396484375, 7.791748046875, -25.9381103515625, 46.32634735107422, 1.74176025390625, -82.09466552734375, 0.259796142578125, -69.14447021484375, -13.89501953125, -50.997222900390625, 25.5457763671875, 82.478759765625, 0.0, 15.388961791992188, -30.131011962890625, 59.923614501953125, -0.0036468505859375, 104.94921875, -10.454437255859375, -41.69231414794922, 27.912689208984375, 15.519439697265625, -89.69622802734375, 91.40126037597656, -64.58183288574219, 8.858963012695312, -13.036941528320312, 42.42352294921875, -9.417816162109375, 38.04771423339844, 84.234130859375, 92.89181518554688, 45.14997863769531, 59.82301330566406, 2.9807281494140625, 0.9461669921875, 88.53877258300781, 5.715389251708984, -64.95925903320312, 5.6019287109375, 31.098255157470703, 7.480409622192383, 133.87237548828125, 10.583305358886719, 79.65758514404297, -25.083709716796875, 2.3435630798339844, 11.03790283203125, 10.461410522460938, -28.25408935546875, 96.04579162597656, 15.013671875, 66.52607727050781, 44.536285400390625, 24.776519775390625, 50.868247985839844, -78.55947875976562, 97.7998046875, 84.01258850097656, -15.32562255859375, 72.13854217529297, 6.7115478515625, 105.799072265625, 10.28839111328125, 3.4370574951171875, 40.11717224121094, 76.298095703125, 8.073638916015625, 46.009918212890625, -2.236572265625, 88.02125549316406, 77.44903564453125, -15.701652526855469, 72.1180419921875, 78.9912109375, 107.02947998046875, -17.23699951171875, 16.274330139160156, -61.01689147949219, 74.24331665039062, 77.59239196777344, -8.713180541992188, 21.576248168945312, 73.27826690673828, 15.666259765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000207.npy"} {"epoch": 0.43350785340314135, "step": 208, "batch_size": 128, "mean": 26.568960189819336, "std": 44.099281311035156, "min": -107.11105346679688, "p10": -18.262093353271485, "median": 19.01504898071289, "p90": 78.69919433593749, "max": 138.64422607421875, "pos_frac": 0.7421875, "sample": [5.047760009765625, -8.634994506835938, -31.314453125, 17.745468139648438, -82.35726928710938, 5.030418395996094, -12.439142227172852, 24.851348876953125, 4.6417388916015625, 75.88133239746094, -18.732364654541016, 9.56964111328125, 44.37469482421875, 42.12638854980469, 60.95246505737305, 114.60305786132812, 18.016860961914062, 11.64145278930664, 19.17010498046875, 124.43399047851562, -82.93746948242188, 77.33602905273438, -7.1262664794921875, 54.33202362060547, 13.52996826171875, 68.10836791992188, 105.84451293945312, 68.80792236328125, 14.972686767578125, -52.963104248046875, 138.64422607421875, 50.057533264160156, 53.6207275390625, 69.73141479492188, -15.277824401855469, 100.03961181640625, -107.11105346679688, 14.185806274414062, 26.258392333984375, 9.12420654296875, 52.349639892578125, -89.20782470703125, 75.76478576660156, -8.727298736572266, 75.9853515625, 29.88818359375, 33.483367919921875, 102.06494140625, 0.0, 124.82745361328125, -30.827713012695312, -6.503416061401367, 15.652252197265625, -18.386245727539062, 2.939697265625, 122.59088134765625, 12.131973266601562, 44.60627746582031, 30.911651611328125, 90.017822265625, 6.277988433837891, 34.59220886230469, -10.446998596191406, 39.6968994140625, -12.643768310546875, 4.375518798828125, 4.127960205078125, -2.047943115234375, 26.07648468017578, 17.074508666992188, 123.470947265625, 0.0, 6.0454864501953125, 53.36114501953125, 46.514183044433594, 77.65316772460938, -6.290977478027344, 81.13992309570312, 31.328857421875, 62.32958221435547, 37.3187255859375, 41.316619873046875, 64.48804473876953, 89.06954956054688, -7.6467437744140625, 53.386077880859375, -9.6722412109375, -0.7131729125976562, 9.394485473632812, 38.970184326171875, -9.46246337890625, 22.770965576171875, 45.10186767578125, 11.092010498046875, 69.45843505859375, 32.985107421875, 62.804908752441406, 20.243621826171875, 68.7171630859375, -13.032218933105469, 10.701797485351562, 26.22525405883789, 5.099639892578125, 12.233177185058594, 16.94965362548828, 12.292095184326172, -1.7572708129882812, 70.67730712890625, 16.853843688964844, 27.201095581054688, 18.85999298095703, -23.782501220703125, 62.46129608154297, 31.303558349609375, 7.41021728515625, -48.919769287109375, 16.716156005859375, 2.4238948822021484, -26.302749633789062, 46.20281982421875, 57.679595947265625, 39.1663818359375, -18.208885192871094, 81.78497314453125, 58.589508056640625, -36.29132080078125, -4.082275390625, 46.76531982421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000208.npy"} {"epoch": 0.4356020942408377, "step": 209, "batch_size": 128, "mean": 22.836122512817383, "std": 44.23064041137695, "min": -99.78048706054688, "p10": -32.44401550292969, "median": 13.680587768554688, "p90": 80.83384704589844, "max": 118.91986083984375, "pos_frac": 0.625, "sample": [111.06605529785156, -8.90728759765625, 70.57884979248047, 69.05484771728516, -9.624298095703125, -4.1846466064453125, 29.278839111328125, -32.10467529296875, -35.52301025390625, 18.94683837890625, 77.6593017578125, -9.944259643554688, -4.72149658203125, 8.071945190429688, 100.79736328125, 84.83397674560547, 0.5072021484375, 6.580549240112305, 7.989898681640625, 76.64962768554688, 84.00152587890625, -3.392059326171875, 4.023773193359375, -58.760711669921875, 21.574127197265625, 10.567642211914062, 48.38056564331055, 10.878509521484375, -24.639678955078125, 72.24557495117188, 78.08499145507812, -48.15167236328125, -8.030548095703125, -52.34553527832031, -17.49383544921875, -51.2054443359375, 109.00885009765625, 3.22589111328125, 0.0, 11.870086669921875, -23.49395751953125, -40.4168701171875, 58.25701904296875, -30.96197509765625, 28.153717041015625, 50.300750732421875, 57.26200485229492, 4.850959777832031, 9.2733154296875, 96.70014953613281, 38.704010009765625, 31.307708740234375, 35.3397216796875, -33.235809326171875, 27.1732177734375, -3.5574092864990234, -9.242828369140625, 26.62480926513672, 65.05302429199219, -10.407381057739258, 29.948699951171875, -47.89404296875, -5.732757568359375, 80.56756591796875, 81.6759033203125, 27.90789794921875, -9.913726806640625, -20.788818359375, 93.4854736328125, 1.371856689453125, 20.99383544921875, 86.86566162109375, 15.4910888671875, 58.42280197143555, -4.5826263427734375, 0.0, 55.0106201171875, 61.5672607421875, -20.87866973876953, 59.595794677734375, -17.8516845703125, -5.1171417236328125, 57.55773162841797, -0.793121337890625, -1.1334228515625, 81.45516967773438, 73.22018432617188, -37.14776611328125, 11.19415283203125, 7.767822265625, 39.31897735595703, -43.37236022949219, -18.802215576171875, 64.96122741699219, -2.3463821411132812, -0.85516357421875, 26.50677490234375, 47.95740509033203, 118.91986083984375, -99.78048706054688, -11.258377075195312, 6.644115447998047, -36.647674560546875, -7.262241363525391, -0.9545440673828125, -74.12539672851562, -10.0863037109375, 54.30458068847656, 66.75497436523438, 69.44842529296875, 51.78564453125, 72.33448791503906, 67.22360229492188, 36.36724853515625, 57.315216064453125, 48.10333251953125, 35.65443420410156, 73.418701171875, 4.026275634765625, 104.4840087890625, 31.5548095703125, -29.2042236328125, 53.172943115234375, 46.347412109375, 85.22311401367188, 76.65707397460938, -10.247772216796875, 72.6845703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000209.npy"} {"epoch": 0.437696335078534, "step": 210, "batch_size": 128, "mean": 26.872543334960938, "std": 44.61758804321289, "min": -96.81561279296875, "p10": -18.180931091308594, "median": 24.433792114257812, "p90": 82.48713073730468, "max": 161.77490234375, "pos_frac": 0.6953125, "sample": [-10.669044494628906, 0.10466766357421875, 40.2843017578125, -5.29986572265625, 68.73507690429688, 29.204864501953125, 101.982421875, 101.56964111328125, 50.1102294921875, 55.382568359375, 80.64332580566406, 60.69179153442383, 3.337005615234375, 62.03688049316406, 57.08673095703125, -2.7976303100585938, 62.023529052734375, -0.31542205810546875, 54.26387023925781, 4.314422607421875, 45.54500198364258, -1.7363433837890625, -58.51898193359375, 24.745986938476562, 58.163963317871094, 45.901214599609375, 8.66339111328125, 161.77490234375, -50.9041748046875, -7.0617218017578125, 53.157257080078125, 23.866790771484375, 14.224945068359375, -1.1039276123046875, 71.77435302734375, -15.83599853515625, 13.338069915771484, 48.41485595703125, 1.9750328063964844, 25.66009521484375, -5.030731201171875, 72.50515747070312, 15.068031311035156, -8.151870727539062, 11.335060119628906, 60.1873779296875, 29.050079345703125, -3.77252197265625, 71.16033935546875, 14.950422286987305, 24.635009765625, -3.9474334716796875, 59.475921630859375, 81.99200439453125, 11.476737976074219, 58.23688507080078, 40.313385009765625, -40.70341491699219, 93.8453369140625, -7.037445068359375, 55.27239990234375, 58.776519775390625, 77.16856384277344, 61.69358444213867, 28.134002685546875, 43.24444580078125, -81.2088623046875, -8.781448364257812, 10.2486572265625, 50.3892822265625, -67.38722229003906, -27.41607666015625, 83.64242553710938, 53.448486328125, 94.49966430664062, 83.82450866699219, 4.727422714233398, 7.165275573730469, -13.7503662109375, -96.81561279296875, 65.1999282836914, 18.712127685546875, -1.0043010711669922, 96.8438720703125, 47.084022521972656, 75.03738403320312, 116.62786865234375, 62.43743896484375, -3.21160888671875, -16.1517333984375, -18.260986328125, 14.499996185302734, 3.0646896362304688, -24.518356323242188, 28.286848068237305, 58.5831298828125, -18.146621704101562, 106.04238891601562, -15.792022705078125, 50.55866241455078, 56.024932861328125, 113.697021484375, 10.533329010009766, -12.092514038085938, 94.74716186523438, 19.23711395263672, 84.00031280517578, -32.54719543457031, 10.68048095703125, -17.926788330078125, 48.65647888183594, 24.232574462890625, -5.2079620361328125, 16.402191162109375, 15.184402465820312, -89.09773254394531, 14.393280029296875, 54.881072998046875, 31.46417236328125, -68.870849609375, 57.19939422607422, -7.277801513671875, 58.82470703125, 50.4715576171875, -27.942108154296875, -7.7996826171875, -5.7581787109375, 40.48524475097656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000210.npy"} {"epoch": 0.4397905759162304, "step": 211, "batch_size": 128, "mean": 34.97750473022461, "std": 40.57498550415039, "min": -68.38410949707031, "p10": -14.331985473632812, "median": 34.15081787109375, "p90": 88.5794448852539, "max": 131.98886108398438, "pos_frac": 0.8203125, "sample": [62.26092529296875, 27.4788818359375, 104.0966796875, 37.77117919921875, 71.51739501953125, 86.71343994140625, 6.114116668701172, 6.9378662109375, -14.29486083984375, 59.63482666015625, 63.14471435546875, -66.76467895507812, 40.03082275390625, 42.004791259765625, -14.845222473144531, 80.99874877929688, 2.1860885620117188, 0.7269039154052734, 2.316650390625, -14.6622314453125, 12.37255859375, 63.924781799316406, 20.554306030273438, 45.18011474609375, 69.56825256347656, 25.5775146484375, 8.001571655273438, -51.65704345703125, 68.18104553222656, -9.01025390625, 0.6520271301269531, 71.78025817871094, 4.6767578125, 59.205223083496094, -3.5219802856445312, 63.365936279296875, 5.422811508178711, -21.773399353027344, 19.26134490966797, 78.38943481445312, 5.6739501953125, 13.37689208984375, 9.9453125, -1.00067138671875, -2.35845947265625, 113.093505859375, 88.65971374511719, 51.76104736328125, 5.2527313232421875, 14.40243148803711, 62.868133544921875, -18.560348510742188, 0.936065673828125, 0.4986572265625, 131.98886108398438, 68.48175048828125, 84.11297607421875, -24.47930145263672, 20.24188232421875, 59.79052734375, -7.174072265625, -25.34320068359375, 5.1968536376953125, 119.35043334960938, -14.418609619140625, 7.2533111572265625, -3.99981689453125, 66.90682983398438, 88.5450439453125, 13.619049072265625, -23.04736328125, -1.2078857421875, -2.385274887084961, 59.715110778808594, 11.504146575927734, 14.769350051879883, 71.71903991699219, 60.82209777832031, 69.36620330810547, 38.07164001464844, 49.57391357421875, 51.775634765625, 5.309501647949219, 70.9373779296875, 65.28305053710938, 120.96847534179688, 17.282958984375, 1.4409332275390625, 65.07481384277344, 19.606788635253906, -11.51800537109375, 108.54098510742188, 97.24197387695312, -30.192657470703125, 88.8509521484375, 37.069366455078125, 78.65260314941406, 54.313568115234375, 20.931594848632812, 13.56549072265625, 69.55807495117188, 96.36566162109375, 39.74310302734375, 10.85867691040039, -68.38410949707031, 5.086421966552734, 43.97857666015625, 42.202850341796875, 94.81332397460938, 44.19905090332031, 48.563079833984375, 14.939643859863281, 75.61410522460938, 37.783050537109375, 31.037811279296875, 31.232269287109375, 99.96414184570312, 91.2752685546875, 64.92039489746094, 55.30525207519531, 21.618270874023438, 23.06873321533203, -22.799545288085938, 81.60015869140625, 48.987876892089844, 41.052276611328125, 1.967926025390625, 80.39212036132812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000211.npy"} {"epoch": 0.4418848167539267, "step": 212, "batch_size": 128, "mean": 26.284400939941406, "std": 44.63903045654297, "min": -121.296875, "p10": -31.317195129394527, "median": 22.796493530273438, "p90": 82.1748908996582, "max": 120.28167724609375, "pos_frac": 0.7265625, "sample": [12.484561920166016, -0.224365234375, 27.351531982421875, 53.993553161621094, -17.649627685546875, -33.165985107421875, 1.1291961669921875, 13.71484375, 4.542549133300781, -34.9697265625, 93.11237335205078, 7.2144012451171875, 116.1865234375, 4.534115791320801, 47.0577392578125, -44.04057312011719, 32.62749481201172, 69.15530395507812, 68.35140991210938, 29.973785400390625, 64.93516540527344, 21.186248779296875, -22.208770751953125, 17.82781982421875, 70.36911010742188, 14.02508544921875, -30.524856567382812, -63.493896484375, 43.791412353515625, 51.026390075683594, -13.918743133544922, -52.24082946777344, 8.679275512695312, 17.791793823242188, 48.74475860595703, 53.8851318359375, 24.371490478515625, 89.08204650878906, 57.27677917480469, -5.1843414306640625, 75.74627685546875, -85.77789306640625, -1.214019775390625, -6.982479095458984, -10.9827880859375, 69.84162902832031, 53.047607421875, 69.91232299804688, -9.70672607421875, -44.30647277832031, 59.606414794921875, 17.4442138671875, -17.733489990234375, 98.0140380859375, 43.73030090332031, -7.091583251953125, -1.361053466796875, 63.28106689453125, 8.92413330078125, -5.7166748046875, -7.51007080078125, 4.389642715454102, 70.4775161743164, 27.1329345703125, -3.0471553802490234, 70.39807891845703, -11.284942626953125, 55.80720520019531, 52.642822265625, 86.33973693847656, 41.72465515136719, 74.16183471679688, -56.7938232421875, 41.72900390625, 75.05697631835938, 19.909652709960938, 51.161766052246094, 63.205474853515625, -2.0005435943603516, 97.2730712890625, 48.37359619140625, 98.09963989257812, 23.532058715820312, 54.49884033203125, -104.92997741699219, 20.073516845703125, -1.94970703125, -36.33447265625, 37.76347351074219, 7.456157684326172, 71.38018798828125, 46.7210693359375, 9.941619873046875, 6.616039276123047, 113.099365234375, 59.62590026855469, 20.774658203125, 87.97295379638672, 56.24878692626953, 49.57762145996094, 45.08610534667969, 120.28167724609375, 53.14691162109375, -6.412567138671875, 19.920745849609375, -121.296875, 65.89126586914062, 97.764892578125, -35.30000305175781, -11.805328369140625, -8.454856872558594, -40.88677978515625, 31.766510009765625, 13.101964950561523, 43.971160888671875, 1.5010986328125, 33.99787902832031, 22.060928344726562, 39.6651496887207, 13.262702941894531, 83.31820678710938, 11.016693115234375, 48.00275421142578, 81.68489837646484, 9.29754638671875, 3.955657958984375, 103.0283203125, 12.04638671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000212.npy"} {"epoch": 0.44397905759162304, "step": 213, "batch_size": 128, "mean": 24.671600341796875, "std": 44.34601974487305, "min": -76.57232666015625, "p10": -25.649998474121094, "median": 14.640951156616211, "p90": 89.03671569824218, "max": 134.7562255859375, "pos_frac": 0.6953125, "sample": [95.32403564453125, 72.83842468261719, -40.91925048828125, 80.17413330078125, -30.938232421875, 26.532241821289062, 67.95234680175781, 60.355987548828125, -4.0870361328125, 28.053070068359375, 5.63037109375, 12.961456298828125, 23.18218994140625, -37.13517761230469, 49.037574768066406, 120.50942993164062, 29.72472381591797, 19.675933837890625, 20.92578125, 53.16473388671875, 27.45245361328125, -11.050872802734375, 120.70162963867188, -4.910400390625, 60.10227966308594, 6.07177734375, -5.58770751953125, 16.19140625, 28.394500732421875, 14.7496337890625, 5.3779449462890625, -26.77056884765625, -14.72955322265625, -3.3125762939453125, 86.11094665527344, 30.260635375976562, 68.242431640625, -3.462921142578125, 10.781265258789062, -23.39642333984375, 84.81556701660156, 42.04985046386719, 0.371063232421875, 67.33450317382812, -18.217132568359375, 76.41754150390625, 54.889404296875, 6.823974609375, -40.9466552734375, 39.64752197265625, 19.249832153320312, 68.95645141601562, 91.70822143554688, 14.532268524169922, 102.2586669921875, -25.542648315429688, -8.090957641601562, 62.66670227050781, -20.972991943359375, 94.89457702636719, 15.588958740234375, 13.998046875, 24.03717803955078, -61.48548889160156, 7.8863525390625, 34.6253662109375, 102.99871826171875, -25.900482177734375, -15.38995361328125, -10.910598754882812, 70.17887878417969, -15.34246826171875, 26.266815185546875, -76.57232666015625, 8.24493408203125, 2.2962722778320312, 53.83385467529297, 87.89178466796875, 8.416629791259766, 32.9644775390625, -19.6815185546875, 6.260709762573242, -2.398080825805664, 2.911529541015625, -0.7859573364257812, 68.9132080078125, -41.893524169921875, -5.897705078125, 26.379486083984375, 114.68911743164062, 63.40802764892578, 93.23101806640625, 5.060321807861328, 9.3306884765625, 39.771026611328125, -9.955066680908203, 96.16064453125, -43.469451904296875, -25.40972900390625, 134.7562255859375, 1.6301422119140625, 7.692840576171875, -6.61737060546875, 75.36504364013672, -32.23150634765625, 107.869140625, 77.85545349121094, -10.471763610839844, 44.95304870605469, 0.0, 12.34085464477539, 52.98046875, 6.8331146240234375, 85.84317016601562, 12.010627746582031, -75.2412109375, 71.72915649414062, 25.312824249267578, -6.3118743896484375, 19.208984375, 3.1598167419433594, 73.5789794921875, -28.745147705078125, 0.747802734375, 103.96701049804688, 16.040267944335938, -24.690216064453125, 1.1228256225585938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000213.npy"} {"epoch": 0.44607329842931936, "step": 214, "batch_size": 128, "mean": 22.22649383544922, "std": 43.38418197631836, "min": -84.15560913085938, "p10": -22.50760803222656, "median": 17.49911117553711, "p90": 77.7585952758789, "max": 148.61441040039062, "pos_frac": 0.7109375, "sample": [53.33030700683594, 63.03955078125, -17.779823303222656, 6.0855712890625, 26.76336669921875, 82.07308959960938, 19.748092651367188, -10.02423095703125, -44.32318115234375, -17.182373046875, 27.18450927734375, 25.346466064453125, -14.509185791015625, 28.513458251953125, 39.0908203125, 53.37835693359375, -84.15560913085938, 72.12415313720703, 4.3331756591796875, 23.525482177734375, -3.8416671752929688, -15.594970703125, 21.438880920410156, 61.01103210449219, 32.086700439453125, 17.885482788085938, 61.89533233642578, -22.393798828125, 5.4608154296875, 2.2775497436523438, 6.1611785888671875, 4.544189453125, 15.781013488769531, 2.000885009765625, 57.10658264160156, 21.73291015625, 148.61441040039062, 60.22196960449219, 93.56246948242188, -69.71090698242188, 6.5666351318359375, -40.333892822265625, 5.173496246337891, 25.20526123046875, 28.208984375, 6.7856903076171875, -3.8712158203125, 83.70166015625, 19.5078125, -7.23432731628418, 62.73577117919922, 27.20269775390625, -48.94642639160156, 13.344985961914062, -4.6552276611328125, 70.60957336425781, 39.0767822265625, 50.543701171875, -17.360260009765625, 74.6676025390625, 67.56034851074219, 96.92288208007812, 59.74040985107422, 27.75506591796875, 22.477645874023438, -53.72291564941406, -9.915847778320312, -14.225830078125, 2.401611328125, 0.0, -22.773162841796875, 5.228057861328125, -10.07794189453125, 14.2484130859375, 64.1783447265625, -3.2530517578125, 41.0843505859375, -16.948040008544922, 6.844322204589844, 23.32830047607422, -16.54852294921875, -80.20855712890625, 83.87518310546875, 6.817535400390625, 69.01986694335938, -29.561553955078125, 80.0740966796875, 11.900077819824219, 22.041015625, 17.11273956298828, -77.03414916992188, 4.099128723144531, 27.414215087890625, 60.638824462890625, 77.374755859375, 10.92724609375, -14.7418212890625, 67.08158874511719, 12.34259033203125, 63.71928405761719, 78.65422058105469, 109.1734619140625, 5.49884033203125, 67.73342895507812, 0.141632080078125, 85.03396606445312, 86.5377197265625, 27.776519775390625, -21.762405395507812, 71.26445007324219, 131.43914794921875, 17.978073120117188, -14.1722412109375, -38.456268310546875, -25.34527587890625, 70.90342712402344, 13.269895553588867, -53.748382568359375, 3.651611328125, -2.234649658203125, -13.940177917480469, 23.27149200439453, 10.54010009765625, 50.44895935058594, -0.46822357177734375, 137.1187744140625, 43.317649841308594, 56.4375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000214.npy"} {"epoch": 0.4481675392670157, "step": 215, "batch_size": 128, "mean": 24.527870178222656, "std": 48.55804443359375, "min": -100.33366394042969, "p10": -34.2912841796875, "median": 19.47370147705078, "p90": 86.84089508056641, "max": 149.60784912109375, "pos_frac": 0.6875, "sample": [46.31542205810547, 27.926177978515625, 52.636505126953125, 53.134368896484375, 108.0450439453125, 68.57806396484375, 77.7530517578125, 34.891502380371094, 63.0242919921875, -10.736404418945312, 96.87741088867188, 86.7919921875, 92.58343505859375, -25.243087768554688, 7.7325592041015625, 24.6656494140625, -25.881134033203125, 124.74322509765625, 126.4376220703125, 70.16421508789062, 72.5986328125, 0.233489990234375, 0.0, -26.27557373046875, -5.33587646484375, -12.101325988769531, 33.71783447265625, 65.41741943359375, 122.2271728515625, -9.89361572265625, 78.5557861328125, 0.985595703125, -12.593841552734375, -43.37510681152344, -0.09952354431152344, 11.157974243164062, 79.4231948852539, -2.6326904296875, 16.087448120117188, 122.06497192382812, -52.062408447265625, 55.694580078125, -26.9537353515625, -29.274505615234375, 1.2159309387207031, -8.951568603515625, -46.297271728515625, 81.72084045410156, 71.86102294921875, 17.990379333496094, 37.20106506347656, 149.60784912109375, 17.9278564453125, 24.3609619140625, -48.697052001953125, 7.3750457763671875, 7.909423828125, -40.110931396484375, 0.0, 76.78445434570312, -19.739944458007812, 5.271518707275391, 47.76263427734375, -34.750274658203125, 49.28240966796875, 53.99168395996094, -66.15615844726562, 4.8780364990234375, 86.95500183105469, 24.239227294921875, 25.989395141601562, 8.513843536376953, -13.826744079589844, 76.2099609375, -58.05029296875, 70.50582885742188, 6.2752685546875, -100.33366394042969, 44.570343017578125, 42.662872314453125, 18.979415893554688, 30.44690704345703, 19.967987060546875, -14.822235107421875, -67.84449005126953, -11.065820693969727, 73.1376953125, 13.57476806640625, 87.76657104492188, -34.094573974609375, -10.971458435058594, 26.6400146484375, 15.922615051269531, -2.07672119140625, 6.771324157714844, 29.22620391845703, 2.4276485443115234, 31.584793090820312, 107.530517578125, 140.4442138671875, -31.43634033203125, 25.45708465576172, -64.40501403808594, 0.770233154296875, -65.46121215820312, -11.6246337890625, 62.99652099609375, 82.67876434326172, -30.89190673828125, 81.78591918945312, 21.083892822265625, -7.887119293212891, 9.862972259521484, 18.0877685546875, 15.373992919921875, 31.80462646484375, 21.324172973632812, 75.91397094726562, -47.312744140625, 27.6229248046875, 56.54774475097656, -0.253448486328125, 13.488037109375, 87.72406005859375, 21.342742919921875, 38.742279052734375, 22.615966796875, 77.916015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000215.npy"} {"epoch": 0.450261780104712, "step": 216, "batch_size": 128, "mean": 26.709692001342773, "std": 51.42607116699219, "min": -127.19464111328125, "p10": -39.17650299072265, "median": 22.450897216796875, "p90": 95.9166046142578, "max": 120.01168823242188, "pos_frac": 0.71875, "sample": [7.9924774169921875, 34.38597106933594, 61.477691650390625, 30.896697998046875, 16.3570556640625, -35.50837707519531, -127.19464111328125, 34.90863037109375, 63.7103271484375, -12.42281723022461, 73.24824523925781, 83.20541381835938, 66.8250732421875, -19.31446075439453, 7.6578521728515625, 37.41755676269531, 116.5386962890625, 9.580192565917969, 99.7393798828125, 110.84869384765625, 22.09539794921875, -80.63848876953125, 26.996597290039062, -1.2770843505859375, -22.26584243774414, 18.831756591796875, -88.84844970703125, 11.965225219726562, 60.75367736816406, 73.4033203125, -39.91554260253906, -51.25260925292969, -93.3504638671875, 1.3651809692382812, 65.08876037597656, 50.2989501953125, 74.37356567382812, -46.079681396484375, 101.39175415039062, 69.42662048339844, -48.67477798461914, 81.67329406738281, 114.79913330078125, 70.65071868896484, 70.61175537109375, 60.671051025390625, 36.2735595703125, 39.96405029296875, 4.122356414794922, -2.906646728515625, -46.359130859375, -6.282257080078125, 120.01168823242188, 38.2537841796875, 9.808952331542969, 33.110137939453125, 14.295745849609375, 100.65362548828125, -8.938568115234375, 5.724239349365234, 0.0, 58.65147399902344, -18.3919677734375, 48.001068115234375, 95.4122314453125, 85.99951171875, 88.8563232421875, -14.302215576171875, -6.251739501953125, 14.753173828125, 60.149688720703125, 38.9989013671875, -108.46035766601562, 14.0855712890625, 107.48506164550781, 94.7783203125, 9.605926513671875, 97.09347534179688, 62.7947998046875, 11.547637939453125, 22.806396484375, 8.076469421386719, 83.73974609375, 21.662445068359375, 97.58544921875, 73.79924011230469, -23.632186889648438, -8.54925537109375, 102.07131958007812, 106.71466064453125, -40.086212158203125, -38.859771728515625, 0.82855224609375, 54.2501220703125, -7.519432067871094, -14.6165771484375, 14.3114013671875, -17.858489990234375, 34.39311218261719, 19.3037109375, 114.182861328125, 12.836135864257812, 0.0, 62.476219177246094, 71.88529968261719, -69.01094055175781, 4.224582672119141, 25.50110626220703, 64.08464813232422, 55.1900634765625, 0.5701694488525391, 52.781982421875, 77.08653259277344, -8.90765380859375, 61.82550811767578, -8.22857666015625, 32.59587097167969, 36.4478759765625, 7.917236328125, 51.8638916015625, -2.9178466796875, -94.38491821289062, 75.51734161376953, 75.4891357421875, 4.04010009765625, 1.77056884765625, -20.915069580078125, 5.515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000216.npy"} {"epoch": 0.4523560209424084, "step": 217, "batch_size": 128, "mean": 37.165504455566406, "std": 49.88722610473633, "min": -82.72177124023438, "p10": -20.502569580078124, "median": 35.783782958984375, "p90": 99.99547576904295, "max": 189.1502685546875, "pos_frac": 0.78125, "sample": [9.588043212890625, -2.154296875, -5.92840576171875, 74.32992553710938, 98.80009460449219, -71.24063110351562, 3.832733154296875, 91.14263916015625, -13.23624038696289, 39.44317626953125, 94.62071228027344, -7.30364990234375, -1.0083465576171875, 28.555938720703125, -13.546142578125, 91.95147705078125, 71.57029724121094, 79.81011962890625, 67.80633544921875, 37.50750732421875, 56.287567138671875, 38.89312744140625, 4.031108856201172, 68.710693359375, 157.5545654296875, 61.011749267578125, 80.67367553710938, 70.6039047241211, 82.65289306640625, 88.85590362548828, 33.31854248046875, 60.037109375, -15.648849487304688, 120.13397216796875, 42.59083557128906, 24.0318603515625, 189.1502685546875, 74.13323974609375, 85.15103149414062, 39.126220703125, -82.72177124023438, 86.7069091796875, 35.982666015625, 139.31005859375, -69.58659362792969, 45.235565185546875, 11.186660766601562, -25.43671417236328, 38.1473388671875, 20.3165283203125, 3.8692169189453125, 3.4704666137695312, 3.1788330078125, 38.55499267578125, 41.36479187011719, -47.724456787109375, 1.016204833984375, 81.39002990722656, 55.027565002441406, -15.283935546875, -7.936943054199219, 44.053802490234375, -5.4115447998046875, 114.02169799804688, 3.636077880859375, 51.22929382324219, 16.770883560180664, 93.12754821777344, 66.54644775390625, 11.986251831054688, 64.81385040283203, 75.35994720458984, 16.410560607910156, -3.20941162109375, 95.32489013671875, -39.12237548828125, -25.492141723632812, -29.124656677246094, 57.80657958984375, 148.53878784179688, 11.100997924804688, 102.78469848632812, 35.58489990234375, 3.8452415466308594, -78.29367065429688, 114.8695068359375, 103.09829711914062, 103.0328369140625, 13.22198486328125, -22.420989990234375, 14.00897216796875, 81.03263854980469, -11.146209716796875, 74.43154907226562, 4.6599578857421875, 74.64981079101562, 10.694252014160156, 48.941497802734375, 14.119956970214844, 77.31112670898438, -34.47711181640625, 0.0, 16.50091552734375, 26.099777221679688, -27.270904541015625, 121.19134521484375, 67.64697265625, -12.3323974609375, 51.2470703125, 112.25413513183594, 54.239227294921875, 25.68414306640625, 26.6552734375, 126.55352783203125, 76.53253173828125, 11.927658081054688, 6.569496154785156, 20.130218505859375, -29.9560546875, 4.7624053955078125, 57.926177978515625, 41.077762603759766, 46.34210205078125, 21.66204833984375, 22.263763427734375, -19.680389404296875, 16.3868408203125, 2.5461692810058594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000217.npy"} {"epoch": 0.4544502617801047, "step": 218, "batch_size": 128, "mean": 29.91568374633789, "std": 43.48790740966797, "min": -77.0053939819336, "p10": -21.08526039123535, "median": 25.625152587890625, "p90": 86.26727600097657, "max": 133.59857177734375, "pos_frac": 0.7578125, "sample": [46.7835693359375, 12.415573120117188, -4.595062255859375, 84.49609375, -6.3294677734375, 3.423553466796875, -14.76220703125, 100.8734130859375, 36.89902114868164, 19.56359100341797, -54.541259765625, 12.917205810546875, 61.84352111816406, 61.389984130859375, 44.92926025390625, -70.32550048828125, -21.00909423828125, 75.028076171875, 63.53471374511719, 64.92486572265625, -21.262981414794922, 54.34996032714844, 40.885711669921875, 25.62835693359375, 34.074371337890625, 26.721664428710938, 9.4761962890625, -14.24493408203125, -19.842987060546875, 60.00769805908203, 64.35108184814453, 111.90081787109375, 5.877227783203125, 83.26981353759766, -23.42694091796875, 10.299560546875, 73.17404174804688, -28.099700927734375, 70.12583923339844, 60.607177734375, 133.59857177734375, 42.71044921875, 19.626296997070312, 0.11195564270019531, -15.220321655273438, 86.73548889160156, 21.036113739013672, -55.399139404296875, -33.480316162109375, 75.76310729980469, 35.50587463378906, -26.23853302001953, 73.213623046875, 55.71898651123047, 86.32638549804688, 1.8554248809814453, 86.241943359375, 40.1944580078125, 84.17095947265625, -20.89385986328125, 25.6219482421875, -8.531051635742188, -16.600425720214844, 10.746490478515625, 0.0, 60.02606201171875, 5.3035125732421875, 0.57269287109375, -59.50457763671875, 78.6768798828125, 14.846298217773438, 49.36448669433594, -45.3985481262207, 77.06411743164062, 32.2679443359375, 86.90130615234375, -2.0229110717773438, 12.126327514648438, 24.6568603515625, 13.900215148925781, 101.61599731445312, 84.83251953125, 105.9788818359375, 4.5914306640625, -3.815582275390625, 18.79833984375, -6.3765411376953125, 7.613189697265625, 103.791748046875, 30.53863525390625, 15.918556213378906, 54.56011962890625, 19.05291748046875, -59.422607421875, 78.26238250732422, 57.44200897216797, 83.34587097167969, 97.247802734375, -40.0252685546875, 43.68914794921875, 4.6236572265625, 74.56097412109375, -7.800563812255859, 3.51373291015625, 48.832183837890625, -10.36199951171875, -77.0053939819336, 53.67143249511719, 8.327392578125, 63.04559326171875, 10.519500732421875, 17.0645751953125, 40.961822509765625, 59.193634033203125, -19.485076904296875, 20.38018798828125, 106.07220458984375, 1.295745849609375, 25.400901794433594, -12.4173583984375, 86.5545654296875, 52.8505859375, 61.924224853515625, 35.6600341796875, 42.76861572265625, 91.00726318359375, 47.21923828125, 10.263092041015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000218.npy"} {"epoch": 0.45654450261780105, "step": 219, "batch_size": 128, "mean": 29.59929084777832, "std": 46.269309997558594, "min": -108.97320556640625, "p10": -20.57692222595215, "median": 22.534347534179688, "p90": 93.3864532470703, "max": 123.48696899414062, "pos_frac": 0.7734375, "sample": [-108.97320556640625, 20.596282958984375, 37.1466064453125, 10.832992553710938, 81.81300354003906, 45.14171600341797, -15.256622314453125, 52.554931640625, 79.18890380859375, 29.62615966796875, 70.46939086914062, -1.287872314453125, 39.3470458984375, -23.287567138671875, 6.491081237792969, 35.70256042480469, 57.3316650390625, -11.747772216796875, 17.756423950195312, 69.26315307617188, 46.3470458984375, 7.458564758300781, 44.40203857421875, 113.6485595703125, -3.3170852661132812, 22.382354736328125, 66.20285034179688, 9.884780883789062, 67.87554931640625, -26.378036499023438, -18.36412811279297, -20.720874786376953, -65.65426635742188, 40.92182922363281, 9.594329833984375, 88.33919525146484, -7.0303955078125, 23.553218841552734, 18.50677490234375, 84.28492736816406, 102.59408569335938, 122.7301025390625, 7.562450408935547, 50.598785400390625, 13.150894165039062, -47.065643310546875, 8.364105224609375, 69.01885986328125, 8.85296630859375, 111.26431274414062, 0.0, 24.668685913085938, 19.264114379882812, 25.793258666992188, -43.64410400390625, 97.906982421875, -20.515228271484375, 94.19781494140625, 102.72531127929688, 14.039047241210938, 58.21012878417969, 19.046630859375, 18.882293701171875, 5.5269775390625, 2.200629234313965, -8.60089111328125, 3.359893798828125, 15.418670654296875, 7.4340667724609375, 67.36900329589844, -9.881195068359375, 20.785400390625, 14.834426879882812, 108.18368530273438, 18.21502685546875, -0.9238128662109375, -75.68391418457031, 7.14300537109375, 55.4793701171875, 93.03872680664062, 1.240438461303711, 87.13302612304688, 22.68634033203125, 26.32769775390625, -11.172815322875977, -34.401214599609375, 107.18585205078125, 18.840721130371094, 105.68486022949219, -3.677093505859375, 86.429443359375, 64.8687744140625, 27.971954345703125, 65.103515625, 71.84796905517578, 84.141357421875, 47.66987609863281, 47.2315673828125, 31.77093505859375, 83.98013305664062, -12.955947875976562, 81.40992736816406, 56.31072998046875, -40.66978454589844, 83.11155700683594, 57.667236328125, 53.938323974609375, 3.3064727783203125, 23.51251220703125, 27.970001220703125, 42.2935791015625, -0.1240997314453125, 41.987701416015625, -105.344482421875, 121.99093627929688, 98.21075439453125, 18.942489624023438, -34.5556640625, 14.4774169921875, 11.615158081054688, 23.9202880859375, 14.94775390625, 46.0408935546875, 11.620429992675781, 123.48696899414062, -7.2228546142578125, 2.553314208984375, -76.7867431640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000219.npy"} {"epoch": 0.4586387434554974, "step": 220, "batch_size": 128, "mean": 31.120248794555664, "std": 49.218135833740234, "min": -119.32574462890625, "p10": -32.695465087890625, "median": 31.051971435546875, "p90": 97.3589096069336, "max": 132.9591827392578, "pos_frac": 0.71875, "sample": [103.08779907226562, 104.22247314453125, 60.43266296386719, 57.311920166015625, 22.522140502929688, 64.684814453125, 4.40277099609375, 73.05889892578125, 27.44384765625, 62.81060791015625, -8.847213745117188, 1.4647216796875, 70.91460418701172, -50.4498291015625, 10.446044921875, 30.746337890625, 46.8197021484375, 96.79949951171875, -24.23468017578125, 13.8572998046875, 49.558929443359375, -6.15948486328125, 23.81407928466797, 2.678363800048828, 0.3301277160644531, 132.9591827392578, -3.16693115234375, 97.09059143066406, -56.30517578125, 19.734359741210938, 51.72601318359375, 3.41827392578125, -7.2664794921875, -32.995758056640625, 28.71014404296875, 9.248046875, -21.2132568359375, 51.0587158203125, 1.4775543212890625, 74.255615234375, 54.96764373779297, 27.8216552734375, 83.58059692382812, 41.02410888671875, 104.63359069824219, 53.766685485839844, -33.21064758300781, -119.32574462890625, 61.93450927734375, 41.9962158203125, 57.543212890625, 74.50151062011719, -8.643798828125, 105.047119140625, -21.011322021484375, 30.6187744140625, -9.881134033203125, -21.931114196777344, 6.216156005859375, 42.36824035644531, -33.01573181152344, 52.98479461669922, 37.67266845703125, 71.78457641601562, 13.422233581542969, -13.702751159667969, -8.970504760742188, 50.76605224609375, 1.7913360595703125, 80.37881469726562, 95.73039245605469, 106.47676086425781, -4.1807403564453125, 23.012496948242188, 70.15234375, 65.8012924194336, -21.6585693359375, 29.87628173828125, 53.822105407714844, -9.99169921875, -15.438339233398438, 14.512893676757812, 76.85200500488281, 45.788055419921875, -5.7285003662109375, 110.00582122802734, -68.38055419921875, 12.385406494140625, 118.31045532226562, 116.52342224121094, -34.7969970703125, 31.35760498046875, -82.3746337890625, -0.8289680480957031, 50.852447509765625, 112.2034912109375, 97.9849853515625, 16.8914794921875, 67.84280395507812, -31.838180541992188, 72.83889770507812, 40.54515075683594, -36.839874267578125, -32.58038330078125, 1.1062240600585938, -12.728317260742188, 73.71505737304688, -28.849014282226562, 104.33465576171875, 96.05635070800781, 65.61558532714844, 33.60400390625, 48.87725830078125, -32.9639892578125, 115.251220703125, -34.2840576171875, 89.662841796875, 67.3099365234375, 12.746292114257812, 77.86526489257812, -89.16316223144531, 40.79669189453125, 54.091026306152344, 25.331377029418945, 77.8731689453125, -13.825904846191406, 72.0267333984375, 42.238525390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000220.npy"} {"epoch": 0.4607329842931937, "step": 221, "batch_size": 128, "mean": 24.50934600830078, "std": 46.832088470458984, "min": -104.96768188476562, "p10": -28.392285156249997, "median": 17.51431655883789, "p90": 87.32121124267577, "max": 118.83291625976562, "pos_frac": 0.7265625, "sample": [6.452850341796875, 49.67266845703125, -15.85711669921875, 20.609405517578125, 37.294952392578125, 28.274749755859375, 17.93585968017578, 42.185699462890625, -12.882186889648438, -12.590534210205078, 56.803382873535156, 50.708892822265625, 2.777557373046875, 54.6771240234375, -2.9144210815429688, 48.523468017578125, 10.763809204101562, -59.33009338378906, 8.78042221069336, -21.16104507446289, 62.9962158203125, 32.408477783203125, 118.83291625976562, -25.90209197998047, 27.513153076171875, 73.47119903564453, 92.89373779296875, 64.17669677734375, 84.20782470703125, -0.324005126953125, 13.47271728515625, 12.096282958984375, 108.96478271484375, -104.96768188476562, -36.4298095703125, 14.439361572265625, 37.29327392578125, 74.572265625, 10.90667724609375, 99.75927734375, -35.300575256347656, 6.92803955078125, 5.4532623291015625, 30.60064697265625, -63.76457214355469, -2.890106201171875, 81.64549255371094, -59.2293701171875, 72.18348693847656, 32.477325439453125, -0.3704414367675781, 75.79917907714844, 17.0927734375, -15.32012939453125, -8.297760009765625, 65.1890869140625, 43.6146240234375, -60.047882080078125, 0.1075439453125, 44.68031311035156, 8.9622802734375, 11.599533081054688, 54.995452880859375, 7.733699798583984, 38.86883544921875, 1.0780391693115234, 51.308502197265625, 78.3984375, -2.161945343017578, 75.62548828125, 40.581687927246094, -9.954025268554688, 60.259033203125, 57.29046630859375, 94.20018005371094, 58.25439453125, 99.011962890625, 22.430084228515625, -51.47486114501953, -67.9483642578125, 75.8438720703125, 95.468505859375, -13.26544189453125, 11.068733215332031, -7.87774658203125, 86.10679626464844, 12.396484375, -20.196319580078125, 29.9200439453125, 64.76406860351562, 104.30343627929688, 94.49795532226562, 1.2632904052734375, 7.191633224487305, -29.427459716796875, 6.099220275878906, -17.748458862304688, 111.09512329101562, -92.6199951171875, 40.509063720703125, 4.601165771484375, -70.9608154296875, -2.0964813232421875, 104.88529968261719, 14.274383544921875, 10.14959716796875, 25.600677490234375, 90.15484619140625, 83.89794921875, 10.195976257324219, 47.72901916503906, 49.2294921875, -27.948638916015625, -18.79959487915039, 7.253517150878906, 7.6851348876953125, 85.80024719238281, 29.876220703125, 75.7838134765625, -8.19677734375, 28.20037078857422, 75.42483520507812, -20.44000244140625, 90.27926635742188, 23.734588623046875, 10.426122665405273, -90.33053588867188, 10.67718505859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000221.npy"} {"epoch": 0.46282722513089003, "step": 222, "batch_size": 128, "mean": 32.49159622192383, "std": 44.75177001953125, "min": -141.17498779296875, "p10": -17.754545593261717, "median": 22.826396942138672, "p90": 94.63554992675782, "max": 120.8824462890625, "pos_frac": 0.7578125, "sample": [54.251129150390625, 14.187255859375, 83.34748840332031, 68.79244995117188, -26.20831298828125, -27.245712280273438, 75.28073120117188, 37.11960983276367, 18.276641845703125, 7.914482116699219, -9.127655029296875, 22.64697265625, 82.1937255859375, 48.3607177734375, 24.262908935546875, 45.05323028564453, 67.76580047607422, 110.95672607421875, 59.188507080078125, 39.56990051269531, 8.788604736328125, 4.4485626220703125, -22.533203125, -51.891326904296875, 54.82147216796875, 119.31451416015625, -11.58709716796875, -5.20758056640625, -67.45411682128906, 9.370655059814453, 9.026657104492188, 43.39678955078125, 120.8824462890625, 2.9098892211914062, 17.34805679321289, 9.78085708618164, 19.71832275390625, 58.384918212890625, 72.15208435058594, 38.854827880859375, 5.71942138671875, -35.65025329589844, 58.826507568359375, 2.876800537109375, 48.789764404296875, 7.840610504150391, -17.989669799804688, 58.91493225097656, -141.17498779296875, 42.81597900390625, 91.78166198730469, 64.54460144042969, -0.01210784912109375, 93.26956176757812, -10.297138214111328, 119.0498046875, 20.68486785888672, 94.55792236328125, 1.8508491516113281, 3.40966796875, 22.050048828125, 23.005821228027344, 25.23345947265625, 95.05328369140625, -3.575897216796875, 17.2215576171875, 79.74869537353516, 118.82589721679688, 60.79083251953125, 58.06315612792969, 90.9390869140625, 77.63468933105469, 0.4180908203125, 78.14505004882812, -19.158889770507812, 3.751953125, 17.46014404296875, -0.035640716552734375, 94.81668090820312, -9.1793212890625, 6.742256164550781, 29.645254135131836, -6.184823989868164, 46.57298278808594, -17.653778076171875, -0.860931396484375, 96.85688781738281, -22.481399536132812, -6.7733001708984375, 62.870574951171875, 79.56608581542969, 81.21148681640625, 33.11937713623047, 45.74076843261719, 6.03190803527832, -20.70849609375, 4.691780090332031, 43.09834289550781, 24.84014892578125, 101.73541259765625, 109.9095458984375, -9.001441955566406, -2.93646240234375, 37.27156066894531, 99.381591796875, -4.494476318359375, 85.5474853515625, 43.63194274902344, 14.9091796875, 89.568359375, 21.06439208984375, 41.774993896484375, -27.14056396484375, 9.579864501953125, 0.6728057861328125, -5.802978515625, -14.22650146484375, 76.14962768554688, -18.6456298828125, 71.1646728515625, 53.086578369140625, 9.49542236328125, 107.45101928710938, 76.23011779785156, 18.337554931640625, -0.36577415466308594, 13.819961547851562, 106.30172729492188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000222.npy"} {"epoch": 0.4649214659685864, "step": 223, "batch_size": 128, "mean": 35.54475402832031, "std": 49.6721076965332, "min": -88.95553588867188, "p10": -31.27990264892578, "median": 33.86731719970703, "p90": 97.14186096191406, "max": 139.52734375, "pos_frac": 0.765625, "sample": [10.982307434082031, -11.595352172851562, 7.139274597167969, 79.64218139648438, 10.717414855957031, 23.524219512939453, 1.623260498046875, -14.88876724243164, 14.360090255737305, 92.64140319824219, 25.612548828125, 90.21586608886719, 63.83918762207031, 5.875274658203125, 25.97314453125, 129.74566650390625, 97.61322021484375, -88.95553588867188, 76.23666381835938, -36.24192810058594, 28.291717529296875, 111.78753662109375, 1.111083984375, 49.32257080078125, 29.128173828125, -33.43992614746094, 16.927398681640625, 16.988189697265625, 105.62936401367188, -14.6807861328125, 53.14019775390625, 58.190399169921875, 72.37176513671875, 91.381103515625, 68.80520629882812, 87.22518920898438, 131.09344482421875, -79.85284423828125, 60.05766296386719, -72.32038879394531, 14.490585327148438, 46.84332275390625, 9.216644287109375, 65.88343811035156, 95.12619018554688, 82.42074584960938, 105.20199584960938, 19.52801513671875, 109.20721435546875, 41.085723876953125, -1.0899658203125, 20.23492431640625, 112.1776123046875, 120.1986083984375, 75.69813537597656, 58.25640869140625, 2.0660400390625, 96.93984985351562, 20.288963317871094, 46.29949951171875, 41.757537841796875, 21.117042541503906, 46.66461181640625, 139.52734375, 38.39323425292969, 46.56150817871094, 24.8267822265625, 69.9554443359375, 70.85739135742188, -4.0523834228515625, -13.343093872070312, 39.53254699707031, 10.88726806640625, 90.30767822265625, -40.65021514892578, 26.12019157409668, -24.517234802246094, -31.442886352539062, 76.68341064453125, -13.925956726074219, 30.62737274169922, -6.700103759765625, 1.308197021484375, 28.837848663330078, 83.75257873535156, 59.8807373046875, -40.784820556640625, -13.75433349609375, 66.46856689453125, -12.498039245605469, -78.38410186767578, 25.267044067382812, 23.230911254882812, 90.849609375, 79.35443115234375, 114.10791015625, 74.56253051757812, 70.61862182617188, 62.60919952392578, 11.04925537109375, -1.6958961486816406, 59.077911376953125, 46.98919677734375, 67.00997924804688, 53.8033447265625, -10.47808837890625, 96.36825561523438, 126.39678955078125, -1.8950920104980469, 12.501266479492188, 100.62960815429688, -52.044334411621094, -10.6912841796875, 6.750587463378906, 92.183837890625, 53.110870361328125, -61.067256927490234, -1.246368408203125, 60.27195739746094, 6.74041748046875, -60.69789123535156, 81.74134826660156, 49.64833068847656, 42.4364013671875, 37.107261657714844, 18.770355224609375, -31.210052490234375, -41.737884521484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000223.npy"} {"epoch": 0.46701570680628274, "step": 224, "batch_size": 128, "mean": 34.9310417175293, "std": 46.84828567504883, "min": -91.28790283203125, "p10": -23.440516662597656, "median": 30.8892822265625, "p90": 95.81590270996094, "max": 136.11862182617188, "pos_frac": 0.796875, "sample": [115.87503051757812, 136.11862182617188, 29.505767822265625, 53.20509338378906, 2.4095706939697266, 5.4939422607421875, -59.465179443359375, 37.433563232421875, 76.62506103515625, 13.669136047363281, 6.854095458984375, 69.35894775390625, -0.9503021240234375, 66.17080688476562, 26.8544921875, 30.538787841796875, 10.04583740234375, 77.4556884765625, 12.093101501464844, -49.771087646484375, 68.5931396484375, -23.861557006835938, 109.06951904296875, 108.77536010742188, 45.8673095703125, 63.31361389160156, 71.04258728027344, 19.98004150390625, 16.983078002929688, 72.36920166015625, 107.86355590820312, 82.04635620117188, -91.28790283203125, 65.97300720214844, 33.99974060058594, 21.629425048828125, 33.25688934326172, 67.80286407470703, 66.50189208984375, 5.55841064453125, 11.3546142578125, -15.557403564453125, 79.48365783691406, 61.93402099609375, 51.062042236328125, 31.38243865966797, 51.377227783203125, -38.76020812988281, 96.40283203125, -0.8897457122802734, 66.76129150390625, 50.30841064453125, 0.0, 102.04299926757812, 81.75627136230469, 13.943912506103516, 8.875762939453125, -63.65623474121094, 4.5557861328125, 30.501953125, 52.06358337402344, 18.293731689453125, 1.5301666259765625, 77.68365478515625, -1.0000457763671875, 96.6964111328125, -19.61016845703125, -30.306640625, 44.941741943359375, -34.50306701660156, 17.406097412109375, 118.66168212890625, 20.568557739257812, 95.56436157226562, 17.9464111328125, -8.941650390625, -10.147064208984375, 61.5133056640625, 13.315841674804688, 103.70785522460938, -20.741958618164062, -13.424980163574219, -67.25361633300781, 10.650505065917969, 15.527580261230469, 89.06365966796875, 44.946075439453125, -36.49957275390625, 93.46377563476562, -79.46146392822266, 87.03697204589844, 43.505462646484375, 57.93428039550781, 93.11947631835938, 28.458251953125, 50.74674987792969, 92.79766845703125, 50.177825927734375, 38.4188232421875, 5.823036193847656, -28.615570068359375, 26.895751953125, 6.164276123046875, 17.2745361328125, 21.334091186523438, 49.84996032714844, 133.9068603515625, -23.26007080078125, 103.790771484375, 66.65921020507812, 16.874664306640625, -36.27374267578125, 10.087301254272461, 4.458587646484375, -6.9501953125, 31.239776611328125, 77.01644897460938, 6.594871520996094, 55.84375, -15.54052734375, 7.432674407958984, 90.85324096679688, 66.9530029296875, 66.27371215820312, 91.39549255371094, 70.18880462646484, 1.50140380859375, 111.6739501953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000224.npy"} {"epoch": 0.46910994764397906, "step": 225, "batch_size": 128, "mean": 24.599689483642578, "std": 55.67556381225586, "min": -103.30731201171875, "p10": -48.14630126953124, "median": 22.47669219970703, "p90": 104.73589477539062, "max": 134.58682250976562, "pos_frac": 0.6328125, "sample": [106.85302734375, -26.74462890625, 40.309326171875, 102.83721923828125, 92.66838073730469, 6.8570709228515625, -67.9041748046875, -70.18569946289062, 70.82977294921875, 102.57080078125, 26.74676513671875, 77.25250244140625, 24.7884521484375, -10.95697021484375, 110.53555297851562, 28.252197265625, 119.21011352539062, 76.60678100585938, 3.650165557861328, 27.947921752929688, -10.35723876953125, -12.831741333007812, 20.876243591308594, -33.42341613769531, 94.12860107421875, 10.22479248046875, 108.33767700195312, -13.152297973632812, 1.0313720703125, -58.55804443359375, -4.7381591796875, -76.75833129882812, 76.85018157958984, 81.03094482421875, 8.66796875, 11.043128967285156, 73.91629028320312, -2.411670684814453, -1.048736572265625, -44.85484313964844, 51.6470947265625, 112.52474975585938, -44.723731994628906, -60.06645202636719, -80.94735717773438, -60.987640380859375, 48.04057312011719, 40.872802734375, -7.928371429443359, 35.22271728515625, -46.21211242675781, 21.37213134765625, 66.13724517822266, 81.40444946289062, -10.50872802734375, 16.417266845703125, -38.70147705078125, 32.7640380859375, 96.65023803710938, 71.34922790527344, 78.14480590820312, 110.50416564941406, 121.5616455078125, -22.76922607421875, 27.80059814453125, -10.481399536132812, -61.62451171875, 60.292205810546875, 46.42826843261719, 124.28489685058594, -0.432373046875, 103.748779296875, 26.889404296875, 72.31897735595703, -7.524383544921875, 44.60783386230469, 92.0773696899414, 115.52032470703125, -22.674560546875, 79.38214111328125, -30.027145385742188, -56.82452392578125, -38.01520538330078, 10.597248077392578, -76.54646301269531, -8.395416259765625, 23.989431381225586, 0.5698585510253906, 42.41957092285156, -103.30731201171875, -0.312408447265625, -26.39453125, -15.74713134765625, -12.8756103515625, 32.66302490234375, -28.43865966796875, 0.7954559326171875, 29.37328338623047, 134.58682250976562, 59.14732360839844, 4.512092590332031, 66.3784408569336, 107.72512817382812, 102.21588134765625, -26.23626708984375, 22.368179321289062, -62.245338439941406, 28.586929321289062, 12.5433349609375, 124.36767578125, -14.538482666015625, 67.10269165039062, 76.02578735351562, 120.38020324707031, 5.91412353515625, 59.25823974609375, 31.585750579833984, -18.23931884765625, 51.207725524902344, -52.65940856933594, 22.585205078125, 1.9002418518066406, -10.054008483886719, -2.793701171875, -41.716522216796875, 25.802505493164062, 33.2181396484375, 103.82855224609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000225.npy"} {"epoch": 0.4712041884816754, "step": 226, "batch_size": 128, "mean": 24.098691940307617, "std": 48.852195739746094, "min": -87.01376342773438, "p10": -45.319305419921875, "median": 14.937576293945312, "p90": 88.67530212402343, "max": 121.22177124023438, "pos_frac": 0.6875, "sample": [20.622314453125, 121.22177124023438, -11.1807861328125, 59.001312255859375, 88.65277099609375, 103.48153686523438, 7.348052978515625, -1.3425006866455078, 56.72682189941406, 82.90170288085938, 102.37966918945312, 50.91657638549805, -0.20892333984375, 104.9197998046875, -2.52337646484375, 12.860370635986328, 38.46343994140625, -62.091796875, 84.23681640625, 94.47274780273438, 9.283836364746094, 0.13909149169921875, 78.71206665039062, 60.09632110595703, 3.9928436279296875, 21.376846313476562, -10.938699722290039, -75.12338256835938, 76.62620544433594, 56.217041015625, 0.36911582946777344, 84.94380187988281, 56.03070068359375, 88.1495361328125, -4.3323822021484375, 14.4820556640625, 88.47602844238281, -53.9163818359375, -0.640625, 65.95087432861328, -71.71420288085938, 40.05885314941406, -48.72718811035156, -0.6376743316650391, -10.391876220703125, 43.619232177734375, 13.212860107421875, -16.134124755859375, -87.01376342773438, 64.8592529296875, 31.654571533203125, -46.57904052734375, 87.27175903320312, 1.9944534301757812, 9.472671508789062, -66.292236328125, 3.0418624877929688, 23.1875, 1.9097518920898438, -0.35224151611328125, -7.855560302734375, 14.795944213867188, 34.716712951660156, -82.692138671875, 52.955413818359375, -7.61749267578125, 12.14996337890625, 94.64236450195312, 13.971071243286133, 9.4237060546875, -28.8399658203125, 7.219635009765625, 35.9420166015625, -25.651809692382812, 43.862796783447266, 2.40966796875, 10.820526123046875, 69.242431640625, 5.6179351806640625, 102.22567749023438, -65.01246643066406, 88.26195526123047, 107.59048461914062, -68.67339324951172, 103.10272216796875, 0.13623046875, 14.243057250976562, 2.2962570190429688, 28.8695068359375, 82.02099609375, 63.53631591796875, 40.09698486328125, 83.14300537109375, 61.29229736328125, 88.72787475585938, 15.079208374023438, -16.726318359375, -1.821075439453125, 73.29341888427734, 30.471908569335938, 107.45329284667969, 16.79718017578125, 103.10089111328125, -0.7587890625, 7.123046875, 21.366439819335938, 72.34056091308594, 33.270233154296875, 69.80818176269531, -20.383193969726562, 40.14927673339844, -44.7794189453125, -0.7409210205078125, 64.88665771484375, 41.66144561767578, -4.6955718994140625, 60.076507568359375, -14.330680847167969, -24.388671875, 109.4700927734375, 31.4183349609375, -56.32971954345703, -44.24705505371094, 15.469524383544922, -50.00624084472656, -26.790481567382812, 17.3372802734375, -10.1070556640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000226.npy"} {"epoch": 0.4732984293193717, "step": 227, "batch_size": 128, "mean": 35.07008361816406, "std": 50.52546691894531, "min": -112.30352783203125, "p10": -27.50895080566406, "median": 33.521915435791016, "p90": 101.36470489501953, "max": 140.00167846679688, "pos_frac": 0.765625, "sample": [-75.05987548828125, 36.932098388671875, 19.295013427734375, 27.06964111328125, 66.92013549804688, 43.20586395263672, 107.083740234375, 20.109771728515625, 5.122001647949219, 33.536781311035156, 101.03045654296875, 58.47016906738281, 14.016433715820312, -3.9130859375, 13.876335144042969, 37.44352722167969, 0.0, 106.0126953125, -101.03570556640625, -60.823062896728516, 65.4105224609375, -59.0447998046875, 80.60838317871094, 97.78569030761719, 129.9674835205078, 140.00167846679688, 95.56549072265625, -61.362335205078125, 88.57968139648438, -16.956451416015625, -7.675102233886719, 87.24201965332031, 13.55096435546875, 13.783599853515625, 33.507049560546875, -11.24102783203125, 30.797752380371094, 56.843605041503906, 53.412994384765625, 109.83041381835938, 113.84652709960938, 82.094970703125, 14.416473388671875, 68.052001953125, 34.152557373046875, -24.339584350585938, 31.74072265625, 51.021881103515625, -8.728874206542969, 29.398895263671875, 101.2149658203125, 26.618736267089844, -10.363983154296875, -27.270172119140625, 40.602783203125, 123.2662353515625, 5.86663818359375, 69.42474365234375, 101.93888854980469, 28.35684585571289, 105.10087585449219, -45.71208190917969, -28.06610107421875, 77.70297241210938, -112.30352783203125, 25.14410400390625, 27.908660888671875, 116.69219970703125, -39.851715087890625, 34.699859619140625, 69.2813720703125, 36.70097351074219, 2.8383350372314453, 72.53010559082031, 100.8563232421875, 45.127532958984375, 48.0994873046875, 73.86642456054688, -39.340576171875, 79.494873046875, 112.94549560546875, 5.873291015625, 21.37230682373047, 100.15394592285156, 23.27752685546875, 34.71684265136719, 40.564239501953125, 29.592193603515625, 27.584869384765625, 47.29339599609375, -30.17431640625, 57.97948455810547, 17.351547241210938, -54.99467468261719, 88.25262451171875, 39.783172607421875, 77.64408874511719, 9.876708984375, -13.375541687011719, 93.85041046142578, 0.27593994140625, 113.9337158203125, -2.409639358520508, 101.71409606933594, -51.095375061035156, 89.17581176757812, -2.4394874572753906, -0.798431396484375, 4.0046844482421875, 3.68609619140625, 51.57623291015625, 12.911697387695312, 88.00662231445312, 90.54707336425781, 14.421600341796875, 31.75921630859375, 34.7738037109375, 43.452667236328125, 85.56582641601562, -6.73870849609375, -17.299835205078125, -18.357666015625, 22.899078369140625, 58.66992950439453, -20.59918212890625, 5.982542037963867, 57.454246520996094, 76.3485107421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000227.npy"} {"epoch": 0.47539267015706804, "step": 228, "batch_size": 128, "mean": 35.06806945800781, "std": 49.31890869140625, "min": -65.85211181640625, "p10": -22.08679656982422, "median": 27.771257400512695, "p90": 97.4923843383789, "max": 128.11309814453125, "pos_frac": 0.71875, "sample": [-58.6568603515625, -22.92102813720703, 63.8502197265625, 86.8028564453125, 67.66610717773438, -4.617748260498047, 117.47042846679688, 28.9473876953125, 63.0587158203125, 95.97738647460938, 9.183517456054688, -18.988929748535156, -10.112564086914062, 106.32667541503906, 46.746612548828125, 57.680572509765625, -8.0443115234375, 77.96573638916016, -39.06640625, 18.796371459960938, 88.48313903808594, 20.029747009277344, -65.85211181640625, -17.41094970703125, 16.073776245117188, 87.03549194335938, 23.3360595703125, 0.09929656982421875, -17.556732177734375, 36.35919952392578, 13.076400756835938, 22.3741455078125, -44.03483581542969, 89.7542724609375, 54.363128662109375, 88.99931335449219, 56.837066650390625, 62.49253845214844, -19.67681884765625, -9.377227783203125, 95.63333129882812, -6.752044677734375, -16.802536010742188, -22.18218994140625, 70.97637939453125, 10.30499267578125, -4.472198486328125, 16.196762084960938, 40.717254638671875, 86.46511840820312, 35.42694091796875, 115.507568359375, 59.183929443359375, 18.715465545654297, 100.81576538085938, 36.619171142578125, 79.5914306640625, 70.74896240234375, 84.24981689453125, 109.52734375, -0.406982421875, -62.53971862792969, -18.8812255859375, -0.42996978759765625, 77.19276428222656, 7.919677734375, 0.0, 22.789459228515625, 27.44049072265625, 15.421791076660156, 12.752098083496094, 125.43048095703125, 17.28270721435547, -4.7874755859375, -33.962982177734375, 85.07319641113281, -18.6781005859375, 94.02386474609375, -12.567291259765625, 98.21414184570312, -14.865631103515625, -58.43035125732422, 18.05145263671875, 55.08599853515625, 93.48117065429688, 107.39767456054688, 2.25140380859375, 0.051025390625, 113.09745788574219, 40.95245361328125, -22.045913696289062, 95.0523681640625, 16.870086669921875, 84.4599609375, 34.195068359375, 63.51250457763672, -42.635528564453125, 4.5219268798828125, 17.029659271240234, 97.18305969238281, 93.730712890625, -65.26235961914062, -13.751319885253906, 37.760284423828125, 73.04072570800781, 60.04986572265625, 86.20718383789062, -52.8819580078125, -16.94158935546875, -3.388275146484375, 13.910614013671875, 91.4569091796875, 10.845077514648438, 75.6107177734375, 59.6409912109375, 113.33132934570312, 115.2545166015625, 59.433197021484375, 10.918403625488281, -38.813873291015625, 70.77438354492188, 78.0770263671875, 109.91114807128906, 59.32209777832031, 128.11309814453125, 10.665740966796875, 13.152824401855469, 28.10202407836914], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000228.npy"} {"epoch": 0.4774869109947644, "step": 229, "batch_size": 128, "mean": 46.418540954589844, "std": 51.57893371582031, "min": -84.90034484863281, "p10": -17.799031066894525, "median": 52.74453353881836, "p90": 110.12562561035155, "max": 161.30081176757812, "pos_frac": 0.8125, "sample": [75.57537841796875, -21.00103759765625, 81.43563842773438, 69.37469482421875, -7.881256103515625, -82.40571594238281, 57.27192687988281, 88.31146240234375, 38.26434326171875, -56.377685546875, 138.0838623046875, 115.46627807617188, 70.19596862792969, 2.098896026611328, 121.9522705078125, 92.6123046875, 9.50006103515625, 9.03582763671875, 18.28948974609375, 0.0751953125, 33.98713684082031, 21.070205688476562, 94.0411376953125, -6.886077880859375, 19.66376495361328, -48.968505859375, 10.166229248046875, 8.855178833007812, 106.15121459960938, 68.28961181640625, 49.75517272949219, 88.18685913085938, 113.94915771484375, 80.02545166015625, 23.709091186523438, 120.04736328125, -0.9127197265625, 17.2640380859375, 135.30511474609375, -84.90034484863281, 11.93438720703125, -57.532379150390625, 55.53901672363281, 84.8414306640625, 57.426517486572266, 126.31280517578125, 161.30081176757812, 97.54975128173828, 82.67805480957031, 86.30886840820312, 67.56573486328125, -1.0558948516845703, 55.841278076171875, 8.407470703125, 80.05489349365234, 101.17268371582031, 88.2491455078125, 11.489471435546875, 9.950119018554688, -21.133682250976562, 21.18926239013672, -69.87931823730469, 94.45767211914062, 93.32513427734375, -9.590194702148438, -9.483779907226562, 90.3564453125, 15.290245056152344, 72.09727478027344, 0.555023193359375, 3.378448486328125, 52.49859619140625, 0.27314186096191406, -1.706451416015625, 93.34126281738281, 57.764434814453125, 46.440650939941406, 85.93785858154297, 109.3033447265625, 42.90679931640625, 102.80000305175781, 70.20588684082031, 33.078369140625, 87.37483215332031, 16.710147857666016, 10.2830810546875, -33.31910705566406, 93.94770812988281, 122.22259521484375, 52.99047088623047, 125.89208984375, 88.74197387695312, 11.100543975830078, 65.1285400390625, 14.96417236328125, 97.0267333984375, 41.1031494140625, 105.57150268554688, 35.595458984375, 50.10194396972656, 75.49191284179688, -6.911712646484375, 42.416046142578125, 112.04428100585938, -1.187286376953125, 8.045989990234375, 43.521392822265625, -16.426742553710938, 10.444942474365234, 107.1607666015625, 60.327064514160156, -4.0305633544921875, 50.475128173828125, 46.454376220703125, 72.04721069335938, 68.77011108398438, 88.22174072265625, -22.372161865234375, 77.544677734375, -37.1202392578125, 120.42080688476562, 58.549591064453125, 67.08010864257812, 53.53204345703125, 64.2996826171875, -58.084476470947266, 128.9091796875, -23.60401153564453], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000229.npy"} {"epoch": 0.47958115183246075, "step": 230, "batch_size": 128, "mean": 35.20207595825195, "std": 52.890159606933594, "min": -157.99972534179688, "p10": -19.552645874023437, "median": 27.971099853515625, "p90": 104.88194885253905, "max": 144.11764526367188, "pos_frac": 0.75, "sample": [106.47174072265625, 72.25421142578125, 32.42291259765625, 85.85763549804688, 143.35372924804688, 62.46868896484375, 109.70075988769531, -4.22979736328125, 104.63082885742188, 43.34175109863281, 16.269744873046875, 28.360931396484375, 12.976238250732422, -68.00589752197266, -157.99972534179688, 85.99551391601562, 88.5526123046875, -1.7274971008300781, -3.8848876953125, -70.808349609375, 4.91192626953125, 12.473682403564453, 113.02706909179688, -22.951553344726562, 0.0, 94.67633056640625, -2.901885986328125, 16.109539031982422, 61.1937255859375, 88.20014953613281, 82.65853118896484, 94.98257446289062, 3.69744873046875, 39.94633483886719, 0.3159942626953125, -64.40362548828125, 95.0111083984375, 81.90812683105469, 66.09716796875, -17.842361450195312, 0.05755615234375, -34.41204833984375, 12.117340087890625, 30.5306396484375, 15.660545349121094, 93.391845703125, 62.078147888183594, 32.208892822265625, 84.71282958984375, -19.63568115234375, -2.503366470336914, 45.8077392578125, -33.960693359375, 63.245208740234375, 90.40597534179688, -38.28684997558594, 41.77409362792969, 12.049789428710938, 92.44082641601562, 25.3370361328125, -60.288475036621094, 76.87112426757812, 118.150146484375, 75.71029663085938, 15.376922607421875, 5.3583984375, -13.166595458984375, 86.24362182617188, -9.26800537109375, 88.156982421875, 44.75140380859375, 6.829315185546875, -19.517059326171875, -12.455093383789062, 36.29083251953125, 21.4368896484375, 8.25640869140625, 24.221027374267578, 119.82220458984375, -4.619354248046875, -1.9441814422607422, 58.92713928222656, 24.48760986328125, -16.185022354125977, 22.337066650390625, -115.52938842773438, -1.4384384155273438, -20.06121063232422, 117.85760498046875, 75.46728515625, 128.02249145507812, 43.82040786743164, 10.4500732421875, 8.759689331054688, 105.4678955078125, 102.26153564453125, 1.2795028686523438, 29.00537109375, 88.53487396240234, 83.03861999511719, 98.66336059570312, 0.3343982696533203, 70.47077178955078, 3.00921630859375, -21.57061767578125, 82.33251953125, -2.33172607421875, 115.50341796875, 68.93153381347656, 5.523597717285156, 73.216552734375, 109.11595153808594, 61.96087646484375, 2.1862659454345703, 62.75367736816406, 33.31122589111328, -0.17445945739746094, 117.60594177246094, 144.11764526367188, 3.2808380126953125, 0.16024017333984375, 44.012451171875, 3.0178298950195312, 9.872039794921875, 27.581268310546875, 38.43730163574219, -0.1888427734375, -4.1185455322265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000230.npy"} {"epoch": 0.4816753926701571, "step": 231, "batch_size": 128, "mean": 28.295738220214844, "std": 47.70357894897461, "min": -94.17134094238281, "p10": -28.854757690429683, "median": 25.800933837890625, "p90": 84.04267959594726, "max": 133.73095703125, "pos_frac": 0.75, "sample": [58.69960021972656, 20.902618408203125, -33.49530029296875, -37.909454345703125, 83.44060516357422, 77.59622192382812, 44.649932861328125, -5.6413726806640625, 68.81512451171875, -94.17134094238281, 94.44282531738281, -32.7264404296875, 53.817352294921875, 72.4119873046875, -5.4118194580078125, 77.58169555664062, 10.519729614257812, 8.59963607788086, 36.59971618652344, -6.713470458984375, -25.905059814453125, -53.41349792480469, -10.602508544921875, 42.30517578125, 11.918716430664062, 40.541839599609375, 28.710113525390625, 83.98290252685547, 1.53662109375, -51.012176513671875, 32.578857421875, 83.23611450195312, 75.04290771484375, 3.1025543212890625, 62.74627685546875, -10.367576599121094, 114.56439208984375, 45.43585205078125, 10.706375122070312, 39.5567626953125, 3.3927001953125, 37.82476806640625, -0.176422119140625, 1.4751739501953125, 100.2410888671875, 81.62457275390625, -7.853515625, 38.50135803222656, 13.93060302734375, 31.810089111328125, 80.67327880859375, -24.85516357421875, 13.431289672851562, 13.7498779296875, 6.36431884765625, 16.7857666015625, 83.04135131835938, -17.82061767578125, 25.017913818359375, 127.09536743164062, -43.0382080078125, 4.455783843994141, -5.0112762451171875, 45.964630126953125, 2.283550262451172, 76.33139038085938, 35.75836181640625, 26.583953857421875, 19.22161865234375, 53.047576904296875, 49.83042907714844, 52.16595458984375, 122.21759033203125, -27.195465087890625, -84.41795349121094, -1.694915771484375, 29.20063018798828, -10.856719970703125, 78.46575927734375, 47.196197509765625, 17.9072265625, 14.94207763671875, 34.166900634765625, 41.92431640625, 1.4183349609375, 18.642471313476562, 8.35186767578125, 98.038818359375, 51.608123779296875, 10.878219604492188, 69.3524169921875, 65.950927734375, 77.86656188964844, 57.45648193359375, -65.31626892089844, 71.19927978515625, 18.629390716552734, 65.41293334960938, 133.73095703125, 0.6903514862060547, -80.5115966796875, 23.1888427734375, 36.52137756347656, 125.35369873046875, 75.45211791992188, -5.96392822265625, -0.06517791748046875, 57.09136962890625, 9.664016723632812, 32.69237518310547, -6.6697845458984375, 7.684900283813477, 34.93943786621094, -78.8740234375, 61.4300537109375, 123.29841613769531, 10.69647216796875, 100.2623291015625, 9.507522583007812, 56.588104248046875, -16.740127563476562, 3.38934326171875, 89.78636169433594, -60.521240234375, 84.18215942382812, 0.0, -75.05137634277344, 100.23617553710938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000231.npy"} {"epoch": 0.4837696335078534, "step": 232, "batch_size": 128, "mean": 32.44244384765625, "std": 51.018619537353516, "min": -82.89576721191406, "p10": -24.906243896484376, "median": 22.693073272705078, "p90": 96.74694061279297, "max": 156.3505859375, "pos_frac": 0.6875, "sample": [-24.122802734375, 93.98953247070312, -2.1139907836914062, 103.62994384765625, 29.3056640625, 63.00531005859375, 84.89483642578125, 42.448089599609375, -0.11474609375, -8.576797485351562, 70.45967102050781, 96.42431640625, 12.681610107421875, -23.513381958007812, -24.497573852539062, 21.461669921875, 7.772026062011719, 13.761871337890625, 19.116455078125, 21.300750732421875, 27.14453125, 9.475975036621094, -75.22626495361328, 64.17257690429688, 90.64639282226562, 76.97125244140625, 30.70677947998047, 19.71868896484375, -0.9151268005371094, 119.91986083984375, 54.472930908203125, -5.810005187988281, -38.092529296875, 55.106842041015625, 71.98715209960938, 89.56256103515625, -50.350250244140625, 35.179107666015625, 16.0521240234375, 96.62936401367188, 60.00761413574219, -48.606285095214844, -0.42780303955078125, 23.19098663330078, 9.130104064941406, 2.389120101928711, 81.52987670898438, 11.995452880859375, 90.49104309082031, 0.9955215454101562, 26.904327392578125, -24.854736328125, 85.41279602050781, 20.4659423828125, 2.3907928466796875, 88.24588012695312, 22.195159912109375, 89.66524505615234, -1.3527069091796875, 79.77790832519531, 11.56744384765625, 126.5296630859375, 97.02128601074219, -25.02642822265625, 83.64115905761719, 88.86534118652344, -5.32476806640625, 24.989501953125, -5.557861328125, -10.55682373046875, 13.329002380371094, 106.50469970703125, 8.962982177734375, -33.701080322265625, 120.82272338867188, 114.91810607910156, -2.56121826171875, -44.452392578125, 39.874969482421875, 65.86685180664062, 1.011453628540039, 2.7042388916015625, -17.8934326171875, 77.37518310546875, 91.71905517578125, -17.65780258178711, -6.092414855957031, 69.83406066894531, -5.19598388671875, -26.65789794921875, 81.81596374511719, 58.1656494140625, 0.0, 109.07077026367188, 64.77232360839844, 9.926048278808594, 61.553009033203125, 26.961151123046875, 86.16950988769531, 61.1297607421875, 83.02523040771484, 20.038848876953125, 120.29095458984375, -73.83778381347656, 78.95710754394531, 156.3505859375, 10.90234375, -20.51922607421875, 87.42355346679688, 58.822052001953125, -5.46142578125, 109.89825439453125, -73.3119888305664, -13.480987548828125, -16.836898803710938, -48.2545166015625, 37.340370178222656, -23.27682113647461, 38.338714599609375, -51.420013427734375, 99.20669555664062, -4.022796630859375, -82.89576721191406, 109.66854858398438, 71.7646484375, 69.78238677978516, 27.87640380859375, -22.36993408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000232.npy"} {"epoch": 0.48586387434554973, "step": 233, "batch_size": 128, "mean": 34.73678207397461, "std": 50.65744400024414, "min": -86.33265686035156, "p10": -24.37328872680664, "median": 24.70745849609375, "p90": 99.73003387451172, "max": 149.235595703125, "pos_frac": 0.796875, "sample": [-46.57035827636719, 87.94481658935547, 0.7047119140625, -3.25286865234375, 13.471771240234375, 98.55160522460938, 37.4453125, 43.92083740234375, 3.461923599243164, 13.460777282714844, 41.56134033203125, 24.396106719970703, 103.58380126953125, 47.51287841796875, 11.485710144042969, 0.01873779296875, 57.9364013671875, 98.560302734375, 26.847015380859375, 141.26229858398438, 92.07595825195312, 85.42992401123047, 98.90164184570312, 85.27842712402344, 60.184654235839844, 100.68669128417969, 92.07745361328125, 12.398635864257812, 82.06906127929688, -24.025733947753906, 33.2918701171875, -6.925323486328125, 85.42939758300781, 2.1063385009765625, 103.04896545410156, 30.910552978515625, 22.405670166015625, -68.60791015625, 69.40660095214844, 117.70286560058594, -21.143455505371094, -3.713735580444336, 5.0207061767578125, 23.0987548828125, -19.664215087890625, 133.283203125, 66.44293212890625, 17.479854583740234, 29.365570068359375, 10.999618530273438, 120.685546875, 125.35919189453125, 20.4932861328125, -35.33830261230469, 37.32928466796875, -9.720184326171875, 134.368408203125, 26.14935302734375, 90.42829895019531, 8.6702880859375, 43.48345947265625, 90.361328125, -43.689056396484375, 34.623531341552734, 8.146041870117188, -22.860748291015625, -25.184249877929688, 52.20040512084961, -59.6082763671875, -67.77774047851562, 91.276611328125, 99.221435546875, 109.10406494140625, 61.279144287109375, 133.777587890625, 50.328033447265625, 19.009307861328125, 12.717681884765625, 11.94552230834961, 4.491424560546875, 4.055213928222656, 5.6737060546875, 12.8302001953125, 66.8388671875, -27.731231689453125, 13.594911575317383, 149.235595703125, 51.1978759765625, 3.721527099609375, 38.061309814453125, 20.75955581665039, -31.773056030273438, -78.5501708984375, 122.39035034179688, 23.320411682128906, -21.18389892578125, 1.0308837890625, 82.37484741210938, -3.2882080078125, 93.73634338378906, -13.53570556640625, 7.7557220458984375, -11.936447143554688, -86.33265686035156, 74.32614135742188, 74.17083740234375, 24.53167724609375, 72.36322784423828, 24.88323974609375, -65.27284240722656, 2.5582351684570312, 79.87796020507812, 51.387603759765625, -0.6279754638671875, 39.377166748046875, 25.637969970703125, -35.035987854003906, 99.32003784179688, 46.623207092285156, 57.266082763671875, 21.360441207885742, 4.192024230957031, 4.64599609375, 56.70159912109375, 72.0068359375, 16.03240966796875, 18.535079956054688, 20.6126708984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000233.npy"} {"epoch": 0.48795811518324606, "step": 234, "batch_size": 128, "mean": 35.18009948730469, "std": 59.196083068847656, "min": -98.07229614257812, "p10": -49.15576019287109, "median": 31.980422973632812, "p90": 108.82376403808593, "max": 172.41156005859375, "pos_frac": 0.7578125, "sample": [-71.99588012695312, -90.3841552734375, -49.226287841796875, -0.010990142822265625, 41.49267578125, 104.46673583984375, 107.5037841796875, 44.7203369140625, -25.953155517578125, 127.33981323242188, 94.51893615722656, -98.07229614257812, 89.27719116210938, 84.75639343261719, -21.8138427734375, 59.6934814453125, 33.772010803222656, 29.6470947265625, 61.62715148925781, 114.52169799804688, 80.09794616699219, 54.33979797363281, 62.958892822265625, 122.68472290039062, 14.308013916015625, 7.3777313232421875, 83.04928588867188, 79.26223754882812, 84.48387145996094, 71.8507080078125, 19.345855712890625, 49.432373046875, 93.79904174804688, 125.8411865234375, 83.6044921875, 22.661415100097656, 30.00823974609375, 7.683135986328125, 0.792572021484375, 47.98822021484375, 59.87530517578125, 17.583953857421875, 5.70709228515625, -49.12553405761719, 23.521636962890625, -17.608963012695312, 77.34716796875, 78.44994354248047, -3.0271530151367188, 102.38443756103516, 6.7744140625, 7.2843017578125, 123.666015625, 55.566951751708984, -5.82781982421875, 146.358642578125, 86.09150695800781, 55.23682403564453, 91.64231872558594, 66.28717041015625, 7.474246978759766, 5.758161544799805, 114.65853881835938, 10.309234619140625, 7.562688827514648, -2.216583251953125, -31.095962524414062, 0.0, 29.591552734375, -12.437065124511719, 14.831634521484375, 34.657012939453125, -87.16354370117188, 10.849639892578125, 94.24667358398438, 131.45611572265625, 96.31637573242188, 26.867477416992188, -71.73916625976562, 52.055572509765625, -0.8061962127685547, 86.38882446289062, 105.37211608886719, -17.69512939453125, 63.845947265625, 30.18883514404297, 77.03594970703125, 21.27752685546875, -89.33648681640625, 123.9730224609375, -69.51382446289062, 129.40972900390625, 19.534820556640625, 22.291900634765625, 84.53971099853516, 1.73345947265625, 80.48562622070312, 34.42388916015625, 0.7314529418945312, 8.7930908203125, 83.66598510742188, 0.0, 37.114253997802734, 46.32453918457031, 172.41156005859375, 123.47113037109375, -93.59831237792969, -34.78558349609375, 111.90371704101562, 63.619232177734375, 68.10704803466797, 1.48394775390625, -92.186279296875, 19.06512451171875, 57.639404296875, 98.06077575683594, -32.69732666015625, -25.6632080078125, -59.521148681640625, 1.79095458984375, 47.77734375, 3.9941253662109375, -62.6708984375, 11.359519958496094, -80.02621459960938, 97.04693603515625, -13.133636474609375, 102.2042236328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000234.npy"} {"epoch": 0.4900523560209424, "step": 235, "batch_size": 128, "mean": 32.91307067871094, "std": 55.42068862915039, "min": -113.188232421875, "p10": -30.819060516357418, "median": 25.95016098022461, "p90": 98.61420440673828, "max": 143.598388671875, "pos_frac": 0.6953125, "sample": [21.741302490234375, 7.552906036376953, -5.21563720703125, 1.092620849609375, 6.880516052246094, -95.31759643554688, -41.60293960571289, -3.3426513671875, 85.26554870605469, 88.1199951171875, 10.074623107910156, 134.58212280273438, 73.91357421875, 48.057281494140625, 68.69805908203125, 21.03060531616211, 85.04534912109375, 56.8802490234375, -2.6666908264160156, -6.77301025390625, 72.31619262695312, 23.07073974609375, 37.044647216796875, 65.0845947265625, 124.54156494140625, 121.16754150390625, 68.23065185546875, -16.462326049804688, 81.455078125, -6.760627746582031, 104.0115966796875, -0.8853759765625, 65.15237426757812, 94.00553131103516, 33.90301513671875, 88.02674865722656, 121.683837890625, 93.2003173828125, 7.2351531982421875, 131.9832763671875, -12.412406921386719, 56.954864501953125, -29.93671417236328, -36.66229248046875, 62.96543884277344, 124.920654296875, -96.6275634765625, -24.30697250366211, -76.3199462890625, 33.82615661621094, 59.61199951171875, -15.459564208984375, -2.0128040313720703, -19.870849609375, 34.138916015625, 85.24545288085938, -15.812576293945312, 89.56526184082031, -8.772796630859375, -57.867919921875, 10.167106628417969, 25.6097412109375, 98.19766235351562, -19.120880126953125, 132.18020629882812, 54.92511749267578, -9.984130859375, -86.35993957519531, 6.887458801269531, 40.03448486328125, -26.869903564453125, 51.171234130859375, -34.87312316894531, 85.83805847167969, 6.9912567138671875, 11.616241455078125, 7.05291748046875, 109.79917907714844, 7.60772705078125, 46.531402587890625, 52.73565673828125, 24.1070556640625, -112.44210815429688, 28.9654541015625, 24.95697021484375, -9.024421691894531, 80.30049896240234, 92.86648559570312, 68.5555419921875, 3.2337799072265625, 137.57723999023438, 82.29202270507812, 64.53620910644531, 143.598388671875, 99.58613586425781, 91.12649536132812, 22.7105712890625, 22.08428955078125, 92.06626892089844, -7.489587783813477, -3.7826995849609375, 97.27188110351562, -53.18677520751953, 55.33135986328125, -0.0704345703125, 79.0432357788086, -4.7748565673828125, 43.307037353515625, 84.57275390625, -5.41510009765625, -3.22930908203125, 19.130218505859375, 3.87774658203125, -32.87786865234375, 26.29058074951172, 13.42367172241211, 94.65464782714844, -113.188232421875, -47.33611297607422, 3.2082748413085938, 75.95315551757812, 74.65448760986328, 22.009185791015625, 104.22589111328125, 68.18528747558594, 52.29613494873047, 26.42425537109375, -0.05702781677246094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000235.npy"} {"epoch": 0.49214659685863876, "step": 236, "batch_size": 128, "mean": 36.752315521240234, "std": 52.96216583251953, "min": -110.46636962890625, "p10": -28.124702453613278, "median": 32.79732131958008, "p90": 101.1372085571289, "max": 149.05035400390625, "pos_frac": 0.7890625, "sample": [54.982513427734375, 24.130767822265625, -69.96435546875, 56.06437683105469, 65.68756103515625, 78.89303588867188, 9.77996826171875, 97.03030395507812, -14.012176513671875, -36.48423767089844, -77.9349365234375, 33.043434143066406, -55.76548767089844, 57.941131591796875, 65.3958740234375, 82.88894653320312, -0.6827411651611328, 85.86448669433594, 70.33345031738281, 19.87298583984375, 24.101898193359375, 75.57433319091797, 37.986572265625, -46.16658020019531, 42.66365051269531, 85.66973876953125, 83.81394958496094, 4.559061050415039, 59.818145751953125, 96.05229187011719, 19.690597534179688, 30.69970703125, 42.674598693847656, 21.232139587402344, 11.464881896972656, 19.061691284179688, 10.1346435546875, 125.7197265625, 93.32994079589844, 1.7412261962890625, 11.354049682617188, 93.99454498291016, 98.27383422851562, 34.91038513183594, 117.42523193359375, 1.3126411437988281, -49.3201904296875, 6.43829345703125, 19.870223999023438, 0.0, -4.44378662109375, 23.802017211914062, 85.26162719726562, 89.80484771728516, 9.908788681030273, -26.2398681640625, -91.19754028320312, 95.05801391601562, 11.09765625, -31.1573486328125, 72.92127990722656, 107.17889404296875, 49.67120361328125, 58.843299865722656, 1.4898681640625, -26.824996948242188, 77.76104736328125, -110.46636962890625, 62.00958251953125, 3.2819366455078125, -17.7508544921875, 149.05035400390625, 43.20243835449219, 131.857421875, 2.6678619384765625, -6.017311096191406, -45.3701171875, -2.047027587890625, 101.73307800292969, 66.5782470703125, 38.70097351074219, 62.3062744140625, -39.240966796875, 77.90817260742188, 86.71975708007812, 32.55120849609375, 104.39701843261719, 123.36669921875, -5.078643798828125, 73.89793395996094, 55.96435546875, 100.03579711914062, 72.6912841796875, 18.964462280273438, 3.8850860595703125, 100.8818359375, 98.23541259765625, 57.81440734863281, 86.60892486572266, 107.88784790039062, 93.34159851074219, 29.8355712890625, -23.666961669921875, 12.014450073242188, 0.0, 40.59954833984375, 6.753814697265625, 29.00469970703125, -93.43849182128906, 96.55390930175781, 11.589706420898438, 0.4556999206542969, -61.78315734863281, -20.070022583007812, 45.7177734375, 93.77385711669922, 110.73780822753906, 12.398239135742188, 7.149269104003906, -19.48047637939453, 112.03157043457031, 83.44461059570312, 110.23751831054688, 18.20013427734375, 2.143157958984375, 106.89236450195312, 7.041133880615234, 31.510910034179688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000236.npy"} {"epoch": 0.4942408376963351, "step": 237, "batch_size": 128, "mean": 31.670757293701172, "std": 53.110530853271484, "min": -116.90667724609375, "p10": -20.20259094238281, "median": 23.244537353515625, "p90": 105.24160079956054, "max": 177.26141357421875, "pos_frac": 0.7421875, "sample": [38.19708251953125, 12.771896362304688, 33.4393196105957, 109.81533813476562, -11.885009765625, 30.319976806640625, 67.62841033935547, 79.14055633544922, 38.496368408203125, 25.178512573242188, 0.61376953125, 42.80641174316406, 126.84967041015625, -55.66191864013672, -15.693031311035156, 0.5797615051269531, 21.22161865234375, 45.3707275390625, 177.26141357421875, 82.03235626220703, 71.07005310058594, 30.592636108398438, 101.79962158203125, 63.29985809326172, 11.427764892578125, 3.3525390625, 70.83604431152344, 0.0, 68.52378845214844, 75.17068481445312, -2.0404281616210938, 14.52301025390625, -55.23353576660156, 22.02210235595703, -16.564620971679688, 36.863525390625, 4.9131011962890625, 1.1266021728515625, -116.90667724609375, -18.199363708496094, -3.2580184936523438, -1.141815185546875, 8.316719055175781, 86.84222412109375, 7.447052001953125, 27.291259765625, 103.21090698242188, 28.0010986328125, 39.20928955078125, 2.5154991149902344, 4.143096923828125, 132.89315795898438, 59.588226318359375, 22.26976776123047, 11.193511962890625, 77.16497802734375, 100.07742309570312, 55.268798828125, 38.955718994140625, 20.23126220703125, 115.85888671875, 35.32337188720703, 24.352020263671875, 20.755035400390625, 56.41050720214844, -8.011993408203125, 113.48782348632812, 93.5396728515625, -47.988739013671875, -59.73358154296875, 0.59466552734375, 120.723876953125, -70.34443664550781, 8.78009033203125, 99.25619506835938, -3.1330947875976562, -58.743621826171875, -3.8256072998046875, 137.50099182128906, -19.37298583984375, 22.696319580078125, 25.654129028320312, 109.28045654296875, 26.384918212890625, 108.1910400390625, -59.01446533203125, -4.8008270263671875, 3.207733154296875, 107.32366180419922, 24.03021240234375, 10.849807739257812, 136.1441650390625, -4.162895202636719, -0.502288818359375, -37.83624267578125, 121.12652587890625, -55.39422607421875, 2.930755615234375, 97.33075714111328, 71.79779052734375, 16.02313232421875, 9.389625549316406, 87.07366943359375, -1.92047119140625, 7.615013122558594, 3.1575088500976562, 88.720947265625, -22.138336181640625, -14.678009033203125, -10.65618896484375, 2.3841552734375, 104.34928894042969, -28.236297607421875, 11.470924377441406, 49.07300567626953, 36.458648681640625, -114.71359252929688, 57.53857421875, 84.8880615234375, 101.68682861328125, 88.36431884765625, -17.326904296875, 23.792755126953125, 80.30471801757812, -1.5539474487304688, 32.031524658203125, 20.569305419921875, 64.24237060546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000237.npy"} {"epoch": 0.4963350785340314, "step": 238, "batch_size": 128, "mean": 35.530426025390625, "std": 55.328060150146484, "min": -81.6469497680664, "p10": -23.749249267578115, "median": 25.095138549804688, "p90": 110.46835250854491, "max": 155.8702392578125, "pos_frac": 0.6875, "sample": [-35.05348205566406, 1.1395187377929688, 5.9694976806640625, 11.058464050292969, -49.25812530517578, 139.8375244140625, 92.23233032226562, 45.1591796875, 83.45355224609375, -47.580810546875, -1.438720703125, 71.27137756347656, -5.9548492431640625, 14.405471801757812, -7.755226135253906, -66.60835266113281, 112.623291015625, 98.09880065917969, -2.014284133911133, 39.935203552246094, 98.54643249511719, 72.24630737304688, 37.35821533203125, 99.64344024658203, 135.73866271972656, 16.30744171142578, -13.69677734375, 3.959259033203125, 25.65118408203125, -78.51802062988281, -77.35055541992188, -81.6469497680664, 122.73297119140625, -45.03619384765625, -6.124267578125, 119.11306762695312, 31.61163330078125, 91.43081665039062, 4.2060089111328125, 84.27593994140625, 133.62548828125, 6.12109375, 29.089248657226562, 142.1990966796875, 73.88619995117188, 105.42626953125, 63.6092529296875, 55.4654541015625, -5.913177490234375, 34.645294189453125, 106.84617614746094, 144.83843994140625, 66.29977416992188, -20.083290100097656, 19.341064453125, 13.7828369140625, -11.86187744140625, 4.556264877319336, -10.344192504882812, 69.0489730834961, -0.850128173828125, -5.230926513671875, -21.1092529296875, -80.946044921875, 92.55438232421875, 60.690643310546875, 61.5936279296875, 99.1567153930664, 79.20162963867188, 101.10501861572266, -45.68159484863281, -29.90924072265625, 31.277626037597656, -4.6605072021484375, -2.4042434692382812, 65.14218139648438, 113.687255859375, -13.207061767578125, 76.240478515625, 91.975341796875, 74.45707702636719, 113.71795654296875, -14.722042083740234, -8.335456848144531, 3.81207275390625, 24.89501953125, 86.10877227783203, -5.066293716430664, 57.92238235473633, 35.145751953125, 0.7981948852539062, -10.3729248046875, -16.886245727539062, 15.96221923828125, 155.8702392578125, -9.4154052734375, 32.347442626953125, 58.15583801269531, 19.312049865722656, 10.567127227783203, 25.090042114257812, 118.45620727539062, 109.54480743408203, 29.3206787109375, 58.96449279785156, 3.585906982421875, -15.937973022460938, 101.64292907714844, 24.288848876953125, -38.488555908203125, 25.100234985351562, 104.52279663085938, 17.187721252441406, 12.685821533203125, 93.24713134765625, -12.92144775390625, 5.65087890625, 97.69512939453125, -6.4328155517578125, -9.140350341796875, 102.64768981933594, 130.02877807617188, 32.902801513671875, -30.334686279296875, 34.28129577636719, 6.680908203125, 78.41969299316406, -4.24176025390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000238.npy"} {"epoch": 0.49842931937172774, "step": 239, "batch_size": 128, "mean": 30.310585021972656, "std": 57.58256149291992, "min": -134.1534423828125, "p10": -38.02674560546874, "median": 18.03122329711914, "p90": 115.84369735717773, "max": 155.2279052734375, "pos_frac": 0.6953125, "sample": [-56.57593536376953, -47.883544921875, 129.01828002929688, 83.43075561523438, -14.78021240234375, -0.6461181640625, -134.1534423828125, 68.6605453491211, 124.59942626953125, 0.0001678466796875, 61.91156005859375, -121.9417724609375, -87.69058227539062, -4.250732421875, 83.33797454833984, 52.770263671875, 91.905029296875, 115.61534881591797, 34.57048034667969, -8.9549560546875, 0.4594879150390625, 103.747802734375, 1.8681373596191406, 24.691024780273438, -2.6825485229492188, 78.54884338378906, 0.48288917541503906, 88.4243392944336, 18.459075927734375, -16.048858642578125, -42.81207275390625, 89.23373413085938, 79.51715850830078, -1.367645263671875, 135.47750854492188, -12.98614501953125, 24.571548461914062, 90.94683837890625, 155.2279052734375, 108.5753173828125, 80.70623016357422, -6.29638671875, 24.366851806640625, 0.242401123046875, 138.65838623046875, -11.180152893066406, -13.76025390625, 66.5411376953125, -70.91928100585938, 87.79296875, -60.74713134765625, 67.94178771972656, 17.603370666503906, 52.798309326171875, 36.804542541503906, -14.6378173828125, 45.876190185546875, 6.526206970214844, 12.342330932617188, -35.97589111328125, 17.546966552734375, 108.85537719726562, 7.809925079345703, 70.48495483398438, 22.478675842285156, 27.6395263671875, 92.89936065673828, 2.5185546875, 6.94140625, 125.60162353515625, 1.286569595336914, 12.227777481079102, 11.084457397460938, 39.0843505859375, 96.51107788085938, -9.3106689453125, 88.51153564453125, 14.988471984863281, -57.10868835449219, 10.471923828125, -1.504425048828125, 27.038536071777344, 131.2886505126953, -18.501708984375, -3.881927490234375, 124.1727294921875, -54.31402587890625, 28.78827667236328, 10.560653686523438, -47.4307861328125, 103.55020141601562, -3.768207550048828, -11.279891967773438, 100.51541900634766, 4.2611541748046875, 68.849609375, 11.198432922363281, -1.211181640625, 129.73599243164062, 34.388153076171875, 0.4851493835449219, 128.59576416015625, 16.113479614257812, -33.522003173828125, 49.7490234375, -2.7485885620117188, -29.520950317382812, 28.405120849609375, 38.814910888671875, 121.62139892578125, 4.4700164794921875, 38.8946533203125, 51.995361328125, 132.47528076171875, 116.37651062011719, 0.0, 25.589447021484375, 1.0422534942626953, 100.71319580078125, -0.296051025390625, -44.513580322265625, 76.83421325683594, 19.88995361328125, 21.98260498046875, 34.32427978515625, -73.43984985351562, -2.396514892578125, 16.880233764648438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000239.npy"} {"epoch": 0.5005235602094241, "step": 240, "batch_size": 128, "mean": 38.885643005371094, "std": 52.24014663696289, "min": -101.66752624511719, "p10": -18.745858764648435, "median": 32.26154327392578, "p90": 107.986083984375, "max": 152.25238037109375, "pos_frac": 0.765625, "sample": [7.635955810546875, 104.60525512695312, 62.33026123046875, 97.05120849609375, -17.212814331054688, 3.9624252319335938, 62.62086486816406, 38.16693115234375, 0.0, 126.13389587402344, 32.75341796875, -26.2650146484375, -77.05146789550781, 0.542694091796875, -20.795486450195312, 93.06586456298828, 50.38089370727539, 70.19066619873047, 79.9586410522461, 38.155029296875, 18.351165771484375, 0.723480224609375, 74.31546020507812, 95.96966552734375, 18.154998779296875, 95.70437622070312, 73.54804992675781, -44.315704345703125, 110.92892456054688, 105.50657653808594, 38.748382568359375, 5.514892578125, 123.51690673828125, 69.7498779296875, 11.218994140625, -12.564453125, 118.07150268554688, 80.23992919921875, 88.0863037109375, 10.830848693847656, 25.27752685546875, 44.807373046875, 110.59159851074219, 0.41902923583984375, 73.7159423828125, 49.43243408203125, 85.62786865234375, 40.62202453613281, 82.75151062011719, 79.11117553710938, 110.08837890625, 57.069541931152344, 0.0, 65.54998016357422, -101.66752624511719, 25.212799072265625, -3.043212890625, -5.4874114990234375, -11.074310302734375, 0.103973388671875, 62.455322265625, 90.31672668457031, -6.675079345703125, -12.1175537109375, 107.74346923828125, 94.15455627441406, 45.244140625, 79.7094497680664, 99.73680114746094, 35.142974853515625, 125.0859375, -27.7198486328125, 0.0, 17.519378662109375, -17.867446899414062, -11.878753662109375, 136.32318115234375, 0.1349029541015625, 14.637565612792969, 28.406097412109375, -10.122169494628906, 35.882080078125, 82.17112731933594, 23.768997192382812, 74.44706726074219, 152.25238037109375, -35.28729248046875, 149.34341430664062, -92.69802856445312, 107.60482788085938, 6.011749267578125, 5.427703857421875, 18.008865356445312, 93.113037109375, 89.78103637695312, 39.85113525390625, 31.769668579101562, 77.48847961425781, 15.699079513549805, -1.8725452423095703, -15.698867797851562, 7.369386672973633, 2.6102237701416016, 17.425277709960938, 91.13761901855469, 35.142059326171875, 74.12673950195312, 108.55218505859375, 31.461822509765625, 9.238006591796875, 109.52488708496094, 23.964111328125, 28.399169921875, -65.62287139892578, 28.984390258789062, 3.7069969177246094, -44.3072509765625, 70.10203552246094, -32.912994384765625, 90.27775573730469, 0.0, -3.5672225952148438, 62.04383087158203, 24.583740234375, 21.36895751953125, -58.47108840942383, -26.16851806640625, 123.46124267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000240.npy"} {"epoch": 0.5026178010471204, "step": 241, "batch_size": 128, "mean": 29.411834716796875, "std": 57.10769271850586, "min": -105.41104888916016, "p10": -37.35388488769531, "median": 16.122947692871094, "p90": 105.84218292236328, "max": 163.41830444335938, "pos_frac": 0.703125, "sample": [0.4119873046875, 108.55926513671875, 78.76690673828125, 59.712066650390625, 37.84861755371094, -30.19049072265625, 22.335296630859375, 0.5185871124267578, 61.80572509765625, 90.36663818359375, 95.49343872070312, 0.0, -11.54754638671875, 106.0867919921875, 84.9190902709961, 135.0447998046875, 7.824249267578125, -3.4347686767578125, 16.88018798828125, 111.83612060546875, 42.76907730102539, 95.77550506591797, -12.5408935546875, -12.6048583984375, 89.55118560791016, 16.723190307617188, -39.73486328125, 17.75189208984375, 92.51303100585938, -11.3389892578125, 111.48380279541016, -36.333465576171875, 96.52664184570312, 54.14410400390625, 12.856857299804688, 5.4053955078125, -9.17279052734375, -7.3278350830078125, 5.207307815551758, 99.97555541992188, 10.455612182617188, 18.431381225585938, 86.04483795166016, 134.33319091796875, 12.31793212890625, 75.19378662109375, 54.891693115234375, -102.71591186523438, 12.56304931640625, 24.532424926757812, 20.93310546875, 2.3664913177490234, 2.3858299255371094, 84.68478393554688, -23.451004028320312, -8.746139526367188, 100.4820556640625, 7.68182373046875, 19.599700927734375, -60.93608093261719, 34.076202392578125, 80.50045776367188, 44.085906982421875, 94.76022338867188, -6.779876708984375, 28.7506103515625, -14.073211669921875, 49.56184387207031, 0.9039535522460938, 85.695068359375, -81.49866485595703, 6.4967803955078125, 121.23670959472656, 70.9041748046875, -18.9345703125, -47.002685546875, 114.74945068359375, 76.86749267578125, 3.6318359375, -1.41058349609375, 3.878326416015625, 2.9167098999023438, 131.8200225830078, 27.8101806640625, 32.32020568847656, 11.944808959960938, -57.123870849609375, 94.2493667602539, -8.607650756835938, 105.73735046386719, 5.978515625, -65.57156372070312, -1.6959075927734375, -0.0168609619140625, 97.49700927734375, 66.93244934082031, 3.1666622161865234, 4.7323150634765625, 114.13150024414062, -26.729522705078125, 78.48908996582031, -73.25030517578125, 21.48553466796875, 2.4962005615234375, 103.58563232421875, 97.5015869140625, -77.28567504882812, 6.455680847167969, 117.85934448242188, 56.19342803955078, -21.665283203125, -105.41104888916016, -29.015960693359375, -6.9422607421875, 84.24312591552734, -0.2572956085205078, 124.41403198242188, 49.82917785644531, -64.15577697753906, 67.37002563476562, 163.41830444335938, 3.211883544921875, 15.522705078125, -99.82913208007812, 9.9903564453125, -22.086090087890625, 66.59371948242188, -45.85260009765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000241.npy"} {"epoch": 0.5047120418848168, "step": 242, "batch_size": 128, "mean": 35.68629455566406, "std": 50.412628173828125, "min": -111.12823486328125, "p10": -20.80233917236328, "median": 36.36164855957031, "p90": 105.9078384399414, "max": 135.94418334960938, "pos_frac": 0.765625, "sample": [87.93864440917969, 10.274887084960938, 76.44989013671875, 9.603710174560547, 22.158905029296875, 38.86675262451172, 34.302581787109375, 69.90402221679688, -25.83343505859375, 10.918685913085938, 128.46408081054688, 83.861328125, 42.47435760498047, -13.88372802734375, -92.61547088623047, 52.20861053466797, 30.208206176757812, 95.45264434814453, -111.12823486328125, 44.56467056274414, 12.235164642333984, 122.55668640136719, 115.75848388671875, 38.43236541748047, 50.09283447265625, 126.36373901367188, 2.1079559326171875, 120.52655029296875, -2.6863555908203125, -0.11165237426757812, 66.75094604492188, 93.53652954101562, 0.6676025390625, 113.95199584960938, 37.36199951171875, 135.94418334960938, 115.78150939941406, 3.4634933471679688, -36.362815856933594, 1.5630645751953125, 85.91873168945312, -20.551361083984375, -2.8768310546875, 55.372711181640625, -21.387954711914062, -4.025430679321289, 31.4658203125, 71.32093811035156, 62.16305923461914, 3.8969802856445312, -10.452316284179688, -8.843090057373047, -8.458732604980469, 42.08274841308594, -7.998725891113281, -2.855466842651367, 96.89460754394531, 4.5908203125, -10.96868896484375, 88.99685668945312, 45.998077392578125, 0.24473953247070312, -25.0274658203125, 113.77719116210938, 6.0139312744140625, -3.5300140380859375, 134.257568359375, 60.717559814453125, 40.14442443847656, 20.9039306640625, 32.480438232421875, 56.070953369140625, -22.440353393554688, 15.545539855957031, 81.1612548828125, 33.53839874267578, 77.6032943725586, 20.483890533447266, 17.59014892578125, -59.914894104003906, 39.40101623535156, 78.68403625488281, 25.2529296875, 42.630523681640625, -59.509521484375, 15.141082763671875, 6.469062805175781, 41.99675750732422, 39.451019287109375, 83.56703186035156, 135.30599975585938, 6.357177734375, 42.49664306640625, 28.887615203857422, -3.02532958984375, 19.063201904296875, 66.45686340332031, 92.05792236328125, 68.00550079345703, 17.162353515625, 2.1984405517578125, 46.74658203125, -35.34184646606445, -16.446121215820312, 81.94219970703125, 46.841957092285156, 34.252166748046875, -99.38348388671875, 56.1212158203125, 42.49742126464844, 35.361297607421875, 7.515350341796875, 105.30149841308594, 118.04344177246094, 76.19027709960938, 76.20504760742188, 107.3226318359375, 43.667022705078125, -55.694740295410156, 83.57470703125, -6.869384765625, 43.18983459472656, 12.817276000976562, -44.56121826171875, 55.272300720214844, 89.48727416992188, -17.264129638671875, 90.978759765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000242.npy"} {"epoch": 0.506806282722513, "step": 243, "batch_size": 128, "mean": 42.169410705566406, "std": 51.12571334838867, "min": -103.80557250976562, "p10": -20.319784545898436, "median": 34.447242736816406, "p90": 110.59898223876952, "max": 154.0341796875, "pos_frac": 0.765625, "sample": [-48.5333251953125, -20.8021240234375, 30.825454711914062, 36.27378845214844, 97.62496948242188, 68.56924438476562, 27.87432861328125, 119.25422668457031, -57.658966064453125, 87.5804443359375, 25.66864013671875, 134.35589599609375, 18.75774383544922, -24.896446228027344, 52.609893798828125, -9.365829467773438, -20.113067626953125, 45.339874267578125, 122.6129150390625, -54.44416809082031, 12.903121948242188, 17.60980224609375, -25.723541259765625, 0.01129150390625, 58.360816955566406, 105.9583740234375, 154.0341796875, 103.5853271484375, -5.9315185546875, 41.4510498046875, 9.40097427368164, 87.74737548828125, 17.59143829345703, 62.09852600097656, 102.8492431640625, 7.52032470703125, 113.44148254394531, 15.627182006835938, 59.620452880859375, 114.19790649414062, -21.053909301757812, 91.94100952148438, 52.2330322265625, 57.159820556640625, 65.28788757324219, 89.36334228515625, 115.450439453125, -37.651611328125, 52.462249755859375, 60.53777313232422, 49.206787109375, -103.80557250976562, -1.6897506713867188, 37.810882568359375, 16.53765869140625, 4.1773681640625, 31.737327575683594, 16.35394287109375, 14.661453247070312, -0.3765869140625, 133.92440795898438, 78.36126708984375, 84.02101135253906, -13.108444213867188, -7.425689697265625, 85.86552429199219, 23.623382568359375, 94.77767944335938, 82.11231231689453, -3.099365234375, 27.106735229492188, -21.3902587890625, -12.982421875, 107.48921203613281, 79.40438842773438, 51.34498596191406, 1.916839599609375, 61.828826904296875, 29.324722290039062, 21.515960693359375, -8.427438735961914, 83.9985122680664, -25.212181091308594, 56.57441711425781, 7.931884765625, 0.0, 3.9210662841796875, 40.87760925292969, 90.5905532836914, 99.26181030273438, 48.964378356933594, 72.98126220703125, 76.185546875, -5.090351104736328, 104.71243286132812, 106.3521728515625, 115.6290283203125, 13.1287841796875, 3.2476119995117188, 32.620697021484375, 20.796875, -1.72027587890625, -11.909286499023438, 27.64068603515625, -17.9361572265625, 109.38076782226562, 124.6082763671875, 83.30540466308594, 3.3323440551757812, 28.769866943359375, 18.686481475830078, 10.714607238769531, 103.94559478759766, -13.206817626953125, -34.576560974121094, 145.988037109375, 17.82485580444336, 107.14569091796875, -10.31744384765625, 20.279434204101562, 55.48216247558594, 99.12738037109375, 138.57473754882812, 66.66683197021484, 128.09954833984375, 49.38616943359375, 51.323516845703125, -20.81805419921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000243.npy"} {"epoch": 0.5089005235602094, "step": 244, "batch_size": 128, "mean": 42.29181671142578, "std": 56.15134048461914, "min": -123.54232788085938, "p10": -18.520826721191398, "median": 37.93273162841797, "p90": 116.67251892089844, "max": 196.21011352539062, "pos_frac": 0.7578125, "sample": [-123.54232788085938, 35.663726806640625, 48.474822998046875, -0.873565673828125, 30.013717651367188, 12.737743377685547, 62.630615234375, -29.324661254882812, 134.56134033203125, 78.21644592285156, 96.19256591796875, 73.152587890625, 12.987380981445312, -6.433074951171875, 10.318565368652344, 19.425262451171875, 81.2000732421875, 82.79460144042969, 82.89164733886719, 82.52728271484375, 19.48248291015625, 95.9986801147461, -1.3213119506835938, -5.7689971923828125, 100.05453491210938, -2.2147216796875, -29.36370849609375, -54.398353576660156, 102.49759674072266, -23.43536376953125, 7.634185791015625, -16.414596557617188, -5.513481140136719, -4.5440826416015625, -93.70672607421875, 11.197883605957031, 132.96075439453125, 7.40399169921875, 68.97024536132812, 101.36493682861328, 3.0494842529296875, 53.39520263671875, 120.01007080078125, 54.29667663574219, 127.6329345703125, 64.07917785644531, 1.031463623046875, 56.19189453125, 28.760421752929688, 71.79389953613281, 55.50494384765625, 86.30615234375, 0.0, 86.66336059570312, -15.2041015625, -43.41455078125, -50.2772216796875, 90.54978942871094, -3.403411865234375, -4.7600860595703125, 116.86984252929688, -31.471969604492188, 3.45172119140625, 75.1741714477539, 5.487186431884766, 116.58795166015625, 107.66798400878906, 123.36257934570312, 196.21011352539062, 16.587860107421875, 46.891021728515625, 62.27630615234375, -35.9359130859375, 95.38681030273438, 38.092926025390625, 1.8859405517578125, 121.92205810546875, -9.753570556640625, 119.9254150390625, 79.36224365234375, 0.051361083984375, 68.01126098632812, 10.598190307617188, -61.63885498046875, 21.22418212890625, 73.716064453125, -5.20697021484375, 95.01947021484375, 93.19691467285156, 7.344512939453125, 5.0268707275390625, -2.44073486328125, 36.858642578125, 63.5333251953125, 132.61129760742188, 117.76449584960938, 105.17057800292969, 4.270027160644531, 142.21185302734375, 49.2220458984375, 1.70623779296875, 88.27227783203125, 37.77253723144531, 83.50112915039062, -8.47705078125, 82.54261779785156, 30.920883178710938, 18.239578247070312, 114.59770965576172, 52.65476989746094, -25.3050537109375, 58.979248046875, 2.913818359375, 87.2193603515625, 68.9625244140625, 143.2491455078125, -2.5928001403808594, 72.88201904296875, 81.1424560546875, -9.1036376953125, 22.190086364746094, 13.696670532226562, 96.00995635986328, -99.3712158203125, 27.813873291015625, 80.30068969726562, 108.30319213867188, 1.1021728515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000244.npy"} {"epoch": 0.5109947643979058, "step": 245, "batch_size": 128, "mean": 42.688087463378906, "std": 56.82699966430664, "min": -120.67355346679688, "p10": -22.424763488769525, "median": 37.649169921875, "p90": 107.20875396728515, "max": 155.63946533203125, "pos_frac": 0.78125, "sample": [20.175140380859375, -13.026512145996094, 32.55253601074219, 29.174827575683594, 4.244354248046875, 25.50323486328125, 82.07403564453125, 4.379638671875, 5.474822998046875, 79.26333618164062, 36.839202880859375, -15.282073974609375, 37.352333068847656, 123.3785400390625, 77.8143310546875, 15.222686767578125, 143.75921630859375, 133.747802734375, 18.637191772460938, 154.3243408203125, -120.67355346679688, -27.7532958984375, -60.13775634765625, -27.01318359375, 32.82978820800781, 66.11077880859375, 0.0, 34.28131103515625, 74.0703353881836, -7.7952423095703125, 86.16342163085938, 88.6322021484375, 85.13783264160156, 97.852783203125, 2.786651611328125, 133.7059326171875, 3.8421707153320312, 27.303176879882812, 88.29499053955078, 69.62069702148438, 123.5716552734375, 12.45587158203125, 48.606719970703125, 80.7440185546875, 9.585845947265625, 89.9837646484375, 27.68536376953125, 45.7989501953125, 46.989688873291016, -0.4727935791015625, -27.4945068359375, 50.694496154785156, 6.484710693359375, 105.63192749023438, 22.539947509765625, -8.03570556640625, 80.77532958984375, 55.7503662109375, 51.37548828125, 110.88801574707031, 139.37969970703125, 6.779499053955078, 1.9642715454101562, 65.58741760253906, 49.88713073730469, 91.47193145751953, 27.775367736816406, -39.27018737792969, 53.571075439453125, -78.26693725585938, 54.72050476074219, 93.00594329833984, 78.15618896484375, 83.11090087890625, 98.06122589111328, -13.135009765625, -59.2769775390625, 32.6856689453125, 154.90228271484375, 102.40780639648438, 6.18549919128418, 97.24378204345703, -59.853118896484375, 7.830329895019531, 11.5386962890625, -54.44770812988281, 7.410736083984375, 104.85659790039062, 96.00799560546875, 96.06050109863281, 78.80934143066406, 15.25830078125, -5.840080261230469, 137.38494873046875, 52.530120849609375, 148.88665771484375, 13.87158203125, 53.268646240234375, 54.96216583251953, 83.42049407958984, -85.4639892578125, 103.66610717773438, 73.38443756103516, -18.18341064453125, -3.644866943359375, 76.13739013671875, 31.303619384765625, 95.3831787109375, 74.35548400878906, 93.91851806640625, 10.453277587890625, -12.366287231445312, 37.745269775390625, 37.553070068359375, 90.9422607421875, 16.05938720703125, -1.478851318359375, -20.458297729492188, 130.38064575195312, -0.5096511840820312, 105.29080200195312, 23.429550170898438, 73.2125244140625, 155.63946533203125, -108.61965942382812, -44.156951904296875, 0.0, 64.77207946777344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000245.npy"} {"epoch": 0.5130890052356021, "step": 246, "batch_size": 128, "mean": 27.993988037109375, "std": 58.96973419189453, "min": -119.1807861328125, "p10": -37.66013526916503, "median": 15.491479873657227, "p90": 114.37769470214843, "max": 155.71392822265625, "pos_frac": 0.640625, "sample": [59.37834167480469, 85.29571533203125, 78.17630767822266, -2.134246826171875, 74.43685913085938, -6.698066711425781, 155.71392822265625, 9.399627685546875, 16.052021026611328, 47.06401062011719, -42.4406852722168, 73.91734313964844, 40.01641845703125, 17.547531127929688, -50.742950439453125, -59.05000305175781, -12.667800903320312, 127.92294311523438, 129.71331787109375, -53.74842834472656, 109.23042297363281, 135.7347412109375, 18.80841064453125, 91.02784729003906, 39.045806884765625, -3.460540771484375, 10.869110107421875, -51.917144775390625, 24.466400146484375, -5.275177001953125, 38.55633544921875, -27.534210205078125, 113.6884765625, 31.937408447265625, 19.476898193359375, -1.6023101806640625, -96.25885009765625, 112.97769165039062, 4.40911865234375, 0.7732429504394531, 14.70538330078125, -105.38873291015625, -20.979888916015625, 115.98587036132812, -47.61808776855469, -22.56640625, -65.31907653808594, -27.758026123046875, 5.59552001953125, 73.22909545898438, 29.94061279296875, 134.1883544921875, -35.611328125, 18.628387451171875, 118.23977661132812, -15.739765167236328, 37.71636962890625, -21.706329345703125, -119.1807861328125, 29.257877349853516, 128.04379272460938, 63.80255126953125, 55.360931396484375, 53.575927734375, -34.154693603515625, 60.653587341308594, 130.65231323242188, 75.30172729492188, -13.102081298828125, 124.34426879882812, 108.05355834960938, 30.35968017578125, -33.24357604980469, -2.271270751953125, 21.37548828125, 1.404327392578125, 100.29148864746094, 14.896835327148438, -24.406051635742188, -35.369476318359375, -50.9178466796875, -8.075164794921875, 90.9547119140625, 105.13385009765625, -24.623321533203125, -2.308502197265625, -69.378173828125, -62.64141845703125, 16.81402587890625, 6.750511169433594, 47.355621337890625, 110.69920349121094, 74.1624755859375, -21.579864501953125, 65.29792022705078, 23.737640380859375, 0.93780517578125, -9.69970703125, 0.0, 105.58544921875, 119.26493835449219, 97.20928955078125, -14.580848693847656, -23.207923889160156, 87.98443603515625, 0.83624267578125, 1.44635009765625, 79.79830932617188, 13.165306091308594, 67.81172180175781, 43.716346740722656, 4.812644958496094, 108.8724365234375, 65.32498168945312, 14.930938720703125, 116.9788818359375, -6.103290557861328, -32.190696716308594, -22.959991455078125, -20.886474609375, 124.06854248046875, -3.085235595703125, 1.836456298828125, 58.47416687011719, 11.811058044433594, 13.552764892578125, -3.64263916015625, 106.49234008789062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000246.npy"} {"epoch": 0.5151832460732985, "step": 247, "batch_size": 128, "mean": 37.06138610839844, "std": 53.82144546508789, "min": -119.35711669921875, "p10": -22.798159790039062, "median": 29.80614471435547, "p90": 113.03985900878905, "max": 163.6724853515625, "pos_frac": 0.75, "sample": [100.90757751464844, -30.1513671875, 79.20454406738281, 69.69087982177734, -11.995529174804688, 50.536407470703125, 110.88870239257812, 116.24504089355469, 85.57481384277344, 61.85502624511719, 0.0, 88.46636199951172, 114.20906066894531, 17.349517822265625, 8.011024475097656, 36.80560302734375, -21.08094024658203, 7.796600341796875, -44.81854248046875, 82.27749633789062, -6.7188720703125, 115.407470703125, -22.55731201171875, -64.77417755126953, 52.8775634765625, 85.9014892578125, 43.99127197265625, -4.64141845703125, 92.01675415039062, 110.4678955078125, 63.5030517578125, 105.10308074951172, 13.732101440429688, 137.4425048828125, 69.17295837402344, -28.63531494140625, 1.110107421875, 127.314697265625, -6.2978515625, 9.032272338867188, -10.010704040527344, -24.52086639404297, 24.227874755859375, -10.507156372070312, 12.994720458984375, 70.71710205078125, -2.8429946899414062, 35.94342041015625, 3.6753158569335938, -66.9266357421875, 3.2316627502441406, 75.59466552734375, 53.922630310058594, 59.765533447265625, -31.438873291015625, 101.1502456665039, 142.55242919921875, 17.3321533203125, 1.6594085693359375, -13.291061401367188, -19.120811462402344, -10.401771545410156, -6.9423675537109375, 124.60391235351562, 0.415557861328125, 0.5434417724609375, 77.72433471679688, -77.91021728515625, 8.697555541992188, -23.360137939453125, 22.417984008789062, 145.287109375, -119.35711669921875, 0.1312255859375, 98.40283203125, 14.588218688964844, 25.964012145996094, 8.184837341308594, 30.371490478515625, 72.59378051757812, 36.93450927734375, 82.55107116699219, 163.6724853515625, 13.729644775390625, -25.311721801757812, -7.402687072753906, 2.1779251098632812, 40.171661376953125, 91.324951171875, 46.02520751953125, 25.960205078125, 83.76644897460938, 55.03385925292969, 23.4339599609375, 15.120040893554688, 88.9420166015625, 119.664794921875, 112.53877258300781, 132.06048583984375, -15.828521728515625, 29.328598022460938, 22.315956115722656, 55.7567138671875, 40.009490966796875, 85.493408203125, 44.73295593261719, 30.28369140625, -29.99346923828125, 124.01708984375, 12.37603759765625, -90.56600952148438, 89.50162506103516, 54.59241485595703, -21.950469970703125, 27.324462890625, 32.858543395996094, 75.40078735351562, -6.697551727294922, 30.531036376953125, -0.89959716796875, 80.05115509033203, 37.415069580078125, 120.06907653808594, 2.7780399322509766, 91.06492614746094, 7.1041412353515625, 23.50885009765625, 61.60247802734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000247.npy"} {"epoch": 0.5172774869109947, "step": 248, "batch_size": 128, "mean": 30.509273529052734, "std": 57.5780143737793, "min": -161.71823120117188, "p10": -37.90616149902343, "median": 25.25189971923828, "p90": 110.73850097656248, "max": 168.5152587890625, "pos_frac": 0.7578125, "sample": [13.918006896972656, 103.0665283203125, -4.075347900390625, 10.748085021972656, 131.64459228515625, -21.63702392578125, 21.310150146484375, 4.60467529296875, 81.600341796875, 15.303199768066406, 25.858367919921875, 56.014862060546875, 70.73541259765625, 28.20229721069336, 23.913856506347656, 34.24853515625, 127.70925903320312, 22.050491333007812, 0.262451171875, 82.30500793457031, -17.17455291748047, 24.645431518554688, 78.05294799804688, 71.06038665771484, 120.66061401367188, 35.55072021484375, 26.99197006225586, 34.43205261230469, 2.49822998046875, 68.36737060546875, 22.137557983398438, -39.583045959472656, 132.88153076171875, -17.806976318359375, 4.569742202758789, 35.61521911621094, 115.3619384765625, -161.71823120117188, 27.690277099609375, -3.356657028198242, -27.90771484375, 35.494720458984375, 32.35198974609375, -25.095184326171875, -8.306324005126953, 35.483642578125, -3.664937973022461, -99.54159545898438, 36.64738464355469, 43.189697265625, 14.8515625, 132.1306915283203, 33.54743957519531, 21.95960807800293, 146.8499755859375, -49.942108154296875, 116.82867431640625, 0.6196918487548828, -3.10784912109375, 124.78628540039062, 42.49674987792969, 90.6772689819336, -26.2506103515625, -68.10519409179688, 16.876327514648438, 13.217864990234375, 168.5152587890625, 108.9888916015625, 76.04965209960938, 82.5511474609375, 78.090087890625, -39.29603576660156, 1.9270172119140625, 73.8087158203125, 1.45306396484375, -89.87518310546875, 0.0, 30.23284912109375, 117.25341796875, 108.96420288085938, 97.21502685546875, 70.13943481445312, 67.24519348144531, 53.4892578125, 9.37750244140625, 33.13291931152344, -25.610595703125, -79.98541259765625, -7.008613586425781, 9.561271667480469, -37.31050109863281, 12.51513671875, 114.8209228515625, 92.27383422851562, 105.83824157714844, -81.84844970703125, 33.675018310546875, 22.73754119873047, 7.446342468261719, 97.15277099609375, 0.9553451538085938, -49.09375, 33.86976623535156, -17.50640869140625, 90.63548278808594, 34.52960205078125, 86.18484497070312, 8.725112915039062, 10.097564697265625, -96.22181701660156, 24.103912353515625, 20.57537841796875, 41.74427032470703, -3.6374969482421875, 141.41629028320312, 72.747802734375, 66.89984130859375, 53.16975402832031, 44.465118408203125, -75.32275390625, 4.307167053222656, -5.7663726806640625, 15.860733032226562, 27.418014526367188, 85.90238952636719, -69.21343994140625, 10.313323974609375, 19.759185791015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000248.npy"} {"epoch": 0.5193717277486911, "step": 249, "batch_size": 128, "mean": 35.011043548583984, "std": 57.36042022705078, "min": -128.556884765625, "p10": -26.46621398925781, "median": 25.815505981445312, "p90": 108.13047332763671, "max": 157.42633056640625, "pos_frac": 0.7109375, "sample": [-11.844940185546875, 8.59494400024414, -65.40144348144531, -72.8504638671875, 36.72441101074219, 4.7170867919921875, -16.216262817382812, 10.948348999023438, -0.517333984375, 20.817596435546875, -2.4228057861328125, -21.58185577392578, 90.359130859375, 132.24514770507812, -9.743057250976562, 72.48211669921875, 33.231109619140625, 111.9814453125, 27.091400146484375, 104.29421997070312, 13.832412719726562, -125.6717529296875, 102.05558013916016, 19.14019775390625, -34.89134216308594, 138.49859619140625, 79.17105102539062, -19.27789306640625, 39.0343017578125, 68.84226989746094, 18.22411346435547, -11.018280029296875, 35.59710693359375, -12.898902893066406, -2.8868408203125, 29.3714599609375, 118.24400329589844, 32.26298522949219, 126.25094604492188, -31.89886474609375, 92.15327453613281, 99.1826171875, -11.54583740234375, 106.48005676269531, 142.34567260742188, 25.938247680664062, -88.40074157714844, 86.96076965332031, 54.93896484375, 4.862571716308594, -11.748748779296875, 41.228355407714844, 153.7915802001953, 73.64530181884766, 96.84814453125, -2.580810546875, 12.571578979492188, 67.41834259033203, -28.472412109375, 85.48849487304688, 34.97503662109375, -1.49493408203125, 0.94708251953125, 3.678192138671875, 90.85833740234375, 9.505325317382812, 84.53756713867188, 1.9417381286621094, 77.13639831542969, 83.65104675292969, -24.19402313232422, -62.689598083496094, 25.44317626953125, 71.72430419921875, 57.43315124511719, 80.68667602539062, 157.42633056640625, 129.32437133789062, 98.36271667480469, 90.75867462158203, 140.11053466796875, 72.6349105834961, 18.57166290283203, -54.873207092285156, 84.76309204101562, 129.43980407714844, -2.875202178955078, -23.40228271484375, -13.314395904541016, 84.34027099609375, 18.306076049804688, -46.42626953125, 74.9532470703125, 82.67933654785156, 24.676589965820312, 86.554443359375, 40.269744873046875, 25.692764282226562, 2.5361175537109375, -40.169769287109375, 113.4761962890625, 1.06524658203125, 33.68079376220703, 60.939056396484375, 22.863052368164062, 7.2735595703125, -5.500221252441406, -128.556884765625, -35.760894775390625, -12.782341003417969, -25.606414794921875, 8.650102615356445, 51.473480224609375, 1.8036956787109375, 18.11248779296875, -2.81787109375, 76.01416015625, 38.255706787109375, 96.34857177734375, 4.50555419921875, 101.30148315429688, 77.4130859375, 68.90480041503906, -24.134307861328125, -4.8177032470703125, 41.0108642578125, 22.662500381469727, 125.16136169433594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000249.npy"} {"epoch": 0.5214659685863874, "step": 250, "batch_size": 128, "mean": 35.56262969970703, "std": 61.859371185302734, "min": -107.17129516601562, "p10": -44.017482376098634, "median": 33.284393310546875, "p90": 118.5779541015625, "max": 161.6846466064453, "pos_frac": 0.7109375, "sample": [27.1734619140625, 15.099967956542969, -57.259307861328125, 113.0709228515625, 85.57058715820312, 159.6527099609375, 145.18434143066406, 83.03155517578125, 38.31683349609375, -87.18750762939453, 80.87545776367188, 21.900903701782227, 117.98956298828125, 32.08624267578125, 19.890960693359375, 126.64311218261719, 59.74822998046875, -9.394073486328125, 122.1351318359375, 95.4022216796875, 110.96418762207031, 63.09210205078125, 27.029495239257812, 92.3382568359375, -29.45867919921875, 3.3779296875, -107.17129516601562, 20.498214721679688, -6.73394775390625, -0.1614990234375, 122.0965576171875, -88.90777587890625, 48.971099853515625, 151.45941162109375, -9.2933349609375, 9.88616943359375, 45.51458740234375, 1.351480484008789, 122.68572998046875, -100.939697265625, -1.9432334899902344, 49.809715270996094, 112.17434692382812, 27.901771545410156, 85.79214477539062, 63.55296325683594, -43.50860595703125, -76.34552001953125, -1.5291709899902344, 70.22340393066406, 86.34902954101562, 5.105094909667969, 57.5802001953125, -5.07806396484375, 34.631103515625, 12.27777099609375, -6.475067138671875, -33.1441650390625, 63.87133026123047, 3.0512237548828125, 95.12884521484375, 125.5021743774414, 80.30316162109375, -0.1980419158935547, 23.610374450683594, 41.078125, 142.80905151367188, 96.07225036621094, -9.79351806640625, 36.34385681152344, 82.28128814697266, 91.32844543457031, -23.25640869140625, 109.57414245605469, 35.4329833984375, -24.987884521484375, 67.50494384765625, 103.61236572265625, 119.95086669921875, 36.577606201171875, 17.412654876708984, -22.4097900390625, 34.4825439453125, -24.6968994140625, 104.035400390625, 12.11508560180664, -1.853851318359375, -95.26463317871094, 76.93771362304688, 132.31411743164062, -7.953857421875, 161.6846466064453, -67.32899475097656, -59.065673828125, -99.61785888671875, -0.1152801513671875, 14.275886535644531, -50.6832275390625, -86.7078857421875, 93.13934326171875, 71.61311340332031, 95.9074935913086, 86.14361572265625, 75.01983642578125, 88.357666015625, 45.65074920654297, -30.11163330078125, 35.849029541015625, 15.47256088256836, -4.535776138305664, 5.482452392578125, 69.77894592285156, 140.40745544433594, -8.993255615234375, 39.044097900390625, 66.07044982910156, 96.61207580566406, 7.8151702880859375, 7.426177978515625, 0.7358856201171875, -24.324607849121094, 86.71327209472656, 43.003936767578125, 0.059844970703125, 6.783660888671875, 24.58612060546875, 26.252822875976562, -45.20486068725586], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000250.npy"} {"epoch": 0.5235602094240838, "step": 251, "batch_size": 128, "mean": 35.70658874511719, "std": 57.07754898071289, "min": -121.51762390136719, "p10": -28.681391525268555, "median": 25.20256805419922, "p90": 115.12506103515624, "max": 219.8980712890625, "pos_frac": 0.75, "sample": [69.59724426269531, 58.601226806640625, 25.893417358398438, -14.952865600585938, -44.0689697265625, -41.532989501953125, -65.86669921875, 43.459259033203125, 11.161529541015625, 20.42059326171875, 8.304187774658203, 59.8358154296875, 93.77066040039062, 82.15059661865234, 4.0963592529296875, -3.6725387573242188, 27.293212890625, -1.976959228515625, 5.1720123291015625, 124.72344970703125, 1.3467254638671875, 2.955596923828125, -30.2657470703125, 106.34942626953125, 81.1939697265625, 120.30450439453125, -49.67889404296875, 95.63145446777344, 122.62200927734375, 21.949188232421875, 98.57965087890625, 4.960060119628906, -26.806259155273438, 41.30401611328125, 29.74639892578125, -17.701068878173828, 53.7899169921875, 0.6136512756347656, -28.520751953125, -29.056217193603516, 87.30535888671875, 43.83538818359375, 99.1937255859375, 118.88674926757812, 118.66297912597656, 5.151275634765625, -4.77593994140625, -43.91389465332031, 3.2105674743652344, 4.4862518310546875, -31.17517852783203, 17.65667724609375, 219.8980712890625, 36.2388916015625, 71.311767578125, 90.93718719482422, 18.16064453125, 105.57437133789062, 7.936561584472656, 63.816688537597656, 79.95394134521484, -121.51762390136719, 27.79297637939453, 83.580078125, 24.123458862304688, -35.001243591308594, 7.1179351806640625, -3.5051040649414062, 24.51171875, -3.2520694732666016, 98.29949951171875, 81.00054931640625, 9.492645263671875, 4.029640197753906, 53.501129150390625, -26.13232421875, 2.7989501953125, 27.72064208984375, 5.537876129150391, 75.66720581054688, 2.44696044921875, 64.266845703125, 2.1607770919799805, -4.1438140869140625, 63.929718017578125, 91.77234649658203, 117.0865478515625, -80.49815368652344, -3.15142822265625, 1.9790802001953125, 56.885345458984375, 125.04571533203125, 96.74264526367188, 62.023529052734375, 139.26272583007812, -3.5882091522216797, 180.432861328125, 114.284423828125, 83.3447494506836, -5.727497100830078, 61.75030517578125, 29.45361328125, 68.29718017578125, 44.8634033203125, 61.009307861328125, 31.75311279296875, 108.61507415771484, 0.69384765625, 75.20964050292969, -67.51382446289062, 0.24005126953125, 34.546539306640625, 12.786270141601562, -15.474586486816406, 4.940238952636719, -67.66720581054688, -25.9237060546875, -19.482765197753906, 3.1536407470703125, 138.87591552734375, 99.43215942382812, -22.199264526367188, 120.11610412597656, 126.53921508789062, -11.732757568359375, 4.281837463378906, 74.06665802001953, 85.413818359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000251.npy"} {"epoch": 0.5256544502617801, "step": 252, "batch_size": 128, "mean": 39.24053192138672, "std": 64.23287200927734, "min": -136.84884643554688, "p10": -31.83067932128906, "median": 34.042266845703125, "p90": 124.02001495361328, "max": 193.46578979492188, "pos_frac": 0.7265625, "sample": [151.52630615234375, 127.33745574951172, 124.39373779296875, 64.11061096191406, -74.70967102050781, 63.6319694519043, -71.89107513427734, 11.856369018554688, 8.165706634521484, -16.051666259765625, -1.5391616821289062, -29.703201293945312, 17.15899658203125, 32.34381103515625, 193.46578979492188, 35.027862548828125, 116.55770874023438, 73.09539031982422, 105.33456420898438, 96.96527099609375, 6.148012161254883, 52.81524658203125, -21.201202392578125, 123.85984802246094, -136.84884643554688, 5.347900390625, 5.1683197021484375, 70.46782684326172, -32.230133056640625, 43.745758056640625, 0.41468048095703125, 117.72933959960938, 108.91845703125, 64.52764892578125, 77.89237213134766, 20.952301025390625, -49.843170166015625, -11.306591033935547, 63.123565673828125, 190.435302734375, 91.21005249023438, 1.40594482421875, 79.49995422363281, 27.082550048828125, 54.03559875488281, 15.418777465820312, 29.972579956054688, -9.871917724609375, 0.2017822265625, 14.071525573730469, 102.55313110351562, 2.74542236328125, 124.41543579101562, 82.98826599121094, -18.56494140625, -3.5572509765625, 70.1832275390625, -45.60528564453125, 30.5782470703125, 29.098793029785156, 83.00672912597656, -20.123809814453125, 29.95025634765625, 131.17828369140625, 58.920433044433594, 69.81765747070312, 111.7659912109375, 34.94122314453125, 77.9112548828125, 118.27215576171875, 1.71173095703125, 98.09242248535156, 132.69830322265625, -5.691062927246094, -53.35083770751953, 81.2803955078125, -91.6422119140625, 9.424751281738281, 74.36114501953125, 86.46923828125, 80.42752075195312, 3.08587646484375, 43.447021484375, 92.01590728759766, -7.9891815185546875, -1.1217117309570312, 0.0, 87.33834075927734, -98.7613525390625, 106.59589385986328, -0.19298553466796875, 98.02944946289062, -7.956634521484375, -21.88836669921875, 33.41766357421875, 130.5023193359375, -2.7384185791015625, 128.65609741210938, 48.349365234375, 111.33132934570312, -5.4364013671875, 81.29390716552734, 37.93769454956055, 1.55889892578125, 125.9698486328125, -0.736907958984375, 90.93722534179688, -113.47918701171875, 18.4322509765625, 6.2227935791015625, -28.085296630859375, -22.971832275390625, 34.6668701171875, -103.0377197265625, -95.99884033203125, 6.20587158203125, 90.2120361328125, 126.8114013671875, -49.66314697265625, 99.9364013671875, 66.633544921875, 137.33529663085938, 84.10842895507812, -31.65948486328125, 12.433319091796875, 80.91848754882812, 11.604145050048828, 104.04066467285156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000252.npy"} {"epoch": 0.5277486910994764, "step": 253, "batch_size": 128, "mean": 30.13889503479004, "std": 59.6779899597168, "min": -129.76800537109375, "p10": -37.416743469238284, "median": 17.282363891601562, "p90": 116.47612915039062, "max": 149.2388916015625, "pos_frac": 0.734375, "sample": [140.23553466796875, -7.12005615234375, 10.537673950195312, 56.18762969970703, 15.96759033203125, 24.113845825195312, 2.3643150329589844, 99.8653564453125, 145.67254638671875, 24.882568359375, 37.71234130859375, 117.15449523925781, 6.937164306640625, -38.3682861328125, 68.683837890625, -95.93795776367188, 55.94896697998047, -60.56664276123047, 100.449951171875, -29.61114501953125, 1.233673095703125, -10.39349365234375, 34.08202362060547, 19.029800415039062, 0.3475341796875, -33.754638671875, 31.24237060546875, 122.222412109375, 94.50421905517578, 87.62615966796875, -7.872365951538086, -11.832538604736328, -94.052001953125, 25.000778198242188, 149.2388916015625, 6.616790771484375, -13.894180297851562, 111.87130737304688, -129.76800537109375, -28.0762939453125, 101.490234375, 31.03985595703125, 116.31878662109375, 23.30872344970703, 19.36248779296875, -37.68205261230469, 103.36799621582031, 4.227020263671875, -9.387924194335938, 16.49078369140625, 116.84326171875, 33.78717041015625, 113.18655395507812, 3.556793212890625, -20.339332580566406, 102.154052734375, 2.7671585083007812, 10.357521057128906, -36.013885498046875, 81.77272033691406, 25.97308349609375, -37.30303955078125, -0.3155364990234375, 98.17481994628906, 21.29119873046875, 107.77461242675781, -41.726104736328125, -15.668045043945312, 56.01234436035156, -42.050994873046875, 18.073944091796875, -11.51617431640625, -104.8564224243164, -0.336517333984375, -104.8570556640625, 34.317718505859375, -69.05374145507812, 3.9922237396240234, -5.508338928222656, -30.5501708984375, 70.12995910644531, 1.4499931335449219, 54.77740478515625, 91.76336669921875, 36.566680908203125, -1.653594970703125, 26.68633270263672, 7.304290771484375, -50.66265869140625, 134.28109741210938, -46.11324691772461, 7.957878112792969, 89.55654907226562, 5.164581298828125, 4.5494384765625, 103.072509765625, -36.54503631591797, 138.6845703125, 8.682640075683594, 76.45150756835938, 79.03126525878906, 11.266014099121094, -34.54205322265625, 0.914306640625, 36.683746337890625, 116.89804077148438, 13.262481689453125, 124.71829223632812, 14.1318359375, 6.63102912902832, 20.26116943359375, 124.7772216796875, 84.37801361083984, 1.9839019775390625, 89.08837127685547, 14.153839111328125, 9.81768798828125, 14.703643798828125, 30.25738525390625, 143.26611328125, 6.653404235839844, 41.537841796875, 127.64617919921875, 92.2327880859375, 11.2239990234375, 85.20443725585938, 64.04103088378906, 98.494384765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000253.npy"} {"epoch": 0.5298429319371728, "step": 254, "batch_size": 128, "mean": 38.86888122558594, "std": 54.846858978271484, "min": -131.05133056640625, "p10": -22.543753051757808, "median": 30.373798370361328, "p90": 112.02701492309569, "max": 149.34620666503906, "pos_frac": 0.75, "sample": [136.15399169921875, -1.6361236572265625, -75.57635498046875, 29.77942657470703, -9.52569580078125, -3.5624046325683594, 20.9151611328125, 17.19353485107422, -1.7653961181640625, 19.403152465820312, -34.42205047607422, -15.705657958984375, -12.258819580078125, -1.441253662109375, 6.46435546875, 116.70010375976562, 6.59564208984375, 100.25225830078125, 46.7142333984375, -31.29595947265625, -30.310028076171875, 120.89251708984375, 17.764259338378906, 9.17333984375, -10.637603759765625, 35.24031066894531, 94.22650146484375, -61.76802062988281, 94.20611572265625, 65.9693603515625, 95.68072509765625, 130.63662719726562, -84.6324462890625, 73.04959106445312, 0.7160739898681641, 19.7237548828125, -20.203659057617188, 50.66814041137695, 10.824722290039062, 9.590972900390625, 63.46490478515625, 51.73944091796875, 81.2225341796875, 37.394195556640625, -62.0093994140625, 30.968170166015625, 70.19345092773438, 55.4666748046875, 62.162353515625, 95.38508605957031, 59.614898681640625, 4.647674560546875, 39.206626892089844, 107.55160522460938, 77.3648681640625, 27.791534423828125, 82.75650024414062, -33.07232666015625, 19.54071044921875, 76.146484375, 2.8004684448242188, 15.967418670654297, 129.03839111328125, 7.8534393310546875, 44.292877197265625, 113.14620208740234, 43.16815185546875, 108.69378662109375, 111.54736328125, 29.623947143554688, 9.429054260253906, 38.298675537109375, 121.48330688476562, 5.588226318359375, 48.046630859375, -20.781509399414062, -4.957977294921875, 32.152923583984375, 3.414867401123047, 8.741241455078125, -2.5858154296875, -10.313621520996094, 125.54461669921875, 109.9610595703125, 76.54305267333984, 135.10443115234375, 66.54388427734375, 98.27580261230469, 17.52032470703125, 109.71609497070312, 24.422828674316406, 109.91255187988281, 107.77273559570312, 90.62159729003906, 101.7805404663086, 25.03314208984375, 0.1513042449951172, 97.84217071533203, -4.8390655517578125, -7.1622772216796875, -13.725120544433594, 48.200927734375, 41.64778137207031, -1.8750228881835938, 0.0, -26.655654907226562, 118.350830078125, 74.80828857421875, 33.0849609375, 25.508209228515625, 9.152488708496094, 23.444435119628906, 36.212669372558594, 110.85260009765625, 76.15802001953125, -0.20928955078125, -131.05133056640625, -57.97930145263672, 135.62896728515625, 24.487037658691406, -43.263092041015625, -32.7646484375, 5.13067626953125, 149.34620666503906, 142.0189666748047, 77.39065551757812, 55.60711669921875, 94.98768615722656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000254.npy"} {"epoch": 0.5319371727748691, "step": 255, "batch_size": 128, "mean": 42.32177734375, "std": 64.53144073486328, "min": -109.36534118652344, "p10": -24.784732055664044, "median": 29.648033142089844, "p90": 125.4062515258789, "max": 173.158203125, "pos_frac": 0.765625, "sample": [7.8111419677734375, -60.88800048828125, 135.56256103515625, 67.72093200683594, 38.224830627441406, -101.95193481445312, 1.86297607421875, 117.5672607421875, -11.88653564453125, 103.6209487915039, -9.0360107421875, 9.7830810546875, 18.62856674194336, 8.38812255859375, 143.69345092773438, 122.447021484375, 10.684959411621094, 15.50958251953125, 135.21185302734375, 8.780426025390625, 108.65704345703125, -82.4447021484375, -2.30792236328125, 98.61843872070312, 39.59674072265625, 121.57003784179688, 25.765914916992188, 100.29086303710938, -19.649383544921875, 30.387420654296875, 77.56884765625, 8.663421630859375, 21.48253631591797, 13.1650390625, 38.92509460449219, 31.615615844726562, 118.37258911132812, 57.033233642578125, -81.06939697265625, -54.746185302734375, 173.158203125, 146.20904541015625, 124.99739074707031, 25.97418212890625, -86.94136810302734, -18.899948120117188, -107.67361450195312, 128.3092041015625, 90.93201446533203, 82.64776611328125, 114.92282104492188, 103.53536987304688, 86.60659790039062, 88.01235961914062, 140.81028747558594, -6.8882904052734375, 50.571014404296875, -84.67071533203125, -38.595794677734375, 103.08958435058594, 23.740127563476562, 113.71755981445312, 102.31008911132812, 96.68386840820312, 36.2662353515625, 87.63702392578125, 5.755897521972656, -95.38172149658203, 121.98910522460938, 17.853713989257812, 138.85031127929688, 11.075775146484375, 32.670040130615234, -2.8324661254882812, 3.99713134765625, 7.940312385559082, 23.44122314453125, 28.908645629882812, 8.134613037109375, 93.38452911376953, -6.4588775634765625, 81.80233764648438, -109.36534118652344, 136.39202880859375, 113.08355712890625, 92.48237609863281, -7.60076904296875, 50.9765625, 18.170120239257812, 15.054527282714844, 126.36026000976562, 104.53363037109375, 71.03662109375, 121.62010955810547, -16.92926025390625, 147.26870727539062, 110.15206909179688, 3.315460205078125, 1.4947509765625, -14.230300903320312, 0.37515735626220703, -36.7672119140625, 74.29029846191406, 4.8446197509765625, 128.84613037109375, 41.2567138671875, 5.750244140625, 101.3211669921875, -1.7292251586914062, 64.470458984375, 4.2532958984375, 8.925140380859375, -12.115852355957031, 114.25686645507812, 20.765625, 34.96089172363281, -8.138809204101562, -11.327850341796875, 101.56944274902344, 43.70062255859375, 21.235023498535156, 103.80467987060547, -3.0762939453125, 83.98345947265625, -53.804931640625, 150.10519409179688, 22.22552490234375, -11.430252075195312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000255.npy"} {"epoch": 0.5340314136125655, "step": 256, "batch_size": 128, "mean": 43.082672119140625, "std": 67.35308074951172, "min": -128.73216247558594, "p10": -48.29477233886719, "median": 47.76301574707031, "p90": 124.56605224609375, "max": 158.31927490234375, "pos_frac": 0.7265625, "sample": [5.102630615234375, 112.458984375, 124.52423095703125, -38.473876953125, 19.891937255859375, 77.6163330078125, 111.58465576171875, 86.36587524414062, 100.797119140625, -2.69659423828125, 94.63999938964844, -57.36027526855469, -36.000152587890625, -96.12535858154297, 74.89973449707031, 119.27363586425781, 117.34112548828125, -21.3524169921875, 80.63427734375, 124.66363525390625, 0.9055404663085938, 106.92298889160156, -50.39007568359375, 158.31927490234375, 17.1768798828125, 118.91520690917969, 105.5533447265625, -61.4390869140625, 133.33377075195312, 13.7156982421875, 133.36273193359375, 32.26925277709961, 53.5194091796875, 37.29949951171875, 36.974365234375, 9.570343017578125, -41.557525634765625, -102.87384033203125, 94.67904663085938, 58.760498046875, 137.7279052734375, 25.761474609375, 79.14118957519531, 123.91352844238281, 52.82475280761719, 51.6065673828125, -34.73945617675781, 10.841331481933594, -46.3990478515625, 151.83477783203125, 5.3128662109375, -43.760475158691406, 127.66616821289062, 130.45361328125, -9.482803344726562, 3.1847705841064453, -127.42788696289062, -33.12249755859375, 107.44564819335938, 75.89228820800781, 95.20298767089844, 16.90826416015625, 95.57916259765625, 133.46719360351562, -2.218794822692871, 90.24525451660156, 61.026123046875, -94.91341400146484, 97.22079467773438, -3.2824859619140625, -48.126251220703125, -90.2802734375, 80.76922607421875, 53.908294677734375, 39.22113037109375, 106.73565673828125, -14.394256591796875, 112.27252197265625, 25.394683837890625, 130.58425903320312, 143.70169067382812, 85.2173843383789, 89.88484954833984, 1.28887939453125, -1.62384033203125, 60.74078369140625, -89.73146057128906, 11.825794219970703, 46.280426025390625, 122.04708862304688, -49.4779052734375, 117.84904479980469, 95.59635925292969, 126.0858154296875, 14.05657958984375, 96.2680892944336, 51.06141662597656, -9.78582763671875, 21.42218780517578, 27.893310546875, 37.786529541015625, 35.503028869628906, 124.103515625, 76.4339599609375, 34.862457275390625, 49.24560546875, 35.10325622558594, 115.5994873046875, -50.10002517700195, -128.73216247558594, 6.794952392578125, 122.67811584472656, 126.79684448242188, 15.008354187011719, -9.605979919433594, -0.18829345703125, 120.38156127929688, -32.665191650390625, -13.458717346191406, -48.68798828125, 93.13877868652344, -5.05047607421875, -2.9759063720703125, 60.7735595703125, 18.93621826171875, 73.33009338378906, 65.3360595703125, 106.83648681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000256.npy"} {"epoch": 0.5361256544502618, "step": 257, "batch_size": 128, "mean": 39.083473205566406, "std": 59.79700469970703, "min": -111.80107116699219, "p10": -26.269088745117188, "median": 28.60904312133789, "p90": 119.0592041015625, "max": 169.08949279785156, "pos_frac": 0.6875, "sample": [72.91993713378906, 62.289276123046875, 169.08949279785156, -20.937541961669922, -2.0613174438476562, 164.35629272460938, 22.083160400390625, 120.16290283203125, 42.11566162109375, 102.42281341552734, 63.90899658203125, 138.51849365234375, 104.18869018554688, 67.44512939453125, -0.5028553009033203, 30.23859405517578, 19.255859375, -61.084678649902344, -25.14379119873047, 57.530120849609375, 119.3394775390625, -34.94312286376953, 82.73249053955078, 154.88674926757812, -18.221923828125, -59.37548828125, 20.39324951171875, 26.641281127929688, -3.51806640625, -43.061668395996094, -18.75926971435547, -30.1029052734375, 109.80755615234375, 26.9794921875, 57.731414794921875, 118.36492919921875, 166.1885986328125, 4.3327484130859375, 129.61178588867188, 91.67520904541016, -20.39886474609375, 3.9127426147460938, -19.958518981933594, 101.50088500976562, 123.34158325195312, 7.17230224609375, -2.7492752075195312, 32.44207763671875, 25.377944946289062, -111.80107116699219, 83.595458984375, 143.24932861328125, 63.178016662597656, 61.43815612792969, 111.55810546875, 49.924896240234375, 76.21073913574219, -14.671768188476562, 60.22999572753906, 24.5400390625, -6.273529052734375, 126.00741577148438, 98.69456481933594, 15.377159118652344, -54.09776306152344, -14.682708740234375, 39.0074462890625, 2.7589378356933594, 9.31622314453125, 67.01425170898438, 0.0255889892578125, 96.99984741210938, 74.1188735961914, 38.623046875, 19.19900894165039, 11.403854370117188, -64.60725402832031, -0.053741455078125, 96.61201477050781, 118.9390869140625, 2.7010955810546875, 3.486419677734375, 78.35589599609375, 30.587127685546875, 12.190433502197266, -16.337448120117188, 2.45184326171875, 117.3121337890625, 55.90771484375, -8.939544677734375, -15.599365234375, 86.08135986328125, -30.838592529296875, 118.82711791992188, 66.29476928710938, 104.55050659179688, 107.7828369140625, 37.01995849609375, -17.141571044921875, 8.803886413574219, 41.612762451171875, 13.83172607421875, 111.88839721679688, 35.01141357421875, -5.8264312744140625, -15.992462158203125, -3.728607177734375, -88.97062683105469, -31.113677978515625, 124.01498413085938, 22.634246826171875, 107.244873046875, -20.0018310546875, 159.6035919189453, -27.162353515625, 0.32000732421875, 65.986328125, -25.886260986328125, 86.68292236328125, 33.2772216796875, 90.94192504882812, -5.353549957275391, -10.369720458984375, 69.03697204589844, -10.13876724243164, -12.15380859375, -49.69560241699219, 103.52099609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000257.npy"} {"epoch": 0.5382198952879581, "step": 258, "batch_size": 128, "mean": 43.163719177246094, "std": 55.949790954589844, "min": -136.08169555664062, "p10": -21.025630187988277, "median": 33.21513366699219, "p90": 114.62417907714844, "max": 152.74966430664062, "pos_frac": 0.765625, "sample": [101.7725830078125, 20.408594131469727, 17.410919189453125, 3.7628936767578125, 55.71730041503906, 26.0892333984375, 82.87490844726562, -24.268951416015625, 23.253137588500977, 128.06103515625, -34.878936767578125, 35.1087646484375, 55.156524658203125, 14.21923828125, 110.30599975585938, 70.15724182128906, -23.83465576171875, 18.272796630859375, 113.263916015625, 90.674560546875, 123.44122314453125, 47.98174285888672, -54.05476379394531, 11.37750244140625, 141.4771728515625, -5.1146697998046875, -136.08169555664062, 110.25363159179688, 87.93435668945312, 34.951171875, 107.62387084960938, 99.27243041992188, -33.348907470703125, 63.59521484375, 79.77801513671875, 32.582122802734375, 4.309711456298828, -24.146697998046875, 65.38397979736328, 97.35821533203125, 112.7547607421875, 83.87542724609375, 6.034629821777344, 37.40254211425781, 14.564956665039062, 20.766510009765625, 23.3544921875, 24.62640380859375, 18.453033447265625, 13.238357543945312, 67.70108032226562, 8.18034553527832, 102.32028198242188, 24.77783203125, -116.07508850097656, 82.6866683959961, 31.98553466796875, -7.044464111328125, -49.40350341796875, 6.45867919921875, -19.821762084960938, 21.109336853027344, 33.84814453125, 35.73931884765625, 115.3040771484375, 1.4542694091796875, 142.08926391601562, 90.1229248046875, -7.399627685546875, 96.33030700683594, 61.526458740234375, 63.28167724609375, -3.9456634521484375, -3.4211273193359375, -1.096207618713379, 77.3958740234375, 126.53060913085938, -51.58599853515625, 81.60892486572266, 27.4119873046875, -8.909467697143555, -25.660919189453125, 92.8647689819336, 102.76583862304688, -7.404052734375, -0.6841583251953125, 90.432373046875, 32.35797119140625, 114.33279418945312, -33.51165771484375, 40.58942794799805, 9.212303161621094, 19.336517333984375, 10.451725006103516, -5.5030670166015625, 55.32012939453125, 92.69754028320312, 108.82695007324219, 9.124752044677734, -14.507476806640625, 7.8507537841796875, -4.3962249755859375, 72.33958435058594, -2.8030853271484375, 152.74966430664062, -1.5482940673828125, 105.23765563964844, 126.10545349121094, 117.62158203125, 98.57722473144531, 42.156158447265625, -16.871917724609375, 145.33230590820312, 84.48812866210938, 130.38287353515625, 128.4708251953125, 76.24116516113281, 11.691864013671875, 68.10975646972656, 88.06504821777344, -13.1312255859375, 89.09466552734375, 80.11289978027344, 136.88735961914062, 3.25848388671875, 26.402069091796875, 22.767578125, -63.61033630371094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000258.npy"} {"epoch": 0.5403141361256545, "step": 259, "batch_size": 128, "mean": 39.34259796142578, "std": 59.88593673706055, "min": -112.82418823242188, "p10": -30.15855865478515, "median": 39.00863075256348, "p90": 117.24154510498046, "max": 157.26373291015625, "pos_frac": 0.71875, "sample": [93.08218383789062, -48.86894226074219, 122.48721313476562, 9.636577606201172, -28.584625244140625, 10.630630493164062, 14.185699462890625, 134.9810028076172, 114.02203369140625, -10.80877685546875, 55.599609375, 15.5157470703125, 45.50785827636719, 99.55477905273438, 20.77386474609375, 68.09075927734375, 97.4705810546875, -103.14649963378906, 76.24935913085938, 108.91256713867188, 99.38388061523438, 142.55224609375, -59.264984130859375, 33.73369216918945, 129.60614013671875, 22.868789672851562, 72.56456756591797, 85.13141632080078, -62.115570068359375, -8.131523132324219, 6.851837158203125, 33.164215087890625, -19.582855224609375, 90.42547607421875, 50.47552490234375, 44.2835693359375, 64.62310791015625, 13.3675537109375, 17.17938232421875, 78.39010620117188, -38.4483642578125, 20.78704833984375, 115.86270141601562, 9.330177307128906, 2.3524169921875, -56.50244140625, 74.56890106201172, 130.0888671875, -2.3389053344726562, 62.343231201171875, 86.23655700683594, 68.0660400390625, 114.3121337890625, 8.575927734375, 14.908241271972656, 13.363059997558594, -0.236297607421875, 132.68585205078125, 80.62530517578125, 15.227134704589844, 157.26373291015625, 76.6526870727539, 9.049652099609375, 1.9858360290527344, -6.522705078125, -92.60132598876953, 48.530303955078125, -33.83106994628906, -36.27183532714844, 129.08651733398438, 73.38211059570312, -8.45062255859375, -4.95709228515625, 62.275604248046875, 120.45884704589844, -7.401763916015625, 46.30140686035156, 0.0, -9.92645263671875, 111.66830444335938, 104.24192810058594, 104.71954345703125, -18.52338409423828, 15.00628662109375, 49.9638671875, -2.3277435302734375, 126.5068359375, -84.94593048095703, -1.7278366088867188, 48.799102783203125, -112.82418823242188, -13.760894775390625, 122.82145690917969, 136.22415161132812, 24.969329833984375, 74.2547607421875, 69.9341812133789, -20.21624755859375, 102.57132720947266, -11.66156005859375, 47.27275085449219, 61.916748046875, -15.447769165039062, 112.48565673828125, -21.76019287109375, 56.835174560546875, 109.76065826416016, -68.93133544921875, 55.2196044921875, 112.99072265625, 31.063568115234375, 111.5279769897461, 12.333892822265625, -24.125946044921875, 9.434063911437988, 18.116073608398438, 132.8064727783203, 3.58258056640625, 55.50001525878906, -18.35174560546875, 97.14094543457031, -76.04446411132812, -24.951431274414062, 62.50177001953125, 3.51019287109375, 73.95651245117188, 99.66993713378906, 84.51698303222656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000259.npy"} {"epoch": 0.5424083769633508, "step": 260, "batch_size": 128, "mean": 18.143917083740234, "std": 59.92117691040039, "min": -144.1800537109375, "p10": -52.144554901123044, "median": 12.615232467651367, "p90": 106.31556930541991, "max": 147.2276611328125, "pos_frac": 0.640625, "sample": [121.76187133789062, 4.4888763427734375, 90.16753387451172, 8.844161987304688, -116.093017578125, 93.37246704101562, 45.22113037109375, 119.05401611328125, 21.28765869140625, 31.806610107421875, -17.0491943359375, 18.76214599609375, 0.0, -18.112213134765625, 17.836883544921875, 4.036821365356445, 2.8033447265625, 147.2276611328125, 32.04150390625, 29.224842071533203, 98.60269165039062, -144.1800537109375, -4.150123596191406, 13.46209716796875, 140.58163452148438, -51.36114501953125, 14.656368255615234, 88.24472045898438, 31.937957763671875, 16.23406982421875, 32.984405517578125, -50.29917907714844, 15.356491088867188, 4.044523239135742, 5.178182601928711, 138.83786010742188, -38.124786376953125, -66.09249877929688, -13.437240600585938, 14.905105590820312, 2.125885009765625, 24.24713134765625, 32.50213623046875, -36.18096923828125, -27.67169189453125, 94.37214660644531, -77.4659423828125, -15.03017807006836, -59.826934814453125, 12.683895111083984, 12.54656982421875, -57.64495849609375, 112.39154052734375, 114.347412109375, -12.28696060180664, 10.975372314453125, 113.46127319335938, -99.03280639648438, 99.51235961914062, 83.18107604980469, -27.85467529296875, -3.471599578857422, 105.07530212402344, -7.36639404296875, 6.074810028076172, -75.35598754882812, 61.68747329711914, 103.468994140625, -4.7091064453125, 119.25234985351562, 36.4769287109375, 61.419158935546875, -32.31944274902344, 9.78863525390625, 1.3708114624023438, -19.732833862304688, -16.067092895507812, 33.52897644042969, 13.302825927734375, 14.523193359375, -86.80120849609375, -37.549468994140625, 134.6844482421875, 0.342254638671875, 0.7958984375, 31.92681884765625, 23.91259765625, -19.857009887695312, -10.246444702148438, -105.50387573242188, -27.33428955078125, 2.345367431640625, -101.81361389160156, 129.70266723632812, 23.881790161132812, 13.977935791015625, -7.380771636962891, 45.97993469238281, -25.750701904296875, 110.09982299804688, 71.54800415039062, 59.669921875, 20.713394165039062, 49.148162841796875, 58.4276123046875, 4.7298583984375, -6.6606597900390625, 109.20952606201172, -53.972511291503906, 73.93328857421875, -11.473541259765625, 56.925384521484375, 101.45556640625, 13.27239990234375, -46.54522705078125, 2.0768890380859375, -19.74560546875, -109.6785888671875, -12.292999267578125, 33.186431884765625, 91.6468505859375, -42.4921875, 30.583526611328125, 9.40437126159668, 88.6602554321289, -3.4547500610351562, 86.21660614013672, -17.845672607421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000260.npy"} {"epoch": 0.5445026178010471, "step": 261, "batch_size": 128, "mean": 33.958717346191406, "std": 64.58592987060547, "min": -114.5362548828125, "p10": -54.04058761596679, "median": 31.08465576171875, "p90": 115.8014175415039, "max": 186.5089111328125, "pos_frac": 0.7265625, "sample": [-11.903549194335938, -114.5362548828125, 65.88154602050781, -27.584487915039062, 32.724700927734375, 18.8277587890625, 93.86505126953125, -19.94866943359375, 7.846405029296875, 108.55087280273438, 125.05929565429688, 0.0, -18.28702735900879, -25.040374755859375, 1.0717430114746094, 39.911712646484375, -20.23309326171875, 40.461517333984375, 115.33345031738281, 35.29962158203125, 71.483154296875, -84.96846008300781, 106.7649154663086, -95.8892822265625, 78.67396545410156, -53.72325897216797, -9.255096435546875, 116.89334106445312, -35.676841735839844, -89.12220764160156, 84.72314453125, 111.56312561035156, 117.4920654296875, 76.51422119140625, 125.06166076660156, 54.085418701171875, -14.607421875, 15.462276458740234, -48.029754638671875, 74.8512191772461, 13.804740905761719, 11.97418212890625, 127.3505859375, -24.57989501953125, 10.407867431640625, 15.742767333984375, -0.5764617919921875, 113.64360809326172, -84.49047088623047, 91.39201354980469, 22.360107421875, -25.433135986328125, 40.634765625, 135.95550537109375, 111.80615234375, 50.246917724609375, 43.955894470214844, -39.67445373535156, 89.732421875, 111.15771484375, 110.1318359375, 115.00161743164062, 78.83682250976562, 14.041702270507812, 80.9227294921875, -0.6962013244628906, 63.10999298095703, -54.78102111816406, 25.744720458984375, -1.29473876953125, 89.2413330078125, 186.5089111328125, 3.972442626953125, 143.27603149414062, 42.30055236816406, 7.659027099609375, 8.90582275390625, 3.42828369140625, 3.866424560546875, 9.881362915039062, 29.444610595703125, -6.227142333984375, 104.10212707519531, -64.03289794921875, 144.2056884765625, 49.69647216796875, 34.896209716796875, 70.6737060546875, 18.311424255371094, -88.07000732421875, 16.99706268310547, 17.751861572265625, -90.2376708984375, 1.948211669921875, -99.2601318359375, 125.78376770019531, 105.03536987304688, 106.316162109375, 111.32490539550781, 57.81132507324219, -69.33438110351562, 54.07062530517578, 0.43122100830078125, 86.247314453125, 20.570144653320312, 49.579345703125, -65.82734680175781, 146.37149047851562, 120.99957275390625, -25.68225860595703, 72.62179565429688, -6.366050720214844, 44.18047332763672, -113.39579772949219, 46.66510009765625, 10.029273986816406, 12.130523681640625, 51.04034423828125, 85.99807739257812, 46.475502014160156, 71.71043395996094, 146.55255126953125, 11.124465942382812, -31.97576904296875, 1.3902454376220703, 46.305145263671875, 6.0369415283203125, 87.23065948486328], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000261.npy"} {"epoch": 0.5465968586387434, "step": 262, "batch_size": 128, "mean": 42.56793212890625, "std": 58.155364990234375, "min": -105.40226745605469, "p10": -25.532705688476558, "median": 30.031585693359375, "p90": 121.05472717285156, "max": 155.51431274414062, "pos_frac": 0.75, "sample": [128.57415771484375, 28.08636474609375, 66.37686157226562, 27.208847045898438, 93.71688842773438, 3.7540817260742188, 148.35430908203125, 124.37255859375, 25.8134765625, 132.74624633789062, 153.30731201171875, -20.90252685546875, 40.19207000732422, 12.640090942382812, 14.682098388671875, 4.972698211669922, 99.77074432373047, 31.342437744140625, -0.5833282470703125, 82.66580200195312, 117.19981384277344, -7.7882537841796875, -28.603038787841797, 113.33114624023438, 70.18801879882812, 134.596435546875, 18.038841247558594, 26.505584716796875, 100.2811279296875, 135.29608154296875, 25.05218505859375, -105.40226745605469, 30.19097900390625, -15.543014526367188, -1.905487060546875, -95.57908630371094, -29.450897216796875, 42.337158203125, 3.042633056640625, 110.5189208984375, 28.628021240234375, -5.983102798461914, -9.827301025390625, -28.115814208984375, 36.65571594238281, 57.74187469482422, 74.34988403320312, -8.93414306640625, 35.99884033203125, -19.507110595703125, 60.762054443359375, 155.51431274414062, 105.83917236328125, 81.46051025390625, -28.48980712890625, 0.20374679565429688, 75.59707641601562, 39.4776611328125, -24.4256591796875, 16.39892578125, 73.39419555664062, -18.2137451171875, -9.351913452148438, 70.89459228515625, -5.321075439453125, 15.21807861328125, 126.51095581054688, 12.4901123046875, 117.90084838867188, 33.54270935058594, -66.31538391113281, 0.071624755859375, 124.00222778320312, 0.794677734375, 62.43342590332031, -35.666107177734375, -29.338340759277344, 117.17330932617188, 2.762451171875, -31.85199737548828, -95.02113342285156, -10.76213264465332, 0.0, 112.3965835571289, 147.23330688476562, 39.88206481933594, 14.910232543945312, 93.07340240478516, 14.280906677246094, 17.696975708007812, 102.83990478515625, 110.216552734375, 106.15632629394531, 21.20379638671875, 122.16195678710938, 75.632568359375, -5.434173583984375, 65.54312133789062, 55.719085693359375, 87.45172119140625, 3.311124801635742, -4.640472412109375, 57.437255859375, 109.585205078125, -31.04888916015625, 12.447608947753906, 29.8721923828125, 27.98370361328125, 120.5802001953125, 83.48382568359375, -3.810588836669922, 19.55127716064453, 141.94137573242188, 84.26193237304688, 101.67436218261719, 21.23974609375, 119.70828247070312, 68.37689208984375, 114.047607421875, 105.24238586425781, 25.07672119140625, -67.20740509033203, 37.9000244140625, 74.87619018554688, 10.242279052734375, 85.95289611816406, 5.7614593505859375, -24.20654296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000262.npy"} {"epoch": 0.5486910994764398, "step": 263, "batch_size": 128, "mean": 35.809974670410156, "std": 62.19670486450195, "min": -135.46563720703125, "p10": -26.75738525390625, "median": 19.516090393066406, "p90": 123.2640869140625, "max": 170.76144409179688, "pos_frac": 0.6953125, "sample": [127.34274291992188, 61.55769348144531, 121.89404296875, 26.71484375, -5.542533874511719, 81.34249877929688, -27.030181884765625, 5.41986083984375, 78.01615142822266, 59.42828369140625, -15.506402969360352, 88.19844055175781, 28.153167724609375, 25.49560546875, -43.30633544921875, 3.754955291748047, 0.2829093933105469, -70.45831298828125, 52.237884521484375, 0.0758056640625, 56.725311279296875, -11.2147216796875, 1.2065277099609375, 82.64256286621094, -10.140472412109375, 117.30349731445312, 42.25360107421875, 69.0967025756836, -1.4599609375, -26.2825927734375, -19.346641540527344, 46.166900634765625, 128.72076416015625, 120.44467163085938, 135.95654296875, -12.476333618164062, -8.472747802734375, -15.832611083984375, -46.83494567871094, 119.75285339355469, 10.6915283203125, -5.36700439453125, -11.623825073242188, 137.04721069335938, 72.80242919921875, 13.294570922851562, 92.63641357421875, 2.431488037109375, 111.5223388671875, 22.48571014404297, 63.61846923828125, -26.640472412109375, -18.463668823242188, 87.8389892578125, 123.3134765625, 82.04571533203125, -101.40480041503906, 8.88836669921875, 110.16311645507812, 112.11030578613281, 0.0, 32.79241943359375, -135.46563720703125, 13.858333587646484, 63.556732177734375, 8.599456787109375, 0.474456787109375, 2.1459426879882812, 13.621307373046875, 36.177093505859375, 132.6559600830078, 4.1014404296875, -47.13177490234375, -49.142181396484375, 37.23229217529297, 11.2606201171875, 86.60552978515625, 65.82366943359375, 69.51116943359375, 2.5989532470703125, -4.573516845703125, 9.095062255859375, -50.20819091796875, 13.021942138671875, 21.976760864257812, 65.65606689453125, -86.66658020019531, 135.58651733398438, 127.72561645507812, 18.417274475097656, -20.4305419921875, 133.0523681640625, 11.857414245605469, -105.682861328125, -7.1279144287109375, -7.365203857421875, -88.98213195800781, 130.25189208984375, 20.09490966796875, -42.240570068359375, 96.08192443847656, -1.9006805419921875, 0.0, 9.571380615234375, 85.042724609375, 79.61764526367188, 170.76144409179688, 65.90486907958984, 166.61489868164062, -6.890083312988281, 43.41064453125, 8.991630554199219, 123.242919921875, 0.0, 98.36865234375, 99.15917205810547, 18.937271118164062, 28.128433227539062, 104.18413543701172, 3.9881362915039062, 116.35562896728516, 111.46498107910156, 52.72532653808594, -2.631683349609375, 167.366455078125, -9.72174072265625, 94.75177001953125, -16.2535400390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000263.npy"} {"epoch": 0.5507853403141362, "step": 264, "batch_size": 128, "mean": 44.558963775634766, "std": 59.259090423583984, "min": -118.67984008789062, "p10": -16.34728088378906, "median": 29.142242431640625, "p90": 129.46962890625, "max": 186.02374267578125, "pos_frac": 0.7578125, "sample": [-5.151267051696777, -11.686126708984375, 153.3260498046875, 65.28457641601562, 142.18017578125, 104.41541290283203, 79.07096862792969, 89.04449462890625, 8.527498245239258, 10.6280517578125, -2.9305686950683594, 133.59896850585938, 37.2559814453125, 82.12364196777344, -27.709716796875, 101.37919616699219, 74.35984802246094, 113.59504699707031, -7.524089813232422, 0.03148651123046875, 121.9730224609375, 108.0916748046875, -24.00970458984375, 78.56127166748047, -68.02095794677734, 155.79632568359375, 112.45181274414062, 0.8862705230712891, 59.3511962890625, 61.064910888671875, -30.009902954101562, 96.34353637695312, 121.90530395507812, 114.26200866699219, 141.16165161132812, 0.0, 3.46002197265625, 99.6197509765625, 66.30458068847656, 14.914505004882812, -22.324005126953125, 124.43814086914062, -32.6605224609375, 64.92144775390625, 77.58026885986328, 115.754638671875, -32.285308837890625, 145.131103515625, -62.05511474609375, 34.78900146484375, 2.057403564453125, 11.98046875, 120.33984375, 55.47544860839844, 106.85643005371094, -62.229888916015625, 119.28739929199219, 94.90185546875, 51.361568450927734, 18.983924865722656, 50.97467041015625, 3.3845787048339844, 52.133514404296875, 32.667640686035156, 15.491607666015625, -17.19171142578125, -5.71734619140625, 22.09307861328125, 19.318313598632812, 116.23904418945312, -0.7939605712890625, 13.66876220703125, -118.67984008789062, -13.411590576171875, 51.21433639526367, -2.9367218017578125, 17.205791473388672, 10.266952514648438, 131.32977294921875, 130.74737548828125, 96.62197875976562, 1.6444244384765625, 0.4715137481689453, 14.678390502929688, -2.52838134765625, 186.02374267578125, -6.278388977050781, 0.0, 28.17047119140625, 20.123268127441406, -44.572998046875, 33.34562683105469, 104.73090362548828, 129.40219116210938, 18.60247802734375, 99.85369873046875, -4.941783905029297, 41.943115234375, 87.09941101074219, 16.638107299804688, 129.62698364257812, 67.63433837890625, 6.0932769775390625, -15.985382080078125, 0.454925537109375, -1.971099853515625, 30.114013671875, 8.271392822265625, -15.73974609375, 98.64147186279297, 0.3723907470703125, 51.580963134765625, 45.197479248046875, 164.3666229248047, 24.6103515625, 141.01068115234375, -7.3570556640625, 21.580360412597656, 133.80206298828125, -6.139732360839844, -54.666419982910156, 12.970687866210938, 55.68010711669922, 11.901260375976562, 110.11758422851562, 1.9754486083984375, 39.06121826171875, 15.0799560546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000264.npy"} {"epoch": 0.5528795811518324, "step": 265, "batch_size": 128, "mean": 42.360435485839844, "std": 63.121673583984375, "min": -108.90484619140625, "p10": -38.82763671875, "median": 38.100921630859375, "p90": 127.32800903320312, "max": 177.4473876953125, "pos_frac": 0.734375, "sample": [4.523710250854492, 145.73126220703125, 115.56060791015625, -65.42718505859375, 44.621612548828125, 7.045654296875, 106.87718963623047, 79.56950378417969, 77.82366943359375, -75.10272216796875, 34.81602478027344, 41.630828857421875, 95.34855651855469, -13.07159423828125, 131.27032470703125, 99.83735656738281, 43.312713623046875, 129.45257568359375, 105.76351928710938, 86.56076049804688, 177.4473876953125, 136.62933349609375, 145.413330078125, 21.887725830078125, 158.35308837890625, 93.50320434570312, 69.34927368164062, 135.83050537109375, 129.53128051757812, -5.1658477783203125, -48.866065979003906, 95.77458190917969, 148.031494140625, 8.74237060546875, 114.13356018066406, 126.41748046875, 73.92242431640625, 105.04171752929688, 79.2005615234375, 125.60101318359375, 0.849090576171875, 0.733978271484375, -42.24755859375, 35.17152404785156, -9.217384338378906, -17.799514770507812, -19.9044189453125, -41.31475830078125, -3.4933319091796875, 94.22862243652344, 0.888763427734375, -16.852874755859375, -35.03288269042969, 2.3009033203125, -33.80023956298828, 92.51348876953125, 109.04656982421875, 19.555419921875, 75.65707397460938, -108.90484619140625, 22.696483612060547, 102.24368286132812, 5.3548583984375, 88.21867370605469, 37.48675537109375, -32.935546875, 123.22554016113281, 19.5198974609375, 82.42921447753906, -22.764297485351562, -41.193939208984375, 121.49984741210938, 110.47602844238281, 7.39727783203125, 0.0, -5.11872673034668, -89.62332153320312, -80.5992431640625, 86.45382690429688, 113.14042663574219, 38.715087890625, 5.105560302734375, 82.77188873291016, 0.509857177734375, 99.43801879882812, 27.443817138671875, 0.0, -6.235725402832031, -13.80340576171875, 68.0050048828125, 70.4915771484375, -31.1414794921875, 0.723968505859375, 0.22457122802734375, 60.7591552734375, 5.176277160644531, 51.62493896484375, 157.2470703125, 33.04803466796875, -42.430877685546875, 23.207496643066406, 54.895263671875, 74.61257934570312, 67.77044677734375, 61.78216552734375, 73.3189697265625, -75.40118408203125, 132.7757568359375, 103.66860961914062, -47.4635009765625, -6.102973937988281, 20.67633056640625, 9.9271240234375, -10.689468383789062, 62.153656005859375, -37.813507080078125, 0.7034378051757812, 132.02670288085938, 80.68496704101562, 88.29829406738281, 98.54949951171875, -11.3612060546875, 39.77912139892578, 28.3782958984375, 22.43467140197754, -78.4637451171875, 16.966293334960938, 51.940582275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000265.npy"} {"epoch": 0.5549738219895288, "step": 266, "batch_size": 128, "mean": 42.627052307128906, "std": 59.789161682128906, "min": -134.07598876953125, "p10": -11.16141242980957, "median": 33.24847412109375, "p90": 118.59085083007812, "max": 159.96810913085938, "pos_frac": 0.828125, "sample": [41.65887451171875, 93.20195770263672, 159.96810913085938, 21.301300048828125, -11.072479248046875, -12.695732116699219, 90.44625854492188, 4.34112548828125, -16.10674285888672, 13.966667175292969, 27.824310302734375, 0.804931640625, -3.4594573974609375, 52.01287841796875, 56.51289367675781, -9.664794921875, -9.077362060546875, 46.783843994140625, 28.372074127197266, -87.11956787109375, 114.77796936035156, 34.313751220703125, 4.2670135498046875, 31.463348388671875, 125.65780639648438, 34.13189697265625, -0.831298828125, 38.45587158203125, 135.58026123046875, 115.36532592773438, 1.8048973083496094, 152.222412109375, -7.189178466796875, 44.677276611328125, 5.990596771240234, 104.57183837890625, -42.98419189453125, 10.478702545166016, 74.4564208984375, 65.90365600585938, 2.2838821411132812, 38.5816650390625, 16.21875762939453, 4.7184600830078125, 113.35045623779297, 103.16619873046875, 85.10215759277344, 115.0396499633789, 16.04645538330078, 29.193191528320312, 13.294403076171875, 99.26080322265625, -107.13543701171875, -115.72917175292969, 0.31126976013183594, -74.76206970214844, 6.036760330200195, 40.16607666015625, 30.464569091796875, 27.008087158203125, 119.7379150390625, 89.67744445800781, 14.897056579589844, 16.181015014648438, 20.953994750976562, 96.28982543945312, 90.55189514160156, 111.30584716796875, 11.96331787109375, 31.40899658203125, 32.36505126953125, 135.6004180908203, 63.20989990234375, 40.6182861328125, 109.8019790649414, 159.78225708007812, 41.3829345703125, -2.951873779296875, -75.28451538085938, -93.93314361572266, -134.07598876953125, 135.1075897216797, 11.227691650390625, 116.67564392089844, -13.382568359375, 50.211822509765625, 40.388153076171875, 16.927886962890625, 124.58712768554688, 69.43899536132812, -72.330322265625, 5.9405517578125, -4.098907470703125, 25.69598388671875, 112.02853393554688, 56.715087890625, 95.93013000488281, 98.43603515625, 58.020263671875, 8.781951904296875, 20.29248046875, 50.784881591796875, 26.39947509765625, 141.28945922851562, 15.202802658081055, 10.356674194335938, -11.36892318725586, 2.9869842529296875, 37.938995361328125, -5.7167510986328125, 109.63235473632812, 129.94522094726562, 142.89207458496094, 9.831207275390625, 114.58537292480469, 13.9833984375, 100.3328857421875, 82.01907348632812, 12.7371826171875, 35.04986572265625, 118.39520263671875, 11.345947265625, 112.67274475097656, 10.934280395507812, 102.79754638671875, 100.02518463134766, 52.35595703125, 119.04736328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000266.npy"} {"epoch": 0.5570680628272251, "step": 267, "batch_size": 128, "mean": 35.53943634033203, "std": 67.58745574951172, "min": -133.79095458984375, "p10": -41.84925918579101, "median": 32.194549560546875, "p90": 117.16329650878906, "max": 201.6844482421875, "pos_frac": 0.7109375, "sample": [3.3854141235351562, 109.40179443359375, 96.18000793457031, 89.65093994140625, 158.91705322265625, 36.595794677734375, -0.559600830078125, 29.550506591796875, -70.2598876953125, 3.2821044921875, -0.2755546569824219, 27.817962646484375, -37.9193115234375, 67.16267395019531, 59.39850616455078, -24.539337158203125, 0.8107833862304688, 139.63134765625, 138.79458618164062, 34.55572509765625, -91.25335693359375, 167.31243896484375, 112.85067749023438, 3.8368377685546875, 38.11777114868164, -16.075653076171875, 61.48370361328125, 92.95256042480469, 74.12152099609375, -120.45843505859375, 71.7322998046875, 21.01116943359375, 27.278839111328125, -16.16412353515625, 14.771408081054688, 88.74822998046875, -9.141777038574219, 134.38027954101562, 117.33233642578125, 84.89305877685547, -80.36549377441406, 97.95513916015625, 63.64403533935547, 89.67510986328125, 14.038185119628906, 175.02838134765625, 49.1116943359375, -24.60552978515625, 119.69357299804688, 118.81916809082031, -99.60884857177734, -92.67739868164062, 81.33099365234375, 99.45349884033203, 10.069839477539062, 167.527587890625, 99.54505157470703, 91.22750854492188, 135.40353393554688, 11.722085952758789, 92.63359069824219, 78.79608154296875, -103.55767822265625, 12.364288330078125, 43.190185546875, 2.2213592529296875, 66.07635498046875, -61.14322280883789, 70.48104858398438, 103.33822631835938, 13.216033935546875, 37.398406982421875, 70.28515625, -133.79095458984375, 57.00666809082031, 81.15521240234375, 4.959625244140625, 62.760986328125, 4.863609313964844, -0.9254722595214844, -40.31896209716797, 71.03216552734375, -9.406654357910156, 42.97088623046875, -19.091018676757812, 28.633193969726562, -45.419952392578125, 59.41099548339844, 13.1309814453125, 87.88949584960938, -12.412361145019531, -39.71461486816406, 54.04154968261719, -20.4757080078125, 4.3692169189453125, 103.87240600585938, 47.386451721191406, 71.21609497070312, 50.68156433105469, -11.896148681640625, 114.7750244140625, 29.8333740234375, -26.98406982421875, 72.54949951171875, -11.5343017578125, 27.12811279296875, 26.204269409179688, 112.29617309570312, -5.713737487792969, 201.6844482421875, -23.932601928710938, -80.54393005371094, 93.77928924560547, 28.30950927734375, 89.21281433105469, 1.7928619384765625, 117.09085083007812, 105.69380187988281, -11.32110595703125, -19.8968505859375, 120.96998596191406, -131.1529998779297, -1.3480224609375, 0.6070919036865234, 48.323463439941406, -86.36712646484375, -34.43336486816406, 8.49664306640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000267.npy"} {"epoch": 0.5591623036649215, "step": 268, "batch_size": 128, "mean": 41.41619873046875, "std": 61.56035614013672, "min": -106.64271545410156, "p10": -11.36273193359375, "median": 28.576507568359375, "p90": 124.26514282226563, "max": 171.02435302734375, "pos_frac": 0.734375, "sample": [33.22526550292969, 10.116897583007812, 85.15943908691406, 108.0428466796875, 118.62527465820312, 7.73504638671875, -1.4687728881835938, -41.506629943847656, -53.97307586669922, -58.46209716796875, -1.722259521484375, 154.69723510742188, 20.59820556640625, 126.57337188720703, 1.014373779296875, 85.90237426757812, 171.02435302734375, -6.217329025268555, -1.4661865234375, 99.32098388671875, 109.46375274658203, 112.66749572753906, 32.571311950683594, -106.64271545410156, 72.97186279296875, 30.683517456054688, -1.9559001922607422, 39.68438720703125, 60.807373046875, 20.12509536743164, 8.828413009643555, 94.92898559570312, 97.43634033203125, 93.30995178222656, 33.76741027832031, 0.6400222778320312, -106.14860534667969, -75.65042114257812, -26.541152954101562, 167.35980224609375, 103.47378540039062, 30.572509765625, 26.5076904296875, 124.412841796875, -1.312255859375, 13.354721069335938, 65.73971557617188, -98.294921875, -44.699951171875, -7.464313507080078, 28.79754638671875, 87.71966552734375, 54.586273193359375, 121.67642211914062, 4.600788116455078, 39.25006103515625, 148.7283935546875, -7.3244476318359375, 76.3858642578125, 0.86163330078125, 159.1142578125, 12.83062744140625, 115.38172149658203, -4.39996337890625, -11.125244140625, 19.540420532226562, 125.84091186523438, 37.87409973144531, -3.9372100830078125, -10.27545166015625, 94.99000549316406, 17.665267944335938, 109.24957275390625, 109.0916748046875, 19.070587158203125, -9.393501281738281, 44.8172607421875, 135.88665771484375, 124.20184326171875, 48.66912841796875, 74.70382690429688, 74.08741760253906, 69.3128662109375, -5.829002380371094, 79.64637756347656, 108.57090759277344, -67.39326477050781, 48.714141845703125, 47.66532897949219, 8.233978271484375, 28.35546875, 113.67997741699219, 143.7047119140625, 74.28955078125, 1.5040283203125, 120.13037109375, 19.717788696289062, 5.355712890625, 22.289703369140625, 42.6453857421875, -9.48675537109375, 26.192543029785156, 56.7947998046875, -3.05706787109375, 5.550130844116211, 107.79624938964844, 12.0548095703125, -11.9168701171875, -0.69122314453125, 34.32008361816406, 158.55258178710938, 11.25344467163086, 94.69509887695312, 121.74295043945312, -1.634765625, 140.5252685546875, -78.37467956542969, -9.777694702148438, 164.99359130859375, 13.468597412109375, 0.43023109436035156, -38.489837646484375, 19.92913055419922, 37.16313171386719, 10.719970703125, 18.150123596191406, -7.206634521484375, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000268.npy"} {"epoch": 0.5612565445026177, "step": 269, "batch_size": 128, "mean": 37.16244125366211, "std": 66.26826477050781, "min": -177.79141235351562, "p10": -28.54334487915039, "median": 29.5723876953125, "p90": 120.01938323974609, "max": 167.99365234375, "pos_frac": 0.6953125, "sample": [-99.93695068359375, 50.6142578125, 106.76956176757812, -61.34516906738281, 52.05055236816406, -10.316131591796875, 13.352432250976562, 63.1473388671875, 9.36993408203125, -53.380615234375, 95.67131805419922, -3.8895339965820312, -8.089920043945312, 79.63992309570312, -30.1712646484375, 109.18655395507812, 5.99755859375, -4.08697509765625, 117.2457046508789, 45.133995056152344, 167.99365234375, 150.071044921875, 25.876556396484375, 137.4012451171875, 151.36294555664062, -21.5887451171875, 29.725189208984375, 113.10673522949219, -11.6253662109375, 68.8968505859375, 3.6524429321289062, 78.20927429199219, 106.30322265625, -82.4195785522461, 101.41571044921875, 0.0, 33.0955810546875, -27.845664978027344, 81.41827392578125, 100.00750732421875, -72.45313262939453, 19.938812255859375, 132.803466796875, 60.00559997558594, 6.293952941894531, 82.3744888305664, 136.48681640625, 0.0, 89.13763427734375, 89.69351196289062, 100.11567687988281, -19.89617919921875, 120.00285339355469, -113.7884521484375, -78.09698486328125, 118.12138366699219, 94.16907501220703, -15.846382141113281, -64.82766723632812, -0.5749149322509766, 118.763427734375, 63.75358581542969, 50.116943359375, 9.5125732421875, 99.83869934082031, -22.24627685546875, 13.76666259765625, 97.06057739257812, 112.18161010742188, 14.677558898925781, 51.272613525390625, 99.95355224609375, 0.0, 129.91966247558594, 15.646675109863281, 20.15966796875, -128.11679077148438, 3.184894561767578, 46.80316162109375, 5.118232727050781, -5.70269775390625, -20.62420654296875, 128.55560302734375, 20.853836059570312, 91.59004211425781, 29.419586181640625, 29.132568359375, 56.470428466796875, 33.124595642089844, 125.05696105957031, 47.611297607421875, 36.964019775390625, 13.696990966796875, -20.33648681640625, 114.13140869140625, 33.107086181640625, -16.226486206054688, 110.40153503417969, 6.147117614746094, 88.38903045654297, 65.6676025390625, 26.74219512939453, 3.2255172729492188, -18.537826538085938, 3.92437744140625, -100.1148681640625, 83.74735260009766, 129.00140380859375, -0.3700408935546875, 89.12287139892578, -25.404296875, 13.889381408691406, 85.27053833007812, -0.803863525390625, 101.16714477539062, -82.99668884277344, -27.1019287109375, -5.177467346191406, 25.302261352539062, -1.3623580932617188, 30.58721923828125, 127.93574523925781, 64.29127502441406, -177.79141235351562, -17.466888427734375, 28.444473266601562, 120.05795288085938, 146.73458862304688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000269.npy"} {"epoch": 0.5633507853403141, "step": 270, "batch_size": 128, "mean": 45.71007537841797, "std": 69.5350341796875, "min": -167.22708129882812, "p10": -36.75619812011719, "median": 39.30906677246094, "p90": 132.28585968017578, "max": 201.53021240234375, "pos_frac": 0.7421875, "sample": [-133.65899658203125, 69.87368774414062, 110.314453125, 104.15668487548828, -29.21551513671875, 143.897216796875, -9.97031021118164, 28.119110107421875, 75.33393859863281, -49.017730712890625, 118.29889678955078, 22.307952880859375, 19.08642578125, 42.37408447265625, 22.830001831054688, 107.93182373046875, -13.827911376953125, 132.64418029785156, 76.1343765258789, -61.41145324707031, -36.856964111328125, 36.971900939941406, 1.3565826416015625, 40.2442626953125, 10.812080383300781, -0.5662841796875, 112.6365966796875, 108.62140655517578, -17.509246826171875, 1.546865463256836, 38.373870849609375, 119.36656188964844, 51.224517822265625, 2.4184837341308594, 7.913200378417969, 9.140838623046875, -45.46502685546875, 22.93182373046875, -136.77587890625, 49.373992919921875, -41.162567138671875, 132.98492431640625, 28.79241943359375, 13.991806030273438, 116.74627685546875, 0.0855712890625, -4.57368278503418, 131.6219482421875, 87.85549926757812, -79.83502197265625, -62.41534423828125, 121.63882446289062, 123.52984619140625, 127.87461853027344, 45.11194610595703, 100.36698913574219, 138.0584716796875, -34.88189697265625, 118.61245727539062, 7.8448486328125, 29.3448486328125, -18.637115478515625, 11.001800537109375, 23.322376251220703, 123.73516845703125, 113.67568969726562, 136.27627563476562, 4.346405029296875, 119.281494140625, 37.49418640136719, 90.92330932617188, -1.30877685546875, -127.3170166015625, 115.97998809814453, -46.206268310546875, 132.13229370117188, 59.92997741699219, 50.535465240478516, 99.66314697265625, 32.164703369140625, -16.271575927734375, 11.3719482421875, 59.83154296875, 116.00634765625, 22.706096649169922, -5.0245361328125, -1.232269287109375, 11.669342041015625, 1.3956832885742188, -4.063072204589844, -36.7130126953125, -45.65545654296875, 91.316650390625, 149.2335205078125, -7.75750732421875, 31.432941436767578, 90.192138671875, 133.11669921875, 56.1611328125, -14.490478515625, 113.76268005371094, 145.2799072265625, 24.057403564453125, 88.7424545288086, 44.604976654052734, -167.22708129882812, -14.009552001953125, -7.12896728515625, 139.81964111328125, 9.60693359375, 40.723846435546875, 107.82611083984375, 66.91615295410156, 104.38992309570312, 130.27493286132812, 140.2130126953125, 134.64700317382812, 57.121421813964844, 9.599029541015625, 142.92172241210938, -19.5257568359375, 201.53021240234375, 120.48602294921875, 121.5673828125, 92.3374252319336, 104.9381103515625, 92.15582275390625, -28.51318359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000270.npy"} {"epoch": 0.5654450261780105, "step": 271, "batch_size": 128, "mean": 42.33795928955078, "std": 60.90958023071289, "min": -92.81884002685547, "p10": -19.97142639160156, "median": 28.418777465820312, "p90": 117.24331665039062, "max": 200.6953125, "pos_frac": 0.7578125, "sample": [102.1678237915039, -92.81884002685547, -8.945648193359375, 30.8389892578125, 14.599578857421875, 0.571136474609375, 7.250465393066406, -9.99173355102539, 33.21966552734375, -5.633289337158203, 2.5703125, 28.573654174804688, 117.21694946289062, 0.4928741455078125, 32.57441711425781, 40.366943359375, 83.718994140625, 24.588516235351562, 30.278244018554688, 13.705070495605469, 51.49270248413086, 10.2225341796875, 114.199951171875, 102.06682586669922, 2.3722763061523438, 3.2326202392578125, -36.03656005859375, 104.66806030273438, 18.653942108154297, -19.271514892578125, 105.00704956054688, 135.68502807617188, -3.64715576171875, 43.3836669921875, 1.51513671875, 27.26141357421875, 28.485809326171875, 20.222373962402344, 86.337890625, -70.36660766601562, 109.32516479492188, 47.61506652832031, 0.032745361328125, 116.12216186523438, 100.74771881103516, 18.6197509765625, 46.8472900390625, -0.094757080078125, 106.86418151855469, 12.497650146484375, -4.053680419921875, 78.09098815917969, 126.85298156738281, 61.87261962890625, 200.6953125, 148.25811767578125, 0.8470096588134766, 128.16778564453125, 28.35174560546875, 107.76448059082031, 95.59555053710938, 9.27508544921875, -9.047386169433594, 180.69375610351562, 101.09402465820312, -69.38558959960938, -21.60455322265625, 125.67269897460938, -7.699676513671875, -1.728302001953125, 116.18624877929688, 163.15744018554688, -27.7579345703125, 77.13644409179688, -13.434326171875, 21.10919189453125, -54.32179260253906, 29.291748046875, 38.52532958984375, 87.18075561523438, -24.53369140625, -77.7017822265625, 117.30484008789062, -10.195655822753906, 158.80145263671875, 95.38247680664062, -4.15667724609375, -41.829864501953125, 32.6357307434082, 20.275390625, 6.085086822509766, 129.98382568359375, 102.87797546386719, -12.722564697265625, 3.177734375, 99.60762023925781, 3.0772342681884766, 99.61750030517578, 115.55093383789062, 33.41016387939453, 9.836013793945312, 22.359542846679688, -7.83624267578125, 96.85614013671875, 146.43972778320312, -5.5112457275390625, -34.53533935546875, 16.7265625, 19.619232177734375, 65.18310546875, 33.867919921875, 105.22140502929688, 106.86709594726562, 10.77838134765625, 40.626068115234375, 14.223907470703125, 110.45266723632812, 112.2586669921875, -1.0758209228515625, -10.284210205078125, 114.07032775878906, -86.82272338867188, 115.46511840820312, 142.8719482421875, 20.481246948242188, 0.459228515625, 36.93878173828125, -69.11952209472656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000271.npy"} {"epoch": 0.5675392670157068, "step": 272, "batch_size": 128, "mean": 38.38246154785156, "std": 63.34762191772461, "min": -127.3031005859375, "p10": -25.399012756347656, "median": 28.419662475585938, "p90": 126.02685546875, "max": 181.19442749023438, "pos_frac": 0.75, "sample": [47.570159912109375, 0.0, -25.728302001953125, 63.53523254394531, -88.05410766601562, 52.87158203125, 154.22711181640625, 55.568023681640625, 114.6531982421875, -22.34320068359375, -25.257888793945312, 21.825166702270508, 48.8214111328125, 111.5643310546875, 54.693603515625, 143.06082153320312, 139.37725830078125, 107.54769897460938, 31.6395263671875, 13.600814819335938, 110.15150451660156, 10.299758911132812, -9.45458984375, -5.775909423828125, 106.49710083007812, -16.91907501220703, 5.06620979309082, 157.27557373046875, 1.0063114166259766, 11.6011962890625, 110.410400390625, 50.797401428222656, 66.85403442382812, 2.051013946533203, 128.34237670898438, 82.56793212890625, 150.71685791015625, 124.4576416015625, 12.586639404296875, -0.7962188720703125, 77.95510864257812, -41.8099365234375, -57.48193359375, 66.37625122070312, -10.700225830078125, 32.76365661621094, 106.21209716796875, -46.0732421875, -1.8967514038085938, 1.7165184020996094, 133.89208984375, 107.1851806640625, -14.983613967895508, 20.03734588623047, 74.87141418457031, 142.71844482421875, 56.108154296875, 44.530853271484375, 94.60372924804688, 0.4365272521972656, -2.006929397583008, 9.190078735351562, 10.278549194335938, 16.714035034179688, 89.25117492675781, 44.34027099609375, 181.19442749023438, 29.296112060546875, 125.89208984375, 11.57574462890625, 120.77484130859375, -43.94041442871094, 79.5341796875, 10.4326171875, -123.72332763671875, -49.027862548828125, -7.9439697265625, 38.76200866699219, 5.189544677734375, 96.12664794921875, 83.54310607910156, 48.775238037109375, -9.431915283203125, 150.0313262939453, 3.67889404296875, -127.3031005859375, 23.650346755981445, 13.595169067382812, 30.468414306640625, 2.589019775390625, 39.70269775390625, 85.37760925292969, 6.0840606689453125, -69.25369262695312, 86.48599243164062, 56.6588134765625, 50.55596923828125, 105.49337005615234, -17.09442138671875, 3.6900100708007812, 0.0, -24.169403076171875, 11.040616989135742, -98.90924072265625, -22.301666259765625, 17.1640625, 28.697540283203125, 1.3908157348632812, -3.1715240478515625, 9.8385009765625, 88.10054016113281, 179.75152587890625, -88.7000732421875, 57.809226989746094, 113.9368896484375, -81.71347045898438, 11.758247375488281, 2.8631458282470703, 7.169748306274414, -8.27752685546875, 131.8388671875, 60.49579620361328, 27.333587646484375, 28.14178466796875, 103.63653564453125, 126.34130859375, 59.1072998046875, 81.20487976074219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000272.npy"} {"epoch": 0.5696335078534032, "step": 273, "batch_size": 128, "mean": 43.01234436035156, "std": 67.04242706298828, "min": -134.63165283203125, "p10": -38.43522033691406, "median": 34.04368591308594, "p90": 128.81041259765624, "max": 194.8330078125, "pos_frac": 0.6796875, "sample": [70.51681518554688, 154.96612548828125, 143.31103515625, 10.578054428100586, 152.99594116210938, -10.825180053710938, 173.7872314453125, 23.4974365234375, 38.46482849121094, 7.745979309082031, -24.5311279296875, 17.985435485839844, -0.045886993408203125, 89.75506591796875, 194.8330078125, -83.40139770507812, 64.51568603515625, 165.4249267578125, 15.2939453125, 62.78753662109375, -73.32766723632812, 88.87145233154297, 115.13890075683594, 27.543170928955078, 134.88424682617188, 25.594024658203125, -107.1201171875, 33.202789306640625, 108.7852783203125, -12.331787109375, 18.026630401611328, -36.473236083984375, -19.871803283691406, 114.71017456054688, -21.615158081054688, -0.740814208984375, 10.09832763671875, -2.25982666015625, -6.1199798583984375, 117.06564331054688, -6.039867401123047, -10.28338623046875, 1.3424758911132812, 46.65985107421875, 51.43925476074219, 24.13983154296875, 97.64459991455078, -0.28172874450683594, -56.736106872558594, 5.873809814453125, 63.9827880859375, -24.14508056640625, 139.384521484375, -19.791046142578125, 65.68794250488281, -6.862335205078125, -0.334014892578125, 66.54513549804688, 14.184829711914062, 93.83769989013672, -43.01318359375, 107.23053741455078, 5.0009613037109375, 97.19337463378906, -19.332366943359375, 140.67236328125, 91.49494934082031, 1.0901336669921875, 96.11143493652344, 0.8082275390625, 112.52206420898438, -1.9162750244140625, -48.7265625, -14.659774780273438, 55.26043701171875, 124.37948608398438, 34.88458251953125, 119.73165893554688, 107.01033020019531, -62.6224365234375, 96.05960083007812, 108.19307708740234, 28.238922119140625, -51.30469512939453, 57.6334228515625, 106.4208755493164, 28.340972900390625, 43.559532165527344, 133.43301391601562, 46.38609313964844, 42.653656005859375, -55.577239990234375, -6.0432281494140625, 110.38784790039062, -134.63165283203125, 32.46044921875, -94.98162841796875, 113.56355285644531, 128.50680541992188, -3.2321319580078125, 144.18612670898438, 92.23931884765625, 123.51641845703125, 121.78054809570312, 122.068603515625, -5.323212623596191, -63.115570068359375, 19.3841552734375, 115.40298461914062, 98.63335418701172, 55.70928955078125, 52.62347412109375, -2.7235565185546875, 36.095855712890625, 22.045982360839844, 59.692535400390625, -48.543548583984375, 135.2698974609375, 24.740936279296875, -6.727165222167969, -11.104846954345703, 125.21563720703125, 45.76556396484375, 108.30780792236328, 129.51882934570312, -33.159759521484375, 109.627685546875, -26.699249267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000273.npy"} {"epoch": 0.5717277486910994, "step": 274, "batch_size": 128, "mean": 45.82276916503906, "std": 65.53291320800781, "min": -135.07403564453125, "p10": -17.040953063964842, "median": 28.95368194580078, "p90": 124.1667594909668, "max": 192.253662109375, "pos_frac": 0.7890625, "sample": [115.11388397216797, 115.4892578125, -45.74822998046875, 80.54772186279297, -2.9148025512695312, 62.064208984375, 62.8009033203125, 40.66346740722656, 98.8327407836914, 9.972426414489746, 26.46978759765625, 105.94037628173828, 160.13795471191406, 24.456619262695312, 5.972633361816406, 98.80284881591797, 168.8514404296875, 142.7451629638672, -76.78997802734375, -12.523643493652344, 23.5186767578125, 138.02578735351562, 117.50991821289062, 40.275177001953125, -4.39599609375, -4.40283203125, -91.64151763916016, -26.413848876953125, 12.835622787475586, 29.982879638671875, 18.83984375, 26.887481689453125, 101.03028869628906, 45.68951416015625, 113.61846923828125, 47.42633056640625, 41.55552673339844, 106.24305725097656, 13.604782104492188, 77.3126220703125, 140.71951293945312, -13.441856384277344, -7.9140167236328125, 5.922523498535156, -16.714813232421875, 120.08982849121094, 19.31903076171875, 22.17047882080078, 66.03030395507812, 146.56060791015625, -10.309707641601562, 8.270965576171875, 109.191650390625, 62.02813720703125, -120.11744689941406, 27.924484252929688, -17.197463989257812, -34.487457275390625, 1.0980377197265625, -18.499755859375, 19.122838973999023, 5.062685012817383, -10.997352600097656, 85.66954040527344, -68.40554809570312, -1.6899299621582031, 119.35322570800781, 0.10106658935546875, 77.849853515625, 113.2796630859375, 13.551132202148438, -135.07403564453125, 134.1614990234375, 37.63829040527344, 60.0972900390625, 52.201995849609375, 3.602874755859375, 48.84333801269531, 192.253662109375, 8.136100769042969, 175.10052490234375, 99.89726257324219, 23.00360107421875, 119.82440185546875, -1.555816650390625, 60.65338134765625, 109.00018310546875, 125.46331787109375, 71.9473876953125, 123.61109161376953, 110.88108825683594, 24.705169677734375, -44.29937744140625, 83.94822692871094, 115.85293579101562, 102.35009002685547, 9.78857421875, 130.62652587890625, 147.4542236328125, 24.977752685546875, -86.51004028320312, 16.9473876953125, 9.7626953125, -14.972686767578125, 42.67686462402344, 109.41291809082031, 8.285545349121094, 122.18162536621094, 14.3660888671875, 52.35205078125, 122.63522338867188, 102.65604400634766, 4.0123291015625, 26.444961547851562, 13.470184326171875, -16.973876953125, 6.1407470703125, 3.098785400390625, 14.74163818359375, 51.77703857421875, 11.879650115966797, -133.1436767578125, 112.99395751953125, -1.2428474426269531, 23.703998565673828, 101.08905029296875, 113.2344970703125, 167.2762451171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000274.npy"} {"epoch": 0.5738219895287958, "step": 275, "batch_size": 128, "mean": 53.71851348876953, "std": 63.02543258666992, "min": -111.99423217773438, "p10": -8.911138916015622, "median": 43.17262268066406, "p90": 132.70290832519532, "max": 166.03280639648438, "pos_frac": 0.7890625, "sample": [165.835205078125, -7.89056396484375, -2.8600635528564453, -103.232421875, 129.87506103515625, 76.34974670410156, 127.62158203125, 33.88726806640625, 29.653045654296875, -111.99423217773438, 5.6371002197265625, 143.86016845703125, -37.303955078125, 95.1474609375, 0.0, 7.4866943359375, 18.509384155273438, 90.88388061523438, 3.364185333251953, 3.4853057861328125, 0.0, 0.0, 110.55130004882812, 114.6844253540039, 94.13832092285156, 4.3086090087890625, 140.05673217773438, 97.90350341796875, 19.04071044921875, 78.83381652832031, 69.78097534179688, 6.3455657958984375, 111.0373306274414, 125.1180419921875, 145.52345275878906, 143.4356689453125, 1.2334403991699219, -1.3922119140625, -109.14544677734375, 42.656494140625, 132.27587890625, 29.584075927734375, 76.36483001708984, 28.467300415039062, 16.6734619140625, -5.401193618774414, -1.7828216552734375, 121.63339233398438, 132.60015869140625, 17.413970947265625, 132.94265747070312, -54.45654296875, 2.6805648803710938, 35.23040771484375, 91.98944091796875, 15.720794677734375, 91.03852844238281, 25.21282958984375, 128.37393188476562, 0.0, 22.551239013671875, 22.260944366455078, 26.95550537109375, 128.28817749023438, -7.07843017578125, 153.51498413085938, 152.22039794921875, 83.95590209960938, 49.60359191894531, 0.0, 8.405290603637695, 138.41571044921875, 73.76119995117188, 104.79360961914062, -6.475360870361328, 59.08477783203125, 131.54171752929688, 166.03280639648438, 49.37059020996094, 7.660129547119141, 108.84991455078125, 118.79141235351562, 80.2733154296875, 99.99917602539062, 68.31863403320312, 2.4512901306152344, 22.820098876953125, 107.52203369140625, 85.94325256347656, 32.431427001953125, -14.8701171875, 67.92891693115234, 8.75311279296875, 133.07373046875, 4.561767578125, 122.60546875, -14.5203857421875, 9.459609985351562, 161.84024047851562, 49.10707092285156, 43.688751220703125, 28.837799072265625, 121.51812744140625, -7.888664245605469, 127.46002197265625, 115.17842864990234, 12.569995880126953, -46.65966796875, -0.43536376953125, -11.29248046875, 128.563720703125, -77.48125457763672, 108.02738952636719, 1.9742431640625, 35.44207763671875, 6.270292282104492, 23.03070068359375, 72.9583740234375, 158.70639038085938, 109.7679443359375, -21.504852294921875, 90.68350219726562, 70.04135131835938, -17.721725463867188, 92.531005859375, -21.83935546875, 117.93002319335938, 116.4229736328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000275.npy"} {"epoch": 0.5759162303664922, "step": 276, "batch_size": 128, "mean": 50.00370788574219, "std": 62.45387268066406, "min": -95.46786499023438, "p10": -20.998130798339844, "median": 37.36650848388672, "p90": 131.76508483886718, "max": 197.88726806640625, "pos_frac": 0.7734375, "sample": [197.88726806640625, 30.314071655273438, 108.35649871826172, -5.0220794677734375, 21.377044677734375, 0.007171630859375, -1.207733154296875, 62.882415771484375, 46.27978515625, 61.893592834472656, 94.34188842773438, 3.863861083984375, 10.527008056640625, 0.6759662628173828, -0.503143310546875, 97.31453704833984, 57.48382568359375, 21.575592041015625, -33.875, 56.41259765625, 139.41912841796875, 82.35935974121094, 7.806463241577148, 136.30157470703125, 86.36471557617188, -26.8824462890625, -21.124755859375, -4.48089599609375, -8.79376220703125, 113.69772338867188, 38.034332275390625, 51.51298522949219, 90.91033935546875, 107.9546890258789, 79.971923828125, 13.030059814453125, 27.48907470703125, -1.6014785766601562, 49.81529235839844, 45.42430114746094, 19.482147216796875, 23.592010498046875, 122.73477172851562, 31.037384033203125, 6.226169586181641, 148.7098388671875, 74.7677001953125, 134.29708862304688, 36.8721923828125, -8.43121337890625, 7.843528747558594, 108.87509155273438, 100.94815063476562, 30.6983642578125, 125.43241119384766, 12.955535888671875, 66.30357360839844, 25.208648681640625, 127.24331665039062, -40.10113525390625, 0.0, -46.519500732421875, 42.98773193359375, -86.78445434570312, 8.4664306640625, -39.459869384765625, 59.19187927246094, 157.590087890625, -2.1741714477539062, -20.137542724609375, 35.586002349853516, -15.58026123046875, 97.47415161132812, 124.2423095703125, 103.50177764892578, 130.09442138671875, 129.28298950195312, -20.943862915039062, 37.86082458496094, 113.32450866699219, 127.56898498535156, 110.1197509765625, 108.87088012695312, 42.13897705078125, 13.736251831054688, 77.09440612792969, -93.40457153320312, 105.5154037475586, -3.7612686157226562, 30.549163818359375, 0.4822883605957031, 182.437255859375, 65.38311767578125, 19.37884521484375, 10.796707153320312, 10.1099853515625, -1.7089385986328125, 147.20782470703125, 16.702407836914062, 129.30938720703125, 64.38168334960938, 31.838088989257812, 138.89862060546875, 0.5816650390625, -2.4844970703125, 100.64540100097656, -16.70444679260254, -34.350608825683594, 30.484100341796875, 131.40554809570312, 5.6153106689453125, 108.51437377929688, 18.21471405029297, 102.9461669921875, 132.60400390625, -71.59249877929688, 120.82666015625, 20.672607421875, 103.76002502441406, -43.168914794921875, 150.5323944091797, 19.113739013671875, 95.20004272460938, -95.46786499023438, -29.88091278076172, 135.45819091796875, 163.54989624023438, 119.86109924316406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000276.npy"} {"epoch": 0.5780104712041885, "step": 277, "batch_size": 128, "mean": 36.87504577636719, "std": 71.04310607910156, "min": -148.25436401367188, "p10": -39.75881652832031, "median": 33.12363243103027, "p90": 122.6701446533203, "max": 200.50152587890625, "pos_frac": 0.6953125, "sample": [4.3939208984375, 6.422027587890625, -6.045305252075195, 115.18402099609375, 33.25579833984375, -22.34259033203125, -94.61639404296875, -9.273956298828125, 59.87310791015625, 43.75787353515625, 22.462158203125, 67.73338317871094, 114.78103637695312, 114.68052673339844, -6.951751708984375, 124.1754150390625, 90.13543701171875, -119.9208984375, 60.52667236328125, 3.351165771484375, 30.09442138671875, 9.430524826049805, 98.92461395263672, 50.789337158203125, 98.48590087890625, 140.77383422851562, -0.7348966598510742, -5.9566650390625, 122.48895263671875, 52.030189514160156, 81.9217758178711, -141.476806640625, -4.136507034301758, -37.748634338378906, -13.833793640136719, 18.500457763671875, 114.22773742675781, 6.35162353515625, 78.0683822631836, 64.80438232421875, -147.32470703125, 1.654052734375, -22.61328887939453, -30.752273559570312, 9.196563720703125, 81.54461669921875, 31.763458251953125, 112.06461334228516, -6.4578399658203125, 34.853851318359375, 9.09918212890625, -10.105849266052246, 1.671661376953125, 200.50152587890625, 99.95761108398438, 0.0, 47.631195068359375, 97.8510513305664, 107.96940612792969, 26.56927490234375, 153.55398559570312, 90.642333984375, -112.94537353515625, -5.332027435302734, -0.901153564453125, 4.607696533203125, 90.9059066772461, -91.18917846679688, 179.49871826171875, 175.7620849609375, -37.75429153442383, 41.195159912109375, 150.98358154296875, 48.33961486816406, 9.072296142578125, -148.25436401367188, 73.395263671875, -11.6937255859375, -41.37461853027344, 32.9914665222168, 20.21099853515625, 98.23627471923828, 48.279510498046875, -10.030670166015625, 76.0885009765625, 99.58622741699219, 101.29523468017578, 118.15847778320312, -86.09613037109375, 128.2459716796875, 1.867177963256836, 9.939041137695312, -13.503265380859375, 66.3836898803711, 73.34390258789062, -5.256988525390625, 72.30752563476562, 94.75680541992188, 75.77410888671875, 117.46072387695312, 12.432830810546875, 15.517745971679688, 114.84317016601562, 123.09292602539062, 34.40447235107422, 116.2152099609375, 73.04974365234375, 130.08941650390625, -53.75456237792969, 0.0, 130.96749877929688, 168.5001220703125, 2.4638595581054688, 57.543914794921875, -49.294281005859375, -10.738571166992188, -116.73837280273438, 124.69482421875, 22.416091918945312, 82.96652221679688, 91.34236145019531, -107.4910888671875, -1.1244945526123047, -1.5480270385742188, 7.325260162353516, 43.8524169921875, 49.83232116699219, -39.06632995605469], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000277.npy"} {"epoch": 0.5801047120418849, "step": 278, "batch_size": 128, "mean": 31.309158325195312, "std": 73.78474426269531, "min": -168.42251586914062, "p10": -76.34132995605468, "median": 18.34959316253662, "p90": 123.39800720214842, "max": 202.01416015625, "pos_frac": 0.7109375, "sample": [119.00762939453125, 94.89306640625, -83.95608520507812, 42.154205322265625, -4.5713653564453125, 36.42047119140625, 55.7078857421875, 10.531585693359375, 12.822351455688477, 129.92164611816406, 43.62548828125, 116.02042388916016, 92.34001159667969, 45.59486389160156, 95.64610290527344, 86.35308837890625, -112.70150756835938, -19.282577514648438, 25.360977172851562, -4.50457763671875, 19.22339630126953, 73.94790649414062, -4.782989501953125, -22.534210205078125, 32.643768310546875, 171.27197265625, 145.67019653320312, 93.36487579345703, 9.224258422851562, -20.8203125, -168.42251586914062, 181.50396728515625, -95.05590057373047, 6.687164306640625, 56.0059814453125, 83.72039794921875, 76.35404968261719, -57.5147705078125, 144.19857788085938, 23.82396697998047, 131.40625, 146.969970703125, 87.41217041015625, 12.800537109375, -8.418434143066406, 47.11114501953125, -1.8339385986328125, 148.32046508789062, 69.32384490966797, 73.0798110961914, 9.048612594604492, 0.97393798828125, 2.7727317810058594, -90.56831359863281, -18.04583740234375, -135.82473754882812, 2.8871517181396484, 154.07180786132812, -79.57502746582031, -97.59709930419922, 32.84519958496094, 76.081787109375, 28.599212646484375, 3.6961669921875, 12.915740966796875, 122.126220703125, -43.8021240234375, 110.97798156738281, 115.3631362915039, 30.120010375976562, 156.9501953125, 202.01416015625, -18.961381912231445, -24.161453247070312, 74.69757080078125, -99.06367492675781, 98.91303253173828, 13.7652587890625, 89.31214904785156, 34.63037109375, -128.1248321533203, 8.591285705566406, 2.9383468627929688, 51.115966796875, 102.49945831298828, -13.192550659179688, 12.67169189453125, 64.38764190673828, 74.3726806640625, 104.16807556152344, 0.2952880859375, -21.177490234375, 113.73741149902344, 169.9927978515625, -20.2127685546875, 122.16091918945312, -74.95545959472656, 8.898941040039062, 33.07323455810547, 0.3589935302734375, -12.133468627929688, 6.7379608154296875, 20.6312255859375, -94.12081146240234, -14.89697265625, 24.84027099609375, 126.2845458984375, -21.117172241210938, 5.37115478515625, 11.660335540771484, 108.63751983642578, 4.11713981628418, -4.023223876953125, 89.4578857421875, 0.5626850128173828, -29.84404754638672, 17.47579002380371, -122.88807678222656, 88.44718933105469, -31.9603271484375, 8.048896789550781, 16.447738647460938, 15.40631103515625, 120.20419311523438, -111.96310424804688, 77.77752685546875, -20.412185668945312, 115.99571228027344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000278.npy"} {"epoch": 0.5821989528795811, "step": 279, "batch_size": 128, "mean": 49.97075653076172, "std": 66.02855682373047, "min": -119.51927947998047, "p10": -16.46661376953125, "median": 35.33892822265625, "p90": 138.8604309082031, "max": 212.50091552734375, "pos_frac": 0.796875, "sample": [38.9176025390625, 109.55970764160156, 43.25636291503906, -22.555511474609375, 18.17706298828125, -2.7176513671875, 69.01945495605469, 50.16168212890625, 6.065269470214844, 121.54473876953125, 66.5531005859375, 157.8062744140625, 37.22076416015625, -11.228271484375, -3.7239990234375, 32.06793212890625, 97.67202758789062, 103.79521942138672, 33.0408935546875, 54.403961181640625, 18.005615234375, 29.943702697753906, 32.92242431640625, 34.4395751953125, 38.322227478027344, 26.513580322265625, 18.424942016601562, 6.07110595703125, -110.47972106933594, 5.400016784667969, 82.24258422851562, 161.85003662109375, 123.7821044921875, 130.69577026367188, 82.75381469726562, -16.8148193359375, 5.9112548828125, 102.74961853027344, -26.218795776367188, 10.531997680664062, 106.65149688720703, 121.10983276367188, 109.98844909667969, 145.8639678955078, 12.252220153808594, 138.53521728515625, -12.458694458007812, 11.030677795410156, -41.629058837890625, 90.0856704711914, 24.18829345703125, 34.36016845703125, 8.185455322265625, 31.485849380493164, 68.88740539550781, -8.730087280273438, 212.50091552734375, -119.51927947998047, 105.13238525390625, 157.79501342773438, -14.474624633789062, -6.532470703125, 14.993942260742188, 23.611083984375, -16.3173828125, -10.868789672851562, -23.237899780273438, 122.07563781738281, -2.068958282470703, 178.31005859375, -46.891448974609375, 187.3153076171875, 45.25523376464844, 87.67754364013672, 9.70565414428711, -113.20596313476562, 9.11700439453125, 126.29937744140625, 166.50628662109375, -97.94369506835938, 6.493011474609375, 23.33074951171875, 70.55569458007812, 91.95819091796875, 88.84652709960938, 85.87899780273438, 27.911285400390625, 7.283252716064453, 135.1943817138672, 10.786666870117188, 101.79669189453125, -20.990203857421875, 106.21600341796875, 46.590789794921875, 22.223434448242188, 17.60736083984375, 136.89987182617188, 1.4957695007324219, 51.21588134765625, 128.9279327392578, 89.072021484375, 139.6192626953125, -11.125411987304688, 123.7659912109375, 148.11517333984375, 135.9910888671875, 16.1070556640625, 8.91314697265625, 12.904327392578125, 140.4371795654297, -4.130340576171875, 36.10308837890625, 48.74884033203125, 166.906982421875, 108.95845031738281, 54.4285888671875, 22.13043212890625, 43.28424072265625, -97.29913330078125, -21.07122039794922, 152.3619384765625, -6.494110107421875, 84.9918212890625, 2.5460357666015625, 122.61672973632812, 44.39154052734375, 34.57476806640625, 68.06024169921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000279.npy"} {"epoch": 0.5842931937172775, "step": 280, "batch_size": 128, "mean": 44.95423126220703, "std": 68.92656707763672, "min": -141.93824768066406, "p10": -31.161740875244135, "median": 31.30915069580078, "p90": 140.75210571289062, "max": 175.3480224609375, "pos_frac": 0.75, "sample": [109.42401123046875, 111.37760925292969, 0.893218994140625, 122.11929321289062, -5.190309524536133, -141.93824768066406, 137.7130584716797, 120.80072021484375, 151.197509765625, 58.97023010253906, -21.847518920898438, 35.681236267089844, 22.37232208251953, 85.98516845703125, 16.903076171875, -25.769668579101562, 147.8194580078125, 91.1895751953125, 148.78228759765625, -4.31488037109375, 6.539825439453125, 142.0543212890625, -65.14552307128906, 143.10263061523438, -27.766754150390625, 105.85226440429688, 25.912981033325195, 19.1031494140625, -5.311737060546875, 114.67120361328125, 0.0, 22.051910400390625, 41.350860595703125, -34.047943115234375, 46.23150634765625, 175.3480224609375, -34.80902862548828, 134.21817016601562, 10.175640106201172, 1.341796875, -6.923736572265625, 87.970947265625, 10.207778930664062, 21.05682373046875, 92.98319244384766, 36.584320068359375, 104.63262939453125, -58.71284484863281, 144.26736450195312, 18.2938232421875, 55.1827392578125, 9.503440856933594, 7.2574462890625, 6.275205612182617, 102.74691772460938, 145.11376953125, 71.23492431640625, -26.0235595703125, 1.7998199462890625, 116.11892700195312, 130.692626953125, 28.569869995117188, 19.115150451660156, -82.93648529052734, -8.780426025390625, 8.7313232421875, -76.4552001953125, 135.49005126953125, 7.6756134033203125, 89.18757629394531, 125.5965576171875, 51.26268768310547, 140.33563232421875, 141.723876953125, 149.34881591796875, -24.84613037109375, 100.68319702148438, 81.86630249023438, 25.98883056640625, 52.982452392578125, 103.49383544921875, 118.31524658203125, -25.500640869140625, 110.61557006835938, 131.44815063476562, -29.92479705810547, 14.99481201171875, 16.909576416015625, -38.52323913574219, -1.06134033203125, 93.3433837890625, 42.185546875, 82.35371398925781, -69.7154541015625, 50.762935638427734, 154.38348388671875, -7.5990142822265625, 94.41622924804688, 114.12548828125, 34.048431396484375, 127.74232482910156, 13.389785766601562, 118.134521484375, -19.851272583007812, 17.500534057617188, 4.783100128173828, 0.009674072265625, -0.491607666015625, 149.71444702148438, 7.85028076171875, 44.116790771484375, 40.30622100830078, 5.799537658691406, 95.60623168945312, -61.609344482421875, 100.83766174316406, 114.09518432617188, 168.33404541015625, -9.8521728515625, -136.40098571777344, -5.9879150390625, 9.811405181884766, -90.52447509765625, 13.28615951538086, 34.2894287109375, 3.5621280670166016, 76.19924926757812, -74.4249267578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000280.npy"} {"epoch": 0.5863874345549738, "step": 281, "batch_size": 128, "mean": 44.79418182373047, "std": 65.54524993896484, "min": -129.05422973632812, "p10": -21.376190185546875, "median": 30.179670333862305, "p90": 135.0844497680664, "max": 243.7869873046875, "pos_frac": 0.703125, "sample": [99.42120361328125, 45.97691345214844, -0.221710205078125, 3.3721237182617188, 76.88204956054688, 9.649017333984375, 50.603485107421875, 53.14185333251953, 8.527618408203125, 142.80157470703125, 132.09185791015625, -78.39157104492188, -108.1707763671875, 149.76095581054688, 17.821834564208984, 85.2071533203125, 105.18538665771484, 82.71429443359375, 4.6840667724609375, -8.446907043457031, 135.63768005371094, 11.540119171142578, -6.0312957763671875, -6.3986358642578125, 70.5389404296875, -9.192138671875, -46.781951904296875, 141.93695068359375, 17.480072021484375, 60.880615234375, 0.0, 47.787506103515625, -3.036895751953125, 66.09512329101562, 30.42556381225586, 128.46400451660156, 41.06524658203125, -22.09588623046875, 145.13487243652344, 79.01882934570312, -129.05422973632812, 59.57342529296875, 120.93804931640625, -14.672586441040039, 0.0, 28.68377685546875, 26.681182861328125, 119.98519897460938, -26.55694580078125, 33.14772033691406, 72.467529296875, -43.854705810546875, 8.569164276123047, 39.21293640136719, 1.613250732421875, 162.64010620117188, -9.77459716796875, 30.493560791015625, -1.2589263916015625, 76.29249572753906, 16.5576171875, 41.532562255859375, -21.0677490234375, -1.857757568359375, 101.09912109375, 66.48416137695312, 24.83081817626953, -26.546676635742188, -4.8431396484375, 5.037353515625, 17.45703125, 131.0771026611328, -1.3349609375, 134.19900512695312, 124.08311462402344, 54.99591064453125, 133.76123046875, 5.883514404296875, 37.02447509765625, 56.08087921142578, 26.819351196289062, 101.13386535644531, 243.7869873046875, 138.62289428710938, 123.70831298828125, -30.49188232421875, -0.248321533203125, 103.68324279785156, 16.575881958007812, 2.7891845703125, 165.68484497070312, 31.929962158203125, 100.32475280761719, -5.391201019287109, 21.566146850585938, -33.71630859375, -7.764007568359375, 151.2359619140625, 119.5545654296875, 40.36077880859375, -25.683441162109375, 44.34466552734375, 0.0, 149.72796630859375, -1.8464241027832031, -18.433685302734375, 98.98529052734375, -92.43499755859375, 153.64434814453125, 148.23806762695312, 117.20213317871094, 134.84735107421875, 128.66473388671875, -3.75665283203125, 131.94644165039062, 12.5010986328125, -16.155410766601562, 72.2227783203125, 12.34079360961914, 22.595809936523438, 25.722183227539062, -87.59957885742188, -8.26666259765625, 130.80010986328125, 17.143402099609375, 46.83277893066406, 29.93377685546875, -4.685211181640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000281.npy"} {"epoch": 0.5884816753926702, "step": 282, "batch_size": 128, "mean": 41.26219940185547, "std": 73.16077423095703, "min": -152.541259765625, "p10": -43.56532745361328, "median": 29.38542366027832, "p90": 134.98310699462888, "max": 201.219970703125, "pos_frac": 0.75, "sample": [134.04270935058594, -71.82293701171875, 36.83502197265625, 201.219970703125, 70.66268920898438, 6.527008056640625, -152.541259765625, 137.1773681640625, 98.33369445800781, 0.3458709716796875, 8.568466186523438, 88.32598114013672, 23.576416015625, -41.594970703125, 13.156494140625, 10.009452819824219, 91.815673828125, -38.7568359375, 113.82839965820312, -11.408164978027344, -48.38677978515625, 53.962371826171875, -89.40153503417969, 40.394256591796875, 13.34640121459961, 191.62149047851562, 125.08445739746094, 62.4713134765625, -0.43731689453125, 12.36016845703125, -38.45123291015625, 21.650054931640625, 121.72900390625, 72.67720031738281, -4.67633056640625, 63.42864227294922, 96.889892578125, 13.185550689697266, 45.975250244140625, 32.29248046875, 53.99180603027344, 85.1988525390625, 6.521087646484375, -64.0245361328125, 121.21478271484375, -34.46089172363281, -111.9373779296875, 40.43702697753906, 148.30621337890625, 5.818902969360352, 24.09808349609375, 16.5140380859375, -25.084930419921875, 12.793922424316406, -35.387115478515625, 167.87692260742188, 38.1036376953125, 3.7390594482421875, 16.795635223388672, -65.46934509277344, 3.68084716796875, 2.76629638671875, -19.61724853515625, 120.69384765625, 133.81210327148438, 102.59235382080078, 95.84171295166016, -10.291267395019531, -4.0158538818359375, 130.15924072265625, -88.99859619140625, -6.777618408203125, 125.71755981445312, 58.776275634765625, 12.01531982421875, 163.44464111328125, 117.9298095703125, -77.4895248413086, 75.85980224609375, -28.978973388671875, -27.610641479492188, 10.00088119506836, 22.62628173828125, 161.3981475830078, 60.459228515625, -44.23793029785156, 19.055999755859375, 15.629440307617188, 146.9410400390625, 133.55462646484375, 165.22311401367188, 84.7608642578125, 15.810516357421875, 79.22552490234375, -116.2740478515625, 40.610992431640625, 113.85968017578125, 142.84864807128906, 8.308752059936523, -5.4567413330078125, 126.7452392578125, 196.1199951171875, 113.671875, 29.389850616455078, 111.71406555175781, 1.2148284912109375, -135.07565307617188, 20.2972412109375, 42.65240478515625, 29.978214263916016, 128.0897216796875, 29.380996704101562, 52.995758056640625, 94.95687103271484, 7.096015930175781, 132.5626220703125, 146.04306030273438, 13.885650634765625, 141.32151794433594, 46.26300048828125, 59.24853515625, -82.71063232421875, -4.452484130859375, -43.277069091796875, 105.3389892578125, -9.693267822265625, 14.903488159179688, 69.98388671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000282.npy"} {"epoch": 0.5905759162303665, "step": 283, "batch_size": 128, "mean": 44.826019287109375, "std": 69.4393081665039, "min": -115.14447021484375, "p10": -46.89275512695312, "median": 33.60362243652344, "p90": 130.55355072021484, "max": 193.5828857421875, "pos_frac": 0.7578125, "sample": [-26.872222900390625, 24.791915893554688, 132.12762451171875, -28.28399658203125, 30.62108039855957, -51.97113037109375, 125.47322082519531, -68.66215515136719, 60.8045654296875, 3.777984619140625, 1.897186279296875, 126.89093017578125, 80.68428802490234, 91.3887710571289, -1.7171268463134766, -55.745361328125, -17.2637939453125, 94.91675567626953, 6.5338897705078125, -36.533935546875, 185.70855712890625, 148.25750732421875, 34.01495361328125, -36.21240234375, 137.546875, 114.81912231445312, 91.64999389648438, -9.625625610351562, 64.31260681152344, 15.816925048828125, 75.63957214355469, 74.76492309570312, 24.929916381835938, 116.79620361328125, 48.140174865722656, 99.68547821044922, 7.3433380126953125, 0.23473739624023438, 61.59912109375, 68.5601806640625, -9.6168212890625, 113.00146484375, -32.70506286621094, -82.43453979492188, 107.79835510253906, 111.564697265625, 19.4521484375, 7.6396484375, -94.14608764648438, 164.79873657226562, 97.47088623046875, 131.5907745361328, 113.42280578613281, 101.21006774902344, 153.5530242919922, 5.441764831542969, 8.84527587890625, 37.13398742675781, 25.97064208984375, 91.38920593261719, 55.86834716796875, 103.1064453125, 140.3272705078125, 138.6085205078125, 128.6873779296875, -2.6898193359375, 130.15234375, 8.00726318359375, 35.39289855957031, 20.756805419921875, -66.08788299560547, 109.68455505371094, -90.04823303222656, 128.57119750976562, 34.78314208984375, 67.18486022949219, 131.4897003173828, 109.71468353271484, 0.0, 3.70001220703125, -69.86780548095703, 12.908767700195312, 104.35975646972656, -105.01960754394531, 123.70489501953125, 90.20709228515625, 84.11463928222656, 26.0831298828125, -44.71630859375, 32.09761047363281, 0.4746551513671875, -0.1726226806640625, 10.47967529296875, -21.19500732421875, 31.561187744140625, 118.05877685546875, 112.01007080078125, 61.5413818359375, -115.14447021484375, 93.39764404296875, 1.331644058227539, 31.09991455078125, 193.5828857421875, -4.840549468994141, 33.192291259765625, 23.910751342773438, 111.91421508789062, -95.67776489257812, 21.529815673828125, 1.5038604736328125, 141.17568969726562, 22.60516357421875, -0.3969287872314453, 2.977113723754883, 0.0, 91.69793701171875, 103.62476348876953, 191.40350341796875, 17.12657928466797, 5.053134918212891, 59.34306335449219, 109.8369369506836, 66.2698974609375, 40.21342468261719, -62.89208984375, 124.44322204589844, -83.08901977539062, -29.523887634277344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000283.npy"} {"epoch": 0.5926701570680628, "step": 284, "batch_size": 128, "mean": 43.82318115234375, "std": 79.51141357421875, "min": -147.40234375, "p10": -76.40488891601562, "median": 36.66746520996094, "p90": 143.21266479492186, "max": 244.40924072265625, "pos_frac": 0.7421875, "sample": [-49.58056640625, 89.07281494140625, -83.83660125732422, -147.40234375, 111.33590698242188, 34.80499267578125, 98.46826171875, 141.1251220703125, -87.66575622558594, 44.13341522216797, 88.52006530761719, 142.96875, 88.75736236572266, 27.088699340820312, 28.038360595703125, 95.99150085449219, 130.23033142089844, 63.1158447265625, 145.598388671875, 8.491233825683594, 123.9947509765625, 209.2998046875, -18.865631103515625, 101.48141479492188, -7.237886428833008, 15.98846435546875, 26.899322509765625, -77.56332397460938, 13.97119140625, -44.3160400390625, -75.90841674804688, 39.2061767578125, 62.153961181640625, -29.657371520996094, 7.4661102294921875, 37.31695556640625, 115.80157470703125, 85.2879638671875, 79.8011245727539, 0.719146728515625, -119.0586929321289, 11.443954467773438, 81.30418395996094, 36.017974853515625, -9.121940612792969, -13.971099853515625, -78.24903869628906, 47.4493408203125, 11.98785400390625, 185.2022705078125, 132.7797088623047, -96.16322326660156, 46.15045166015625, 20.43359375, 117.87042236328125, 90.12002563476562, 9.79547119140625, -5.563255310058594, 145.05545043945312, 65.41217041015625, 104.69790649414062, 158.6619415283203, 21.935653686523438, -5.98114013671875, 133.94273376464844, 55.55023193359375, 166.17376708984375, 26.5108642578125, 88.80633544921875, 11.184297561645508, 85.56918334960938, 18.742889404296875, -17.911611557006836, 65.3184814453125, 3.442331314086914, 244.40924072265625, 6.4673309326171875, 15.500907897949219, 11.6572265625, -1.0214996337890625, 168.5789031982422, 130.30657958984375, 103.54228973388672, 148.87860107421875, -58.198760986328125, -134.197021484375, 11.411096572875977, -20.036895751953125, 53.51191711425781, -60.53832244873047, 102.03437805175781, 59.35516357421875, 99.47540283203125, 120.12916564941406, 1.9275932312011719, -1.1104583740234375, 60.09063720703125, -108.11268615722656, -2.18914794921875, 100.52371215820312, 143.78179931640625, 35.697967529296875, 104.39778137207031, 11.165679931640625, 0.9883861541748047, 129.65335083007812, 21.493942260742188, 24.002197265625, 23.00604248046875, 100.2293701171875, 130.66555786132812, 190.5662841796875, 53.09503173828125, -23.805877685546875, -131.50656127929688, -106.47503662109375, 101.92630767822266, -88.10153198242188, 1.4362163543701172, -20.4273681640625, 172.254150390625, 50.03562927246094, 106.7275390625, 186.604736328125, 111.57499694824219, -2.3924407958984375, -97.84417724609375, 97.587646484375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000284.npy"} {"epoch": 0.5947643979057592, "step": 285, "batch_size": 128, "mean": 34.614463806152344, "std": 67.75348663330078, "min": -130.81781005859375, "p10": -46.56424560546874, "median": 25.58003044128418, "p90": 125.39397888183593, "max": 175.16806030273438, "pos_frac": 0.7109375, "sample": [16.81048583984375, 92.93716430664062, 109.64764404296875, 30.171600341796875, 122.2900390625, 128.18280029296875, 113.09686279296875, -94.55154418945312, 121.21322631835938, 32.83457946777344, 60.592041015625, -16.6951904296875, 9.644111633300781, -87.93437194824219, -11.5706787109375, 36.055198669433594, 17.49932861328125, -34.81463623046875, -16.86578369140625, -11.629058837890625, -8.83526611328125, 68.548583984375, -97.6913833618164, 16.035537719726562, 29.923545837402344, 81.54301452636719, 150.33636474609375, 50.669921875, 89.22642517089844, 36.562408447265625, 175.16806030273438, -52.141387939453125, 9.696502685546875, 17.383407592773438, 131.68228149414062, 132.2602081298828, 118.8662109375, -37.29115295410156, -31.847244262695312, 10.91558837890625, -44.174041748046875, -29.610321044921875, -20.743484497070312, -1.769927978515625, 5.357114791870117, 61.96075439453125, 53.9124755859375, 117.36727905273438, 1.7257041931152344, 9.346099853515625, 131.9977569580078, 9.641410827636719, 11.705162048339844, -61.280372619628906, 146.36630249023438, 2.069591522216797, -6.5333404541015625, -3.974050521850586, 116.89456176757812, 16.19892120361328, 41.45849609375, 88.06036376953125, 124.41117858886719, 76.01251220703125, 45.7386474609375, 48.57452392578125, -79.33843994140625, 14.474964141845703, -95.62639617919922, 44.4683837890625, -122.85150146484375, 122.50244140625, 29.959327697753906, 6.502342224121094, 97.31607055664062, 9.479507446289062, 53.23876953125, 24.56728744506836, 43.93928527832031, 14.2113037109375, 92.57881927490234, 113.895751953125, 11.580108642578125, 1.5665435791015625, -15.450149536132812, 7.4061279296875, 34.5509033203125, 127.68717956542969, 146.27783203125, 108.51885986328125, 101.18663024902344, 118.10856628417969, -3.8740768432617188, -26.22784423828125, -24.236053466796875, 14.562973022460938, 35.356201171875, 58.600677490234375, -57.30169677734375, 18.052734375, 94.21774291992188, -61.33351135253906, 97.78462219238281, -34.83648681640625, -31.4222412109375, -26.878005981445312, 7.0599365234375, -13.94100570678711, 139.59637451171875, 66.66641235351562, -130.81781005859375, 22.856826782226562, 30.543853759765625, -44.109710693359375, -5.146339416503906, -117.26449584960938, 122.19668579101562, 5.9407806396484375, 119.63285827636719, 58.95281982421875, 26.5927734375, 83.13681030273438, 84.41241455078125, 132.613525390625, -65.00210571289062, 95.83804321289062, 155.17459106445312, 143.864013671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000285.npy"} {"epoch": 0.5968586387434555, "step": 286, "batch_size": 128, "mean": 34.54685974121094, "std": 72.92725372314453, "min": -133.25704956054688, "p10": -66.7594612121582, "median": 27.75689697265625, "p90": 132.22933349609374, "max": 191.84588623046875, "pos_frac": 0.6953125, "sample": [-12.044158935546875, -21.55804443359375, 50.856048583984375, 19.064773559570312, 11.436073303222656, 26.331314086914062, 8.532928466796875, 9.957206726074219, 9.6322021484375, 81.61546325683594, 53.454246520996094, 18.83709716796875, -6.617424011230469, 66.1142578125, -117.79046630859375, 153.1553192138672, 11.1881103515625, 146.23898315429688, 108.50395202636719, 141.97222900390625, -36.789703369140625, 134.72586059570312, 22.558868408203125, 31.404312133789062, 18.672569274902344, 23.29559326171875, -16.056060791015625, -5.636985778808594, 89.81442260742188, -17.24158477783203, -133.25704956054688, 0.1988372802734375, 126.57563018798828, 118.99263000488281, -28.0406494140625, 29.522531509399414, -1.21661376953125, 21.567779541015625, 9.35177230834961, 80.62303161621094, -8.69525146484375, 129.7989501953125, 110.45904541015625, -92.47476959228516, 77.10702514648438, -44.73382568359375, 48.78717041015625, 191.84588623046875, 25.14404296875, -14.65679931640625, 128.09454345703125, -127.98809814453125, 131.15939331054688, 95.49978637695312, 40.8577880859375, 105.701171875, 3.072357177734375, 97.98715209960938, -52.25206756591797, 154.23287963867188, -10.206718444824219, 0.0, -5.222900390625, -97.56277465820312, 51.845340728759766, 52.723541259765625, -5.427314758300781, -31.893753051757812, -57.608238220214844, 45.19084930419922, -53.999114990234375, 111.19601440429688, 113.41799926757812, 47.510101318359375, -66.3265380859375, 34.046966552734375, 125.62286376953125, 84.76980590820312, -72.7164306640625, 20.499988555908203, 58.103759765625, 10.097122192382812, 3.538393020629883, 188.78399658203125, 126.23529052734375, 36.92399597167969, 29.656494140625, -79.77615356445312, 107.99200439453125, -104.34149169921875, 70.63050079345703, 148.51162719726562, 32.10028076171875, 60.21783447265625, 136.12802124023438, 65.94244384765625, 45.71124267578125, 99.22063446044922, 97.56171417236328, 95.48262023925781, 159.96923828125, 29.182479858398438, 138.79971313476562, 0.8716278076171875, 61.259979248046875, -45.12213134765625, 151.17030334472656, 39.66283416748047, -12.944976806640625, -52.541019439697266, -89.406494140625, 127.32012939453125, -4.778568267822266, -67.76961517333984, -100.1013412475586, 179.80172729492188, -84.0953369140625, -68.89450073242188, 25.062171936035156, 12.358818054199219, 37.69828796386719, 80.66726684570312, 0.058597564697265625, 15.969100952148438, 50.72576904296875, -22.492034912109375, 106.44631958007812, 11.647682189941406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000286.npy"} {"epoch": 0.5989528795811518, "step": 287, "batch_size": 128, "mean": 40.54359436035156, "std": 69.98181915283203, "min": -159.74081420898438, "p10": -48.734320068359374, "median": 41.643043518066406, "p90": 125.0415557861328, "max": 166.87472534179688, "pos_frac": 0.703125, "sample": [101.15898132324219, -21.831504821777344, -55.916778564453125, -10.381099700927734, -28.49249267578125, -56.55131530761719, 125.98318481445312, 117.54733276367188, 67.82555389404297, 146.14593505859375, 110.13388061523438, 17.55999755859375, 5.1981658935546875, 124.02293395996094, 5.872314453125, 16.892303466796875, -26.389328002929688, -71.64862060546875, -26.047882080078125, 35.35192108154297, -26.07330322265625, 166.87472534179688, 120.38134765625, -0.425537109375, 80.61099243164062, -23.304428100585938, -13.092964172363281, 65.10028076171875, 65.3824462890625, 43.410064697265625, -9.036247253417969, 132.79544067382812, 97.35092163085938, 80.52020263671875, -19.23040771484375, 3.29034423828125, 84.44020080566406, 109.82656860351562, -1.7852706909179688, 113.98188781738281, 112.60785675048828, 99.24301147460938, 20.396392822265625, 76.4920654296875, 133.1222686767578, 10.618440628051758, -4.410491943359375, 124.63800048828125, 56.23558044433594, 153.26953125, 40.37693786621094, 120.66323852539062, 116.69430541992188, 65.90339660644531, 26.58502197265625, 42.909149169921875, 36.431190490722656, -23.83636474609375, 99.25979614257812, -55.1431884765625, 114.1096420288086, 89.98519897460938, 44.18592834472656, 36.63637924194336, 2.3393592834472656, 30.726547241210938, 93.60842895507812, 104.126708984375, 58.38787841796875, 60.496551513671875, -125.64349365234375, -6.911464691162109, 112.58489990234375, 137.40548706054688, 50.45428466796875, 127.67916870117188, 117.36474609375, 108.7012939453125, 15.7064208984375, 98.174560546875, 159.90538024902344, -69.95822143554688, -96.94676208496094, 5.094810485839844, -46.968170166015625, 78.97956085205078, -30.046844482421875, 72.98112487792969, -86.680908203125, 14.595413208007812, 105.05516052246094, 147.02613830566406, -159.74081420898438, -13.003204345703125, 66.14783477783203, -17.217803955078125, 78.86380004882812, 4.024961471557617, -112.34819030761719, 39.22053527832031, 8.06951904296875, 15.797027587890625, 7.417640686035156, 55.250030517578125, -30.333587646484375, -50.371612548828125, 111.1745834350586, -122.06011962890625, 150.27841186523438, -11.471382141113281, 119.24945068359375, -6.5463714599609375, 115.26922607421875, 45.03163146972656, 139.82164001464844, 8.892059326171875, 19.017372131347656, 86.75971984863281, 60.787994384765625, -0.015186309814453125, 22.783071517944336, 8.046577453613281, -101.34396362304688, -48.032623291015625, 56.23895263671875, 130.8603515625, -9.240203857421875, 99.64048767089844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000287.npy"} {"epoch": 0.6010471204188481, "step": 288, "batch_size": 128, "mean": 40.37410354614258, "std": 70.03819274902344, "min": -129.6103515625, "p10": -39.37781982421875, "median": 25.192001342773438, "p90": 128.9158172607422, "max": 198.4910888671875, "pos_frac": 0.7421875, "sample": [47.321319580078125, 4.228096008300781, -23.754638671875, -12.154052734375, 90.4483642578125, 4.248321533203125, -17.37488555908203, -19.935592651367188, 29.677200317382812, 101.45211791992188, 24.5753173828125, 157.58428955078125, -0.807708740234375, 98.01461791992188, 74.65097045898438, -8.11456298828125, 6.693626403808594, 48.33323669433594, 23.73193359375, 0.0, 22.385589599609375, -9.87738037109375, 37.956512451171875, 9.99148178100586, 60.74214172363281, 3.2706298828125, -30.677459716796875, 6.150390625, -73.61129760742188, 18.9013671875, 163.833251953125, 112.12777709960938, 49.86051940917969, 87.23359680175781, 6.967987060546875, 0.0, 1.22503662109375, 38.19915008544922, 20.3013916015625, 17.177154541015625, -4.3403472900390625, 109.7684097290039, -91.286376953125, -97.57949829101562, 13.081398010253906, 112.8790283203125, 157.11276245117188, 148.53921508789062, -129.6103515625, 0.8478622436523438, 122.9639892578125, -8.124092102050781, 117.79974365234375, 16.229652404785156, 25.808685302734375, -122.54046630859375, 3.740863800048828, 48.14251708984375, 117.62159729003906, -61.361419677734375, 198.4910888671875, -88.32261657714844, -38.636993408203125, 92.67707824707031, 173.86822509765625, 15.834991455078125, -11.143508911132812, -49.50048828125, 125.82351684570312, 16.649635314941406, 123.46331787109375, 31.54022216796875, -2.1483306884765625, 70.51881408691406, 12.65032958984375, 11.019683837890625, 18.8447265625, 80.19309997558594, -30.473907470703125, 58.75248718261719, 94.19026184082031, 36.977752685546875, -16.954940795898438, -76.02557373046875, -22.155433654785156, -75.1653823852539, 15.432170867919922, 149.1824951171875, 128.50277709960938, 9.411773681640625, 41.41851806640625, 2.8240966796875, 98.22015380859375, -5.428314208984375, 92.01858520507812, 115.1346435546875, 96.44070434570312, -71.29876708984375, 137.18316650390625, 26.89813232421875, 50.70265197753906, 80.41154479980469, -93.94671630859375, 177.82443237304688, 8.81998062133789, 31.888671875, 15.579160690307617, 52.3134765625, 41.3067626953125, -14.49481201171875, -41.106414794921875, 79.75625610351562, 65.610595703125, 124.23660278320312, 98.59518432617188, 117.10147094726562, 28.2872314453125, 13.47076416015625, 17.8077392578125, 119.05964660644531, 41.66104507446289, 198.25082397460938, 97.35050964355469, 161.03924560546875, 129.87957763671875, 23.035186767578125, 196.0601806640625, 109.80720520019531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000288.npy"} {"epoch": 0.6031413612565445, "step": 289, "batch_size": 128, "mean": 29.019657135009766, "std": 71.9331283569336, "min": -146.64419555664062, "p10": -61.3064598083496, "median": 16.014404296875, "p90": 125.79322509765625, "max": 164.404296875, "pos_frac": 0.6484375, "sample": [-120.74993896484375, -20.941650390625, 14.38250732421875, -72.93492889404297, -7.01336669921875, 0.97528076171875, 11.29351806640625, -23.60870361328125, 63.86627197265625, -75.5035400390625, 122.16461181640625, -4.122520446777344, 1.3214445114135742, 42.7286376953125, -0.9404449462890625, 73.53298950195312, -32.29534912109375, 136.218505859375, 16.334228515625, -56.25474548339844, 164.404296875, 138.27032470703125, 70.59463500976562, 19.869979858398438, -3.80145263671875, -19.612686157226562, -9.774078369140625, 3.3510475158691406, 8.41143798828125, 82.54112243652344, -12.056228637695312, 111.7149658203125, -3.897308349609375, 34.407196044921875, 2.5642013549804688, 0.0, 8.184883117675781, 86.31265258789062, -27.275482177734375, 15.5435791015625, 13.118705749511719, -39.0347900390625, -109.00506591796875, -125.4649658203125, 75.9593505859375, -28.841522216796875, 51.846435546875, 110.2139892578125, -6.93511962890625, 0.163360595703125, -8.293716430664062, 115.39578247070312, -130.095703125, -65.21090698242188, 15.694580078125, 29.261505126953125, -6.97593879699707, -8.334022521972656, 50.34107208251953, 155.96841430664062, -93.8009033203125, 113.7996826171875, 51.5423583984375, 68.50283813476562, 45.10101318359375, 115.0325698852539, -87.14907836914062, -112.93670654296875, -59.63312530517578, 5.442962646484375, 127.15365600585938, -30.880538940429688, -0.029876708984375, 107.53300476074219, 22.22539520263672, 17.169570922851562, 58.33306121826172, 43.25408935546875, 85.67343139648438, 5.67718505859375, 75.54180908203125, -20.84759521484375, -146.64419555664062, 134.12396240234375, -15.061141967773438, 94.5497055053711, -9.4910888671875, 126.11137390136719, -40.05989074707031, 132.87454223632812, 23.9239501953125, 153.90103149414062, 118.12522888183594, 10.46795654296875, 34.87242126464844, 96.9453125, 94.98468017578125, 8.29833984375, 91.39093017578125, 86.3955078125, 131.92507934570312, 85.6294937133789, 150.5762939453125, 100.33604431152344, -22.39495849609375, 113.45176696777344, 15.61273193359375, 50.33985900878906, 90.67092895507812, 164.38665771484375, -109.19024658203125, 116.27029418945312, 123.78288269042969, -35.047088623046875, 125.65687561035156, -59.37548828125, -106.52197265625, 50.217979431152344, -18.311920166015625, 124.27434539794922, 33.590484619140625, 29.2108154296875, 147.93360900878906, 7.7895965576171875, 18.981842041015625, -6.720573425292969, 5.77044677734375, 25.27972412109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000289.npy"} {"epoch": 0.6052356020942409, "step": 290, "batch_size": 128, "mean": 35.72801208496094, "std": 67.07257080078125, "min": -138.56866455078125, "p10": -32.91852722167968, "median": 21.38829803466797, "p90": 129.16998291015625, "max": 175.16085815429688, "pos_frac": 0.6640625, "sample": [139.22634887695312, 13.29656982421875, -6.73681640625, 108.9287109375, 59.575042724609375, -138.56866455078125, 1.3420257568359375, 95.72003173828125, -10.571533203125, 36.3043212890625, 162.34930419921875, 124.16000366210938, 46.7918701171875, -13.853645324707031, 119.29107666015625, 87.75816345214844, 4.0784149169921875, 34.793975830078125, 133.90191650390625, 112.44784545898438, 121.67343139648438, -11.467880249023438, 1.5838165283203125, -0.3187103271484375, 175.16085815429688, 0.134490966796875, -86.87620544433594, -97.65509033203125, -18.87701416015625, 0.6160888671875, 116.729248046875, 1.7635993957519531, 2.801177978515625, 51.7437744140625, 137.0206298828125, 153.81585693359375, 132.47987365722656, 59.27740478515625, -4.03363037109375, -70.01736450195312, 104.0313720703125, 33.13621520996094, 1.0517578125, 119.27969360351562, 41.042510986328125, -1.3098068237304688, -0.457366943359375, -25.6317138671875, 45.50059127807617, 62.001373291015625, -15.159088134765625, -2.85540771484375, 23.92205810546875, 86.66388702392578, 18.021560668945312, -29.33343505859375, 47.711761474609375, 102.5399169921875, 9.0189208984375, 21.257965087890625, 12.84423828125, 15.353099822998047, -10.211448669433594, 106.50518798828125, 20.1490478515625, 26.823883056640625, 21.518630981445312, 72.55615234375, 10.40673828125, 101.4158935546875, -32.243682861328125, 34.77513122558594, 1.560211181640625, -8.667160034179688, -18.40185546875, 144.5599365234375, -7.054840087890625, -46.66455078125, -0.85302734375, 90.60198974609375, -1.163665771484375, -12.020561218261719, -29.441253662109375, 126.6556396484375, 112.32901000976562, 118.87716674804688, 128.25267028808594, -13.938720703125, -133.2794189453125, 6.812347412109375, 64.43058776855469, -36.14360046386719, 75.60052490234375, -2.043182373046875, -77.64691162109375, 5.844963073730469, -23.383636474609375, -11.743988037109375, -76.81422424316406, 107.39300537109375, 9.184638977050781, -78.6282730102539, 41.823272705078125, 171.67828369140625, 133.3101806640625, 67.3763427734375, 130.13043212890625, 128.75836181640625, 134.62225341796875, 26.953323364257812, -60.68122863769531, 21.5294189453125, 104.57952880859375, 45.143890380859375, 22.387420654296875, -55.51220703125, -14.756141662597656, 7.731170654296875, -34.4931640625, 111.2479248046875, -16.98974609375, -9.926727294921875, 24.196151733398438, 92.84278106689453, -26.041275024414062, 52.257904052734375, 137.03912353515625, 127.64947509765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000290.npy"} {"epoch": 0.6073298429319371, "step": 291, "batch_size": 128, "mean": 51.652286529541016, "std": 72.10135650634766, "min": -160.60751342773438, "p10": -33.84735717773437, "median": 43.10198974609375, "p90": 142.32596130371093, "max": 181.40521240234375, "pos_frac": 0.7265625, "sample": [149.61660766601562, 0.7224044799804688, 173.57217407226562, 14.0142822265625, 166.99679565429688, 46.951385498046875, 140.98934936523438, 125.05569458007812, 42.04981994628906, -37.775543212890625, -4.16009521484375, 102.1719970703125, 33.77533721923828, -3.21307373046875, 146.05419921875, 11.320846557617188, 141.46420288085938, -16.361236572265625, 110.03964233398438, 28.305044174194336, 102.58840942382812, 79.35031127929688, 109.91940307617188, 20.45867919921875, -10.445541381835938, 125.99551391601562, 44.15415954589844, 5.7772216796875, 103.3372802734375, -28.306915283203125, -130.81239318847656, -21.386219024658203, -53.00099182128906, 21.972492218017578, -160.60751342773438, 27.96331787109375, 105.4603271484375, 87.7374267578125, 37.27294921875, 5.866241455078125, 142.48141479492188, -30.497268676757812, -83.57933044433594, 65.24050903320312, 138.29396057128906, 33.652069091796875, -10.947486877441406, 93.65547180175781, 56.75457763671875, 30.470855712890625, 139.06414794921875, 83.41752624511719, 93.72528839111328, -18.066925048828125, 171.2725830078125, 110.61143493652344, 67.22967529296875, -32.163848876953125, -51.52497863769531, 114.34039306640625, 150.63516235351562, 10.86126708984375, -7.91046142578125, -7.60345458984375, 134.31448364257812, -49.74603271484375, 6.340850830078125, 74.59561157226562, -64.21382141113281, 142.25933837890625, -62.94615173339844, 16.593994140625, 39.30936813354492, 181.40521240234375, 48.367279052734375, 25.5406494140625, 95.5142822265625, 121.4678955078125, 138.19384765625, -8.291107177734375, 71.0968017578125, 1.2933292388916016, -12.583648681640625, -19.54681396484375, 31.724685668945312, 152.21923828125, 88.34074401855469, -69.0404052734375, 0.0, -17.78008270263672, -53.55705261230469, 113.7779541015625, 22.512374877929688, -70.77111053466797, 176.98831176757812, 111.97181701660156, 92.38247680664062, 37.499114990234375, 144.509521484375, -26.02557373046875, 72.93191528320312, 124.63858032226562, 29.38995361328125, 112.01303100585938, -0.48760986328125, 68.10955810546875, 17.115875244140625, 34.41180419921875, 175.11767578125, 41.322509765625, -81.41448974609375, 115.24261474609375, -9.513038635253906, 85.88056182861328, 116.85845947265625, 63.359710693359375, 119.14019775390625, -1.229379653930664, 154.1021728515625, 134.14764404296875, 66.59674072265625, 29.960006713867188, 4.0450439453125, -11.078289031982422, 137.53298950195312, 138.63760375976562, 106.27369689941406, 50.37324523925781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000291.npy"} {"epoch": 0.6094240837696335, "step": 292, "batch_size": 128, "mean": 42.20623016357422, "std": 68.78230285644531, "min": -174.30303955078125, "p10": -40.3101203918457, "median": 34.96632385253906, "p90": 128.06856994628905, "max": 206.80810546875, "pos_frac": 0.7421875, "sample": [15.067779541015625, 2.494140625, 108.87564849853516, -61.75201416015625, 122.84688568115234, 109.28121185302734, 124.28591918945312, 180.0072021484375, 21.004913330078125, 124.29489135742188, 188.23492431640625, 95.62278747558594, -29.44329833984375, -4.774444580078125, 52.86297607421875, 26.720947265625, -50.466957092285156, -9.742246627807617, 15.15557861328125, -3.881591796875, 25.271163940429688, 60.99188232421875, 35.19366455078125, 66.8221435546875, 101.51705169677734, 57.391632080078125, 115.62373352050781, 26.803070068359375, 25.401275634765625, -4.433612823486328, 121.26904296875, 105.14810180664062, 2.333843231201172, 39.57379150390625, 104.45407104492188, 32.96091842651367, 31.57098388671875, -64.85812377929688, -171.08941650390625, 12.344024658203125, 104.08998107910156, 57.062767028808594, 49.627777099609375, 41.51092529296875, -5.508697509765625, 7.192619323730469, 63.78569030761719, -92.44210815429688, 0.5918197631835938, 136.95333862304688, -28.65869140625, 151.11810302734375, -69.68124389648438, -15.453857421875, -5.123291015625, 35.914886474609375, 0.809844970703125, -31.537002563476562, 24.66644287109375, 13.628616333007812, 109.55599212646484, 18.35546875, 122.48793029785156, -47.1597900390625, -11.876495361328125, 12.981796264648438, 12.704147338867188, 66.663330078125, 206.80810546875, 54.6156005859375, 6.934967041015625, 40.729705810546875, 79.25143432617188, 34.738983154296875, 109.29234313964844, 164.65740966796875, 24.354839324951172, 138.0369415283203, 27.471088409423828, -12.44998550415039, -53.441162109375, -174.30303955078125, 164.56341552734375, 87.93717956542969, 128.88766479492188, 91.22038269042969, 103.79046630859375, 48.50115966796875, 6.6678009033203125, 130.94976806640625, -35.795074462890625, 19.020294189453125, -24.11138916015625, 17.719482421875, 77.51911926269531, -38.940452575683594, -5.1031341552734375, -23.655838012695312, 18.93203353881836, 142.7962646484375, 92.20884704589844, 54.58434295654297, 20.13482666015625, 35.912559509277344, 28.849815368652344, 79.740966796875, 45.56184387207031, -1.2402496337890625, -1.0594329833984375, 13.343658447265625, -45.65325927734375, 62.767242431640625, 49.717254638671875, -0.15325927734375, 127.717529296875, 120.47109985351562, 176.25057983398438, -72.67611694335938, 83.39157104492188, 57.60820007324219, 144.64871215820312, -81.5516357421875, -43.506011962890625, 103.56234741210938, 91.03132629394531, 117.10723876953125, 67.0634765625, 45.722503662109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000292.npy"} {"epoch": 0.6115183246073298, "step": 293, "batch_size": 128, "mean": 43.290348052978516, "std": 70.330322265625, "min": -127.41998291015625, "p10": -29.310296630859373, "median": 27.16082763671875, "p90": 140.26486053466797, "max": 191.0693359375, "pos_frac": 0.7265625, "sample": [136.23841857910156, -11.772857666015625, 126.042724609375, 150.05780029296875, 23.85101318359375, -56.46429443359375, 18.002349853515625, 98.80450439453125, 3.1639480590820312, 44.728759765625, -5.3239593505859375, 106.64348602294922, 40.61102294921875, 23.223724365234375, 38.515541076660156, 14.927040100097656, -58.921119689941406, -0.619476318359375, -77.16757202148438, -21.948699951171875, 11.192543029785156, -2.4176025390625, 150.66485595703125, 116.66928100585938, 18.466094970703125, 0.0, 151.60821533203125, 39.80059814453125, 149.031494140625, 25.16521453857422, 116.29254150390625, -28.6962890625, -11.912826538085938, 6.8021697998046875, 16.23040771484375, 25.513916015625, -20.89739990234375, 117.70211791992188, -3.379486083984375, 25.345809936523438, 129.2115478515625, -1.2896099090576172, 98.2083740234375, 125.23220825195312, 37.86224365234375, -127.41998291015625, 138.88983154296875, 132.4861602783203, 170.75448608398438, 19.953189849853516, 66.46107482910156, 39.51409912109375, 116.02290344238281, 77.39422607421875, 4.67498779296875, 104.05186462402344, 26.466033935546875, -14.50091552734375, 5.818328857421875, 20.29107666015625, -6.598411560058594, 141.2880401611328, 61.4697265625, -60.95355224609375, -121.18047332763672, 48.85680389404297, 12.9132080078125, -30.74298095703125, 96.38337707519531, -9.293907165527344, -122.61550903320312, -35.7916259765625, 102.01158142089844, 191.0693359375, -99.34783935546875, 11.627655029296875, -9.67193603515625, 114.0521240234375, -23.408096313476562, 39.56072998046875, 27.855621337890625, 13.8260498046875, 46.93060302734375, 25.78636932373047, 139.82635498046875, 26.26043701171875, 47.086822509765625, 122.97811889648438, 115.8055419921875, -25.544845581054688, -18.55487060546875, 154.4295654296875, 13.450935363769531, 123.88688659667969, 49.469146728515625, 57.704803466796875, 66.85350036621094, -109.73765563964844, 40.10687255859375, 33.292205810546875, 87.83799743652344, 171.72219848632812, 184.653076171875, 158.38186645507812, 62.82318115234375, 122.60282897949219, 28.94287872314453, 33.28025817871094, -45.39778137207031, 3.2929916381835938, -0.5552597045898438, -18.782073974609375, -1.62481689453125, 75.01513671875, 186.73965454101562, 182.9254150390625, 23.23651123046875, 26.248733520507812, 112.70660400390625, -75.67555236816406, 90.81732177734375, -6.680824279785156, 7.508636474609375, 72.23728942871094, 18.869699478149414, 57.392059326171875, 39.843414306640625, 25.606231689453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000293.npy"} {"epoch": 0.6136125654450262, "step": 294, "batch_size": 128, "mean": 37.99412155151367, "std": 70.27153778076172, "min": -124.68060302734375, "p10": -50.259970855712886, "median": 20.918479919433594, "p90": 134.75824890136718, "max": 198.88671875, "pos_frac": 0.71875, "sample": [120.74151611328125, 112.6943359375, 121.62313842773438, -2.6472015380859375, -4.3813629150390625, 3.4411544799804688, 1.236612319946289, 5.6634521484375, 16.29871368408203, 135.02230834960938, 47.71649169921875, 149.75314331054688, 109.52920532226562, -6.0812835693359375, -4.673736572265625, 0.9934940338134766, 127.68684387207031, 158.35914611816406, 129.28933715820312, 23.630401611328125, 45.541259765625, 105.09169006347656, 100.10792541503906, -52.92247772216797, 13.970176696777344, 124.4993896484375, 12.889007568359375, -12.546875, 6.088134765625, 28.70281219482422, 158.07675170898438, 21.22998046875, 10.291549682617188, 93.02301025390625, -28.924102783203125, 22.946502685546875, -26.9322509765625, -91.73040771484375, 66.84153747558594, 20.606979370117188, 70.04261779785156, 154.11334228515625, -54.144775390625, -101.4331283569336, 141.06097412109375, 44.732093811035156, 101.01988983154297, -5.38671875, -5.007133483886719, 42.309478759765625, 139.96163940429688, -7.892616271972656, 149.30474853515625, 131.72488403320312, 57.862735748291016, 139.7752227783203, 139.21192932128906, -25.70916748046875, 17.990509033203125, 11.566764831542969, 101.3516845703125, 27.04241943359375, 26.855987548828125, 123.20526123046875, -117.27529907226562, 89.63101196289062, 105.48046875, 54.7244873046875, 14.624526977539062, 6.513427734375, 12.287445068359375, 156.8914337158203, 101.84136962890625, 0.0, 128.8553466796875, 198.88671875, -85.42903137207031, -8.042434692382812, -13.656005859375, 44.772186279296875, -6.101457595825195, 120.63946533203125, 7.323352813720703, -124.55967712402344, -14.930206298828125, 80.97906494140625, -34.811737060546875, 4.49420166015625, 100.37216186523438, -49.118896484375, 27.833724975585938, 7.498863220214844, -4.300743103027344, -121.8514404296875, 14.784011840820312, 83.39175415039062, 21.697616577148438, 2.3132095336914062, 13.6954345703125, -16.066848754882812, 13.897727966308594, 57.50017547607422, 15.052490234375, 18.48265838623047, 98.21484375, 61.22712707519531, -81.77301025390625, -78.05419921875, 32.32373046875, -0.6133766174316406, -64.79490661621094, 102.26289367675781, 94.19056701660156, 16.651336669921875, 79.99333190917969, 72.02267456054688, 3.79217529296875, -25.838340759277344, 95.19126892089844, -18.677650451660156, 12.045379638671875, -61.454315185546875, 73.65099334716797, 2.7667369842529297, 141.72439575195312, -124.68060302734375, 107.80602264404297, 134.64508056640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000294.npy"} {"epoch": 0.6157068062827226, "step": 295, "batch_size": 128, "mean": 44.62049102783203, "std": 73.995361328125, "min": -127.8211669921875, "p10": -31.52671661376953, "median": 23.76665496826172, "p90": 138.15762329101562, "max": 219.6497802734375, "pos_frac": 0.7578125, "sample": [150.50381469726562, 20.937835693359375, -32.06059265136719, -107.20573425292969, -4.365997314453125, 59.37724304199219, -28.177528381347656, 130.783447265625, -95.09408569335938, -9.926254272460938, -16.04901885986328, 13.068885803222656, 134.46759033203125, 19.64532470703125, -11.228363037109375, -14.9144287109375, -115.68482971191406, 103.03207397460938, -105.86773681640625, 34.444580078125, 84.05696105957031, 173.40447998046875, -15.192626953125, -127.8211669921875, -10.746063232421875, 108.22135162353516, 21.3856201171875, -82.32139587402344, 97.76861572265625, -2.1839447021484375, 5.27655029296875, 24.560043334960938, 50.17584228515625, 123.63595581054688, 120.12722778320312, 8.31390380859375, 107.61318969726562, 22.9732666015625, 73.79925537109375, 87.88482666015625, 176.65618896484375, -17.686185836791992, 67.69573974609375, 76.38278198242188, 1.4921875, 79.986572265625, 5.6630096435546875, 162.98931884765625, 132.52609252929688, 113.58291625976562, -115.89675903320312, -80.84129333496094, 132.77369689941406, 219.6497802734375, 132.5800323486328, 181.8934326171875, -23.101409912109375, 97.16510772705078, 53.273193359375, 15.804328918457031, -15.427520751953125, 14.60137939453125, 35.650634765625, 2.7808303833007812, 114.45635986328125, 57.41687774658203, 18.483421325683594, 67.28427124023438, 115.07640075683594, 19.844818115234375, 5.667900085449219, 9.793731689453125, 0.8233261108398438, 136.46719360351562, -31.29791259765625, 96.82008361816406, 134.2874755859375, 119.04270935058594, 13.748764038085938, 93.37610626220703, 9.807823181152344, 32.315673828125, 45.3629150390625, 35.64519500732422, 7.7619476318359375, 22.63916015625, 0.0, 75.16661071777344, 20.20123291015625, 15.531967163085938, 137.98248291015625, 165.27322387695312, 58.24041748046875, 41.320648193359375, 137.17276000976562, 1.5368995666503906, -62.0955810546875, 151.538330078125, 24.924644470214844, 137.44798278808594, 107.85206604003906, -16.870101928710938, 10.140174865722656, 21.05865478515625, 26.462554931640625, -96.87754821777344, 124.95502471923828, 13.190145492553711, 0.824859619140625, 65.69987487792969, -25.724132537841797, 130.9222869873047, 183.72833251953125, 138.5662841796875, 5.909820556640625, 164.54043579101562, 2.978168487548828, 7.242271423339844, 21.10376739501953, -45.61920166015625, -40.03021240234375, 5.1036376953125, 142.99468994140625, -5.17669677734375, 114.75262451171875, 141.8185577392578, 64.95367431640625, -28.9554443359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000295.npy"} {"epoch": 0.6178010471204188, "step": 296, "batch_size": 128, "mean": 60.175418853759766, "std": 73.32363891601562, "min": -146.03890991210938, "p10": -30.744609069824218, "median": 59.43311882019043, "p90": 152.5388168334961, "max": 179.256103515625, "pos_frac": 0.78125, "sample": [112.82430267333984, 60.931461334228516, 24.38397216796875, 114.30870056152344, 117.02159118652344, -16.58000946044922, -12.200592041015625, -1.7304553985595703, 79.5001449584961, 7.511928558349609, 97.86917114257812, -85.06991577148438, 44.68186950683594, 41.10382080078125, 9.17657470703125, 124.314697265625, 131.40777587890625, 92.5164794921875, 116.89945220947266, -39.7115478515625, 129.0018310546875, 93.35092163085938, -7.68865966796875, 122.93705749511719, 62.134033203125, 151.97918701171875, -6.117279052734375, 14.258602142333984, 141.7744140625, -41.238433837890625, 22.1280517578125, 57.934776306152344, 165.8414306640625, -0.08618927001953125, -116.37831115722656, 159.12261962890625, 48.54364013671875, 115.75846862792969, 137.89736938476562, 32.347557067871094, -15.940444946289062, 22.88427734375, 158.36968994140625, 127.4299087524414, 5.284889221191406, -31.4903564453125, 46.1436767578125, -30.425003051757812, 173.18963623046875, 17.637802124023438, 158.62356567382812, 72.14410400390625, -10.902252197265625, 140.79669189453125, 154.3544921875, 174.42474365234375, 10.4327392578125, 162.03945922851562, 48.85101318359375, 109.92950439453125, 11.29083251953125, -61.832969665527344, 106.76139831542969, -73.80499267578125, 54.44342041015625, 81.99320983886719, 29.26275634765625, 140.46307373046875, -119.58445739746094, 78.67791748046875, 133.85357666015625, 37.3240966796875, -37.297393798828125, 123.44518280029297, 90.26335144042969, 40.314910888671875, 165.15267944335938, 176.308837890625, -1.57940673828125, 111.43862915039062, 44.30548095703125, 125.030517578125, 17.255950927734375, -146.03890991210938, 149.8204345703125, 107.9095458984375, 18.424789428710938, -39.02012634277344, 149.94973754882812, -6.57855224609375, 117.06758117675781, 11.08636474609375, 99.7216796875, -41.620513916015625, 65.71356201171875, 0.0, 114.74932861328125, 120.4390869140625, 118.78811645507812, 83.02301025390625, 23.4705810546875, 130.22320556640625, 102.31698608398438, 124.5234375, 14.240623474121094, 48.808624267578125, 13.454437255859375, 12.15924072265625, 43.04656982421875, 123.37091064453125, 165.01708984375, 87.93870544433594, 17.642581939697266, -19.19647216796875, -2.976348876953125, 131.43756103515625, 126.50384521484375, 1.1183624267578125, 3.54339599609375, 153.84461975097656, 18.017982482910156, 82.9775390625, 23.27264404296875, 147.45123291015625, 90.65798950195312, 179.256103515625, -127.7420654296875, -5.2860870361328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000296.npy"} {"epoch": 0.6198952879581152, "step": 297, "batch_size": 128, "mean": 46.464027404785156, "std": 70.72743225097656, "min": -140.07858276367188, "p10": -26.389439392089837, "median": 32.03057861328125, "p90": 143.7706787109375, "max": 177.80816650390625, "pos_frac": 0.6796875, "sample": [169.53878784179688, 20.54802703857422, 125.66413879394531, 143.6063232421875, -15.990188598632812, -6.508171081542969, 129.3389892578125, 49.85697937011719, 142.74310302734375, -21.067901611328125, -1.42919921875, -4.48529052734375, -35.56889343261719, 4.4717254638671875, 155.13436889648438, -78.04063415527344, -9.358601570129395, 38.41448974609375, 102.87205505371094, 96.12338256835938, 10.870071411132812, 31.95599365234375, -4.566566467285156, 47.42938232421875, -82.25338745117188, -33.1993408203125, -22.6982421875, -140.07858276367188, 15.106689453125, -16.09195327758789, -46.05773162841797, 106.58383178710938, 27.708755493164062, 83.59967803955078, -12.14215087890625, 5.6617889404296875, 138.06576538085938, 1.250335693359375, 176.93310546875, 18.812469482421875, -0.6270217895507812, 29.806190490722656, 83.30587768554688, -30.68060302734375, 4.53594970703125, -18.50799560546875, -2.85455322265625, 24.38201904296875, -105.92263793945312, 124.3353271484375, 19.91571044921875, -22.390411376953125, -1.01202392578125, -33.4356689453125, 0.67315673828125, 177.80816650390625, 132.4576416015625, 109.58589172363281, -4.5365753173828125, -24.550369262695312, 72.716064453125, 123.83331298828125, -109.20846557617188, 26.273788452148438, 159.33413696289062, 39.12200927734375, 19.518218994140625, -72.3216552734375, 152.810302734375, -6.514862060546875, 31.88128662109375, 104.2725830078125, 147.79412841796875, 146.71180725097656, 166.50408935546875, -59.79486083984375, 47.67268753051758, 4.575828552246094, 64.84490966796875, -7.908882141113281, -21.435394287109375, 120.32176208496094, 67.73568725585938, -7.614776611328125, 2.160795211791992, 32.10516357421875, 34.281036376953125, -8.004142761230469, 137.96353149414062, 7.852294921875, 117.9996337890625, -11.1973876953125, 68.64871215820312, 119.74200439453125, 100.16949462890625, 92.92071533203125, 8.04168701171875, 134.89768981933594, -3.989410400390625, 159.1055908203125, 109.96389770507812, 108.05694580078125, 61.23396301269531, 131.7586669921875, -8.167243957519531, 101.96054077148438, 32.109344482421875, 36.68829345703125, 106.84137725830078, 125.25814819335938, 163.053955078125, 120.41793823242188, 20.207855224609375, 40.729034423828125, 0.26812744140625, -24.375778198242188, -5.682281494140625, 81.0411376953125, 121.249267578125, 32.4261474609375, 41.43163299560547, 128.26486206054688, 158.36065673828125, -39.948455810546875, -5.24658203125, 70.21417236328125, 116.26323699951172, 144.1541748046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000297.npy"} {"epoch": 0.6219895287958115, "step": 298, "batch_size": 128, "mean": 52.4852294921875, "std": 77.847900390625, "min": -150.09295654296875, "p10": -40.774072265625, "median": 32.941162109375, "p90": 149.76993103027343, "max": 285.65899658203125, "pos_frac": 0.7265625, "sample": [196.93406677246094, 120.68392944335938, -20.47857666015625, 12.121170043945312, -40.28759765625, -41.9091796875, -50.99713134765625, -16.68670654296875, -24.600830078125, 150.23513793945312, 71.84123229980469, -12.100067138671875, 7.8542022705078125, 15.895904541015625, 113.92076110839844, -67.51858520507812, 131.17666625976562, 94.25482177734375, 22.7923583984375, 102.20830535888672, 136.70016479492188, 133.49984741210938, 37.451904296875, 192.8916015625, 7.280517578125, -33.71624755859375, 123.55912780761719, -92.06619262695312, 120.43534851074219, 50.00990295410156, 139.60980224609375, -35.350616455078125, 121.30120849609375, 84.73579406738281, 33.4342041015625, -55.3577880859375, 8.541259765625, 69.69049072265625, 133.69241333007812, -45.01756286621094, 118.06819152832031, -2.2637939453125, 119.26309204101562, -21.859588623046875, 41.091278076171875, 161.56912231445312, 41.4822998046875, 32.94268798828125, -46.04142761230469, 2.3480072021484375, 61.87062072753906, 130.98150634765625, 20.76287841796875, 140.78497314453125, -75.86337280273438, 43.611663818359375, 122.0411605834961, 147.8197021484375, 14.1295166015625, 113.10154724121094, 178.73068237304688, -150.09295654296875, 126.00625610351562, -12.082260131835938, -16.717796325683594, 186.11770629882812, 151.0065155029297, 68.66241455078125, -42.9964599609375, 0.0, 5.766532897949219, 8.347358703613281, 174.334716796875, 18.657989501953125, 27.814598083496094, -0.5125617980957031, 119.13874816894531, 61.348785400390625, 8.398551940917969, 126.68423461914062, -34.97764587402344, -12.607831954956055, 14.744361877441406, 83.95948791503906, 123.30663299560547, 6.009130477905273, 140.77989196777344, 131.65509033203125, 131.4268798828125, 7.238922119140625, 200.985107421875, 41.20489501953125, -64.26668548583984, -2.367584228515625, 12.766830444335938, 32.93963623046875, 82.5653076171875, 13.56744384765625, -7.976173400878906, -0.76300048828125, -8.66107177734375, 112.654052734375, 56.40232849121094, 132.70947265625, 188.22061157226562, 164.01779174804688, -18.21331787109375, 102.73811340332031, 285.65899658203125, 18.791091918945312, 138.13772583007812, 32.62213134765625, 3.9044189453125, 31.385177612304688, 149.570556640625, 21.28594970703125, -11.996185302734375, 114.08883666992188, 21.30426025390625, 167.9783935546875, 5.063880920410156, 96.03610229492188, 17.488677978515625, 20.03631591796875, -80.14822387695312, -36.4957275390625, 68.19522094726562, -75.97477722167969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000298.npy"} {"epoch": 0.6240837696335079, "step": 299, "batch_size": 128, "mean": 41.026763916015625, "std": 66.04656219482422, "min": -146.24032592773438, "p10": -31.010995483398435, "median": 26.363754272460938, "p90": 130.17902526855468, "max": 176.54702758789062, "pos_frac": 0.7109375, "sample": [89.50752258300781, 131.89657592773438, 14.126091003417969, 87.84845733642578, 27.612991333007812, 2.53375244140625, 134.16281127929688, -97.697265625, 110.2032470703125, -146.24032592773438, 153.28713989257812, 52.347991943359375, 19.28296661376953, 105.656982421875, -35.7462158203125, 127.07382202148438, -2.095794677734375, 23.15106964111328, 11.033050537109375, -125.99253845214844, 123.49361419677734, -1.533416748046875, 69.68203735351562, -33.40623474121094, 136.987060546875, -4.912841796875, -50.84751892089844, -29.249359130859375, 113.46730041503906, 35.18085861206055, 83.62860107421875, -5.4202880859375, 137.10081481933594, -2.3083648681640625, -91.43649291992188, 50.62495803833008, 111.55133056640625, -11.941757202148438, -71.97724151611328, 109.6070556640625, 28.136627197265625, 159.01954650878906, 25.407257080078125, 10.567007064819336, -6.474540710449219, -0.12579345703125, 139.14334106445312, -4.562658309936523, -15.30780029296875, -90.33668518066406, -89.240234375, 11.14837646484375, 7.33613395690918, 16.245651245117188, 106.34173583984375, -37.485687255859375, 141.08160400390625, -5.99969482421875, 147.37713623046875, 45.666481018066406, 114.18933868408203, 1.528961181640625, 18.0584716796875, 88.21026611328125, -3.3349647521972656, 176.54702758789062, 8.095794677734375, 9.64979362487793, 104.742431640625, 89.55368041992188, 92.53416442871094, 9.217464447021484, 58.62481689453125, 7.516204833984375, 114.47065734863281, 107.22120666503906, -3.1448745727539062, -21.189632415771484, -6.4614105224609375, 122.39581298828125, -2.55963134765625, 78.45989990234375, 35.25971984863281, 14.686737060546875, 41.28825378417969, -8.499221801757812, 133.6982421875, 44.36614990234375, 73.63041687011719, 99.68132019042969, 16.877166748046875, 87.74276733398438, 123.4993896484375, 9.78497314453125, 14.98638916015625, -9.6031494140625, 40.47199630737305, 5.295066833496094, 116.02397918701172, -16.451515197753906, -30.13995361328125, 53.06343078613281, 33.92701721191406, -46.6724853515625, 103.71275329589844, 172.65484619140625, 17.45819091796875, 129.44293212890625, 27.32025146484375, -2.61083984375, 15.62799072265625, 2.312845230102539, 117.3542709350586, -33.043426513671875, 7.7603759765625, 11.276824951171875, -3.4503936767578125, 32.856201171875, 55.75907897949219, 41.78692626953125, -14.00396728515625, 101.82327270507812, 144.80577087402344, 110.00619506835938, 36.32733154296875, 117.87362670898438, 23.72991943359375, 99.22219848632812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000299.npy"} {"epoch": 0.6261780104712041, "step": 300, "batch_size": 128, "mean": 42.55999755859375, "std": 75.57134246826172, "min": -123.240966796875, "p10": -55.01024475097656, "median": 39.346492767333984, "p90": 133.54898376464843, "max": 245.50372314453125, "pos_frac": 0.6796875, "sample": [96.30264282226562, 110.69601440429688, 19.580509185791016, -82.93406677246094, 38.532310485839844, -26.609893798828125, 111.07046508789062, 50.46827697753906, 19.74540901184082, -47.34228515625, 154.6114501953125, 130.78567504882812, 137.63436889648438, 5.953395843505859, 1.3782730102539062, 135.10067749023438, -106.32949829101562, 32.388092041015625, 63.021575927734375, -8.134429931640625, 150.27890014648438, 62.96356201171875, 56.621185302734375, 83.48982238769531, 87.46498107910156, 8.96514892578125, 40.160675048828125, -17.528076171875, 14.85003662109375, -32.558265686035156, 109.58905029296875, 109.84869384765625, 130.32125854492188, 79.33917236328125, 99.76034545898438, -57.35185241699219, -123.240966796875, 16.294647216796875, 4.096936225891113, 180.51931762695312, 67.1187744140625, -20.535888671875, -38.3958740234375, -0.358642578125, 102.20549011230469, 127.39938354492188, -54.00669860839844, 130.966552734375, 100.79539489746094, 165.90872192382812, 20.82512664794922, -115.76963806152344, -108.55830383300781, 58.62921142578125, 31.705806732177734, -4.998006820678711, 49.336212158203125, 52.17094421386719, 133.9075927734375, -11.262809753417969, 122.86038970947266, 125.95550537109375, -47.76666259765625, 153.78550720214844, 52.7705078125, -10.0714111328125, 3.5325927734375, 98.86427307128906, 98.86933898925781, 127.85646057128906, 58.59674072265625, 100.35137939453125, 0.16964340209960938, 24.980560302734375, -19.401046752929688, -91.75807189941406, 163.68983459472656, 54.51904296875, -74.46940612792969, 130.51229858398438, 133.39529418945312, 13.460407257080078, 48.7342529296875, -9.130874633789062, 68.04265594482422, 22.38275146484375, -34.6961669921875, 0.0, -13.640243530273438, 32.28675842285156, 165.03500366210938, -0.53076171875, -120.55612182617188, 58.189208984375, 2.8902435302734375, 133.13662719726562, 78.08760070800781, -2.82244873046875, 102.1890869140625, 123.96282958984375, 7.9232330322265625, 245.50372314453125, 6.334991455078125, 82.49980163574219, -64.59320068359375, 99.98077392578125, 133.31167602539062, -92.4932861328125, -36.17022705078125, -16.450729370117188, 98.07534790039062, -9.603006362915039, 67.1185302734375, -22.802642822265625, 52.5548095703125, 113.49330139160156, 84.03347778320312, 13.628005981445312, -66.10617065429688, -2.9781494140625, 172.4222412109375, -0.43463134765625, -68.45260620117188, -28.09112548828125, 128.50250244140625, 149.83938598632812, -7.320587158203125, 6.803680419921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000300.npy"} {"epoch": 0.6282722513089005, "step": 301, "batch_size": 128, "mean": 42.32103729248047, "std": 72.37611389160156, "min": -177.36947631835938, "p10": -39.295484924316405, "median": 27.774635314941406, "p90": 144.22729187011717, "max": 206.86431884765625, "pos_frac": 0.7421875, "sample": [15.188308715820312, 168.20751953125, 128.24114990234375, 82.96392822265625, 21.637962341308594, 131.15414428710938, 46.583274841308594, -48.20301818847656, 83.07969665527344, 8.369401931762695, 3.4848499298095703, -0.0283203125, 114.21246337890625, 11.726531982421875, 1.4110107421875, 5.368743896484375, -57.9998779296875, -41.99896240234375, 107.52314758300781, 159.7015380859375, 15.982791900634766, 17.220596313476562, 68.12335205078125, 25.52978515625, 34.53886413574219, 17.5291748046875, -1.47393798828125, 54.076904296875, 11.976776123046875, 52.03009033203125, 82.37847900390625, -7.6185302734375, 0.0, 157.26763916015625, 125.881103515625, 80.20655059814453, -13.6348876953125, 35.98040771484375, -26.969860076904297, 121.85710144042969, -22.29058837890625, 118.51770782470703, 156.0923309326172, 22.669578552246094, -147.6876220703125, 1.57794189453125, -51.55859375, 110.35409545898438, -30.5164794921875, 24.654296875, 44.517921447753906, -12.135009765625, -39.68174743652344, 139.6737823486328, 162.30023193359375, 77.98560333251953, 37.40046691894531, 156.6981201171875, 1.386688232421875, 206.86431884765625, 26.259483337402344, 113.53152465820312, 19.126800537109375, 40.65313720703125, -50.75250244140625, -68.04396057128906, 16.7923583984375, -177.36947631835938, -121.06997680664062, 80.27516174316406, 48.69750213623047, 63.21051025390625, -18.677825927734375, 26.856128692626953, 10.344482421875, 27.8363037109375, 137.11831665039062, 31.728729248046875, 147.86065673828125, -79.3857421875, 1.5099029541015625, 0.7685699462890625, 171.9224090576172, 114.7755126953125, 148.89947509765625, 29.2689208984375, -4.811305999755859, -30.455429077148438, 114.41555786132812, -111.20494079589844, 134.19235229492188, 35.824462890625, -19.07305908203125, 2.5294647216796875, 122.139404296875, -20.5399169921875, 73.34121704101562, 45.625762939453125, 84.54157257080078, 110.19976806640625, 113.39973449707031, 134.45257568359375, 152.14773559570312, -10.356689453125, 28.56037139892578, 142.67013549804688, 77.87945556640625, -23.69500160217285, 157.4847412109375, -22.9993896484375, 21.89105224609375, 13.380157470703125, -10.257080078125, 8.213638305664062, 4.565757751464844, 100.30304718017578, -58.803497314453125, 31.431516647338867, 3.8977394104003906, 130.19052124023438, 164.224609375, -39.12994384765625, -19.790802001953125, 80.3732681274414, 27.712966918945312, 70.42498779296875, 103.27000427246094, 8.46051025390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000301.npy"} {"epoch": 0.6303664921465969, "step": 302, "batch_size": 128, "mean": 50.051368713378906, "std": 74.1887435913086, "min": -144.90164184570312, "p10": -32.13536834716796, "median": 46.70146179199219, "p90": 141.17579040527343, "max": 180.135986328125, "pos_frac": 0.7421875, "sample": [140.9652557373047, 53.74017333984375, 107.61407470703125, 136.2890625, 22.921539306640625, 99.24626922607422, 0.0, -8.372634887695312, -1.123321533203125, -58.50910949707031, -127.2130126953125, 110.31756591796875, 88.37076568603516, 135.52011108398438, -13.72064208984375, 26.876991271972656, 62.69293212890625, 58.12017822265625, -17.012649536132812, -35.58116149902344, 151.71383666992188, 81.2552490234375, 162.3056640625, -15.11871337890625, 120.49685668945312, 130.9151611328125, 180.135986328125, 17.080841064453125, -100.592041015625, 107.2027816772461, 22.502273559570312, 108.28628540039062, 119.61399841308594, 107.11441040039062, 74.0999755859375, 91.122802734375, 167.65341186523438, 85.912109375, 46.09564208984375, 46.72003173828125, 96.89590454101562, -2.024749755859375, -144.90164184570312, -2.21148681640625, 85.54017639160156, 75.89373779296875, 172.80291748046875, 35.895233154296875, 160.55322265625, 138.9295654296875, -13.63958740234375, 124.86534118652344, 13.665580749511719, 3.371917724609375, -8.767646789550781, -82.76222229003906, 97.00732421875, 11.108856201171875, 112.33480834960938, 36.75361633300781, 92.6204833984375, 90.5493392944336, 41.6158447265625, 8.797767639160156, 38.396240234375, 7.06085205078125, 14.37518310546875, 11.145965576171875, -5.516155242919922, -93.7187271118164, 111.4069595336914, 148.10708618164062, -19.095977783203125, -37.98468017578125, 136.553955078125, 115.58999633789062, -6.145355224609375, 93.28741455078125, 108.21275329589844, 90.8221206665039, 25.629684448242188, 51.665374755859375, 130.33786010742188, -107.739501953125, 143.10174560546875, -8.14642333984375, -138.5762939453125, 54.21983337402344, -30.658599853515625, -103.23314666748047, 18.115257263183594, -106.36444091796875, 6.24169921875, 30.15802001953125, 4.97041130065918, 46.682891845703125, 96.97483825683594, 141.7755126953125, 139.10702514648438, -52.89311218261719, 3.7506103515625, 24.1630859375, 58.19091033935547, -16.752853393554688, 120.39606475830078, 1.0424461364746094, 31.945587158203125, 174.48834228515625, 61.46256637573242, 109.91500854492188, -4.56829833984375, 12.154335021972656, 49.53192138671875, -15.604095458984375, 109.79000854492188, 34.97019958496094, 174.64736938476562, 3.2217140197753906, 124.23944091796875, 38.015106201171875, 21.25152587890625, 172.898681640625, 88.32046508789062, 134.5401153564453, 95.31800842285156, -10.798370361328125, 141.6670379638672, -16.04144287109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000302.npy"} {"epoch": 0.6324607329842932, "step": 303, "batch_size": 128, "mean": 45.449241638183594, "std": 76.21871948242188, "min": -147.5452880859375, "p10": -39.19031066894531, "median": 45.45636749267578, "p90": 133.36600646972656, "max": 196.1505126953125, "pos_frac": 0.78125, "sample": [16.623748779296875, 42.86389923095703, -133.87216186523438, -114.25888061523438, 113.38534545898438, -10.373825073242188, 108.11102294921875, 59.506134033203125, 127.32272338867188, 106.79419708251953, -30.78277587890625, 123.29679107666016, 140.86666870117188, 129.5526123046875, 39.905120849609375, -82.76394653320312, -18.652008056640625, 110.48471069335938, -6.923397064208984, 148.06903076171875, 104.82321166992188, 35.80682373046875, 115.06497955322266, 42.861236572265625, 53.870147705078125, -38.479888916015625, 76.94505310058594, -10.603790283203125, 155.11843872070312, 9.748527526855469, 144.6651611328125, 130.41514587402344, -40.84796142578125, 128.94032287597656, -12.757217407226562, 11.99609375, 20.093414306640625, 50.4989013671875, 0.0, 13.688072204589844, -101.1437759399414, 31.441848754882812, -13.084480285644531, 9.748794555664062, 10.351112365722656, 5.9757080078125, 126.36976623535156, 44.912139892578125, 90.04414367675781, 111.66629028320312, 123.28105163574219, -116.85685729980469, 118.92904663085938, -141.53961181640625, 113.12326049804688, 134.6265106201172, 146.61746215820312, 86.16851806640625, 44.01727294921875, 118.3411865234375, 0.8025932312011719, -3.7473602294921875, 22.321044921875, -10.38165283203125, -66.183349609375, 6.4671630859375, 98.46621704101562, 73.310546875, 132.82579040527344, 81.22347259521484, 149.13388061523438, -117.09213256835938, 17.913970947265625, 5.6876220703125, 128.07965087890625, 107.78239440917969, 0.243011474609375, 81.59164428710938, 132.58724975585938, 188.5965576171875, 113.65534210205078, -147.5452880859375, 107.21015930175781, -124.49380493164062, 5.416492462158203, -4.16876220703125, 106.92108154296875, 50.766754150390625, 59.7825927734375, 82.1824951171875, 11.664337158203125, -7.3018798828125, 142.97698974609375, 3.27239990234375, 142.58367919921875, 2.4831085205078125, 6.249931335449219, 49.80426025390625, -29.72674560546875, 7.543060302734375, -29.052825927734375, -142.817626953125, 52.33746337890625, 0.01416015625, 5.882347106933594, 98.03008270263672, 136.632568359375, 93.58425903320312, 130.02052307128906, 102.36343383789062, 49.989105224609375, 62.060272216796875, 51.28611755371094, 117.95761108398438, 0.513763427734375, 196.1505126953125, 74.09939575195312, -97.68295288085938, 92.68051147460938, 17.982574462890625, 153.433837890625, 16.401275634765625, 112.5711669921875, 8.394668579101562, 31.867340087890625, 8.0196533203125, 15.887542724609375, 46.00059509277344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000303.npy"} {"epoch": 0.6345549738219896, "step": 304, "batch_size": 128, "mean": 40.347557067871094, "std": 73.30216217041016, "min": -137.34942626953125, "p10": -51.00401153564453, "median": 28.0328369140625, "p90": 140.25103454589845, "max": 175.166748046875, "pos_frac": 0.6953125, "sample": [0.0, 31.059478759765625, -13.741912841796875, 55.16375732421875, -30.561904907226562, -1.816192626953125, -128.17391967773438, 120.48974609375, 131.18084716796875, -22.24102783203125, 77.4122543334961, 123.71728515625, 84.72822570800781, 85.05256652832031, -88.7739486694336, 43.28607177734375, 15.73333740234375, 144.78903198242188, 33.25629425048828, 11.3858642578125, 12.51617431640625, 28.435012817382812, -0.5235443115234375, -13.082748413085938, 103.45909118652344, 125.87640380859375, 119.41888427734375, 21.225494384765625, 15.604736328125, 2.5332794189453125, -84.89190673828125, 14.488363265991211, -2.6287384033203125, 6.488288879394531, 35.18089294433594, 23.51104736328125, 41.70020294189453, -17.41094207763672, 118.25814819335938, 98.43954467773438, -13.752105712890625, 99.64228820800781, 122.23005676269531, -49.281585693359375, 164.75698852539062, 23.25994873046875, -4.841102600097656, 104.8870849609375, -82.14358520507812, 40.03363037109375, 36.7578125, -25.883682250976562, 44.521728515625, -137.34942626953125, 0.0, 141.08493041992188, -66.07354736328125, 96.06805419921875, 7.72930908203125, -26.889617919921875, -58.72406005859375, 23.097000122070312, -105.74713134765625, 84.92152404785156, 20.46221923828125, 44.225311279296875, -35.766937255859375, 46.8572998046875, 175.166748046875, 120.71533203125, -108.05441284179688, 20.7740478515625, 39.15287780761719, -50.9405517578125, -52.57586669921875, 155.88113403320312, 147.09852600097656, -3.6808700561523438, 140.77084350585938, -99.83642578125, 129.5421142578125, 109.12905883789062, -22.35517120361328, -5.066989898681641, 168.780517578125, 16.892898559570312, 141.07382202148438, -5.9310302734375, 129.23397827148438, -127.2479248046875, 19.031463623046875, -0.43096160888671875, 24.14752197265625, 140.02825927734375, 148.42567443847656, 10.272214889526367, -18.602996826171875, 119.8951416015625, 132.5229949951172, 145.84854125976562, 32.067291259765625, 34.41253662109375, 50.407981872558594, 9.83563232421875, 44.32872772216797, 133.184326171875, 27.630661010742188, 160.81204223632812, 122.10542297363281, 126.80349731445312, 104.95147705078125, 47.55194091796875, -51.15208435058594, 4.32275390625, 79.12022399902344, 25.878631591796875, -8.6995849609375, -14.259506225585938, 118.58743286132812, 0.8020172119140625, 37.732582092285156, 104.98522186279297, 1.44091796875, 126.01373291015625, -17.271896362304688, 130.11752319335938, 22.097442626953125, 156.393798828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000304.npy"} {"epoch": 0.6366492146596858, "step": 305, "batch_size": 128, "mean": 59.329959869384766, "std": 76.53275299072266, "min": -179.36276245117188, "p10": -17.009441375732408, "median": 50.55120086669922, "p90": 149.05870361328124, "max": 274.6192626953125, "pos_frac": 0.8203125, "sample": [9.130355834960938, 18.18252182006836, -12.556350708007812, 195.02468872070312, 36.3795166015625, 4.6224365234375, 109.42300415039062, -56.04420471191406, 122.21861267089844, 21.38945770263672, 72.21674346923828, 30.651504516601562, 0.9009857177734375, 46.46931457519531, 18.618728637695312, -72.5528564453125, -77.40847778320312, 25.543060302734375, 50.30900573730469, 0.5968666076660156, 64.98785400390625, 140.00448608398438, 2.9260330200195312, 65.18000793457031, 6.4720458984375, 18.214584350585938, -108.82064819335938, -4.1074066162109375, 31.763320922851562, 114.84085083007812, 21.11639404296875, 89.37286376953125, 116.88177490234375, 83.66748046875, 144.30645751953125, 19.901870727539062, -0.48593902587890625, -3.94927978515625, 130.66946411132812, -1.221588134765625, -138.1392364501953, 92.36886596679688, -4.463714599609375, 105.10120391845703, 146.882568359375, 122.98951721191406, 50.79339599609375, 153.49075317382812, 134.72842407226562, 38.114532470703125, 0.2693939208984375, -110.45631408691406, 161.38259887695312, 153.75442504882812, 36.282684326171875, 51.56158447265625, 274.6192626953125, 82.33502197265625, 175.39801025390625, 144.90933227539062, 30.973373413085938, 1.3223152160644531, -8.82598876953125, 147.6695556640625, 38.776641845703125, 110.79218292236328, 150.58807373046875, 121.52973937988281, 120.3404541015625, 120.64126586914062, -27.399986267089844, 120.16605377197266, -27.666481018066406, 38.25212097167969, 126.36683654785156, -179.36276245117188, 30.130325317382812, 123.01856994628906, 10.58856201171875, 103.92282104492188, 57.32550048828125, 148.40325927734375, 156.8523406982422, 10.2783203125, 69.81402587890625, 113.4197998046875, -30.482635498046875, 194.3673095703125, 3.339252471923828, -4.3717193603515625, 154.8284912109375, 72.95263671875, 56.299835205078125, 44.97126770019531, 102.54318237304688, 126.64801025390625, 80.54713439941406, 125.33294677734375, 174.429931640625, 55.131591796875, 35.087799072265625, -2.98773193359375, 122.74017333984375, 118.76046752929688, 193.0234375, 24.509735107421875, 32.16064453125, 28.6077880859375, 7.768157958984375, 31.71697998046875, 147.21636962890625, 3.0271873474121094, 63.85206604003906, 3.61767578125, 46.45216369628906, -81.18543243408203, 148.27108764648438, 4.217704772949219, 11.051864624023438, -53.744842529296875, 131.35079956054688, 52.629150390625, 186.84097290039062, 129.483154296875, 130.4510498046875, -70.23112487792969, -5.392822265625, 143.7282257080078], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000305.npy"} {"epoch": 0.6387434554973822, "step": 306, "batch_size": 128, "mean": 51.227684020996094, "std": 73.60459899902344, "min": -142.0059814453125, "p10": -20.01675338745117, "median": 53.16266632080078, "p90": 145.62318420410156, "max": 181.93423461914062, "pos_frac": 0.734375, "sample": [18.316802978515625, -100.9049072265625, -101.99462127685547, 179.11221313476562, -142.0059814453125, 175.807861328125, -11.545225143432617, 55.08234405517578, -11.780792236328125, 135.74322509765625, 22.66350555419922, 24.37384033203125, 138.97767639160156, 67.62295532226562, -3.5284576416015625, 11.942977905273438, -37.7928466796875, 122.509765625, -4.209342956542969, 25.0269775390625, 43.58282470703125, 47.37602996826172, -56.226959228515625, 52.81781005859375, -137.0332489013672, 162.93280029296875, 158.940673828125, 124.75439453125, 112.22283935546875, -0.7935104370117188, 59.328460693359375, 87.53762817382812, 23.999053955078125, 124.42666625976562, 145.98046875, 124.15300750732422, 166.7471923828125, 124.02899169921875, 76.14157104492188, 100.79379272460938, 7.612483978271484, 113.76311492919922, -6.345916748046875, 9.203811645507812, 79.88095092773438, 55.713165283203125, 60.147491455078125, 112.75532531738281, 54.045074462890625, -81.89483642578125, -12.010833740234375, 140.83607482910156, -19.58264923095703, 178.93804931640625, 54.940826416015625, -136.01702880859375, 83.21704864501953, 65.33636474609375, -22.516250610351562, 175.36895751953125, -3.3077926635742188, 61.58941650390625, 94.11285400390625, 31.570648193359375, 120.08174896240234, 4.388986587524414, 6.510345458984375, 4.081745147705078, 106.75094604492188, -49.81993103027344, 107.06361389160156, -101.30068969726562, 108.70867919921875, 41.29180908203125, 174.14056396484375, -1.5952930450439453, -1.381072998046875, 27.31787109375, 17.855545043945312, 32.82257080078125, 24.131439208984375, 57.900054931640625, 104.62271118164062, 96.05459594726562, 146.939453125, -15.753768920898438, -18.9888916015625, 60.404937744140625, 82.78497314453125, 181.93423461914062, 64.85488891601562, 53.50752258300781, 18.984634399414062, -21.0296630859375, 22.54974365234375, 98.77462768554688, 114.247802734375, -9.004440307617188, 107.99844360351562, -17.963899612426758, 140.21856689453125, 68.82260131835938, 134.6664581298828, 145.47006225585938, 73.40045166015625, 27.238731384277344, 10.371391296386719, -7.2213287353515625, -2.33099365234375, -4.4095916748046875, -8.400775909423828, 4.50665283203125, 138.6887664794922, 117.842529296875, 13.301025390625, 169.96923828125, 10.77961540222168, -3.2961273193359375, 1.320037841796875, 0.7216796875, 112.07327270507812, 115.24696350097656, -0.1138458251953125, -89.8885726928711, 105.11304473876953, 38.689178466796875, 143.63821411132812, 148.37509155273438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000306.npy"} {"epoch": 0.6408376963350786, "step": 307, "batch_size": 128, "mean": 56.46784973144531, "std": 71.49482727050781, "min": -178.42752075195312, "p10": -20.163752746582027, "median": 52.971479415893555, "p90": 140.2003402709961, "max": 187.7293701171875, "pos_frac": 0.78125, "sample": [132.76467895507812, 133.74012756347656, 14.539535522460938, 159.28494262695312, -63.67413330078125, -5.97296142578125, 79.7986831665039, 103.7781982421875, 97.364013671875, 130.36468505859375, -158.30023193359375, -3.5021419525146484, 73.64434814453125, 1.0745391845703125, -13.235595703125, 117.39593505859375, 131.11300659179688, 116.65855407714844, 3.3112220764160156, 18.713623046875, 59.417755126953125, -40.32989501953125, 119.74545288085938, 10.005598068237305, 170.189208984375, -25.933349609375, 61.2303466796875, 11.6490478515625, 121.12176513671875, 159.58251953125, 91.58616638183594, 24.201141357421875, 28.65185546875, 83.62602996826172, 106.60752868652344, -17.260940551757812, 142.84657287597656, 68.17947387695312, 19.17138671875, -6.304985046386719, 24.011505126953125, 86.24734497070312, 119.75432586669922, -13.3433837890625, 129.2569580078125, 97.8177490234375, 137.8071746826172, -9.98004150390625, 1.5638580322265625, 25.675445556640625, 82.90922546386719, 0.8322582244873047, 139.4164276123047, 8.828351974487305, 70.83082580566406, 120.59056091308594, 109.83392333984375, 147.8352813720703, 17.586448669433594, 3.9535675048828125, 16.47430419921875, 79.61161041259766, -64.56948852539062, 109.90011596679688, -6.4886474609375, -3.6995811462402344, 127.10813903808594, 40.750694274902344, -7.009788513183594, 6.575366973876953, -8.753812789916992, 133.26683044433594, -39.510009765625, -34.54864501953125, 31.3834228515625, -19.465530395507812, 26.097747802734375, 130.75279235839844, 102.53662109375, 25.609588623046875, 139.8693084716797, -21.792938232421875, 46.525203704833984, 152.18145751953125, 161.07583618164062, 123.15646362304688, -29.877552032470703, 125.33990478515625, 12.333080291748047, 91.78118896484375, 19.615684509277344, -41.26615905761719, 78.65669250488281, 166.16317749023438, 101.29209899902344, 17.46240234375, 129.30557250976562, -130.72006225585938, -10.073165893554688, 125.078125, 6.18212890625, 176.15560913085938, 32.3055419921875, 151.010009765625, -178.42752075195312, 6.4501953125, 3.15338134765625, 96.23667907714844, 37.919342041015625, 90.38909149169922, -42.84088134765625, 23.6126708984375, 187.7293701171875, 117.68486785888672, 29.830413818359375, 100.02859497070312, -0.5904541015625, -9.21044921875, 38.58396911621094, 131.94207763671875, 92.2108154296875, 14.584548950195312, 118.22360229492188, 133.07577514648438, 151.074462890625, 7.340839385986328, 140.97274780273438, 111.85781860351562], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000307.npy"} {"epoch": 0.6429319371727749, "step": 308, "batch_size": 128, "mean": 37.36277389526367, "std": 73.1423110961914, "min": -162.86709594726562, "p10": -46.71914138793945, "median": 20.499969482421875, "p90": 129.48658752441406, "max": 197.42544555664062, "pos_frac": 0.7265625, "sample": [143.62437438964844, 45.00749969482422, -65.23297119140625, 141.2390594482422, 98.569580078125, -97.17070007324219, -49.8092041015625, 50.888397216796875, 9.383758544921875, 35.502471923828125, 5.15155029296875, 106.27114868164062, 0.8072280883789062, -76.67327880859375, -5.84527587890625, 3.22894287109375, 117.99978637695312, 1.0776519775390625, 120.33587646484375, 86.3350830078125, 124.6513442993164, 34.50221252441406, -0.641326904296875, -76.08611297607422, 12.394195556640625, 104.45442199707031, -150.734375, -34.74089813232422, 16.587066650390625, 141.63504028320312, 113.68257141113281, 22.977890014648438, 36.5838623046875, 88.85087585449219, -16.39776611328125, -7.61810302734375, 125.3031005859375, 6.6038818359375, 103.9207763671875, 18.09197998046875, 9.738212585449219, 39.93547821044922, 184.26934814453125, 5.16748046875, -35.561767578125, -6.03411865234375, -0.517578125, 134.5322265625, -4.2092132568359375, 118.2261962890625, 5.53076171875, 23.354202270507812, 4.352054595947266, 109.25281524658203, -7.181007385253906, 28.698089599609375, 129.22406005859375, 21.40190887451172, -162.86709594726562, 123.04351806640625, 63.41987609863281, 15.943328857421875, -7.37835693359375, -16.2425537109375, 197.42544555664062, 115.86375427246094, -2.350709915161133, -122.07485961914062, 14.135406494140625, 2.9644088745117188, 99.98365783691406, -89.92416381835938, 141.00149536132812, 77.99234771728516, 11.180572509765625, 28.404205322265625, 159.58389282226562, 179.76861572265625, 145.78628540039062, -1.6936073303222656, 110.04692077636719, -14.107177734375, 1.9264106750488281, 6.8511962890625, 71.93609619140625, -126.79547119140625, 19.59803009033203, 8.719207763671875, -52.571136474609375, 102.78656005859375, 21.57781982421875, 15.966560363769531, 35.58599853515625, 10.520126342773438, 118.34202575683594, 110.87641143798828, 152.78640747070312, 130.09915161132812, 33.98683166503906, 120.80462646484375, 27.16400146484375, 75.0140380859375, 174.89157104492188, -45.39482879638672, 41.33385467529297, 14.7308349609375, 12.85345458984375, -38.69556427001953, 89.36817932128906, 113.5328369140625, -93.76132202148438, 11.69927978515625, -24.6270751953125, 3.120647430419922, -2.2134628295898438, 96.89971923828125, -28.440826416015625, 16.762496948242188, -4.66162109375, 4.84271240234375, -42.069244384765625, 99.72199249267578, 72.75918579101562, 108.3193359375, -90.37366485595703, 70.99240112304688, 106.073974609375, 64.79916381835938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000308.npy"} {"epoch": 0.6450261780104712, "step": 309, "batch_size": 128, "mean": 55.450714111328125, "std": 79.67029571533203, "min": -128.80511474609375, "p10": -40.304216384887695, "median": 45.873077392578125, "p90": 158.21244506835936, "max": 281.7535705566406, "pos_frac": 0.734375, "sample": [31.238388061523438, 126.62826538085938, 135.11398315429688, 115.93527221679688, -1.5944900512695312, -16.620330810546875, -7.22430419921875, -95.63060760498047, 134.9443359375, 45.927886962890625, 157.68612670898438, 17.27147674560547, -12.49542236328125, 6.0912628173828125, -40.24967956542969, 95.18329620361328, 14.797119140625, 18.62023162841797, -8.558609008789062, 40.395599365234375, -12.171157836914062, 110.84647369384766, -1.87261962890625, -100.61845397949219, 9.5692138671875, 97.13973999023438, 1.48516845703125, 281.7535705566406, 109.33847045898438, 46.790283203125, -8.164901733398438, 163.532470703125, 105.25949096679688, 1.9600296020507812, -5.066192626953125, -11.451400756835938, 84.92857360839844, -56.75225830078125, 197.32351684570312, -75.594482421875, 181.03231811523438, -79.61697387695312, -3.9427108764648438, 30.421356201171875, 126.5455322265625, -56.95490264892578, 0.36041259765625, -37.297576904296875, -42.660308837890625, 41.26731872558594, 132.24427795410156, -14.848121643066406, 25.581069946289062, -100.1177978515625, 149.94680786132812, 122.67691040039062, 126.42105102539062, 16.44488525390625, 95.2197265625, 15.705810546875, 159.44052124023438, 4.619476318359375, 51.816795349121094, 222.0194091796875, -12.135650634765625, 167.54302978515625, 10.973556518554688, 105.14733123779297, 25.066734313964844, 59.676177978515625, 23.29156494140625, 26.882972717285156, -2.0456314086914062, 177.0833740234375, 2.9420623779296875, 146.53659057617188, 135.7808837890625, 89.13177490234375, 11.46295166015625, 162.515625, 126.11801147460938, 131.9148712158203, 6.911659240722656, 42.51957702636719, 120.50402069091797, 143.8289794921875, -3.8306961059570312, 65.94720458984375, 39.2843017578125, -40.43146896362305, 129.6177978515625, -128.80511474609375, 56.91181945800781, -17.18230628967285, 161.75172424316406, 91.18086242675781, 17.792449951171875, 131.47171020507812, 70.17965698242188, 9.867145538330078, -100.12069702148438, 85.34019470214844, 130.54119873046875, 157.52667236328125, 139.79421997070312, -2.353302001953125, 186.62503051757812, 31.84912109375, 52.30126953125, -121.04229736328125, 152.10476684570312, 60.23626708984375, 26.921142578125, 133.49777221679688, 77.93800354003906, 122.22991943359375, 50.5257568359375, 163.9814453125, -30.27398681640625, 140.13558959960938, 162.83380126953125, 57.54248809814453, 109.09785461425781, -68.36761474609375, 85.290283203125, 45.818267822265625, -3.2880687713623047, 103.5799560546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000309.npy"} {"epoch": 0.6471204188481675, "step": 310, "batch_size": 128, "mean": 45.40919494628906, "std": 68.7824478149414, "min": -118.8370361328125, "p10": -37.70216064453125, "median": 30.457672119140625, "p90": 135.0939178466797, "max": 219.29296875, "pos_frac": 0.7265625, "sample": [219.29296875, 16.358673095703125, -1.5149307250976562, 180.6759033203125, 32.94195556640625, 30.85931396484375, 22.92474365234375, 2.078214645385742, 124.60238647460938, 67.54632568359375, 112.02497863769531, 163.619384765625, 88.11042785644531, 120.6553955078125, 25.28826904296875, 35.811614990234375, 9.27154541015625, -5.607940673828125, 37.580810546875, -65.31822204589844, -7.48443603515625, 37.299041748046875, 19.357421875, 54.405975341796875, 66.1034927368164, -86.90536499023438, 89.22740173339844, -41.66461181640625, 121.08024597167969, -0.6051025390625, -6.859445571899414, -5.628120422363281, -42.52056884765625, 130.98092651367188, -1.98638916015625, -3.5419464111328125, 56.268402099609375, 161.9349365234375, -60.72039794921875, -97.36968994140625, 111.8892822265625, -2.1521148681640625, 16.860519409179688, 20.853179931640625, 96.22103881835938, -9.20184326171875, 95.67221069335938, 31.840003967285156, 54.0968017578125, 4.9942779541015625, -10.90045166015625, 51.077728271484375, 148.01287841796875, 111.59808349609375, 85.31576538085938, 135.92779541015625, 174.77001953125, 14.230484008789062, -3.9016952514648438, -118.8370361328125, -0.5330696105957031, 157.77682495117188, 72.9947280883789, 21.35198974609375, -3.579132080078125, -62.86064910888672, 7.7352142333984375, 17.791473388671875, 110.4490966796875, 3.920583724975586, -18.45899200439453, 15.573104858398438, 52.13157653808594, 27.988525390625, 72.25666809082031, 98.83612060546875, 124.94659423828125, 162.05877685546875, 53.5162353515625, -10.090225219726562, -23.154529571533203, 99.08711242675781, 111.50436401367188, 5.2656707763671875, -33.05076599121094, 115.42406463623047, 127.76134490966797, 3.1757354736328125, 114.57466888427734, -7.67523193359375, 37.8072509765625, -49.3798828125, 124.93295288085938, -79.80712890625, 46.7318000793457, 2.8728256225585938, 12.701446533203125, 132.08163452148438, 71.54739379882812, 30.0560302734375, 50.108482360839844, 150.67164611816406, 2.00677490234375, -32.58539581298828, 131.54318237304688, -9.381927490234375, 7.8905487060546875, -41.38690185546875, -71.86807250976562, 92.92132568359375, -50.73350524902344, 134.73654174804688, 18.747604370117188, 62.298797607421875, 7.69049072265625, -36.12298583984375, 18.21826171875, 28.651805877685547, 7.4972991943359375, 101.22743225097656, 42.048370361328125, 6.71131706237793, 132.5553741455078, 145.12466430664062, 196.20526123046875, 113.86650085449219, 141.01556396484375, 109.51554870605469], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000310.npy"} {"epoch": 0.6492146596858639, "step": 311, "batch_size": 128, "mean": 49.66218948364258, "std": 82.5586929321289, "min": -182.84237670898438, "p10": -49.2296806335449, "median": 36.012752532958984, "p90": 146.87303771972657, "max": 207.28622436523438, "pos_frac": 0.734375, "sample": [6.302558898925781, -4.2657318115234375, 122.80780029296875, 118.31998443603516, 143.91802978515625, 28.52317237854004, 27.804275512695312, -6.9099578857421875, 141.4229736328125, 117.618408203125, 136.04466247558594, 7.627933502197266, -16.47878646850586, -24.877517700195312, 36.31074523925781, -1.092529296875, 140.24420166015625, 57.84148406982422, -69.70780944824219, 34.182159423828125, 172.74502563476562, 135.76361083984375, 4.382659912109375, 107.94744873046875, -7.081184387207031, -13.587310791015625, 153.00735473632812, 116.3558349609375, 6.434741973876953, -38.762237548828125, -84.30362701416016, -177.52752685546875, -182.84237670898438, 114.22598266601562, 11.560638427734375, 155.67620849609375, -124.33270263671875, -71.03558349609375, 131.12942504882812, 52.390777587890625, -63.402976989746094, 95.89396667480469, 86.54611206054688, 4.60870361328125, -9.641244888305664, 51.474517822265625, -16.483604431152344, 123.94309997558594, -114.40910339355469, 19.68317413330078, 101.39351654052734, 31.2191162109375, 1.5782470703125, 29.471588134765625, 110.93428039550781, -14.930374145507812, 90.14872741699219, 17.5921630859375, -19.3829345703125, 157.74868774414062, 157.91546630859375, 134.73257446289062, 142.44354248046875, 117.01121520996094, 121.37200927734375, -81.744873046875, 34.156402587890625, 12.984329223632812, 40.62117004394531, 35.714759826660156, -101.58355712890625, -12.791309356689453, 169.95513916015625, 32.5523681640625, 20.008926391601562, 78.89119720458984, 145.2587890625, 114.6767578125, 20.72372817993164, 125.14181518554688, 207.28622436523438, 4.26190185546875, -102.86434936523438, 57.51982116699219, 0.73663330078125, -28.623809814453125, 162.45980834960938, 147.78009033203125, 135.36813354492188, 35.27679443359375, 115.20671081542969, 98.27647399902344, 106.07063293457031, 57.46307373046875, 171.2452392578125, -117.82493591308594, 168.89422607421875, -6.986602783203125, -124.17404174804688, 116.19917297363281, 8.380735397338867, 146.747802734375, -23.994110107421875, 57.66033935546875, 11.555328369140625, -6.185016632080078, 147.16525268554688, 141.47506713867188, 9.561742782592773, 47.2357177734375, 116.34832763671875, 106.33261108398438, 118.09173583984375, 29.320037841796875, -36.21995544433594, 7.636589050292969, 104.66925048828125, 139.20303344726562, 135.3346710205078, 2.3888778686523438, -12.2357177734375, -4.319099426269531, 5.7114105224609375, 190.8170166015625, 95.22380065917969, 126.65328216552734, 81.976318359375, -43.15541076660156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000311.npy"} {"epoch": 0.6513089005235602, "step": 312, "batch_size": 128, "mean": 49.094608306884766, "std": 77.32024383544922, "min": -165.51336669921875, "p10": -43.19018096923828, "median": 44.00213623046875, "p90": 144.71826171875, "max": 196.54928588867188, "pos_frac": 0.7265625, "sample": [78.06878662109375, -4.52435302734375, -16.30316162109375, 132.45468139648438, 134.19302368164062, -165.51336669921875, 136.017578125, 121.1944580078125, -35.02149200439453, -29.285064697265625, -59.249725341796875, -36.35895538330078, 88.1199951171875, 108.169677734375, 50.878448486328125, 155.81857299804688, 102.38163757324219, 130.65660095214844, -19.485321044921875, -103.5947265625, 6.8175048828125, 121.28913879394531, 124.63037109375, -1.436126708984375, 6.952156066894531, 59.48339080810547, 54.45635986328125, 196.54928588867188, -42.05914306640625, -13.53628158569336, 149.54547119140625, -4.969978332519531, 4.78253173828125, 156.21417236328125, 60.34613037109375, 35.83506774902344, 84.74555969238281, 0.7247047424316406, 97.008056640625, 111.80322265625, 169.80703735351562, 148.55738830566406, 48.786460876464844, 120.3096923828125, 113.66415405273438, 110.24227905273438, 59.560333251953125, 109.19961547851562, 125.32133483886719, 33.54975891113281, 133.80471801757812, 135.58450317382812, 18.112045288085938, 3.73162841796875, -35.28851318359375, 43.367767333984375, 44.636505126953125, 36.186920166015625, 0.4678497314453125, 185.62762451171875, 9.313796997070312, -27.6412353515625, 144.39599609375, -112.6383056640625, 49.210296630859375, -35.3642578125, 163.18710327148438, 107.01959228515625, -54.073638916015625, 123.02577209472656, 26.379608154296875, 125.6090087890625, 27.244125366210938, -5.079475402832031, 75.49166870117188, 95.24810791015625, 4.824134826660156, 145.8970947265625, -0.017822265625, 111.86944580078125, 119.24368286132812, 145.47021484375, -80.48171997070312, 123.14028930664062, -3.218442916870117, 0.0, 59.70709228515625, -62.08845520019531, -115.134033203125, -124.49156188964844, 15.591552734375, 73.79983520507812, 5.090549468994141, 37.92607116699219, 100.19781494140625, 0.9122467041015625, -112.90901184082031, 17.16788101196289, 116.3472900390625, 1.82257080078125, 40.50286865234375, 26.796920776367188, 129.18710327148438, 105.02932739257812, -22.00006103515625, 103.49734497070312, 128.6705322265625, -0.57269287109375, 172.03524780273438, 35.95758056640625, 135.44277954101562, 21.582550048828125, 38.63238525390625, -5.058738708496094, -71.63992309570312, 98.31423950195312, 66.28652954101562, 10.68524169921875, -20.907333374023438, 167.1544189453125, -103.97714233398438, 152.511962890625, -30.088775634765625, 118.27484130859375, 111.76309967041016, 22.257049560546875, 24.578941345214844, -45.82926940917969], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000312.npy"} {"epoch": 0.6534031413612565, "step": 313, "batch_size": 128, "mean": 45.740333557128906, "std": 80.57459259033203, "min": -150.30938720703125, "p10": -57.82401733398437, "median": 32.402801513671875, "p90": 150.56586761474608, "max": 221.9212646484375, "pos_frac": 0.71875, "sample": [150.67288208007812, 141.70928955078125, -28.6744384765625, -23.0462646484375, -60.9737548828125, -4.027801513671875, 60.78045654296875, 131.94601440429688, 156.732666015625, -15.84283447265625, -10.541183471679688, 135.21304321289062, -91.093505859375, 141.72349548339844, 98.24083709716797, 152.0036163330078, 131.96966552734375, 41.160369873046875, -19.009796142578125, -27.6053466796875, 23.376724243164062, 17.36297607421875, 4.353496551513672, 139.1012420654297, 116.81851196289062, -9.045257568359375, -2.8818359375, 28.027664184570312, 106.22576904296875, 176.88516235351562, 168.01300048828125, 202.909423828125, 35.25944519042969, 73.56097412109375, -47.7025146484375, 57.59912109375, 100.88333892822266, 81.88742065429688, 174.23077392578125, 37.23603820800781, -59.4705810546875, 178.77020263671875, 118.17164611816406, 20.27581787109375, -112.12272644042969, 46.98101806640625, 16.099578857421875, -57.11834716796875, 113.76913452148438, 119.79534912109375, 51.03935241699219, 94.62382507324219, -109.57504272460938, 15.756778717041016, 20.069915771484375, -8.071044921875, 151.92364501953125, 59.72706604003906, 26.075820922851562, 3.8029937744140625, 120.51304626464844, 2.47467041015625, 122.22198486328125, 14.694221496582031, 10.725486755371094, 71.7509765625, -25.986968994140625, 112.18557739257812, -101.75553894042969, -45.193359375, 161.25135803222656, -14.818939208984375, 23.87494659423828, 150.31753540039062, 19.2908935546875, -7.53289794921875, 29.760528564453125, 137.10435485839844, -108.32574462890625, 104.14128112792969, 11.41827392578125, -105.79327392578125, 96.81961059570312, -45.05986022949219, 38.8660888671875, 126.55973815917969, 85.88198852539062, -8.133056640625, 11.372442245483398, 5.915061950683594, 28.661376953125, 221.9212646484375, 114.58512878417969, 74.1932373046875, 121.98831176757812, 148.11712646484375, 24.223220825195312, 160.432373046875, -40.725860595703125, -141.92300415039062, 56.81597900390625, 43.431251525878906, 7.660390853881836, 147.777587890625, 91.32832336425781, -78.57286071777344, -1.1702194213867188, -2.4826202392578125, 131.6373748779297, 103.09954833984375, 24.81585693359375, 42.91438293457031, 35.045074462890625, 3.78619384765625, -2.0100975036621094, -11.329902648925781, 119.29660034179688, 28.757553100585938, -104.04086303710938, -105.84419250488281, 3.4626693725585938, -150.30938720703125, 95.77775573730469, 69.63138580322266, 25.452362060546875, 150.52000427246094, 28.50653076171875, 158.8290557861328], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000313.npy"} {"epoch": 0.6554973821989529, "step": 314, "batch_size": 128, "mean": 46.924015045166016, "std": 79.40814971923828, "min": -168.65872192382812, "p10": -58.76440811157226, "median": 41.52961730957031, "p90": 142.30984191894532, "max": 204.38165283203125, "pos_frac": 0.71875, "sample": [36.527610778808594, 40.815643310546875, 116.46730041503906, 42.24359130859375, -11.803421020507812, 21.738983154296875, -85.58741760253906, 89.65036010742188, -85.69335174560547, -109.26325988769531, -21.065139770507812, 66.68269348144531, 140.30072021484375, 134.05577087402344, 156.9517822265625, -63.80657958984375, -106.04568481445312, 173.22268676757812, 124.03572845458984, 148.96661376953125, 79.11029052734375, 3.7420654296875, 83.39471435546875, 120.53123474121094, 1.73974609375, 14.174247741699219, 12.100090026855469, 161.2318115234375, 7.435882568359375, 171.82110595703125, 5.682659149169922, -18.957351684570312, 7.75872802734375, -94.44650268554688, 134.19671630859375, 68.53372192382812, 16.95611572265625, 142.85784912109375, 40.508544921875, 143.75076293945312, -8.416183471679688, 110.961669921875, -64.65432739257812, 113.3935546875, 4.653472900390625, 49.71772766113281, 176.54339599609375, 122.89741516113281, 9.88836669921875, 5.99951171875, 133.83071899414062, -0.38776397705078125, 28.697708129882812, 100.76898193359375, -1.1142005920410156, 103.81765747070312, 45.9403076171875, 19.7763671875, -5.38623046875, -91.45858764648438, 122.02635192871094, 139.6138916015625, 116.69392395019531, 0.0, -0.07177734375, -64.87391662597656, -159.13946533203125, 123.81637573242188, 25.26812744140625, -18.418182373046875, -43.495208740234375, 19.373992919921875, 70.73834228515625, 124.615966796875, 150.92349243164062, -56.603477478027344, -14.080650329589844, 159.22314453125, -30.298980712890625, -168.65872192382812, 120.9673080444336, 43.097259521484375, 132.22671508789062, 89.68777465820312, 0.0, 141.99050903320312, -6.2447509765625, 74.49827575683594, 131.99835205078125, 15.5284423828125, 29.050933837890625, 131.01226806640625, 130.7571258544922, 34.304229736328125, 81.07034301757812, 142.07498168945312, 95.00984954833984, 19.67230224609375, 45.05438232421875, 8.15191650390625, 11.582199096679688, -35.46881103515625, -4.648681640625, -30.31524658203125, 69.56948852539062, 8.82672119140625, 94.84284210205078, 119.64064025878906, 49.243408203125, 119.2682876586914, 164.62527465820312, 70.33726501464844, -77.56924438476562, -1.7667694091796875, 4.050262451171875, 204.38165283203125, 132.83282470703125, -15.305488586425781, 19.198856353759766, 88.86260986328125, 129.87669372558594, 60.974586486816406, -51.256317138671875, 75.22177124023438, -158.12924194335938, 166.10018920898438, 73.5916748046875, -4.841405868530273], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000314.npy"} {"epoch": 0.6575916230366492, "step": 315, "batch_size": 128, "mean": 52.806121826171875, "std": 77.61270141601562, "min": -156.26780700683594, "p10": -29.84362869262695, "median": 27.923603057861328, "p90": 153.43193359375, "max": 236.50445556640625, "pos_frac": 0.765625, "sample": [0.333648681640625, 24.384033203125, 2.1510162353515625, 7.5045928955078125, 46.82679748535156, -39.68194580078125, -27.8675537109375, 12.819900512695312, 114.54559326171875, -10.869049072265625, 27.940353393554688, -11.83380126953125, 120.34410858154297, 20.762176513671875, 118.8970947265625, -36.182830810546875, 85.02412414550781, 150.16555786132812, 54.247589111328125, -24.3531494140625, -57.83795166015625, -29.044151306152344, 0.845367431640625, -2.2470474243164062, 175.28839111328125, 13.5203857421875, 115.11392211914062, 158.20394897460938, 189.19635009765625, 1.7078399658203125, 21.693988800048828, 46.55859375, 143.74700927734375, 152.63677978515625, 32.61377716064453, 175.056884765625, 152.349365234375, 142.62808227539062, 102.17474365234375, -4.868896484375, 2.000579833984375, -31.709075927734375, 28.34014892578125, 136.47488403320312, -41.10032653808594, 15.841934204101562, 7.542724609375, 6.831329345703125, 14.3818359375, 7.33746337890625, 138.72613525390625, -41.061344146728516, 144.353759765625, 158.23233032226562, 139.5662841796875, 141.6548309326172, 26.9901123046875, -12.727806091308594, 28.24334716796875, 73.26184844970703, 12.21807861328125, -0.332122802734375, -6.822345733642578, 48.88397979736328, 126.55780029296875, 141.66299438476562, 83.21347045898438, 106.50577545166016, 9.947219848632812, 236.50445556640625, -80.73110961914062, 32.33149719238281, 118.07025146484375, 22.8656005859375, 133.77659606933594, -41.435638427734375, -138.1754150390625, 170.116455078125, 6.310371398925781, 161.0107421875, 127.17034912109375, 14.254653930664062, -3.9496612548828125, 134.72991943359375, 184.638671875, -6.117301940917969, 36.4471435546875, 108.82255554199219, 20.593994140625, 174.38250732421875, 19.110824584960938, 16.555450439453125, -126.52008056640625, -26.470916748046875, 146.5933837890625, 164.94863891601562, 76.9139404296875, 27.90685272216797, -156.26780700683594, 149.67800903320312, 41.72166442871094, -8.60626220703125, -1.693328857421875, 102.89513397216797, 9.43170166015625, -107.9666748046875, -67.79644775390625, 52.4095458984375, 23.007991790771484, 18.31048583984375, 147.24618530273438, 17.376998901367188, 130.33758544921875, 194.5158233642578, 15.93658447265625, 101.98912048339844, 155.28729248046875, 11.981498718261719, -9.918212890625, 14.54730224609375, 96.4693832397461, -6.204490661621094, 133.25669860839844, 38.454612731933594, 131.84197998046875, 83.71807861328125, 13.702621459960938, 93.3543930053711], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000315.npy"} {"epoch": 0.6596858638743456, "step": 316, "batch_size": 128, "mean": 55.85517883300781, "std": 78.15308380126953, "min": -127.50042724609375, "p10": -30.890634155273435, "median": 39.24249267578125, "p90": 160.8062957763672, "max": 250.77630615234375, "pos_frac": 0.71875, "sample": [112.15380859375, 71.76019287109375, 51.819091796875, 37.40057373046875, -22.5296630859375, -31.607040405273438, -8.5380859375, 164.502197265625, 81.14797973632812, 167.0679931640625, 23.686782836914062, -37.317108154296875, 143.1562042236328, 38.48252868652344, -43.85844421386719, 15.972028732299805, 120.00323486328125, -30.583602905273438, 72.71075439453125, 9.32940673828125, 141.46185302734375, 151.42294311523438, -25.34228515625, 122.18344116210938, 132.25582885742188, 232.9879150390625, 42.436279296875, 132.13888549804688, 42.475006103515625, 180.49107360839844, -5.0554962158203125, 137.9373779296875, -24.28167724609375, 123.86347961425781, 160.04605102539062, 12.142471313476562, 105.5156021118164, 146.80697631835938, 155.82814025878906, 22.223602294921875, -3.598907470703125, 84.68441772460938, 6.62677001953125, 63.456634521484375, 24.03338623046875, 47.67906951904297, -35.916900634765625, 152.39617919921875, 18.36181640625, 0.547515869140625, 80.97366333007812, 3.711639404296875, -127.50042724609375, -57.00494384765625, -27.501434326171875, 186.07452392578125, 111.54281616210938, 88.48249816894531, -4.14501953125, -1.7833175659179688, -10.960817337036133, 69.22109985351562, 174.45260620117188, 137.81536865234375, 24.852142333984375, 22.66131591796875, -19.52276611328125, 250.77630615234375, 38.394493103027344, 115.269287109375, 7.7654266357421875, -2.9798736572265625, -31.7288818359375, -53.00677490234375, -0.2396087646484375, 140.84896850585938, 82.62506103515625, 140.6326446533203, 5.243995666503906, 137.32150268554688, 122.43041229248047, 119.2657470703125, 162.5802001953125, -4.458930969238281, -6.785848617553711, 68.70947265625, 37.457122802734375, 182.7525634765625, -78.59961700439453, 88.11669921875, -92.803955078125, 87.2293701171875, 0.0, -98.412109375, 20.9173583984375, 166.92227172851562, 167.52853393554688, 26.173095703125, 0.3778076171875, 179.486328125, 97.7822265625, 15.744171142578125, -111.87669372558594, 29.334854125976562, -10.3837890625, 61.0050048828125, 148.37173461914062, -5.503814697265625, 86.4928970336914, 100.02633666992188, 172.7161865234375, 20.68560791015625, 152.08074951171875, -24.95831298828125, 19.779815673828125, 26.5767822265625, -21.712936401367188, -125.69044494628906, 39.95343017578125, -2.837005615234375, 39.146942138671875, 114.93411254882812, 114.96287536621094, 39.338043212890625, 111.30854034423828, 31.927169799804688, 129.99176025390625, -11.477523803710938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000316.npy"} {"epoch": 0.6617801047120419, "step": 317, "batch_size": 128, "mean": 51.44091033935547, "std": 74.76044464111328, "min": -145.53134155273438, "p10": -26.81642837524414, "median": 41.217735290527344, "p90": 146.8070327758789, "max": 236.40545654296875, "pos_frac": 0.7421875, "sample": [128.95716857910156, -123.8223648071289, 0.0, 7.774360656738281, 187.02471923828125, -27.36609649658203, 157.2635040283203, 81.27680969238281, 9.96185302734375, 132.11590576171875, 16.7325439453125, 148.91656494140625, 60.066131591796875, -13.638900756835938, 136.67413330078125, -93.0738525390625, -25.635055541992188, 141.64535522460938, 119.1497802734375, 26.63567352294922, 49.16754150390625, 6.8333740234375, 118.47262573242188, 114.67105102539062, 14.294776916503906, 130.2332763671875, 6.716407775878906, 0.8102016448974609, -10.537200927734375, 23.548187255859375, -17.643447875976562, -39.008941650390625, 35.133148193359375, -6.522003173828125, -26.580856323242188, 102.10176086425781, 94.91637420654297, 128.58941650390625, 117.26004028320312, -33.055389404296875, -36.787994384765625, 32.64735412597656, 158.6102294921875, -117.06170654296875, -123.2760009765625, 145.5400390625, 169.08880615234375, 145.35888671875, 17.622825622558594, 0.0, 113.2799301147461, 89.80484008789062, 20.91583251953125, 179.89688110351562, -125.88198852539062, -14.4306640625, 46.16444396972656, 50.92448425292969, 2.4955291748046875, 27.813720703125, 17.5108642578125, 148.16021728515625, 138.14112854003906, 105.90985107421875, 0.0, 12.914794921875, 16.4378662109375, 1.0074462890625, 40.680301666259766, 57.366943359375, 131.35302734375, -5.295829772949219, -22.782821655273438, -145.53134155273438, 160.00811767578125, 66.87474060058594, 152.114013671875, -64.2376708984375, 90.51087951660156, -15.1885986328125, 64.97093200683594, 119.72022247314453, 110.74324035644531, -13.181610107421875, 60.98394775390625, 91.072998046875, 48.2286376953125, -28.8924560546875, 110.23567199707031, 146.54940795898438, -2.42041015625, 15.343963623046875, 155.5703887939453, 11.400108337402344, 9.511672973632812, -0.6733226776123047, 3.0244293212890625, 130.61166381835938, 82.64625549316406, 236.40545654296875, 3.4243087768554688, 111.42079162597656, 141.14990234375, -16.37353515625, -33.563385009765625, 32.11749267578125, 20.54339599609375, 35.909912109375, 0.2784557342529297, 69.15057373046875, 106.55662536621094, 4.739725112915039, 133.43841552734375, 7.212368011474609, 90.83984375, 132.65975952148438, 109.21949005126953, 43.39091110229492, 157.64724731445312, 41.75516891479492, -7.073341369628906, 41.80108642578125, -4.9721527099609375, 147.4081573486328, 69.79829406738281, 121.70481872558594, -3.6788406372070312, 127.34222412109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000317.npy"} {"epoch": 0.6638743455497382, "step": 318, "batch_size": 128, "mean": 50.49152374267578, "std": 72.35525512695312, "min": -156.06011962890625, "p10": -30.173898315429675, "median": 47.17645454406738, "p90": 144.541015625, "max": 187.05419921875, "pos_frac": 0.7421875, "sample": [86.51516723632812, 23.696243286132812, 138.44912719726562, 130.144287109375, 21.674224853515625, -7.5201416015625, 44.31976318359375, 36.436859130859375, 143.8924560546875, 42.36845397949219, 5.1900634765625, -19.93224334716797, 146.0543212890625, 56.932220458984375, 167.8843994140625, 20.505401611328125, 126.18560791015625, -13.34820556640625, 104.5670394897461, 84.41419982910156, -23.731842041015625, -23.1572265625, 128.52798461914062, 149.73312377929688, -18.303070068359375, 43.76776123046875, 99.80441284179688, 134.3837432861328, -114.59381866455078, 29.515625, 1.975341796875, -45.486778259277344, 73.57247924804688, 66.63299560546875, 59.10423278808594, -20.856597900390625, 9.200126647949219, 56.167327880859375, 135.55355834960938, 174.3518829345703, 67.75146484375, 82.89486694335938, 72.96478271484375, 55.89697265625, 10.313858032226562, 6.77215576171875, -136.50135803222656, 112.09249114990234, -3.321136474609375, 100.74261474609375, -156.06011962890625, 12.65572738647461, 82.94914245605469, 150.66754150390625, 119.42059326171875, 44.489044189453125, -13.676826477050781, 117.18070983886719, 122.71890258789062, -16.44415283203125, 97.23233032226562, 99.78778076171875, 166.586669921875, 54.6109619140625, -46.96859359741211, 47.25750732421875, -43.470550537109375, 4.218902587890625, 151.39462280273438, -2.949188232421875, 125.27548217773438, 27.141815185546875, 7.061492919921875, 88.4684066772461, 118.27157592773438, 122.95925903320312, 73.88116455078125, 127.13673400878906, 31.19970703125, -10.414398193359375, 74.01322174072266, 63.25762939453125, 29.953125, 133.3800048828125, 71.77362060546875, -44.69671630859375, 46.75457763671875, 161.15093994140625, -40.533470153808594, -8.829025268554688, -7.92718505859375, 45.58660888671875, 33.19122314453125, 96.44731140136719, 60.72119140625, 49.578155517578125, 75.67828369140625, -51.14118957519531, 97.59835815429688, -21.44989013671875, 116.3990249633789, 187.05419921875, 161.43853759765625, 27.199935913085938, -38.9766845703125, 122.78155517578125, 43.5723876953125, 1.5406341552734375, -16.486221313476562, -67.02902221679688, 150.6650390625, 32.01513671875, 63.022125244140625, 149.4208221435547, 32.090667724609375, 133.5503387451172, 47.095401763916016, 141.46734619140625, -26.401275634765625, 114.69488525390625, -123.50588989257812, -3.9336395263671875, -3.1563644409179688, 30.105361938476562, 152.22576904296875, 0.7869873046875, -156.00881958007812, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000318.npy"} {"epoch": 0.6659685863874345, "step": 319, "batch_size": 128, "mean": 55.999759674072266, "std": 65.786376953125, "min": -153.92958068847656, "p10": -20.948791503906246, "median": 48.052751541137695, "p90": 140.3038116455078, "max": 195.76336669921875, "pos_frac": 0.828125, "sample": [97.50269317626953, 2.3610172271728516, 58.24261474609375, 53.515289306640625, 30.08348846435547, 195.76336669921875, 95.9346923828125, -42.98211669921875, 86.06224060058594, 25.339111328125, -11.870445251464844, 11.775968551635742, 103.84796142578125, 155.6876220703125, 85.71044921875, 151.1383056640625, 7.19866943359375, 0.0, 47.854400634765625, 99.7855224609375, 11.055984497070312, 19.85723876953125, 90.76748657226562, 9.769561767578125, 6.42266845703125, 124.25033569335938, 9.405181884765625, 54.9635009765625, -73.36822509765625, 105.81846618652344, -1.618408203125, 11.606674194335938, 89.3433837890625, 5.389533996582031, 120.40654754638672, 67.7216796875, 7.27032470703125, 116.12057495117188, 137.01171875, 38.443878173828125, 4.687349319458008, 135.2366943359375, 44.71551513671875, -103.38580322265625, 3.5088577270507812, 100.39166259765625, 155.34912109375, 9.558990478515625, -51.115447998046875, 139.81866455078125, 108.72781372070312, 100.15672302246094, 12.267227172851562, 21.462310791015625, 123.41156005859375, 24.94207763671875, 25.283416748046875, -53.55156707763672, 58.43053436279297, 134.7321014404297, -153.92958068847656, 75.49182891845703, 107.40381622314453, -58.705108642578125, 57.29931640625, -19.633407592773438, 49.342681884765625, 27.38188362121582, 64.71853637695312, 110.12823486328125, 4.402107238769531, 93.61588287353516, 111.80197143554688, 141.43582153320312, -4.2737884521484375, -11.9835205078125, 116.61181640625, 114.82217407226562, 142.2469482421875, -24.018020629882812, 16.043067932128906, 127.94705200195312, 125.98426055908203, 37.886627197265625, 30.91082763671875, 1.6504669189453125, 117.2122802734375, 193.0433349609375, 87.80963134765625, 19.524810791015625, -11.386016845703125, 94.40866088867188, 76.42864990234375, -25.62591552734375, 106.495849609375, 39.40972900390625, 28.352569580078125, 146.16677856445312, -42.2161865234375, 157.8076171875, 34.383056640625, 27.981857299804688, 160.4971923828125, 32.57891845703125, -53.646209716796875, 11.45843505859375, 107.63174438476562, 73.97940063476562, 137.87872314453125, -6.759265899658203, 39.40972900390625, 23.510013580322266, -7.517385482788086, 134.29782104492188, -25.504119873046875, 117.26658630371094, 118.07243347167969, 171.85223388671875, -31.994834899902344, 41.4189453125, 154.39697265625, 15.686309814453125, 48.251102447509766, 1.0559844970703125, 139.29446411132812, 51.93115234375, 34.42509460449219, 145.9024658203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000319.npy"} {"epoch": 0.6680628272251309, "step": 320, "batch_size": 128, "mean": 51.38296127319336, "std": 82.73348999023438, "min": -152.81585693359375, "p10": -42.392517089843736, "median": 45.134517669677734, "p90": 151.1662796020508, "max": 195.71588134765625, "pos_frac": 0.734375, "sample": [111.47906494140625, 71.56266021728516, -71.36874389648438, -141.19107055664062, 118.06015014648438, 148.8043212890625, -75.19635009765625, 6.93896484375, 143.55413818359375, 147.43484497070312, 18.127899169921875, 117.505859375, 35.915283203125, 11.807159423828125, 153.216552734375, 84.73489379882812, -19.8521728515625, -0.6214141845703125, 13.30279541015625, 22.402015686035156, 15.823837280273438, 137.80686950683594, 43.86181640625, 18.807540893554688, 1.23687744140625, -75.44839477539062, 168.24942016601562, 114.37513732910156, 59.00929260253906, 28.76018524169922, 96.18274688720703, 107.387451171875, 150.9207763671875, 66.71440124511719, 52.5196533203125, 136.27548217773438, -14.716644287109375, 151.01434326171875, 12.128082275390625, -1.1113739013671875, 5.849445343017578, 11.507431030273438, 112.6888427734375, -27.561126708984375, 158.16310119628906, 174.23587036132812, 91.4610595703125, 6.938934326171875, 99.11083984375, 99.34983825683594, -35.411712646484375, 1.82135009765625, -108.98855590820312, -8.2681884765625, 158.8826904296875, -15.56256103515625, 84.28012084960938, 6.50732421875, 92.20281219482422, 149.21505737304688, 125.91558837890625, 148.45672607421875, 13.2423095703125, 141.2552490234375, 151.75869750976562, -34.185279846191406, 143.52682495117188, -17.663818359375, 162.47769165039062, 137.74435424804688, 36.263633728027344, -6.524810791015625, 24.610061645507812, -4.6678466796875, -152.81585693359375, 86.25997924804688, 11.52337646484375, 145.1766357421875, 7.947998046875, 127.65040588378906, -124.995849609375, 38.243316650390625, 2.9550933837890625, -6.13726806640625, -33.21672058105469, 46.40721893310547, 100.28543090820312, 138.5777587890625, 85.50289916992188, -2.8917160034179688, -92.78529357910156, 48.77947998046875, 184.02142333984375, -7.55615234375, 121.88247680664062, -25.683162689208984, 12.460769653320312, 84.6075210571289, -25.7808837890625, 158.12118530273438, 84.17633056640625, -52.4195556640625, 81.6563720703125, 151.5207977294922, 136.81646728515625, 195.71588134765625, 133.72108459472656, -29.404922485351562, 95.43444061279297, -119.926513671875, 61.83738708496094, 102.07638549804688, -134.23556518554688, -5.8937530517578125, 160.70281982421875, 36.302734375, 142.55545043945312, 42.54380798339844, 120.15605163574219, 107.59663391113281, -38.09521484375, 20.37225341796875, 43.521484375, -126.13243865966797, 176.24649047851562, 38.73108673095703, -139.4156951904297, 117.2723388671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000320.npy"} {"epoch": 0.6701570680628273, "step": 321, "batch_size": 128, "mean": 44.387664794921875, "std": 74.98474884033203, "min": -165.52481079101562, "p10": -41.594924926757805, "median": 27.168060302734375, "p90": 147.12580413818358, "max": 182.97705078125, "pos_frac": 0.7265625, "sample": [140.93246459960938, -59.63812255859375, 75.60577392578125, -145.5999755859375, 26.91387939453125, 160.90521240234375, 19.6495361328125, 0.97747802734375, -55.30186462402344, 167.64218139648438, 120.95756530761719, 21.419021606445312, -73.91554260253906, 88.32950592041016, 100.91641235351562, 14.4935302734375, 4.903923034667969, 136.8798065185547, 160.93780517578125, 9.992652893066406, 74.63188934326172, -66.4638671875, 2.827190399169922, -8.250091552734375, 154.760009765625, 105.21188354492188, -1.1620941162109375, 4.406036376953125, 105.3868408203125, -1.147674560546875, 122.47520446777344, 56.969017028808594, 104.8406982421875, 16.1312255859375, 40.028472900390625, -56.53271484375, -38.8780517578125, 63.54582214355469, 7.1743011474609375, 130.2396240234375, 76.54974365234375, 29.494873046875, -6.3370513916015625, 49.982421875, 10.4422607421875, 158.53904724121094, -37.11628723144531, 143.74526977539062, 182.97705078125, 64.47357177734375, 20.37458038330078, 21.026092529296875, -17.508289337158203, 39.53399658203125, -5.233085632324219, 8.150405883789062, 145.55206298828125, -9.95550537109375, 143.17147827148438, -30.321014404296875, 21.158233642578125, 100.36431884765625, 37.261207580566406, 116.63111114501953, 74.17266845703125, 150.83685302734375, -31.7530517578125, 68.74032592773438, -70.47370147705078, -5.23016357421875, -12.016372680664062, 59.404205322265625, -33.29273986816406, 158.23175048828125, -148.51690673828125, 118.09481811523438, 162.60635375976562, 179.71774291992188, 27.4222412109375, 121.74615478515625, 21.538375854492188, -31.3427734375, 101.087890625, 91.65998840332031, 9.28009033203125, 139.76718139648438, -49.65814208984375, 150.79786682128906, -89.11039733886719, 126.14248657226562, 11.720832824707031, 15.996728897094727, 105.53123474121094, -0.7523117065429688, -1.4895744323730469, 166.14044189453125, 2.08154296875, 59.28594970703125, 32.65238952636719, 104.19866943359375, 113.43071746826172, -3.97393798828125, 92.9215087890625, 39.70697021484375, 0.73382568359375, 116.58663940429688, 5.6074981689453125, 2.190736770629883, 1.146291732788086, -1.3415069580078125, -16.783218383789062, -47.934295654296875, 9.978506088256836, -23.641250610351562, 56.57611846923828, 168.89602661132812, -165.52481079101562, 141.74447631835938, 36.607818603515625, 70.98294830322266, 10.851531982421875, 77.07305908203125, 17.301116943359375, 0.7837295532226562, 141.59228515625, 140.711181640625, 0.0, -85.9730224609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000321.npy"} {"epoch": 0.6722513089005235, "step": 322, "batch_size": 128, "mean": 54.306304931640625, "std": 75.97172546386719, "min": -250.581298828125, "p10": -28.67817001342773, "median": 50.23182678222656, "p90": 153.16424560546875, "max": 192.5150146484375, "pos_frac": 0.7265625, "sample": [-55.827545166015625, -14.34088134765625, -68.91082763671875, 91.88101959228516, 75.28025817871094, 12.433944702148438, 51.64555358886719, -38.071044921875, 135.7894287109375, 80.98675537109375, -14.749313354492188, -0.20098876953125, 136.51861572265625, 15.5693359375, 144.82626342773438, 116.72463989257812, 27.5181884765625, -1.56158447265625, 115.96768188476562, 75.48388671875, 58.250823974609375, -16.94622802734375, 164.5704345703125, 2.530120849609375, 1.48309326171875, 43.816375732421875, -27.966720581054688, 106.36361694335938, 5.559173583984375, 114.84233093261719, 48.81809997558594, 132.37701416015625, 0.6060333251953125, -15.253448486328125, 127.48361206054688, 31.386444091796875, 17.427894592285156, 138.94393920898438, -59.47320556640625, 122.86429595947266, 134.13687133789062, 115.82090759277344, -76.26165771484375, 16.52825927734375, -4.060791015625, 26.476760864257812, 69.23274993896484, -32.03901672363281, 124.8360595703125, -128.58335876464844, 121.35142517089844, 3.02178955078125, 137.09228515625, 184.53231811523438, -15.98681640625, 127.05836486816406, 88.33177185058594, -42.156494140625, 29.526138305664062, 119.2049560546875, 159.49691772460938, -3.8931884765625, 183.17941284179688, 192.5150146484375, -15.06512451171875, 122.07459259033203, 88.25706481933594, 29.024688720703125, 104.63916015625, 12.61602783203125, 173.92691040039062, 151.07891845703125, 112.72198486328125, 59.049163818359375, 9.74456787109375, 0.0, -59.53118896484375, 47.63525390625, 74.670654296875, -33.04383850097656, 155.3043212890625, 78.73469543457031, 133.0013427734375, 152.2470703125, -20.819381713867188, 85.82955932617188, 53.92431640625, -5.18011474609375, 3.82281494140625, 87.84803771972656, 166.11734008789062, -3.247039794921875, -15.034088134765625, -40.92486572265625, -13.9718017578125, 1.0879344940185547, 94.40792846679688, 2.7739105224609375, -250.581298828125, 70.030029296875, 136.0336151123047, 56.526153564453125, 52.38153076171875, 166.38607788085938, 128.0179443359375, -30.338218688964844, -1.7096023559570312, 186.70216369628906, 183.42266845703125, 5.845912933349609, -11.940322875976562, 123.94786071777344, 25.959808349609375, 41.20652770996094, -9.325393676757812, 13.335269927978516, -16.325157165527344, 18.1575927734375, 26.71337890625, 106.55413818359375, 81.02877807617188, 173.96145629882812, 90.85360717773438, -7.11883544921875, 18.1427001953125, 143.926025390625, 63.34123229980469, 158.34262084960938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000322.npy"} {"epoch": 0.6743455497382199, "step": 323, "batch_size": 128, "mean": 37.82035827636719, "std": 84.04615783691406, "min": -196.29461669921875, "p10": -64.58278961181638, "median": 23.604372024536133, "p90": 144.93175201416017, "max": 184.2239227294922, "pos_frac": 0.6796875, "sample": [14.215522766113281, -31.689361572265625, -129.5620574951172, -113.55062866210938, 123.18205261230469, 132.14747619628906, -77.1712646484375, 84.7847900390625, 32.13494873046875, 177.45648193359375, -86.42054748535156, 134.18463134765625, 25.23748779296875, -19.82568359375, 1.9296722412109375, 107.8764877319336, 147.80458068847656, -7.2883148193359375, 81.8359375, -24.95513916015625, 178.84368896484375, 105.42327880859375, 21.341796875, 134.85006713867188, 22.470184326171875, 162.91636657714844, 18.164794921875, 131.30410766601562, -87.64376831054688, 139.814453125, -16.61529541015625, -167.12945556640625, -7.113227844238281, 40.08770751953125, -0.06360435485839844, 102.37472534179688, 22.516799926757812, 19.822982788085938, 81.65875244140625, 18.62738609313965, 27.499130249023438, 127.37774658203125, -29.1624755859375, 121.0997314453125, 132.8941650390625, -14.047393798828125, -19.04559326171875, -196.29461669921875, 148.25921630859375, 27.980064392089844, 7.199346542358398, 7.111026763916016, -52.334442138671875, 8.696285247802734, 62.09642028808594, 70.73825073242188, 47.367889404296875, 142.08834838867188, 184.2239227294922, -25.903419494628906, 77.31488037109375, 85.88714599609375, 123.84610748291016, -24.7626953125, 154.07485961914062, 22.209747314453125, 124.45791625976562, -18.52416229248047, -47.060791015625, 128.0635986328125, 32.563720703125, 2.6309947967529297, -59.18772888183594, 3.318878173828125, 124.08584594726562, 76.46371459960938, 171.17120361328125, -19.1630859375, 166.98838806152344, 4.088623046875, 122.70104217529297, -120.76176452636719, 30.06527328491211, 30.674591064453125, 17.29876708984375, 173.35946655273438, -51.06193542480469, 143.25048828125, -132.04000854492188, 23.79758071899414, 134.526611328125, 25.36811065673828, -47.32847595214844, 113.19042205810547, 0.4632568359375, -8.22348403930664, -88.9956283569336, -19.53813934326172, -7.961944580078125, -6.73321533203125, 21.108489990234375, -96.85342407226562, -10.526123046875, 3.1468963623046875, 28.774063110351562, 102.792236328125, 20.325170516967773, -17.00830841064453, 128.56121826171875, 171.024169921875, -2.581533432006836, 32.32818603515625, 135.3998565673828, 144.9223175048828, -129.87843322753906, -131.78622436523438, 32.52543640136719, 23.411163330078125, 108.572998046875, 25.528465270996094, 14.264165878295898, -6.56494140625, 101.16639709472656, 144.95376586914062, 99.71499633789062, -17.19720458984375, 0.3970069885253906, 182.1446533203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000323.npy"} {"epoch": 0.6764397905759162, "step": 324, "batch_size": 128, "mean": 59.99622344970703, "std": 71.21031188964844, "min": -99.99766540527344, "p10": -22.8324592590332, "median": 41.741661071777344, "p90": 154.06925354003906, "max": 199.650146484375, "pos_frac": 0.765625, "sample": [68.39735412597656, 156.56427001953125, 14.87860107421875, 125.57763671875, 22.766876220703125, 120.7908935546875, 3.039947509765625, 2.6342639923095703, 137.32498168945312, 32.53197479248047, 31.382080078125, -0.6980209350585938, -51.00178527832031, 158.353759765625, 4.875885009765625, 128.87216186523438, 122.27505493164062, 9.499893188476562, -24.263397216796875, 0.0, 9.97698974609375, 4.62939453125, 130.34469604492188, -5.933418273925781, 4.735157012939453, 31.848663330078125, 128.4185028076172, -48.43281555175781, 126.84396362304688, -33.00456237792969, 6.208112716674805, 137.35760498046875, 143.5144805908203, 154.35333251953125, 131.10150146484375, 123.78511810302734, 18.651031494140625, 49.719390869140625, 78.45993041992188, 0.0, -6.742588043212891, 141.31634521484375, 157.73944091796875, 100.23731994628906, 96.52047729492188, 129.1197509765625, 0.0, -2.3770751953125, 107.6678466796875, -11.725799560546875, -64.77691650390625, 40.53636169433594, 42.94696044921875, 19.93902587890625, 111.69580078125, 93.37179565429688, 9.119415283203125, 122.41436767578125, 158.97055053710938, 7.04168701171875, -50.81147003173828, 160.79678344726562, 11.363876342773438, 129.31396484375, 102.91935729980469, -1.8718395233154297, 116.66825866699219, 126.1456298828125, 18.7698974609375, -99.99766540527344, 154.097900390625, 45.099830627441406, -91.91352844238281, 126.52867126464844, -4.6016998291015625, -4.7200927734375, 29.886253356933594, 47.533416748046875, 12.161941528320312, 199.650146484375, 180.72589111328125, 7.453388214111328, -29.09307861328125, 56.183197021484375, 181.23651123046875, 143.85513305664062, 15.92095947265625, 22.563568115234375, -13.717193603515625, 5.93646240234375, 134.292236328125, 35.98675537109375, 61.044044494628906, -45.5540771484375, -22.219200134277344, 54.30401611328125, 36.77012252807617, 154.05697631835938, 128.92849731445312, 37.726715087890625, -40.56989288330078, -4.652153015136719, 8.19696044921875, 49.33183288574219, -48.506591796875, 180.0516357421875, 176.28762817382812, 143.3568115234375, 88.00119018554688, 5.264404296875, 149.446533203125, 52.54874038696289, -29.25731658935547, 129.9534454345703, 20.4342041015625, 131.42401123046875, 155.5928955078125, 127.83009338378906, -0.6334190368652344, 145.7904510498047, 133.4879150390625, -0.33876800537109375, 132.65078735351562, 139.4588165283203, 139.56617736816406, 10.545501708984375, 5.2236328125, -3.7844390869140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000324.npy"} {"epoch": 0.6785340314136126, "step": 325, "batch_size": 128, "mean": 48.876914978027344, "std": 88.97467803955078, "min": -148.19216918945312, "p10": -43.947073364257804, "median": 25.525318145751953, "p90": 168.79157409667968, "max": 247.03179931640625, "pos_frac": 0.6484375, "sample": [77.31649780273438, 143.61337280273438, -115.968017578125, -29.9296875, 113.7275390625, 22.10113525390625, 73.08837890625, 145.65673828125, 38.27972412109375, -1.80023193359375, 1.14752197265625, 15.9737548828125, 141.02793884277344, 141.19656372070312, -10.505279541015625, 4.916790008544922, 157.65097045898438, 131.5369873046875, 82.306640625, 173.54275512695312, 137.82913208007812, 170.39920043945312, 98.33294677734375, -1.9178295135498047, 131.95913696289062, 4.133003234863281, 14.057510375976562, 78.43975830078125, 130.1719512939453, 9.623992919921875, -98.16047668457031, 9.130989074707031, 53.76314163208008, 158.91065979003906, 157.25250244140625, 97.85743713378906, 21.014062881469727, 162.0192413330078, 37.41876220703125, -6.228515625, 127.31390380859375, -141.88735961914062, 154.7648468017578, -0.21966552734375, -93.91044616699219, -0.474212646484375, 247.03179931640625, -127.34632873535156, 179.6581573486328, -3.389678955078125, -1.1759033203125, 23.561767578125, 157.24517822265625, 11.99169921875, 8.683563232421875, -109.25800323486328, -8.830490112304688, -118.53414154052734, -8.400238037109375, -50.10894775390625, 55.52253723144531, 93.04583740234375, -1.9669265747070312, -6.99822998046875, 76.35908508300781, 56.40458679199219, -2.3821258544921875, -117.92684936523438, 189.10888671875, -10.894287109375, 83.70364379882812, 21.435317993164062, 239.540771484375, -11.89239501953125, -10.9351806640625, -0.41827392578125, -10.380218505859375, -14.045051574707031, 80.55105590820312, 131.765380859375, -121.64327239990234, -10.645538330078125, -41.90362548828125, 170.7327880859375, -126.75634765625, 53.313507080078125, 126.53005981445312, 44.244873046875, 44.68739318847656, -22.57781982421875, -48.715118408203125, 113.87889099121094, 170.20962524414062, -32.09916687011719, 1.7622394561767578, -148.19216918945312, -15.61883544921875, 174.91351318359375, 58.33903503417969, 138.91607666015625, 168.183837890625, 28.446701049804688, 98.00776672363281, 196.8790283203125, 156.2293701171875, -38.098876953125, 5.395284652709961, 27.488868713378906, 23.093338012695312, -1.72772216796875, -21.231887817382812, 147.46807861328125, 44.8939208984375, 135.66690063476562, 7.9660491943359375, 124.77855682373047, 124.73117065429688, 184.50299072265625, 14.091888427734375, 1.5951004028320312, -2.0050201416015625, -29.064788818359375, 200.70257568359375, -23.387176513671875, -15.227691650390625, 198.38473510742188, 64.08786010742188, 147.81785583496094], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000325.npy"} {"epoch": 0.680628272251309, "step": 326, "batch_size": 128, "mean": 50.89619445800781, "std": 81.46173858642578, "min": -196.91265869140625, "p10": -47.925097656249996, "median": 38.24091339111328, "p90": 146.6186279296875, "max": 175.53819274902344, "pos_frac": 0.71875, "sample": [138.43643188476562, 121.7847900390625, 126.21556091308594, 173.56715393066406, 60.28355407714844, 63.452972412109375, -46.50921630859375, -1.47381591796875, 132.56085205078125, -88.87657165527344, 114.1234130859375, -17.6304931640625, 134.71180725097656, 139.2913055419922, 23.32213592529297, 2.6687240600585938, 4.785612106323242, 150.35491943359375, 175.53819274902344, 14.554916381835938, 67.82821655273438, 135.61550903320312, 6.1361083984375, -20.63580322265625, 63.0345458984375, 106.18795776367188, 120.23802185058594, 92.55035400390625, 14.203472137451172, 18.320388793945312, -119.83553314208984, 147.0711669921875, 0.579620361328125, 105.21597290039062, -125.08737182617188, -42.163116455078125, 21.851261138916016, 85.32183074951172, -30.90826416015625, 121.882080078125, 67.57586669921875, 108.61625671386719, -22.342926025390625, 144.59234619140625, 32.206756591796875, 107.98439025878906, -37.12425231933594, 137.85296630859375, -23.00250244140625, 161.90272521972656, 125.5364990234375, 71.7998046875, 141.76168823242188, 16.996620178222656, 146.4246826171875, -123.54955291748047, -35.51678466796875, 11.324020385742188, 126.16864013671875, -11.429641723632812, 124.68418884277344, 111.59415435791016, 19.9063720703125, 97.87832641601562, 0.0, -196.91265869140625, 13.879457473754883, -0.61773681640625, 126.81242370605469, -56.60723876953125, 9.461822509765625, 160.95242309570312, 172.30584716796875, 78.96615600585938, 38.49259948730469, -56.642974853515625, 32.822853088378906, -71.77362060546875, 32.59864807128906, 101.80352020263672, -1.903900146484375, 157.2471923828125, -4.242431640625, 30.516326904296875, 12.244346618652344, 135.83685302734375, -36.252716064453125, -102.34133911132812, 109.19581604003906, 12.218429565429688, 135.2696533203125, 18.793792724609375, 161.06622314453125, -9.80902099609375, 86.7186050415039, 32.08624267578125, 108.97028350830078, 152.1038818359375, 23.66180419921875, 126.56747436523438, 154.18231201171875, 0.0, 130.48150634765625, 37.989227294921875, 59.956787109375, 128.31500244140625, -5.1663818359375, 144.952880859375, 103.13037109375, 0.0, 131.02127075195312, -104.48236083984375, -36.848846435546875, -114.6589126586914, 22.213943481445312, -115.40997314453125, 155.676513671875, -8.39166259765625, 142.45611572265625, 19.081100463867188, 68.83403778076172, 16.5318603515625, -51.22882080078125, -36.01642608642578, 130.00881958007812, 151.895263671875, 29.97955322265625, 136.31149291992188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000326.npy"} {"epoch": 0.6827225130890052, "step": 327, "batch_size": 128, "mean": 56.87364959716797, "std": 81.80422973632812, "min": -152.65081787109375, "p10": -27.427358245849597, "median": 39.7120361328125, "p90": 159.91478881835937, "max": 214.27252197265625, "pos_frac": 0.7734375, "sample": [32.21836853027344, 135.57708740234375, 120.56779479980469, 20.696685791015625, 190.7186279296875, -141.84051513671875, 92.29815673828125, 143.17730712890625, 5.69158935546875, 114.00961303710938, -0.47614479064941406, 180.76779174804688, -6.60040283203125, -138.88406372070312, 27.711769104003906, -16.2867431640625, -80.60992431640625, 8.04058837890625, 172.60061645507812, 169.32672119140625, 111.76527404785156, 109.5853042602539, 21.845260620117188, 53.31108093261719, 104.22747802734375, -37.6466064453125, 17.958389282226562, 111.78861999511719, 11.603973388671875, 121.64091491699219, 47.1534423828125, 152.21392822265625, -4.711244583129883, -38.908729553222656, 62.16395568847656, 112.52630615234375, 166.3155517578125, 9.255584716796875, 214.27252197265625, 103.73564147949219, -22.63397216796875, -8.631742477416992, -10.646820068359375, 111.02326202392578, 33.19038391113281, -16.347259521484375, 139.52786254882812, 37.672760009765625, 114.23025512695312, 13.906646728515625, 146.93258666992188, 3.954193115234375, -9.27752685546875, -152.65081787109375, 121.78461456298828, 8.490966796875, 138.35101318359375, 126.20346069335938, 177.50875854492188, 38.791107177734375, 61.041839599609375, 76.60177612304688, 112.45780944824219, 43.92315673828125, 156.24798583984375, 146.56890869140625, 165.0357666015625, 1.412506103515625, 95.00115966796875, -20.502532958984375, 30.25433349609375, 7.253448486328125, -11.3896484375, -42.64971160888672, 116.69044494628906, 9.471099853515625, 33.41926574707031, 58.50762939453125, 168.99932861328125, 9.900299072265625, -22.829620361328125, 5.399261474609375, 26.3629150390625, 118.88873291015625, 20.185089111328125, 161.74453735351562, 15.464569091796875, 0.0, 148.788818359375, 23.838516235351562, 26.138702392578125, 159.9764404296875, 185.16610717773438, 20.74066162109375, 131.2421875, 12.529922485351562, 134.54214477539062, 34.58674621582031, 141.5824737548828, 139.62060546875, 71.1326904296875, 12.165267944335938, -146.314697265625, -23.32611846923828, 107.603271484375, 116.07976531982422, -95.50526428222656, 40.632965087890625, 60.9356689453125, 132.3396453857422, -131.55813598632812, 119.008544921875, 166.16461181640625, 8.786468505859375, -36.996917724609375, 135.00726318359375, -5.452800750732422, 159.88836669921875, 1.6916675567626953, -9.984115600585938, 157.72000122070312, 143.79794311523438, 136.03536987304688, 6.263702392578125, -103.84649658203125, 139.04104614257812, 19.793746948242188, -73.67098999023438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000327.npy"} {"epoch": 0.6848167539267016, "step": 328, "batch_size": 128, "mean": 50.621185302734375, "std": 76.75634765625, "min": -143.55795288085938, "p10": -24.362550354003908, "median": 33.798797607421875, "p90": 146.6781478881836, "max": 205.14984130859375, "pos_frac": 0.765625, "sample": [25.991729736328125, 120.26564025878906, 145.97535705566406, 37.66876220703125, 182.14990234375, -3.2120361328125, 2.2969970703125, 121.92683410644531, 121.39979553222656, 114.02738952636719, 97.37220764160156, 50.16508102416992, 5.2951202392578125, 113.57025146484375, 54.46989440917969, 0.20471763610839844, 8.681640625, -9.11846923828125, 46.95355224609375, 93.39231872558594, 22.341018676757812, 68.57208251953125, 78.32781982421875, -25.556007385253906, 177.72561645507812, 66.63677978515625, -0.8193817138671875, 41.267852783203125, 110.79827880859375, 118.8870620727539, -4.153594970703125, 122.23413848876953, 34.51824951171875, 158.75924682617188, 194.12109375, 98.07242584228516, 169.40467834472656, 6.1168365478515625, 171.06472778320312, -0.10350990295410156, 4.1702880859375, 139.69451904296875, 87.03070068359375, -0.1719970703125, 7.530487060546875, -6.481721878051758, -22.5615234375, 27.258270263671875, -24.307861328125, 136.19418334960938, 36.740203857421875, 9.54876708984375, -24.236846923828125, 164.946533203125, 10.667266845703125, 31.6427001953125, 12.479400634765625, 11.598770141601562, 43.97029113769531, -35.771080017089844, 11.431869506835938, 22.57958221435547, 113.99176025390625, 148.3179931640625, 83.85601806640625, 156.7525634765625, 121.9774398803711, -72.7801513671875, 135.78668212890625, -97.07286834716797, 13.287887573242188, 118.7572021484375, 121.91251373291016, 125.25749206542969, 23.96240234375, 192.33840942382812, 176.55691528320312, 22.987060546875, -1.19366455078125, 120.84597778320312, -143.55795288085938, 8.438697814941406, 140.26461791992188, 40.29023742675781, -113.01321411132812, 8.456268310546875, 120.01377868652344, 123.68721771240234, 121.653564453125, -56.045562744140625, -14.125152587890625, 138.5606689453125, 5.544963836669922, -0.9175186157226562, 4.70428466796875, -20.845703125, 1.576751708984375, 21.887298583984375, -112.19087219238281, 10.438226699829102, 47.16212463378906, 33.079345703125, 115.53372192382812, 2.610260009765625, -3.8556900024414062, 142.7930908203125, 141.23046875, -142.72906494140625, -3.5211944580078125, -24.490158081054688, 35.76220703125, 20.614715576171875, 141.69430541992188, -55.76019287109375, 28.912109375, 20.729248046875, 2.254058837890625, 122.9141845703125, -5.652463912963867, -132.62220764160156, 16.263748168945312, 148.51974487304688, 108.30381774902344, -100.46383666992188, 66.016845703125, 205.14984130859375, 124.36982727050781, 52.68567657470703], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000328.npy"} {"epoch": 0.6869109947643979, "step": 329, "batch_size": 128, "mean": 42.954986572265625, "std": 87.05113983154297, "min": -171.00335693359375, "p10": -61.30440063476562, "median": 31.299049377441406, "p90": 156.4231018066406, "max": 211.5380859375, "pos_frac": 0.6875, "sample": [-9.766998291015625, 6.6488189697265625, -6.072357177734375, 44.57928466796875, 2.6740951538085938, -12.941814422607422, 8.408218383789062, 25.421524047851562, 3.9493942260742188, 31.957916259765625, 141.6417694091797, -16.811904907226562, -109.83892822265625, 112.34457397460938, 101.80924987792969, 175.41726684570312, -7.747283935546875, -55.55718994140625, 125.04850769042969, 104.42157745361328, 53.1614990234375, 20.54595947265625, 153.76565551757812, -13.50262451171875, 122.70213317871094, 152.7742156982422, -131.0825958251953, 115.89999389648438, 172.16746520996094, 58.662109375, 0.42657470703125, 22.672164916992188, 32.975341796875, 153.57781982421875, -39.1390380859375, 41.493011474609375, 102.07537841796875, 105.1815185546875, 17.700210571289062, 107.45747375488281, 1.04302978515625, 168.4495849609375, 1.297393798828125, -41.463134765625, 92.43850708007812, -96.0306396484375, -14.334991455078125, 152.58087158203125, 14.0938720703125, -171.00335693359375, -4.7118072509765625, 169.36709594726562, -64.96572875976562, -18.8641357421875, 112.449462890625, 30.640182495117188, 44.69879150390625, 37.3314208984375, 35.35813903808594, -117.66436767578125, 8.669120788574219, 150.19497680664062, -1.51043701171875, -28.777023315429688, 0.5761260986328125, -4.815797805786133, -0.46048736572265625, -59.735260009765625, 197.07257080078125, 43.04686737060547, 33.11671447753906, 1.600372314453125, 124.70442199707031, 207.2569580078125, 13.510457992553711, 137.39736938476562, 67.01724243164062, -25.262619018554688, -85.32781982421875, 146.78924560546875, 136.5588836669922, -11.136749267578125, -167.641357421875, 115.91294860839844, -14.18133544921875, 169.82826232910156, 107.34889221191406, 1.75030517578125, 90.58071899414062, 0.8366546630859375, 181.9547119140625, 1.2992782592773438, -116.53092956542969, 173.24884033203125, 128.954345703125, -123.1044921875, 204.88414001464844, 162.62380981445312, -3.2288169860839844, 123.20843505859375, -5.4255523681640625, 13.884002685546875, 114.94461059570312, 33.336029052734375, -0.12249755859375, 0.8950004577636719, 205.3875732421875, 51.87493896484375, 141.34544372558594, 211.5380859375, -14.893798828125, 114.56765747070312, 128.5639190673828, -8.621429443359375, 37.1998291015625, -95.50670623779297, 24.4375, -20.42303466796875, 47.106414794921875, -5.94769287109375, 77.46420288085938, 128.49169921875, 1.6281166076660156, 54.01647186279297, 136.7969512939453, 46.283416748046875, -105.3292236328125, -149.29315185546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000329.npy"} {"epoch": 0.6890052356020943, "step": 330, "batch_size": 128, "mean": 37.453285217285156, "std": 80.18151092529297, "min": -132.31239318847656, "p10": -67.46213378906249, "median": 24.753948211669922, "p90": 153.5240753173828, "max": 251.9822998046875, "pos_frac": 0.7421875, "sample": [-4.8975067138671875, 135.341552734375, 18.441726684570312, -101.4705810546875, 92.96576690673828, 130.51898193359375, 38.367095947265625, 175.80740356445312, 133.2994842529297, 39.80194091796875, -64.31414794921875, 58.052001953125, 6.26153564453125, -104.0499267578125, 59.47308349609375, -31.8797607421875, 155.7421875, 29.840667724609375, 3.80029296875, 0.158599853515625, 156.86636352539062, -132.31239318847656, -31.368606567382812, 7.448150634765625, 3.544189453125, 114.75244140625, 26.65325927734375, -7.51318359375, 10.252349853515625, -0.5677490234375, 60.00653076171875, 82.77584838867188, 170.93109130859375, 103.54020690917969, 8.854934692382812, -7.01416015625, -1.04998779296875, -95.03602600097656, -74.80743408203125, 16.42529296875, 45.52899169921875, 166.51406860351562, 36.347412109375, 41.07696533203125, 66.19009399414062, 82.55349731445312, 17.926189422607422, 187.63919067382812, 13.994834899902344, 105.23446655273438, 2.9575862884521484, 152.57345581054688, 81.86428833007812, -123.18869018554688, 9.243011474609375, 24.02813720703125, -121.21558380126953, 34.685699462890625, 92.1282958984375, 28.418350219726562, 14.188003540039062, 169.40545654296875, 12.721565246582031, 112.14956665039062, -122.02314758300781, 4.555694580078125, 130.878173828125, -95.501220703125, 5.236492156982422, 21.652267456054688, 44.058929443359375, -55.88322448730469, 172.03466796875, 72.65008544921875, -2.18450927734375, 86.96823120117188, -131.29696655273438, 172.00680541992188, -42.932281494140625, 6.324371337890625, 23.08551025390625, 91.45314025878906, -4.246406555175781, -18.133270263671875, 24.187896728515625, 5.2974395751953125, 30.59027099609375, 151.06735229492188, 98.81771850585938, 9.645591735839844, 168.44009399414062, 13.4888916015625, 103.21674346923828, 42.736053466796875, 21.57354736328125, -32.89385986328125, 15.975067138671875, 14.774578094482422, 65.54290008544922, 136.64556884765625, 173.419189453125, 32.2056884765625, -118.77008056640625, 62.63494873046875, -123.92721557617188, 146.263671875, -14.4130859375, 25.31999969482422, -19.379173278808594, 78.01434326171875, 116.70193481445312, -107.71359252929688, 1.2592239379882812, -55.549835205078125, 1.9906005859375, 180.96420288085938, 43.28839111328125, 45.85089874267578, 1.52996826171875, 133.79920959472656, -24.55560302734375, 251.9822998046875, 77.69288635253906, -8.24920654296875, -26.8612060546875, 135.376708984375, 90.90176391601562, 29.828369140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000330.npy"} {"epoch": 0.6910994764397905, "step": 331, "batch_size": 128, "mean": 45.837799072265625, "std": 78.2320785522461, "min": -165.2740936279297, "p10": -33.23793334960937, "median": 26.33740234375, "p90": 157.24199676513672, "max": 190.07269287109375, "pos_frac": 0.765625, "sample": [2.0461959838867188, 183.33538818359375, 155.53802490234375, -93.43997192382812, 188.9176025390625, 90.94882202148438, -1.6278076171875, -7.468597412109375, 44.03765869140625, 104.68548583984375, -1.2933502197265625, -68.29576110839844, 73.45321655273438, -89.26617431640625, 181.427978515625, 15.703094482421875, 4.0844268798828125, 3.646350860595703, -9.957000732421875, -72.12289428710938, 14.266777038574219, 50.762420654296875, -31.78216552734375, 24.6617431640625, 22.948379516601562, 153.33364868164062, 120.06399536132812, 33.12293243408203, 0.8750457763671875, 128.46951293945312, 36.4813232421875, 43.3758544921875, -13.370849609375, 26.39129638671875, 25.378990173339844, 157.11380004882812, 10.86444091796875, 79.56805419921875, 167.1884765625, 124.72628021240234, -47.17253112792969, 13.12359619140625, -16.991989135742188, -158.3829803466797, 4.252540588378906, 7.841709136962891, 157.54112243652344, 22.953155517578125, 48.943115234375, 0.0, 36.90087890625, -0.0814208984375, 127.19125366210938, -8.974090576171875, 10.151824951171875, 77.54019165039062, 9.337158203125, 145.2373809814453, 35.871368408203125, 54.397735595703125, 26.28350830078125, 49.539306640625, -33.89593505859375, -48.2861328125, 176.29681396484375, 48.78157043457031, 162.18978881835938, 103.98611450195312, 21.300209045410156, 27.708396911621094, 4.92425537109375, 27.54248046875, 150.34161376953125, -6.23077392578125, 37.45102310180664, 16.41913604736328, -51.958404541015625, 5.121673583984375, 10.099296569824219, -32.9559326171875, 42.93597412109375, 30.212860107421875, 120.70806884765625, 30.778091430664062, -165.2740936279297, 183.43017578125, 8.15216064453125, 156.9312744140625, -28.2490234375, 31.499664306640625, 177.8118896484375, 123.44992065429688, 104.93358612060547, 180.22076416015625, 46.69627380371094, 161.87637329101562, 146.1173858642578, 13.229804992675781, -16.69921875, -8.978118896484375, -47.3695068359375, 24.63226318359375, 151.09654235839844, 121.34661865234375, -138.0511474609375, 23.544647216796875, 190.07269287109375, 2.956989288330078, 128.30758666992188, -2.0612258911132812, 3.363250732421875, 12.968269348144531, 137.04879760742188, 13.212158203125, 116.20014190673828, 129.48907470703125, 1.03765869140625, 72.3472900390625, 18.304912567138672, 123.14041137695312, -146.37454223632812, 23.22747039794922, 174.72018432617188, 129.37403869628906, 53.524017333984375, -30.269439697265625, 134.2381591796875, 14.296432495117188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000331.npy"} {"epoch": 0.6931937172774869, "step": 332, "batch_size": 128, "mean": 57.80183029174805, "std": 83.78578186035156, "min": -178.2465362548828, "p10": -38.16046142578124, "median": 47.54609680175781, "p90": 158.8087951660156, "max": 228.09005737304688, "pos_frac": 0.75, "sample": [70.54306030273438, 44.863250732421875, 10.722496032714844, 182.78445434570312, 124.58078002929688, 134.64813232421875, 77.96546936035156, 89.1003189086914, 187.47259521484375, 132.0404815673828, -0.7859649658203125, 10.310602188110352, 9.181354522705078, -5.310413360595703, 93.39460754394531, 107.26568603515625, -32.894073486328125, 21.536453247070312, 142.96759033203125, -75.19178771972656, 119.00802612304688, -1.0569610595703125, 44.38525390625, 16.881778717041016, 0.8665695190429688, 23.357513427734375, -35.972259521484375, 4.4978485107421875, 8.550865173339844, 12.027389526367188, 147.171630859375, 163.775634765625, 2.787872314453125, 2.664398193359375, 153.60244750976562, 114.8094253540039, 138.72528076171875, 164.233154296875, 94.7713623046875, 104.48516845703125, 43.393157958984375, 7.7335205078125, 128.80728149414062, -0.2609443664550781, 138.6304931640625, 30.718841552734375, 160.42510986328125, 139.58358764648438, -2.8899688720703125, 27.79901123046875, 148.59381103515625, 180.26376342773438, -106.46676635742188, -96.52471923828125, 181.78778076171875, 127.19844818115234, -67.0311279296875, 72.70654296875, 150.58389282226562, -158.24981689453125, 41.611656188964844, 154.23489379882812, 143.4375, -2.994607925415039, 13.054664611816406, 6.82275390625, 107.87909698486328, 22.1134033203125, 12.05589485168457, -43.266265869140625, -7.425811767578125, 158.1160888671875, 110.95576477050781, 18.453460693359375, -2.367034912109375, 168.70123291015625, 92.61105346679688, 118.69295501708984, -64.31195831298828, 77.99124145507812, -20.513137817382812, 119.94728088378906, -24.654861450195312, 1.9648590087890625, 119.79067993164062, -107.1556396484375, 86.55120849609375, 20.843780517578125, 5.3841552734375, 115.17239379882812, 152.47067260742188, 107.37750244140625, -94.5700454711914, 146.5296630859375, 71.31195068359375, -11.5506591796875, 140.94451904296875, -22.680824279785156, -15.006851196289062, 40.59510803222656, 161.215576171875, 149.44561767578125, -116.998779296875, -44.47662353515625, 228.09005737304688, 185.7999267578125, 127.46739196777344, 19.39794921875, 13.001678466796875, 16.814682006835938, 153.01124572753906, 62.612518310546875, 122.8990249633789, 176.94021606445312, 53.244384765625, 131.35919189453125, -25.31097412109375, 145.87652587890625, -9.59842300415039, 121.02965545654297, 111.49481201171875, 173.985107421875, -7.454566955566406, -104.44398498535156, 0.0, 10.562652587890625, -178.2465362548828, 50.22894287109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000332.npy"} {"epoch": 0.6952879581151833, "step": 333, "batch_size": 128, "mean": 42.738197326660156, "std": 75.06159973144531, "min": -135.649169921875, "p10": -55.83508758544921, "median": 30.221660614013672, "p90": 140.49305114746093, "max": 174.1746826171875, "pos_frac": 0.7109375, "sample": [122.59615325927734, -127.32229614257812, -104.62522888183594, 164.4814453125, 140.5291748046875, -42.7540283203125, 84.38095092773438, 49.84222412109375, 57.436798095703125, 118.157470703125, -52.82231140136719, 9.48779296875, -16.882431030273438, 152.16546630859375, 140.47756958007812, 136.8009033203125, 126.79647064208984, 20.334320068359375, 103.83275604248047, 49.766082763671875, -22.797012329101562, 25.976703643798828, 10.636871337890625, 144.05908203125, 135.67286682128906, 69.33062744140625, 56.46282958984375, 104.32135009765625, 131.862060546875, -4.118621826171875, 26.023345947265625, 166.33824157714844, -11.1456298828125, 74.17779541015625, -135.649169921875, 3.988262176513672, 53.937103271484375, -11.25518798828125, 132.558349609375, 125.14706420898438, 49.51409912109375, 20.50455665588379, 10.962158203125, 12.09893798828125, 80.96438598632812, 136.46824645996094, 4.0869293212890625, 111.23178100585938, 25.41802978515625, -67.42155456542969, -64.5338134765625, 72.36276245117188, 0.0, -4.52813720703125, 134.9935760498047, 19.748291015625, 146.91375732421875, 161.39935302734375, 140.7557373046875, 83.54556274414062, -0.3056640625, 44.98716735839844, -43.025726318359375, 135.30545043945312, 84.29107666015625, -29.983047485351562, 139.93576049804688, 174.1746826171875, 44.65147399902344, 33.06182861328125, -111.21208190917969, 76.95208740234375, 35.62591552734375, -64.60699462890625, 5.222846984863281, 132.71917724609375, -2.634124755859375, -3.82000732421875, 5.96624755859375, 127.65738677978516, 37.23091125488281, 44.78363800048828, 28.125274658203125, 154.02874755859375, -62.864898681640625, 146.25596618652344, 147.382568359375, -7.64166259765625, -84.605712890625, 146.48019409179688, 30.531227111816406, 91.73419189453125, -9.365928649902344, 108.43099975585938, -24.583534240722656, -19.3765869140625, 23.508636474609375, 29.912094116210938, -122.27932739257812, 108.77444458007812, -12.44342041015625, 92.12505340576172, -111.58511352539062, -5.5974884033203125, 108.13916015625, 131.204833984375, 1.14697265625, 49.196380615234375, 139.79922485351562, 11.954132080078125, 122.47875213623047, 3.3310375213623047, 123.85122680664062, 29.4927978515625, 14.699188232421875, -5.49212646484375, -95.70582580566406, 59.96155548095703, -18.457275390625, 29.058563232421875, -8.318511962890625, 6.4169921875, -116.01017761230469, 18.019142150878906, -0.0164794921875, 47.04307556152344, 19.6363525390625, 2.4449520111083984], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000333.npy"} {"epoch": 0.6973821989528796, "step": 334, "batch_size": 128, "mean": 50.78205490112305, "std": 86.75420379638672, "min": -186.7730255126953, "p10": -50.4369903564453, "median": 35.087623596191406, "p90": 162.9324737548828, "max": 273.5806884765625, "pos_frac": 0.71875, "sample": [126.47241973876953, 99.4385986328125, -25.09503173828125, -10.44891357421875, 0.0, -7.974822998046875, 5.544677734375, -59.14384460449219, 32.589996337890625, 1.591156005859375, 29.4622802734375, 33.202850341796875, 14.229034423828125, 125.52706909179688, -112.11688232421875, 130.89529418945312, -0.71453857421875, 165.21743774414062, 47.99247741699219, -14.497802734375, 117.44648742675781, 64.74163818359375, 203.0642547607422, 161.99945068359375, -86.0181884765625, 35.04646301269531, 20.410446166992188, 25.60321044921875, -3.0965957641601562, 14.080184936523438, 77.5474853515625, -10.646842956542969, -0.1802997589111328, 103.27764892578125, 130.62844848632812, 35.1287841796875, -1.5629196166992188, 117.23460388183594, -165.29354858398438, 141.53713989257812, -1.7864761352539062, 3.943695068359375, 20.70318603515625, 18.6634521484375, 29.493324279785156, 74.31329345703125, 10.90771484375, -57.5677490234375, 132.7761688232422, -1.5010986328125, 97.74075317382812, -123.23468017578125, 153.02195739746094, 94.34490966796875, 191.13531494140625, 135.52767944335938, -5.235040664672852, 86.04951477050781, 8.739273071289062, 18.200881958007812, 17.585365295410156, 6.523521423339844, 190.02398681640625, 65.61605834960938, -38.85020446777344, -121.01535034179688, 48.188934326171875, -120.14932250976562, 2.7204742431640625, -26.120559692382812, 172.6358642578125, 32.083953857421875, -16.17059326171875, 178.34918212890625, 141.823974609375, 128.63058471679688, 90.25709533691406, 17.089134216308594, 11.2889404296875, 131.481201171875, 41.012939453125, 54.828426361083984, -5.40484619140625, 127.1888427734375, 108.98056030273438, 171.50733947753906, 87.95365142822266, 54.76470947265625, 140.4595947265625, 70.27398681640625, 135.14288330078125, 130.86416625976562, 139.59425354003906, -19.27099609375, 1.3647537231445312, 26.136276245117188, -23.550613403320312, -47.380950927734375, 155.00656127929688, 173.85110473632812, 165.17672729492188, 27.157623291015625, 8.055492401123047, -115.42791748046875, 146.7027587890625, -83.322265625, 95.22252655029297, 123.76455688476562, 165.10952758789062, 140.90496826171875, -186.7730255126953, 194.9393310546875, 153.66143798828125, 0.4610595703125, 128.13754272460938, -151.59408569335938, -11.37811279296875, 91.01327514648438, -30.856040954589844, 56.698211669921875, 35.625701904296875, 273.5806884765625, 134.8270263671875, 75.31454467773438, 65.80326843261719, -11.897445678710938, -65.64361572265625, 190.16900634765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000334.npy"} {"epoch": 0.6994764397905759, "step": 335, "batch_size": 128, "mean": 45.94326400756836, "std": 86.73538970947266, "min": -170.67352294921875, "p10": -56.88970947265624, "median": 35.49650955200195, "p90": 157.20345764160157, "max": 207.50296020507812, "pos_frac": 0.6875, "sample": [-95.47805786132812, -143.88674926757812, 32.05132293701172, -43.83259582519531, 108.646728515625, -92.69058227539062, 179.5611572265625, -9.916290283203125, 38.94169616699219, 136.51239013671875, 95.72512817382812, -148.5743408203125, 43.19000244140625, -4.602163314819336, -3.5282440185546875, 23.1068115234375, 6.850791931152344, -1.9090709686279297, -27.39306640625, -60.856689453125, -16.6505126953125, 173.30520629882812, -4.244384765625, 6.1597900390625, 13.936767578125, -1.11260986328125, 82.06913757324219, 68.77558135986328, -34.382598876953125, 133.9434814453125, -170.67352294921875, 187.8956298828125, -22.449493408203125, 96.12850952148438, 47.86895751953125, 187.32754516601562, 8.75958251953125, 129.43771362304688, 150.0687713623047, 9.617687225341797, 158.06301879882812, 136.6903839111328, 152.47967529296875, 121.95726013183594, 180.608642578125, 157.04147338867188, 73.36734008789062, 127.2914810180664, 142.671875, 160.2525634765625, -7.561279296875, 113.61344909667969, 13.825004577636719, 118.90369415283203, 130.09249877929688, 131.19540405273438, 190.85165405273438, 85.303466796875, 4.215919494628906, 39.7489013671875, 47.395721435546875, -134.07275390625, 0.78466796875, -39.08056640625, -114.00186157226562, 146.5650634765625, 82.48408508300781, 147.25189208984375, 90.01067352294922, 14.962234497070312, 27.024200439453125, 0.80010986328125, -19.303451538085938, 124.71493530273438, 188.05996704101562, 135.946533203125, 4.940399169921875, -77.69392395019531, 11.066268920898438, 22.278181076049805, 79.21241760253906, 94.91842651367188, -9.263092041015625, 117.50270080566406, 161.4899139404297, -16.07050323486328, -55.1895751953125, 4.4863739013671875, 87.31521606445312, 31.122222900390625, -40.35125732421875, -1.6537017822265625, 119.11761474609375, 3.0838470458984375, -19.68341827392578, -15.096588134765625, 13.982986450195312, 21.429664611816406, 81.69619750976562, -7.0787811279296875, 176.71585083007812, -135.18316650390625, 47.67779541015625, 74.07276153564453, 139.13803100585938, 65.94342041015625, 0.0, 150.94744873046875, 157.5814208984375, 130.02108764648438, -102.49261474609375, 123.85621643066406, 138.31851196289062, -118.83233642578125, -123.21559143066406, 207.50296020507812, 6.3159942626953125, -1.9679183959960938, 10.605039596557617, -29.7537841796875, 10.287109375, 111.35354614257812, 48.29736328125, 105.22251892089844, -52.223731994628906, -15.77197265625, 81.907470703125, 54.99971008300781], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000335.npy"} {"epoch": 0.7015706806282722, "step": 336, "batch_size": 128, "mean": 44.32357406616211, "std": 77.17157745361328, "min": -148.72064208984375, "p10": -35.633280944824214, "median": 28.004947662353516, "p90": 152.93724060058594, "max": 246.3145751953125, "pos_frac": 0.6875, "sample": [60.81390380859375, 4.48590087890625, 82.81759643554688, -27.569580078125, 137.9638214111328, 40.79566955566406, -20.81292724609375, 124.15628051757812, 123.38299560546875, 147.8187255859375, -9.919685363769531, 24.381103515625, 81.1511459350586, 52.488616943359375, 3.098644256591797, 61.336334228515625, -1.83905029296875, 174.8459930419922, 8.238777160644531, 153.90249633789062, 163.84898376464844, 26.319747924804688, 109.27061462402344, 131.66107177734375, -56.07147216796875, -16.453277587890625, 5.78564453125, 17.61658477783203, 15.176223754882812, 51.748291015625, 136.18264770507812, 46.335411071777344, -2.5166091918945312, 133.87081909179688, 165.9259033203125, 156.39598083496094, 47.51422119140625, 205.06076049804688, 17.16179656982422, 9.73193359375, -114.74211120605469, 134.43316650390625, -42.856536865234375, 142.374267578125, 97.46063232421875, -1.5139999389648438, -1.2592716217041016, 137.66842651367188, -37.66300964355469, 23.329837799072266, 97.94487762451172, 0.75543212890625, 20.88629150390625, 98.17453002929688, 37.55397033691406, -51.026458740234375, -13.689208984375, -13.467926025390625, 99.60445404052734, -14.536109924316406, 91.45155334472656, 31.237762451171875, -79.8365478515625, 77.05010986328125, -25.7020263671875, 94.08490753173828, 46.912078857421875, -50.007568359375, 95.24430847167969, 134.10540771484375, 3.978515625, -10.202621459960938, 142.67372131347656, 66.04106140136719, -5.012809753417969, 6.921875, -6.68463134765625, 18.75665283203125, 41.22593688964844, -50.79792785644531, 2.9503421783447266, 38.4967041015625, 46.254119873046875, -124.51235961914062, 125.35765075683594, 37.598236083984375, 116.46890258789062, 59.342681884765625, 21.389076232910156, 152.5235595703125, -20.73907470703125, -121.3182601928711, 163.76991271972656, 161.25662231445312, 68.848876953125, 117.614990234375, 173.78182983398438, 29.690147399902344, -5.918182373046875, 71.69465637207031, -5.103513717651367, 159.8878631591797, 3.67425537109375, -30.92919921875, -129.94070434570312, 246.3145751953125, 16.554443359375, 141.40098571777344, 188.5823974609375, -148.72064208984375, 4.351070404052734, -8.450370788574219, -0.1316375732421875, 149.2470245361328, 25.9505615234375, 10.093643188476562, 30.622802734375, -15.43865966796875, -96.3101806640625, 0.0, -17.42669677734375, 23.69781494140625, 39.154205322265625, 158.62969970703125, -34.763397216796875, -22.40869140625, -16.400985717773438, 79.75767517089844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000336.npy"} {"epoch": 0.7036649214659686, "step": 337, "batch_size": 128, "mean": 51.21649169921875, "std": 83.5705795288086, "min": -130.2268524169922, "p10": -64.62302246093749, "median": 40.58372497558594, "p90": 161.44701538085937, "max": 197.5855712890625, "pos_frac": 0.7109375, "sample": [34.469512939453125, -50.98945617675781, 7.3211669921875, 0.9812850952148438, 14.763492584228516, -3.759613037109375, -12.954200744628906, 177.41714477539062, 100.9068603515625, 197.5855712890625, -2.7783966064453125, 92.619384765625, 19.147613525390625, 42.19964599609375, 174.12615966796875, 160.48025512695312, -63.057525634765625, 45.59440612792969, 2.7044677734375, 19.337112426757812, -113.07382202148438, -78.19485473632812, -38.29075622558594, -44.35821533203125, 28.307518005371094, 3.4963836669921875, 21.45025634765625, 154.0275421142578, -0.6330070495605469, 144.47903442382812, 9.93658447265625, 174.17919921875, 78.67465209960938, 155.9293212890625, 151.888427734375, 128.42672729492188, 41.77130126953125, 9.318168640136719, 100.02426147460938, 110.34628295898438, -11.499458312988281, -14.979110717773438, -32.71502685546875, 142.88772583007812, 57.952484130859375, 145.10064697265625, -9.961959838867188, -121.36337280273438, 13.957763671875, 32.6876220703125, 19.026275634765625, 131.57310485839844, -14.591384887695312, 120.72427368164062, 112.07125854492188, -100.36217498779297, 80.93939208984375, 106.71025085449219, 65.8173828125, 43.95343017578125, 1.2345733642578125, -5.460121154785156, 56.43568420410156, 169.60589599609375, 143.78872680664062, -28.122314453125, 136.35433959960938, 133.01675415039062, -130.2268524169922, 27.24859619140625, 91.50592041015625, 169.37155151367188, 39.396148681640625, 28.01416015625, 121.13252258300781, 2.7891693115234375, -5.70660400390625, -25.32056427001953, 160.0313720703125, 114.58138275146484, 14.415651321411133, 72.0440673828125, 103.4461441040039, 151.25750732421875, 129.02432250976562, 163.70278930664062, -17.04776954650879, 110.32803344726562, -37.99766540527344, -98.55155944824219, 56.55549621582031, 148.00672912597656, 38.25494384765625, 185.62423706054688, -86.63497924804688, 180.515625, -68.27584838867188, -7.9666748046875, 114.2863998413086, 119.51611328125, 81.1523208618164, -16.44573974609375, -107.36456298828125, 165.17491149902344, -49.577789306640625, -90.86380767822266, 175.67425537109375, 83.67111206054688, 4.680942535400391, 2.443265914916992, -28.393112182617188, 32.17534637451172, 146.8536376953125, 171.2132568359375, -84.9134521484375, 163.90711975097656, -0.65325927734375, 3.631561279296875, -80.34161376953125, 131.7357177734375, 133.58465576171875, 111.33969116210938, -77.81704711914062, 47.65330505371094, 14.31842041015625, 152.18246459960938, 148.38465881347656, 90.38140869140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000337.npy"} {"epoch": 0.7057591623036649, "step": 338, "batch_size": 128, "mean": 59.165679931640625, "std": 79.74364471435547, "min": -172.79173278808594, "p10": -27.584963989257812, "median": 41.29652404785156, "p90": 165.36796264648436, "max": 213.65789794921875, "pos_frac": 0.7734375, "sample": [167.271728515625, -9.646272659301758, 7.656463623046875, 128.0498809814453, 157.83261108398438, 174.82476806640625, 124.0019302368164, 128.24530029296875, 1.9727859497070312, -43.768798828125, 46.27187728881836, 141.51629638671875, 35.8475341796875, 14.814306259155273, 153.47610473632812, 27.126235961914062, 133.45816040039062, 74.82373046875, -8.004409790039062, -33.078582763671875, 162.35430908203125, 76.51460266113281, -1.0372066497802734, 180.40191650390625, 5.950714111328125, 57.990814208984375, 142.95877075195312, 84.8470687866211, -7.869232177734375, 1.6978168487548828, 41.113433837890625, 10.71002197265625, -3.230987548828125, 18.7830810546875, 62.259246826171875, 124.30436706542969, 69.6650390625, 157.03872680664062, 26.688430786132812, 38.65773010253906, 96.09027099609375, 33.705780029296875, 97.57051086425781, 152.00933837890625, 166.91583251953125, 19.217681884765625, -31.43682861328125, 147.22494506835938, 60.48980712890625, 41.4796142578125, 92.48089599609375, -23.29248046875, 161.72592163085938, -128.53912353515625, 35.398712158203125, -5.290565490722656, -11.3458251953125, 176.85214233398438, -106.83578491210938, 161.93023681640625, 132.6809539794922, 213.65789794921875, 149.3746337890625, 44.24713134765625, 84.51815795898438, 34.911529541015625, 1.5637283325195312, -12.802078247070312, -109.59698486328125, -5.408672332763672, 7.568576812744141, 129.4401397705078, 177.16473388671875, 22.437957763671875, -27.05059814453125, 3.585355758666992, 29.251361846923828, 156.503173828125, 160.91702270507812, -1.9508285522460938, 68.42987060546875, 1.3111419677734375, 128.56434631347656, -2.35491943359375, 124.59378051757812, 23.1229248046875, -36.768463134765625, 10.161895751953125, 109.35333251953125, 136.8643798828125, 117.31011962890625, 177.4737548828125, 14.9627685546875, -7.02606201171875, -124.58137512207031, 155.19100952148438, 21.39892578125, 187.91598510742188, -102.01673889160156, 84.18888092041016, 32.83216857910156, -28.831817626953125, 17.177818298339844, 70.23440551757812, 11.113937377929688, 164.70458984375, 114.91664123535156, 44.16331481933594, -39.30858612060547, 182.1641845703125, 45.89422607421875, 32.25199890136719, 0.6899223327636719, 36.690185546875, -6.40496826171875, 173.25875854492188, 40.940826416015625, 95.38642883300781, 117.45733642578125, 45.00140380859375, -172.79173278808594, 2.734283447265625, 28.97613525390625, -9.002693176269531, 185.66656494140625, 181.96730041503906, -36.24066162109375, 153.61512756347656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000338.npy"} {"epoch": 0.7078534031413612, "step": 339, "batch_size": 128, "mean": 60.243106842041016, "std": 84.36206817626953, "min": -177.2384033203125, "p10": -46.610952758789054, "median": 57.250091552734375, "p90": 155.86547088623047, "max": 218.61114501953125, "pos_frac": 0.765625, "sample": [68.40061950683594, -44.45635986328125, 86.4574966430664, 110.24127197265625, 128.08486938476562, -10.842361450195312, 29.31658935546875, -59.207916259765625, 125.06886291503906, -3.953826904296875, 180.94622802734375, 19.3919677734375, 7.28472900390625, 6.054557800292969, 142.89321899414062, 8.354034423828125, 51.803680419921875, 101.56991577148438, 97.85810852050781, -59.21502685546875, 138.3140869140625, 12.504364013671875, 64.8072509765625, 121.43775939941406, 0.0, 186.04693603515625, -51.638336181640625, 63.49456787109375, 0.0, 20.99935531616211, 49.95392608642578, -36.293792724609375, 182.69488525390625, 136.4627685546875, 7.98333740234375, -144.1260528564453, 152.76873779296875, 162.2608642578125, 45.51823425292969, 56.1424560546875, 95.91989135742188, 143.899658203125, -30.49603271484375, 127.88385009765625, 104.4366455078125, 75.07659912109375, 124.832275390625, 0.8051528930664062, -4.277961730957031, 131.05792236328125, 28.445953369140625, 10.56451416015625, 218.61114501953125, 111.90054321289062, 140.95919799804688, 137.54434204101562, 20.480201721191406, 18.764442443847656, 146.16073608398438, 139.95413208007812, 155.70413208007812, -3.18609619140625, -36.96293640136719, -83.559326171875, -1.6541099548339844, 173.54959106445312, 6.0242919921875, 117.99716186523438, 123.317138671875, 30.19500732421875, 52.529296875, 17.294219970703125, -120.13957214355469, 88.97613525390625, 134.40890502929688, 27.9522705078125, 152.454345703125, 157.2274169921875, -66.53713989257812, 125.06109619140625, 2.9943008422851562, 134.01748657226562, 156.24192810058594, -10.4686279296875, 148.65560913085938, 216.74923706054688, 62.56549072265625, 0.0, 58.35772705078125, 110.75595092773438, 135.93548583984375, 126.80764770507812, 137.39837646484375, 152.8565673828125, 155.61026000976562, -10.448516845703125, 66.90074157714844, -7.67108154296875, -7.5006103515625, 157.70254516601562, 105.54754638671875, -89.43089294433594, 28.18206787109375, 9.22515869140625, 10.998138427734375, 135.9292449951172, 139.25161743164062, 8.61419677734375, -53.82344055175781, -6.196434020996094, 175.09832763671875, 117.88275146484375, -177.2384033203125, 150.5968475341797, 26.43353271484375, 11.628829956054688, 10.269378662109375, 25.512008666992188, 157.64764404296875, -115.69180297851562, -164.6076202392578, 187.70684814453125, 44.233673095703125, 33.19622802734375, 146.52996826171875, 7.3839874267578125, -94.5207748413086, 144.74200439453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000339.npy"} {"epoch": 0.7099476439790576, "step": 340, "batch_size": 128, "mean": 48.548011779785156, "std": 82.39824676513672, "min": -139.78439331054688, "p10": -48.394879150390615, "median": 33.497093200683594, "p90": 153.76051025390626, "max": 217.74810791015625, "pos_frac": 0.71875, "sample": [197.8325653076172, 78.39765167236328, -32.01594543457031, 140.60113525390625, 1.6475677490234375, 149.91224670410156, -53.86102294921875, -18.28113555908203, -54.5408935546875, 148.0424041748047, -104.2896728515625, 127.57508850097656, -72.70596313476562, 3.2951812744140625, 0.0, 153.66494750976562, 32.09197998046875, -46.05224609375, 114.2611312866211, 155.48460388183594, 143.94210815429688, 11.787261962890625, -43.82557678222656, 154.56515502929688, -79.33006286621094, -16.419601440429688, -24.69476318359375, 136.22842407226562, 8.015335083007812, 150.87620544433594, 189.23980712890625, 2.7605018615722656, 68.23818969726562, 150.96759033203125, 26.56402587890625, -12.901107788085938, 0.47779083251953125, -97.686767578125, 99.87178039550781, 15.07293701171875, 146.34951782226562, 132.92874145507812, -110.88595581054688, 46.25994873046875, 56.808555603027344, 155.7582244873047, -30.226394653320312, 88.63249206542969, 20.69012451171875, -16.296981811523438, 115.456787109375, 182.53677368164062, -9.207441329956055, -89.9315185546875, 9.095930099487305, -10.829315185546875, 135.67153930664062, 101.3541259765625, 31.555694580078125, -31.256561279296875, 7.331451416015625, -139.78439331054688, 158.60560607910156, 0.0, 189.37918090820312, 28.907806396484375, 50.57769775390625, 17.327674865722656, 146.24871826171875, 36.0667724609375, 15.0692138671875, -9.853260040283203, 134.65269470214844, -103.99482727050781, 123.61030578613281, 34.82275390625, 44.89996337890625, -108.88385009765625, 154.69427490234375, 12.1915283203125, 150.95184326171875, 212.04119873046875, -82.222412109375, 29.731231689453125, -19.012908935546875, 115.48442077636719, 63.152099609375, 39.57810974121094, 115.17333984375, 22.78765869140625, 39.876434326171875, -13.109048843383789, 82.48701477050781, -13.855361938476562, 133.087890625, 120.55682373046875, -17.912277221679688, 5.8980712890625, 52.691314697265625, 50.955780029296875, 217.74810791015625, 8.81442642211914, 142.559326171875, -44.09727478027344, 153.98348999023438, -33.00299072265625, 15.250389099121094, -132.38116455078125, 138.83056640625, 114.67381286621094, 133.22262573242188, 37.620933532714844, 12.408084869384766, 108.18550109863281, 156.26473999023438, 152.8625030517578, 17.849822998046875, 39.825416564941406, 13.28717041015625, 69.12728881835938, 32.17143249511719, 65.9915771484375, -18.739013671875, 20.883411407470703, 42.039878845214844, 15.638351440429688, 133.36917114257812, -9.725662231445312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000340.npy"} {"epoch": 0.7120418848167539, "step": 341, "batch_size": 128, "mean": 40.67928695678711, "std": 86.64032745361328, "min": -155.04217529296875, "p10": -71.18021850585937, "median": 21.985122680664062, "p90": 151.90023040771484, "max": 213.24371337890625, "pos_frac": 0.6796875, "sample": [6.0282135009765625, -15.763702392578125, 108.16946411132812, 143.261962890625, 151.39657592773438, -42.66584777832031, 48.40423583984375, 18.483154296875, 40.53788757324219, 4.6468353271484375, 138.07855224609375, 39.75103759765625, 145.26919555664062, 10.45975112915039, 184.13916015625, 163.02342224121094, -67.80755615234375, 24.743896484375, 18.616729736328125, 32.772186279296875, 141.810546875, 15.667388916015625, 129.54840087890625, -124.18302917480469, 180.914794921875, 53.79039001464844, 12.118160247802734, 16.429901123046875, 8.541336059570312, -2.24713134765625, -2.344482421875, 37.26531982421875, -14.7884521484375, -9.264251708984375, -101.068603515625, 30.032379150390625, -61.8819580078125, -99.87677001953125, -0.5390052795410156, 12.4354248046875, 3.8690185546875, 104.96250915527344, 168.09738159179688, 22.105133056640625, 132.6262969970703, 4.76116943359375, 146.4559326171875, 85.91473388671875, -1.8445549011230469, 125.72422790527344, 105.69601440429688, 125.78289794921875, 184.886474609375, -22.180419921875, 147.68927001953125, 124.72802734375, 113.5369873046875, 100.5850830078125, 13.923995971679688, -11.802490234375, 53.07891845703125, -111.33203125, 213.24371337890625, -8.729776382446289, 114.1279296875, -121.37820434570312, -89.88235473632812, 143.54974365234375, 162.87034606933594, 6.388397216796875, 38.1422119140625, 149.28366088867188, 110.5447998046875, -20.03118133544922, 127.53153991699219, 143.51901245117188, 143.6395263671875, 16.687911987304688, 19.165603637695312, -138.9751739501953, 65.48403930664062, 5.232879638671875, -12.583511352539062, -51.14654541015625, 195.25595092773438, 29.05780029296875, -20.34674072265625, 47.17694091796875, 51.6654052734375, 153.07542419433594, -102.943603515625, 88.96929931640625, 182.3070068359375, 137.80044555664062, -99.22187805175781, -12.971954345703125, -54.011985778808594, -1.0658226013183594, -10.793182373046875, 29.66644287109375, -23.350860595703125, 146.4102783203125, 21.8651123046875, -12.833740234375, -124.20875549316406, 34.19752502441406, 36.03533935546875, -45.891937255859375, -155.04217529296875, -29.545318603515625, 16.8004150390625, -68.95147705078125, -12.38330078125, 157.38970947265625, -148.6085662841797, 15.569747924804688, 130.85818481445312, 155.54873657226562, 132.3494110107422, 50.55311584472656, 143.58941650390625, 7.335060119628906, 31.488380432128906, -76.380615234375, 189.52459716796875, 14.102783203125, -13.32342529296875, 12.376670837402344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000341.npy"} {"epoch": 0.7141361256544503, "step": 342, "batch_size": 128, "mean": 50.56574249267578, "std": 78.61018371582031, "min": -173.94204711914062, "p10": -29.079689025878906, "median": 30.737354278564453, "p90": 154.54483947753906, "max": 211.02545166015625, "pos_frac": 0.7421875, "sample": [-5.007843017578125, 23.018325805664062, 46.60723876953125, 20.84893798828125, 110.80999755859375, 201.3988037109375, 19.148616790771484, 83.85958862304688, 6.454566955566406, 37.6683349609375, -28.979415893554688, 13.768096923828125, 0.9768524169921875, 129.54322814941406, 112.25579833984375, 132.37896728515625, -7.325553894042969, 60.791015625, -119.61724090576172, 3.9627532958984375, 50.17549133300781, 7.166107177734375, 13.70037841796875, 30.356101989746094, -49.713706970214844, 10.23980712890625, 12.196121215820312, -43.81829833984375, 46.647308349609375, 127.95816040039062, 30.35009002685547, 120.88555908203125, 105.75410461425781, 77.33609008789062, -8.444488525390625, 15.25201416015625, 15.926651000976562, -12.251947402954102, -49.44966125488281, 33.36602783203125, 39.80957794189453, 174.17823791503906, 19.11865234375, -97.01847839355469, -15.128341674804688, -22.610496520996094, -21.417144775390625, 58.29461669921875, 58.1956787109375, 1.239227294921875, 124.7886962890625, -2.0230712890625, -31.046722412109375, 158.47805786132812, 5.5827789306640625, 183.29196166992188, 150.27188110351562, 0.10321044921875, 88.65338134765625, 122.54632568359375, 153.1441650390625, 157.81307983398438, 114.395751953125, 123.72183990478516, 98.92918395996094, -119.07077026367188, -21.975997924804688, 0.0, -29.31365966796875, 43.0311279296875, 171.65420532226562, 2.7801055908203125, 173.918701171875, 4.295557022094727, 51.94407653808594, 128.52780151367188, 158.64505004882812, 52.52483367919922, 176.3418426513672, 3.943084716796875, -17.800048828125, 136.36407470703125, 31.118606567382812, 108.62165832519531, -64.985107421875, -9.652061462402344, -173.94204711914062, -16.9249267578125, 113.9266357421875, 107.17478942871094, 148.6855010986328, 114.931396484375, -0.9336700439453125, 10.768035888671875, 38.96034240722656, 122.59468078613281, 124.92373657226562, 20.104202270507812, 10.0341796875, -110.35324096679688, 19.977218627929688, 134.5584716796875, -17.264434814453125, 211.02545166015625, 16.96661376953125, 150.3076629638672, -76.7200927734375, 11.778900146484375, 86.40423583984375, -107.65462493896484, 160.02117919921875, 126.483642578125, 52.16804504394531, 143.75082397460938, 132.79531860351562, 147.74456787109375, 16.865982055664062, 20.50616455078125, 133.8499298095703, 145.1895751953125, -13.578697204589844, -12.287109375, -21.5924072265625, 189.9361572265625, 123.68408203125, -6.990100860595703, 179.01910400390625, 15.1014404296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000342.npy"} {"epoch": 0.7162303664921466, "step": 343, "batch_size": 128, "mean": 45.383689880371094, "std": 81.31710815429688, "min": -202.1668701171875, "p10": -42.86712493896484, "median": 28.321182250976562, "p90": 149.64791412353514, "max": 206.67221069335938, "pos_frac": 0.6953125, "sample": [-34.18132019042969, 87.21124267578125, 141.02439880371094, 3.1533260345458984, 40.58026123046875, 68.17462158203125, 117.46932983398438, 20.38970947265625, -51.990753173828125, 28.0748291015625, 147.07176208496094, -40.103118896484375, -8.0802001953125, -40.54705810546875, 104.54058837890625, 172.13958740234375, 158.38316345214844, 46.5235595703125, -10.934799194335938, -7.144350051879883, 199.32379150390625, 19.358261108398438, 156.20571899414062, 148.6920623779297, 111.82769012451172, 58.51664733886719, -1.7557754516601562, 83.24462890625, 26.1558837890625, 130.22389221191406, 71.81890869140625, -104.7723159790039, 21.061920166015625, -9.13836669921875, -121.90824890136719, 39.634666442871094, 15.35479736328125, 142.88522338867188, -65.74334716796875, 6.76531982421875, -42.04066467285156, 176.12606811523438, 40.1275634765625, 75.73541259765625, 0.0, -99.27291870117188, 24.31329345703125, 110.13664245605469, 51.144371032714844, 204.30731201171875, -125.87689208984375, 35.647178649902344, 40.1109619140625, 11.291790008544922, 144.3857421875, -35.633453369140625, 97.19587707519531, 68.41033935546875, -2.3658828735351562, 113.08795928955078, 126.66880798339844, 7.9906463623046875, 4.711666107177734, 4.2242584228515625, 121.04981994628906, 139.7783966064453, -15.3546142578125, 146.52398681640625, 19.0361328125, 79.34176635742188, 93.21726989746094, -20.795745849609375, 104.04681396484375, -27.602630615234375, -0.19757843017578125, 55.89043426513672, 7.537872314453125, 206.67221069335938, 123.2845458984375, 146.09681701660156, 34.904090881347656, 82.7105712890625, -68.2503662109375, 11.990478515625, 130.57489013671875, -18.306427001953125, 4.165069580078125, 34.94598388671875, 13.2811279296875, -28.4678955078125, 8.49125862121582, 144.28460693359375, 143.37075805664062, 197.841064453125, 8.6011962890625, -14.33380126953125, 124.82164001464844, 79.11102294921875, 138.1812744140625, 24.23957061767578, 159.56727600097656, 98.59165954589844, -5.97320556640625, 119.7156982421875, -202.1668701171875, 151.87823486328125, -18.39569091796875, -5.907691955566406, -45.4169921875, -44.7955322265625, 10.766910552978516, 129.8529052734375, 154.0589599609375, 2.793975830078125, -78.64860534667969, -11.33184814453125, -8.116256713867188, -113.19912719726562, 47.969024658203125, 184.33746337890625, 89.23966979980469, -0.16840744018554688, 28.567535400390625, 8.158393859863281, 7.64825439453125, 164.51309204101562, -13.692352294921875, -131.347900390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000343.npy"} {"epoch": 0.7183246073298429, "step": 344, "batch_size": 128, "mean": 55.33774185180664, "std": 82.58492279052734, "min": -167.19415283203125, "p10": -38.125546264648435, "median": 37.33015441894531, "p90": 167.29758148193358, "max": 209.121826171875, "pos_frac": 0.734375, "sample": [26.182586669921875, -9.29345703125, 124.21682739257812, 82.38421630859375, -20.105804443359375, 124.86917114257812, 143.4188690185547, 25.43804931640625, 136.7744598388672, -20.795654296875, 123.38021850585938, 47.86439514160156, 124.68962097167969, -16.678543090820312, 146.15188598632812, 176.97451782226562, 17.751861572265625, 66.87216186523438, -154.02560424804688, 135.90585327148438, -95.70620727539062, 2.510711669921875, 0.0408935546875, -33.608428955078125, 184.29827880859375, 101.52326965332031, -15.971269607543945, 169.96194458007812, 138.00320434570312, 11.63055419921875, 169.44317626953125, -37.401611328125, -7.829010009765625, 134.06845092773438, -167.19415283203125, 61.50959777832031, 142.14691162109375, 125.2811050415039, 13.916595458984375, -9.24102783203125, -149.85890197753906, 24.063140869140625, 32.951171875, -9.709678649902344, -5.152069091796875, 21.484603881835938, -72.5157470703125, 86.1588363647461, 177.54754638671875, 49.43609619140625, -31.959014892578125, -44.963623046875, 1.932952880859375, -48.497352600097656, 105.55502319335938, 102.63338470458984, 173.23037719726562, -11.674102783203125, 187.22344970703125, 166.98057556152344, 146.9058837890625, 144.19619750976562, 206.14913940429688, 143.73959350585938, 155.73350524902344, -8.344039916992188, -8.890228271484375, 177.27236938476562, -13.102859497070312, 47.97454833984375, -51.256561279296875, 14.52471923828125, 13.201324462890625, 197.91085815429688, 62.275299072265625, 37.203125, 16.60394287109375, 80.20309448242188, 30.5499267578125, 97.53106689453125, 31.68535614013672, 21.170806884765625, 57.269012451171875, 137.721435546875, -9.553192138671875, 76.2725830078125, 6.7153167724609375, 145.08624267578125, 20.813552856445312, 59.566558837890625, 188.6837158203125, 108.7191390991211, 18.70855712890625, -54.7391357421875, -10.470001220703125, 158.58694458007812, 76.836669921875, -0.4915733337402344, 35.25929260253906, -39.814727783203125, -7.687578201293945, 32.82183837890625, 17.237945556640625, 209.121826171875, -13.851470947265625, -79.64447021484375, 146.89456176757812, 37.457183837890625, 15.474601745605469, 5.84893798828125, 2.77716064453125, 114.63485717773438, 7.1474456787109375, 163.31051635742188, 102.58645629882812, 42.91062927246094, 115.15386962890625, 80.91856384277344, 36.627716064453125, 168.03726196289062, 125.12939453125, 30.111724853515625, 65.26882934570312, 150.0460205078125, 88.9979248046875, -40.792869567871094, 161.18917846679688, -137.12948608398438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000344.npy"} {"epoch": 0.7204188481675393, "step": 345, "batch_size": 128, "mean": 43.91633224487305, "std": 72.97171783447266, "min": -186.2037353515625, "p10": -26.666807556152342, "median": 22.416091918945312, "p90": 151.38170166015624, "max": 186.88522338867188, "pos_frac": 0.7578125, "sample": [49.07476806640625, 95.42591094970703, 79.75128173828125, -124.282470703125, 173.95782470703125, 121.84933471679688, 11.1005859375, -41.38909912109375, 15.072662353515625, -1.279287338256836, 57.15931701660156, 12.118942260742188, -0.22168731689453125, 3.6314239501953125, 36.806549072265625, -3.7779541015625, 162.68585205078125, 15.245697021484375, 12.61602783203125, -6.43646240234375, 15.791015625, -24.236236572265625, 127.0580825805664, 144.9313201904297, 3.628133773803711, 40.26141357421875, 131.876220703125, 31.339447021484375, -35.4356689453125, 0.6138763427734375, -10.190673828125, 25.065643310546875, -16.54623794555664, 28.864227294921875, 0.781280517578125, 73.43411254882812, 57.65625, 6.83050537109375, 18.152008056640625, 14.93133544921875, 31.620758056640625, -8.173467636108398, -32.65928268432617, 133.26708984375, 13.048828125, 145.2046356201172, 13.341400146484375, 0.5081043243408203, 45.08556365966797, 183.2073974609375, 128.17520141601562, 132.354736328125, 82.59552001953125, 157.4590606689453, 166.3203582763672, 4.430328369140625, 19.56524658203125, -27.6988525390625, 15.074325561523438, -9.224617004394531, 152.52996826171875, 53.155029296875, 54.89463806152344, 42.83172607421875, 22.07366943359375, -72.17019653320312, -186.2037353515625, 15.314109802246094, -111.50582885742188, -20.618438720703125, 161.86032104492188, 167.13897705078125, 144.9201202392578, 21.646072387695312, 34.029510498046875, -35.476043701171875, 186.88522338867188, -74.74658203125, 38.551849365234375, -2.842010498046875, 150.88958740234375, 157.55328369140625, 119.85620880126953, -116.1041259765625, 81.579345703125, 14.584362030029297, -26.224502563476562, 179.54031372070312, 38.73931884765625, 86.1498031616211, 163.11573791503906, 134.79110717773438, 21.996124267578125, 25.18792724609375, 110.00604248046875, 15.500579833984375, 50.001747131347656, 140.7198486328125, 141.26040649414062, -19.6707763671875, 147.100341796875, -45.63507080078125, 127.09285736083984, -91.69845581054688, 38.3797607421875, 43.069488525390625, 3.953521728515625, 22.758514404296875, -9.456161499023438, -12.52593994140625, -13.147476196289062, 118.0520248413086, 18.417999267578125, 33.94984436035156, 154.62933349609375, 129.847900390625, 13.924835205078125, 17.08919906616211, 0.0, 24.175186157226562, 0.5977554321289062, 19.183868408203125, 6.927314758300781, -0.06937408447265625, 140.77084350585938, 5.9814453125, 98.98440551757812, 69.77841186523438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000345.npy"} {"epoch": 0.7225130890052356, "step": 346, "batch_size": 128, "mean": 53.35143280029297, "std": 83.0159912109375, "min": -186.8486328125, "p10": -36.22458190917968, "median": 38.756004333496094, "p90": 161.61268615722656, "max": 188.28860473632812, "pos_frac": 0.734375, "sample": [-48.702972412109375, -33.27783203125, 1.0138912200927734, -47.42185974121094, 107.8265380859375, 66.97723388671875, 169.24996948242188, 168.41134643554688, 129.568603515625, -7.8659210205078125, 133.80690002441406, -19.54632568359375, -30.85809326171875, 27.0205078125, 138.48114013671875, -133.36572265625, -90.67222595214844, 2.794189453125, 22.67803955078125, 161.31430053710938, 22.091110229492188, -7.5197296142578125, 34.61680603027344, -45.487213134765625, 47.696598052978516, 129.01840209960938, 9.826385498046875, 184.5263671875, -14.421478271484375, -3.609649658203125, 38.612274169921875, 57.32550811767578, -109.04994201660156, 28.666259765625, -2.3033447265625, 154.52633666992188, 89.06549072265625, 164.0325927734375, 12.95611572265625, 18.6669921875, 75.60162353515625, 123.16151428222656, -183.51019287109375, 19.283477783203125, -35.91035461425781, 61.167236328125, -10.98431396484375, 1.5194549560546875, 27.486541748046875, 64.72866821289062, 125.31845092773438, 15.86114501953125, 128.58401489257812, -36.95777893066406, 30.413604736328125, 93.800048828125, 180.72079467773438, 9.123275756835938, -4.613716125488281, 136.52871704101562, 164.22607421875, 171.32769775390625, 29.9981689453125, 13.429924011230469, 22.750534057617188, 42.51299285888672, 166.67388916015625, -16.651382446289062, 140.9959716796875, 102.54598999023438, 103.86184692382812, 116.666015625, 136.91270446777344, 183.38726806640625, 161.761474609375, 145.62322998046875, -108.373291015625, 161.54891967773438, 56.06317138671875, 32.885009765625, 114.1297607421875, -123.84466552734375, 154.7720947265625, 6.42529296875, -6.401824951171875, -11.135852813720703, -35.032554626464844, 5.083314895629883, -28.260711669921875, 163.91455078125, 160.74566650390625, 37.3226318359375, -15.772598266601562, 150.83352661132812, -1.8137702941894531, -13.769432067871094, 122.03318786621094, -186.8486328125, 2.5956802368164062, 154.53387451171875, 1.3269329071044922, 8.148635864257812, 124.61711120605469, 122.73249816894531, 171.9760284423828, 69.24142456054688, 124.55728149414062, 136.59866333007812, 9.760345458984375, 5.72021484375, 0.0, 67.40687561035156, -4.123321533203125, 78.28529357910156, 133.26800537109375, 92.46900939941406, 38.89973449707031, 134.6343994140625, 44.98419952392578, -86.97792053222656, 128.7743682861328, 45.4759521484375, -38.43193054199219, 124.14822387695312, 150.4635009765625, 126.70965576171875, 4.384542465209961, 188.28860473632812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000346.npy"} {"epoch": 0.724607329842932, "step": 347, "batch_size": 128, "mean": 45.90258026123047, "std": 79.77628326416016, "min": -200.86587524414062, "p10": -24.01744461059569, "median": 28.743453979492188, "p90": 155.88534545898438, "max": 179.34683227539062, "pos_frac": 0.71875, "sample": [1.4413299560546875, 76.0733642578125, 137.73733520507812, 15.86334228515625, 65.80853271484375, 110.54096984863281, 10.713424682617188, 153.37161254882812, 14.995445251464844, 148.48081970214844, -200.86587524414062, 86.0545654296875, 147.4297332763672, 24.5367431640625, 162.72817993164062, -85.86331176757812, -39.21759033203125, -2.9909515380859375, 137.95050048828125, 165.54344177246094, 122.41679382324219, -41.04693603515625, 70.950927734375, 11.146953582763672, 33.96250915527344, 38.358154296875, 144.53253173828125, 18.1854248046875, 12.102283477783203, -3.869384765625, 0.61151123046875, 59.99029541015625, 176.20204162597656, 33.05467224121094, 140.1312255859375, 27.088289260864258, 155.42098999023438, -53.310935974121094, 54.9818115234375, 34.093666076660156, 8.027381896972656, 160.1627197265625, 175.8403778076172, 26.099609375, 31.724517822265625, 99.81239318847656, -16.193130493164062, 12.3441162109375, 20.685333251953125, -122.27804565429688, -2.92584228515625, -5.6105499267578125, 74.256591796875, -11.278182983398438, 76.91531372070312, 110.969482421875, -1.4547271728515625, 142.25006103515625, 48.35850524902344, -80.98080444335938, 118.84503173828125, 12.397918701171875, -0.0769805908203125, -17.947040557861328, 0.4980621337890625, 105.63093566894531, 118.3773193359375, 51.02824401855469, 178.38772583007812, 48.50670623779297, 28.256744384765625, -129.89578247070312, -16.01617431640625, 8.705461502075195, 119.42123413085938, -7.956825256347656, -117.23731231689453, -13.793109893798828, 17.731887817382812, 4.8880615234375, 4.394279479980469, 119.72161865234375, 161.78656005859375, -13.7576904296875, 11.919269561767578, -146.27764892578125, 109.25951385498047, 179.34683227539062, 45.51593017578125, 81.90776062011719, -11.839492797851562, 131.6959228515625, -57.956268310546875, 143.32046508789062, -20.031822204589844, -15.521621704101562, 36.20332336425781, -2.0928802490234375, 156.96884155273438, -9.2005615234375, -171.24807739257812, -3.0419921875, 176.15240478515625, 145.67617797851562, 0.0, 23.86627197265625, 20.887176513671875, 34.42095947265625, 123.68075561523438, 161.5877685546875, 29.23016357421875, 5.8856658935546875, 145.33648681640625, 129.0869903564453, -9.788818359375, 17.452590942382812, 40.03302001953125, -5.230194091796875, 133.71649169921875, 37.316864013671875, 6.21453857421875, 161.19598388671875, 175.4878387451172, 36.167816162109375, 15.607994079589844, -33.317230224609375, 153.86029052734375, -1.8816452026367188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000347.npy"} {"epoch": 0.7267015706806282, "step": 348, "batch_size": 128, "mean": 41.151248931884766, "std": 75.0673599243164, "min": -193.0721435546875, "p10": -36.67547912597655, "median": 14.596870422363281, "p90": 148.18421936035156, "max": 194.63075256347656, "pos_frac": 0.6484375, "sample": [60.93096923828125, 3.167144775390625, 3.572183609008789, 4.3590087890625, -6.846038818359375, -5.303985595703125, 139.57701110839844, -8.507640838623047, 78.89773559570312, 120.80522918701172, 12.50982666015625, 145.3203582763672, 31.08258056640625, 132.7816162109375, 95.70941162109375, 12.53192138671875, 180.33834838867188, 120.7293701171875, -75.6572494506836, 0.0, 15.36279296875, 51.3302001953125, 110.7695083618164, -1.2984580993652344, -44.1358642578125, 176.02122497558594, -2.9402618408203125, -1.1491851806640625, -11.548736572265625, 106.03256225585938, 116.35163879394531, 116.34310913085938, -10.423980712890625, -16.579055786132812, 98.0743408203125, -4.55328369140625, 147.84140014648438, -4.346925735473633, 107.949951171875, 158.601806640625, 2.830228805541992, -17.097412109375, -79.2957763671875, 28.49053955078125, -1.50750732421875, 142.53338623046875, 37.0780029296875, 18.730453491210938, 30.75689697265625, 84.28231048583984, 47.71473693847656, 95.553466796875, -58.906982421875, -0.41941070556640625, -12.82037353515625, 89.35475158691406, 158.92684936523438, 181.69793701171875, 157.7470703125, -88.73355102539062, -8.021697998046875, -5.4935302734375, -43.763427734375, 76.44454956054688, 1.3388710021972656, 30.679367065429688, 57.3062744140625, -2.7491722106933594, 6.93707275390625, 40.8359375, 133.0281982421875, -26.497711181640625, -135.59375, -4.777576446533203, -78.961669921875, 44.858734130859375, -47.029327392578125, 7.2582855224609375, 99.94252014160156, 7.1146240234375, 183.61383056640625, 8.50592041015625, 85.85263061523438, 194.63075256347656, 1.30987548828125, 168.4449462890625, -33.637786865234375, -26.43218994140625, -31.01495361328125, 21.59979248046875, -10.170232772827148, 152.87747192382812, 132.78997802734375, -8.895576477050781, 193.27877807617188, -55.873016357421875, 7.402763366699219, 6.636699676513672, -60.276214599609375, 7.960693359375, 148.984130859375, 12.690231323242188, 116.16508483886719, -10.46342658996582, 17.988670349121094, -0.8165740966796875, 144.2449951171875, 67.26327514648438, -12.61907958984375, 139.30108642578125, 114.49693298339844, -29.730255126953125, 76.55340576171875, -59.24903869628906, 9.026962280273438, 48.0745849609375, 161.51930236816406, 34.82708740234375, 145.02340698242188, 33.7760009765625, 123.91485595703125, 3.4770889282226562, -3.9889907836914062, 13.830947875976562, -16.05609130859375, 37.46525573730469, -193.0721435546875, 82.655029296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000348.npy"} {"epoch": 0.7287958115183246, "step": 349, "batch_size": 128, "mean": 56.03184127807617, "std": 83.35742950439453, "min": -195.90673828125, "p10": -24.463891601562498, "median": 34.072540283203125, "p90": 164.6174377441406, "max": 229.381591796875, "pos_frac": 0.7890625, "sample": [136.0618133544922, 150.80320739746094, 26.734619140625, -195.90673828125, -66.52474975585938, 0.5337944030761719, 229.381591796875, 119.59616088867188, -112.22853088378906, 58.8662109375, 43.20310974121094, -64.54310607910156, 138.36953735351562, 175.05279541015625, 114.93927001953125, 137.61297607421875, 189.9693603515625, 10.454360961914062, 125.74325561523438, 176.31472778320312, 73.60543823242188, 2.20684814453125, -23.63677978515625, -21.00543212890625, -5.1180419921875, 52.177825927734375, -18.837890625, -0.11595344543457031, 171.04638671875, 170.8905792236328, 52.93089294433594, 130.3579559326172, 119.68429565429688, -45.99676513671875, 32.96490478515625, 38.32386779785156, 185.1352996826172, -7.33013916015625, 18.231292724609375, 49.401885986328125, 7.625331878662109, 1.3021087646484375, 146.82589721679688, 2.7664947509765625, 10.581405639648438, 72.15069580078125, -9.078460693359375, 205.44064331054688, 24.30828857421875, 5.340728759765625, -59.10542297363281, -0.8674411773681641, -57.39508056640625, 163.2674560546875, 151.40609741210938, 16.442520141601562, 7.1537933349609375, 6.7139434814453125, -108.38543701171875, 71.62553405761719, 0.0, 199.04623413085938, 3.4571533203125, 41.82501220703125, 134.55545043945312, 29.50738525390625, 145.33935546875, 160.20510864257812, 146.850830078125, 7.74644660949707, 156.04461669921875, -25.7305908203125, -108.05964660644531, 204.5987548828125, 2.12359619140625, 48.295135498046875, -6.76995849609375, 120.44999694824219, 2.972381591796875, 27.685775756835938, 19.2406005859375, 119.10053253173828, 167.76739501953125, 101.67425537109375, 147.437255859375, 7.841094970703125, 65.58580017089844, 25.728458404541016, 148.205078125, 46.789642333984375, 62.84770202636719, 7.465087890625, 147.23377990722656, 9.617202758789062, -21.63323974609375, 6.32012939453125, 147.763671875, 106.111572265625, 28.28436279296875, -123.48684692382812, 35.18017578125, 31.171607971191406, 53.47618103027344, -1.292755126953125, -132.218017578125, 20.08538818359375, 2.754669189453125, 5.360712051391602, 21.871368408203125, 138.92984008789062, 21.46575927734375, 141.23330688476562, 28.4925537109375, 128.04335021972656, 141.7262420654297, 173.7618865966797, 151.56060791015625, -23.9210205078125, 96.870361328125, 187.69363403320312, 9.096221923828125, 75.09896850585938, 137.42294311523438, 117.79385375976562, -72.58706665039062, -2.86431884765625, 14.394287109375, 133.96917724609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000349.npy"} {"epoch": 0.7308900523560209, "step": 350, "batch_size": 128, "mean": 49.204673767089844, "std": 86.64383697509766, "min": -154.36822509765625, "p10": -75.10137176513672, "median": 39.04535484313965, "p90": 161.23150024414062, "max": 211.0478515625, "pos_frac": 0.7890625, "sample": [161.22613525390625, 49.40699768066406, 0.0242156982421875, -5.2500762939453125, 9.78530502319336, 18.126419067382812, 34.1285400390625, 180.75253295898438, 2.4439468383789062, 128.96380615234375, 32.736572265625, 14.16790771484375, 154.46945190429688, 0.3113861083984375, 27.426055908203125, 161.15093994140625, 148.16586303710938, 161.2440185546875, 18.38127899169922, -49.3359489440918, 147.4694061279297, 14.378097534179688, 5.410369873046875, 118.86575317382812, 105.53495025634766, 135.50685119628906, 46.28492736816406, 106.28128051757812, 193.3558349609375, 182.90750122070312, 54.300323486328125, 101.10562133789062, 18.274822235107422, -3.2029361724853516, -46.62705993652344, 81.3412094116211, 14.595123291015625, 18.99134063720703, 53.920135498046875, 34.5159912109375, 31.47186279296875, 108.75886535644531, 125.3243408203125, -138.11959838867188, 155.68588256835938, -14.03887939453125, 6.0401153564453125, 49.108489990234375, -154.36822509765625, 3.5552825927734375, 43.67326354980469, 141.42333984375, 203.7113037109375, 103.71546936035156, 26.7032470703125, 59.470672607421875, 177.331787109375, 141.45587158203125, 59.659088134765625, 187.9207763671875, 94.05368041992188, -70.526123046875, 158.0912322998047, 9.410186767578125, -12.752288818359375, -96.00703430175781, 42.28157043457031, 58.113189697265625, -138.95751953125, 8.098905563354492, 128.80111694335938, 108.17510986328125, 3.3984146118164062, -109.35577392578125, -13.28955078125, 117.36538696289062, -9.690086364746094, 37.34508514404297, 16.078765869140625, 83.47085571289062, 177.39471435546875, 148.83721923828125, 110.76062774658203, 29.93670654296875, 31.633758544921875, 143.68017578125, 17.5120849609375, 169.94879150390625, -97.47306823730469, 9.520294189453125, 177.47723388671875, 48.35505676269531, 26.613800048828125, 56.504364013671875, 133.282470703125, 105.78802490234375, 30.91876220703125, 138.2225799560547, 211.0478515625, 119.92935180664062, 156.19436645507812, -93.08615112304688, 67.63153076171875, 40.74562454223633, -150.6099853515625, -110.76801300048828, -82.12338256835938, 15.432693481445312, -132.67437744140625, 46.423431396484375, -145.16038513183594, -73.93034362792969, 13.058013916015625, 176.15841674804688, 88.69535827636719, 47.21783447265625, 8.880645751953125, 21.092147827148438, 165.05471801757812, -77.83377075195312, -9.049591064453125, 34.559288024902344, -22.44189453125, -11.36151123046875, 160.03802490234375, -57.95390319824219, 47.621978759765625, 22.373756408691406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000350.npy"} {"epoch": 0.7329842931937173, "step": 351, "batch_size": 128, "mean": 34.47273635864258, "std": 84.98603820800781, "min": -205.5682373046875, "p10": -49.05882835388183, "median": 15.93316650390625, "p90": 159.26180419921874, "max": 225.4014892578125, "pos_frac": 0.65625, "sample": [0.0, 140.595947265625, 13.0357666015625, 193.34539794921875, -0.6541290283203125, 179.28021240234375, 12.020744323730469, -54.00365447998047, 59.07611083984375, -5.273406982421875, -0.4482421875, -75.73141479492188, 15.66558837890625, 40.33567810058594, 158.55624389648438, -121.15126037597656, 25.546024322509766, 108.62250518798828, -33.56659698486328, 18.237548828125, 26.158226013183594, 135.73802185058594, -18.936508178710938, 198.23760986328125, 63.62385559082031, 155.90550231933594, 6.112295150756836, -17.438812255859375, -5.2679443359375, 0.0717315673828125, 169.02783203125, 177.6653594970703, 29.983795166015625, 43.14333724975586, -5.4224700927734375, -59.79302978515625, -2.812479019165039, 135.44744873046875, -36.024559020996094, -46.93961715698242, -97.14666748046875, 131.36117553710938, 12.918228149414062, 47.63372802734375, -61.47149658203125, 131.8893280029297, 8.410850524902344, 16.20074462890625, -22.75860595703125, 16.462615966796875, -36.221221923828125, -27.962493896484375, 17.01873779296875, -4.8687896728515625, 16.8988037109375, -0.675506591796875, 154.96969604492188, -36.2178955078125, -12.261322021484375, 16.919189453125, 164.60665893554688, 222.10110473632812, 23.33991050720215, 194.015380859375, 152.7439422607422, -16.401580810546875, 31.4862060546875, 15.365264892578125, 123.37542724609375, -84.78010559082031, -3.0858306884765625, 117.95584869384766, -27.32293701171875, 14.17633056640625, 160.90811157226562, -174.72958374023438, 29.653900146484375, 1.443603515625, 8.915252685546875, 25.61932373046875, 0.0, -0.188720703125, 23.175552368164062, 129.14749145507812, 140.5443572998047, -0.7652111053466797, 1.4781341552734375, 17.385353088378906, 123.4976806640625, 1.122100830078125, 32.522064208984375, 7.6223297119140625, 39.18304443359375, -14.131114959716797, 0.0, 172.898681640625, 168.2174072265625, 19.838134765625, 25.14896011352539, 1.387847900390625, 8.058456420898438, -205.5682373046875, 118.694580078125, 157.36874389648438, 45.97357177734375, 5.2397613525390625, 225.4014892578125, 32.239501953125, 153.2101287841797, -2.1746902465820312, 0.3462200164794922, 5.78912353515625, -13.04571533203125, 49.571502685546875, 8.563045501708984, 146.41293334960938, 20.97107696533203, -45.727813720703125, -93.30587768554688, -149.75296020507812, 39.40776062011719, 83.61648559570312, -173.79595947265625, 83.316162109375, -41.368682861328125, -59.04717254638672, 212.7962646484375, 38.782318115234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000351.npy"} {"epoch": 0.7350785340314137, "step": 352, "batch_size": 128, "mean": 56.02505874633789, "std": 82.20909118652344, "min": -155.1124267578125, "p10": -39.93561706542968, "median": 43.48783874511719, "p90": 166.16051635742187, "max": 201.97900390625, "pos_frac": 0.734375, "sample": [116.42318725585938, 19.28662109375, 162.66448974609375, -42.00970458984375, 10.044448852539062, 186.94482421875, 113.16039276123047, 98.15045166015625, 114.97654724121094, 2.7660140991210938, -56.08935546875, 130.0078125, -139.81927490234375, 14.355224609375, 143.0846710205078, 4.707118988037109, 32.667755126953125, 174.53793334960938, 141.36550903320312, 112.19024658203125, 194.66522216796875, -29.159059524536133, -39.046722412109375, 176.95046997070312, 34.2952880859375, -155.1124267578125, -17.67193603515625, 36.51893615722656, 17.33758544921875, 118.95487213134766, -0.8457851409912109, 58.9549560546875, 171.912353515625, 15.661224365234375, 101.77935791015625, -105.08724212646484, 11.037704467773438, 91.4412841796875, -18.627349853515625, 14.187652587890625, 68.15509033203125, -130.3253173828125, -28.061920166015625, 84.17794799804688, -28.2725830078125, 24.159591674804688, 0.0, 107.36215209960938, 18.625457763671875, 168.43177795410156, -1.50714111328125, 181.3455810546875, 122.45718383789062, 31.04302978515625, 146.42050170898438, 70.8529052734375, 20.914703369140625, 0.0, 65.75033569335938, 94.61834716796875, 109.84550476074219, -0.8250579833984375, 17.109962463378906, 94.980712890625, -102.99691772460938, -19.392913818359375, 10.244697570800781, 76.73924255371094, -96.87664794921875, 6.986297607421875, 108.83558654785156, 189.32388305664062, 70.4739990234375, 18.031227111816406, 8.130393981933594, -50.311676025390625, 173.126220703125, 4.616146087646484, -81.47225952148438, -86.11602783203125, 201.97900390625, 169.7781524658203, 165.18711853027344, 24.172592163085938, -3.138641357421875, 141.1006317138672, 12.010101318359375, -42.7347412109375, -0.25087738037109375, 102.77590942382812, 46.120574951171875, -10.430343627929688, 55.96478271484375, 21.113250732421875, 40.8551025390625, 46.82275390625, 140.949462890625, 129.98587036132812, -8.70892333984375, 178.01231384277344, 50.31956481933594, 105.84349822998047, 137.18162536621094, 164.13369750976562, 150.32081604003906, -30.94525146484375, 142.4342041015625, 151.86553955078125, 97.89871215820312, -3.28411865234375, 24.77569580078125, 126.58418273925781, 162.2052001953125, 1.3079833984375, -31.471588134765625, -18.04998779296875, 29.788623809814453, 163.94827270507812, 130.53131103515625, 174.19078063964844, -25.871353149414062, -61.454620361328125, 133.7269287109375, 144.85507202148438, 83.7694320678711, 113.1471939086914, 22.950782775878906, 134.78005981445312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000352.npy"} {"epoch": 0.7371727748691099, "step": 353, "batch_size": 128, "mean": 55.81460952758789, "std": 83.35114288330078, "min": -160.74636840820312, "p10": -27.90453491210937, "median": 48.038307189941406, "p90": 152.99041290283202, "max": 235.80096435546875, "pos_frac": 0.7578125, "sample": [35.66509246826172, 137.1219482421875, 66.19410705566406, -54.663360595703125, 98.17312622070312, 164.38412475585938, -114.97797393798828, 150.92816162109375, 134.98318481445312, 13.848785400390625, 67.23609924316406, 23.969314575195312, 8.56817626953125, 62.20306396484375, -6.91339111328125, 15.745574951171875, -17.296112060546875, -26.05712890625, 43.164764404296875, -18.15985107421875, 77.41398620605469, 101.03404235839844, -1.5081863403320312, -160.74636840820312, 137.22976684570312, -21.537841796875, 80.88336181640625, 109.77957153320312, 218.793212890625, -1.52655029296875, 168.91058349609375, 134.92105102539062, 151.26205444335938, -63.58782958984375, 151.88775634765625, -139.392333984375, 181.57061767578125, 87.063232421875, 83.7071533203125, 235.80096435546875, 50.25050354003906, -98.17262268066406, 39.71626281738281, 153.12850952148438, 35.311920166015625, -3.71014404296875, 151.28289794921875, 225.27114868164062, 200.17678833007812, 127.17729949951172, 136.01368713378906, 81.1817398071289, -91.6957778930664, 114.9833984375, 16.0712890625, 206.09860229492188, 3.9312572479248047, -15.317115783691406, 124.08342742919922, 92.28324890136719, 178.9552001953125, 81.660888671875, -33.657379150390625, -106.07806396484375, 1.4279632568359375, 110.23309326171875, -123.00609588623047, -16.75665283203125, 73.72918701171875, 27.105331420898438, 52.501129150390625, 56.71794128417969, 134.7777862548828, 146.16567993164062, 142.1151885986328, 124.17446899414062, 14.836677551269531, 138.47027587890625, 76.6197509765625, 10.041839599609375, 138.28192138671875, 18.487308502197266, 5.955230712890625, 26.8253173828125, 45.82611083984375, 155.32028198242188, 33.73420715332031, 128.4834747314453, 119.08621215820312, 3.9990310668945312, -4.411767959594727, -19.548980712890625, 130.66172790527344, 114.273681640625, 30.635887145996094, -22.373931884765625, -3.7591896057128906, 19.686019897460938, 11.588493347167969, -13.586761474609375, 70.7672119140625, 12.183563232421875, 27.43865966796875, 4.644613265991211, 157.6004180908203, 40.1744384765625, 85.69000244140625, 54.46484375, 145.62611389160156, -103.91555786132812, 27.17601776123047, 0.650146484375, 27.727970123291016, 62.896484375, 6.476726531982422, 0.0, 197.09567260742188, 131.1884765625, -32.21514892578125, -13.11993408203125, 4.56640625, -147.93484497070312, -13.1544189453125, 18.76971435546875, 152.9312286376953, 115.72222900390625, 138.51089477539062, 98.96661376953125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000353.npy"} {"epoch": 0.7392670157068063, "step": 354, "batch_size": 128, "mean": 35.11090850830078, "std": 89.02236938476562, "min": -167.091552734375, "p10": -93.89601898193358, "median": 17.372653007507324, "p90": 149.67210388183594, "max": 221.67990112304688, "pos_frac": 0.65625, "sample": [8.963638305664062, -48.56697082519531, -131.2511444091797, 147.8109588623047, 47.716827392578125, 201.03302001953125, 86.1441650390625, 8.681846618652344, 9.902851104736328, 3.847553253173828, 62.73419189453125, 38.8216552734375, -90.65863037109375, 16.325469970703125, 149.69528198242188, -12.044052124023438, -15.200912475585938, -0.6565017700195312, 6.35406494140625, -4.012786865234375, 115.21571350097656, 134.3363800048828, 93.65670776367188, -109.21652221679688, 27.179588317871094, 21.375839233398438, -10.681377410888672, 3.6975860595703125, 140.8158721923828, 196.657470703125, -91.74134826660156, -3.17059326171875, 0.3626518249511719, 57.6319580078125, 174.80044555664062, 28.327301025390625, 13.374393463134766, -0.486114501953125, 146.13626098632812, 140.33944702148438, 4.1468658447265625, -55.40142822265625, 117.51591491699219, -19.1336669921875, 145.00978088378906, 121.61991119384766, -10.229736328125, 8.191600799560547, -153.7850341796875, 4.2616729736328125, 14.08050537109375, 197.8800048828125, 112.26605224609375, 171.60317993164062, 176.44851684570312, 173.5357666015625, 126.96634674072266, 61.868408203125, 37.429931640625, 76.351318359375, 78.90464782714844, -21.777114868164062, -17.966148376464844, -98.923583984375, 4.75592041015625, 136.22042846679688, 10.307090759277344, 221.67990112304688, -2.3508148193359375, -4.65411376953125, 149.66217041015625, -75.75650024414062, 124.77365112304688, 6.188108444213867, 193.865966796875, 46.69679260253906, 113.18709564208984, -119.46803283691406, -4.2213134765625, 0.8905792236328125, -32.18902587890625, -73.75582885742188, 167.380859375, 39.454833984375, -19.963394165039062, 78.60244750976562, 32.0076904296875, 141.28627014160156, -29.604385375976562, -74.54702758789062, 21.724853515625, -7.102851867675781, 148.243408203125, 7.6927032470703125, 126.05946350097656, 26.036087036132812, -8.681747436523438, -22.150909423828125, -118.17153930664062, -129.97811889648438, 57.87054443359375, 2.892181396484375, -16.44733428955078, 124.38688659667969, 125.94686889648438, 79.45831298828125, 3.78076171875, -6.4044189453125, 21.222503662109375, 183.0599365234375, 142.23504638671875, 86.2740707397461, -25.945960998535156, 153.23162841796875, -133.09207153320312, 72.34393310546875, -167.091552734375, 122.57878112792969, -112.64141845703125, -114.28585815429688, 128.05902099609375, 18.419836044311523, -116.15109252929688, 23.856491088867188, -31.73193359375, -99.11585998535156, 63.76197814941406, 18.488250732421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000354.npy"} {"epoch": 0.7413612565445026, "step": 355, "batch_size": 128, "mean": 53.88383865356445, "std": 83.59260559082031, "min": -199.53570556640625, "p10": -33.33631286621093, "median": 38.00001525878906, "p90": 160.43460083007812, "max": 205.78195190429688, "pos_frac": 0.765625, "sample": [-0.5622596740722656, -199.53570556640625, 154.57769775390625, 96.1087646484375, 13.448577880859375, 11.199260711669922, 68.2930908203125, 5.4345703125, 83.95794677734375, 39.12200927734375, 158.0476837158203, 8.792137145996094, -9.893928527832031, 128.465576171875, 3.3266525268554688, 108.36756896972656, 175.30157470703125, 148.187744140625, -5.349712371826172, 1.55084228515625, -79.47042846679688, 17.21295166015625, 5.1101531982421875, 178.49951171875, 12.2647705078125, 159.5894317626953, -13.759056091308594, -9.34942626953125, 39.45460510253906, 31.048065185546875, 26.291534423828125, 26.64288330078125, 51.0260009765625, 183.478759765625, 176.76953125, 50.27698516845703, 184.0755615234375, 26.565505981445312, -1.130126953125, -48.953369140625, 108.50753021240234, 147.9149627685547, 54.303131103515625, 66.69627380371094, -57.41474914550781, 25.2833251953125, -6.7637939453125, 71.90956115722656, 18.00896453857422, 2.00384521484375, 147.45626831054688, 43.9332275390625, 127.79071044921875, 12.374557495117188, 0.1011962890625, 85.79058837890625, 2.485553741455078, 47.47658157348633, 34.66351318359375, -25.425968170166016, 169.93560791015625, 10.350761413574219, -32.40191650390625, -10.113296508789062, 120.0009536743164, -127.54214477539062, 24.729583740234375, 205.78195190429688, 133.1175537109375, 143.82960510253906, -0.3043861389160156, 168.02096557617188, -156.5775146484375, 150.45486450195312, 0.2236480712890625, 37.5257568359375, 139.49392700195312, 84.74298095703125, 185.5206298828125, 79.80735778808594, 101.94033813476562, 147.73464965820312, -0.347137451171875, 14.209991455078125, 33.44781494140625, -55.84869384765625, 36.74790954589844, -61.138885498046875, 32.92012023925781, 33.19622802734375, 138.00450134277344, -35.516571044921875, -144.67935180664062, 5.499969482421875, -8.64019775390625, 42.488739013671875, 18.330429077148438, 162.4066619873047, -20.7962646484375, 153.21197509765625, 38.474273681640625, -121.26690673828125, 197.099609375, 114.92991638183594, 177.42526245117188, 153.73907470703125, 156.63800048828125, 93.38531494140625, -88.24812316894531, -12.23828125, 13.572479248046875, 36.47099304199219, -8.094711303710938, 146.32601928710938, 129.68954467773438, 135.25704956054688, 146.65072631835938, 127.60088348388672, 131.9808349609375, 117.71565246582031, 48.59180450439453, -3.2486572265625, 44.107696533203125, 168.593017578125, 59.315185546875, -132.03582763671875, 158.7837371826172, 34.568817138671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000355.npy"} {"epoch": 0.743455497382199, "step": 356, "batch_size": 128, "mean": 45.695064544677734, "std": 74.98320770263672, "min": -171.26202392578125, "p10": -35.41514167785645, "median": 33.004241943359375, "p90": 144.2539520263672, "max": 181.7672882080078, "pos_frac": 0.7109375, "sample": [1.435150146484375, -0.7519073486328125, 143.8280029296875, -3.712615966796875, 113.16995239257812, -69.8814697265625, 18.037109375, -40.914154052734375, 3.049154281616211, 17.482223510742188, -9.276756286621094, 10.7952880859375, 27.23388671875, 130.30349731445312, -8.116086959838867, 2.633485794067383, 38.15099334716797, -20.2269287109375, 1.94635009765625, 60.359466552734375, -59.656280517578125, 139.45626831054688, 38.69371032714844, 35.46832275390625, -0.23285388946533203, 171.72308349609375, -113.37208557128906, 135.2324981689453, 2.812744140625, 89.6519775390625, 36.65838623046875, 156.9075927734375, 10.31475830078125, 103.1251449584961, 141.2206573486328, 153.0014190673828, 73.94845581054688, 50.428741455078125, 168.65797424316406, 124.49339294433594, -93.09248352050781, 138.13424682617188, -0.05120849609375, 8.703254699707031, 23.85382080078125, 31.845703125, -8.786430358886719, 173.3685302734375, 85.16323852539062, -5.735603332519531, 58.218711853027344, 107.67276000976562, 149.58448791503906, 167.03570556640625, 0.0008563995361328125, 37.88751220703125, -86.61128234863281, 136.26025390625, 116.22035217285156, 42.739532470703125, 130.14096069335938, -133.01596069335938, 131.61329650878906, 86.45718383789062, 137.28761291503906, 137.77883911132812, 124.4361572265625, -10.768905639648438, 2.0056838989257812, 4.5467681884765625, -8.826065063476562, 22.779205322265625, 110.83497619628906, 12.814109802246094, 1.693603515625, 22.10181427001953, 109.00216674804688, -0.37981414794921875, 158.2762908935547, -62.6705322265625, 40.464630126953125, -34.85734176635742, 176.89608764648438, 34.512176513671875, -32.151214599609375, 154.11477661132812, 132.02798461914062, 111.55714416503906, 28.086750030517578, -1.5752220153808594, -79.39196014404297, -3.5425071716308594, -11.773162841796875, -171.26202392578125, 45.29133605957031, 13.364959716796875, 45.76898193359375, 0.0, 79.8797607421875, 37.535430908203125, 111.80874633789062, 34.16278076171875, -7.167947769165039, -20.940582275390625, 0.42957496643066406, 16.085906982421875, 25.615772247314453, 125.16343688964844, -2.7523193359375, -87.23733520507812, 145.24783325195312, 181.7672882080078, 130.31692504882812, 84.8363037109375, -23.516456604003906, 153.79171752929688, 133.75042724609375, 10.16497802734375, 44.190093994140625, 135.26467895507812, -36.7166748046875, 84.90882873535156, -0.86492919921875, -81.65877532958984, 52.16960144042969, -11.991466522216797, 21.494796752929688, 133.100341796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000356.npy"} {"epoch": 0.7455497382198953, "step": 357, "batch_size": 128, "mean": 49.07231140136719, "std": 84.07088470458984, "min": -174.99972534179688, "p10": -38.08316879272461, "median": 26.1010684967041, "p90": 166.63986816406248, "max": 241.31488037109375, "pos_frac": 0.71875, "sample": [31.811309814453125, 147.692138671875, -14.276277542114258, 0.17474365234375, 153.3536376953125, 40.399078369140625, 0.0, -8.99542236328125, 98.44036865234375, 5.0349273681640625, 144.90036010742188, -138.2232666015625, 3.371969223022461, 241.31488037109375, 194.73077392578125, 203.4959716796875, 8.961589813232422, -3.11077880859375, 96.76719665527344, 25.80816650390625, 15.075363159179688, 6.26690673828125, 75.51502990722656, 149.385498046875, 157.42019653320312, -90.7294921875, 219.44100952148438, 21.869644165039062, 26.393970489501953, 175.0191650390625, -44.7623291015625, 13.09259033203125, -20.846405029296875, 6.44969367980957, 203.03536987304688, 34.55485534667969, -17.681243896484375, -37.84076690673828, 6.81671142578125, 102.32735443115234, 115.5399398803711, 197.1876220703125, 44.5404052734375, 111.46743774414062, 70.85932922363281, -38.648773193359375, 4.7828369140625, -69.05191040039062, 162.88687133789062, 1.99127197265625, -31.343124389648438, 30.150856018066406, 35.44123840332031, 9.67572021484375, 110.98472595214844, 118.24238586425781, 179.02362060546875, 165.1912841796875, 34.55352783203125, -1.4771728515625, 142.062255859375, 61.07275390625, 22.339508056640625, 15.9400634765625, 47.97509765625, -7.7305908203125, 144.54356384277344, 145.90859985351562, 53.48260498046875, 39.026611328125, -11.812469482421875, -92.23626708984375, -37.635528564453125, 5.4419708251953125, 2.0269012451171875, 139.0404510498047, 57.76704406738281, -59.411895751953125, 20.852020263671875, -35.88592529296875, -56.1192626953125, 8.655181884765625, 9.553115844726562, 26.432952880859375, 179.66046142578125, 40.634918212890625, 154.16305541992188, -35.14630126953125, -46.67649841308594, -0.495391845703125, -7.102638244628906, 16.09625244140625, 42.31439208984375, -70.78677368164062, 74.16128540039062, 139.58056640625, 0.35894775390625, -104.76313781738281, 0.0, 170.0198974609375, -134.0571746826172, 0.0, 52.317657470703125, -6.874298095703125, 163.08843994140625, 8.458389282226562, 14.673545837402344, 137.71461486816406, 157.07025146484375, 139.64181518554688, 176.85955810546875, -174.99972534179688, 155.33694458007812, 50.504425048828125, 2.376047134399414, 69.41046142578125, -3.77532958984375, 120.79838562011719, 98.29886627197266, 187.27301025390625, 186.05712890625, 17.261608123779297, -17.91253662109375, -24.816024780273438, 114.83601379394531, -1.0579757690429688, 2.181976318359375, 112.82992553710938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000357.npy"} {"epoch": 0.7476439790575916, "step": 358, "batch_size": 128, "mean": 55.55646896362305, "std": 88.00504302978516, "min": -174.73648071289062, "p10": -52.154116821289065, "median": 34.635162353515625, "p90": 168.16890563964844, "max": 206.4168701171875, "pos_frac": 0.75, "sample": [1.1314697265625, 1.171844482421875, 100.78605651855469, -71.62490844726562, 7.1352386474609375, 39.80181884765625, -32.56317138671875, 116.20169067382812, 179.92315673828125, 206.4168701171875, 8.488204956054688, 4.040607452392578, 26.051040649414062, 151.7895965576172, 169.3414306640625, 153.4365997314453, 24.626754760742188, 176.55648803710938, 169.1220703125, 33.02716064453125, -115.99351501464844, 138.90921020507812, -52.20361328125, -119.41592407226562, 104.44949340820312, 14.448905944824219, 126.24543762207031, -0.96087646484375, 172.14862060546875, -11.774169921875, 9.343460083007812, 111.54402160644531, 42.20213317871094, 128.9728240966797, 90.4390869140625, -8.24847412109375, 197.05450439453125, -69.18588256835938, 34.6842041015625, 127.06452941894531, -114.20573425292969, 159.64962768554688, 157.13670349121094, 169.57672119140625, 28.317344665527344, 153.0111083984375, 122.0058364868164, 150.58470153808594, 153.58447265625, -0.015380859375, -30.81060791015625, -8.621726989746094, -7.1867523193359375, 23.86528778076172, 149.30857849121094, -12.63299560546875, 157.99606323242188, 158.8980712890625, -54.24322509765625, 201.05511474609375, -52.132904052734375, -174.73648071289062, 171.20135498046875, -17.00342559814453, 7.5389404296875, 153.63214111328125, 76.254150390625, 30.75778579711914, 73.3419189453125, 10.158843994140625, 58.063331604003906, 13.252227783203125, 19.529884338378906, 21.958786010742188, 3.233673095703125, -128.28158569335938, 145.1786651611328, 112.32339477539062, -13.590660095214844, 180.2283477783203, 61.90026092529297, -55.8778076171875, 126.0556640625, 50.987457275390625, 167.76040649414062, 34.58612060546875, 129.3743896484375, 27.73297119140625, 41.114070892333984, 164.7158203125, 10.903610229492188, 182.78878784179688, 141.38278198242188, 107.22093200683594, 109.127685546875, 134.347412109375, -29.38470458984375, 9.61505126953125, 7.41387939453125, 107.58882141113281, 188.1519775390625, 23.1407470703125, 124.94381713867188, 167.66390991210938, 67.95635986328125, 164.33041381835938, 154.25570678710938, -158.91363525390625, -15.985027313232422, -120.20339965820312, -114.24072265625, 6.734600067138672, 133.7235565185547, 6.0603790283203125, 25.69464874267578, -33.256858825683594, 7.352485656738281, 12.522014617919922, -1.7385101318359375, 29.214569091796875, 0.0, 79.10353088378906, 59.490997314453125, 46.987060546875, 143.8243408203125, -49.33128356933594, 32.76873779296875, -31.14215087890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000358.npy"} {"epoch": 0.749738219895288, "step": 359, "batch_size": 128, "mean": 52.89460372924805, "std": 81.01226043701172, "min": -195.4248046875, "p10": -29.335751342773435, "median": 33.13288116455078, "p90": 160.1195861816406, "max": 216.8472900390625, "pos_frac": 0.734375, "sample": [51.585670471191406, -55.77862548828125, 135.9199981689453, 176.70736694335938, 106.96713256835938, -74.34890747070312, 91.1864242553711, 15.421730041503906, 35.958740234375, 49.359588623046875, 27.81476593017578, 103.83480834960938, 20.655635833740234, 7.845123291015625, -13.16204833984375, -28.231231689453125, 112.20661163330078, -1.61724853515625, 21.24505615234375, 14.760345458984375, 117.18502044677734, -15.26904296875, 143.90518188476562, 169.87884521484375, -40.93367004394531, 141.30279541015625, 6.83831787109375, 35.00543212890625, 94.22174072265625, -183.09393310546875, -0.31884765625, 199.13180541992188, 4.791961669921875, 65.43749237060547, 2.7650070190429688, 68.21722412109375, -9.229318618774414, 140.01132202148438, 159.00958251953125, 167.06622314453125, -25.138641357421875, -118.53555297851562, 135.2537841796875, 3.5622482299804688, 136.70599365234375, 100.56378173828125, 8.20526123046875, 77.62091064453125, 131.17022705078125, 123.88363647460938, 128.20095825195312, 1.09869384765625, 212.47503662109375, -50.44964599609375, 135.6082763671875, -0.9278335571289062, 56.653350830078125, -159.13031005859375, 114.3414535522461, 42.2735595703125, 190.92669677734375, 116.76809692382812, -195.4248046875, 112.54379272460938, -45.7764892578125, -9.819625854492188, -12.95819091796875, 51.8316650390625, 126.56716918945312, 91.64666748046875, -4.901397705078125, 22.58782958984375, -32.75456237792969, 35.8787841796875, 96.04813385009766, 110.61236572265625, -8.16827392578125, 123.202880859375, 13.322677612304688, 139.73336791992188, 11.105499267578125, -10.7257080078125, 6.25750732421875, 119.744384765625, 24.478927612304688, 14.818004608154297, 18.263031005859375, 146.75900268554688, 33.00482177734375, 17.132827758789062, 164.51820373535156, 13.634841918945312, 168.1441650390625, 128.77249145507812, 22.586395263671875, -28.066162109375, -57.808631896972656, 127.830322265625, -22.786415100097656, -0.061603546142578125, 28.64910888671875, 173.10928344726562, 24.62750244140625, -73.00090026855469, 16.365737915039062, -27.1871337890625, 165.80606079101562, -19.5904541015625, 53.418609619140625, 154.88662719726562, 132.80291748046875, 115.17252349853516, 9.813720703125, 26.303138732910156, -8.649993896484375, 172.78390502929688, 158.5693359375, 88.71014404296875, 98.14651489257812, -1.5762405395507812, -6.853176116943359, 216.8472900390625, 28.3697509765625, 19.703779220581055, 162.7095947265625, 33.26094055175781, 122.06570434570312, -31.9129638671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000359.npy"} {"epoch": 0.7518324607329843, "step": 360, "batch_size": 128, "mean": 46.10019302368164, "std": 75.07502746582031, "min": -205.0806884765625, "p10": -19.46338653564453, "median": 24.29962921142578, "p90": 155.66123809814454, "max": 215.11529541015625, "pos_frac": 0.7109375, "sample": [83.42221069335938, 121.11946105957031, -45.98594665527344, 157.99635314941406, -135.89044189453125, -8.085309982299805, -100.14129638671875, 67.289794921875, 215.11529541015625, -46.34913635253906, 31.9931640625, 77.96464538574219, -20.157821655273438, 138.8614501953125, 48.23577880859375, 100.90959167480469, 14.762847900390625, 140.8223876953125, -16.08892822265625, 48.10450744628906, 169.9676513671875, 23.884536743164062, 17.38763427734375, 108.13044738769531, 125.03801727294922, -10.089553833007812, 3.299407958984375, 19.90032958984375, -14.69036865234375, 160.18341064453125, 35.107330322265625, 177.68601989746094, 82.12678527832031, 17.23114013671875, 78.38323974609375, 7.286956787109375, -20.397933959960938, 133.52841186523438, -31.9765625, 112.12944030761719, 96.01495361328125, 87.302734375, 139.76046752929688, 28.51507568359375, 38.239501953125, 21.304340362548828, 17.693878173828125, 137.22296142578125, -7.18609619140625, 40.3701171875, -12.461273193359375, 147.2769775390625, 117.84878540039062, 94.88121032714844, 189.70462036132812, -3.8126220703125, 168.33753967285156, -37.7022705078125, 146.23260498046875, 12.103828430175781, 48.305023193359375, 39.0201416015625, 150.14129638671875, 49.80974578857422, 19.187164306640625, -6.4888458251953125, 168.42587280273438, 18.706771850585938, 59.55220031738281, -136.86871337890625, -2.25921630859375, 114.20062255859375, 1.4037857055664062, 156.83738708496094, -18.2481689453125, 20.172252655029297, 79.94268798828125, 166.7958984375, 142.40878295898438, 135.2913818359375, -9.115753173828125, 116.69525146484375, 6.689977645874023, -1.7882575988769531, -205.0806884765625, 118.98779296875, 42.911865234375, 9.290863037109375, -19.165771484375, 6.4358673095703125, 147.06378173828125, -35.216522216796875, 179.57412719726562, 48.77948760986328, -0.044281005859375, -4.20281982421875, 5.03912353515625, -138.3914031982422, 8.92779541015625, -0.5141639709472656, 46.99168395996094, 0.0, 24.7147216796875, 17.967559814453125, 155.45298767089844, 4.677644729614258, 65.634765625, 43.7254638671875, -1.335601806640625, 8.211832046508789, 10.468658447265625, -4.50433349609375, -7.586311340332031, 121.98648071289062, 13.542655944824219, 59.365997314453125, -28.058998107910156, -18.87322998046875, 0.0, 10.057319641113281, 46.40972900390625, 23.124008178710938, 37.27006530761719, -18.550559997558594, -14.798904418945312, 19.60003662109375, 156.30926513671875, 156.14715576171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000360.npy"} {"epoch": 0.7539267015706806, "step": 361, "batch_size": 128, "mean": 61.99161911010742, "std": 86.16715240478516, "min": -176.82260131835938, "p10": -33.2955352783203, "median": 58.918006896972656, "p90": 166.38409271240232, "max": 234.425537109375, "pos_frac": 0.734375, "sample": [70.8482437133789, -8.5533447265625, -121.33987426757812, 30.900390625, 29.847259521484375, -7.30413818359375, 0.7500419616699219, -27.802520751953125, 149.16378784179688, 42.2996826171875, 132.0506591796875, 15.058868408203125, 1.9391441345214844, 202.98391723632812, -3.5231990814208984, 157.5814666748047, 0.29608917236328125, 15.026023864746094, -30.500457763671875, 1.9577713012695312, 150.28607177734375, 181.87130737304688, -70.03439331054688, 150.90814208984375, -0.4983978271484375, 182.37832641601562, 126.96714782714844, 137.75067138671875, 47.73155975341797, 186.43411254882812, 9.99908447265625, -5.968040466308594, 112.1551284790039, 81.18011474609375, 14.008209228515625, 142.1212158203125, -126.1319580078125, 85.57666015625, 124.64109802246094, 64.0294189453125, 183.72012329101562, 190.68093872070312, 23.869979858398438, 93.87232971191406, 198.40771484375, 26.187286376953125, -112.99066162109375, -12.831161499023438, 47.85511779785156, -14.82253646850586, -1.169973373413086, 39.38905334472656, 108.28593444824219, 0.0, 35.70426940917969, 13.447662353515625, 144.85617065429688, 73.88046264648438, 3.7420120239257812, -19.1019287109375, 37.29203796386719, 138.32545471191406, -3.129404067993164, 164.58380126953125, 127.26055908203125, -50.48321533203125, 50.92193603515625, 95.66497039794922, 28.08526611328125, -62.28497314453125, 80.4901123046875, -4.203338623046875, -98.30654907226562, -10.062744140625, 90.60790252685547, 0.0, 109.0028076171875, -39.8173828125, 59.91328430175781, -8.406219482421875, 202.5513916015625, 152.84170532226562, 172.05908203125, 76.35708618164062, -176.82260131835938, 140.07948303222656, 6.778564453125, -44.9903564453125, 234.425537109375, 137.9676513671875, -91.74697875976562, 67.93670654296875, -127.4599609375, 156.56558227539062, 170.04730224609375, 110.75637817382812, 69.64469909667969, 33.01194763183594, 152.90682983398438, 128.54144287109375, 160.78460693359375, -11.598501205444336, 203.3540802001953, 102.00732421875, 157.87106323242188, 149.69317626953125, 92.33399963378906, 51.72120666503906, 19.316986083984375, 141.85504150390625, 160.772216796875, 57.9227294921875, -14.8770751953125, 88.40806579589844, 169.654052734375, 144.9156494140625, -4.398895263671875, 152.32925415039062, 164.98268127441406, 116.26605224609375, 9.281631469726562, 128.8332061767578, -5.3219146728515625, 74.10713195800781, 30.351219177246094, 27.18474578857422, 150.15838623046875, -129.95474243164062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000361.npy"} {"epoch": 0.7560209424083769, "step": 362, "batch_size": 128, "mean": 52.66845703125, "std": 81.466064453125, "min": -174.92172241210938, "p10": -37.129893875122065, "median": 34.804664611816406, "p90": 150.811181640625, "max": 245.94644165039062, "pos_frac": 0.7578125, "sample": [18.67963409423828, 90.6619873046875, -5.226978302001953, 44.27781677246094, 143.56765747070312, 24.419479370117188, 135.74752807617188, -12.001739501953125, 58.864990234375, 13.759429931640625, 93.5928955078125, -174.92172241210938, -134.5074005126953, -104.02110290527344, 200.34994506835938, 7.798944473266602, 168.35755920410156, -22.35321044921875, 56.893150329589844, 120.94873046875, 94.73526000976562, 141.58343505859375, 245.94644165039062, 48.87127685546875, 93.14083862304688, 158.51852416992188, 121.57754516601562, -17.181304931640625, -157.18557739257812, 86.55401611328125, -25.543304443359375, -43.396156311035156, 115.6201171875, -6.850982666015625, -80.9083480834961, 131.91766357421875, 31.26068878173828, -12.331966400146484, 153.87210083007812, 14.704429626464844, 200.27099609375, 64.89190673828125, 180.98333740234375, 142.59950256347656, 146.918212890625, 143.84872436523438, 133.04376220703125, -1.34649658203125, -13.33721923828125, 13.849555969238281, -34.3822021484375, 132.32968139648438, 75.57342529296875, -35.83306884765625, 2.0572433471679688, 4.349639892578125, 22.85528564453125, 175.849853515625, -2.76788330078125, 143.8394012451172, 125.21044921875, -20.85943603515625, 31.35137939453125, 111.88153076171875, 111.31468200683594, 117.07000732421875, 38.621376037597656, 18.881439208984375, 114.79017639160156, 30.35186767578125, 170.523193359375, 103.62722778320312, 67.09741973876953, 24.560089111328125, -4.866050720214844, 190.22442626953125, 107.45262145996094, 4.374534606933594, 100.99871826171875, 130.03677368164062, -68.99211120605469, 135.2308349609375, -40.155818939208984, 5.6256256103515625, -98.78817749023438, -58.217041015625, -12.448293685913086, 176.21546936035156, 9.629619598388672, 17.48247528076172, 24.521820068359375, 4.3197479248046875, -7.175102233886719, 14.979804992675781, 28.6220703125, 101.30455017089844, 123.93702697753906, 27.241592407226562, 11.14373779296875, 5.100189208984375, -21.724578857421875, 139.95742797851562, 139.77944946289062, 22.4622802734375, 3.2971038818359375, 54.088226318359375, -124.00177001953125, 16.00469970703125, 88.73614501953125, 20.542152404785156, 175.3356475830078, 15.774698257446289, 120.06439208984375, 96.3018798828125, 29.2418212890625, -10.745777130126953, 90.27294921875, 173.08447265625, 30.9395751953125, 39.05016326904297, -40.7529296875, 87.30400848388672, 129.4520721435547, 149.49935913085938, 23.3857421875, -112.35772705078125, 80.7066650390625, 38.25794982910156], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000362.npy"} {"epoch": 0.7581151832460733, "step": 363, "batch_size": 128, "mean": 59.92329406738281, "std": 84.64720153808594, "min": -192.93832397460938, "p10": -21.494572257995596, "median": 39.501773834228516, "p90": 169.97423400878907, "max": 265.234130859375, "pos_frac": 0.7890625, "sample": [145.09182739257812, -3.421600341796875, -13.6448974609375, 166.5721435546875, 144.3245849609375, 102.14300537109375, -1.0022869110107422, -119.70361328125, 26.64666748046875, 68.57563018798828, -14.572174072265625, 9.464302062988281, 150.10150146484375, 77.868896484375, 6.6381988525390625, 89.38375854492188, 102.8101806640625, 191.0004425048828, 3.65594482421875, -27.39214515686035, 20.24957275390625, 128.79867553710938, 183.46505737304688, 88.55825805664062, 21.427276611328125, 174.4947509765625, 187.2672576904297, 218.10357666015625, 12.375450134277344, 8.713623046875, 147.6947021484375, -0.0197296142578125, -17.99899673461914, -5.6971282958984375, 110.81375885009766, 1.6774120330810547, -3.5793075561523438, 99.52215576171875, 151.43740844726562, 1.6722412109375, 2.6385269165039062, 23.471847534179688, 90.38455200195312, 124.98497772216797, 67.13240051269531, 98.4434814453125, -10.521575927734375, 97.79168701171875, 172.36277770996094, 55.78044128417969, 137.69338989257812, 39.868934631347656, 265.234130859375, 154.78097534179688, -127.69961547851562, 102.51480102539062, 5.0225372314453125, 86.24057006835938, 33.5911865234375, 91.244384765625, 28.819732666015625, 35.98304748535156, 173.68772888183594, 130.6697998046875, 175.71511840820312, 0.5933914184570312, 177.20184326171875, 64.0439453125, 39.134613037109375, 5.379234313964844, 146.19171142578125, 3.028564453125, 175.2882080078125, 222.501953125, 8.651321411132812, 155.778564453125, -85.75332641601562, -11.516799926757812, 58.91950607299805, 60.112518310546875, 11.619815826416016, 16.140167236328125, -34.077789306640625, 21.6746826171875, 53.09712219238281, -14.825393676757812, -1.7305908203125, -5.97607421875, -94.638916015625, 13.997848510742188, 122.85931396484375, -133.2818145751953, 134.03408813476562, 2.4622421264648438, 28.437271118164062, 166.3304443359375, 141.92604064941406, -18.967041015625, 158.6365966796875, 102.02020263671875, 23.041534423828125, 106.9791259765625, -145.53929138183594, 130.1578369140625, 167.39974975585938, 142.08755493164062, 26.582778930664062, 58.79762268066406, 161.95936584472656, 59.676177978515625, -192.93832397460938, -60.70600891113281, -29.668960571289062, 0.9920501708984375, 19.588623046875, 36.636962890625, 9.69110107421875, 12.596115112304688, 130.38722229003906, 169.22579956054688, 163.69891357421875, 155.13943481445312, 12.772041320800781, 171.7205810546875, -41.344078063964844, 5.66571044921875, 24.89910888671875, -27.9627685546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000363.npy"} {"epoch": 0.7602094240837697, "step": 364, "batch_size": 128, "mean": 44.44951629638672, "std": 85.56878662109375, "min": -205.60931396484375, "p10": -42.84691619873047, "median": 26.143173217773438, "p90": 153.45910491943357, "max": 240.4739990234375, "pos_frac": 0.6640625, "sample": [71.09706115722656, -3.4648895263671875, 126.5279541015625, 42.649295806884766, 99.86640930175781, 14.392288208007812, 128.04034423828125, 148.30149841308594, 121.91400146484375, 8.433303833007812, 35.9669189453125, 97.38333129882812, -9.979766845703125, -10.309783935546875, -13.75732421875, 41.621673583984375, 55.894195556640625, -5.7464599609375, -2.968048095703125, 119.36378479003906, -150.2460479736328, -6.0744781494140625, 88.87513732910156, -155.32754516601562, 49.261474609375, -75.85281372070312, 174.17584228515625, 138.22930908203125, -10.984535217285156, 138.022705078125, 7.36907958984375, 17.364425659179688, 72.85096740722656, 6.740802764892578, -24.7548828125, -4.35711669921875, -3.5022430419921875, 135.51060485839844, 47.3843994140625, 147.19566345214844, 115.55438232421875, 9.346282958984375, -178.1715087890625, 60.32304382324219, 131.29006958007812, 175.21530151367188, -41.48920822143555, 87.61468505859375, -41.8580322265625, -26.123458862304688, 138.6336669921875, 157.62841796875, -11.221282958984375, 60.7322998046875, -12.715282440185547, -1.00689697265625, 175.868408203125, -1.3007144927978516, -205.60931396484375, 217.45057678222656, 70.24118041992188, -12.8797607421875, -10.42034912109375, -53.03404235839844, 34.73463439941406, 125.61198425292969, -27.8057861328125, 172.33242797851562, -19.10223388671875, 3.14019775390625, 34.53863525390625, 151.67225646972656, 82.37725830078125, -15.545845031738281, 117.81491088867188, 8.81243896484375, 181.146240234375, 68.72537994384766, 13.7572021484375, 143.91714477539062, 131.3785858154297, -47.45654296875, 28.89910888671875, -6.379486083984375, -5.0039520263671875, 18.77691650390625, 23.387237548828125, -23.80511474609375, 31.186141967773438, 213.66802978515625, 128.3974609375, 1.080902099609375, -31.308013916015625, 0.560791015625, 142.361083984375, 146.0545654296875, 36.4118766784668, 1.1485939025878906, -58.797210693359375, 240.4739990234375, 33.46697998046875, 160.1274871826172, -102.3375244140625, 204.42486572265625, 123.614013671875, -52.409332275390625, 124.07260131835938, 97.96234130859375, 1.783843994140625, -15.021087646484375, 129.82080078125, 0.593994140625, -16.568191528320312, 182.03382873535156, 21.666175842285156, 1.89666748046875, 4.7171630859375, -95.25138854980469, 92.53759765625, -99.30836486816406, 10.218158721923828, 65.17989349365234, 20.46185302734375, -21.949371337890625, -45.15431213378906, 106.16461181640625, 215.32484436035156, 33.133270263671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000364.npy"} {"epoch": 0.762303664921466, "step": 365, "batch_size": 128, "mean": 57.23159408569336, "std": 82.88621520996094, "min": -163.5445556640625, "p10": -25.38856658935546, "median": 34.073585510253906, "p90": 168.16177368164062, "max": 205.46176147460938, "pos_frac": 0.75, "sample": [0.0, 3.0829124450683594, 125.11962890625, 7.209686279296875, 167.65896606445312, 181.54962158203125, 13.515899658203125, 60.07489013671875, 0.0, 33.92889404296875, 23.8304443359375, 5.9464111328125, 14.918914794921875, 173.13107299804688, 13.890037536621094, 12.661026000976562, 19.6402587890625, 66.29059600830078, 16.469467163085938, 162.6850128173828, 44.707672119140625, 162.4912109375, -0.558197021484375, 18.562362670898438, 42.43003845214844, 34.21827697753906, -20.39269256591797, 152.30032348632812, 132.4960174560547, 153.17971801757812, 154.60797119140625, -14.452354431152344, -36.239044189453125, 122.13155364990234, -70.2611083984375, -72.84765625, 139.31146240234375, 70.46405792236328, 128.55381774902344, 62.2510986328125, -18.087739944458008, 141.19439697265625, 176.317138671875, 126.917724609375, -32.04609680175781, 182.03329467773438, -54.30023193359375, 127.98249816894531, 37.50800323486328, 156.8662109375, 136.98069763183594, -22.53533935546875, 1.66314697265625, -4.4423370361328125, 15.685882568359375, 9.02587890625, 112.33262634277344, 164.9491729736328, 152.05242919921875, -56.79411315917969, 25.998291015625, 140.30899047851562, -37.042022705078125, -5.3739013671875, 5.072807312011719, 90.169677734375, 129.67921447753906, 137.9937286376953, 18.79083251953125, 119.49160766601562, 186.42532348632812, -8.64093017578125, 180.64016723632812, 0.0975799560546875, 10.43927001953125, 187.63623046875, 96.54067993164062, -135.39691162109375, 166.7569580078125, 2.390716552734375, 179.74948120117188, 2.5056076049804688, 199.22671508789062, 3.09625244140625, 18.251222610473633, -19.6871337890625, 146.94586181640625, 59.23072814941406, 2.415966033935547, -16.72736358642578, 89.90043640136719, 11.252059936523438, 63.56538391113281, 169.33499145507812, -6.8206024169921875, 8.550155639648438, 5.7974853515625, -20.2681884765625, 135.0782470703125, -10.33685302734375, 25.150253295898438, -145.04588317871094, -18.49139404296875, -2.94012451171875, 139.63381958007812, 0.0, 22.56832504272461, 8.715263366699219, 133.13375854492188, 180.21978759765625, 93.10891723632812, 108.35499572753906, 5.624874114990234, -21.40814208984375, -56.484375, -163.5445556640625, 140.33493041992188, 174.37020874023438, -39.70294189453125, 132.55319213867188, 34.691314697265625, -134.165283203125, 205.46176147460938, 89.12545776367188, 39.924591064453125, 138.619873046875, 139.6678009033203, 105.26715087890625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000365.npy"} {"epoch": 0.7643979057591623, "step": 366, "batch_size": 128, "mean": 55.65153503417969, "std": 81.83153533935547, "min": -191.29061889648438, "p10": -35.039913940429685, "median": 43.519012451171875, "p90": 161.1506881713867, "max": 212.8896484375, "pos_frac": 0.7265625, "sample": [23.873367309570312, 32.15296936035156, 212.8896484375, 143.72373962402344, 122.78844451904297, 138.0223388671875, 166.689453125, 82.91519165039062, 73.10589599609375, 180.27835083007812, 167.39913940429688, 128.20579528808594, 117.44512176513672, -40.139373779296875, 22.5272216796875, 122.84286499023438, 91.63499450683594, 83.16932678222656, -134.87734985351562, 37.48808288574219, -191.29061889648438, 196.811279296875, -117.41516876220703, 12.112483978271484, 51.2188720703125, 137.29953002929688, 130.12802124023438, 74.43280029296875, 148.1195068359375, 105.68426513671875, -13.775177001953125, 151.65017700195312, 19.4208984375, -34.214813232421875, 157.19247436523438, -146.70675659179688, -1.5572509765625, 134.3980712890625, 23.785797119140625, 44.06463623046875, 3.6000747680664062, -85.8111572265625, -25.79974365234375, -16.61396026611328, -0.407684326171875, 76.76155090332031, 148.5187225341797, 7.1571044921875, 114.97627258300781, -14.5830078125, 63.87176513671875, 35.353851318359375, 10.40472412109375, 50.05105972290039, 108.68495178222656, -19.387420654296875, 30.033294677734375, 15.83001708984375, 60.986236572265625, 20.065513610839844, -1.139801025390625, 142.1147918701172, 108.56034851074219, 3.079774856567383, -3.8603439331054688, 169.1221923828125, 45.10317611694336, 104.55889892578125, 95.92264556884766, -72.553955078125, 138.76416015625, 171.49258422851562, -52.47772216796875, 0.0, -28.587127685546875, 169.0757598876953, 59.35797119140625, 111.1617431640625, 123.8482666015625, -1.5664291381835938, 7.6252288818359375, 42.973388671875, 179.423583984375, 83.87593078613281, -4.18084716796875, 54.03987121582031, 19.39453125, 191.33905029296875, -0.0848846435546875, -6.751899719238281, -3.9041748046875, -27.932861328125, 95.68203735351562, 12.573257446289062, 38.33552932739258, 134.2021484375, 41.375640869140625, -6.08868408203125, 54.963897705078125, 158.7769317626953, 143.40719604492188, 18.235031127929688, 4.86187744140625, 190.29244995117188, 140.7621612548828, 27.46739959716797, 127.46702575683594, -18.297607421875, 193.80499267578125, 2.032136917114258, 4.9447021484375, 126.7463607788086, 115.65194702148438, 2.90386962890625, -70.6849594116211, -41.240875244140625, 154.50686645507812, -103.37705993652344, 136.30665588378906, 32.395477294921875, 135.00747680664062, 13.01361083984375, -15.587356567382812, 147.96786499023438, -50.368499755859375, -11.288070678710938, 168.63006591796875, -36.96514892578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000366.npy"} {"epoch": 0.7664921465968586, "step": 367, "batch_size": 128, "mean": 59.29582214355469, "std": 78.16468048095703, "min": -131.7581787109375, "p10": -26.76533966064453, "median": 51.638145446777344, "p90": 168.41812133789062, "max": 197.15182495117188, "pos_frac": 0.7890625, "sample": [132.37319946289062, 28.17425537109375, -16.1524658203125, 177.39422607421875, -1.6675643920898438, 15.833877563476562, 54.536766052246094, -131.7581787109375, -8.3631591796875, 67.32171630859375, 155.49224853515625, 21.25511360168457, 19.42884063720703, 2.229095458984375, 117.64079284667969, 185.80123901367188, 46.37590026855469, 169.037841796875, 0.12603759765625, 13.16937255859375, 1.78363037109375, 193.92459106445312, 91.05026245117188, -43.822845458984375, 81.94843292236328, 185.04608154296875, -118.75827026367188, -12.57208251953125, 39.81535339355469, -20.735260009765625, -25.543441772460938, 106.39459228515625, 3.461742401123047, -43.35832214355469, 145.44705200195312, -7.67431640625, 72.3194580078125, 134.85569763183594, 115.43367004394531, -76.732666015625, 157.25347900390625, 91.35940551757812, 6.937238693237305, 117.5479736328125, 156.683349609375, 4.9783935546875, -80.8641357421875, 17.098297119140625, -1.3731689453125, -31.569808959960938, 50.8773193359375, 7.6571197509765625, 66.8580322265625, 62.70880126953125, 144.26632690429688, 152.60040283203125, 132.2432861328125, 114.92638397216797, 86.61407470703125, -25.715118408203125, -75.49200439453125, -27.853759765625, 191.81527709960938, 105.75897216796875, 16.457008361816406, 171.01626586914062, -26.298873901367188, 129.01556396484375, 197.15182495117188, 77.35292053222656, 67.5020751953125, 30.576171875, 22.700332641601562, 78.99920654296875, 175.51528930664062, 154.0345458984375, 41.07977294921875, 3.209747314453125, 15.591094970703125, -60.638580322265625, 35.8338623046875, 173.7420196533203, 153.24343872070312, 164.4205322265625, 15.1002197265625, 70.02413940429688, 159.16168212890625, 23.59649658203125, -49.364166259765625, 104.8966064453125, 151.6226043701172, 122.56625366210938, 20.773193359375, -8.99444580078125, 13.612070083618164, -4.812995910644531, -98.58661651611328, 168.15252685546875, 56.31413650512695, 167.74209594726562, 3.684722900390625, 179.88525390625, 151.90213012695312, 48.49578857421875, -12.555290222167969, 93.30194091796875, 64.52728271484375, 172.2100830078125, 7.5550537109375, 97.93634033203125, 74.66981506347656, 98.35597229003906, 28.19281005859375, -106.39432525634766, 52.39897155761719, 7.280731201171875, 6.326873779296875, 138.50543212890625, 78.15788269042969, 146.1110076904297, 13.070241928100586, 61.55316162109375, 17.9315185546875, 4.84173583984375, 190.59774780273438, 114.25738525390625, -19.803863525390625, 48.7122802734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000367.npy"} {"epoch": 0.768586387434555, "step": 368, "batch_size": 128, "mean": 65.09085083007812, "std": 81.19083404541016, "min": -151.103759765625, "p10": -22.16866378784179, "median": 51.742610931396484, "p90": 162.9662612915039, "max": 252.10171508789062, "pos_frac": 0.796875, "sample": [11.86322021484375, 179.3216552734375, 106.34977722167969, 151.1400146484375, 18.850189208984375, 159.00946044921875, 27.544021606445312, -3.475616455078125, 4.722412109375, -11.759597778320312, 139.77316284179688, -32.063323974609375, 82.5916748046875, 192.0909423828125, 108.77450561523438, 56.3795166015625, 207.5491943359375, 31.648162841796875, 75.61492919921875, 109.16895294189453, -20.518707275390625, -8.72509765625, 44.380828857421875, -20.584190368652344, 6.11505126953125, 23.259796142578125, -18.254295349121094, 5.177501678466797, 36.593658447265625, 159.567138671875, 2.3361892700195312, 79.392822265625, -48.30223846435547, 40.582244873046875, 10.2872314453125, -150.20587158203125, 4.689056396484375, 252.10171508789062, 131.95458984375, 127.32064819335938, 27.0887451171875, 163.33201599121094, 147.00213623046875, 131.6649169921875, 27.730133056640625, 20.077056884765625, 136.14939880371094, -7.6113739013671875, 25.723962783813477, -144.1956329345703, 6.031219482421875, 138.46533203125, 7.372835159301758, 110.59245300292969, -37.97521209716797, -8.165328979492188, 180.39602661132812, 135.236572265625, -43.3785400390625, 116.43269348144531, 29.099639892578125, 60.497161865234375, 58.76776123046875, 150.4080352783203, 168.14865112304688, 124.68118286132812, -38.6331787109375, 98.20443725585938, 31.369873046875, -8.691192626953125, 0.13287353515625, 186.1793212890625, 157.63217163085938, 41.808013916015625, 32.22938537597656, 131.67982482910156, 149.04364013671875, 95.98773193359375, 160.2707977294922, 138.10501098632812, 9.739669799804688, 120.839599609375, 153.8547821044922, 53.94256591796875, -10.043144226074219, 160.276611328125, 7.728240966796875, 49.54265594482422, -25.865768432617188, 73.85067749023438, -1.1891040802001953, 12.670112609863281, 162.80950927734375, 23.21997833251953, 113.73954010009766, 22.094207763671875, 84.9950942993164, 124.34649658203125, -4.519676208496094, 121.0979232788086, 164.77212524414062, -56.825584411621094, 94.84294128417969, 158.920166015625, -48.2071533203125, 28.30108642578125, -11.5087890625, 125.04025268554688, 201.48953247070312, 170.78860473632812, 27.916961669921875, 145.0266571044922, 44.10858154296875, 11.426322937011719, 15.019859313964844, 144.45860290527344, 165.86065673828125, 155.645751953125, 139.1009521484375, 103.52444458007812, 43.311767578125, -116.66450500488281, -151.103759765625, 6.011810302734375, 116.71932220458984, 165.61195373535156, -83.82205200195312, 147.57955932617188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000368.npy"} {"epoch": 0.7706806282722513, "step": 369, "batch_size": 128, "mean": 48.31496047973633, "std": 83.4413070678711, "min": -174.55078125, "p10": -48.344746398925764, "median": 43.23550033569336, "p90": 156.4318389892578, "max": 224.92227172851562, "pos_frac": 0.7421875, "sample": [128.33094787597656, 59.963172912597656, 137.2471923828125, -74.18289184570312, 27.05170440673828, -7.4232940673828125, 61.675048828125, 47.13397216796875, 134.2379150390625, 0.5203857421875, 64.3345947265625, 42.5252685546875, -28.928592681884766, 60.19500732421875, 179.947509765625, 173.87619018554688, 120.43641662597656, 134.54782104492188, 20.08966064453125, 128.3352813720703, 180.56231689453125, 2.760457992553711, 101.72100830078125, 54.07786560058594, -38.4903564453125, -157.005859375, 116.97610473632812, -17.325225830078125, 85.61647033691406, 2.9552154541015625, 128.4696044921875, 23.9432373046875, 155.26678466796875, 107.655517578125, 75.30902099609375, 140.17800903320312, -25.3902587890625, 68.85443115234375, 148.36727905273438, 56.77326965332031, 17.46307373046875, 4.86212158203125, 154.34564208984375, -35.26092529296875, 13.35357666015625, -90.3592529296875, -3.5504112243652344, 137.35296630859375, -115.8424072265625, 180.82818603515625, -44.27357482910156, -130.93344116210938, 80.59130859375, -90.09701538085938, 166.26580810546875, 123.0715560913086, -112.41583251953125, -9.049415588378906, 138.12911987304688, 7.857452392578125, -130.06192016601562, 0.02789306640625, -111.59095764160156, 154.52000427246094, 166.1321563720703, 58.139808654785156, 93.99545288085938, -2.698760986328125, 98.99267578125, -30.130874633789062, 224.92227172851562, -13.539436340332031, 18.05373191833496, 15.520416259765625, 2.28753662109375, 200.72244262695312, 66.7069091796875, 3.72235107421875, -59.33045196533203, 127.00904083251953, -0.17041778564453125, -174.55078125, 9.147727966308594, 55.711181640625, 16.917724609375, 113.60350799560547, 164.7191162109375, 131.0498504638672, -57.844146728515625, 0.0, 46.34503173828125, 138.80482482910156, -4.015510559082031, 161.38743591308594, -23.230209350585938, 3.21795654296875, 117.9715576171875, -19.72747802734375, 46.827545166015625, 151.1352081298828, 18.64105224609375, 146.06027221679688, 14.998046875, 17.641571044921875, 63.0875244140625, -15.414386749267578, 143.1844482421875, 18.058837890625, 117.1856689453125, -62.939605712890625, 98.65902709960938, 30.010093688964844, 14.218902587890625, 140.75997924804688, 159.15029907226562, 2.4315185546875, 2.20770263671875, 170.45144653320312, -23.221206665039062, 46.41361999511719, 16.561641693115234, 171.51235961914062, 15.006317138671875, 22.715744018554688, 43.94573211669922, -12.968994140625, 49.70361328125, 2.026336669921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000369.npy"} {"epoch": 0.7727748691099476, "step": 370, "batch_size": 128, "mean": 59.354774475097656, "std": 88.26921081542969, "min": -193.46844482421875, "p10": -19.67388191223144, "median": 35.80769348144531, "p90": 161.57047424316406, "max": 315.5024719238281, "pos_frac": 0.78125, "sample": [108.67570495605469, -152.46487426757812, 155.5853271484375, 35.02545166015625, 38.12884521484375, 25.147933959960938, -52.78251647949219, 3.76824951171875, 125.51573181152344, 19.658554077148438, 85.06495666503906, 89.2103271484375, -10.500667572021484, -3.51556396484375, 124.85348510742188, 11.16314697265625, 0.858856201171875, 232.080322265625, 4.175773620605469, 29.9193115234375, 135.6156005859375, 30.253421783447266, 114.22576904296875, 142.04823303222656, -17.564300537109375, -3.5078353881835938, -18.660144805908203, 144.49087524414062, 6.091701507568359, 29.33380126953125, 28.04278564453125, -15.12054443359375, 137.36544799804688, 33.55984115600586, 5.0510101318359375, 0.7766571044921875, 221.80682373046875, 94.62873840332031, 3.824146270751953, 173.92031860351562, 124.1595230102539, -156.57838439941406, 139.584716796875, 8.715375900268555, 118.16127014160156, 99.38423156738281, 81.94192504882812, 39.17970275878906, 38.93133544921875, 0.5002555847167969, 118.26669311523438, 17.228546142578125, 11.06640625, 129.471435546875, 97.568359375, -15.429244995117188, 124.72895812988281, -1.9337158203125, 160.50021362304688, -113.29137420654297, -22.039268493652344, 157.09527587890625, -10.944351196289062, -50.1903076171875, 7.918781280517578, 146.5253448486328, 210.0164794921875, 76.65545654296875, 122.74739074707031, -76.08832550048828, 2.161102294921875, -9.359268188476562, 58.232177734375, 158.5089874267578, 170.69775390625, -25.731658935546875, 175.17120361328125, 171.1553955078125, 107.52755737304688, 155.8706512451172, 167.93341064453125, 142.78587341308594, 34.471710205078125, 207.99752807617188, 141.62448120117188, 149.8267059326172, -24.813499450683594, 7.51898193359375, -165.33480834960938, 151.55682373046875, 130.8662109375, 9.758245468139648, -84.03335571289062, 26.7845458984375, 315.5024719238281, -7.322635650634766, 23.228912353515625, 36.589935302734375, 119.1251220703125, 152.7000732421875, 159.25914001464844, 170.7652587890625, 58.0048828125, 164.0677490234375, -11.484115600585938, 115.64642333984375, 82.73306274414062, 34.4211540222168, 37.074737548828125, 66.28570556640625, 186.04541015625, -193.46844482421875, 8.445594787597656, -7.60052490234375, 2.7817916870117188, 144.2916259765625, 14.457977294921875, 8.663619995117188, 151.4987335205078, 14.593597412109375, 12.45059585571289, -123.18643188476562, 29.21929168701172, 151.37109375, -7.377967834472656, 8.681015014648438, 124.79705047607422, -5.630706787109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000370.npy"} {"epoch": 0.774869109947644, "step": 371, "batch_size": 128, "mean": 47.92729187011719, "std": 80.885498046875, "min": -201.83609008789062, "p10": -33.015710449218744, "median": 33.48386764526367, "p90": 151.09036560058593, "max": 198.6194610595703, "pos_frac": 0.75, "sample": [0.7902030944824219, 97.65284729003906, 50.1824951171875, 142.36569213867188, -12.453760147094727, 0.19074249267578125, 183.85406494140625, 3.42095947265625, 22.598800659179688, 22.229461669921875, 2.4535980224609375, -191.32925415039062, 10.10955810546875, 13.631256103515625, 29.252090454101562, -12.483154296875, -67.895751953125, 98.99337768554688, 9.069366455078125, -20.0120849609375, -26.0958251953125, 39.4747314453125, -9.72268295288086, 164.606689453125, 70.82926940917969, 109.701904296875, -0.27266502380371094, 43.840576171875, -32.04414367675781, 42.286895751953125, 129.33154296875, 164.8795623779297, -65.70989227294922, 18.186111450195312, -201.83609008789062, 71.0596923828125, -40.714508056640625, 1.1011810302734375, 121.79820251464844, 26.632781982421875, 153.4151611328125, 13.46173095703125, -100.14646911621094, -6.2674560546875, 54.752655029296875, 65.58975219726562, 13.563629150390625, 0.0, -41.500213623046875, 6.760917663574219, 171.61720275878906, 147.6634063720703, 83.314453125, -66.29962158203125, -146.18148803710938, -6.211334228515625, 28.573013305664062, 173.1109619140625, 145.66195678710938, 11.420127868652344, 16.98542022705078, 85.287353515625, 68.50917053222656, 192.06149291992188, 31.30413818359375, 69.13006591796875, 45.733428955078125, 137.96725463867188, 162.26815795898438, -21.1898193359375, 31.698726654052734, 145.87109375, 124.51139068603516, -113.21917724609375, 51.965736389160156, 131.51473999023438, 133.3900909423828, 29.072433471679688, 150.09402465820312, 19.699325561523438, -23.730270385742188, -136.50643920898438, 21.297607421875, 198.6194610595703, 97.20172119140625, -20.26751708984375, -86.83919525146484, 133.15203857421875, 130.00238037109375, 48.93254089355469, 0.9988861083984375, -19.207763671875, 77.78427124023438, 103.95845031738281, 32.48060607910156, -2.1855926513671875, 140.4908447265625, 59.343109130859375, 22.445770263671875, 195.70175170898438, 22.954065322875977, -1.7330818176269531, 99.74072265625, 34.868141174316406, 141.6912841796875, 136.92974853515625, -1.311370849609375, 35.186859130859375, 19.189720153808594, 80.61517333984375, 130.60836791992188, -35.28269958496094, 93.05560302734375, 51.485267639160156, 8.23907470703125, 145.07810974121094, 159.6693115234375, 12.598640441894531, 193.14962768554688, 146.70492553710938, 1.7003173828125, -25.588653564453125, 34.48712921142578, -6.313507080078125, 52.92466735839844, 138.36703491210938, 114.4809799194336, 172.61793518066406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000371.npy"} {"epoch": 0.7769633507853403, "step": 372, "batch_size": 128, "mean": 65.39488220214844, "std": 85.91545104980469, "min": -164.06329345703125, "p10": -18.672069549560543, "median": 59.80385971069336, "p90": 171.32655334472656, "max": 258.7939453125, "pos_frac": 0.796875, "sample": [186.892822265625, -6.90350341796875, 134.86569213867188, 70.92030334472656, -1.6175003051757812, 135.35389709472656, 122.33647155761719, -8.78094482421875, 126.99205780029297, 167.9497833251953, 70.5469970703125, 66.62620544433594, 130.85458374023438, 132.7877197265625, -21.613632202148438, 128.42889404296875, 14.78375244140625, 164.202392578125, -7.398967742919922, 18.378501892089844, 80.82955932617188, 30.113853454589844, 126.54527282714844, -29.2989501953125, -14.262939453125, 132.375244140625, 197.05145263671875, 96.78901672363281, 149.63796997070312, 130.25404357910156, -28.835556030273438, -28.23870849609375, 176.95797729492188, -12.8931884765625, 99.67303466796875, 124.53689575195312, 186.9918212890625, 11.469619750976562, 90.46514892578125, 60.17534637451172, -17.411399841308594, 176.59829711914062, 62.94549560546875, 114.03636932373047, 187.52041625976562, -164.06329345703125, 189.1610107421875, 94.34811401367188, 12.390159606933594, 38.345458984375, 19.661949157714844, 8.132476806640625, 117.47966003417969, 0.302398681640625, 24.222137451171875, 30.053466796875, 109.95068359375, 151.73654174804688, 125.72412109375, 29.818817138671875, 127.62809753417969, 30.246307373046875, 128.10113525390625, 17.806884765625, 6.1322021484375, 62.03334045410156, 149.09033203125, 26.15460205078125, 163.41094970703125, -12.571392059326172, 1.763885498046875, 12.75750732421875, 14.996551513671875, 154.1865234375, 169.95852661132812, 187.709716796875, -91.86322021484375, 179.4420166015625, 24.5960693359375, -145.80015563964844, 147.1300506591797, 61.883575439453125, 37.837890625, 57.245849609375, -138.7147674560547, 53.99774169921875, 58.47198486328125, 44.73652648925781, -5.705951690673828, 157.73402404785156, -12.17935562133789, 5.514984130859375, 151.638427734375, 68.60806274414062, 41.2783203125, -163.6959228515625, -80.13864135742188, 35.153961181640625, 125.42620849609375, -61.80457305908203, 158.18377685546875, -4.9997711181640625, 8.008636474609375, 8.474868774414062, 168.7101593017578, 136.17526245117188, -156.19775390625, 174.51861572265625, 258.7939453125, 185.8133544921875, 33.65826416015625, 207.08294677734375, 163.83766174316406, 2.0868911743164062, 53.54583740234375, 167.632080078125, -0.8498191833496094, 59.432373046875, -93.2366943359375, 158.86167907714844, 101.361572265625, 137.84408569335938, 2.539997100830078, 3.83831787109375, -1.5907096862792969, 62.6812744140625, 43.59991455078125, 23.643707275390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000372.npy"} {"epoch": 0.7790575916230367, "step": 373, "batch_size": 128, "mean": 60.56721496582031, "std": 85.52291870117188, "min": -199.98028564453125, "p10": -33.476933288574216, "median": 47.601924896240234, "p90": 172.48021850585934, "max": 198.4237060546875, "pos_frac": 0.7890625, "sample": [11.506591796875, 46.75884246826172, 158.9227294921875, 31.734909057617188, 0.7721328735351562, -15.682891845703125, 17.83087158203125, 187.63467407226562, 16.07806396484375, 170.1761474609375, 145.7548828125, -15.840087890625, 23.830902099609375, 158.86392211914062, 103.29507446289062, 86.88076782226562, 2.436473846435547, 49.771697998046875, -34.445404052734375, 141.09170532226562, 61.502471923828125, 2.4758262634277344, 105.78843688964844, 77.25921630859375, -43.430999755859375, 39.6094970703125, -4.263771057128906, -33.06187438964844, 36.91339111328125, 190.72164916992188, 4.083061218261719, -24.217483520507812, 46.178253173828125, 24.025177001953125, 61.835693359375, 141.98272705078125, 152.51593017578125, -18.7691650390625, 158.8675537109375, 81.49783325195312, 67.9144287109375, 143.24673461914062, 165.05422973632812, 50.09888458251953, 16.74725341796875, 117.65087890625, 10.18389892578125, -85.37918090820312, 48.44500732421875, 177.85638427734375, -32.123477935791016, 153.69406127929688, -69.42842864990234, -41.9024658203125, 70.196533203125, -74.3966064453125, 190.46043395996094, -6.688507080078125, 27.315948486328125, 67.11277770996094, 4.194284439086914, -199.98028564453125, 194.20037841796875, 30.395294189453125, 158.31280517578125, 14.004791259765625, -115.92955017089844, 164.22994995117188, 18.691482543945312, 129.61778259277344, 164.3839111328125, 158.27943420410156, 12.05743408203125, 76.7744140625, 31.730789184570312, 75.60124206542969, 60.37115478515625, 0.3197021484375, 95.76171875, 5.29107666015625, -142.37802124023438, -16.375289916992188, 185.29513549804688, 56.801856994628906, 42.40118408203125, 164.94720458984375, 124.05947875976562, 189.610595703125, 195.0189208984375, 104.74824523925781, 6.430328369140625, 169.7740936279297, -3.79754638671875, 21.256393432617188, 164.22413635253906, 162.37213134765625, 18.98431396484375, 187.070556640625, -119.56684875488281, -130.6321258544922, 78.94696807861328, 152.79429626464844, 198.4237060546875, 154.85110473632812, 70.83926391601562, 14.52691650390625, 129.5016632080078, -14.945770263671875, 195.75399780273438, 15.885223388671875, 14.159626007080078, 14.360832214355469, 21.003326416015625, 94.51731872558594, -57.313175201416016, -5.7117919921875, 132.15786743164062, 4.4653472900390625, -34.71342468261719, 114.23146057128906, -3.136197090148926, 178.5343017578125, -18.27396011352539, 11.28338623046875, 166.07034301757812, 179.66644287109375, 128.455322265625, 12.77203369140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000373.npy"} {"epoch": 0.7811518324607329, "step": 374, "batch_size": 128, "mean": 41.904762268066406, "std": 84.56990814208984, "min": -183.440673828125, "p10": -55.74536285400391, "median": 23.001625061035156, "p90": 162.93541412353517, "max": 227.76837158203125, "pos_frac": 0.734375, "sample": [211.9462890625, 0.0, 6.34759521484375, 215.70123291015625, 2.89764404296875, -87.38137817382812, -11.278953552246094, 23.189987182617188, 165.61997985839844, 126.21534729003906, 7.827101707458496, 201.50027465820312, 16.73480224609375, -55.85972595214844, 39.09461975097656, 35.552947998046875, 124.44702911376953, 127.59014892578125, 22.071075439453125, 8.504196166992188, -121.27374267578125, -41.237640380859375, 103.40702056884766, -123.79830932617188, 158.06700134277344, 27.951698303222656, 17.94512939453125, 10.94036865234375, 5.526975631713867, -23.153823852539062, 11.830093383789062, 7.5995635986328125, -23.9521484375, 52.90761947631836, -7.712158203125, 36.25011444091797, 21.37701416015625, -16.769866943359375, 180.66058349609375, 138.82968139648438, 47.681884765625, 9.30340576171875, 227.76837158203125, 0.19017791748046875, 65.0521240234375, 2.78143310546875, 0.570526123046875, 40.63726806640625, -183.440673828125, -16.63878631591797, 32.986968994140625, -45.0928955078125, 162.66427612304688, 79.06621551513672, 22.03759765625, 81.76870727539062, 9.93157958984375, 68.00689697265625, 70.50047302246094, -0.4423942565917969, 27.539718627929688, 180.5362548828125, 185.142578125, -29.473602294921875, -0.716796875, 64.84949493408203, 50.67974853515625, 61.9254150390625, -21.129119873046875, 141.03892517089844, 18.544708251953125, 55.60760498046875, 130.96878051757812, 123.7066650390625, -146.03814697265625, 35.22991180419922, -55.69635009765625, 10.3814697265625, 25.334442138671875, 144.261962890625, 120.41797637939453, 11.98565673828125, 46.894012451171875, 148.5224609375, 0.0, 149.11471557617188, 9.762786865234375, -80.5482177734375, -38.233978271484375, -97.191162109375, 16.396766662597656, 125.69866943359375, 96.97125244140625, 163.5680694580078, 22.813262939453125, -19.47039794921875, 199.100830078125, 21.408599853515625, 181.23834228515625, 58.4468994140625, 17.365966796875, -105.898681640625, -98.7564697265625, 28.860626220703125, 16.582578659057617, 1.6521720886230469, 70.69427490234375, 184.15414428710938, -13.116096496582031, -83.07292175292969, 0.59832763671875, 80.78118896484375, 2.481658935546875, -5.3359527587890625, 153.916015625, -2.10528564453125, 133.79510498046875, 67.79487609863281, -117.23458862304688, 100.90513610839844, 210.91006469726562, 142.58544921875, 146.81365966796875, 35.314666748046875, 33.591705322265625, -4.591419219970703, 73.98280334472656, -119.89813232421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000374.npy"} {"epoch": 0.7832460732984293, "step": 375, "batch_size": 128, "mean": 52.82130432128906, "std": 85.59053039550781, "min": -175.688720703125, "p10": -33.711169433593746, "median": 33.141502380371094, "p90": 166.66869506835937, "max": 246.58685302734375, "pos_frac": 0.7578125, "sample": [51.29742431640625, 46.236846923828125, 94.32986450195312, -4.420928955078125, 33.9462890625, 4.171012878417969, 25.163818359375, -116.16909790039062, 6.6527557373046875, 36.533660888671875, 177.8497314453125, 0.6569843292236328, 159.9505615234375, 171.91378784179688, 36.80201721191406, -9.567535400390625, 87.56983947753906, 118.76434326171875, 5.704898834228516, 6.881614685058594, 155.82977294921875, 166.2056884765625, 14.941848754882812, 41.58953857421875, 63.81572723388672, 153.97828674316406, -13.678321838378906, 9.878662109375, -38.0120849609375, 62.95732879638672, 15.160163879394531, 14.88399887084961, -4.057647705078125, -93.25140380859375, 11.162895202636719, 159.554443359375, 106.934814453125, 167.59124755859375, 22.42432403564453, -26.68646240234375, 33.31199645996094, 12.522216796875, -120.78628540039062, 0.991607666015625, -175.688720703125, 115.03256225585938, 148.3929443359375, 145.86260986328125, 15.211097717285156, -139.7503662109375, -134.63238525390625, 20.5423583984375, 93.75100708007812, -4.1993255615234375, 32.97100830078125, -3.9037322998046875, 134.34579467773438, -10.556549072265625, -32.34527587890625, -121.937744140625, 172.50177001953125, 161.206298828125, -19.3148193359375, 208.60418701171875, 26.208438873291016, 24.9493408203125, -5.6753997802734375, 3.4998016357421875, 246.58685302734375, -8.973844528198242, 66.5177001953125, 12.615280151367188, 24.018936157226562, 93.58179473876953, 42.178192138671875, 180.408447265625, 48.769927978515625, 1.9078369140625, 51.72686767578125, 61.15968322753906, 130.6956787109375, 107.03323364257812, -36.89825439453125, -50.94004821777344, -0.6845855712890625, 122.1531982421875, 19.559173583984375, 127.771240234375, 83.51956176757812, 150.63970947265625, -142.11526489257812, 168.97891235351562, 197.89276123046875, -14.741836547851562, 199.57391357421875, 215.68939208984375, -45.006072998046875, 160.67832946777344, 107.39694213867188, 166.2733154296875, -12.423370361328125, -52.58281707763672, 162.23013305664062, -0.31732177734375, 141.0751953125, 2.293842315673828, 129.81427001953125, 171.940673828125, 64.8948974609375, 4.80364990234375, -4.1948699951171875, 175.91586303710938, 113.25852966308594, 36.13694763183594, 156.51522827148438, -29.14935302734375, 3.4164581298828125, 159.6341094970703, 1.8085479736328125, 18.772918701171875, 15.104522705078125, 24.375221252441406, 98.64892578125, 124.3074951171875, 10.979217529296875, 139.85548400878906, 9.705780029296875, 99.73414611816406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000375.npy"} {"epoch": 0.7853403141361257, "step": 376, "batch_size": 128, "mean": 48.91350173950195, "std": 87.66482543945312, "min": -160.67672729492188, "p10": -55.028053283691406, "median": 37.994415283203125, "p90": 165.8668426513672, "max": 255.64398193359375, "pos_frac": 0.6875, "sample": [85.85853576660156, 62.859161376953125, 166.9433135986328, 255.64398193359375, 122.21017456054688, 224.13116455078125, 40.96490478515625, 89.80000305175781, 47.72737121582031, -109.63009643554688, 97.3072509765625, -24.691299438476562, -80.86279296875, 27.585418701171875, -20.6556396484375, 60.82615661621094, -0.6753158569335938, 42.341766357421875, 125.5799560546875, 166.86734008789062, 120.19631958007812, -148.39886474609375, -1.0014724731445312, 11.805206298828125, 66.5081787109375, 24.89798355102539, -26.208450317382812, 175.5799560546875, -2.8669281005859375, 146.91204833984375, 147.38906860351562, -4.8558502197265625, 165.75497436523438, 37.41259765625, 1.875823974609375, -48.24951171875, 125.45095825195312, 74.5344467163086, 166.12786865234375, 10.574188232421875, 0.0, 20.341140747070312, 19.162643432617188, 119.89014434814453, -48.734596252441406, 188.13958740234375, -90.67904663085938, 138.19436645507812, 105.34693908691406, 126.25570678710938, -135.82086181640625, 132.901123046875, 45.090301513671875, 6.6588134765625, 149.22573852539062, 9.648468017578125, 27.5716552734375, 102.83100891113281, -61.784332275390625, -10.0201416015625, 111.826904296875, -3.0963134765625, 210.45443725585938, 196.7308349609375, 18.09442138671875, 184.08160400390625, 159.37579345703125, 70.43121337890625, -34.38531494140625, -27.19451904296875, 80.3450927734375, -54.69500732421875, 27.83807373046875, -38.73033905029297, -5.322395324707031, -1.24853515625, -123.63819885253906, 11.520347595214844, -73.37088012695312, 169.5507049560547, 118.120361328125, 57.183135986328125, 159.684326171875, 37.68511962890625, 151.99884033203125, 125.0587158203125, 7.538543701171875, -160.67672729492188, 11.877838134765625, -101.96849060058594, 48.069580078125, 2.816516876220703, 143.76165771484375, -18.664772033691406, 8.336406707763672, 132.9417724609375, 98.529296875, -10.998992919921875, -114.1748046875, 159.65296936035156, -2.822509765625, 101.48284912109375, -55.80516052246094, 137.37704467773438, 18.92633056640625, 21.90524673461914, 98.69353485107422, 35.169677734375, 54.731964111328125, 144.14385986328125, -17.88360595703125, -41.87457275390625, -36.72792434692383, 39.8475341796875, 122.574951171875, -13.825538635253906, 33.74627685546875, -34.218666076660156, 59.819915771484375, -1.2488174438476562, 5.456647872924805, 74.61561584472656, 80.52969360351562, 38.3037109375, 171.5772705078125, -131.17344665527344, 191.61117553710938, 162.8675537109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000376.npy"} {"epoch": 0.787434554973822, "step": 377, "batch_size": 128, "mean": 46.79558563232422, "std": 82.21345520019531, "min": -151.08172607421875, "p10": -31.862603759765623, "median": 25.901596069335938, "p90": 165.4336486816406, "max": 211.02584838867188, "pos_frac": 0.7109375, "sample": [178.70452880859375, 115.34815979003906, 124.76686096191406, -13.153038024902344, 81.486572265625, 9.809921264648438, 4.76806640625, 76.31671142578125, 139.60914611816406, 9.73779296875, 27.589073181152344, 49.63397216796875, 119.339111328125, -58.49261474609375, 118.51675415039062, 135.77708435058594, 172.12701416015625, -28.54150390625, 1.5537757873535156, -32.11993408203125, 136.28778076171875, 2.267547607421875, 10.554458618164062, 41.4406623840332, -14.480375289916992, -14.175743103027344, 68.61996459960938, -12.22906494140625, 162.43907165527344, 28.173934936523438, 107.06208038330078, 21.77703857421875, 104.91545104980469, -121.786376953125, -9.445358276367188, 150.77444458007812, 3.2630577087402344, -5.28851318359375, -9.95703125, 41.995574951171875, 16.73590087890625, 105.40000915527344, 12.950202941894531, -70.125732421875, 132.39088439941406, 8.51611328125, 188.13623046875, 60.06597900390625, 11.120773315429688, -106.7449951171875, 11.10723876953125, 194.8231201171875, -27.800758361816406, 0.533843994140625, 174.1566162109375, 4.596593856811523, -31.7523193359375, -0.5431556701660156, 137.65603637695312, 155.84298706054688, 153.8096923828125, 27.0595703125, 99.99435424804688, 116.3262939453125, -22.479888916015625, 145.0315704345703, 85.36021423339844, 0.0, -35.85296630859375, 31.149574279785156, 134.80368041992188, -7.836395263671875, 116.7552490234375, 141.6074676513672, -7.126258850097656, 41.047637939453125, 9.430034637451172, -14.953704833984375, -0.7235813140869141, 140.35035705566406, 21.211761474609375, 205.64886474609375, 5.856414794921875, -24.776123046875, -15.55340576171875, -84.64398956298828, 41.60078430175781, 4.166759490966797, 162.5650634765625, 40.945037841796875, 63.4075927734375, 29.729400634765625, 8.833908081054688, 139.95111083984375, -85.5023193359375, 174.12660217285156, 11.602409362792969, 24.743621826171875, 45.188629150390625, 132.94613647460938, 44.87188720703125, -119.89779663085938, -12.62933349609375, 47.285736083984375, 185.76104736328125, -64.13847351074219, 195.0087890625, 32.15423583984375, 21.751373291015625, 140.49639892578125, 15.847389221191406, 11.447280883789062, 28.8294677734375, 6.1932373046875, 177.65945434570312, 139.27200317382812, 175.4619140625, -128.93194580078125, -13.481842041015625, 211.02584838867188, 16.394577026367188, -14.042999267578125, -131.8873748779297, 181.036376953125, -11.409805297851562, -9.24908447265625, -151.08172607421875, 58.235748291015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000377.npy"} {"epoch": 0.7895287958115184, "step": 378, "batch_size": 128, "mean": 46.32257080078125, "std": 81.33554077148438, "min": -160.870849609375, "p10": -41.589586639404295, "median": 24.190841674804688, "p90": 165.75372924804688, "max": 198.163818359375, "pos_frac": 0.7109375, "sample": [129.34890747070312, 3.30633544921875, 19.30364990234375, 127.24017333984375, -23.60595703125, 7.713106155395508, -28.60089111328125, 9.057373046875, 171.2137451171875, 80.312255859375, -7.702117919921875, 1.610443115234375, 27.213897705078125, 135.0684814453125, 134.07608032226562, -46.28106689453125, -160.870849609375, 174.06631469726562, 187.89208984375, -25.5550537109375, 37.701995849609375, 23.559783935546875, 11.610504150390625, -44.578338623046875, 92.9271240234375, 176.35064697265625, 198.15591430664062, 143.7848358154297, 23.3936767578125, 131.56689453125, -0.3022613525390625, 140.39230346679688, 115.93284606933594, 23.38916015625, 10.65899658203125, -136.47128295898438, 2.2535343170166016, 2.819183349609375, 57.37261962890625, 90.7784423828125, 0.0, 155.32937622070312, -6.325347900390625, 131.52072143554688, 184.97604370117188, -48.80426025390625, -8.72903823852539, 198.163818359375, -21.652889251708984, 165.45904541015625, 91.3896484375, 114.73866271972656, -10.048171997070312, 5.94439697265625, 8.313262939453125, 5.511474609375, -30.987548828125, 119.13027954101562, -19.451751708984375, 145.98727416992188, 75.42913818359375, -1.088766098022461, 134.4716796875, 144.04931640625, 38.454925537109375, 126.32440185546875, -12.675567626953125, -70.0177001953125, 168.75686645507812, 55.45270538330078, 80.86959838867188, 31.10040283203125, 0.20648193359375, -16.319488525390625, -118.8778076171875, 165.94757080078125, -70.92684936523438, -91.82711791992188, 32.39801025390625, -2.4722747802734375, 40.453636169433594, 5.176849365234375, 94.091796875, 107.77745056152344, 9.3204345703125, 98.35354614257812, 175.07601928710938, 8.383926391601562, 15.466796875, 39.0662841796875, 150.77249145507812, 151.2646942138672, -24.28253173828125, 7.451141357421875, 24.8218994140625, -40.308692932128906, -79.87939453125, 9.971649169921875, 12.36090087890625, -21.860931396484375, -76.385009765625, 173.6123046875, -5.2755126953125, 52.09480285644531, -137.94996643066406, -3.81927490234375, 26.5458984375, 124.49237823486328, 0.0389404296875, 34.394378662109375, 46.632781982421875, 43.15362548828125, -17.036300659179688, -6.4782257080078125, 190.617919921875, 118.74383544921875, 165.670654296875, 159.80738830566406, 173.30694580078125, 16.439254760742188, -76.51235961914062, 82.3862533569336, 131.50653076171875, 47.13555908203125, 32.83311462402344, -29.925140380859375, 13.7742919921875, 0.182586669921875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000378.npy"} {"epoch": 0.7916230366492146, "step": 379, "batch_size": 128, "mean": 47.82709503173828, "std": 87.95182800292969, "min": -173.80328369140625, "p10": -59.25342712402342, "median": 29.80011749267578, "p90": 165.14521331787108, "max": 205.550048828125, "pos_frac": 0.7421875, "sample": [10.190292358398438, 9.548751831054688, 2.5354461669921875, 9.033111572265625, -1.3954238891601562, -94.95533752441406, 18.712013244628906, 149.65475463867188, 30.442153930664062, 2.6708831787109375, 143.13589477539062, 117.06259155273438, 50.19696807861328, 50.51806640625, 13.786102294921875, 26.746795654296875, 70.87106323242188, -120.4312744140625, 176.99703979492188, 19.501998901367188, 140.94357299804688, 108.73309326171875, 152.26353454589844, -33.94342041015625, 12.021629333496094, -6.7372283935546875, 154.3011016845703, 3.44927978515625, 77.799072265625, -6.3224945068359375, 34.48858642578125, 164.95213317871094, 168.94064331054688, 56.455322265625, 43.64543151855469, 8.28131103515625, -17.121543884277344, 179.06552124023438, 143.71908569335938, 7.442558288574219, -6.8904876708984375, -73.87164306640625, 29.1580810546875, 16.343795776367188, 5.3734130859375, 191.94985961914062, 12.663421630859375, -71.80508422851562, -22.068572998046875, 28.896759033203125, 133.64532470703125, -10.072224617004395, 35.865631103515625, -143.41372680664062, -1.764373779296875, 9.58990478515625, 163.94232177734375, 145.60870361328125, 128.53982543945312, 126.7469482421875, 69.55615997314453, -103.32015991210938, 23.58795166015625, 46.612945556640625, 146.99899291992188, 205.550048828125, 12.085845947265625, 167.2535400390625, 5.6648101806640625, -26.973968505859375, -52.19059753417969, 0.0, -6.676239013671875, 49.46821594238281, 21.44317626953125, 175.04898071289062, 53.628997802734375, 150.42758178710938, 169.66583251953125, 162.38523864746094, 8.58587646484375, 123.51190185546875, 11.354324340820312, 35.37175750732422, 108.13619995117188, 149.97076416015625, -8.664665222167969, 120.62596130371094, -10.812362670898438, 143.06028747558594, 85.93731689453125, -6.336036682128906, 177.33428955078125, 2.423917770385742, 117.35751342773438, -11.6484375, 2.351837158203125, 85.86886596679688, -127.50532531738281, 201.8775634765625, 167.43939208984375, -173.80328369140625, 8.207763671875, 10.918731689453125, 164.72244262695312, 79.36883544921875, 59.67620849609375, -47.929443359375, 97.39466857910156, 159.19711303710938, 178.42684936523438, 75.45355224609375, 160.09002685546875, 60.272796630859375, -116.3324203491211, -135.77430725097656, 31.600914001464844, 3.3393726348876953, -0.004791259765625, 164.74867248535156, -53.8741455078125, 116.86246490478516, -144.0303192138672, -96.41966247558594, -137.69097900390625, 20.22861671447754, 49.527679443359375, 165.59573364257812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000379.npy"} {"epoch": 0.793717277486911, "step": 380, "batch_size": 128, "mean": 50.725990295410156, "std": 83.70243835449219, "min": -136.97653198242188, "p10": -45.47111663818359, "median": 36.059974670410156, "p90": 162.92108764648438, "max": 253.83294677734375, "pos_frac": 0.7265625, "sample": [135.19357299804688, 62.062957763671875, 79.14637756347656, 61.49931335449219, 122.00381469726562, 18.164459228515625, -123.39573669433594, 4.466156005859375, 1.7452049255371094, -61.352386474609375, 92.5771484375, 96.8714599609375, 147.20196533203125, -116.32762908935547, 39.8494873046875, 13.387458801269531, 112.85479736328125, 42.86383056640625, 143.73428344726562, -3.7042694091796875, 0.0, 44.139190673828125, 10.452598571777344, 27.594970703125, 23.028564453125, 159.38922119140625, 7.4541015625, 82.128173828125, -0.0553741455078125, 10.94482421875, 50.83465576171875, 206.27032470703125, 12.840362548828125, 124.5430908203125, -2.414520263671875, -136.97653198242188, 128.3300323486328, -5.841312408447266, 40.04155731201172, -17.2762451171875, 45.7869873046875, 63.18995666503906, 163.66458129882812, -39.61260986328125, -18.416725158691406, 15.37493896484375, 16.353668212890625, 178.3966064453125, 179.04364013671875, -49.907928466796875, -77.77102661132812, 41.81269836425781, 20.405593872070312, -0.797210693359375, 22.654937744140625, -72.88811492919922, -23.48998260498047, 13.93359375, 177.9525146484375, 146.33230590820312, -58.74371337890625, 158.63275146484375, 91.72312927246094, 39.80487060546875, 128.08990478515625, 161.95645141601562, -88.68692779541016, -5.79278564453125, 18.342071533203125, 40.74972152709961, 145.7801513671875, 183.77325439453125, 1.818359375, 123.86006927490234, 157.4040985107422, -5.2278289794921875, 70.50332641601562, -37.913330078125, 41.74444580078125, -128.16734313964844, 106.24552917480469, 253.83294677734375, -34.6392822265625, 133.7582244873047, 136.47671508789062, 79.04563903808594, 144.37844848632812, 160.91717529296875, 162.60244750976562, 79.78912353515625, 36.72821044921875, 215.64791870117188, -71.50607299804688, 71.83340454101562, 18.783248901367188, 61.139869689941406, 3.27423095703125, -42.71563720703125, 180.39898681640625, 245.07821655273438, 34.24430465698242, 47.62074279785156, 23.929824829101562, 160.65179443359375, 35.39173889160156, 16.513465881347656, 26.865478515625, 47.120330810546875, -32.843292236328125, -62.8605842590332, 8.25921630859375, -43.56962585449219, 8.914459228515625, 20.27152442932129, 24.613021850585938, 115.30621337890625, -20.66485595703125, 85.84356689453125, 89.3402328491211, 225.986083984375, 30.452392578125, -14.00811767578125, -35.49737548828125, 173.60076904296875, -2.4774322509765625, -68.06719970703125, 198.97781372070312, -17.9959716796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000380.npy"} {"epoch": 0.7958115183246073, "step": 381, "batch_size": 128, "mean": 56.75567626953125, "std": 79.83351135253906, "min": -126.24275207519531, "p10": -33.42230377197264, "median": 44.266536712646484, "p90": 163.204345703125, "max": 188.61172485351562, "pos_frac": 0.7109375, "sample": [131.40652465820312, 186.5400390625, 172.87295532226562, 48.81431579589844, 178.26634216308594, -41.350250244140625, 165.16510009765625, 179.27084350585938, 123.7666015625, 5.961097717285156, 183.60272216796875, 24.933395385742188, 15.138168334960938, 146.59515380859375, 119.9925537109375, 157.1817626953125, 0.0, 10.773101806640625, 89.7457275390625, 112.51007080078125, 126.82673645019531, 48.16339111328125, 15.1265869140625, 188.61172485351562, 51.730255126953125, 79.42375946044922, 93.95027923583984, -126.24275207519531, 0.0, -71.84466552734375, 101.81182861328125, 112.67864227294922, -12.4930419921875, 142.3531951904297, -7.524021148681641, 55.29052734375, 16.88739013671875, 54.07366943359375, 11.180999755859375, 114.671875, 184.0408935546875, 89.31077575683594, -18.572235107421875, 135.0301055908203, 132.345703125, 8.49407958984375, 164.28143310546875, -4.367362976074219, -73.10214233398438, 129.19981384277344, 148.66990661621094, 160.1517791748047, 43.72868347167969, 1.3323516845703125, 154.75128173828125, 162.74273681640625, 12.04510498046875, 137.2462158203125, 167.6812744140625, 72.48665618896484, -65.70661926269531, 48.050933837890625, 84.23374938964844, 8.963104248046875, -7.511688232421875, 38.35203170776367, 154.58511352539062, -2.940032958984375, -20.462432861328125, -25.245361328125, -2.9706649780273438, -0.05444526672363281, -6.643524169921875, 146.95016479492188, -21.50146484375, 1.009857177734375, -1.8074378967285156, 99.56611633300781, 53.11865234375, -97.60231018066406, 159.24462890625, -30.024612426757812, 142.30014038085938, 44.134033203125, 172.81216430664062, -4.268310546875, -22.90350341796875, -8.476455688476562, 158.07691955566406, 11.900146484375, -4.566497802734375, -51.916412353515625, 137.89356994628906, 124.59495544433594, 114.45648193359375, -47.607666015625, 3.305553436279297, 7.23699951171875, 22.959991455078125, -106.19234466552734, 107.66154479980469, 7.4523162841796875, -90.86660766601562, 111.14348602294922, -95.15786743164062, 96.81785583496094, -23.9561767578125, 7.5765228271484375, -17.356292724609375, 143.4888916015625, -7.5830078125, 178.0350799560547, -49.09881591796875, 185.9876708984375, 58.600555419921875, 36.93065643310547, 20.676475524902344, 9.469154357910156, 133.38424682617188, 23.567306518554688, 152.29922485351562, 156.5189208984375, 44.39904022216797, -27.8177490234375, -62.783782958984375, 103.19241333007812, 23.71697998046875, 19.724822998046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000381.npy"} {"epoch": 0.7979057591623037, "step": 382, "batch_size": 128, "mean": 69.7794189453125, "std": 86.35844421386719, "min": -129.10279846191406, "p10": -35.5057502746582, "median": 74.12223052978516, "p90": 171.2456848144531, "max": 211.01364135742188, "pos_frac": 0.765625, "sample": [-115.44403076171875, 27.864105224609375, 68.70599365234375, 48.769775390625, 97.73858642578125, 155.23025512695312, 156.53192138671875, -19.921661376953125, 23.437498092651367, 124.23576354980469, 108.59036254882812, -110.2591552734375, 115.56532287597656, 14.807212829589844, 176.65115356445312, 72.78866577148438, 59.887115478515625, 118.44825744628906, 137.1314697265625, -13.560211181640625, 52.2130126953125, -129.10279846191406, 159.81092834472656, -5.1390380859375, 3.8159637451171875, 177.65890502929688, 42.815277099609375, -27.59869384765625, 140.10507202148438, 110.96929168701172, -10.34771728515625, 139.6814727783203, 105.98446655273438, 33.44348907470703, 151.75677490234375, 100.3967056274414, -6.710273742675781, 69.44939422607422, -5.83100700378418, 168.17031860351562, 28.350189208984375, 30.6522216796875, 165.91799926757812, 26.77978515625, 161.7736358642578, 166.02053833007812, 178.58290100097656, 47.8394775390625, 16.994232177734375, 11.053863525390625, 170.31341552734375, 130.1210479736328, 9.763397216796875, 117.17694091796875, 0.432373046875, 115.66912841796875, 211.01364135742188, 198.41925048828125, -78.05136108398438, -9.97186279296875, -40.19140625, -128.21392822265625, 209.28555297851562, 53.298301696777344, 2.8365631103515625, 102.10148620605469, 195.30154418945312, 69.480224609375, 180.5404052734375, 28.99884033203125, 189.27810668945312, 23.961441040039062, 46.98783874511719, -5.117706298828125, 161.95388793945312, 115.48348999023438, 132.74766540527344, 20.771987915039062, 116.82380676269531, 164.19091796875, 41.397705078125, -58.522216796875, 56.7052001953125, 143.3282470703125, -117.49774169921875, -3.011749267578125, -86.00721740722656, 161.2987060546875, -20.22454833984375, 162.5231475830078, 174.6435546875, -24.559173583984375, 13.347671508789062, 172.86915588378906, 163.35446166992188, 82.31330871582031, 123.00414276123047, -100.63726806640625, 7.3423004150390625, -13.904266357421875, 30.8099365234375, 71.7635498046875, 116.11723327636719, 148.75953674316406, 123.82284545898438, -66.55563354492188, 32.86231994628906, 132.6838836669922, -99.43032836914062, 138.84010314941406, 133.4781494140625, -0.9541397094726562, 139.05462646484375, 121.1572265625, 75.45579528808594, 131.97132873535156, -94.40872192382812, 83.99554443359375, -33.49761199951172, 119.64129638671875, 165.92941284179688, -20.982433319091797, 199.84759521484375, 128.8302459716797, 173.30642700195312, 96.30120849609375, 170.54991149902344, -23.45721435546875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000382.npy"} {"epoch": 0.8, "step": 383, "batch_size": 128, "mean": 50.702659606933594, "std": 83.1688232421875, "min": -149.606201171875, "p10": -56.95737915039062, "median": 31.09466552734375, "p90": 161.06887817382812, "max": 208.92984008789062, "pos_frac": 0.71875, "sample": [-1.0521240234375, 43.0252685546875, 144.46560668945312, 129.10015869140625, -51.478912353515625, 24.359710693359375, -31.961029052734375, 30.287872314453125, -0.87139892578125, 54.78474426269531, 133.10751342773438, -121.91520690917969, 139.16067504882812, 31.0445556640625, 0.0, -12.135719299316406, 21.688125610351562, -58.748992919921875, 50.216552734375, -56.189544677734375, 19.8282470703125, 175.74188232421875, 135.59312438964844, 99.69268798828125, -116.1875, 174.38343811035156, -85.4495849609375, -5.322864532470703, -73.72494506835938, 11.029891967773438, 73.39202880859375, 10.519775390625, -14.407062530517578, -46.06938934326172, 149.2362060546875, -61.027587890625, 187.28494262695312, 158.861328125, 35.80445861816406, 138.7924041748047, 82.6180419921875, 5.144096374511719, 98.02340698242188, 208.92984008789062, -90.36839294433594, -1.006256103515625, 44.68772888183594, 161.99151611328125, 54.8214111328125, 8.0361328125, 163.73504638671875, 2.658843994140625, -4.963994979858398, 144.46505737304688, 35.043365478515625, 22.6778564453125, 4.0805206298828125, -74.36483001708984, 0.1144561767578125, -126.50027465820312, 130.74191284179688, 124.2906494140625, 11.276470184326172, 168.56405639648438, -26.064483642578125, 26.526992797851562, 160.6734619140625, 0.26253509521484375, -2.25836181640625, 17.245697021484375, 110.17647552490234, 171.45616149902344, 132.87808227539062, 133.52081298828125, 113.92990112304688, 55.08464431762695, -10.225067138671875, -0.2326507568359375, 34.823089599609375, 141.1753387451172, 2.4266204833984375, 117.70696258544922, 131.51600646972656, 58.34117126464844, 156.8609619140625, 30.112319946289062, 26.09881591796875, 22.297386169433594, 55.95957946777344, -149.606201171875, 163.55178833007812, 8.80752944946289, 59.00294494628906, 74.99883270263672, 31.144775390625, 5.383724212646484, -41.40960693359375, 188.96217346191406, 135.96820068359375, 129.65756225585938, 140.13150024414062, -42.06732177734375, 208.1480712890625, -85.29026794433594, -72.6075439453125, 23.99585723876953, 107.26702880859375, 19.4237060546875, 0.0, 127.9716796875, 149.5057373046875, -3.071012496948242, 57.16363525390625, 117.3944091796875, 89.40762329101562, -20.718185424804688, -44.03118896484375, 14.521383285522461, 171.0980224609375, 118.17739868164062, 9.373001098632812, 141.73580932617188, 189.84506225585938, -99.52760314941406, 27.7496337890625, -4.008026123046875, 137.2770538330078, 134.76513671875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000383.npy"} {"epoch": 0.8020942408376963, "step": 384, "batch_size": 128, "mean": 37.57197570800781, "std": 87.96893310546875, "min": -166.14208984375, "p10": -77.21373825073242, "median": 22.54499053955078, "p90": 157.37059936523437, "max": 213.66998291015625, "pos_frac": 0.6484375, "sample": [191.00128173828125, -166.14208984375, 98.88458251953125, -21.11431884765625, -22.88995361328125, 142.79507446289062, 54.493438720703125, 213.66998291015625, 23.021743774414062, -14.7545166015625, 142.1418914794922, -18.121578216552734, -26.686309814453125, -82.96981811523438, 33.779632568359375, -117.71957397460938, 153.5679473876953, -3.4482498168945312, 176.58868408203125, -26.3541259765625, 4.463262557983398, 175.36862182617188, 137.13778686523438, 134.57476806640625, -6.1439056396484375, 146.6575927734375, 23.560195922851562, 134.35409545898438, 17.29876708984375, -82.00271606445312, 3.569833755493164, -55.19683837890625, -10.180389404296875, 65.6187744140625, -25.5848388671875, 98.56983947753906, 23.456756591796875, 34.64617919921875, 143.567626953125, -18.05957794189453, 13.0079345703125, -122.2755126953125, -2.7678298950195312, 16.12397003173828, 135.60052490234375, 149.7877960205078, 149.75616455078125, -158.85646057128906, 3.8349609375, 20.201927185058594, 121.69927215576172, 182.41452026367188, -68.56523132324219, -8.36431884765625, -139.37533569335938, 49.59809112548828, 88.91790771484375, 82.37420654296875, -121.65849304199219, -25.115997314453125, 16.592742919921875, -127.220458984375, 123.27130126953125, 133.24615478515625, 169.5760498046875, 44.14628601074219, 45.80657958984375, 54.62677001953125, 83.59405517578125, -4.241146087646484, -10.626522064208984, 96.98050689697266, 156.58740234375, 12.706588745117188, 126.21810913085938, 162.38763427734375, -11.185211181640625, 17.3985595703125, -24.306774139404297, 83.23291015625, 41.46501159667969, -103.88209533691406, 107.7987060546875, 86.1856689453125, 144.556884765625, 205.22927856445312, 25.17108154296875, 9.665008544921875, 78.71145629882812, 39.075286865234375, 173.255859375, -8.727912902832031, 147.29736328125, 63.998992919921875, -6.15380859375, 4.304779052734375, -36.0321044921875, 171.7329559326172, 9.806182861328125, -33.125640869140625, -3.96405029296875, -48.909027099609375, 22.0682373046875, -61.509796142578125, -136.72216796875, 31.137258529663086, 0.69500732421875, 29.429792404174805, 11.21603012084961, 57.74847412109375, 186.42343139648438, -129.8302001953125, 55.31915283203125, -4.269428253173828, -4.742565155029297, -12.65032958984375, 169.359375, 159.19805908203125, -78.64179229736328, 126.57452392578125, 134.84951782226562, -60.74072265625, 3.030590057373047, 16.17138671875, 29.0235595703125, 71.13128662109375, 13.536529541015625, -76.60171508789062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000384.npy"} {"epoch": 0.8041884816753927, "step": 385, "batch_size": 128, "mean": 49.97581481933594, "std": 82.09783172607422, "min": -158.13006591796875, "p10": -37.64782104492187, "median": 37.26817321777344, "p90": 163.03512420654297, "max": 221.68719482421875, "pos_frac": 0.734375, "sample": [-41.145599365234375, 57.323978424072266, 162.9984893798828, 91.50799560546875, 40.27593994140625, 27.29168701171875, 154.67608642578125, 28.530914306640625, 160.35443115234375, 140.97003173828125, 28.667816162109375, 0.081817626953125, -51.984832763671875, 22.32305908203125, 26.997589111328125, 54.26707458496094, 100.78106689453125, 125.28338623046875, -66.77197265625, -16.654312133789062, 163.12060546875, -46.934844970703125, 140.18496704101562, -108.94139099121094, 40.104736328125, 160.7483673095703, 127.30655670166016, 12.282310485839844, 128.72216796875, 26.626251220703125, 3.8685760498046875, 201.52838134765625, 132.8577880859375, 25.37921142578125, 173.19302368164062, 221.68719482421875, 148.72079467773438, 24.176010131835938, 83.20452117919922, 214.6873016357422, 49.982177734375, -32.77900695800781, 56.355712890625, 71.16754150390625, 71.9464111328125, -121.09074401855469, 133.39492797851562, 140.6830596923828, -126.484619140625, -10.05389404296875, 212.37185668945312, 173.44268798828125, 213.25082397460938, -9.523765563964844, 126.05854797363281, 25.08184814453125, 64.93496704101562, 1.2688026428222656, -158.13006591796875, 62.348854064941406, 16.8309326171875, 182.31524658203125, -10.6300048828125, 158.75436401367188, 5.804718017578125, 3.2655029296875, -26.833343505859375, 130.38421630859375, -4.9959716796875, 63.47479248046875, 99.30912780761719, 3.1721725463867188, -31.77789306640625, 10.796428680419922, 20.873580932617188, -23.629043579101562, 109.81692504882812, 137.34317016601562, 123.60379028320312, -100.34536743164062, 12.47802734375, 84.4280776977539, 140.37741088867188, 67.98186492919922, 122.1226806640625, 12.074172973632812, -13.109207153320312, 27.115768432617188, -131.0201416015625, 3.518077850341797, 56.6966552734375, -4.819766998291016, -7.625640869140625, 40.40180206298828, 166.13616943359375, 47.991851806640625, 0.0, 39.38322448730469, 68.90933227539062, 21.505889892578125, 1.3258132934570312, 12.53839111328125, 47.000396728515625, -36.148773193359375, 182.73333740234375, -6.777091979980469, -104.7811279296875, 114.43452453613281, -28.930511474609375, 186.72125244140625, 45.06396484375, 0.0, -27.368865966796875, -93.82376861572266, 127.513671875, 3.108306884765625, 87.2978515625, -41.30158233642578, -1.225921630859375, -2.74200439453125, -19.364212036132812, 19.61602783203125, 72.03479766845703, 36.509063720703125, 115.36715698242188, 38.02728271484375, 170.25689697265625, 15.21673583984375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000385.npy"} {"epoch": 0.806282722513089, "step": 386, "batch_size": 128, "mean": 64.05810546875, "std": 88.26769256591797, "min": -224.65231323242188, "p10": -17.702905273437494, "median": 47.971527099609375, "p90": 174.9167999267578, "max": 233.61077880859375, "pos_frac": 0.765625, "sample": [118.01568603515625, -5.030052185058594, 5.179473876953125, 127.68610382080078, 65.48883056640625, 73.96527099609375, 67.226318359375, 168.9718017578125, 114.07553100585938, 142.10324096679688, 140.00778198242188, 233.61077880859375, -36.20068359375, 31.637176513671875, 189.99447631835938, 29.06292724609375, 28.53094482421875, 30.73492431640625, 4.08393669128418, 113.30743408203125, 113.31661987304688, 148.11785888671875, 22.6812744140625, 60.171630859375, -11.66552734375, -141.24220275878906, 38.5648193359375, -11.472732543945312, 146.59873962402344, 85.0567626953125, 31.559051513671875, -122.43583679199219, 185.2105712890625, 34.11279296875, -10.993194580078125, 16.24639892578125, 15.599807739257812, 1.1712799072265625, 50.53753662109375, 141.0518798828125, 223.9029541015625, 156.6171875, 87.33578491210938, 38.50422668457031, 117.67984008789062, 124.37762451171875, 187.0123291015625, -70.8006591796875, 45.9892578125, 202.24224853515625, 94.0540771484375, -22.116058349609375, -82.33343505859375, -37.741294860839844, 192.02639770507812, 138.3401641845703, -10.208976745605469, 0.0, -40.4833984375, -15.811553955078125, -144.3643798828125, -98.92657470703125, 124.97518920898438, 38.1739501953125, 2.090433120727539, -9.01507568359375, 51.297332763671875, -2.020719528198242, 154.2145233154297, -175.5490264892578, 180.63653564453125, -7.0654296875, 138.52725219726562, 198.5150146484375, 198.15716552734375, 162.0977783203125, 8.86569595336914, -1.991943359375, 160.23439025878906, 144.30813598632812, 136.6045379638672, 10.410247802734375, -224.65231323242188, 47.57012939453125, -0.889495849609375, 164.49365234375, 148.19125366210938, 13.13763427734375, 137.4927978515625, 1.481475830078125, 51.1220703125, 18.847183227539062, -1.0496826171875, 0.0, 191.96331787109375, 82.67735290527344, -9.41802978515625, 7.539360046386719, 81.10052490234375, 141.58047485351562, 159.61337280273438, 168.46168518066406, 48.3729248046875, 162.1225128173828, 12.267250061035156, 172.07696533203125, 30.6650390625, -23.055984497070312, -2.16552734375, 153.4012451171875, 180.63954162597656, 43.83148193359375, 184.4032440185547, 100.83604431152344, 5.50970458984375, 24.32696533203125, 29.987106323242188, 9.479339599609375, 90.80731201171875, 142.97720336914062, 28.48898696899414, 172.46548461914062, 13.70132827758789, 9.573783874511719, 156.48455810546875, 0.0, 140.88864135742188, 98.68873596191406], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000386.npy"} {"epoch": 0.8083769633507853, "step": 387, "batch_size": 128, "mean": 57.81619644165039, "std": 88.4186782836914, "min": -187.13134765625, "p10": -37.95851135253906, "median": 38.549346923828125, "p90": 174.16193542480468, "max": 209.8009033203125, "pos_frac": 0.7109375, "sample": [57.13618469238281, -11.129241943359375, 147.60498046875, 38.06706619262695, 8.12799072265625, 177.42718505859375, 192.7104949951172, -78.47518920898438, -145.99624633789062, -77.90237426757812, 0.0, 180.98471069335938, -11.797637939453125, 193.65371704101562, 133.43417358398438, 101.57275390625, 7.125335693359375, -14.66510009765625, -19.737213134765625, 36.546905517578125, -38.946197509765625, 18.54444122314453, 150.7400360107422, -90.34939575195312, 45.62651824951172, 162.50030517578125, -96.69842529296875, -21.656646728515625, 162.6640167236328, 1.343698501586914, 153.2684326171875, -43.274383544921875, 155.940185546875, 36.1688232421875, 146.75094604492188, 144.24542236328125, -15.86273193359375, 104.75369262695312, 51.81090545654297, -9.00543212890625, -37.53521728515625, 135.54571533203125, 38.108673095703125, -14.80780029296875, 181.47189331054688, 47.2147216796875, 194.55035400390625, 79.10659790039062, -37.20294952392578, 0.1753387451171875, -3.5257644653320312, 179.70443725585938, -44.159576416015625, -15.052978515625, 20.2998046875, 23.19879150390625, 14.74627685546875, -106.40402221679688, 141.1878662109375, 153.1428680419922, 113.65142822265625, 35.637451171875, 62.09112548828125, 26.318817138671875, 137.08518981933594, -124.59759521484375, 38.03485107421875, 38.990020751953125, 7.761627197265625, -12.776206970214844, -19.629837036132812, -19.49090576171875, 50.80474853515625, 185.03067016601562, 148.33494567871094, 126.23251342773438, 78.37283325195312, 39.64673614501953, -7.7574920654296875, 109.080322265625, 12.88165283203125, 1.2259750366210938, 165.87742614746094, 19.924560546875, 85.349609375, 172.76254272460938, 128.16001892089844, 8.109039306640625, 114.19561767578125, 140.3876953125, 2.2884674072265625, 166.703125, 73.36406707763672, 22.478836059570312, 199.41107177734375, 82.92086791992188, 147.35113525390625, -20.921112060546875, 150.00857543945312, 152.4710235595703, 6.731172561645508, 36.53729248046875, -27.05413818359375, 143.29812622070312, -5.67280387878418, -63.4215087890625, -187.13134765625, -15.564834594726562, 140.90121459960938, 209.8009033203125, 17.582977294921875, 27.355941772460938, 77.57312774658203, 117.38363647460938, 190.17266845703125, 140.64431762695312, 146.6099090576172, 41.72894287109375, 200.52203369140625, 13.6085205078125, -15.21380615234375, -128.41693115234375, 169.97091674804688, 187.531494140625, -10.652748107910156, 149.1395721435547, -18.583026885986328, 134.302490234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000387.npy"} {"epoch": 0.8104712041884817, "step": 388, "batch_size": 128, "mean": 45.28251647949219, "std": 89.58303833007812, "min": -174.39761352539062, "p10": -48.978573608398435, "median": 22.309014320373535, "p90": 160.79660339355468, "max": 275.9781494140625, "pos_frac": 0.6796875, "sample": [99.79583740234375, -111.58273315429688, 141.6126708984375, 0.0, 275.9781494140625, 152.02000427246094, -4.62921142578125, 179.72552490234375, 50.74687194824219, 160.94522094726562, -144.05499267578125, 140.21047973632812, -133.48538208007812, 4.30866813659668, 5.8963623046875, 25.654754638671875, 41.963356018066406, 121.33145141601562, 200.64422607421875, -24.490020751953125, 134.73098754882812, 204.90557861328125, -99.859130859375, -5.150238037109375, -13.619537353515625, -42.04948425292969, 95.61710357666016, 128.52780151367188, 165.93609619140625, 167.04208374023438, 56.85462951660156, 16.96497344970703, 0.0, 11.250579833984375, 17.550643920898438, -96.155029296875, -16.63702392578125, 14.823577880859375, 110.96875, 101.080322265625, -78.29850769042969, 21.82830047607422, 94.44401550292969, 139.4998016357422, -27.257354736328125, -40.74853515625, 166.30795288085938, -8.234535217285156, -7.537731170654297, 6.6501312255859375, 99.65725708007812, -14.68988037109375, 136.05181884765625, -34.25114440917969, 13.754379272460938, 129.36090087890625, -5.4075927734375, 173.4547119140625, 87.71893310546875, -34.51190185546875, -31.621810913085938, 53.578033447265625, 142.62962341308594, 6.5797271728515625, 19.203054428100586, -50.0399169921875, 99.95590209960938, -172.95858764648438, 180.9053955078125, -109.82737731933594, 50.582366943359375, 157.30831909179688, 129.61767578125, 110.78461456298828, 28.03009033203125, -0.0037841796875, -0.3084716796875, 160.73291015625, 10.7362060546875, -136.193603515625, -14.789993286132812, 176.61993408203125, 156.6315460205078, -19.7025146484375, 19.97067642211914, 10.577377319335938, 29.085769653320312, 132.925537109375, 82.73197937011719, 22.182130813598633, -79.59225463867188, 39.594024658203125, 6.0564727783203125, -174.39761352539062, 45.26736831665039, 180.53724670410156, 156.08709716796875, 142.61398315429688, 32.31700134277344, 143.12701416015625, 5.023323059082031, 120.53939819335938, 160.24127197265625, 187.73968505859375, -48.523712158203125, 10.6002197265625, 38.45237731933594, 22.435897827148438, -28.312057495117188, 1.9015235900878906, 159.81179809570312, 13.35235595703125, -33.75621032714844, 17.359588623046875, 20.320560455322266, -39.68793487548828, 127.623291015625, -3.0267715454101562, 142.599853515625, 50.672454833984375, -106.48934173583984, -22.570297241210938, 30.67919921875, 7.848114013671875, 26.020050048828125, 129.8139190673828, 124.50101470947266, -9.705848693847656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000388.npy"} {"epoch": 0.812565445026178, "step": 389, "batch_size": 128, "mean": 59.41819763183594, "std": 86.6992416381836, "min": -166.07241821289062, "p10": -37.82183990478515, "median": 62.13578796386719, "p90": 162.1979766845703, "max": 197.9437255859375, "pos_frac": 0.7265625, "sample": [-2.4198760986328125, -108.17178344726562, -16.794227600097656, 99.00872802734375, 140.4293212890625, 144.272705078125, 33.040496826171875, 26.506866455078125, -9.140830993652344, 179.88922119140625, 28.167999267578125, 147.22093200683594, 48.62774658203125, 184.12802124023438, 71.34612274169922, 140.92654418945312, -4.5672760009765625, 144.73818969726562, 100.44662475585938, 188.30569458007812, -58.54203796386719, 130.51480102539062, 174.91802978515625, 100.1988525390625, 4.722496032714844, 150.39903259277344, -154.56927490234375, 179.9716796875, 36.43864440917969, 168.15182495117188, 34.231689453125, -142.84622192382812, 126.4134521484375, -31.875076293945312, -47.8284912109375, -4.968475341796875, -54.63758087158203, 132.7684326171875, 178.38128662109375, 197.9437255859375, 142.69192504882812, 0.0, -106.06504821777344, -5.194999694824219, 131.37796020507812, -93.67108154296875, 161.750732421875, 137.3123779296875, 147.46177673339844, 19.926025390625, -5.820823669433594, 144.71774291992188, 33.085601806640625, 16.993730545043945, 122.27284240722656, 106.05245971679688, 28.132217407226562, 55.288665771484375, -36.453460693359375, -12.986236572265625, 46.35820007324219, 44.99213409423828, 85.08665466308594, 151.1923065185547, -90.77435302734375, 137.54754638671875, 83.42922973632812, 140.3347625732422, 104.83900451660156, 78.04576110839844, 138.28269958496094, -13.142913818359375, 40.79547119140625, 154.8880615234375, 3.881591796875, 191.10040283203125, 94.02548217773438, 72.3436279296875, -29.649627685546875, -8.06219482421875, 70.19488525390625, 100.41940307617188, 16.56574249267578, 53.761871337890625, 16.40172576904297, 152.8077392578125, -149.81845092773438, 136.3712921142578, 107.84860229492188, -19.400238037109375, 183.5020751953125, 27.225830078125, 0.122100830078125, 3.7689857482910156, -4.823692321777344, 122.8944320678711, 8.92022705078125, -166.07241821289062, -25.296356201171875, 163.24154663085938, -28.975341796875, 146.0517578125, -13.47100830078125, -41.01472473144531, 99.7633056640625, 174.77615356445312, 0.0, 85.74604797363281, 121.898193359375, 187.6797332763672, 8.00429916381836, 20.088363647460938, -2.644805908203125, 68.98291015625, 71.39726257324219, 159.63352966308594, 130.527099609375, 10.186485290527344, -21.660934448242188, 19.556724548339844, 96.98536682128906, 150.52542114257812, -137.31884765625, 14.193283081054688, 132.05865478515625, 0.035003662109375, 131.9363555908203, 153.81927490234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000389.npy"} {"epoch": 0.8146596858638744, "step": 390, "batch_size": 128, "mean": 56.33869934082031, "std": 91.56771087646484, "min": -172.89141845703125, "p10": -62.32678527832031, "median": 63.59966278076172, "p90": 169.7503479003906, "max": 198.5389404296875, "pos_frac": 0.75, "sample": [141.88671875, 1.0628814697265625, 54.209442138671875, 16.374786376953125, 37.2418212890625, 138.68777465820312, 1.9268226623535156, 154.51263427734375, 126.9361572265625, 63.82514953613281, -36.01231384277344, 8.073898315429688, 1.2929763793945312, -66.708984375, 33.10130310058594, 128.83749389648438, 0.0620269775390625, 179.82357788085938, 152.92147827148438, 101.07568359375, 106.65524291992188, -41.90338134765625, 161.10098266601562, 9.940170288085938, 69.3726806640625, 81.47573852539062, -129.287841796875, 151.400390625, 146.27493286132812, 63.697906494140625, 159.84158325195312, 133.32608032226562, -110.20561218261719, 44.899452209472656, 146.672119140625, 54.05572509765625, 84.84813690185547, 21.407577514648438, -1.7840118408203125, 63.55012512207031, -95.8359375, 146.36935424804688, -60.448699951171875, 127.98291015625, 90.98516082763672, -104.70077514648438, -26.603755950927734, -167.25628662109375, 138.40625, -35.251922607421875, 65.74589538574219, -11.771240234375, -144.6656494140625, 33.43682861328125, 192.5333251953125, 17.23410415649414, 95.39138793945312, 164.2957000732422, 126.79855346679688, -144.1304931640625, -97.24374389648438, 95.93710327148438, 167.74647521972656, 17.5234375, 184.1849365234375, -141.9805908203125, 195.60504150390625, 83.36318969726562, -17.112579345703125, 143.76010131835938, 62.44097900390625, 188.01483154296875, 111.36602783203125, 0.0, -17.758209228515625, -10.386077880859375, 146.4599609375, -24.986465454101562, 141.09481811523438, 106.04960632324219, -30.38304901123047, 122.032958984375, 182.8546142578125, -172.89141845703125, -16.41143798828125, 158.63473510742188, 198.5389404296875, 154.25938415527344, 5.5020751953125, 7.743318557739258, 25.84699249267578, 178.47720336914062, 79.19338989257812, 63.649200439453125, 69.33360290527344, 124.6965103149414, -12.051353454589844, 170.57489013671875, 30.034332275390625, -59.4127197265625, 5.293794631958008, -50.962554931640625, 177.93984985351562, 124.38497924804688, -27.395339965820312, 38.39984130859375, 186.0701904296875, 5.09417724609375, 1.5040016174316406, 4.592529296875, -72.66238403320312, 4.7583465576171875, -6.756385803222656, 124.507568359375, 67.04063415527344, 25.95733642578125, 161.033203125, 152.79324340820312, 64.37419128417969, 169.39697265625, 138.9678192138672, 30.6865234375, 14.5274658203125, 171.4964599609375, 173.20172119140625, -96.74154663085938, 121.91651153564453, 24.64916229248047], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000390.npy"} {"epoch": 0.8167539267015707, "step": 391, "batch_size": 128, "mean": 54.425926208496094, "std": 79.53353881835938, "min": -150.41357421875, "p10": -31.184666442871094, "median": 37.92572021484375, "p90": 157.8633026123047, "max": 310.62451171875, "pos_frac": 0.734375, "sample": [-0.9992904663085938, -55.854644775390625, 5.30183219909668, 100.79755401611328, 36.602020263671875, 9.08270263671875, -0.473876953125, 5.670326232910156, 17.33405303955078, 109.31099700927734, 3.2418441772460938, 14.461395263671875, -42.708656311035156, 90.77812194824219, 181.0294189453125, 56.167327880859375, 74.68803405761719, 99.16730499267578, -95.49342346191406, -12.013618469238281, -16.3175048828125, 154.90579223632812, 75.37460327148438, 118.25918579101562, -45.731964111328125, 43.55485534667969, 157.85671997070312, 75.16354370117188, 32.751399993896484, 28.400543212890625, 3.722076416015625, -54.59765625, 146.9263916015625, -9.066787719726562, 50.98486328125, 121.8111572265625, 126.48835754394531, 9.399011611938477, -13.702484130859375, 59.6307373046875, 12.122493743896484, 133.6599578857422, 150.7097625732422, 208.237060546875, 149.70263671875, -9.4544677734375, 157.878662109375, 12.68359375, -10.347661972045898, -1.235107421875, 200.12744140625, 88.25594329833984, -62.026405334472656, 53.71124267578125, -15.729568481445312, 110.89865112304688, 131.95816040039062, 119.03868103027344, 138.18325805664062, -30.98974609375, 44.9918212890625, -90.42744445800781, 135.56527709960938, 21.610626220703125, 32.54998779296875, 13.156707763671875, 101.6998291015625, 8.03720474243164, 59.21800994873047, 4.4983978271484375, 1.81378173828125, 167.033203125, 23.32537841796875, 122.468505859375, 123.7334976196289, -7.380096435546875, 127.23663330078125, 39.249420166015625, -6.7088623046875, 152.56375122070312, -1.678192138671875, 75.1083984375, 157.08111572265625, 310.62451171875, 174.22235107421875, 11.60565185546875, -9.37945556640625, -25.052597045898438, 150.7621307373047, 1.678436279296875, 12.45184326171875, -0.7269191741943359, -147.53402709960938, 153.73809814453125, 40.054779052734375, 54.7159423828125, -31.639480590820312, 125.54971313476562, 175.85528564453125, 170.747314453125, -34.5374755859375, -47.866455078125, 3.6251373291015625, -44.650390625, 3.140228271484375, -1.7792243957519531, 51.99229431152344, 93.46128845214844, 136.08599853515625, 161.58395385742188, 12.114837646484375, 129.18341064453125, 111.01229095458984, -14.243331909179688, 65.90422821044922, 63.567230224609375, 174.89413452148438, 147.37945556640625, 205.9810791015625, 16.466140747070312, -29.58599853515625, 9.430923461914062, 33.1988525390625, 6.149543762207031, -8.704658508300781, 42.1109619140625, -150.41357421875, 159.3096923828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000391.npy"} {"epoch": 0.818848167539267, "step": 392, "batch_size": 128, "mean": 69.54386901855469, "std": 80.4658203125, "min": -151.2593994140625, "p10": -12.660944366455077, "median": 45.43226623535156, "p90": 183.7215087890625, "max": 230.45053100585938, "pos_frac": 0.7890625, "sample": [156.8981170654297, 126.10126495361328, 42.79890441894531, 44.30340576171875, -13.903274536132812, 4.328559875488281, 31.54058837890625, 108.119140625, 77.750244140625, 129.6619873046875, 16.58831787109375, 83.14007568359375, 120.62246704101562, 19.32177734375, -11.348602294921875, 17.275253295898438, -151.2593994140625, 219.032958984375, 104.46903991699219, -22.289947509765625, 183.54510498046875, -6.084217071533203, 8.340484619140625, 67.81076049804688, 137.15390014648438, 192.17633056640625, 0.0, -41.6395263671875, 161.2061767578125, 142.18368530273438, 7.386711120605469, 29.28802490234375, 185.11962890625, 45.90966796875, 65.98297119140625, 186.98529052734375, 183.2132568359375, 163.83523559570312, 137.7447052001953, 65.87690734863281, 143.11294555664062, 155.98873901367188, 14.197738647460938, 12.89215087890625, -9.078399658203125, 22.062095642089844, -7.600105285644531, 2.40167236328125, 111.58001708984375, 160.01995849609375, 44.954864501953125, 168.09320068359375, 193.34353637695312, 113.37690734863281, 183.52789306640625, 135.0421600341797, 37.657928466796875, 97.78152465820312, 211.206787109375, -5.988067626953125, -12.128517150878906, 117.90848541259766, 2.423065185546875, 181.8858642578125, 33.70875930786133, 36.708518981933594, -27.103652954101562, 132.602294921875, 197.75588989257812, -0.13201904296875, 54.47955322265625, 175.81613159179688, 191.0623779296875, 135.52984619140625, 52.632659912109375, 186.65713500976562, 44.011505126953125, -95.69476318359375, -15.561477661132812, -4.852344512939453, 11.116607666015625, 154.25747680664062, 5.53955078125, 65.74578857421875, 38.83673095703125, -7.507904052734375, -7.79736328125, 6.5828857421875, 22.576675415039062, -24.815567016601562, -3.04901123046875, 0.0, 137.9861297607422, 21.96246337890625, 154.28591918945312, 193.91925048828125, 57.859375, 109.8057861328125, 184.13311767578125, 0.3068084716796875, 171.95248413085938, 0.1301746368408203, 51.79057312011719, 2.046854019165039, 7.09130859375, 230.45053100585938, -2.9710617065429688, 189.978515625, 19.862442016601562, 85.97187042236328, -112.34452819824219, 158.90670776367188, -22.43102264404297, 9.914077758789062, 141.093994140625, 25.63531494140625, 58.5142822265625, -49.656158447265625, 5.5712890625, 128.5664825439453, 44.357147216796875, 140.39846801757812, -34.04632568359375, 151.32925415039062, -25.18109130859375, 23.753623962402344, 153.81971740722656, 33.89764404296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000392.npy"} {"epoch": 0.8209424083769633, "step": 393, "batch_size": 128, "mean": 50.47279357910156, "std": 86.22425842285156, "min": -187.1534423828125, "p10": -44.73474197387695, "median": 34.623046875, "p90": 157.48641815185547, "max": 292.164794921875, "pos_frac": 0.7421875, "sample": [39.98726272583008, -85.69442749023438, 3.5612945556640625, 54.59716796875, 105.78885650634766, -63.9189453125, 141.49734497070312, 94.84259033203125, 157.23184204101562, 20.906982421875, 12.1051025390625, -3.4591064453125, 63.5460205078125, -8.220115661621094, 130.4473876953125, 27.358154296875, -103.66331481933594, 115.59059143066406, 95.54642486572266, 73.01348876953125, 164.03857421875, 51.287506103515625, 10.735626220703125, 47.08348083496094, -40.56964111328125, 169.83065795898438, 158.08042907714844, 152.9307861328125, 19.399749755859375, 14.958221435546875, 148.9349365234375, 37.10142517089844, 76.4503173828125, 135.4359130859375, 31.940826416015625, 32.14466857910156, -1.2012939453125, 169.80828857421875, 26.595184326171875, 140.71665954589844, 41.15350341796875, -53.8204345703125, 126.66094970703125, 160.06268310546875, -95.42129516601562, 160.23553466796875, -46.31505584716797, -30.751922607421875, -102.45906066894531, 83.4400634765625, 249.983154296875, 82.11221313476562, 10.279266357421875, 20.30059814453125, 13.780433654785156, 71.6728515625, 0.9325027465820312, -49.28367614746094, 108.62106323242188, 92.06846618652344, -176.66427612304688, 15.991485595703125, 14.033950805664062, -15.349029541015625, 22.007537841796875, 149.94309997558594, 106.92262268066406, 154.69570922851562, -154.505859375, -57.983795166015625, -38.416259765625, 28.205307006835938, 20.275428771972656, 51.04742431640625, 16.205402374267578, 58.665191650390625, 196.56893920898438, -14.618453979492188, 49.408782958984375, -9.61050033569336, 9.518566131591797, -9.541099548339844, 141.80184936523438, -156.88075256347656, 31.324203491210938, 113.37014770507812, 147.38038635253906, 97.06632995605469, 138.8462677001953, 151.2000274658203, -27.248931884765625, 173.36312866210938, 141.66986083984375, 159.69021606445312, 158.2010040283203, 28.399200439453125, 31.87542724609375, -40.95947265625, 134.2919464111328, -187.1534423828125, 9.532356262207031, 37.38148498535156, 0.56292724609375, -4.657318115234375, 292.164794921875, 88.81808471679688, 54.667144775390625, 0.0, 10.239913940429688, 51.927764892578125, 205.73699951171875, -44.057464599609375, 6.9138336181640625, 9.928482055664062, -7.933563232421875, 100.67842102050781, 148.15695190429688, -11.938566207885742, -19.168685913085938, -11.24932861328125, 0.8150310516357422, -28.17742919921875, 150.1498260498047, 155.44052124023438, 1.4213027954101562, 83.8599853515625, 120.26351928710938, 145.98446655273438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000393.npy"} {"epoch": 0.8230366492146597, "step": 394, "batch_size": 128, "mean": 49.8526611328125, "std": 82.49048614501953, "min": -153.3340606689453, "p10": -43.032711791992185, "median": 30.462646484375, "p90": 165.2355728149414, "max": 217.28765869140625, "pos_frac": 0.6796875, "sample": [164.28427124023438, 49.810302734375, 168.91024780273438, 126.1429443359375, 46.9775390625, 139.10595703125, -13.085205078125, -6.453559875488281, 22.029541015625, 170.44744873046875, 52.4493408203125, -14.403472900390625, 34.654052734375, 87.630126953125, -19.26434326171875, -2.806049346923828, 156.39077758789062, -51.05742645263672, 164.5771484375, -22.623592376708984, 8.360076904296875, 151.66757202148438, 202.40762329101562, -17.25775146484375, 0.8554763793945312, -79.90219116210938, 8.18475341796875, 100.08146667480469, 86.556640625, 63.08154296875, -34.76786804199219, 0.0, -44.68871307373047, -48.94517517089844, 165.2418212890625, 13.056289672851562, 16.6754150390625, 111.6798324584961, 4.09429931640625, 38.30018615722656, 70.408203125, 99.27583312988281, 46.546600341796875, -153.3340606689453, 19.15056610107422, 120.3260498046875, 162.75411987304688, 35.28554153442383, 2.125732421875, 16.3680419921875, -3.695587158203125, -42.0662841796875, 58.809417724609375, -14.4571533203125, -21.91180419921875, 175.95802307128906, 30.103469848632812, 86.72369384765625, 10.103927612304688, 55.175537109375, 62.880401611328125, -52.958251953125, -0.26238250732421875, -36.51858901977539, 161.33743286132812, 2.5556869506835938, 25.631675720214844, 106.579345703125, 25.89019012451172, 119.30828857421875, 151.7062530517578, 165.23289489746094, 174.468994140625, 3.8191967010498047, 199.7392578125, -3.44671630859375, -62.592132568359375, -31.2447509765625, -7.3701934814453125, 217.28765869140625, 9.282268524169922, -121.26336669921875, 82.96636962890625, -14.2156982421875, -42.600830078125, 19.2821044921875, 40.28125, 30.821823120117188, 5.9841766357421875, 76.85684204101562, -0.271087646484375, -3.3296966552734375, 100.30612182617188, 21.784034729003906, -24.307586669921875, -122.88217163085938, 83.08633422851562, 44.3516845703125, 98.46665954589844, 71.084228515625, -124.59501647949219, 148.18309020996094, 29.97528076171875, -28.489700317382812, 183.0615997314453, 207.736328125, -26.631484985351562, 43.13128662109375, 122.89714050292969, 170.1416015625, 118.94418334960938, 132.38992309570312, 157.49417114257812, 145.45693969726562, 163.91043090820312, -44.040435791015625, 198.4058837890625, 20.962646484375, -25.139892578125, 170.77371215820312, 4.21343994140625, 158.70266723632812, -56.12432861328125, 131.15646362304688, -46.97541046142578, -13.773757934570312, -36.878936767578125, 146.47756958007812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000394.npy"} {"epoch": 0.8251308900523561, "step": 395, "batch_size": 128, "mean": 52.472930908203125, "std": 81.13288116455078, "min": -141.719970703125, "p10": -33.501316833496084, "median": 27.064292907714844, "p90": 168.6466110229492, "max": 204.84619140625, "pos_frac": 0.75, "sample": [52.083740234375, 3.82086181640625, -24.86504364013672, -7.6102142333984375, -7.15167236328125, -15.69732666015625, 3.1864013671875, 199.8868408203125, 190.6231689453125, 23.768295288085938, 19.454139709472656, 69.09054565429688, 29.1923828125, 16.7943115234375, 1.5368022918701172, 54.18659973144531, 0.45047760009765625, -71.78094482421875, 72.2177734375, 19.56512451171875, 13.261016845703125, 155.73049926757812, -19.713775634765625, -58.962188720703125, 177.64260864257812, 121.03005981445312, 11.179641723632812, -0.9828529357910156, 132.91629028320312, 37.38493347167969, 131.48968505859375, 126.88577270507812, 111.18942260742188, 110.07855224609375, 29.622711181640625, 165.8262939453125, 73.803466796875, 47.96551513671875, 0.20114898681640625, -12.303199768066406, 92.09939575195312, 12.798858642578125, -13.7509765625, -18.662078857421875, 158.34951782226562, 120.51608276367188, -11.4488525390625, -82.00145721435547, 132.760009765625, 171.68972778320312, 186.993408203125, -77.06787109375, 64.91111755371094, 145.372314453125, 166.6009521484375, 103.83444213867188, 107.63882446289062, 168.01589965820312, 16.36395263671875, 109.70204162597656, -41.67816162109375, -85.37725830078125, 20.948715209960938, -17.797042846679688, 155.2425537109375, -24.42449951171875, 20.05683135986328, 78.53895568847656, -16.240936279296875, 10.0181884765625, -30.391998291015625, 12.853019714355469, 5.393280029296875, -42.66571044921875, 31.157440185546875, 131.7687225341797, 138.1686553955078, -2.2889328002929688, 36.609832763671875, 4.786376953125, 23.455657958984375, 20.892250061035156, 204.84619140625, 166.86883544921875, 161.88363647460938, -21.629852294921875, 5.742919921875, -73.42352294921875, 0.7923583984375, 36.860443115234375, 8.379261016845703, 56.41749572753906, -129.33499145507812, 102.79902648925781, -12.426055908203125, 20.851226806640625, 178.5458984375, 179.9645233154297, 172.98265075683594, 166.69342041015625, -9.563835144042969, 113.14541625976562, 27.361114501953125, -40.75639343261719, 120.51998901367188, 69.84048461914062, 159.69898986816406, 7.524658203125, 152.6746826171875, -125.92866516113281, -2.496124267578125, 142.46078491210938, 178.8837890625, -67.2691650390625, 9.448760986328125, -141.719970703125, 32.86187744140625, 5.952175140380859, 177.191650390625, 175.66049194335938, 162.96795654296875, 170.11827087402344, 22.33197021484375, 8.977630615234375, 26.767471313476562, 80.02378845214844, 7.9616241455078125, 58.343505859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000395.npy"} {"epoch": 0.8272251308900523, "step": 396, "batch_size": 128, "mean": 52.53191375732422, "std": 86.39521789550781, "min": -197.44491577148438, "p10": -47.17958068847656, "median": 46.053192138671875, "p90": 163.3347213745117, "max": 218.98583984375, "pos_frac": 0.703125, "sample": [127.83557891845703, 17.06121826171875, -8.3468017578125, 167.3243408203125, 148.5545654296875, 16.512374877929688, -6.686927795410156, 17.56732940673828, -41.02001953125, 179.82406616210938, 86.04058837890625, -22.812271118164062, 105.30406188964844, 44.844085693359375, 1.6889190673828125, 83.32516479492188, 183.71755981445312, -80.76202392578125, 7.059562683105469, 123.66796875, -32.8248291015625, -12.962776184082031, -11.33099365234375, 218.98583984375, -5.4709014892578125, 33.375091552734375, 64.22003173828125, 137.06661987304688, 97.46522521972656, 126.22187042236328, 74.5728759765625, -114.1185302734375, 69.54376220703125, 135.40512084960938, -123.97958374023438, 166.04995727539062, 11.527091979980469, -2.25555419921875, -6.074176788330078, 96.59178161621094, 28.34991455078125, 137.59375, 75.9893798828125, 136.52952575683594, 181.80621337890625, -46.2950439453125, 12.11440658569336, 9.350067138671875, -5.0084228515625, 131.70953369140625, 191.60333251953125, 181.40231323242188, 213.20904541015625, 6.061439514160156, 161.91818237304688, 88.90200805664062, 148.9593505859375, 10.832496643066406, 15.474609375, 25.129480361938477, 8.131156921386719, -23.61669921875, -57.75048828125, -70.32369995117188, 113.48309326171875, -80.85975646972656, 51.336273193359375, -139.68145751953125, 12.096298217773438, 50.7093505859375, 141.33944702148438, -6.260009765625, -24.800857543945312, -5.200569152832031, 145.06826782226562, 102.30758666992188, 61.696380615234375, 137.64474487304688, 22.79705810546875, -151.0796356201172, 64.01727294921875, 113.89163970947266, 41.016746520996094, 135.96499633789062, 55.424560546875, 164.5687255859375, 22.56740951538086, 185.25242614746094, -88.86729431152344, 153.069091796875, -24.1287841796875, 87.79412841796875, -49.243499755859375, 72.64737701416016, 54.40424346923828, 115.40066528320312, 133.940673828125, -1.743011474609375, 17.02166748046875, 44.23291015625, 162.8058624267578, 105.34336853027344, -0.122467041015625, -197.44491577148438, -25.000030517578125, 160.53968811035156, -9.775680541992188, 178.01031494140625, 157.32611083984375, 169.6223602294922, 0.0, 20.7342529296875, 153.46847534179688, 39.243682861328125, 20.22174072265625, 9.083976745605469, 0.0, 47.262298583984375, 99.01478576660156, -17.914077758789062, 62.487152099609375, -132.95721435546875, 73.20230102539062, 156.5561065673828, 146.21139526367188, -96.75753784179688, -36.36732482910156, 116.68321228027344], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000396.npy"} {"epoch": 0.8293193717277487, "step": 397, "batch_size": 128, "mean": 50.62960433959961, "std": 86.58203887939453, "min": -176.0137939453125, "p10": -54.06404571533202, "median": 39.174896240234375, "p90": 165.1750701904297, "max": 202.97967529296875, "pos_frac": 0.7109375, "sample": [-8.224853515625, 7.3665771484375, -12.090553283691406, 20.85710906982422, 172.060546875, 26.475250244140625, -97.73233795166016, 48.84423828125, -5.11962890625, 18.22869873046875, 8.505462646484375, 14.85321044921875, 73.13848114013672, 14.79681396484375, 26.955184936523438, 166.93728637695312, 54.42924499511719, 109.17013549804688, 198.383544921875, 134.52764892578125, -134.0941162109375, 24.91278076171875, -20.07122802734375, 71.17208862304688, 126.83611297607422, 152.733154296875, 0.0, 1.60430908203125, 126.245361328125, 40.62640380859375, 0.0, -52.1671142578125, 137.43328857421875, -11.271408081054688, -3.95208740234375, 2.9346466064453125, 10.798828125, 154.9403839111328, -73.67727661132812, 143.17446899414062, 174.08810424804688, 63.36125946044922, 105.90301513671875, 165.78665161132812, -29.776275634765625, 7.611976623535156, 135.42724609375, 151.28585815429688, 202.97967529296875, -41.3345947265625, 200.06076049804688, -39.94952392578125, 181.32876586914062, -176.0137939453125, -107.05645751953125, -41.217620849609375, 46.93817138671875, -126.26974487304688, -5.037910461425781, -149.85150146484375, -35.09747314453125, 85.16398620605469, 164.9129638671875, 139.5572509765625, 92.75987243652344, 28.259639739990234, 132.66757202148438, -18.199676513671875, 34.272003173828125, -1.5716361999511719, 54.32720947265625, 188.77691650390625, 77.84085083007812, 96.057861328125, 146.52755737304688, 148.0987548828125, 137.53707885742188, -68.76068115234375, 89.36872863769531, 38.676055908203125, 148.90406799316406, 31.06231689453125, 172.6796875, 169.84902954101562, 39.673736572265625, 1.159311294555664, 142.8077850341797, -5.2979736328125, 11.291580200195312, 20.192108154296875, 143.6322021484375, 16.009628295898438, 98.00093078613281, 12.574073791503906, 150.73046875, 143.04275512695312, 23.373676300048828, 27.380531311035156, 169.29345703125, 100.80120849609375, 17.79681396484375, -81.8472900390625, 39.955230712890625, -5.634426116943359, 181.40708923339844, 164.71072387695312, 126.02568054199219, 5.35369873046875, -13.842628479003906, -19.00457763671875, -25.978042602539062, 160.9122314453125, 130.39816284179688, -72.72808837890625, 52.20947265625, 39.8992919921875, -125.7276611328125, -12.7298583984375, 140.43113708496094, 24.194259643554688, -40.150604248046875, -58.49021911621094, 65.6322021484375, 104.58736419677734, 64.37603759765625, -121.40853881835938, 118.1812744140625, 84.92037963867188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000397.npy"} {"epoch": 0.831413612565445, "step": 398, "batch_size": 128, "mean": 39.04682922363281, "std": 93.49176025390625, "min": -168.19293212890625, "p10": -69.10674133300779, "median": 23.63241958618164, "p90": 164.8377243041992, "max": 243.8153076171875, "pos_frac": 0.6640625, "sample": [15.16754150390625, 17.851593017578125, 148.5706329345703, -8.7890625, 243.8153076171875, 5.3647308349609375, 57.8887939453125, -51.553955078125, 92.42770385742188, 165.00057983398438, -140.22216796875, -57.64178466796875, -161.40570068359375, 23.165313720703125, 7.042320251464844, -28.47216796875, 37.326202392578125, -4.006317138671875, 161.52481079101562, 171.57122802734375, 164.76792907714844, -136.28536987304688, -111.96337890625, 134.54745483398438, 37.16278076171875, 116.60458374023438, 20.956451416015625, 140.13192749023438, 151.84945678710938, 59.263763427734375, -108.68099975585938, 6.579233169555664, 36.772369384765625, -168.19293212890625, -62.74603271484375, 138.1103057861328, 5.99041748046875, 132.66854858398438, 3.0621337890625, -117.7435531616211, 2.045370101928711, -83.94839477539062, 24.73798370361328, 131.97586059570312, -0.6410655975341797, -49.3341064453125, 0.2604217529296875, 135.23170471191406, 21.447052001953125, -8.770172119140625, 45.14501953125, -18.397064208984375, 80.63298034667969, -29.880416870117188, 129.1378936767578, 41.7313232421875, 18.300268173217773, -130.54061889648438, 25.845352172851562, -15.224822998046875, 25.163421630859375, 40.591094970703125, 163.64169311523438, -35.86322021484375, 179.726806640625, 5.387542724609375, 24.099525451660156, -137.15911865234375, 122.91043090820312, 172.45155334472656, 24.490249633789062, 197.27801513671875, 34.48338317871094, 207.89871215820312, -16.08380126953125, -18.9974365234375, 9.624797821044922, 163.5616455078125, -29.78076171875, -14.935943603515625, 175.36056518554688, 176.92965698242188, 3.9179229736328125, 32.66716003417969, 40.331329345703125, 7.0836181640625, -48.24369812011719, 143.12779235839844, 52.2027587890625, -49.594818115234375, 132.87191772460938, 106.47378540039062, -11.370559692382812, -150.46011352539062, 77.2601318359375, -7.707855224609375, 157.09759521484375, -24.802032470703125, -11.418212890625, 90.02227783203125, -17.259933471679688, -11.852996826171875, 140.86578369140625, 131.5646514892578, 13.288116455078125, 14.306594848632812, 87.03480529785156, 167.894775390625, 159.70278930664062, -33.107086181640625, -93.61823272705078, 182.8466796875, 142.013671875, -58.548927307128906, 159.8897705078125, -4.8272552490234375, -48.80873107910156, -167.89697265625, 14.451950073242188, 168.37161254882812, 123.61981201171875, -7.9344482421875, 7.022180557250977, 215.38607788085938, 55.022216796875, 32.1234130859375, 117.96334838867188, 41.007354736328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000398.npy"} {"epoch": 0.8335078534031414, "step": 399, "batch_size": 128, "mean": 48.13529968261719, "std": 82.3637466430664, "min": -170.49374389648438, "p10": -37.89505615234375, "median": 31.18694305419922, "p90": 163.80902709960935, "max": 206.79226684570312, "pos_frac": 0.7109375, "sample": [-18.633041381835938, 141.84303283691406, 142.63577270507812, 147.12680053710938, 9.6385498046875, -39.577056884765625, 0.716552734375, 37.383758544921875, -11.78424072265625, 174.50137329101562, 75.92086029052734, -48.73468017578125, -24.939544677734375, 153.4433135986328, -2.7901077270507812, 39.352996826171875, 104.31935119628906, 3.868682861328125, 151.4818572998047, 39.86714172363281, -3.340545654296875, -99.591796875, -19.187347412109375, 162.06866455078125, 12.97137451171875, -1.72186279296875, 26.592880249023438, -131.3857421875, -170.49374389648438, 129.55296325683594, -3.840789794921875, 166.6324462890625, 17.121944427490234, 174.32318115234375, 18.324180603027344, -9.131362915039062, 148.4595947265625, -7.628349304199219, 10.093505859375, 182.1326904296875, 35.781005859375, 160.52308654785156, 206.79226684570312, 155.446044921875, 23.93121337890625, -11.823097229003906, 162.61285400390625, -60.455841064453125, 4.244985580444336, 146.62799072265625, -37.602691650390625, 37.60718536376953, 102.3953857421875, 103.64190673828125, 185.9930419921875, -7.92926025390625, 40.09977340698242, 10.14276123046875, 14.703948974609375, -12.858524322509766, 56.30128479003906, 166.60009765625, 4.3381195068359375, 24.202499389648438, 152.46484375, 37.81760787963867, 149.68592834472656, 21.20684814453125, 96.26872253417969, 43.76458740234375, 64.32283020019531, 121.51473999023438, -11.672515869140625, -127.50971984863281, -151.04104614257812, -139.70750427246094, -13.10595703125, 35.90625, 167.0950927734375, 169.48794555664062, 21.523712158203125, 144.08465576171875, 71.76171875, 140.20684814453125, 125.96768188476562, 62.252655029296875, 12.69140625, -12.001998901367188, 78.0595703125, 83.40673828125, 13.139450073242188, -38.577239990234375, 22.209854125976562, 98.8680419921875, 8.557144165039062, 17.954559326171875, -57.7451171875, -0.4920692443847656, 170.89849853515625, 10.132843017578125, 74.54290771484375, 2.60870361328125, 53.218292236328125, -57.10919189453125, 1.816925048828125, 167.37246704101562, -19.032989501953125, 196.67022705078125, 117.15316772460938, 57.58599853515625, 176.77598571777344, -120.48117065429688, -34.509735107421875, 40.821441650390625, 24.24473762512207, 23.481460571289062, 107.71293640136719, 113.6873779296875, -8.185470581054688, 140.2047119140625, -23.70916748046875, 37.55726623535156, 123.99234008789062, 87.96139526367188, -26.507781982421875, -1.785400390625, 107.55879211425781, 19.365005493164062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000399.npy"} {"epoch": 0.8356020942408376, "step": 400, "batch_size": 128, "mean": 55.14772033691406, "std": 84.36236572265625, "min": -179.0472412109375, "p10": -44.303045654296874, "median": 38.06087112426758, "p90": 163.30621490478515, "max": 202.84506225585938, "pos_frac": 0.734375, "sample": [6.1013336181640625, -55.65242004394531, 3.014984130859375, 39.399322509765625, 65.2783203125, 14.823417663574219, 122.58267974853516, 120.06349182128906, 58.935882568359375, -98.37439727783203, 4.732582092285156, 84.02255249023438, -179.0472412109375, 42.882568359375, 168.9517059326172, 15.794143676757812, -13.254638671875, 134.12362670898438, 126.62664794921875, 155.9239959716797, 193.27838134765625, -0.191650390625, 23.223934173583984, 63.588623046875, 157.63729858398438, 85.46725463867188, -44.4144287109375, 147.1663360595703, -21.47528076171875, 109.735107421875, 55.441070556640625, 9.069656372070312, 18.70044708251953, 127.99615478515625, 44.0784912109375, 166.77777099609375, 168.206298828125, 106.3446044921875, 22.553070068359375, 13.413253784179688, -135.63339233398438, 144.69708251953125, 127.16134643554688, 163.02658081054688, 22.158782958984375, 158.09231567382812, -2.678131103515625, 106.54898071289062, 8.6903076171875, 16.894805908203125, 36.09442138671875, 137.77386474609375, 171.44287109375, -120.26336669921875, 0.36871337890625, -29.277488708496094, 6.8525896072387695, 36.72241973876953, 178.19967651367188, -1.855813980102539, -1.4340782165527344, 49.480377197265625, 161.50949096679688, 139.38995361328125, 29.575714111328125, -26.65020751953125, 21.09710693359375, 153.28433227539062, 69.59814453125, 172.99839782714844, -46.568878173828125, 4.025306701660156, -25.001113891601562, 23.935272216796875, -44.25531005859375, -81.76300048828125, 161.54403686523438, 172.42539978027344, 2.16925048828125, -46.757598876953125, 99.71249389648438, 55.189910888671875, 123.4473876953125, 118.66738891601562, -3.393556594848633, 161.88812255859375, 134.46298217773438, -14.853057861328125, 32.97532653808594, 52.537132263183594, 117.9441146850586, -11.594482421875, -6.9802703857421875, -4.603384017944336, 3.295806884765625, 150.875, -32.411285400390625, 126.87776184082031, 2.8501453399658203, -107.04580688476562, 53.094146728515625, 172.1820831298828, 131.63214111328125, 24.442901611328125, -34.6893310546875, 16.05517578125, 131.71066284179688, -7.528076171875, 4.16107177734375, 163.9586944580078, 143.9595184326172, 3.8468856811523438, 152.55075073242188, -13.1278076171875, 153.35641479492188, 160.73899841308594, 151.60838317871094, -17.495025634765625, 202.84506225585938, 178.90338134765625, 169.93312072753906, 155.3330078125, 0.0, -73.283447265625, -60.577392578125, 3.851104736328125, -132.92547607421875, 47.385528564453125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000400.npy"} {"epoch": 0.837696335078534, "step": 401, "batch_size": 128, "mean": 66.28703308105469, "std": 81.91929626464844, "min": -209.3651123046875, "p10": -21.462426757812498, "median": 59.98199462890625, "p90": 166.93818359374998, "max": 234.362548828125, "pos_frac": 0.78125, "sample": [0.701629638671875, 6.332061767578125, 27.0333251953125, -9.317230224609375, 19.65966796875, 108.25982666015625, 22.8577880859375, 120.78011322021484, 88.63470458984375, -5.982421875, 149.87576293945312, 25.073211669921875, 45.89115905761719, 164.468994140625, 77.8702392578125, 31.855941772460938, 81.4954833984375, 141.60299682617188, -4.877227783203125, 133.7726287841797, -6.491729736328125, 61.51202392578125, -19.45012664794922, 81.382080078125, -6.013763427734375, 164.60757446289062, 90.546630859375, 1.6463031768798828, 150.06689453125, 53.67596435546875, 177.48077392578125, 205.27114868164062, 142.0480499267578, 59.72149658203125, 15.716064453125, 16.2176513671875, 130.9125213623047, 138.65890502929688, 10.560844421386719, 141.71902465820312, 234.362548828125, -18.39764404296875, 158.6536102294922, 172.71563720703125, 184.41091918945312, 150.4403076171875, 107.03649139404297, 15.36785888671875, 148.090576171875, 113.59268188476562, -93.0223388671875, 147.47976684570312, 73.41302490234375, 9.112297058105469, -209.3651123046875, 105.0487060546875, 11.3175048828125, 156.58154296875, 107.43101501464844, -38.38017272949219, -48.09356689453125, 136.35858154296875, 180.615234375, 141.66207885742188, 92.10678100585938, 85.144775390625, -86.88616943359375, -102.59803009033203, 157.59478759765625, -19.9813232421875, -2.300994873046875, 2.6211318969726562, 196.65576171875, 55.0892333984375, 35.38995361328125, 4.0989837646484375, 156.8744354248047, 90.3096923828125, 50.544281005859375, 65.07814025878906, 82.491455078125, -23.68206787109375, 147.36672973632812, 168.6217041015625, 163.30691528320312, 143.73406982421875, 51.877899169921875, -3.797088623046875, 60.24249267578125, -22.754852294921875, 91.52838134765625, 23.60693359375, -8.31268310546875, -21.235137939453125, 26.751758575439453, -24.9906005859375, -27.833831787109375, 177.09487915039062, 202.88059997558594, 25.21240234375, 40.47515869140625, 1.762237548828125, 170.15603637695312, 159.37191772460938, 174.289794921875, 21.983306884765625, 11.036376953125, 42.813690185546875, -22.599632263183594, 62.040496826171875, 31.026031494140625, 2.443511962890625, 36.59187316894531, 16.145801544189453, -21.992767333984375, -2.2846450805664062, 96.18099975585938, 142.32310485839844, 115.05961608886719, -21.220558166503906, 125.99397277832031, 154.9917755126953, 160.8272705078125, -4.2272491455078125, 197.8020782470703, -173.31814575195312, 6.7912139892578125, 166.2166748046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000401.npy"} {"epoch": 0.8397905759162304, "step": 402, "batch_size": 128, "mean": 52.98896026611328, "std": 85.94712829589844, "min": -155.85330200195312, "p10": -41.38859748840331, "median": 42.094871520996094, "p90": 161.41001739501954, "max": 236.859619140625, "pos_frac": 0.75, "sample": [-5.5254058837890625, -4.3943023681640625, -129.64669799804688, 47.932281494140625, 151.81089782714844, -19.1270751953125, -78.07345581054688, -31.01312255859375, 106.31765747070312, -155.85330200195312, -136.02490234375, 135.41326904296875, 13.96319580078125, -45.25307846069336, 5.050575256347656, 213.44210815429688, 142.56268310546875, 131.34664916992188, 45.27647399902344, -29.0694580078125, 139.71719360351562, 2.1605072021484375, 12.568382263183594, 182.1619873046875, -134.0662384033203, 3.561431884765625, 1.234527587890625, 145.7012939453125, -153.95840454101562, 32.926116943359375, 136.27978515625, 2.69512939453125, 66.41790771484375, 120.8092041015625, 8.52850341796875, 0.0, 213.002685546875, 156.36453247070312, -16.042236328125, 145.34698486328125, 89.19937133789062, 30.5648193359375, 138.72561645507812, 13.326461791992188, 110.4386215209961, 149.36383056640625, -98.84088134765625, -10.190155029296875, 174.3551025390625, 134.98477172851562, 26.71258544921875, -87.8016357421875, -37.07904052734375, 74.15054321289062, 123.12393188476562, 30.037643432617188, 123.16496276855469, 99.11729431152344, 3.4267654418945312, 71.5085678100586, 11.733177185058594, 196.37451171875, 162.587646484375, 13.174118041992188, 123.23430633544922, -23.038375854492188, -106.11811828613281, 68.05812072753906, 14.372222900390625, -141.49227905273438, 162.7388916015625, 148.72076416015625, 50.28596496582031, -2.41510009765625, 139.1811981201172, -5.0811004638671875, 236.859619140625, 3.8257408142089844, 68.41888427734375, 170.13937377929688, 158.1953125, 160.9053192138672, 30.1260986328125, 92.71698760986328, 88.8460693359375, 39.514373779296875, 167.9947509765625, 116.98504638671875, -49.401885986328125, 127.88067626953125, 72.52794647216797, 45.4755859375, 11.949073791503906, -0.741607666015625, 44.45062255859375, 109.77813720703125, 6.236946105957031, 41.1142578125, 9.7738037109375, 158.9982452392578, 154.27191162109375, 163.20184326171875, 137.5916748046875, -27.815261840820312, 198.8797607421875, -3.1501312255859375, 43.07548522949219, 12.898162841796875, 28.50372314453125, 166.55215454101562, 14.107704162597656, 134.73681640625, 94.1827392578125, 87.9014892578125, -49.18572998046875, 6.3314208984375, -0.8535614013671875, 51.452484130859375, 34.302947998046875, 6.519500732421875, 101.62652587890625, 0.9717159271240234, 8.821683883666992, -39.732391357421875, -11.6541748046875, 46.5220947265625, 116.62086486816406, -1.8134078979492188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000402.npy"} {"epoch": 0.8418848167539267, "step": 403, "batch_size": 128, "mean": 73.11346435546875, "std": 76.15779113769531, "min": -134.47772216796875, "p10": -15.768707275390621, "median": 59.1937255859375, "p90": 167.2520782470703, "max": 224.16363525390625, "pos_frac": 0.8046875, "sample": [-26.721237182617188, 130.10977172851562, -7.5120849609375, 160.37892150878906, 129.58847045898438, 55.087646484375, 160.00344848632812, 93.210693359375, 25.12933349609375, 26.963134765625, 163.75341796875, 162.9857177734375, 46.52435302734375, 36.9324951171875, 10.819007873535156, 80.59481811523438, 20.873672485351562, 125.66537475585938, 37.2711181640625, 14.46636962890625, 165.7192840576172, -18.06365203857422, -34.76495361328125, 31.50592041015625, 163.371337890625, 138.5102081298828, 169.7698974609375, 45.5557861328125, 100.25982666015625, 182.21151733398438, 198.32928466796875, 11.67950439453125, -0.00572967529296875, 122.36119842529297, -2.568258285522461, 92.34334564208984, 15.453750610351562, 31.02001953125, 0.0, -14.941802978515625, -0.2280750274658203, 154.04652404785156, 196.71405029296875, 99.44036865234375, 160.67584228515625, 160.83367919921875, 152.82144165039062, 21.2542724609375, 140.75967407226562, 200.79840087890625, 12.84246826171875, 45.828895568847656, 15.0029296875, 21.21938705444336, 172.141357421875, 3.717620849609375, 28.536224365234375, -51.5650634765625, 99.5482177734375, -45.14225769042969, -18.868194580078125, 60.07366943359375, 0.2393951416015625, 177.6011962890625, 147.5121612548828, 18.4287109375, 191.751953125, 1.1728096008300781, 33.872230529785156, 25.845184326171875, 33.69439697265625, 79.46484375, 12.690475463867188, 129.4317626953125, -13.61181640625, 45.94573974609375, 119.53863525390625, 166.39352416992188, 4.475410461425781, 78.7518310546875, -1.01251220703125, 58.31378173828125, 178.543212890625, 16.36090087890625, 133.73077392578125, -17.698150634765625, 4.6363983154296875, 144.43563842773438, 130.50677490234375, 83.57843017578125, 220.70477294921875, 98.557373046875, 224.16363525390625, 169.25537109375, 80.53498840332031, 177.1038818359375, -30.197357177734375, 134.97662353515625, 151.677490234375, 128.19900512695312, 126.8939208984375, 56.04973602294922, 122.17414855957031, 47.39892578125, -134.47772216796875, 164.60276794433594, -25.983856201171875, 153.7032470703125, -1.4087448120117188, 142.82833862304688, -54.889434814453125, -2.4250717163085938, 27.70166015625, 79.53671264648438, -30.2996826171875, 35.39131164550781, 67.54055786132812, -0.152313232421875, 116.91433715820312, -8.109115600585938, -118.5972900390625, 148.41163635253906, 151.12954711914062, 33.903717041015625, 135.17738342285156, 129.39077758789062, 135.91543579101562, 46.31134033203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000403.npy"} {"epoch": 0.8439790575916231, "step": 404, "batch_size": 128, "mean": 51.1851806640625, "std": 90.0434341430664, "min": -203.00128173828125, "p10": -41.224467468261714, "median": 38.45979309082031, "p90": 170.03131713867188, "max": 229.317626953125, "pos_frac": 0.703125, "sample": [172.99127197265625, 158.94833374023438, 36.3590087890625, 176.1446990966797, -136.48724365234375, 160.17745971679688, 182.7691650390625, 3.7687530517578125, -84.32872772216797, 0.0, -148.28164672851562, 70.43982696533203, 26.28154754638672, -21.19732666015625, -82.28182983398438, 143.9030303955078, 10.141693115234375, 171.36676025390625, 1.9536323547363281, -50.769287109375, 163.87612915039062, 53.12040710449219, 21.72711181640625, -2.7813720703125, -14.912261962890625, 161.99374389648438, -203.00128173828125, 155.43096923828125, -44.30415344238281, 8.403221130371094, -3.05389404296875, -39.90460205078125, 43.944305419921875, 11.09051513671875, 101.39155578613281, -9.889045715332031, 3.7220115661621094, -6.181396484375, 229.317626953125, 120.813720703125, 153.72726440429688, -3.795989990234375, -8.67193603515625, -104.42984008789062, 143.02593994140625, 75.30508422851562, 67.6895751953125, 31.931396484375, 4.885250091552734, 18.586944580078125, 101.05760192871094, 7.670616149902344, 165.42074584960938, -26.15972900390625, 43.560699462890625, 188.26129150390625, -134.78390502929688, 16.85748291015625, -13.63189697265625, 17.108810424804688, 197.85452270507812, 178.99449157714844, -130.37384033203125, 139.05191040039062, 93.92793273925781, 144.72402954101562, 148.25067138671875, 5.20062255859375, -7.252227783203125, 154.27743530273438, 144.4229736328125, 3.342977523803711, 11.91351318359375, -6.813505172729492, -39.6339111328125, 127.4613037109375, 6.42645263671875, -16.20806884765625, 184.38824462890625, 40.560577392578125, 84.15682983398438, 29.245651245117188, 47.859039306640625, 64.11163330078125, 11.820892333984375, -31.246017456054688, -17.072845458984375, -15.69439697265625, 176.47409057617188, -107.3892822265625, -5.336341857910156, 126.37905883789062, 154.70614624023438, 43.713623046875, -137.76947021484375, -8.202789306640625, 75.38113403320312, 10.268539428710938, 166.71121215820312, 71.76031494140625, -10.649126052856445, -0.433685302734375, 116.23422241210938, 57.6595458984375, 169.458984375, 100.53924560546875, 46.69763946533203, 60.36698913574219, -120.85620880126953, 179.84970092773438, 48.78082275390625, 193.17593383789062, 154.36932373046875, 55.79341125488281, 26.19512939453125, 1.0875701904296875, 64.63398742675781, -11.212024688720703, 82.13912963867188, 145.76564025878906, 27.276031494140625, -18.486618041992188, 9.205230712890625, 137.50927734375, 143.88262939453125, 123.30152893066406, 167.66015625, 195.048095703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000404.npy"} {"epoch": 0.8460732984293193, "step": 405, "batch_size": 128, "mean": 59.19466781616211, "std": 81.44469451904297, "min": -151.36212158203125, "p10": -34.402978515625, "median": 42.1649055480957, "p90": 174.28792572021484, "max": 217.86962890625, "pos_frac": 0.7578125, "sample": [16.32806396484375, 4.221588134765625, 10.5439453125, 0.0, -39.10052490234375, -4.329925537109375, 9.052734375, 144.2363739013672, 104.41806030273438, 153.6212158203125, 2.791839599609375, 16.673553466796875, 112.37681579589844, 46.709320068359375, 142.43756103515625, -69.46534729003906, -151.36212158203125, 21.107650756835938, 42.12053680419922, 185.9437255859375, 57.223697662353516, 157.82464599609375, -1.6633720397949219, -14.373489379882812, 140.33303833007812, 11.309890747070312, -15.361221313476562, 75.71707153320312, 174.2144317626953, 29.668800354003906, 129.7068328857422, 145.61996459960938, 6.64849853515625, -10.414581298828125, 60.12792205810547, 143.4272003173828, 42.20927429199219, -84.91592407226562, -48.612701416015625, 123.06300354003906, -2.469435691833496, 187.30296325683594, 140.38438415527344, 0.6483020782470703, -11.5380859375, 217.86962890625, 146.97909545898438, 106.43104553222656, 97.28109741210938, 69.28012084960938, 7.87066650390625, 82.99932861328125, 174.87716674804688, 38.358367919921875, 8.443771362304688, 0.41650390625, 132.66177368164062, 157.12115478515625, 138.3101043701172, -34.273651123046875, -76.84636688232422, 88.00852966308594, 70.03326416015625, -8.901092529296875, 13.775604248046875, -6.286186218261719, 62.964935302734375, -5.491485595703125, 20.49444580078125, 205.6538848876953, 52.59205627441406, 141.11505126953125, 156.83731079101562, 112.963623046875, 188.0316162109375, 198.99093627929688, -34.704742431640625, 122.72551727294922, 8.605125427246094, 44.15191650390625, 199.14288330078125, 11.680728912353516, -31.74932861328125, -86.412841796875, 141.93569946289062, 23.53957748413086, 0.1745147705078125, 193.40823364257812, 31.242324829101562, 92.51390075683594, 143.34300231933594, 25.862205505371094, -87.30923461914062, 8.683074951171875, 168.54095458984375, 174.45941162109375, 12.46828842163086, 12.215023040771484, 155.03079223632812, 124.0871353149414, -73.58680725097656, 169.483642578125, 1.10662841796875, 139.13690185546875, 57.636505126953125, 91.78689575195312, 12.74477767944336, 103.03202819824219, 70.86190795898438, 41.236480712890625, 42.52996826171875, 1.9748611450195312, 175.612548828125, -2.831033706665039, -10.366470336914062, -16.076416015625, -37.31867218017578, 6.216928482055664, -29.997390747070312, 2.7928295135498047, -67.035888671875, 178.62493896484375, 206.29022216796875, -9.0496826171875, -58.137046813964844, 170.183837890625, 95.59820556640625, 117.86422729492188], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000405.npy"} {"epoch": 0.8481675392670157, "step": 406, "batch_size": 128, "mean": 49.567203521728516, "std": 88.73446655273438, "min": -187.46524047851562, "p10": -55.54581909179687, "median": 29.1072998046875, "p90": 158.42743835449218, "max": 209.19021606445312, "pos_frac": 0.71875, "sample": [142.20994567871094, -33.17200469970703, -37.20489501953125, 19.58489990234375, -55.78375244140625, 194.796142578125, 128.14816284179688, 150.96267700195312, 99.17086791992188, 157.74742126464844, -4.825923919677734, 0.6995601654052734, -20.806640625, 172.33859252929688, 56.436767578125, -28.155532836914062, 135.35272216796875, -12.34078598022461, 97.13481140136719, 24.9754638671875, -124.78146362304688, -49.16349792480469, 16.225982666015625, 101.98196411132812, 147.38381958007812, 9.67669677734375, 117.05686950683594, 195.02789306640625, 26.382720947265625, 132.44375610351562, 57.395751953125, 157.89646911621094, -146.52066040039062, -2.44464111328125, 23.02734375, -131.55010986328125, -19.618865966796875, -8.752708435058594, -16.98040008544922, 140.55221557617188, -77.66650390625, -47.45545959472656, -104.828369140625, -55.44384765625, 38.046607971191406, 159.82403564453125, 19.6995849609375, 141.67578125, 129.18502807617188, 153.9112548828125, 0.6482772827148438, 22.027008056640625, 149.9225311279297, 120.18956756591797, -187.46524047851562, -9.968780517578125, 129.9647216796875, 159.26614379882812, 110.98467254638672, 17.5772705078125, 4.07855224609375, 154.86355590820312, -120.46694946289062, 176.9913330078125, 22.494583129882812, 29.077789306640625, 8.351287841796875, 2.524566650390625, 5.976409912109375, 143.5892333984375, 209.19021606445312, 151.94049072265625, 11.56671142578125, 37.0057373046875, 172.72238159179688, 35.879302978515625, 4.1790618896484375, 142.06179809570312, 14.6502685546875, 182.7506103515625, 158.0679931640625, -17.67707061767578, -145.5638427734375, 10.814964294433594, 150.1659393310547, 131.14556884765625, 112.90945434570312, -11.370758056640625, 29.136810302734375, 60.11517333984375, 33.80516052246094, -12.525657653808594, 117.7948226928711, 131.14932250976562, -68.2571029663086, 206.06771850585938, 97.9478759765625, 2.6004409790039062, 129.80416870117188, 47.914794921875, 100.60163879394531, 138.35873413085938, 36.7579345703125, 135.57359313964844, -25.77099609375, 52.573638916015625, -133.05828857421875, 182.40350341796875, 1.45245361328125, 36.9215087890625, -39.927490234375, 98.330078125, -34.436431884765625, 117.83856201171875, 200.61080932617188, -0.6798057556152344, 1.9021186828613281, 29.310684204101562, -9.809356689453125, -2.8149871826171875, -57.154541015625, 21.991973876953125, 22.4222412109375, 11.479034423828125, 174.17477416992188, -56.36248779296875, 87.88127136230469, 17.95746612548828], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000406.npy"} {"epoch": 0.8502617801047121, "step": 407, "batch_size": 128, "mean": 54.57229995727539, "std": 85.04371643066406, "min": -130.63040161132812, "p10": -40.46000671386719, "median": 33.31338882446289, "p90": 168.224951171875, "max": 210.00888061523438, "pos_frac": 0.703125, "sample": [-24.627525329589844, 27.0986328125, 97.9544677734375, -13.841812133789062, 152.8275146484375, 10.305042266845703, 161.56591796875, 62.69097900390625, -32.491302490234375, 14.138238906860352, -2.6324996948242188, -56.706878662109375, -130.63040161132812, -3.840606689453125, -86.83265686035156, 121.7935562133789, -2.45391845703125, -1.7091941833496094, 16.67919921875, -27.195281982421875, 137.94650268554688, -1.8909149169921875, -6.768707275390625, 97.25137329101562, 165.41824340820312, -53.86186218261719, 96.14954376220703, 125.984130859375, -14.190986633300781, -12.47869873046875, -119.4654769897461, 43.36351013183594, -40.309906005859375, 33.704681396484375, 117.12799072265625, 171.1744384765625, 161.73623657226562, 174.0179443359375, -59.62298583984375, -1.19244384765625, 67.94058227539062, 25.801605224609375, 33.990966796875, 5.523094177246094, -36.545867919921875, 190.48648071289062, 66.7522201538086, 123.04988861083984, 187.16448974609375, 12.91961669921875, 61.096038818359375, 143.287841796875, 8.534698486328125, 190.24429321289062, 0.92608642578125, 149.91647338867188, -64.51126861572266, -2.9839706420898438, 25.870147705078125, 105.37124633789062, 139.65440368652344, 27.272125244140625, 11.395416259765625, 32.922096252441406, 161.845947265625, 42.8408203125, -85.53659057617188, -19.0716552734375, 120.76559448242188, 167.86968994140625, 20.002044677734375, -25.833053588867188, 141.37310791015625, 162.724365234375, 179.3525390625, 16.994503021240234, 8.114669799804688, 16.58672332763672, 0.0, 198.6500244140625, 154.1768798828125, 11.730964660644531, 116.80099487304688, -40.81024169921875, 165.51129150390625, -93.4254150390625, 187.79730224609375, 100.19625854492188, 16.175994873046875, 135.4791259765625, 163.5924072265625, 0.90008544921875, -28.40191650390625, 111.53152465820312, 44.16168212890625, 19.53502655029297, -36.956787109375, 36.780517578125, 38.817108154296875, 129.078125, 139.00140380859375, -19.357284545898438, 194.2554931640625, -123.124267578125, 65.66030883789062, -71.73092651367188, 89.9327392578125, 85.90771484375, 167.24998474121094, 6.73602294921875, 148.66793823242188, 169.05389404296875, 52.31201171875, -125.49920654296875, 178.81988525390625, 7.4962921142578125, 38.55171203613281, -29.89925193786621, 9.283454895019531, 169.1737060546875, 145.93955993652344, 148.60256958007812, 23.107391357421875, 8.764810562133789, -3.9781036376953125, 210.00888061523438, 158.70538330078125, 0.0], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000407.npy"} {"epoch": 0.8523560209424084, "step": 408, "batch_size": 128, "mean": 56.61176681518555, "std": 83.88692474365234, "min": -188.45315551757812, "p10": -32.52933349609375, "median": 44.85673522949219, "p90": 163.85850524902344, "max": 193.58074951171875, "pos_frac": 0.75, "sample": [-3.2629470825195312, 9.838104248046875, -69.98668670654297, 90.81634521484375, 17.987564086914062, 157.3416748046875, 89.68739318847656, 184.4495849609375, 167.10723876953125, 150.60861206054688, 1.117095947265625, 19.707733154296875, 101.21284484863281, 119.89724731445312, -66.61495971679688, 153.28713989257812, 20.7012939453125, -10.7630615234375, -5.3263702392578125, 70.55489349365234, 0.854156494140625, 19.40701675415039, 17.397247314453125, -10.689727783203125, 183.7354736328125, 150.13800048828125, 184.267333984375, 168.7783203125, -6.844757080078125, -31.5518798828125, -22.712539672851562, 152.00393676757812, 135.31936645507812, 65.89768981933594, -36.24713134765625, 41.6180305480957, 191.69947814941406, 125.64457702636719, 156.78298950195312, 51.739990234375, -14.905242919921875, -181.04339599609375, -188.45315551757812, 10.42950439453125, 172.9375, 138.07623291015625, 4.323814392089844, 35.36542510986328, -34.81005859375, 155.44320678710938, 26.100311279296875, 122.20855712890625, -19.62689208984375, -9.722763061523438, 82.51145935058594, 47.535400390625, 139.0904541015625, 146.79608154296875, -31.476898193359375, 66.75765991210938, 159.37843322753906, 143.0935821533203, 193.18927001953125, 24.94762420654297, -8.8717041015625, 157.931884765625, -39.10284423828125, -27.632461547851562, 138.73828125, -94.78515625, 170.42575073242188, 12.772994995117188, 136.18777465820312, 14.2705078125, 74.5345458984375, 163.64019775390625, -19.811492919921875, 95.13131713867188, 0.131256103515625, 26.772384643554688, 156.315185546875, 49.80992126464844, -21.990859985351562, 42.268402099609375, -24.09844970703125, 4.2332763671875, 142.47393798828125, 125.80789184570312, 118.3919677734375, -60.57691955566406, 6.447784423828125, 14.044090270996094, 65.41470336914062, 156.09140014648438, 158.95144653320312, -0.7265472412109375, -150.359130859375, 64.7923583984375, 167.15499877929688, 47.445068359375, -49.90696716308594, 159.82676696777344, 17.51959228515625, 60.45109558105469, 135.84335327148438, -8.09246826171875, 16.880775451660156, 14.327117919921875, 8.2554931640625, 48.384429931640625, 55.09233856201172, 11.087974548339844, 23.28601837158203, 0.0, 12.13226318359375, 193.58074951171875, 38.86698913574219, 1.1719913482666016, 123.8084716796875, -42.3876953125, -130.97146606445312, 30.0472412109375, 164.36788940429688, 68.48544311523438, 175.45510864257812, 90.66302490234375, 119.99943542480469, 122.16259765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000408.npy"} {"epoch": 0.8544502617801047, "step": 409, "batch_size": 128, "mean": 58.708656311035156, "std": 91.67491912841797, "min": -160.0455322265625, "p10": -42.990170288085935, "median": 35.316322326660156, "p90": 173.9419967651367, "max": 232.267333984375, "pos_frac": 0.7421875, "sample": [152.21002197265625, 128.61822509765625, 146.60536193847656, 46.367523193359375, 183.2572021484375, -41.2218017578125, 125.06106567382812, 178.676025390625, 3.6146240234375, 39.610015869140625, 18.509910583496094, -45.070343017578125, 19.584091186523438, 4.005126953125, 16.00079345703125, 196.95895385742188, 47.92475128173828, 22.9520263671875, 171.12893676757812, -139.7779541015625, -11.21628189086914, -100.69003295898438, -1.0840692520141602, 167.87400817871094, 135.59286499023438, -96.39013671875, 15.6356201171875, -107.50955200195312, 167.65293884277344, -160.0455322265625, 173.1134033203125, 153.40115356445312, -76.34593963623047, 13.412490844726562, -5.412353515625, 190.89208984375, 22.330429077148438, 39.587646484375, 187.16644287109375, 175.28076171875, 117.2008056640625, 166.10745239257812, -104.16201782226562, 30.585556030273438, 1.63177490234375, 35.295501708984375, 89.53579711914062, 160.02944946289062, 12.10296630859375, 182.45736694335938, 150.7198486328125, 140.95277404785156, 106.8135986328125, 146.43984985351562, 232.267333984375, -42.502166748046875, -44.12884521484375, 66.01376342773438, 22.6038818359375, 11.449539184570312, 213.06463623046875, 122.15521240234375, 17.690460205078125, 64.684814453125, 155.49301147460938, 85.58901977539062, 22.59002685546875, -37.9005126953125, -4.2144012451171875, -37.69078063964844, 155.01290893554688, -35.26713562011719, -127.805908203125, 173.3682403564453, 0.2566566467285156, 149.92965698242188, 186.71258544921875, 213.28155517578125, 50.13190460205078, 28.471466064453125, 7.266162872314453, -16.36777114868164, 116.2762451171875, 0.0, 110.49566650390625, 219.38653564453125, 35.33714294433594, 172.636474609375, 146.1915283203125, -140.73748779296875, -38.07853698730469, 157.19143676757812, 27.06341552734375, -1.6896896362304688, 32.97174072265625, -1.2049560546875, 132.81124877929688, 159.0245361328125, 143.556396484375, 140.64768981933594, 30.0330810546875, -9.738044738769531, 92.509765625, 79.1558837890625, 43.63691711425781, -14.969558715820312, 22.823806762695312, -46.2022705078125, -3.1282196044921875, -11.476676940917969, 37.681541442871094, 4.30451774597168, -157.7491455078125, 32.529815673828125, 146.25828552246094, 93.87228393554688, 10.361089706420898, 114.534912109375, 31.372482299804688, -38.94610595703125, 152.23367309570312, 12.160934448242188, 8.21762466430664, 117.25570678710938, -21.58575439453125, 132.07704162597656, 187.96279907226562, 31.51567840576172], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000409.npy"} {"epoch": 0.856544502617801, "step": 410, "batch_size": 128, "mean": 51.6251335144043, "std": 89.58415222167969, "min": -160.54339599609375, "p10": -39.03197326660156, "median": 41.08435249328613, "p90": 168.82719116210936, "max": 232.87173461914062, "pos_frac": 0.703125, "sample": [46.65887451171875, -61.345458984375, 16.228439331054688, -38.957275390625, -17.024150848388672, 152.39651489257812, 30.105880737304688, 26.108001708984375, -29.089263916015625, 117.28532409667969, 214.07431030273438, -74.1097412109375, 72.46095275878906, -36.97802734375, 43.355934143066406, -151.1085205078125, -15.349029541015625, 46.626953125, 164.19635009765625, 88.319091796875, 173.0611572265625, -160.54339599609375, 67.36328887939453, 12.745376586914062, 156.3316192626953, -6.14569091796875, 127.53921508789062, 84.98812866210938, 119.78903198242188, 138.68336486816406, 27.32781982421875, 76.57533264160156, 9.35858154296875, 151.07363891601562, -0.3938407897949219, 129.8060760498047, 40.34129333496094, 71.68142700195312, 37.908050537109375, 167.01263427734375, 163.32269287109375, 62.270751953125, 0.3900909423828125, 160.88787841796875, -129.80230712890625, 108.69940185546875, -148.16989135742188, 37.330535888671875, -16.730300903320312, -37.14759826660156, 97.70269775390625, 141.08294677734375, 122.01719665527344, -39.206268310546875, -27.87218475341797, 51.413818359375, -5.098968505859375, 151.72515869140625, 128.89979553222656, 47.905548095703125, -36.15143585205078, 189.1943359375, 38.732269287109375, 179.1475830078125, -1.759622573852539, 8.93798828125, -29.32647705078125, 0.76434326171875, 121.49765014648438, 53.4117431640625, 129.9456329345703, 146.78878784179688, 41.82741165161133, -9.21868896484375, -124.91265869140625, 0.0, 38.416404724121094, 25.053634643554688, 189.12939453125, 164.76031494140625, 198.24765014648438, 14.866058349609375, 15.396926879882812, -57.599334716796875, -34.84111785888672, 158.13735961914062, 19.14899444580078, 19.4315185546875, -28.632736206054688, 28.602500915527344, -139.03103637695312, 65.29814147949219, -7.266510009765625, 118.62040710449219, -32.33477783203125, 198.407470703125, -119.53677368164062, 152.91387939453125, 12.069007873535156, 33.95759582519531, 73.83081817626953, -24.929664611816406, 76.87004852294922, -147.2882080078125, 54.858001708984375, 33.95210266113281, 89.08152770996094, 143.22442626953125, 221.276123046875, 126.75459289550781, 176.35321044921875, -19.15264892578125, 115.41702270507812, 97.59820556640625, -1.3583984375, 77.652587890625, 199.75531005859375, -84.18804931640625, 45.71650695800781, 14.489181518554688, 4.376642227172852, 179.57162475585938, 200.81427001953125, 30.10003662109375, 85.12794494628906, -21.750946044921875, 232.87173461914062, -3.0797882080078125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000410.npy"} {"epoch": 0.8586387434554974, "step": 411, "batch_size": 128, "mean": 59.97734451293945, "std": 81.34214782714844, "min": -130.39242553710938, "p10": -19.99105758666992, "median": 35.92977523803711, "p90": 166.88221130371093, "max": 295.78338623046875, "pos_frac": 0.7578125, "sample": [124.90315246582031, 166.60745239257812, 35.5650634765625, 140.11456298828125, 25.36822509765625, -0.3635101318359375, -1.1699485778808594, 36.29448699951172, 57.190826416015625, 136.2344207763672, 146.41708374023438, 101.18899536132812, -2.9910812377929688, -19.7154541015625, -97.660888671875, -70.01535034179688, 201.83554077148438, -17.971099853515625, 129.046142578125, 85.60769653320312, 43.109649658203125, 61.57635498046875, 144.85008239746094, 82.61669921875, 137.0269317626953, 42.650421142578125, 71.63461303710938, 295.78338623046875, 124.66244506835938, -17.99920654296875, 175.35235595703125, 35.32621765136719, 30.434730529785156, 21.618545532226562, 131.75921630859375, -3.434051513671875, 173.16229248046875, 135.2400665283203, 23.00701904296875, 22.49726104736328, 9.617721557617188, 31.5972900390625, -2.962066650390625, 70.46333312988281, 4.4755096435546875, -63.964698791503906, -0.10028076171875, 7.410297393798828, -18.96380615234375, 132.7483367919922, 141.40478515625, 25.17315673828125, -12.803970336914062, 4.342746734619141, 15.446907997131348, 167.5233154296875, -20.634132385253906, 42.089111328125, 21.980712890625, 60.57683563232422, -0.31764984130859375, 34.728118896484375, -7.68499755859375, -124.04293823242188, 141.19920349121094, 142.197021484375, 119.70596313476562, 164.7352294921875, 26.04290771484375, 188.24481201171875, 8.371978759765625, 128.4165496826172, 12.021697998046875, 164.41348266601562, 15.567924499511719, 184.82138061523438, 21.41424560546875, 5.576080322265625, 156.689208984375, 139.73477172851562, 41.57514953613281, -126.393310546875, 173.72378540039062, 13.522037506103516, 11.210845947265625, -130.39242553710938, 66.60014343261719, 1.1632347106933594, 148.94219970703125, 192.1776123046875, 116.67105102539062, -43.8297119140625, 30.340301513671875, 33.861480712890625, 28.64990997314453, 98.3804931640625, 109.18026733398438, 203.90928649902344, 109.3060073852539, -57.327972412109375, 107.06771850585938, -21.56640625, -77.72830200195312, -13.9515380859375, 9.070137023925781, 34.4345703125, 68.58517456054688, 46.37225341796875, 32.192657470703125, 128.68231201171875, -38.72112274169922, 73.48089599609375, 156.2033233642578, -8.253974914550781, 4.279327392578125, -15.966522216796875, 71.95346069335938, 209.84521484375, 142.32342529296875, 141.98451232910156, 135.67333984375, 196.75950622558594, 125.85626220703125, 184.51046752929688, 12.924453735351562, -7.38140869140625, -79.78120422363281, -7.63592529296875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000411.npy"} {"epoch": 0.8607329842931937, "step": 412, "batch_size": 128, "mean": 49.418601989746094, "std": 90.441650390625, "min": -183.47857666015625, "p10": -45.87105865478515, "median": 33.06052780151367, "p90": 167.01034545898438, "max": 243.30258178710938, "pos_frac": 0.671875, "sample": [-128.95773315429688, 165.38967895507812, 5.539058685302734, 182.5665283203125, 50.4293212890625, 104.25006103515625, 37.522705078125, 135.8204345703125, -29.051513671875, -13.90625, 6.5147705078125, 35.009033203125, -2.66229248046875, -15.342874526977539, 23.175750732421875, -183.47857666015625, -0.09222412109375, -29.203521728515625, -80.03178405761719, 28.192359924316406, 7.4626617431640625, 2.0330047607421875, 17.1197509765625, 85.08795166015625, 164.40956115722656, 156.72015380859375, 173.8475341796875, 35.573265075683594, -38.67933654785156, 32.75519561767578, 122.91781616210938, 123.4833984375, -19.48388671875, 163.73724365234375, 182.40176391601562, 157.49502563476562, 165.97311401367188, 18.169097900390625, 126.98422241210938, 70.71136474609375, 132.7074737548828, -28.13189697265625, 168.68344116210938, -118.51968383789062, 93.43585205078125, -15.74639892578125, -3.0888671875, -29.33599853515625, -145.03253173828125, 49.929656982421875, 56.1600341796875, 176.42738342285156, 196.34347534179688, -17.8572998046875, 169.51300048828125, 21.198638916015625, -10.370773315429688, 71.6806640625, -87.0458984375, -43.974700927734375, -66.5982666015625, -7.579349517822266, 203.214599609375, 62.3226318359375, -5.4215240478515625, 167.30557250976562, 129.5504150390625, 243.30258178710938, 46.331085205078125, 63.205780029296875, -106.60504150390625, 98.08210754394531, -48.423095703125, -15.305816650390625, 26.817047119140625, 33.36585998535156, 193.8556671142578, 161.79104614257812, 16.816802978515625, -165.17678833007812, 35.379730224609375, 96.18202209472656, 156.4614715576172, 0.566680908203125, 166.88381958007812, 1.8330230712890625, 159.98345947265625, 155.0414276123047, 181.74160766601562, -84.06503295898438, 10.563186645507812, 165.83731079101562, 146.0529022216797, -131.71969604492188, -6.803276062011719, 87.60392761230469, -16.38720703125, 132.9114990234375, -8.53643798828125, 129.40786743164062, -3.360870361328125, -0.36614418029785156, 201.1824951171875, 47.217529296875, 139.90081787109375, 7.306276321411133, -24.53912353515625, 73.89402770996094, 8.490509033203125, 116.6527099609375, 155.840087890625, 25.825393676757812, 0.0, 55.13713073730469, -14.308563232421875, 83.17843627929688, 1.8539276123046875, 30.33844757080078, -105.635986328125, 4.079301834106445, 16.322662353515625, 135.75186157226562, -44.77732849121094, 54.95599365234375, -10.627334594726562, -10.02215576171875, 129.79730224609375, 164.32687377929688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000412.npy"} {"epoch": 0.86282722513089, "step": 413, "batch_size": 128, "mean": 46.20387268066406, "std": 80.52751159667969, "min": -154.59503173828125, "p10": -41.921008300781246, "median": 22.32202911376953, "p90": 160.82718963623046, "max": 239.56558227539062, "pos_frac": 0.71875, "sample": [20.013700485229492, -6.44976806640625, -1.2018966674804688, 0.17420196533203125, 7.150970458984375, 146.90634155273438, 40.72216796875, 19.97833251953125, -154.59503173828125, -43.88200378417969, 174.65164184570312, 38.075897216796875, 179.62799072265625, -49.994964599609375, 32.297637939453125, 15.891510009765625, 10.9610595703125, 175.5115966796875, 62.71484375, -102.64689636230469, 50.23944091796875, 13.533641815185547, 74.25834655761719, 16.7071533203125, 23.071792602539062, -21.845932006835938, 9.385627746582031, 189.04974365234375, -6.388025283813477, 106.83914947509766, 142.3641357421875, -38.802734375, -46.78424835205078, 87.94245910644531, -40.0274658203125, 152.81216430664062, 160.7037811279297, 145.45042419433594, 149.17227172851562, -10.340133666992188, 107.5509033203125, -3.6455535888671875, -87.45703125, 2.94781494140625, 70.9849853515625, -28.746490478515625, -87.92945098876953, 54.53858947753906, 173.91098022460938, 64.73626708984375, -36.76611328125, 4.351409912109375, -19.502059936523438, 61.573486328125, 24.2196044921875, 124.24637603759766, 165.76202392578125, -25.478042602539062, 41.911102294921875, 55.688323974609375, 167.66342163085938, 20.735015869140625, 15.325576782226562, 42.271263122558594, 32.98284912109375, 200.873291015625, -80.41854858398438, 47.44598388671875, 154.3709716796875, -44.09442138671875, 145.663330078125, 61.781715393066406, -21.041152954101562, 239.56558227539062, 17.182811737060547, -20.88153076171875, -4.278045654296875, 25.154037475585938, -20.746742248535156, 93.9783935546875, 6.019775390625, 145.43402099609375, -49.1246337890625, 143.06570434570312, 139.0546112060547, -20.2977294921875, 16.008544921875, 124.548583984375, 7.124998092651367, -60.999542236328125, 157.95968627929688, 38.052391052246094, -41.08058166503906, 134.09271240234375, 21.572265625, -19.486438751220703, 0.3129730224609375, -20.131866455078125, 150.00811767578125, 9.308723449707031, 9.09006118774414, -17.89311981201172, 6.010528564453125, 132.2498779296875, 100.60540771484375, 73.37176513671875, 1.0603675842285156, 12.1109619140625, 129.66995239257812, 54.96055603027344, 31.5877685546875, 14.61737060546875, 115.15819549560547, 155.1845703125, -17.1578369140625, -90.50057983398438, 61.238525390625, 161.11514282226562, -136.86013793945312, 205.7947998046875, 10.850738525390625, 178.45962524414062, 6.365325927734375, -0.01200103759765625, 175.62881469726562, 154.58175659179688, 27.203536987304688, 18.4813232421875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000413.npy"} {"epoch": 0.8649214659685864, "step": 414, "batch_size": 128, "mean": 53.40479278564453, "std": 93.6128158569336, "min": -203.1234130859375, "p10": -54.397735595703125, "median": 35.9854621887207, "p90": 175.64136505126953, "max": 276.49884033203125, "pos_frac": 0.7265625, "sample": [119.79010009765625, -137.455810546875, 62.613800048828125, 116.0555419921875, 1.3287734985351562, 142.13641357421875, 68.10076904296875, 101.09710693359375, 23.75054931640625, -85.8529052734375, -9.5111083984375, 8.359352111816406, 174.25692749023438, 0.760498046875, 7.1825714111328125, -1.4867000579833984, -88.87445068359375, -105.46945190429688, 151.9637451171875, -8.52499008178711, -203.1234130859375, -85.19186401367188, 199.97784423828125, 3.6328125, 115.99172973632812, -135.14364624023438, -45.08673095703125, 3.7490577697753906, 140.0045166015625, 143.11131286621094, -70.64468383789062, -26.107269287109375, 116.14402770996094, 153.53033447265625, -55.32806396484375, 173.4789276123047, 144.27920532226562, -2.6375885009765625, 196.66006469726562, 0.9901275634765625, 25.80120849609375, 159.186279296875, 38.109588623046875, -21.334747314453125, 124.70399475097656, 138.46353149414062, 4.2191162109375, 13.870132446289062, 96.661376953125, 198.22552490234375, -7.0762786865234375, -46.927528381347656, -3.2974777221679688, 203.29061889648438, 132.85406494140625, -8.352081298828125, 150.98980712890625, 175.5784454345703, 176.26522827148438, -123.13040161132812, 115.9068603515625, 2.9156131744384766, 1.9830169677734375, 151.3382568359375, 10.0458984375, 46.27197265625, 88.8456039428711, 107.77580261230469, 57.456787109375, -1.85467529296875, 147.2305908203125, -8.123275756835938, 29.246307373046875, 189.86138916015625, 144.87362670898438, 131.32415771484375, 151.13613891601562, 0.27960205078125, 75.95504760742188, 124.81224822998047, 12.857345581054688, -28.643798828125, -129.81265258789062, 199.61038208007812, 276.49884033203125, 159.95132446289062, 159.98178100585938, 109.74784088134766, 5.936820983886719, 143.16775512695312, -185.65423583984375, 9.989105224609375, 35.37554931640625, 155.885986328125, 11.830833435058594, 175.78817749023438, -50.576873779296875, -2.671478271484375, -25.266799926757812, 36.595375061035156, -69.78652954101562, 60.495086669921875, 221.5982666015625, 36.82658767700195, 7.854331970214844, 14.098243713378906, 152.48902893066406, 69.80181884765625, 98.35993957519531, 19.566131591796875, -47.7974853515625, 65.32748413085938, 194.61148071289062, 40.231597900390625, -2.524714469909668, 10.80255126953125, 120.12158203125, 180.517333984375, 22.95159912109375, 37.95286560058594, 15.85052490234375, 1.80035400390625, -53.9990234375, -10.654052734375, 35.0235595703125, 67.87783813476562, 179.28851318359375, -7.3520660400390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000414.npy"} {"epoch": 0.8670157068062827, "step": 415, "batch_size": 128, "mean": 66.75575256347656, "std": 86.91581726074219, "min": -171.609619140625, "p10": -26.374897003173828, "median": 53.82069396972656, "p90": 175.34359436035155, "max": 240.36236572265625, "pos_frac": 0.7578125, "sample": [18.068695068359375, 132.26417541503906, 13.026496887207031, 39.881736755371094, -36.45001220703125, -2.088775634765625, 147.50283813476562, 1.0531005859375, 140.6304931640625, 11.26580810546875, 53.041664123535156, -2.075946807861328, 176.10330200195312, 109.34503173828125, -26.212631225585938, -9.542289733886719, 162.97976684570312, 2.79498291015625, 144.62640380859375, 40.32073974609375, 150.285888671875, 20.18463134765625, 183.04693603515625, 3.2166748046875, 146.36097717285156, 50.628509521484375, 99.0400390625, 162.31195068359375, 162.46218872070312, 114.475830078125, -129.8921356201172, 189.89947509765625, 180.4530029296875, 25.11492919921875, -135.77166748046875, 24.5804443359375, 54.23284912109375, 173.56201171875, -10.631515502929688, 94.32391357421875, 117.17630767822266, 12.372634887695312, 0.2112884521484375, 19.36163330078125, 48.775482177734375, -20.956741333007812, 159.7650146484375, 50.1697998046875, 175.01800537109375, 140.03160095214844, -13.531776428222656, 30.24212646484375, -96.9798583984375, 71.98190307617188, 151.0125732421875, 170.7738037109375, -26.681365966796875, 164.14947509765625, 48.47760009765625, -27.1083984375, 75.57992553710938, 155.40736389160156, -96.8353271484375, 118.18817138671875, 240.36236572265625, -19.43902587890625, 22.468090057373047, 71.84587097167969, -10.36370849609375, -1.0370750427246094, -25.339431762695312, 187.30764770507812, 140.83889770507812, 54.85090637207031, -5.781707763671875, -6.4022369384765625, 156.60494995117188, 195.31719970703125, 121.33778381347656, -42.7177734375, 73.8876953125, 9.004730224609375, 109.09442138671875, -63.4173583984375, 172.760498046875, -24.719146728515625, -83.90460968017578, 78.28638458251953, 106.6456298828125, -25.30780029296875, 0.3555755615234375, 1.31329345703125, 86.55410766601562, 155.85772705078125, 121.04010009765625, -27.4666748046875, 49.7425537109375, 86.57101440429688, 158.11221313476562, -171.609619140625, 197.63446044921875, 68.48859405517578, 97.6524658203125, 18.405654907226562, 28.980560302734375, 182.52072143554688, 53.408538818359375, -21.242050170898438, 163.21731567382812, 139.49951171875, 147.57928466796875, 52.17237091064453, 141.38221740722656, 130.7301788330078, -109.67420196533203, 215.12890625, 23.035812377929688, 184.6911163330078, 188.58544921875, -5.8695068359375, 7.29376220703125, 22.958221435546875, 151.02774047851562, 1.529296875, 158.79171752929688, 124.61735534667969, -26.243553161621094, 214.7601318359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000415.npy"} {"epoch": 0.8691099476439791, "step": 416, "batch_size": 128, "mean": 62.88777542114258, "std": 88.96063232421875, "min": -187.84869384765625, "p10": -40.02023010253906, "median": 57.337127685546875, "p90": 171.6307388305664, "max": 227.24652099609375, "pos_frac": 0.7421875, "sample": [164.32925415039062, 202.84426879882812, 176.98724365234375, 28.977203369140625, 3.9014511108398438, 3.6053085327148438, -41.413116455078125, 51.73625183105469, -14.64068603515625, 117.08168029785156, -39.42327880859375, 227.24652099609375, 71.290771484375, 29.546432495117188, 91.31906127929688, 19.86199188232422, -84.78738403320312, 182.69683837890625, 208.976318359375, 99.971923828125, 125.57687377929688, 151.20449829101562, 56.57098388671875, 170.7779998779297, -7.7745361328125, 39.235198974609375, -59.23634338378906, 139.4990234375, -105.35655212402344, 31.637008666992188, 126.54273986816406, 157.44012451171875, 145.60107421875, -6.3892822265625, -2.216419219970703, 174.49795532226562, -136.6103973388672, 160.05667114257812, 20.149024963378906, 17.9000244140625, 27.277252197265625, 40.13676452636719, -19.990875244140625, -187.84869384765625, -27.035137176513672, 174.93746948242188, 26.988815307617188, 130.3648223876953, 137.59146118164062, 172.87078857421875, 42.382232666015625, 171.0992889404297, 12.412353515625, 21.631248474121094, -77.8052978515625, 154.13778686523438, 168.83526611328125, 121.67169189453125, -6.4402313232421875, -10.986572265625, 127.89508056640625, 60.298065185546875, -133.8337860107422, -31.375, 193.96075439453125, 81.5718994140625, -7.434196472167969, 134.25201416015625, -5.516021728515625, 139.1021728515625, 0.3857574462890625, -26.155532836914062, 69.66201782226562, 143.9522705078125, 205.241943359375, 138.50550842285156, 116.10202026367188, 151.54751586914062, -24.961715698242188, 0.12291717529296875, 104.005615234375, 27.17474365234375, 126.45050048828125, -23.119247436523438, -87.86834716796875, 110.36224365234375, 22.2109375, 92.1033935546875, -7.620719909667969, 23.583580017089844, 60.31526184082031, 140.53953552246094, 70.91773223876953, 101.43572998046875, -160.508056640625, 134.81854248046875, 58.806060791015625, 2.5261077880859375, 62.1778564453125, 163.3822479248047, -63.2735595703125, 141.03814697265625, 44.2760009765625, 5.510284423828125, 58.103271484375, -11.6556396484375, 188.07183837890625, 120.0510025024414, 202.85452270507812, -16.907333374023438, 3.4464950561523438, 202.0098876953125, 47.58441162109375, 22.81787109375, -12.574127197265625, 156.46728515625, 113.83462524414062, 156.66067504882812, 54.74474334716797, -8.03451919555664, 38.49371337890625, -93.98919677734375, 165.01669311523438, -43.61138153076172, 148.43124389648438, 16.224349975585938, 130.38636779785156, 157.19630432128906], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000416.npy"} {"epoch": 0.8712041884816754, "step": 417, "batch_size": 128, "mean": 53.8004264831543, "std": 88.6214370727539, "min": -164.44424438476562, "p10": -63.21251525878906, "median": 43.517887115478516, "p90": 171.22719116210936, "max": 218.77420043945312, "pos_frac": 0.71875, "sample": [-1.1998825073242188, 42.72239685058594, 35.18211364746094, 98.2598876953125, -29.912200927734375, 120.18939208984375, 29.506729125976562, 4.392669677734375, 181.45941162109375, 184.14230346679688, 28.5775146484375, 58.19830322265625, 5.234426498413086, 128.56712341308594, -139.43402099609375, 6.9426727294921875, 10.121566772460938, -6.9197540283203125, -12.83929443359375, 150.29119873046875, 165.6922607421875, -27.598846435546875, 180.532958984375, -72.44412231445312, -63.009552001953125, -141.46499633789062, 100.11114501953125, 6.158622741699219, -38.09747314453125, 196.80715942382812, 58.66870880126953, -8.945770263671875, 11.3360595703125, 77.15511322021484, 115.06513977050781, 172.940185546875, 23.320953369140625, -118.72235107421875, 141.7718048095703, 117.65939331054688, 157.18328857421875, 26.579742431640625, 102.35231018066406, -8.72686767578125, 135.8248291015625, 108.02203369140625, 16.677078247070312, 73.1693115234375, 102.10066223144531, 57.33753967285156, 23.373430252075195, -0.4679603576660156, -12.813369750976562, -77.16830444335938, -70.07896423339844, 39.460205078125, -3.98095703125, 154.6180419921875, 58.404937744140625, -11.7462158203125, 50.0235595703125, 15.0042724609375, 3.2000694274902344, 14.466766357421875, 171.096435546875, 1.834625244140625, 159.70993041992188, -27.9149169921875, -164.44424438476562, 1.9207763671875, -23.640899658203125, 170.4095458984375, 100.3375244140625, 84.47824096679688, 51.89190673828125, -63.68609619140625, 43.81648254394531, 160.9237060546875, 0.0, -1.9248046875, 82.53202819824219, 0.0, 50.394012451171875, 148.686767578125, 117.38731384277344, 135.9906463623047, 52.15081787109375, 4.5799102783203125, 130.26968383789062, -18.66748046875, -87.89138793945312, 168.78021240234375, 155.0903778076172, 155.60269165039062, 171.53228759765625, 89.03958129882812, -157.96646118164062, 89.75013732910156, 216.7344970703125, 7.9354705810546875, 163.385009765625, -114.83838653564453, -65.46597290039062, 55.291133880615234, 154.1181640625, 171.7827606201172, 181.39822387695312, -31.283843994140625, -129.65614318847656, 18.094329833984375, 16.9998779296875, -6.3481597900390625, 27.402753829956055, 161.12942504882812, 182.8111572265625, 164.3167724609375, -14.62115478515625, 49.85675048828125, 173.45175170898438, 7.749570846557617, 68.57333374023438, 35.64021301269531, 43.21929168701172, -5.6976470947265625, 110.12686157226562, 130.66793823242188, 218.77420043945312, 199.60247802734375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000417.npy"} {"epoch": 0.8732984293193717, "step": 418, "batch_size": 128, "mean": 45.0596923828125, "std": 84.50865936279297, "min": -187.7359619140625, "p10": -69.76234512329101, "median": 33.6829833984375, "p90": 157.87162017822266, "max": 229.90234375, "pos_frac": 0.7265625, "sample": [-67.36566162109375, 143.68038940429688, 31.547760009765625, 46.64654541015625, -88.00537109375, -106.60746765136719, 0.310791015625, 185.04925537109375, -44.440277099609375, -187.7359619140625, 39.34588623046875, 93.69583892822266, 30.734130859375, -50.236846923828125, 59.426483154296875, 67.49021911621094, 109.17377471923828, -78.2716064453125, 125.7984619140625, 16.314300537109375, 161.99520874023438, 103.92703247070312, -16.8997802734375, 0.6979942321777344, -0.8319091796875, 9.35736083984375, -23.741683959960938, 134.202392578125, 46.1455078125, 185.34490966796875, 138.93545532226562, 48.19573974609375, 157.6034393310547, 109.7230224609375, 18.222579956054688, 82.44931030273438, -25.189453125, 138.54539489746094, -48.520355224609375, -11.551513671875, 0.21063995361328125, 106.95431518554688, 34.501434326171875, 117.7818603515625, -7.6889495849609375, 20.557830810546875, 0.5903911590576172, 126.47880554199219, 28.241424560546875, -19.08197021484375, 2.6012725830078125, 163.14349365234375, 119.79934692382812, 28.87176513671875, 19.889984130859375, -10.678466796875, 7.04498291015625, 35.428253173828125, 60.909576416015625, 70.31594848632812, -75.37432861328125, 131.6671142578125, 158.49737548828125, 28.331344604492188, 32.864532470703125, 9.128257751464844, 128.92593383789062, -70.27123260498047, -144.61953735351562, 7.708763122558594, -36.822784423828125, 25.457595825195312, 133.99456787109375, 43.57550048828125, 66.33218383789062, 179.34169006347656, 142.7894287109375, -9.865875244140625, 82.26043701171875, -101.81710815429688, 40.9613037109375, 143.00286865234375, -8.39410400390625, 58.5560302734375, 98.07015991210938, 159.1282958984375, 148.5205078125, -70.2764892578125, 2.829010009765625, -70.61029052734375, 14.6048583984375, -28.546966552734375, -21.042266845703125, 130.99862670898438, 147.0937042236328, 16.67272186279297, 169.06289672851562, 90.16526794433594, 56.313575744628906, 152.27171325683594, -72.048583984375, 0.9837837219238281, 8.909814834594727, 11.379180908203125, -117.96356201171875, 39.232086181640625, 64.43429565429688, 140.171875, -148.62884521484375, 165.3638916015625, 34.919620513916016, 18.424774169921875, 50.90049743652344, -4.332592010498047, 156.04913330078125, 69.01111602783203, 229.90234375, 163.08963012695312, -31.190536499023438, 217.20889282226562, 58.27117919921875, -19.676109313964844, -11.9366455078125, 25.622406005859375, -69.54425048828125, 157.25753784179688, 219.45547485351562, 9.855819702148438], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000418.npy"} {"epoch": 0.875392670157068, "step": 419, "batch_size": 128, "mean": 64.07568359375, "std": 94.6686782836914, "min": -224.0828399658203, "p10": -45.63441467285154, "median": 53.55669021606445, "p90": 173.32902221679686, "max": 247.681396484375, "pos_frac": 0.765625, "sample": [70.87835693359375, 16.985458374023438, 2.614349365234375, 170.3197021484375, 28.399757385253906, 27.807708740234375, 8.004951477050781, 19.192428588867188, 54.892333984375, -101.71994018554688, -84.22081756591797, 80.718017578125, -128.42803955078125, 161.00982666015625, 18.615211486816406, 29.244659423828125, 204.05419921875, 12.190505981445312, 118.854248046875, 161.97091674804688, 166.02049255371094, 126.072998046875, 175.34075927734375, 36.61143493652344, 183.7330322265625, 0.16058349609375, 35.54298400878906, 0.315277099609375, -19.761871337890625, -86.56036376953125, 163.05780029296875, 153.39218139648438, 136.09976196289062, -4.666717529296875, 65.85626220703125, -23.597305297851562, 84.52748107910156, -0.7096576690673828, 148.56622314453125, -151.2138671875, 167.09915161132812, 3.360076904296875, 163.01947021484375, 125.06103515625, 247.681396484375, 10.460990905761719, 206.75917053222656, -96.241943359375, 55.701416015625, 203.00021362304688, 130.27450561523438, 98.61770629882812, -19.589767456054688, 54.147621154785156, 43.910552978515625, 159.26458740234375, 206.43609619140625, 151.54132080078125, 6.36944580078125, -12.005096435546875, -116.61421203613281, 11.47882080078125, -28.644691467285156, 2.4316368103027344, 222.935791015625, -100.349365234375, 155.9751739501953, 177.8364715576172, 10.972457885742188, 38.68218994140625, -20.463783264160156, 147.552001953125, 140.98635864257812, 22.043716430664062, 107.19010925292969, 0.0, -95.32867431640625, -15.516143798828125, 15.232986450195312, 0.0, 51.1893310546875, 137.6371307373047, 142.9327850341797, 171.68826293945312, -9.81475830078125, 149.58673095703125, 9.89776611328125, 175.638916015625, 105.63851928710938, 19.067733764648438, 171.5480194091797, 29.495361328125, 174.88265991210938, 66.03057861328125, -5.896240234375, 15.85052490234375, 130.50054931640625, 91.16372680664062, 153.66627502441406, 170.83099365234375, 186.50741577148438, 15.91796875, 217.583251953125, 17.628173828125, 164.31280517578125, -40.357269287109375, 148.7159423828125, 151.3418731689453, 58.79803466796875, 144.70614624023438, -89.29150390625, -57.94775390625, 172.66317749023438, 149.9023895263672, -224.0828399658203, 52.96575927734375, -146.72836303710938, -15.079345703125, 150.93280029296875, -20.2689208984375, 114.70881652832031, 151.47119140625, -11.749786376953125, 21.48052978515625, 9.412002563476562, 86.89071655273438, 156.70596313476562, 41.57373046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000419.npy"} {"epoch": 0.8774869109947644, "step": 420, "batch_size": 128, "mean": 61.713958740234375, "std": 87.83879089355469, "min": -174.2064971923828, "p10": -35.712802505493165, "median": 38.46051788330078, "p90": 165.01202392578125, "max": 223.07171630859375, "pos_frac": 0.78125, "sample": [-104.3025894165039, 14.265151977539062, 31.54278564453125, -172.226318359375, -25.618698120117188, -8.293975830078125, 38.915802001953125, 160.257080078125, 26.822479248046875, 72.42901611328125, 160.82772827148438, 4.17578125, -129.94219970703125, 126.03262329101562, 129.3720245361328, 9.98532485961914, 146.83541870117188, -112.87650299072266, 139.17764282226562, 116.75322723388672, 34.348052978515625, 88.57772064208984, -5.3386993408203125, 72.17523193359375, 5.7560577392578125, 171.11734008789062, 9.7503662109375, 21.230140686035156, 118.3338623046875, 35.024139404296875, 2.1876373291015625, 199.3360595703125, 38.00523376464844, 145.69647216796875, 15.49005126953125, -4.39990234375, 135.10775756835938, 136.2528076171875, 188.2698974609375, 12.244598388671875, 192.0892333984375, 73.3084716796875, 24.51892852783203, 150.3720245361328, 18.401611328125, 32.75592041015625, 117.39096069335938, 4.7361602783203125, -4.6427764892578125, 103.50105285644531, -13.4310302734375, 102.70166015625, 144.43617248535156, 162.63027954101562, 133.9642333984375, 138.08306884765625, 0.0, 32.49147033691406, -36.95777893066406, -136.64410400390625, 134.94735717773438, 133.86195373535156, 10.07769775390625, 16.17995834350586, 15.629364013671875, 31.60186767578125, -9.59979248046875, 188.432861328125, 205.34518432617188, 140.64691162109375, 142.22927856445312, -60.073944091796875, 27.69671630859375, 89.66526794433594, -39.15663146972656, 176.56915283203125, 115.90245056152344, 20.52691650390625, 165.59756469726562, 34.57977294921875, -7.4639739990234375, 12.616127014160156, 223.07171630859375, -46.486572265625, 152.6787567138672, 97.40650939941406, -2.0734100341796875, -35.17924118041992, 163.56060791015625, 164.76107788085938, 76.181884765625, 20.858001708984375, 173.4676513671875, 163.93170166015625, 11.8509521484375, 15.28302001953125, 103.9168701171875, 60.287628173828125, 199.902099609375, -10.11651611328125, 188.41757202148438, 121.17355346679688, -174.2064971923828, -1.0525970458984375, 28.922576904296875, 156.93643188476562, 130.0007781982422, 9.04541015625, -69.36282348632812, -78.33900451660156, 90.95608520507812, 116.68984985351562, 119.34368896484375, 161.5929412841797, 158.59420776367188, 22.2894287109375, 108.62298583984375, -170.55606079101562, 20.72381591796875, 197.11785888671875, 146.72216796875, 89.5965576171875, -34.77447509765625, -7.820423126220703, 3.9840164184570312, 109.65463256835938, 0.5486335754394531, 92.44839477539062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000420.npy"} {"epoch": 0.8795811518324608, "step": 421, "batch_size": 128, "mean": 64.15872192382812, "std": 83.17400360107422, "min": -150.25042724609375, "p10": -27.523069763183592, "median": 53.595237731933594, "p90": 171.08970947265624, "max": 204.04513549804688, "pos_frac": 0.75, "sample": [181.918212890625, 103.11907958984375, 122.29027557373047, 99.78530883789062, 132.77093505859375, 102.98719787597656, 102.3087158203125, 10.900970458984375, -150.25042724609375, -5.667030334472656, 147.94638061523438, -27.138946533203125, 9.44287109375, 120.01646423339844, 40.9547233581543, 130.48785400390625, 37.918121337890625, -141.49737548828125, 44.92266845703125, 125.57325744628906, -32.066017150878906, -66.7017822265625, 67.19187927246094, 13.269447326660156, 140.54525756835938, 200.49530029296875, 120.44046020507812, 42.55035400390625, 10.626659393310547, 70.0468521118164, 0.0, -92.2935791015625, 99.90545654296875, 33.80116271972656, 12.217620849609375, 75.48591613769531, 168.78515625, 149.30844116210938, 26.937454223632812, 157.30670166015625, 179.81057739257812, 151.30133056640625, 43.32489013671875, 14.230712890625, -5.629291534423828, 20.473960876464844, 124.1498031616211, -23.593414306640625, 186.4560546875, 131.57669067382812, 133.6029052734375, -10.475830078125, 67.67636108398438, 93.13101196289062, -6.946533203125, 180.90252685546875, -100.4234390258789, 159.5099639892578, 11.8255615234375, 34.082763671875, 37.741943359375, 75.4714126586914, 199.13262939453125, 186.82809448242188, 161.44618225097656, 76.52918243408203, 147.13880920410156, 157.24310302734375, 158.55772399902344, 148.1311492919922, 153.04173278808594, 178.43035888671875, 95.91903686523438, 156.12213134765625, 73.0889892578125, 49.2205810546875, -24.60015869140625, 23.49713134765625, 156.15365600585938, 12.400291442871094, 137.54232788085938, 7.1619720458984375, 43.843231201171875, 183.35116577148438, -4.2001953125, 65.66018676757812, 36.88623046875, -21.538040161132812, 143.9178466796875, -10.297813415527344, 127.86284637451172, 123.38728332519531, 196.61758422851562, 57.77655029296875, -30.992950439453125, 44.926025390625, 121.145751953125, -102.17924499511719, -142.72268676757812, 49.41392517089844, -25.90411376953125, -38.09642028808594, 36.781219482421875, -11.0579833984375, 24.68646240234375, 170.77960205078125, 8.206207275390625, -7.32452392578125, 102.6010971069336, -92.72921752929688, -28.419357299804688, 162.11819458007812, 45.605072021484375, 140.71612548828125, -22.77825927734375, 171.81329345703125, 110.12622833251953, 0.0, -44.8314208984375, -20.578758239746094, 204.04513549804688, 82.1015625, 33.04881286621094, 0.0, 4.016693115234375, -5.70208740234375, 190.8245849609375, 1.58349609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000421.npy"} {"epoch": 0.881675392670157, "step": 422, "batch_size": 128, "mean": 47.30894470214844, "std": 91.97686767578125, "min": -184.942138671875, "p10": -79.33635025024412, "median": 40.56004333496094, "p90": 167.82152099609374, "max": 210.42929077148438, "pos_frac": 0.71875, "sample": [15.7794189453125, -18.091453552246094, 154.35826110839844, -26.051361083984375, 125.49844360351562, -39.63269805908203, 156.33668518066406, -147.03289794921875, 33.718841552734375, -1.074310302734375, 125.65519714355469, 4.98443603515625, 66.8448486328125, 155.99203491210938, 84.60580444335938, -184.942138671875, 12.99029541015625, 11.93280029296875, -32.70027160644531, 125.06884002685547, 118.5901107788086, 37.019622802734375, 115.3494873046875, 26.396408081054688, -52.791412353515625, 148.9808349609375, -138.9811553955078, 117.88291931152344, 130.05697631835938, 179.27450561523438, 71.38465118408203, 3.905242919921875, -15.712875366210938, 67.53268432617188, 7.988700866699219, 133.95068359375, 6.830835342407227, -115.9880142211914, 181.86985778808594, 8.499740600585938, 8.346176147460938, 164.94309997558594, -31.644989013671875, 48.68072509765625, 47.904388427734375, -166.9493408203125, -11.575576782226562, 210.42929077148438, 76.86383056640625, 156.34906005859375, 46.41973876953125, 179.46945190429688, 128.37069702148438, 14.973087310791016, 102.85379028320312, 7.361236572265625, 19.3533935546875, 15.422647476196289, 20.74676513671875, 83.63705444335938, 141.11978149414062, -93.94676971435547, -37.6474609375, -143.93191528320312, 186.38534545898438, 5.298187255859375, 109.7670669555664, 13.125732421875, 2.8133506774902344, 189.4478759765625, 97.92190551757812, -171.514892578125, 162.98158264160156, -28.147476196289062, 62.103240966796875, 2.2302818298339844, 129.98483276367188, -92.05699920654297, 173.79510498046875, 6.116859436035156, 190.4488525390625, -1.7737960815429688, 131.34280395507812, 51.587127685546875, 180.13726806640625, 57.236724853515625, -73.8846435546875, 19.78363037109375, 167.50381469726562, 0.19840621948242188, -14.756404876708984, -23.19580078125, 93.3857421875, -23.38922119140625, 140.220947265625, 103.61221313476562, -39.15252685546875, -3.1512451171875, 54.503173828125, 120.91490173339844, 142.1839141845703, 13.504043579101562, 191.42120361328125, 139.10565185546875, -103.83659362792969, 139.19937133789062, 61.087860107421875, -130.02862548828125, -13.40081787109375, 101.46270751953125, 0.0, -110.4967041015625, 168.56283569335938, 143.98776245117188, -111.07162475585938, 45.05291748046875, 63.4571533203125, 133.83853149414062, 36.93401336669922, -1.0247726440429688, -24.17401123046875, 100.49047088623047, 4.2403106689453125, 169.41175842285156, 41.083038330078125, 40.03704833984375, -0.3339729309082031, 189.19845581054688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000422.npy"} {"epoch": 0.8837696335078534, "step": 423, "batch_size": 128, "mean": 64.9599838256836, "std": 92.14130401611328, "min": -157.77505493164062, "p10": -39.20352478027343, "median": 59.08330535888672, "p90": 179.44859313964844, "max": 203.16546630859375, "pos_frac": 0.7734375, "sample": [1.788909912109375, 179.17787170410156, -16.167678833007812, 13.493438720703125, 162.0584716796875, 145.0025634765625, 32.86076354980469, -15.052978515625, 2.8511428833007812, 148.14337158203125, 96.61431884765625, 147.0533905029297, 31.188140869140625, 136.8343505859375, -115.60150146484375, -12.05029296875, 160.65399169921875, -98.46542358398438, 180.27239990234375, 12.330886840820312, 9.579586029052734, -32.814697265625, 23.222427368164062, 144.39108276367188, -138.51312255859375, 114.40963745117188, 124.4493408203125, 12.032638549804688, 156.69984436035156, 183.08120727539062, 5.110626220703125, 24.6258544921875, 51.816162109375, 70.4840087890625, 114.54629516601562, 135.99026489257812, 165.21929931640625, 0.3133544921875, -19.84246826171875, 84.40411376953125, 119.4631576538086, 171.26092529296875, -24.811279296875, 157.44808959960938, -2.8966522216796875, 30.406829833984375, 8.6732177734375, 196.27392578125, -42.68145751953125, -36.891357421875, 179.92401123046875, -37.712982177734375, -57.71710205078125, 63.176666259765625, 171.32516479492188, -35.8037109375, -112.334716796875, -128.40805053710938, 179.49676513671875, 166.63134765625, 30.157875061035156, -1.2387542724609375, 140.9422607421875, 12.766845703125, -11.820770263671875, 123.42893981933594, 0.7165908813476562, 161.70208740234375, 7.349281311035156, 29.413177490234375, 147.02163696289062, 156.58151245117188, 123.64224243164062, 135.81353759765625, 38.96208190917969, 61.57427978515625, -64.44625854492188, 166.20309448242188, 203.16546630859375, 62.382598876953125, 18.617706298828125, 177.58868408203125, 15.930511474609375, 136.48226928710938, 187.43536376953125, 56.59233093261719, 92.5057373046875, 24.32330322265625, 16.566680908203125, 188.89871215820312, 141.60154724121094, 151.57070922851562, 184.39727783203125, -1.471649169921875, 198.99395751953125, 138.01214599609375, 2.9183712005615234, -66.05392456054688, 179.42794799804688, -9.46282958984375, 3.2863616943359375, 126.77301025390625, 33.93798828125, 185.49301147460938, 202.276611328125, -35.68182373046875, 1.8260345458984375, 132.52801513671875, 181.0993194580078, -17.50030517578125, -139.202880859375, 43.35865020751953, -93.24172973632812, 69.8746337890625, 17.55813980102539, 165.81271362304688, 76.40640258789062, 144.93292236328125, 172.23419189453125, 33.786376953125, 137.869384765625, 144.68609619140625, 157.92147827148438, -141.009033203125, 30.652496337890625, 90.53010559082031, 0.23832130432128906, -157.77505493164062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000423.npy"} {"epoch": 0.8858638743455497, "step": 424, "batch_size": 128, "mean": 62.45664978027344, "std": 80.75663757324219, "min": -164.33438110351562, "p10": -34.19738311767578, "median": 60.59246826171875, "p90": 161.18330993652344, "max": 212.86941528320312, "pos_frac": 0.75, "sample": [68.85649108886719, 47.47605895996094, 0.4223175048828125, -9.57794189453125, 67.72576904296875, 12.5230712890625, -4.214439392089844, -4.80938720703125, 33.75786590576172, -8.63568115234375, 3.831146240234375, -96.91098022460938, 212.86941528320312, 76.70298767089844, -3.532928466796875, 101.27959442138672, 16.79486083984375, 186.2521514892578, 162.7832794189453, 147.49783325195312, -7.0077362060546875, 5.219970703125, 190.54745483398438, 147.00869750976562, 96.34930419921875, 70.68264770507812, -0.9771308898925781, -11.91717529296875, 129.18634033203125, 124.82906341552734, 102.97467803955078, 191.08981323242188, 152.73158264160156, -38.14375686645508, 126.2357177734375, 134.96212768554688, 35.66473388671875, 142.62896728515625, 167.76055908203125, -79.00845336914062, -10.5533447265625, 155.45712280273438, 138.24642944335938, -164.33438110351562, -0.11084747314453125, 190.3428955078125, 177.12484741210938, 75.14926147460938, -147.79505920410156, 44.1282958984375, -29.53969955444336, 4.30987548828125, -2.589263916015625, 161.67568969726562, 71.55587768554688, 159.68460083007812, 34.13018798828125, 152.99822998046875, 118.88153076171875, 54.179595947265625, 24.481414794921875, -6.6517333984375, 155.40673828125, 59.66033935546875, 188.97393798828125, 134.86868286132812, 61.52459716796875, 33.732177734375, -57.6865234375, -45.44384765625, -38.382568359375, 71.18124389648438, 135.72848510742188, 189.4417724609375, 199.215087890625, 160.9722900390625, 2.9926376342773438, 49.10894012451172, 127.39105224609375, 98.41455078125, 62.3804931640625, 126.84872436523438, -2.0830841064453125, 3.4482650756835938, 153.2592010498047, 89.83306884765625, -57.318450927734375, 0.0, -37.94432067871094, 32.465545654296875, 9.87127685546875, 143.01400756835938, 145.02413940429688, -8.659271240234375, 84.83073425292969, 148.25514221191406, 105.78070068359375, 145.65115356445312, 150.15286254882812, 128.36398315429688, 14.340362548828125, -81.3424072265625, 123.78472900390625, 138.56793212890625, 118.29708862304688, 1.7430095672607422, -9.186098098754883, 7.523681640625, -31.28851318359375, 5.877067565917969, 128.71725463867188, -86.08057403564453, 150.37078857421875, 145.56906127929688, -32.591552734375, 9.055488586425781, 15.880743026733398, 52.1319694519043, 27.408843994140625, 68.4127197265625, 28.05303955078125, 86.91400146484375, -48.74749755859375, 37.829315185546875, 80.0584716796875, 13.926864624023438, 2.1976051330566406, 184.0731201171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000424.npy"} {"epoch": 0.8879581151832461, "step": 425, "batch_size": 128, "mean": 54.80955123901367, "std": 87.17227935791016, "min": -200.94168090820312, "p10": -36.328271484375, "median": 30.634063720703125, "p90": 172.17433166503906, "max": 222.9066162109375, "pos_frac": 0.734375, "sample": [140.09011840820312, 151.36709594726562, 57.4945068359375, 52.88581848144531, -15.75494384765625, 17.410003662109375, 43.002838134765625, 141.9578857421875, 48.09278869628906, -24.292495727539062, 159.31976318359375, 20.865875244140625, 163.73974609375, 87.82151794433594, 133.64175415039062, 0.2296905517578125, 121.764404296875, 140.5244140625, 1.4712448120117188, 22.93408203125, 67.5677490234375, 141.5848846435547, 30.474578857421875, -3.2089691162109375, 154.89013671875, 90.41069793701172, 3.679558753967285, 21.74951171875, -91.62936401367188, -3.8706741333007812, 45.47137451171875, 1.82464599609375, -18.4688720703125, 23.059844970703125, 181.035888671875, -4.337137222290039, 50.431488037109375, 188.9896240234375, 8.513458251953125, 16.015296936035156, -22.12249755859375, 190.43963623046875, 140.05528259277344, 30.623565673828125, -5.16650390625, 5.5376739501953125, -37.3642578125, -1.835540771484375, -149.12744140625, 194.64508056640625, 76.778564453125, 136.17279052734375, -98.5535888671875, 49.93109130859375, 171.49908447265625, 0.5051727294921875, 145.86419677734375, 222.9066162109375, 40.83349609375, 25.65765380859375, -35.88427734375, 30.644561767578125, 8.05538558959961, 39.08338928222656, 179.077392578125, -46.62565612792969, 98.68290710449219, -4.230598449707031, 161.7210693359375, 143.57107543945312, 139.12637329101562, 10.39599609375, 196.69537353515625, 6.831993103027344, -18.214080810546875, -7.727592468261719, -11.624835968017578, 28.896392822265625, -18.060386657714844, 94.65216827392578, 0.16788482666015625, 165.06640625, 27.6053466796875, 116.17341613769531, 174.22129821777344, -104.83100128173828, 14.797439575195312, -48.49928283691406, 32.41473388671875, -28.579345703125, 20.19091796875, -24.769792556762695, 31.575889587402344, -137.07545471191406, -52.4486083984375, -75.1815185546875, 113.4912109375, 15.985183715820312, 138.20294189453125, -125.42410278320312, 44.66473388671875, 206.6436767578125, 21.23651123046875, 154.3747100830078, 83.11524963378906, 107.35836791992188, 166.06314086914062, 0.500457763671875, 5.382530212402344, 149.0201416015625, 0.0, -10.34423828125, 156.14935302734375, 171.1883544921875, 211.81320190429688, -4.949960708618164, 28.97396469116211, 159.9084014892578, -200.94168090820312, 58.57740783691406, 184.32281494140625, 173.9168243408203, 124.3858642578125, 157.9678192138672, 173.74990844726562, -41.57701873779297, -3.8328628540039062, 3.7745132446289062], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000425.npy"} {"epoch": 0.8900523560209425, "step": 426, "batch_size": 128, "mean": 58.55752182006836, "std": 80.0155258178711, "min": -144.88412475585938, "p10": -18.11884307861328, "median": 28.32113265991211, "p90": 166.13714752197265, "max": 198.49307250976562, "pos_frac": 0.7734375, "sample": [143.02609252929688, 8.113021850585938, 1.5072364807128906, -36.433746337890625, 180.30166625976562, 23.609542846679688, 120.44033813476562, 2.1451416015625, -126.111083984375, 20.41912841796875, -121.29388427734375, -45.3201904296875, -49.14320373535156, 117.50296020507812, 153.0582733154297, 163.34857177734375, 146.75131225585938, 28.5645751953125, 138.3197784423828, 27.695480346679688, -53.326263427734375, 52.5106201171875, 180.1717529296875, 20.072052001953125, 3.5774383544921875, 21.589019775390625, 168.6922607421875, 181.234619140625, 143.63916015625, 172.91839599609375, 119.98644256591797, 3.136150360107422, 155.7181396484375, 131.72332763671875, 153.1041259765625, 9.562206268310547, 0.4064788818359375, 24.483482360839844, 165.40185546875, 175.03111267089844, 38.99517822265625, 175.36183166503906, 7.335723876953125, -6.125732421875, 154.23837280273438, 160.28915405273438, -1.596923828125, 117.20680236816406, 11.246292114257812, 9.71826171875, -96.04319763183594, 25.96966552734375, 29.360836029052734, 4.146820068359375, 22.062419891357422, 159.95919799804688, -63.936248779296875, 91.7124252319336, 160.48004150390625, 127.90605163574219, 126.4107666015625, 144.14410400390625, -32.365997314453125, -2.7025222778320312, -3.0333099365234375, 83.46234130859375, 58.45481872558594, 128.86456298828125, 2.8447723388671875, -9.64801025390625, 66.82373046875, 28.07769012451172, -6.246452331542969, 15.065292358398438, 1.09979248046875, 197.82940673828125, -10.365882873535156, -0.97320556640625, 42.93257141113281, 138.11532592773438, -17.808547973632812, 152.65017700195312, 70.90188598632812, 90.2947998046875, 25.932205200195312, 29.8330078125, 24.2919921875, 13.243431091308594, 13.088752746582031, 140.72232055664062, 98.938720703125, 20.190155029296875, 1.0284099578857422, -17.362350463867188, 179.54830932617188, 47.902130126953125, 90.457275390625, 174.85125732421875, -9.06817626953125, -27.440412521362305, 94.16600799560547, 5.28564453125, 0.0, 11.38531494140625, -8.144699096679688, 192.45504760742188, 93.65643310546875, 10.652191162109375, 7.92913818359375, 90.51554870605469, 137.35202026367188, 27.8115234375, -116.59417724609375, -3.5928955078125, -8.722434997558594, 137.35455322265625, -18.842864990234375, 163.85821533203125, 149.1796875, 116.92706298828125, 128.18629455566406, 3.9258127212524414, -144.88412475585938, 107.28695678710938, 198.49307250976562, 39.33349609375, -12.867919921875, 167.8528289794922], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000426.npy"} {"epoch": 0.8921465968586387, "step": 427, "batch_size": 128, "mean": 78.11422729492188, "std": 83.77867126464844, "min": -159.1014404296875, "p10": -23.014344024658204, "median": 74.85758972167969, "p90": 179.56565551757814, "max": 295.40704345703125, "pos_frac": 0.8203125, "sample": [19.788368225097656, 176.28057861328125, 17.213302612304688, -25.825927734375, -70.94646453857422, 9.414215087890625, 133.3593292236328, 89.75360107421875, 91.11068725585938, 38.66239929199219, 115.75958251953125, 40.13160705566406, 184.76260375976562, 11.100227355957031, 156.85693359375, 171.14816284179688, 153.0578155517578, 8.070533752441406, 139.72958374023438, 190.76763916015625, 187.7125244140625, 49.09757995605469, -24.498489379882812, 114.95510864257812, 153.01544189453125, 140.23153686523438, 134.86190795898438, 78.750732421875, 39.197383880615234, 51.331199645996094, 204.49179077148438, 167.53314208984375, 103.5750732421875, 38.734527587890625, 1.7467803955078125, 0.6117057800292969, 123.34575653076172, 13.481765747070312, -2.0135498046875, -35.641876220703125, 170.38526916503906, 22.1368408203125, -89.30105590820312, 63.425018310546875, -115.4853515625, 137.42306518554688, 0.863189697265625, 180.18858337402344, -22.943260192871094, 145.79327392578125, 179.71878051757812, 32.02679443359375, 149.10910034179688, 79.0482177734375, 195.27935791015625, 178.6285400390625, 206.44149780273438, -10.160751342773438, 15.407546997070312, -10.439727783203125, 139.9542236328125, 124.63409423828125, 189.16690063476562, 19.48626708984375, 76.1622314453125, 57.147308349609375, -17.543701171875, 151.38931274414062, 149.2130584716797, -45.94075012207031, -159.1014404296875, 171.564208984375, 118.86298370361328, -23.180206298828125, 131.31201171875, 20.46682357788086, -22.887481689453125, 150.22694396972656, 124.45332336425781, 69.33551025390625, 209.28652954101562, -89.82476806640625, 148.89044189453125, -29.660797119140625, 29.211318969726562, 116.99588012695312, 145.9572296142578, 61.5997314453125, 73.55294799804688, 117.95623779296875, 147.05758666992188, 171.77978515625, -23.38104248046875, 37.35302734375, 163.9783935546875, 139.15975952148438, 179.50003051757812, 55.373291015625, 9.216331481933594, 124.49748992919922, -13.527908325195312, 15.2542724609375, 39.7545166015625, -28.697067260742188, 8.637886047363281, 295.40704345703125, -6.208717346191406, 154.5635528564453, 3.990631103515625, 162.84368896484375, 180.8753662109375, 73.10203552246094, 135.1271209716797, 0.72320556640625, 46.78619384765625, -10.693374633789062, 158.6751708984375, 106.45420837402344, 0.65093994140625, 0.653533935546875, 168.4384765625, 27.746749877929688, -1.1086654663085938, 18.91387939453125, 31.332275390625, 154.78195190429688, 196.07818603515625, 166.58534240722656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000427.npy"} {"epoch": 0.8942408376963351, "step": 428, "batch_size": 128, "mean": 73.86524963378906, "std": 82.64962005615234, "min": -139.06826782226562, "p10": -14.15558700561523, "median": 71.65202331542969, "p90": 166.47731170654296, "max": 260.45477294921875, "pos_frac": 0.8046875, "sample": [32.644287109375, 5.558319091796875, -58.035125732421875, 164.24554443359375, 182.92605590820312, -21.305938720703125, 56.90289306640625, 56.1070556640625, -99.14089965820312, 11.773880004882812, 260.45477294921875, 24.110504150390625, 115.12838745117188, 128.21499633789062, -7.9990234375, 165.56198120117188, 25.05657958984375, -6.384765625, 15.983165740966797, -4.685028076171875, 129.69134521484375, -1.157562255859375, -132.84075927734375, 195.77120971679688, 155.60226440429688, 183.554443359375, 124.68795776367188, 143.845703125, 116.14814758300781, 225.09161376953125, 28.462326049804688, 176.4478759765625, 2.4097213745117188, 72.92831420898438, -17.200164794921875, 14.49294662475586, -1.7301025390625, 135.36651611328125, 37.985137939453125, 125.35382080078125, 58.715087890625, -8.988506317138672, -139.06826782226562, 140.50686645507812, 159.06158447265625, 201.31631469726562, -56.94727325439453, 58.217288970947266, 5.4961395263671875, 12.972747802734375, 150.82061767578125, 122.861572265625, 123.10098266601562, 108.00152587890625, 26.7003173828125, 115.78610229492188, 152.5355682373047, 108.005126953125, -45.48876953125, 216.00344848632812, -135.68873596191406, 198.4290771484375, 149.0012664794922, 45.315948486328125, 152.50390625, 6.596923828125, 137.02081298828125, 157.04872131347656, 117.50845336914062, 54.65960693359375, 17.522079467773438, 38.256866455078125, 154.80567932128906, 153.54159545898438, 223.25439453125, 16.655517578125, 47.91094970703125, -1.95587158203125, 84.09991455078125, 22.656967163085938, 64.55765533447266, 23.136505126953125, 129.88650512695312, 70.375732421875, -2.997039794921875, 8.3343505859375, -19.250747680664062, 25.373779296875, 87.58917236328125, 127.37733459472656, 243.49139404296875, 168.6130828857422, 122.79071044921875, 1.69403076171875, 91.3072509765625, -121.42794799804688, -16.81298828125, 160.2218017578125, 104.9013671875, 0.0, 23.947265625, 162.59815979003906, 137.1310577392578, 22.141799926757812, 152.66445922851562, 19.55896759033203, -13.016700744628906, 109.833984375, 122.29733276367188, 137.954833984375, 76.23336791992188, 139.79470825195312, 39.31169128417969, 64.44979095458984, 160.9947509765625, 73.20562744140625, 9.443328857421875, 74.15521240234375, 122.21357727050781, -1.091888427734375, 48.34271240234375, 25.47943115234375, 175.44149780273438, -27.60392951965332, -2.375457763671875, 153.14828491210938, 140.11480712890625, 128.43978881835938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000428.npy"} {"epoch": 0.8963350785340314, "step": 429, "batch_size": 128, "mean": 57.56501770019531, "std": 99.31901550292969, "min": -190.3032989501953, "p10": -65.53652648925778, "median": 38.46617126464844, "p90": 182.02071533203124, "max": 213.271484375, "pos_frac": 0.703125, "sample": [144.1150360107422, -185.14532470703125, -5.458526611328125, 125.93453979492188, 164.25538635253906, 158.87725830078125, 29.29399871826172, 27.18890380859375, -56.14881896972656, 119.6439208984375, 4.9738922119140625, -120.75230407714844, 3.186737060546875, 155.28575134277344, 157.16796875, 158.97483825683594, 1.208089828491211, 158.189208984375, -13.360504150390625, 23.24188232421875, 122.0223159790039, 134.5911865234375, 11.872133255004883, 159.59872436523438, -86.52059936523438, 171.26220703125, 117.611083984375, 192.5669403076172, 17.407012939453125, 86.740478515625, -125.87661743164062, 143.04852294921875, 138.50613403320312, 164.33236694335938, 29.262344360351562, 26.2708740234375, -123.38079833984375, 10.81842041015625, 162.06085205078125, 213.271484375, 158.18426513671875, 152.1183624267578, -86.60136413574219, 27.653541564941406, 107.5533447265625, -5.365631103515625, 200.16546630859375, 106.25726318359375, 9.21124267578125, -39.1817626953125, 7.482736587524414, 186.18756103515625, 191.85919189453125, 167.59341430664062, 13.831802368164062, 46.3323974609375, 204.45150756835938, -104.68583679199219, -119.10577392578125, 151.8529815673828, -8.88726806640625, 51.406982421875, 166.29759216308594, 181.19183349609375, -34.56005859375, 14.236885070800781, 53.4349365234375, 40.211944580078125, 16.962242126464844, 80.00100708007812, -25.610107421875, -83.21715545654297, -37.955810546875, -55.91905212402344, 151.17132568359375, -190.3032989501953, 174.85296630859375, -17.303443908691406, 17.89306640625, 166.8120574951172, -4.25370979309082, 0.0, 163.25523376464844, 200.88897705078125, 167.64431762695312, 60.270843505859375, 167.16683959960938, 14.640586853027344, -2.6864166259765625, 5.109287261962891, 195.96478271484375, -29.780029296875, 127.61570739746094, -17.20758056640625, 44.09617614746094, 54.875, 21.344974517822266, 4.077003479003906, -5.08740234375, 130.86691284179688, -98.43008422851562, -30.64599609375, -5.120574951171875, 14.230422973632812, -37.466705322265625, 114.04354858398438, 128.72592163085938, 189.30584716796875, -158.07595825195312, 132.1237335205078, 186.83193969726562, 153.026123046875, -57.95911407470703, 178.15785217285156, 183.95477294921875, -18.293701171875, 14.370609283447266, 5.397361755371094, 191.1865234375, 184.89572143554688, -21.710433959960938, 140.75360107421875, -17.221450805664062, -31.738571166992188, 71.28973388671875, 36.72039794921875, 113.16574096679688, -144.64532470703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000429.npy"} {"epoch": 0.8984293193717278, "step": 430, "batch_size": 128, "mean": 50.10704040527344, "std": 93.60381317138672, "min": -165.09530639648438, "p10": -65.08850555419922, "median": 35.477806091308594, "p90": 175.97530212402344, "max": 220.8326416015625, "pos_frac": 0.703125, "sample": [-66.60812377929688, -33.63136291503906, -121.92655944824219, 6.0407562255859375, 123.21422576904297, 127.78091430664062, -145.8141632080078, 118.26434326171875, -28.743194580078125, 35.9725341796875, 141.6246337890625, 9.977184295654297, 32.037109375, 21.844131469726562, 79.43370056152344, -165.09530639648438, 185.02532958984375, 24.107269287109375, 1.7526874542236328, 152.28085327148438, 131.51947021484375, -101.3919677734375, -8.664718627929688, 131.76763916015625, 14.16986083984375, 31.00579833984375, -42.46588134765625, 0.0, 175.84896850585938, 69.8719482421875, 110.43917083740234, -21.328125, 147.68463134765625, 60.00921630859375, 21.05096435546875, 0.0, 68.25418090820312, -2.4518966674804688, 2.46795654296875, 133.2584228515625, -110.7923583984375, 187.77978515625, -12.79693603515625, 151.18234252929688, 184.87118530273438, 119.36416625976562, 76.60563659667969, 19.41590118408203, -23.8785400390625, 220.8326416015625, 134.49436950683594, -9.447021484375, 121.91664123535156, 17.04864501953125, 165.48391723632812, 188.5597686767578, 23.559982299804688, -60.160369873046875, 159.902099609375, -149.15570068359375, 151.26670837402344, 164.8165740966797, 177.10861206054688, 162.55661010742188, 52.73374938964844, 62.623077392578125, 153.515625, 141.04579162597656, 8.867128372192383, 187.33929443359375, 31.56427001953125, 31.024215698242188, 204.9425048828125, 145.6219482421875, -111.24078369140625, -64.43724060058594, -1.727783203125, 35.600494384765625, 1.1722126007080078, 35.993499755859375, 149.6328125, 38.58561706542969, -35.677703857421875, 13.52828598022461, -20.68353271484375, 59.533355712890625, -17.1024169921875, 204.87713623046875, 0.0, 5.672660827636719, 58.974205017089844, -22.234512329101562, 92.4173583984375, 13.882217407226562, 196.84176635742188, 117.82721710205078, 31.73260498046875, 156.93182373046875, 6.910665512084961, -119.82557678222656, 193.6221466064453, 181.9461669921875, 42.957122802734375, -48.868621826171875, -120.68373107910156, 88.45896911621094, 148.4459228515625, 6.213417053222656, 176.27008056640625, 116.72979736328125, -139.27175903320312, 6.5281524658203125, 61.631927490234375, -12.765853881835938, 35.35511779785156, 162.119873046875, 10.53472900390625, -25.509918212890625, -1.5561676025390625, -133.07040405273438, -6.602783203125, -142.10397338867188, 66.57427978515625, 131.0412139892578, 130.740234375, 35.960235595703125, -30.693496704101562, 154.1190948486328], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000430.npy"} {"epoch": 0.900523560209424, "step": 431, "batch_size": 128, "mean": 47.07183837890625, "std": 85.8634033203125, "min": -150.82177734375, "p10": -45.19111251831055, "median": 28.876144409179688, "p90": 170.91746673583984, "max": 236.30682373046875, "pos_frac": 0.6875, "sample": [-9.318878173828125, -150.82177734375, 77.80403137207031, 144.8765106201172, 135.51522827148438, 77.13375854492188, 20.141189575195312, 84.87277221679688, 21.689178466796875, -72.77932739257812, -3.77191162109375, 26.50457763671875, 8.081390380859375, 51.32916259765625, -42.645751953125, 72.3834228515625, -37.629119873046875, 117.80126953125, 114.38572692871094, 15.916265487670898, -0.4991455078125, 121.314453125, 58.45158386230469, -79.087646484375, 107.00537109375, 145.69749450683594, 88.36769104003906, -55.618438720703125, 153.96087646484375, 199.31472778320312, 73.25636291503906, 79.46685791015625, 68.739990234375, -126.77916717529297, -54.05363464355469, 146.88955688476562, 99.65673065185547, -22.868896484375, -4.352012634277344, 108.23480224609375, -36.54288101196289, -145.79583740234375, 7.690401077270508, -7.36480712890625, 136.79196166992188, 56.019622802734375, 180.384521484375, -20.988704681396484, -46.54241180419922, -148.65805053710938, 5.025531768798828, 36.10150146484375, 1.8657665252685547, 142.7506103515625, 1.462860107421875, 149.78317260742188, 46.239532470703125, 188.35838317871094, 159.54037475585938, -20.74200439453125, 0.0029296875, 43.32777404785156, 167.9805450439453, 32.1866455078125, 20.210601806640625, 60.95379638671875, 65.01019287109375, 173.8927001953125, -22.22644805908203, -21.898651123046875, 170.22039794921875, -6.747947692871094, -3.0196533203125, 183.02459716796875, -1.036956787109375, 220.00213623046875, 7.064727783203125, 205.69610595703125, 45.36962890625, 31.214630126953125, 186.383056640625, 236.30682373046875, 44.22747802734375, -78.33428955078125, -0.8407669067382812, -8.21490478515625, 142.31097412109375, 27.32586669921875, -42.583648681640625, 22.308853149414062, 204.23687744140625, 140.7898712158203, 172.54396057128906, -44.61198425292969, 4.4368743896484375, 217.96929931640625, -22.291015625, 162.48788452148438, 2.603487014770508, 10.35101318359375, 37.18641662597656, 47.038726806640625, 6.1607666015625, 5.338134765625, 30.426422119140625, 4.608612060546875, 147.53070068359375, 56.7840576171875, 122.21250915527344, -14.092559814453125, 19.915077209472656, -9.414749145507812, 3.4040069580078125, 137.10455322265625, 23.935516357421875, 98.50431823730469, -10.563262939453125, -53.386993408203125, -102.66697692871094, 33.52992248535156, -17.489181518554688, -12.058380126953125, 11.584541320800781, -11.0045166015625, -104.3350830078125, 203.2952423095703, 143.50042724609375, 35.568603515625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000431.npy"} {"epoch": 0.9026178010471204, "step": 432, "batch_size": 128, "mean": 62.86650848388672, "std": 87.27725219726562, "min": -173.84115600585938, "p10": -29.93966674804687, "median": 43.867095947265625, "p90": 178.56447296142576, "max": 220.6270751953125, "pos_frac": 0.703125, "sample": [-16.307586669921875, 160.94091796875, 3.7131500244140625, 154.3125, 205.3399658203125, 191.38458251953125, 153.11929321289062, 112.5955810546875, -47.9315185546875, -153.37860107421875, 186.889404296875, -7.18011474609375, 162.71112060546875, 37.39459228515625, 14.674240112304688, 13.004981994628906, 34.81202697753906, 177.9766082763672, -49.90858459472656, 6.6302032470703125, 129.25006103515625, -173.84115600585938, 70.25355529785156, -0.21494483947753906, -22.708953857421875, 6.2299604415893555, -21.399169921875, 161.51622009277344, 69.01042175292969, -6.043731689453125, 27.3028564453125, 167.53749084472656, 186.63333129882812, 174.13746643066406, -32.85882568359375, -6.9803314208984375, 132.31039428710938, -37.20166015625, 179.9361572265625, 19.86981201171875, 131.7508544921875, 139.45681762695312, 41.746002197265625, 2.5708389282226562, -3.7517166137695312, -16.788223266601562, 70.623291015625, 14.437751770019531, -120.29244995117188, -3.317962646484375, 127.60710906982422, 59.81999206542969, -6.297431945800781, 182.46209716796875, -0.37842559814453125, 12.50799560546875, -87.8877182006836, 153.11471557617188, -124.03773498535156, 0.40842437744140625, -9.409576416015625, 23.64837646484375, 151.25314331054688, -11.762451171875, 149.4539031982422, 6.287078857421875, 187.00418090820312, 139.71102905273438, 139.07907104492188, 144.3922119140625, -44.94990539550781, 96.0118408203125, 4.751220703125, -12.550350189208984, 17.67491912841797, 191.39895629882812, 137.0242462158203, 199.1024169921875, 30.507476806640625, 134.0889892578125, 45.06158447265625, 3.9455947875976562, 220.6270751953125, -15.23626708984375, 147.1080322265625, 129.68905639648438, 168.97691345214844, 129.91452026367188, -28.6885986328125, 131.64926147460938, 11.508987426757812, -13.452098846435547, 0.0, 142.370849609375, 164.3114013671875, -21.73956298828125, 10.552978515625, -24.2918701171875, -0.79351806640625, 61.619476318359375, 57.320220947265625, -7.3887939453125, -38.36388397216797, 135.5980224609375, 110.48477935791016, 120.55551147460938, 196.68887329101562, 150.1868896484375, 196.6343994140625, 161.05154418945312, -3.3065185546875, 139.63906860351562, 42.672607421875, 195.18792724609375, 1.68157958984375, 51.17877197265625, 74.51324462890625, 95.61727905273438, 3.8639259338378906, 152.82659912109375, 113.10891723632812, 145.88436889648438, -53.79222106933594, 4.929264068603516, 51.76188659667969, 116.34312438964844, -2.8133773803710938, -38.287445068359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000432.npy"} {"epoch": 0.9047120418848168, "step": 433, "batch_size": 128, "mean": 50.07634353637695, "std": 87.06836700439453, "min": -194.47314453125, "p10": -38.59559936523437, "median": 47.44434356689453, "p90": 146.30254821777342, "max": 294.013916015625, "pos_frac": 0.7578125, "sample": [125.27195739746094, -151.94107055664062, -179.09011840820312, 182.82827758789062, 70.96633911132812, 36.86151885986328, 124.93711853027344, -13.33770751953125, -8.221181869506836, 38.71588134765625, 26.11676025390625, 19.913238525390625, 142.27615356445312, 83.17120361328125, 48.3975830078125, 24.90179443359375, -12.1875, 64.9522933959961, 63.865966796875, 18.74639892578125, 39.78607177734375, 15.096343994140625, 133.47552490234375, -41.91961669921875, 0.1265392303466797, 28.031185150146484, 179.51541137695312, 4.896282196044922, 187.4095458984375, 151.54339599609375, 144.74639892578125, -16.170196533203125, 136.72830200195312, 51.31219482421875, 71.42864990234375, 186.27984619140625, 142.01980590820312, 130.40219116210938, -106.92266845703125, 81.7281494140625, -11.253684997558594, 18.965850830078125, 1.41693115234375, 35.69743347167969, 50.125732421875, 88.10433959960938, 109.26863098144531, -1.7404403686523438, 24.51690673828125, 125.88892364501953, 127.96133422851562, 93.24026489257812, 118.54278564453125, 4.742340087890625, 43.746337890625, -117.34381103515625, 46.63409423828125, 137.41744995117188, 15.562301635742188, 10.636260986328125, -17.48675537109375, 62.762474060058594, -105.53150177001953, 119.20973205566406, 165.70306396484375, 76.626708984375, 125.99444580078125, -194.47314453125, 101.57304382324219, 139.94796752929688, -31.134307861328125, 18.244766235351562, 153.75070190429688, -7.4585113525390625, 6.12713623046875, 30.9395751953125, 102.74459075927734, 132.94781494140625, 141.39610290527344, -147.88699340820312, -109.2545166015625, 158.0081787109375, 93.21925354003906, -63.451141357421875, 26.242431640625, -2.513427734375, 149.93356323242188, -5.009586334228516, 68.54440307617188, 39.729400634765625, 139.3430633544922, -3.6303329467773438, -177.29550170898438, 37.885929107666016, 21.818191528320312, 132.62014770507812, 119.37078857421875, 168.2670135498047, 124.55697631835938, -98.06655883789062, 76.35824584960938, -10.307846069335938, 3.1617813110351562, 294.013916015625, -37.1710205078125, 118.41175842285156, 154.5792236328125, 121.81858825683594, 138.89720153808594, 1.4266166687011719, 58.928924560546875, 15.44921875, 61.2020263671875, -23.2991943359375, 48.25459289550781, 100.81283569335938, 52.77740478515625, -5.451297760009766, 120.56910705566406, 197.40185546875, 33.10394287109375, 117.7498779296875, -151.4873046875, 11.44912338256836, -19.40601348876953, -37.1556396484375, 119.52212524414062, 5.058612823486328], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000433.npy"} {"epoch": 0.9068062827225131, "step": 434, "batch_size": 128, "mean": 58.18274688720703, "std": 87.51710510253906, "min": -177.802001953125, "p10": -25.009401893615717, "median": 32.50444030761719, "p90": 166.26902465820314, "max": 234.84661865234375, "pos_frac": 0.7109375, "sample": [74.33302307128906, -2.045989990234375, -6.93853759765625, 5.785980224609375, 164.7912139892578, 21.440597534179688, 84.82172393798828, 132.0912322998047, -19.3857421875, 229.9468994140625, -10.466020584106445, 153.86778259277344, -84.69549560546875, 2.478912353515625, -11.066665649414062, 7.7430419921875, -21.04998779296875, 127.28419494628906, 103.23833465576172, 0.9608154296875, 10.59417724609375, 223.4259033203125, -8.0164794921875, 234.84661865234375, 164.45285034179688, -80.69110107421875, 0.0, -13.978172302246094, 39.49311065673828, 117.79666137695312, 194.0777587890625, 146.91400146484375, -20.29570770263672, 100.05712890625, 132.72845458984375, 141.30665588378906, -9.915283203125, 45.8455810546875, 26.4127197265625, 188.79107666015625, 45.2620849609375, 109.99380493164062, 176.83145141601562, -45.28143310546875, 33.34294128417969, -143.75750732421875, 124.87939453125, 138.68524169921875, 28.38330078125, 127.22161865234375, 169.34640502929688, 165.40992736816406, 116.49856567382812, 225.3968505859375, 31.665939331054688, 121.99681091308594, 152.39459228515625, 10.5367431640625, -43.679290771484375, 161.00946044921875, 23.622276306152344, 121.08502197265625, 199.686279296875, -19.74725341796875, 27.9017333984375, 30.635726928710938, 102.857421875, 161.31094360351562, 20.743316650390625, 20.67087745666504, 189.92430114746094, 5.525177001953125, -38.84991455078125, 173.28970336914062, -63.24444580078125, 2.9131736755371094, -8.9368896484375, 187.27359008789062, 6.5208740234375, 121.72271728515625, 156.2864990234375, -28.1334228515625, 64.73760986328125, 13.652069091796875, 45.555084228515625, 149.69711303710938, 160.53860473632812, -177.802001953125, 93.54519653320312, -22.296295166015625, 65.95272827148438, -17.269577026367188, -115.25186157226562, 0.0, 144.7664794921875, 26.26105499267578, 20.684173583984375, 166.1405029296875, 166.56890869140625, 12.717643737792969, -14.399749755859375, -5.1164703369140625, -27.581518173217773, 115.19891357421875, -9.22119140625, 9.766780853271484, -6.597349166870117, -23.907066345214844, 149.88751220703125, 61.52238464355469, 41.206634521484375, 3.8970870971679688, 0.0, 111.94430541992188, 74.70001220703125, -8.708099365234375, 23.614852905273438, 146.56805419921875, 95.30833435058594, 71.05978393554688, -177.29006958007812, 13.520599365234375, 26.78326416015625, -8.385299682617188, -116.130859375, 165.510498046875, 103.4224853515625, 146.44412231445312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000434.npy"} {"epoch": 0.9089005235602095, "step": 435, "batch_size": 128, "mean": 52.26315689086914, "std": 84.72435760498047, "min": -172.48605346679688, "p10": -37.69355773925779, "median": 35.765296936035156, "p90": 162.36435394287108, "max": 219.17767333984375, "pos_frac": 0.734375, "sample": [131.16238403320312, 14.027191162109375, 6.106006622314453, 46.300621032714844, 32.24519348144531, 179.13832092285156, 152.65432739257812, -120.51710510253906, 129.65431213378906, -5.50274658203125, -10.24032974243164, -23.031707763671875, -172.48605346679688, 43.711952209472656, 6.178853988647461, -20.57269287109375, 72.69476318359375, 69.01260375976562, 161.00479125976562, 115.93295288085938, 16.672943115234375, 25.96929359436035, 164.18753051757812, 0.0, 3.1286849975585938, -15.682159423828125, 69.1737060546875, 25.006553649902344, 130.02645874023438, 16.41497802734375, 138.77767944335938, 151.9236602783203, 111.07559204101562, -54.45550537109375, -8.486625671386719, 196.9739990234375, 116.87493896484375, 92.9136962890625, 42.587005615234375, 12.2186279296875, 169.70367431640625, 168.8130340576172, 168.34243774414062, 28.517044067382812, 139.3228759765625, 23.159454345703125, 145.73162841796875, -29.9736328125, 161.58299255371094, -12.4605712890625, 142.56307983398438, 89.09429931640625, 193.65484619140625, 144.1202392578125, 28.69443702697754, 4.695549011230469, 44.452880859375, 22.20721435546875, 94.15216064453125, -83.95562744140625, 12.149093627929688, 192.69476318359375, 9.91793441772461, -150.54457092285156, 16.616302490234375, 16.1014404296875, 39.285400390625, 167.86474609375, -125.27337646484375, -50.05400085449219, 25.257965087890625, -3.1504364013671875, -30.22735595703125, 183.97695922851562, 178.97552490234375, -21.52142333984375, 141.55162048339844, 158.40223693847656, -93.24899291992188, 127.74348449707031, -2.728759765625, 141.18374633789062, 60.8258056640625, 143.46446228027344, -114.41238403320312, 11.903228759765625, 77.31390380859375, 3.970438003540039, -1.2410430908203125, -1.072296142578125, -9.216346740722656, 137.61923217773438, 99.9581298828125, 219.17767333984375, 107.62712097167969, 158.91943359375, 97.07865905761719, 109.9693603515625, -23.10028839111328, 58.0494384765625, -6.86529541015625, 40.19288635253906, 85.25735473632812, 3.5197906494140625, 30.485137939453125, 15.7628173828125, 39.362152099609375, 8.335006713867188, 133.25660705566406, 6.6688690185546875, -9.269317626953125, -63.20643615722656, -86.46115112304688, -57.94921112060547, -1.21722412109375, 206.72283935546875, 8.42266845703125, 9.23187255859375, 101.28010559082031, -139.15219116210938, 48.91217041015625, 130.1591796875, 146.62332153320312, 0.9751739501953125, -32.39622497558594, 143.5657958984375, 43.3779296875, 127.08992004394531], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000435.npy"} {"epoch": 0.9109947643979057, "step": 436, "batch_size": 128, "mean": 37.98553466796875, "std": 83.2587890625, "min": -170.05093383789062, "p10": -52.697348785400386, "median": 22.237472534179688, "p90": 160.28472900390625, "max": 212.014404296875, "pos_frac": 0.6875, "sample": [53.4971923828125, -45.4200439453125, 197.80398559570312, 135.52769470214844, 3.621061325073242, 30.292213439941406, 158.8027801513672, 56.42362976074219, -33.8206787109375, 97.6383056640625, -101.87588500976562, 57.9451904296875, -64.16378784179688, 34.560150146484375, 0.085357666015625, 185.018798828125, 30.6685791015625, -19.63054656982422, -99.24951171875, -81.5615234375, 3.7181472778320312, 121.9364013671875, 131.32205200195312, 127.97503662109375, 174.63882446289062, -51.22570037841797, 146.083251953125, -0.712188720703125, 138.40402221679688, 19.94219970703125, 68.71302032470703, 25.240188598632812, -4.417642593383789, 22.8387451171875, 163.62559509277344, 124.20074462890625, -7.132698059082031, 39.659934997558594, 1.929351806640625, -63.61372375488281, 124.64090728759766, 12.31781005859375, 18.085304260253906, 75.11448669433594, 167.10464477539062, 27.99298095703125, 4.417724609375, 7.903900146484375, -20.324554443359375, 159.4517822265625, 24.15606689453125, -23.109710693359375, 26.042572021484375, -13.18487548828125, 108.26547241210938, -29.74890899658203, 178.78973388671875, -170.05093383789062, 5.052886962890625, 162.228271484375, 147.01699829101562, 44.29693603515625, 12.530120849609375, 98.49533081054688, 50.12420654296875, 136.24343872070312, -28.007171630859375, -114.89602661132812, 65.62582397460938, 135.3503875732422, 0.9078216552734375, 211.9219970703125, 82.15835571289062, -117.50981140136719, 21.686492919921875, 29.38671875, 15.999670028686523, 127.0185317993164, 10.83038330078125, -2.52520751953125, 73.91671752929688, -15.112701416015625, 15.852752685546875, -6.241355895996094, 60.95721435546875, 114.42715454101562, -2.72027587890625, -15.93731689453125, 39.63082504272461, -69.79510498046875, 166.956298828125, 8.837226867675781, -4.854579925537109, -56.131195068359375, 14.749847412109375, 22.7884521484375, 212.014404296875, -1.8757171630859375, 4.355772018432617, -31.713897705078125, -14.033447265625, -165.28912353515625, 16.857879638671875, 84.144775390625, 89.1595458984375, 9.94549560546875, -131.80255126953125, 143.30703735351562, 211.2589111328125, -13.13140869140625, 11.794715881347656, 28.71393585205078, -27.262451171875, 112.13046264648438, 30.138229370117188, 123.56675720214844, 10.65826416015625, 195.355224609375, -50.03584671020508, 10.960968017578125, -37.7584228515625, 46.4100341796875, -151.38320922851562, 81.77810668945312, -26.04327392578125, 173.92025756835938, 31.492568969726562, -13.898681640625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000436.npy"} {"epoch": 0.9130890052356021, "step": 437, "batch_size": 128, "mean": 63.12446594238281, "std": 89.9930191040039, "min": -158.95452880859375, "p10": -27.63662948608398, "median": 58.280954360961914, "p90": 175.4414276123047, "max": 229.14639282226562, "pos_frac": 0.7578125, "sample": [26.37451171875, 159.4056396484375, 42.4364013671875, 139.16607666015625, -158.95452880859375, 192.73504638671875, 3.0279541015625, 7.7375946044921875, 77.1392822265625, 11.8238525390625, 168.67776489257812, 21.353866577148438, 132.67189025878906, 69.61468505859375, 11.279129028320312, 148.90103149414062, -37.56109619140625, 4.182464599609375, 45.9290771484375, 147.10421752929688, 40.9637451171875, 182.45523071289062, 229.14639282226562, 163.5710906982422, 11.340217590332031, 27.528411865234375, -4.11376953125, 143.90457153320312, 145.5379638671875, 95.32943725585938, -17.52874755859375, 1.307830810546875, 6.409820556640625, 90.613525390625, -2.084197998046875, 28.64997673034668, -134.41004943847656, -23.720718383789062, 10.003585815429688, 29.728271484375, 145.8328399658203, 59.51214599609375, 166.8280029296875, 160.56436157226562, -43.614349365234375, 5.565155029296875, 28.52520751953125, 17.80517578125, 181.07977294921875, 102.42529296875, 133.42523193359375, 197.47337341308594, 128.8544464111328, 159.48995971679688, 8.869670867919922, 16.37628173828125, -25.818023681640625, -16.377262115478516, 185.5560302734375, 175.40545654296875, 155.54400634765625, -2.565216064453125, 169.70880126953125, 186.7735595703125, 211.61721801757812, 99.11984252929688, 2.9104843139648438, 70.33111572265625, -17.494033813476562, -19.566192626953125, 57.04976272583008, 5.249603271484375, -36.306060791015625, 172.02792358398438, -153.56771850585938, 154.3980712890625, -1.4322900772094727, 109.11549377441406, -149.84857177734375, 0.0, 160.54638671875, -77.88868713378906, 6.999755859375, -6.767063140869141, 175.52536010742188, 134.587646484375, -73.23455810546875, 78.65841674804688, -135.85629272460938, 87.93183898925781, 185.11376953125, 117.37322998046875, -2.9058685302734375, 128.01190185546875, 9.009918212890625, -3.988525390625, 122.83749389648438, 25.72736358642578, 83.97567749023438, 114.94618225097656, 177.8152313232422, 180.031494140625, 118.28536987304688, 172.2950439453125, 31.58489990234375, -31.880043029785156, 73.61813354492188, 115.16353607177734, 161.85546875, -127.07145690917969, 3.7883758544921875, 0.714752197265625, 142.54220581054688, -146.6258544921875, -8.600311279296875, 148.06472778320312, 67.8182373046875, 154.08067321777344, 128.06353759765625, -1.7475128173828125, 88.16400146484375, -13.226654052734375, 26.01348876953125, 71.38916015625, 0.0, 12.7333984375, 141.83912658691406, 224.09625244140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000437.npy"} {"epoch": 0.9151832460732985, "step": 438, "batch_size": 128, "mean": 29.452003479003906, "std": 89.81289672851562, "min": -163.06729125976562, "p10": -90.8755630493164, "median": 12.051177978515625, "p90": 150.51754150390624, "max": 223.9576416015625, "pos_frac": 0.6015625, "sample": [8.094932556152344, -161.29684448242188, 119.25387573242188, -11.197982788085938, -58.562644958496094, 148.98809814453125, 52.459747314453125, 36.398895263671875, 0.0, 11.542037963867188, 175.00526428222656, -128.78936767578125, 16.74195098876953, 150.12295532226562, 39.45098876953125, 12.657119750976562, 18.4046630859375, 154.16859436035156, 125.77490997314453, -20.541397094726562, -40.559600830078125, 0.22747802734375, 21.582427978515625, -1.0797958374023438, -107.020751953125, -70.17194366455078, -19.7039794921875, 123.19967651367188, 8.702880859375, 25.018310546875, 14.737314224243164, 172.64895629882812, 122.69094848632812, -11.347640991210938, 30.595947265625, 138.13055419921875, -143.75454711914062, -29.662857055664062, 48.6787109375, -160.60113525390625, -163.06729125976562, -126.37310791015625, 97.45648193359375, 126.90322875976562, -12.933212280273438, -9.486053466796875, 2.42578125, 25.483245849609375, -125.05685424804688, -1.3199653625488281, -1.470367431640625, -138.7206573486328, -5.2137451171875, 11.985214233398438, 5.9600830078125, 36.35191345214844, 112.11865234375, -19.93798828125, 112.12496948242188, 92.5760269165039, 12.117141723632812, -28.124956130981445, -1.1717510223388672, -147.25271606445312, 96.1883544921875, -14.899703979492188, -11.84527587890625, 0.365814208984375, 166.48385620117188, -118.87466430664062, 77.0614013671875, -74.97120666503906, 165.96701049804688, -14.566001892089844, 26.0732421875, 79.71380615234375, 127.74241638183594, -87.66603088378906, 151.24591064453125, 146.66256713867188, -30.414276123046875, 146.64488220214844, -72.8195571899414, -1.710693359375, 193.47076416015625, -14.106185913085938, 139.44998168945312, 141.67343139648438, -109.009765625, 13.331619262695312, 131.2099609375, 15.230262756347656, 2.5686397552490234, 14.6923828125, 169.091796875, 140.03109741210938, -0.5410690307617188, 164.96994018554688, 217.6397705078125, -25.550750732421875, -15.703475952148438, -27.069976806640625, 29.593505859375, 113.66554260253906, 133.8128662109375, -25.8480224609375, 11.525787353515625, 24.64592742919922, -20.223968505859375, 144.7538299560547, 170.26321411132812, 10.339006423950195, -83.84640502929688, 16.813594818115234, 0.0, -0.761199951171875, 82.24857330322266, 9.090461730957031, -25.774269104003906, 118.37306213378906, 167.86309814453125, -98.36447143554688, 7.60710334777832, 150.20538330078125, 223.9576416015625, -5.086341857910156, 12.670196533203125, 28.211212158203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000438.npy"} {"epoch": 0.9172774869109948, "step": 439, "batch_size": 128, "mean": 51.448936462402344, "std": 85.85990905761719, "min": -191.517333984375, "p10": -49.36503601074219, "median": 30.005504608154297, "p90": 164.44971618652343, "max": 212.38424682617188, "pos_frac": 0.7109375, "sample": [-69.91912841796875, -101.10946655273438, -141.04409790039062, 193.03500366210938, -57.51898193359375, 163.49038696289062, 15.7503662109375, 18.051963806152344, 59.23915100097656, 150.395751953125, -3.2809276580810547, 19.88250732421875, 28.010053634643555, 4.791473388671875, 81.78179931640625, -129.818603515625, 10.3121337890625, 113.15362548828125, -2.5556793212890625, 62.567291259765625, 184.053955078125, 150.84747314453125, 9.618621826171875, 25.56756591796875, 3.545562744140625, 144.96786499023438, 44.63465881347656, 145.58773803710938, 168.52389526367188, 20.0623722076416, -96.06028747558594, 112.67056274414062, 12.141685485839844, 195.212158203125, -20.950164794921875, 122.21096801757812, 195.62725830078125, 54.7308349609375, 212.38424682617188, 32.01710510253906, -33.0970458984375, 17.82424545288086, 148.4561767578125, 72.95701599121094, 7.7894287109375, 123.34291076660156, 141.1876220703125, 10.74615478515625, 3.46026611328125, 0.23504638671875, -7.8912353515625, 18.961387634277344, -50.1549072265625, 11.083892822265625, 43.134857177734375, 153.987548828125, 137.24000549316406, -25.142677307128906, 67.32976531982422, 0.0, 27.157485961914062, 0.0, -53.80792236328125, 27.126937866210938, -33.53497314453125, 164.02792358398438, 23.515899658203125, -63.93700408935547, 10.896129608154297, -36.328460693359375, 3.459564208984375, 159.5352783203125, 35.9371337890625, -26.079681396484375, -7.903411865234375, 30.97801971435547, -12.304462432861328, 111.40706634521484, -42.19610595703125, 79.39102172851562, -49.026519775390625, 104.9846420288086, 146.39096069335938, 20.90399169921875, 61.58872985839844, 61.39851379394531, -41.76536560058594, -4.8994293212890625, -73.07250213623047, -3.336669921875, 159.74728393554688, -136.907958984375, 0.0, 172.41192626953125, 115.7269287109375, 99.30350494384766, 185.59701538085938, 147.48046875, 42.14019775390625, -31.602279663085938, 43.6417236328125, -68.88566589355469, 136.64512634277344, 12.497848510742188, 54.416351318359375, 29.032989501953125, 105.602783203125, 125.07310485839844, 132.468505859375, -39.50555419921875, 162.31581115722656, 75.0211181640625, 69.97013854980469, 165.43389892578125, 146.0755615234375, -0.5598583221435547, -191.517333984375, 185.36444091796875, 112.89016723632812, 181.6141357421875, 161.3720703125, 132.14248657226562, 20.382667541503906, -21.38819122314453, 181.65467834472656, 0.0, 171.93746948242188, 153.30426025390625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000439.npy"} {"epoch": 0.9193717277486911, "step": 440, "batch_size": 128, "mean": 39.552093505859375, "std": 88.90399932861328, "min": -178.92584228515625, "p10": -83.63034667968749, "median": 21.791213989257812, "p90": 163.60698089599606, "max": 233.7713623046875, "pos_frac": 0.6796875, "sample": [-36.567596435546875, 0.0, 31.546142578125, -139.13616943359375, -18.501007080078125, 64.24057006835938, -142.05633544921875, 109.97232055664062, -11.361320495605469, 111.59600830078125, -11.61834716796875, 19.870635986328125, -103.86175537109375, 0.32635498046875, 34.429046630859375, -122.23072814941406, 100.95826721191406, 144.734375, 127.59013366699219, 110.4898681640625, 49.353271484375, 180.03216552734375, -2.07647705078125, 22.025634765625, -73.9322509765625, -0.7343521118164062, -178.92584228515625, 116.00163269042969, 46.24896240234375, 117.12762451171875, 179.80224609375, 201.523193359375, 2.4672927856445312, -10.44873046875, -19.20159912109375, 20.624664306640625, 92.01081848144531, -153.05963134765625, 0.0, 55.25732421875, -159.165771484375, -0.630401611328125, 9.18719482421875, 91.1755599975586, 169.03895568847656, -24.27782440185547, -100.4124755859375, -0.0152587890625, 15.602325439453125, 143.76097106933594, 195.89309692382812, -12.899330139160156, 132.2403106689453, 0.8792343139648438, 38.999176025390625, 179.14230346679688, -2.814300537109375, -126.13294982910156, -52.958038330078125, -30.909576416015625, -49.188232421875, 1.09552001953125, 8.893463134765625, -37.85882568359375, 233.7713623046875, 62.2796630859375, 197.10537719726562, 132.1195526123047, 124.39007568359375, 20.760787963867188, 0.051708221435546875, 22.71514892578125, 13.0548095703125, 139.90582275390625, 154.40274047851562, 9.068435668945312, -2.58984375, -90.18759155273438, 17.007598876953125, 15.902130126953125, 32.85008239746094, 117.98011779785156, 142.378173828125, 71.03396606445312, 185.13088989257812, -97.243896484375, 20.68194580078125, 179.6873779296875, 35.250244140625, -93.42477416992188, 184.33273315429688, -90.74673461914062, 147.05722045898438, -33.337310791015625, 137.24374389648438, 52.902069091796875, 101.38272094726562, 29.882549285888672, 117.81423950195312, 42.2252197265625, 161.27899169921875, -17.067916870117188, 19.45751953125, 19.71923828125, 48.65081787109375, 69.53668212890625, 6.3011322021484375, 177.86785888671875, 88.87194061279297, 83.43356323242188, 11.71954345703125, 45.750999450683594, 21.63525390625, 155.93112182617188, -38.863861083984375, 133.92819213867188, 104.58847045898438, 21.947174072265625, 7.0935821533203125, 106.54257202148438, -32.63954162597656, 175.8973388671875, 31.898712158203125, -80.82009887695312, -12.421905517578125, 14.66656494140625, -40.87213897705078, 142.63577270507812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000440.npy"} {"epoch": 0.9214659685863874, "step": 441, "batch_size": 128, "mean": 62.93877410888672, "std": 80.59461975097656, "min": -168.35008239746094, "p10": -12.78462562561035, "median": 46.23578643798828, "p90": 171.31779022216796, "max": 210.1416015625, "pos_frac": 0.78125, "sample": [-11.563461303710938, 34.12762451171875, -128.97451782226562, 115.7603759765625, -0.21188926696777344, -12.337982177734375, 185.67410278320312, -5.539794921875, 4.428070068359375, -6.299560546875, -48.06036376953125, 210.1416015625, 25.531593322753906, 133.087158203125, 26.6064453125, 141.155517578125, 154.70330810546875, 151.29129028320312, 131.36993408203125, 69.19441223144531, 54.3587646484375, 131.25588989257812, 13.174701690673828, 18.076873779296875, 126.14227294921875, 71.62809753417969, 135.33319091796875, 161.55540466308594, 91.2718505859375, 5.51116943359375, 12.8016357421875, 148.4150390625, 86.66966247558594, -13.826793670654297, 141.1947784423828, 104.60736083984375, 14.7166748046875, 163.17642211914062, 128.95030212402344, -68.93811798095703, 158.28445434570312, 143.73941040039062, -31.462501525878906, -3.33111572265625, 185.99383544921875, -1.6954078674316406, 45.297637939453125, 36.68827819824219, 143.58660888671875, 135.35400390625, 141.7296600341797, 5.54986572265625, 49.217071533203125, 18.04175567626953, 18.084014892578125, 14.53656005859375, -2.02783203125, 140.61502075195312, 31.958541870117188, -11.433624267578125, 1.99420166015625, 28.864105224609375, 8.059967041015625, 143.2620849609375, 10.032257080078125, 0.0, 9.239501953125, 88.37960815429688, 47.17393493652344, 194.09756469726562, -66.34146118164062, -64.08181762695312, -0.5908851623535156, 10.018936157226562, 0.0, 139.26622009277344, 40.18310546875, 34.2152099609375, 158.75027465820312, 85.91818237304688, 14.210281372070312, 152.0125732421875, 26.680191040039062, -58.710723876953125, 50.271759033203125, 97.206787109375, 140.92510986328125, 89.800048828125, 1.83270263671875, 23.28826904296875, 1.403717041015625, 62.026123046875, 38.233917236328125, 121.2578125, 139.10507202148438, 12.77471923828125, 171.27403259277344, -96.13957977294922, 175.926513671875, 51.876708984375, 171.41989135742188, -168.35008239746094, 60.299346923828125, 5.397857666015625, -100.85958862304688, 29.604934692382812, 193.226318359375, -110.21405029296875, 154.02561950683594, -4.646697998046875, 29.45685577392578, 177.37014770507812, 175.48928833007812, 174.5560760498047, 184.85833740234375, 109.25991821289062, 81.6254653930664, -5.035400390625, -31.0262451171875, 10.75115966796875, 209.45330810546875, -6.260841369628906, 145.08773803710938, 121.53536987304688, 174.53515625, 107.98994445800781, 134.68338012695312, 33.34765625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000441.npy"} {"epoch": 0.9235602094240838, "step": 442, "batch_size": 128, "mean": 47.666778564453125, "std": 96.55935668945312, "min": -197.40338134765625, "p10": -69.21946563720701, "median": 33.90130615234375, "p90": 169.6794860839844, "max": 246.0108642578125, "pos_frac": 0.7109375, "sample": [139.9041748046875, 13.08526611328125, 20.755569458007812, 45.455501556396484, 210.2989501953125, 162.67356872558594, 1.5314311981201172, -11.205078125, 170.8597412109375, 3.961334228515625, 93.09501647949219, 24.79486083984375, -86.13582611083984, -9.804275512695312, 149.07302856445312, -146.21798706054688, 21.763626098632812, 21.21087646484375, 27.73504638671875, -132.76808166503906, 14.85595703125, 39.393882751464844, -22.243736267089844, -8.61073112487793, 124.44708251953125, -6.8466949462890625, 129.1546630859375, -59.037506103515625, 187.42520141601562, 175.2596435546875, -144.36419677734375, 3.4019622802734375, -40.2913818359375, 148.3632354736328, 105.26963806152344, 145.96336364746094, -8.756973266601562, 24.564666748046875, -143.4139404296875, 8.394058227539062, 27.204055786132812, -92.04656219482422, 116.03668212890625, 11.043571472167969, -24.45477294921875, -163.35330200195312, 17.504226684570312, 37.6142578125, 9.6246337890625, -45.23681640625, 16.7318115234375, -1.1141357421875, 102.43037414550781, 205.04034423828125, 7.29945182800293, 101.4278564453125, 137.8340301513672, 0.0, 84.35964965820312, -141.03915405273438, -57.813720703125, 11.122489929199219, -150.2642364501953, 147.9011688232422, -150.32858276367188, 179.0555419921875, 107.23812866210938, -2.6870880126953125, 66.43185424804688, 132.0719757080078, -0.017137527465820312, -197.40338134765625, 178.25958251953125, -6.33074951171875, 123.74697875976562, 56.523468017578125, 200.96078491210938, -22.7410888671875, 123.81025695800781, 27.214492797851562, 31.914459228515625, 75.96548461914062, -8.458934783935547, 206.30752563476562, 144.46170043945312, 157.7759246826172, -52.5084228515625, -138.45831298828125, -48.5020751953125, -61.96959686279297, 20.23699951171875, 69.10454559326172, 11.346099853515625, 112.6787109375, 17.263275146484375, 139.55010986328125, 117.49433898925781, 141.37203979492188, 111.65177917480469, 152.201416015625, 181.1915283203125, 169.8699951171875, 155.65524291992188, 169.59783935546875, 43.39985656738281, 246.0108642578125, 11.5836181640625, -19.13304901123047, -31.612716674804688, 202.33828735351562, 129.01821899414062, 139.46566772460938, 69.41217041015625, 136.35520935058594, 5.4788055419921875, 35.888153076171875, 49.184532165527344, 64.26239013671875, 137.05612182617188, -57.95819091796875, 62.33131408691406, 143.59646606445312, 159.26336669921875, 25.444793701171875, -140.4947967529297, 56.239654541015625, 137.8647003173828, 53.9588623046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000442.npy"} {"epoch": 0.9256544502617801, "step": 443, "batch_size": 128, "mean": 49.118621826171875, "std": 85.96318054199219, "min": -173.38449096679688, "p10": -47.78830184936522, "median": 29.559188842773438, "p90": 163.66071319580075, "max": 225.61114501953125, "pos_frac": 0.6953125, "sample": [55.53617858886719, 9.89447021484375, 101.29209899902344, 22.23883056640625, -56.988067626953125, 52.11065673828125, 183.6886749267578, 17.310516357421875, 177.75787353515625, -13.746055603027344, 20.195899963378906, 100.92889404296875, -9.499061584472656, 124.09097290039062, -10.58953857421875, -91.17866516113281, -16.94659423828125, 117.655029296875, 161.74526977539062, -5.994266510009766, 42.57093811035156, 82.14605712890625, 65.72380065917969, 40.2498779296875, -6.87274169921875, 23.263931274414062, 128.22921752929688, 150.07266235351562, 33.04792785644531, -41.56450653076172, -27.114990234375, 60.978668212890625, -150.9844970703125, 5.3579864501953125, -10.305755615234375, 29.223724365234375, 143.0238037109375, 32.188018798828125, 160.56396484375, 11.24703598022461, 169.95919799804688, -16.458343505859375, -9.348846435546875, -0.715545654296875, 17.901100158691406, 35.18798828125, 225.61114501953125, 3.6764068603515625, 95.462890625, -56.1087646484375, 66.944091796875, -167.12820434570312, -58.58488464355469, -73.659912109375, 84.99310302734375, 140.78672790527344, 100.10209655761719, 216.26568603515625, 77.925537109375, 182.2755126953125, 119.14501953125, 29.8946533203125, -11.791816711425781, 137.22799682617188, 137.96200561523438, 20.386253356933594, -12.05328369140625, 92.31267547607422, 134.46116638183594, 143.36093139648438, -6.0963897705078125, -6.50689697265625, 180.93902587890625, 12.935638427734375, 168.1300811767578, -2.825958251953125, -1.5369949340820312, 125.43548583984375, -102.31135559082031, 10.469207763671875, -25.818756103515625, 168.919677734375, 161.50314331054688, 92.66231536865234, 70.14495849609375, -44.222389221191406, 15.026161193847656, -74.0820541381836, 135.21424865722656, 199.259765625, 183.8746337890625, 117.91134643554688, 28.611358642578125, 22.939117431640625, -70.57099914550781, 127.45953369140625, 10.3548583984375, 193.09194946289062, 122.77793884277344, 147.490478515625, 80.570556640625, -33.494110107421875, -152.951171875, 160.7468719482422, -2.578704833984375, 0.95257568359375, 94.71214294433594, 128.27301025390625, 58.143463134765625, 9.143081665039062, 68.65803527832031, 8.28692626953125, -38.038238525390625, -3.6391525268554688, 0.4925079345703125, -10.841331481933594, 78.9212646484375, 170.5725555419922, 28.031715393066406, -173.38449096679688, 147.02374267578125, 146.16049194335938, 131.53097534179688, 5.627708435058594, -2.533447265625, -130.81082153320312, 10.818748474121094, 9.068389892578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000443.npy"} {"epoch": 0.9277486910994764, "step": 444, "batch_size": 128, "mean": 53.37458038330078, "std": 90.36863708496094, "min": -192.0924072265625, "p10": -38.649848937988274, "median": 33.607688903808594, "p90": 172.84418334960938, "max": 289.36932373046875, "pos_frac": 0.703125, "sample": [3.1383094787597656, 15.776046752929688, 172.6380615234375, 17.65771484375, 73.49459838867188, 71.07192993164062, -23.176727294921875, 2.8194580078125, 21.735191345214844, 47.3616943359375, -10.082122802734375, 35.221405029296875, -32.628875732421875, 38.30291748046875, 122.29658508300781, 26.03997802734375, 133.81105041503906, 42.27610778808594, -25.4500732421875, 35.118797302246094, 109.44358825683594, 23.918670654296875, 134.592041015625, -92.2080078125, 18.992889404296875, -34.208221435546875, 151.69677734375, 13.267135620117188, 6.036651611328125, 26.891204833984375, 17.00555419921875, 148.92361450195312, 100.95881652832031, -16.025588989257812, 166.91416931152344, -29.524505615234375, -22.151290893554688, 59.35382080078125, 160.15716552734375, 142.4345703125, 158.58389282226562, 199.8154296875, -14.8641357421875, -4.704166412353516, 42.1180419921875, 142.52822875976562, 51.06640625, -133.79266357421875, -122.48129272460938, -37.2830810546875, 216.1846923828125, 130.26565551757812, -67.49288940429688, 199.8043212890625, -4.148220062255859, 70.069091796875, 225.33712768554688, 173.32513427734375, -0.81292724609375, -11.401763916015625, 1.8543148040771484, 289.36932373046875, 137.0640869140625, -34.83418273925781, 75.59388732910156, -8.345001220703125, 183.44793701171875, 116.59977722167969, -83.280029296875, 205.20477294921875, 141.02557373046875, 30.643569946289062, -86.0653076171875, 173.97158813476562, -16.517730712890625, 60.46875, 183.3828125, 159.59530639648438, 139.15020751953125, 75.05206298828125, 168.95547485351562, -9.184904098510742, 1.2515296936035156, 188.18185424804688, -11.236888885498047, 153.14859008789062, -9.320137023925781, 143.17539978027344, 138.99880981445312, -96.26660919189453, 2.510589599609375, 185.46401977539062, -21.020767211914062, 99.28273010253906, 40.982704162597656, 28.6705322265625, 46.1614990234375, -3.270263671875, 10.459793090820312, -58.590057373046875, -20.49560546875, 7.5953216552734375, 172.55850219726562, 4.807121276855469, -95.58544921875, 49.3836669921875, -52.78538513183594, 148.36404418945312, -41.83897399902344, 134.64967346191406, 5.932334899902344, 72.30136108398438, 9.227005004882812, 124.04914855957031, 97.04953002929688, 32.096580505371094, 174.19769287109375, 136.86849975585938, 143.01934814453125, 133.5153350830078, -3.3086318969726562, 12.463871002197266, -13.2763671875, 0.05390357971191406, -169.05020141601562, 25.6551513671875, 128.84744262695312, -192.0924072265625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000444.npy"} {"epoch": 0.9298429319371728, "step": 445, "batch_size": 128, "mean": 42.65665817260742, "std": 88.28392791748047, "min": -160.63592529296875, "p10": -56.25133666992187, "median": 23.714235305786133, "p90": 168.10211791992188, "max": 216.57015991210938, "pos_frac": 0.6875, "sample": [9.608184814453125, -160.0161590576172, 184.50213623046875, 79.01924133300781, 41.302947998046875, 188.18243408203125, -4.9369049072265625, 48.711090087890625, 169.63385009765625, -107.71157836914062, 9.638038635253906, 76.55009460449219, 152.20326232910156, -63.0721435546875, -27.549560546875, 5.05560302734375, 29.560455322265625, 15.470428466796875, 200.09674072265625, 35.566001892089844, 171.32833862304688, -5.95159912109375, 19.826393127441406, -111.84568786621094, 38.786346435546875, 132.03558349609375, 56.45606994628906, -2.4254684448242188, 144.5941162109375, 23.50830078125, 194.06292724609375, 84.35848999023438, 37.2451171875, 0.04347991943359375, -26.366058349609375, -12.733657836914062, 14.360668182373047, 41.3502311706543, 115.88892364501953, 72.68722534179688, 9.61810302734375, 169.2423095703125, 188.66488647460938, -1.2789306640625, 100.39012145996094, 2.894655227661133, 0.1650238037109375, 33.5740966796875, 38.05366516113281, -127.06387329101562, -36.65886688232422, 66.53900146484375, 119.88861083984375, -62.73851013183594, 138.81570434570312, 0.44329071044921875, 50.1248779296875, 172.04293823242188, 16.758056640625, 150.5358123779297, 102.35476684570312, 45.575714111328125, 216.57015991210938, 5.3712158203125, 156.63182067871094, 122.05728149414062, 113.85533142089844, -157.8512420654297, 52.291656494140625, 191.45294189453125, 26.41650390625, 82.76031494140625, -22.818984985351562, 134.04556274414062, 136.95956420898438, -0.4760284423828125, -17.351165771484375, -141.94705200195312, 160.2513427734375, 137.99798583984375, 0.073333740234375, -67.76773071289062, -44.41358947753906, -130.0506591796875, 4.86882209777832, 26.28997802734375, 16.429916381835938, -54.93658447265625, 105.49254608154297, 0.43959808349609375, 97.6392822265625, 137.4674072265625, 187.91696166992188, 151.2169189453125, 42.264404296875, 139.67889404296875, -3.2797412872314453, -141.41534423828125, 171.57254028320312, -2.5208740234375, 6.240486145019531, 77.31195831298828, 166.36648559570312, 23.920169830322266, -6.877685546875, 16.854629516601562, 9.544837951660156, 19.315048217773438, 167.61346435546875, -2.112884521484375, -45.71881103515625, -10.418441772460938, 0.0, -14.104381561279297, 12.319305419921875, -36.851531982421875, -6.093719482421875, -3.5096435546875, 1.1176605224609375, 165.07974243164062, -54.7125244140625, -52.4945068359375, -160.63592529296875, -1.089630126953125, 56.799102783203125, 151.63650512695312, -59.319091796875, 159.7228240966797], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000445.npy"} {"epoch": 0.9319371727748691, "step": 446, "batch_size": 128, "mean": 59.67151641845703, "std": 88.34864044189453, "min": -199.84307861328125, "p10": -35.43881797790527, "median": 46.037384033203125, "p90": 173.752783203125, "max": 233.63833618164062, "pos_frac": 0.7578125, "sample": [127.72689819335938, 210.81048583984375, 10.164146423339844, 3.74102783203125, 1.2118148803710938, 42.577186584472656, 140.01358032226562, 55.79254913330078, 137.26585388183594, -15.405929565429688, 7.3657684326171875, -50.207733154296875, 110.70455932617188, 43.186729431152344, 4.37066650390625, 41.11573028564453, 199.86932373046875, 26.687698364257812, -17.44476318359375, 0.5155029296875, 19.086639404296875, 124.24237060546875, 145.8621826171875, 136.8790283203125, -7.045433044433594, -78.542724609375, 207.22842407226562, 16.581016540527344, 6.958980560302734, -77.77426147460938, 10.482513427734375, 12.6199951171875, -26.4979248046875, 34.582664489746094, -57.1646728515625, 30.83074951171875, 161.60491943359375, 3.9350662231445312, 19.15911865234375, 49.39476776123047, 113.60958862304688, 3.2336502075195312, 99.48140716552734, 21.403060913085938, 45.7183837890625, 138.33615112304688, 174.43475341796875, 167.7628173828125, 166.24124145507812, -199.84307861328125, 114.093017578125, 115.212890625, 199.4622802734375, 3.8393115997314453, 134.63841247558594, 147.83297729492188, 5.1898193359375, 46.35638427734375, -188.10275268554688, 108.46420288085938, 173.46051025390625, -1.2422904968261719, -34.239044189453125, 0.0, 43.394378662109375, 158.89385986328125, 55.286529541015625, 105.20829772949219, 93.68505859375, 145.67674255371094, -10.619659423828125, 151.35308837890625, 122.57855224609375, 145.29046630859375, 146.80703735351562, 11.42529296875, 161.97415161132812, 135.2808380126953, -34.539581298828125, 0.0, 80.44741821289062, -41.28190612792969, 196.01974487304688, -9.3662109375, 28.963333129882812, -70.13177490234375, 89.34054565429688, 207.78399658203125, 73.909423828125, 70.83843994140625, -28.82647705078125, 233.63833618164062, 197.05548095703125, 162.532470703125, 123.24520874023438, 57.46759033203125, 98.1728515625, 138.33514404296875, -37.53703689575195, 88.99896240234375, 19.27216339111328, -13.221893310546875, 191.52105712890625, 191.30075073242188, -181.275634765625, -138.5569305419922, -12.870773315429688, 12.3636474609375, -12.254669189453125, 4.07318115234375, 55.93388366699219, 0.0, -38.319061279296875, -5.64251708984375, 138.58372497558594, -45.493072509765625, 188.62606811523438, 80.12149047851562, 204.45962524414062, 15.895462036132812, 4.751129150390625, 54.89366149902344, 17.713851928710938, -11.545875549316406, 159.05035400390625, 62.81466293334961, 138.6815948486328, 97.94924926757812], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000446.npy"} {"epoch": 0.9340314136125655, "step": 447, "batch_size": 128, "mean": 63.851783752441406, "std": 80.13150787353516, "min": -173.36358642578125, "p10": -27.554870605468743, "median": 53.16579055786133, "p90": 164.02328186035157, "max": 214.55438232421875, "pos_frac": 0.765625, "sample": [-41.625823974609375, 9.913143157958984, 139.9468536376953, 141.66685485839844, 43.97747802734375, 8.787416458129883, 186.21295166015625, 14.3948974609375, -32.28363037109375, 154.38272094726562, 93.70970153808594, 133.17405700683594, 37.50921630859375, 172.02194213867188, 161.09915161132812, 67.36582946777344, 0.0, 11.674468994140625, 177.1451416015625, 11.38033676147461, -0.155914306640625, 33.74786376953125, 146.843505859375, 22.770233154296875, 78.19708251953125, -9.228515625, 2.5074462890625, 137.7936248779297, -133.80996704101562, 135.6224365234375, -106.15631103515625, 24.6131591796875, 124.90081787109375, 92.69586181640625, -13.388435363769531, 22.367958068847656, 161.2089385986328, 141.05557250976562, 36.39093017578125, 135.5224151611328, 7.737281799316406, 144.81243896484375, 173.79637145996094, 169.93048095703125, 39.4774169921875, -0.0441741943359375, 145.92556762695312, 146.50750732421875, 163.63818359375, 0.0, 156.90634155273438, 19.29827880859375, 49.545745849609375, 131.62454223632812, -34.99180603027344, -36.82722473144531, -10.78249740600586, 0.8988037109375, -101.87969207763672, 162.1043243408203, 11.579010009765625, 53.15448760986328, 7.677642822265625, 54.795379638671875, -19.984024047851562, 194.560546875, 134.13040161132812, -2.07220458984375, -33.964630126953125, 15.768417358398438, 66.78439331054688, -34.67877197265625, -25.52825927734375, 9.5789794921875, 83.47634887695312, 0.6018753051757812, 89.63571166992188, -45.0198974609375, 63.246826171875, 163.0996551513672, 214.55438232421875, -2.8646240234375, -3.6556358337402344, 91.47329711914062, -173.36358642578125, 0.0, 7.807945251464844, 137.32644653320312, 50.86393737792969, 163.93032836914062, 53.177093505859375, 37.295379638671875, 28.86004638671875, 129.508544921875, 201.07269287109375, 133.70065307617188, 152.27413940429688, 213.93756103515625, 0.0, 163.3029327392578, 62.83673095703125, 139.01898193359375, 28.4093017578125, 25.84979248046875, 36.513824462890625, 75.56314086914062, 29.278472900390625, -111.95843505859375, 166.17066955566406, 83.37112426757812, -17.318603515625, 134.48121643066406, 38.466522216796875, 164.24017333984375, 121.6212158203125, 86.28762817382812, 14.23004150390625, 130.65589904785156, -10.111618041992188, -8.13525104522705, 61.075836181640625, 167.248291015625, 144.0557403564453, 70.88131713867188, 73.0966796875, 176.73049926757812, -45.862335205078125, 98.65911102294922], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000447.npy"} {"epoch": 0.9361256544502617, "step": 448, "batch_size": 128, "mean": 55.10235595703125, "std": 84.47454071044922, "min": -145.65887451171875, "p10": -30.735872840881346, "median": 41.05787658691406, "p90": 167.00425872802734, "max": 225.0212860107422, "pos_frac": 0.75, "sample": [16.028274536132812, 9.028656005859375, -91.76988220214844, 11.64422607421875, 76.8140869140625, 175.4306640625, 5.221099853515625, -111.1490478515625, 108.09640502929688, 144.9727783203125, -24.7408447265625, -117.67233276367188, 68.99600219726562, 39.019927978515625, 1.1129150390625, 87.79985046386719, 138.33563232421875, 152.2760009765625, 80.0281982421875, -6.602237701416016, 72.5573959350586, 171.22463989257812, 225.0212860107422, 5.2913665771484375, 94.50958251953125, 38.18952178955078, 34.77754211425781, -27.017578125, 17.251953125, 7.962799072265625, -11.58203125, 41.377899169921875, 179.79031372070312, -101.33671569824219, 135.9296875, -40.885498046875, 153.71029663085938, 168.45269775390625, 38.29969787597656, 62.35296630859375, 125.36209106445312, 29.089691162109375, 178.26495361328125, -1.9998626708984375, 77.98150634765625, 5.649944305419922, 65.58668518066406, 135.09634399414062, -4.2155303955078125, 205.08685302734375, -11.04022216796875, -5.30718994140625, -6.8101806640625, 77.59970092773438, 2.6203079223632812, -11.345367431640625, 61.52484130859375, 129.05535888671875, -77.504638671875, 175.45632934570312, 137.19287109375, 64.94296264648438, 16.07598876953125, 0.2803497314453125, 164.33377075195312, 197.81387329101562, 19.787189483642578, 13.926773071289062, 158.48123168945312, 190.93511962890625, 154.27664184570312, 117.53715515136719, 117.25679779052734, 41.409454345703125, 12.504234313964844, 108.19683837890625, 171.53158569335938, 123.77215576171875, 150.6610107421875, 100.32614135742188, -54.31243896484375, 173.56460571289062, -0.6892662048339844, 154.53529357910156, 22.368881225585938, -140.506591796875, -136.91293334960938, -145.65887451171875, 153.38803100585938, 15.71746826171875, -13.371002197265625, -3.43780517578125, 143.21478271484375, 158.4738311767578, 22.905792236328125, -131.49667358398438, 58.14727783203125, -1.4917469024658203, 40.73785400390625, 103.64971923828125, 148.6744384765625, 14.151334762573242, 166.3834991455078, 7.972297668457031, 90.00765991210938, -28.924957275390625, -64.19927978515625, 80.31202697753906, 57.556549072265625, -25.238815307617188, 160.66822814941406, -32.60638427734375, -15.202880859375, 7.93359375, -29.93422508239746, 23.38494873046875, 8.78570556640625, 19.90155029296875, 53.98211669921875, 0.2627296447753906, 119.58643341064453, -7.6378173828125, 10.827533721923828, 159.88922119140625, 219.41558837890625, 124.29615783691406, 47.93499755859375, 75.95152282714844], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000448.npy"} {"epoch": 0.9382198952879581, "step": 449, "batch_size": 128, "mean": 63.145992279052734, "std": 85.28422546386719, "min": -180.13519287109375, "p10": -28.990294265747067, "median": 59.14097595214844, "p90": 173.09559631347656, "max": 209.95849609375, "pos_frac": 0.75, "sample": [142.55685424804688, 75.72372436523438, -39.23924255371094, 122.2375259399414, 58.41168212890625, -14.4443359375, 10.2923583984375, 186.23602294921875, -5.148368835449219, 9.252609252929688, 14.914825439453125, 19.90264892578125, 113.5084228515625, 38.66199493408203, -27.640792846679688, 137.24957275390625, 189.75, 173.64706420898438, 17.56256103515625, -2.2025623321533203, -119.98663330078125, 136.8392333984375, 167.8653564453125, 136.0179443359375, 55.72637939453125, 95.03289031982422, -61.729278564453125, 78.4405517578125, 5.848830223083496, 175.735595703125, 194.05267333984375, 169.18927001953125, 153.30767822265625, 181.3216552734375, 25.2501220703125, 133.0238037109375, 16.609710693359375, 82.18290710449219, 141.47525024414062, 127.18232727050781, 168.25868225097656, 33.93597412109375, 131.70980834960938, 15.043922424316406, 209.95849609375, 197.77560424804688, 1.835479736328125, -1.4202136993408203, 186.99722290039062, 155.01165771484375, -39.642127990722656, 87.27389526367188, 12.6416015625, 46.085418701171875, 7.949277877807617, 140.95201110839844, 11.745132446289062, 50.35273742675781, 4.02777099609375, 85.92854309082031, 4.970703125, 63.474517822265625, 157.00421142578125, 19.667724609375, 106.2513198852539, -17.989028930664062, 60.62261962890625, 59.870269775390625, -9.01531982421875, 157.34628295898438, -61.4515380859375, 93.70429992675781, 154.64840698242188, 148.2669219970703, -180.13519287109375, -1.4866485595703125, 2.633392333984375, 146.11541748046875, 63.24566650390625, 158.33160400390625, 144.58746337890625, 99.90516662597656, 205.41033935546875, 53.31568908691406, 117.87015533447266, 113.29727172851562, 157.6124267578125, 32.43353271484375, 124.203857421875, -28.294391632080078, -5.913322448730469, 38.79339599609375, -3.687255859375, 182.54132080078125, 162.9144287109375, 172.8592529296875, -100.16136169433594, -13.47918701171875, 89.73028564453125, 60.5777587890625, -36.9273681640625, 138.57086181640625, -30.61406707763672, -21.003616333007812, 85.49168395996094, -21.51007080078125, -90.64840698242188, -11.728981018066406, 146.05194091796875, 30.7672119140625, 6.558135986328125, 134.031494140625, -19.290206909179688, -2.7214088439941406, 38.326629638671875, -6.2340240478515625, 151.37997436523438, 85.50260925292969, 2.8248748779296875, 177.13690185546875, 36.362335205078125, -158.59934997558594, 178.65457153320312, 19.307044982910156, 108.2589111328125, -157.9163360595703, -45.35772705078125, -5.619171142578125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000449.npy"} {"epoch": 0.9403141361256544, "step": 450, "batch_size": 128, "mean": 47.307533264160156, "std": 86.08916473388672, "min": -166.40054321289062, "p10": -29.740428161621093, "median": 30.55150604248047, "p90": 168.94144592285156, "max": 209.55059814453125, "pos_frac": 0.703125, "sample": [0.0, -117.0609130859375, 110.90788269042969, -20.48419189453125, 65.70904541015625, 1.8863906860351562, 148.27769470214844, -12.63897705078125, 170.7916259765625, 47.7122802734375, 117.28887939453125, 183.56939697265625, -0.002674102783203125, 45.490631103515625, 147.81735229492188, 10.36496353149414, -6.78399658203125, 123.06982421875, 15.517852783203125, 46.694435119628906, -50.82579040527344, 100.98078918457031, 185.762939453125, 165.4092254638672, 118.02936553955078, 30.1348876953125, -14.724357604980469, 15.5594482421875, 100.19767761230469, 31.38201904296875, 7.022563934326172, -3.4969024658203125, -5.50927734375, 1.85980224609375, 56.07655334472656, 15.7738037109375, 201.91314697265625, -140.07003784179688, 107.02806091308594, -80.3582763671875, -2.5699386596679688, 169.28005981445312, 20.632610321044922, 127.26205444335938, 18.141799926757812, 30.651100158691406, -25.278684616088867, 75.15631103515625, 182.7586669921875, 61.65220642089844, 2.211456298828125, 155.6473846435547, 17.680309295654297, 9.9107666015625, 47.30650329589844, -6.79083251953125, 36.5540771484375, 183.7715301513672, 11.564605712890625, -132.43914794921875, 47.010406494140625, -166.40054321289062, 144.64913940429688, 49.64715576171875, -28.5916748046875, 177.76632690429688, 147.84718322753906, 3.2586193084716797, 130.4024658203125, 91.8067626953125, -3.265533447265625, 120.41151428222656, -0.796142578125, 167.86798095703125, 6.60107421875, 24.802505493164062, 169.66445922851562, 209.55059814453125, -126.35560607910156, 15.710952758789062, 113.58700561523438, 44.761871337890625, -29.076034545898438, 3.533233642578125, 177.74191284179688, 32.9798583984375, -1.1467781066894531, 14.013763427734375, 31.86761474609375, -26.74011993408203, 15.446624755859375, 100.46388244628906, 168.79632568359375, 204.5103759765625, 23.343032836914062, 150.60494995117188, -26.4287109375, -120.10018920898438, -31.290679931640625, -118.4869384765625, 15.436283111572266, 70.47749328613281, 139.81503295898438, 149.14181518554688, 189.5643310546875, 47.450042724609375, -134.9395294189453, 62.35748291015625, 9.85546875, 156.81703186035156, 159.1796417236328, 29.924545288085938, 30.45191192626953, -17.812896728515625, 159.40133666992188, -114.33917999267578, -0.179534912109375, -12.122802734375, 39.932647705078125, 119.469482421875, 0.0, -13.317985534667969, 103.890869140625, 36.639381408691406, -10.15240478515625, 149.58065795898438, -23.291610717773438, -101.20985412597656], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000450.npy"} {"epoch": 0.9424083769633508, "step": 451, "batch_size": 128, "mean": 54.643165588378906, "std": 86.88887786865234, "min": -166.9696807861328, "p10": -34.853845214843744, "median": 38.02293395996094, "p90": 164.09425659179686, "max": 231.3707275390625, "pos_frac": 0.75, "sample": [60.52978515625, 37.96238708496094, -21.929412841796875, 7.678447723388672, 192.59373474121094, 150.81802368164062, 42.36309814453125, 2.4685134887695312, -161.6470947265625, -26.49017333984375, 127.88728332519531, 7.442056655883789, 182.44784545898438, 21.10693359375, 35.542724609375, 118.6505126953125, 106.21340942382812, 11.289093017578125, 176.92257690429688, -53.56550598144531, 152.93121337890625, -2.6658935546875, 11.272293090820312, 48.75407409667969, 199.81405639648438, -1.789398193359375, 228.11630249023438, 9.6365966796875, -12.60028076171875, 33.6876220703125, -71.08181762695312, 11.854743957519531, -166.9696807861328, 170.5084228515625, 73.64410400390625, 49.747825622558594, 21.425331115722656, 104.90768432617188, 35.24139404296875, 136.31454467773438, 18.174480438232422, 148.17483520507812, 123.01678466796875, 143.50982666015625, 138.92202758789062, 160.9581298828125, 10.897842407226562, -0.655303955078125, 190.46365356445312, -2.1776123046875, 7.2703399658203125, 188.79052734375, 20.643890380859375, 205.73410034179688, 140.64501953125, 149.13095092773438, -26.986709594726562, 59.2457275390625, -166.18655395507812, 141.35794067382812, 103.0124282836914, 140.62545776367188, 27.377883911132812, 161.5208740234375, 3.9568328857421875, 38.08348083496094, -113.60198211669922, 0.9127960205078125, 68.12933349609375, 99.54644775390625, 10.161239624023438, 9.234039306640625, -30.878021240234375, 150.56536865234375, 8.695098876953125, -39.42059326171875, 54.205535888671875, 9.11834716796875, 133.71221923828125, 85.61959838867188, -2.03521728515625, -3.56524658203125, -23.771484375, 71.39110565185547, 164.0740966796875, -32.89666748046875, 27.146255493164062, 102.08023071289062, 12.876266479492188, 15.825401306152344, 126.13519287109375, 122.3201675415039, -24.790283203125, 160.98629760742188, 22.353866577148438, -145.8121337890625, 41.97930145263672, -57.30950927734375, -31.13031768798828, 164.14129638671875, 45.254669189453125, 156.85470581054688, 62.388092041015625, 169.68853759765625, -12.0037841796875, -95.86625671386719, -106.1197509765625, -64.522705078125, 125.16375732421875, 61.243499755859375, 146.0867919921875, 30.456695556640625, 170.30987548828125, 158.462646484375, -4.219383239746094, 13.057254791259766, 102.04144287109375, 4.37969970703125, 92.55819702148438, 231.3707275390625, 43.99449157714844, -1.9245758056640625, -27.495498657226562, 3.6546554565429688, 133.4159393310547, 137.57940673828125, -51.64373779296875, 137.6210479736328], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000451.npy"} {"epoch": 0.9445026178010472, "step": 452, "batch_size": 128, "mean": 52.687522888183594, "std": 82.84152221679688, "min": -155.15399169921875, "p10": -29.37108764648437, "median": 32.563114166259766, "p90": 171.04586181640624, "max": 224.217529296875, "pos_frac": 0.75, "sample": [-64.80755615234375, 139.53848266601562, 40.08856201171875, 187.72998046875, -109.36227416992188, 8.000389099121094, 11.390357971191406, 179.87948608398438, -13.010391235351562, -24.586807250976562, 172.01602172851562, 0.4180908203125, 6.28692626953125, 8.284111022949219, -1.8934288024902344, 8.64984130859375, 65.98855590820312, -4.6320648193359375, -8.3646240234375, 13.61944580078125, 0.2543220520019531, -87.61865234375, 113.24111938476562, 149.01788330078125, 135.373779296875, 55.207275390625, 54.535736083984375, 54.18653869628906, -0.16618728637695312, -80.92131042480469, -106.92767333984375, 187.33425903320312, -38.914283752441406, 192.89358520507812, 105.9110107421875, -13.507598876953125, 173.33218383789062, 164.51806640625, -19.128082275390625, 136.93807983398438, 91.6673355102539, 13.17047119140625, 6.535526275634766, 21.9619140625, 136.45245361328125, 11.61878776550293, 34.90283966064453, 98.60791015625, 101.13050842285156, 27.872413635253906, 170.8876953125, 39.7371826171875, 123.1507568359375, -9.31927490234375, 11.050739288330078, -50.09332275390625, 13.133132934570312, 17.92431640625, 17.136444091796875, 22.003082275390625, 11.063468933105469, 43.1318359375, -10.973358154296875, 120.20626831054688, -73.25201416015625, 145.86390686035156, 146.37808227539062, 158.85650634765625, 21.016082763671875, 13.81683349609375, 10.32879638671875, 172.64938354492188, 29.31012725830078, 9.41693115234375, 154.50164794921875, 224.217529296875, 149.3468780517578, -18.549285888671875, 35.995452880859375, 119.14436340332031, -141.81646728515625, 103.09344482421875, 178.9054718017578, -3.0196762084960938, 172.89956665039062, 149.03057861328125, -5.4078216552734375, 144.69151306152344, 163.98556518554688, 30.223388671875, 64.67192077636719, 46.8511962890625, 9.169593811035156, 0.10254669189453125, 37.02085876464844, -62.006866455078125, 147.42059326171875, 53.61174011230469, 58.529510498046875, -16.18157958984375, 207.62741088867188, 53.792877197265625, 24.125030517578125, 142.91818237304688, 69.1890869140625, 1.9809417724609375, -27.61248779296875, 21.189987182617188, 168.44635009765625, 164.67103576660156, 210.5169677734375, 171.4149169921875, -17.022293090820312, 36.86419677734375, 158.35491943359375, 0.0, 55.13676452636719, -1.9128036499023438, -120.19256591796875, -33.4744873046875, 47.350128173828125, 57.115936279296875, -155.15399169921875, 46.52825927734375, 3.548126220703125, 166.68589782714844, 8.19305419921875, -0.8473320007324219], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000452.npy"} {"epoch": 0.9465968586387434, "step": 453, "batch_size": 128, "mean": 60.004150390625, "std": 82.0124282836914, "min": -163.31988525390625, "p10": -27.939923095703122, "median": 49.244651794433594, "p90": 166.43877868652342, "max": 234.78411865234375, "pos_frac": 0.78125, "sample": [27.330116271972656, 140.68629455566406, 3.418426513671875, 114.06063842773438, 167.615478515625, 6.4298858642578125, 20.289649963378906, -2.2776107788085938, 147.35073852539062, 99.47959899902344, 57.94666290283203, 50.98994445800781, 21.903472900390625, 57.35040283203125, -22.83758544921875, -49.028778076171875, 39.14788818359375, -163.31988525390625, 64.05075073242188, 171.69589233398438, -24.043609619140625, 162.68948364257812, 157.240478515625, 137.16847229003906, 49.048583984375, 21.4444580078125, -55.001922607421875, 158.33514404296875, 50.79072570800781, 159.21754455566406, 21.925537109375, 135.81199645996094, 40.63287353515625, 125.40202331542969, 25.244476318359375, 28.30535888671875, 39.96844482421875, -4.6128692626953125, 174.57826232910156, 178.1717529296875, 6.8842010498046875, 169.09405517578125, 31.89727783203125, 70.75387573242188, 140.55987548828125, -124.21231079101562, 14.92578125, 17.732025146484375, 8.05810546875, -42.79302978515625, 6.674407958984375, 144.47396850585938, 3.83978271484375, 19.483062744140625, -153.4056396484375, 50.413177490234375, -4.383087158203125, -45.329376220703125, 49.38810729980469, 8.7506103515625, 123.93310546875, -7.915241241455078, 0.0, 125.67323303222656, 79.06120300292969, 39.05168151855469, 116.91300201416016, 11.264007568359375, 154.44570922851562, 26.346466064453125, 86.00643920898438, 61.33203125, 53.019927978515625, -146.71548461914062, 20.656707763671875, 165.27908325195312, 188.48431396484375, 161.06381225585938, -2.887063980102539, 145.9135284423828, -13.622360229492188, 185.07052612304688, 122.89413452148438, -9.90069580078125, 7.649322509765625, 20.48931884765625, 181.9879150390625, -26.97314453125, -30.19573974609375, 181.9539794921875, 47.144775390625, 136.99627685546875, -49.65576171875, -35.365020751953125, 161.55059814453125, 21.92718505859375, -142.63345336914062, 155.05259704589844, -8.90777587890625, 234.78411865234375, 79.52945709228516, 49.849578857421875, 82.11573028564453, 106.55287170410156, 31.896759033203125, 60.962158203125, -0.3310546875, 4.293098449707031, 165.93447875976562, 78.14176940917969, 157.03253173828125, 128.92352294921875, 25.8453369140625, 131.10067749023438, 49.1011962890625, 109.14909362792969, 176.25482177734375, 177.1431884765625, -9.990341186523438, 12.120208740234375, 130.100341796875, 8.051933288574219, 125.79837036132812, -20.22650146484375, 148.364013671875, -49.50103759765625, 67.90701293945312, 205.83453369140625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000453.npy"} {"epoch": 0.9486910994764398, "step": 454, "batch_size": 128, "mean": 54.647377014160156, "std": 84.60499572753906, "min": -222.65643310546875, "p10": -24.767924499511718, "median": 29.037490844726562, "p90": 173.89588623046873, "max": 225.276611328125, "pos_frac": 0.7890625, "sample": [44.9013671875, 162.90872192382812, -142.63247680664062, -2.34918212890625, -5.7743988037109375, 9.4420166015625, 174.83087158203125, 144.72821044921875, 10.0904541015625, 35.22038269042969, 24.596290588378906, 41.079689025878906, -11.80662727355957, 126.10721588134766, 34.18831253051758, 110.91716766357422, -130.11776733398438, 7.050626754760742, 55.641448974609375, -2.7058753967285156, 113.70193481445312, 52.064002990722656, 21.431182861328125, 3.885589599609375, 209.77886962890625, 225.276611328125, -6.96844482421875, 182.18118286132812, -222.65643310546875, 2.679595947265625, 19.105491638183594, 146.5417022705078, 74.67017364501953, 127.17813873291016, 163.0408477783203, 24.098670959472656, 181.5516357421875, -28.323196411132812, 17.91058349609375, 17.560928344726562, 157.38946533203125, 17.89764404296875, 179.04913330078125, 140.5152587890625, 38.234107971191406, 145.732666015625, 117.05780029296875, -2.5567626953125, 178.41616821289062, -12.09765625, 17.947769165039062, 33.603424072265625, 36.47393798828125, 2.68182373046875, 139.89425659179688, -15.89227294921875, 119.68350219726562, 5.6809844970703125, 161.52581787109375, 177.45587158203125, 118.02435302734375, 146.54248046875, 182.0220947265625, 91.74807739257812, 184.54620361328125, 163.73321533203125, 159.35650634765625, 22.66162109375, 25.574005126953125, 1.077117919921875, 175.36993408203125, 155.79653930664062, 0.263580322265625, -76.09275817871094, 10.591217041015625, 34.2913818359375, -32.236656188964844, 63.3441162109375, 8.348175048828125, -123.77276611328125, -25.519546508789062, 0.36661529541015625, 173.49517822265625, -39.70556640625, -23.48260498046875, 49.4632568359375, 144.9888458251953, 55.49908447265625, -24.44580078125, 156.36883544921875, 33.56208801269531, 14.65887451171875, 18.84637451171875, 177.2707061767578, 203.65054321289062, 2.29638671875, 22.62096405029297, 9.511962890625, 170.1966094970703, 4.792823791503906, -28.845291137695312, 11.562362670898438, 165.9163818359375, 29.45368003845215, 27.39031982421875, 28.86468505859375, 173.4703369140625, 50.535621643066406, -109.43624877929688, 144.49365234375, 2.7390823364257812, 23.247650146484375, 76.49560546875, 2.3073577880859375, 16.195068359375, 134.09237670898438, 74.30755615234375, -2.3155670166015625, -80.23284912109375, 43.157958984375, 17.018600463867188, -9.716690063476562, 29.210296630859375, 166.48703002929688, -13.2938232421875, -1.545623779296875, 16.191177368164062, -78.22872924804688], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000454.npy"} {"epoch": 0.9507853403141361, "step": 455, "batch_size": 128, "mean": 44.61500549316406, "std": 83.44082641601562, "min": -184.99029541015625, "p10": -38.865930175781244, "median": 29.693634033203125, "p90": 156.77572174072264, "max": 224.57998657226562, "pos_frac": 0.6953125, "sample": [141.13726806640625, 95.70399475097656, 23.904678344726562, -22.806655883789062, 128.20999145507812, 71.14773559570312, 119.3296127319336, 163.3507537841797, 21.516860961914062, -27.558727264404297, -1.8857421875, 89.5821533203125, -1.956268310546875, 127.08332061767578, 71.63999938964844, 157.18544006347656, 167.31298828125, 126.12930297851562, 210.23226928710938, -89.03997802734375, 7.692657470703125, 37.2193603515625, -0.10284423828125, 149.13375854492188, 224.57998657226562, 158.39939880371094, 154.07498168945312, 19.955078125, 38.94597244262695, 3.003753662109375, -18.4464111328125, -166.41110229492188, -0.9643211364746094, 22.402236938476562, 24.869949340820312, 9.263702392578125, -12.288749694824219, -25.850418090820312, -71.73848724365234, 19.9866943359375, 34.11767578125, -9.01715087890625, 128.47056579589844, 206.84219360351562, 3.44512939453125, 75.72998046875, 146.27902221679688, -2.907806396484375, 158.91433715820312, 27.52740478515625, 4.04510498046875, 10.4365234375, 39.04035949707031, -7.6392974853515625, -7.0548553466796875, -4.57855224609375, 11.3907470703125, -127.60678100585938, -68.05858612060547, -1.1688766479492188, 19.90179443359375, -36.858245849609375, 5.618442535400391, -35.85493469238281, -30.671112060546875, -51.02709197998047, 115.37503814697266, 56.85984802246094, 142.11822509765625, 110.99269104003906, -154.82421875, 122.50074005126953, 144.7225341796875, 155.75791931152344, 41.54949951171875, -13.6575927734375, -17.183837890625, 161.52276611328125, 156.60012817382812, 72.75328063964844, 42.7877197265625, -2.7432098388671875, 8.965415954589844, -97.49331665039062, 151.40878295898438, -17.495445251464844, -157.90855407714844, 62.04839324951172, -2.988677978515625, 19.958110809326172, -20.373687744140625, 47.643951416015625, 2.0836181640625, 45.17735290527344, 19.130752563476562, 113.62014770507812, 203.14178466796875, 206.3446044921875, 101.41336059570312, -37.80908203125, 2.973358154296875, 6.012489318847656, 58.03692626953125, 45.923248291015625, 6.930694580078125, -2.424083709716797, 14.872657775878906, 89.27743530273438, 27.135009765625, 64.4926986694336, -88.34841918945312, 82.68215942382812, 85.69473266601562, 70.28279113769531, 148.983154296875, -56.074398040771484, -184.99029541015625, 50.06134033203125, -41.3319091796875, 150.53207397460938, 138.3099365234375, 49.902862548828125, 186.11013793945312, 52.645416259765625, 79.58279418945312, 59.78631591796875, 166.5426788330078, 31.85986328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000455.npy"} {"epoch": 0.9528795811518325, "step": 456, "batch_size": 128, "mean": 66.54374694824219, "std": 95.32628631591797, "min": -177.0584716796875, "p10": -39.82637023925779, "median": 50.95353317260742, "p90": 180.89618988037108, "max": 315.53765869140625, "pos_frac": 0.7421875, "sample": [39.3109130859375, -14.040119171142578, -53.0302734375, 140.93760681152344, 54.576759338378906, -14.821189880371094, 151.09498596191406, -89.69210815429688, 178.4522705078125, 115.093994140625, 167.84695434570312, -177.0584716796875, 20.754196166992188, 2.61248779296875, -34.32794189453125, 189.92474365234375, 127.55787658691406, -54.383880615234375, 5.7641754150390625, 14.33319091796875, 171.947998046875, 104.52371215820312, -24.63458251953125, 161.0470428466797, 1.375382423400879, -17.924884796142578, 101.23681640625, 53.17515563964844, 153.01046752929688, -85.66494750976562, 141.79721069335938, -127.54440307617188, 139.18478393554688, -31.938766479492188, 124.61318969726562, 174.70416259765625, 90.45530700683594, 143.67758178710938, 148.96075439453125, 125.58984375, 210.607421875, -18.996978759765625, 177.1077880859375, 23.9378662109375, 181.148681640625, 144.14849853515625, 191.85589599609375, -4.781402587890625, 9.00634765625, 153.4681396484375, 39.9884033203125, 108.84379577636719, -8.175018310546875, 41.956295013427734, -52.656036376953125, 18.974273681640625, 6.421356201171875, 22.28316879272461, 20.3214111328125, 200.6693115234375, 16.80047607421875, -6.466560363769531, 16.812843322753906, -114.30323028564453, -18.249786376953125, 171.98721313476562, 144.48910522460938, -139.0635223388672, 28.076019287109375, 20.34686279296875, 13.547966003417969, 116.20718383789062, -4.218486785888672, 130.62994384765625, 50.939292907714844, 132.19253540039062, 204.03448486328125, 139.5948486328125, -62.041717529296875, 64.88800048828125, 143.30796813964844, 165.0277099609375, -130.58364868164062, 178.32705688476562, 8.8572998046875, -9.49981689453125, 20.348251342773438, 185.48297119140625, 172.34927368164062, 185.40745544433594, 128.80245971679688, 194.861572265625, -3.8096389770507812, 173.750244140625, 0.0, 3.5912227630615234, 119.10108184814453, 169.8765106201172, 130.58877563476562, 122.05035400390625, 14.137054443359375, 9.39990234375, 50.9677734375, -1.70025634765625, 83.62642669677734, -68.57861328125, 139.3609619140625, 4.1681671142578125, 11.786865234375, -18.4859619140625, 1.004180908203125, 112.70103454589844, 52.7200927734375, 123.46832275390625, 180.78797912597656, 0.5220890045166016, 262.66455078125, 4.684967041015625, 315.53765869140625, 212.21676635742188, 166.38189697265625, 31.8480224609375, -16.39141845703125, -6.6251220703125, 163.45167541503906, -28.151199340820312, -120.80377197265625, 188.23260498046875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000456.npy"} {"epoch": 0.9549738219895288, "step": 457, "batch_size": 128, "mean": 59.201019287109375, "std": 89.31880950927734, "min": -189.03939819335938, "p10": -45.302172851562496, "median": 40.05075454711914, "p90": 173.74469909667968, "max": 209.25172424316406, "pos_frac": 0.7109375, "sample": [178.84010314941406, 12.123443603515625, -15.426048278808594, -5.1641998291015625, 119.10260009765625, 34.50665283203125, 157.17173767089844, 168.37411499023438, 167.96417236328125, 21.07354736328125, 158.9659423828125, -0.8537693023681641, 190.28677368164062, 118.80609130859375, 5.275634765625, 195.99124145507812, -4.78369140625, 31.692047119140625, -116.81117248535156, 173.59738159179688, 132.04196166992188, 149.8732452392578, 145.81715393066406, 206.73423767089844, 141.43170166015625, 51.62420654296875, 166.87612915039062, 9.321304321289062, 169.88980102539062, 18.263275146484375, 65.00872802734375, -16.969940185546875, -2.500152587890625, -33.367645263671875, 142.39846801757812, 166.13323974609375, -26.745933532714844, 40.94075012207031, -89.87130737304688, 117.84814453125, -18.391693115234375, 75.92781829833984, -30.062728881835938, -1.35369873046875, 6.000030517578125, -189.03939819335938, 183.30319213867188, 25.231964111328125, 180.04617309570312, -9.6170654296875, -44.90435791015625, 70.18450927734375, 117.9957046508789, -58.6904296875, 143.55859375, 19.970808029174805, -139.270751953125, 119.32476806640625, 57.902618408203125, 26.38958740234375, -23.0908203125, -20.29051971435547, 26.311187744140625, -28.70391082763672, 29.089553833007812, 72.42657470703125, 131.87451171875, 21.254257202148438, 209.25172424316406, -11.417694091796875, -12.445846557617188, 138.46539306640625, 140.31781005859375, 38.65057373046875, 25.462158203125, -145.81101989746094, -47.615997314453125, 34.7164306640625, 39.16075897216797, 89.89463806152344, 22.86669921875, 164.910888671875, 111.83326721191406, 98.66397094726562, 77.80502319335938, -22.026336669921875, 113.55926513671875, -1.960479736328125, 154.03964233398438, 197.0858154296875, 6.27067756652832, -78.84384155273438, 162.828369140625, -102.434814453125, -46.6036376953125, 163.468994140625, -46.23040771484375, 119.31407165527344, -18.44439697265625, 174.08843994140625, 140.91217041015625, 85.9569091796875, 10.499313354492188, 46.472564697265625, -48.53802490234375, 187.44509887695312, 31.0322265625, 125.52196502685547, 12.098983764648438, 113.71455383300781, 179.12289428710938, 94.02385711669922, 6.43475341796875, 173.0527801513672, 162.04336547851562, 136.48260498046875, 4.984039306640625, 3.4271411895751953, -138.99688720703125, 99.65391540527344, 182.5804443359375, 0.8882026672363281, 0.0, 174.13784790039062, -3.6724319458007812, 143.9619598388672, 52.920440673828125, -36.1068115234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000457.npy"} {"epoch": 0.9570680628272251, "step": 458, "batch_size": 128, "mean": 52.134979248046875, "std": 86.82891845703125, "min": -173.61843872070312, "p10": -34.929833984375, "median": 23.80419921875, "p90": 178.00003662109376, "max": 228.2952423095703, "pos_frac": 0.7578125, "sample": [4.9327239990234375, -47.68511962890625, 11.648727416992188, -154.32821655273438, 193.0234832763672, 16.58930206298828, -48.52142333984375, 172.5919189453125, -9.843902587890625, -35.3428955078125, 43.13916015625, -69.84002685546875, -53.474853515625, 0.9538211822509766, 2.1515750885009766, 95.56907653808594, 32.61492919921875, 33.351959228515625, -21.8978271484375, 58.488861083984375, 178.40579223632812, -1.5830879211425781, 119.76766967773438, -0.42584228515625, 25.363571166992188, 3.822805404663086, 156.55844116210938, 180.4298095703125, -5.8209228515625, 141.02749633789062, 22.57708740234375, -18.9984130859375, 135.4125518798828, 208.83200073242188, 39.683990478515625, 17.737884521484375, 20.677947998046875, 80.62466430664062, 167.94296264648438, -28.505844116210938, 177.82614135742188, -3.782804489135742, -82.12255859375, 25.03131103515625, 6.513397216796875, 4.8790740966796875, 8.0015869140625, 12.844924926757812, 7.5788421630859375, 32.1708984375, 4.781803131103516, -50.04127502441406, -18.248382568359375, 161.42596435546875, 142.44107055664062, -66.919189453125, 2.035400390625, -1.0808944702148438, 129.2818603515625, 196.98724365234375, 37.4888916015625, 2.379744529724121, 218.99951171875, 60.0908203125, 122.53184509277344, 5.55859375, 9.680145263671875, -112.89788055419922, -34.7528076171875, 192.81533813476562, 160.35287475585938, 50.474884033203125, -15.520553588867188, 29.0283203125, -72.47943115234375, 150.12046813964844, 98.67108154296875, -15.9571533203125, 4.927947998046875, 131.84950256347656, 155.56060791015625, -26.71856689453125, -9.9569091796875, 13.2637939453125, 193.10586547851562, 22.168777465820312, 228.2952423095703, 19.67523193359375, 57.0223388671875, 152.5474853515625, 30.200286865234375, 191.74197387695312, 145.74147033691406, 133.19764709472656, 205.01173400878906, -4.282798767089844, 137.52496337890625, -6.5814208984375, 6.839103698730469, 127.6756591796875, 41.91473388671875, 11.10357666015625, 6.978363037109375, 160.90548706054688, 106.92434692382812, 29.171546936035156, 8.067527770996094, 9.168258666992188, 156.3332061767578, 172.07908630371094, 33.40850830078125, 192.52528381347656, 152.20184326171875, 41.19989013671875, 141.41946411132812, 18.1270751953125, 103.408447265625, 0.7657012939453125, -169.24029541015625, -173.61843872070312, 134.93719482421875, 178.84317016601562, 16.415103912353516, 4.38165283203125, 131.59556579589844, 34.186309814453125, 6.00799560546875, -28.584495544433594], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000458.npy"} {"epoch": 0.9591623036649215, "step": 459, "batch_size": 128, "mean": 40.66044998168945, "std": 91.09747314453125, "min": -180.21157836914062, "p10": -61.89232635498046, "median": 18.77790069580078, "p90": 167.38443756103516, "max": 224.627685546875, "pos_frac": 0.6640625, "sample": [9.589157104492188, -60.78144836425781, 5.061185836791992, 148.86212158203125, 2.1241455078125, 107.31033325195312, -140.1566925048828, 44.40777587890625, 49.32389831542969, 29.83502960205078, 96.42120361328125, 116.55546569824219, 41.12251663208008, -16.58856201171875, 13.981025695800781, 6.164531707763672, 175.134521484375, 111.65924072265625, -17.32386589050293, 22.356414794921875, 24.582763671875, 18.32666015625, -1.6859130859375, 30.71124267578125, 17.871795654296875, -13.21722412109375, 145.2866668701172, -20.138893127441406, 150.235595703125, -28.715538024902344, -15.880319595336914, 0.0, 5.8425140380859375, -53.02117919921875, 154.58453369140625, -119.92413330078125, 60.148170471191406, -68.32002258300781, -13.793548583984375, 58.07696533203125, 153.6466064453125, -21.14544677734375, 19.4561767578125, 64.32449340820312, -148.6638946533203, 63.521812438964844, -32.40155029296875, 23.121124267578125, -15.897144317626953, 180.28509521484375, 5.2220001220703125, 9.004705429077148, -3.1998138427734375, 212.14840698242188, 61.65110778808594, 15.330101013183594, 97.79556274414062, -143.73301696777344, 166.0234832763672, 224.627685546875, 125.47122192382812, -33.45188903808594, 170.55999755859375, 131.6490936279297, 150.69912719726562, 192.84954833984375, 48.699798583984375, 137.22523498535156, -131.50149536132812, -105.62240600585938, 12.8443603515625, 41.9794921875, 154.17063903808594, 153.5960693359375, -14.151809692382812, -173.59686279296875, 189.10081481933594, 93.77745056152344, 10.0946044921875, -64.484375, -180.21157836914062, -6.9351806640625, 5.794921875, -149.80709838867188, 170.64208984375, 86.097412109375, -17.433837890625, 144.93719482421875, 0.0, 6.702476501464844, 143.7959747314453, -14.4544677734375, 23.466766357421875, 206.81207275390625, -106.10888671875, 194.7608642578125, -5.69140625, 137.32913208007812, 18.313308715820312, 34.188682556152344, -78.81686401367188, 132.19161987304688, 17.41033935546875, -14.110397338867188, -15.271087646484375, 97.56339263916016, 159.8946533203125, 12.35791015625, 4.2509765625, -11.753997802734375, -43.723846435546875, -12.01220703125, -1.81903076171875, -40.010284423828125, 157.5649871826172, 185.9554443359375, 1.924713134765625, 30.15656280517578, 165.29400634765625, 30.0887451171875, -4.463375091552734, 182.52243041992188, 196.78240966796875, 19.229141235351562, 38.3548583984375, 6.65032958984375, 70.35443115234375, 128.71908569335938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000459.npy"} {"epoch": 0.9612565445026178, "step": 460, "batch_size": 128, "mean": 40.687522888183594, "std": 94.77727508544922, "min": -187.42535400390625, "p10": -91.27295379638672, "median": 26.77074432373047, "p90": 159.7171401977539, "max": 224.9964599609375, "pos_frac": 0.7265625, "sample": [13.4586181640625, 57.751800537109375, -122.5260009765625, 152.57748413085938, 60.21942138671875, 159.52792358398438, -151.47512817382812, 133.5254669189453, -44.588623046875, 180.6707763671875, -131.04873657226562, 17.003448486328125, 17.964263916015625, 10.616912841796875, 46.76228713989258, 7.49383544921875, -55.399200439453125, 15.71295166015625, 160.1586456298828, 179.45846557617188, -26.515411376953125, 26.7996826171875, 41.72491455078125, 82.87991333007812, 76.36079406738281, 15.812431335449219, 148.466064453125, -7.0287628173828125, 193.07110595703125, -40.999237060546875, -13.386322021484375, 126.63780212402344, 34.16680908203125, 224.9964599609375, 84.39785766601562, -56.26774597167969, 127.3374252319336, -132.29339599609375, 218.31787109375, -20.20166015625, -171.14651489257812, 176.20025634765625, -103.99240112304688, 8.508602142333984, 90.04330444335938, 150.39862060546875, -48.9478759765625, 134.0323028564453, 8.709716796875, -89.7386474609375, 84.05303955078125, 1.4264945983886719, 1.4483261108398438, 0.9906997680664062, -51.74304962158203, 19.696762084960938, -1.3394775390625, -2.87237548828125, -5.064857482910156, 17.302291870117188, 157.91461181640625, 54.9820556640625, 145.16275024414062, 12.157047271728516, 30.480987548828125, -111.30938720703125, 0.2063751220703125, 50.018157958984375, 155.43258666992188, -0.6165809631347656, 17.81231689453125, 77.3890380859375, 22.0594482421875, 19.9464111328125, 145.20013427734375, 122.06124877929688, 153.29302978515625, 147.94610595703125, 30.84185791015625, 30.344905853271484, 182.27719116210938, 26.741806030273438, -82.7364501953125, 219.5731201171875, 49.4744873046875, 219.06143188476562, 171.56100463867188, 158.53729248046875, -119.95425415039062, 150.22557067871094, 95.55685424804688, 83.49835205078125, -107.73867797851562, 7.0079193115234375, -45.75787353515625, 11.57183837890625, 153.1826629638672, -187.42535400390625, 123.47598266601562, -94.0723876953125, 67.03401184082031, -135.87020874023438, 145.96771240234375, 48.472930908203125, 16.63330841064453, -90.07319641113281, -8.119853973388672, -62.83612060546875, 158.5421600341797, 38.225341796875, 28.849761962890625, 101.09600830078125, 34.05950927734375, 156.30722045898438, -115.12554931640625, 3.406585693359375, 5.695625305175781, 139.44467163085938, 41.38763427734375, 195.00546264648438, 161.27122497558594, 12.028327941894531, 40.636566162109375, -53.94546127319336, -77.91529846191406, 4.078557968139648, 24.887786865234375, 23.367904663085938], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000460.npy"} {"epoch": 0.9633507853403142, "step": 461, "batch_size": 128, "mean": 52.58932113647461, "std": 93.28910827636719, "min": -178.125244140625, "p10": -51.485766983032214, "median": 31.426589965820312, "p90": 178.0827178955078, "max": 217.09295654296875, "pos_frac": 0.703125, "sample": [18.502838134765625, -94.3897705078125, 178.84384155273438, 23.880962371826172, 65.69606018066406, 132.10366821289062, 151.84951782226562, 28.1175537109375, -47.466243743896484, 180.80160522460938, 147.39990234375, 0.0, 14.457244873046875, 217.09295654296875, 123.15003967285156, 137.6563720703125, 137.0625, 71.778564453125, -4.965728759765625, 10.415695190429688, 52.50701141357422, 177.91531372070312, 102.6358413696289, -86.70375061035156, 102.46578979492188, -2.690673828125, 12.41650390625, 188.8612060546875, -38.13899230957031, 149.55807495117188, -0.6535072326660156, 193.70968627929688, 57.99847412109375, 146.3147735595703, -34.63958740234375, 151.6069793701172, 189.98553466796875, 17.83489990234375, -158.181396484375, -74.79095458984375, 178.47332763671875, -128.72354125976562, 1.97412109375, 15.376251220703125, 123.15863037109375, 3.400798797607422, 121.85101318359375, 134.47714233398438, 179.2477264404297, -3.2586669921875, 137.54510498046875, -113.37451171875, 22.4000244140625, -2.0655364990234375, 13.02286148071289, -13.979782104492188, 143.83743286132812, -177.090576171875, -34.746124267578125, 70.10382080078125, 127.51765441894531, 2.0480384826660156, -12.422134399414062, 129.07980346679688, 183.49954223632812, 22.3626708984375, -178.125244140625, 12.179718017578125, 170.3137969970703, 8.03277587890625, 5.934474945068359, 158.2750244140625, -7.059196472167969, 165.12091064453125, -14.033355712890625, 188.92608642578125, 86.51947021484375, 142.11981201171875, -60.864654541015625, -41.67987060546875, 150.06204223632812, 5.746673583984375, 131.5054168701172, 88.23605346679688, 153.42953491210938, 60.404029846191406, -21.97271728515625, -111.23651123046875, -7.676935195922852, 67.84210205078125, 85.60592651367188, 120.53450012207031, 10.065887451171875, 137.48973083496094, 10.1954345703125, 188.2788848876953, -6.5027923583984375, 34.735626220703125, 153.44873046875, -112.20829010009766, 100.42308807373047, 1.6263885498046875, 182.17884826660156, 20.350372314453125, -40.1221923828125, -18.072891235351562, 20.342742919921875, 149.25274658203125, 153.05885314941406, -11.057136535644531, -47.22493362426758, 110.6583480834961, -0.04347038269042969, 107.73844146728516, 19.678314208984375, 1.9466934204101562, 100.71981811523438, 107.61842346191406, -151.3216552734375, -33.54151153564453, 39.09112548828125, 1.1061019897460938, -139.96722412109375, 183.927490234375, -17.419939041137695, 61.783905029296875, 121.79743957519531, 169.548095703125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000461.npy"} {"epoch": 0.9654450261780104, "step": 462, "batch_size": 128, "mean": 67.33772277832031, "std": 93.59368896484375, "min": -170.72918701171875, "p10": -32.01319198608398, "median": 79.45446014404297, "p90": 178.9821365356445, "max": 231.6051025390625, "pos_frac": 0.7734375, "sample": [139.25436401367188, 65.89158630371094, 4.1084442138671875, 118.83648681640625, 173.43161010742188, 49.916290283203125, 145.6503143310547, 112.50872802734375, 78.69023132324219, 183.97869873046875, -141.8175048828125, 122.50933837890625, 188.27337646484375, 46.3043212890625, 177.56832885742188, 6.5076904296875, 103.29701232910156, 145.23779296875, 80.21868896484375, 4.6109619140625, 32.19439697265625, 139.77859497070312, 135.9634246826172, 46.05224609375, 35.38508987426758, 0.157623291015625, 145.6128692626953, -1.082132339477539, 155.2481689453125, -117.75624084472656, -126.8466796875, 108.40975952148438, -6.019378662109375, 32.68206787109375, -4.15155029296875, 153.94918823242188, 183.97848510742188, -88.287109375, -15.18243408203125, 14.58001708984375, 143.12496948242188, 152.36935424804688, 159.05511474609375, 7.58544921875, 221.5811767578125, 108.7057113647461, 111.58653259277344, 169.64132690429688, 124.309814453125, 1.5668792724609375, 182.28102111816406, 7.185249328613281, 138.10130310058594, -113.61973571777344, 13.31689453125, -14.103805541992188, 175.84115600585938, 35.303802490234375, 13.898872375488281, 16.65478515625, 149.89572143554688, 205.01715087890625, -31.64556884765625, 159.41737365722656, -17.548858642578125, 104.87419891357422, -114.18355560302734, 93.9798583984375, 227.631591796875, -22.484115600585938, 161.23391723632812, 165.51092529296875, 124.35488891601562, 132.62744140625, 101.10092163085938, 60.997406005859375, -22.03509521484375, 25.543731689453125, 8.538309097290039, -33.03338623046875, 15.186393737792969, -27.928863525390625, -1.961517333984375, 7.5374755859375, -154.4783935546875, 80.71430969238281, 139.72052001953125, 219.6829833984375, -99.67488098144531, 120.18724060058594, -106.41451263427734, 194.07833862304688, 87.91224670410156, 33.47754669189453, 103.24726867675781, 231.6051025390625, -30.212379455566406, 108.29278564453125, 134.35275268554688, 159.3831787109375, 115.69153594970703, -20.2056884765625, 226.47476196289062, 19.772216796875, -19.187744140625, 19.405548095703125, 142.48587036132812, 43.22235107421875, 30.084930419921875, 13.5469970703125, 142.6197509765625, 7.4951629638671875, 61.6842041015625, 173.84530639648438, -131.87677001953125, 203.96371459960938, -30.413192749023438, 113.07225799560547, 132.52406311035156, -19.235000610351562, 187.21127319335938, 39.81843566894531, 0.825592041015625, -32.87097930908203, 123.74826049804688, 133.0928955078125, 134.63182067871094, -170.72918701171875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000462.npy"} {"epoch": 0.9675392670157068, "step": 463, "batch_size": 128, "mean": 47.96533966064453, "std": 83.33688354492188, "min": -157.11810302734375, "p10": -34.147885131835935, "median": 29.654674530029297, "p90": 162.62679138183591, "max": 216.05889892578125, "pos_frac": 0.7109375, "sample": [98.04360961914062, 49.91770935058594, 60.07042694091797, 5.161338806152344, 11.76904296875, -2.4326744079589844, 187.857177734375, 4.150634765625, -156.17147827148438, 80.4989013671875, 157.32440185546875, -1.0207366943359375, -141.90940856933594, -39.60625457763672, -2.705902099609375, 179.29806518554688, 98.52703857421875, 89.74819946289062, 17.26263427734375, 5.712564468383789, 18.75971221923828, -127.85552978515625, -20.577117919921875, 5.988334655761719, 72.27728271484375, -2.3288726806640625, -89.2861328125, 32.94866943359375, 167.13699340820312, 172.51226806640625, -152.2150421142578, 107.66425323486328, 15.291900634765625, -30.200347900390625, -35.93084716796875, 186.6871337890625, 173.2203369140625, 153.77491760253906, 216.05889892578125, -60.46189880371094, 107.51797485351562, -38.877349853515625, 113.50833129882812, 199.529541015625, 65.17408752441406, 8.11761474609375, 190.708251953125, -10.307655334472656, 27.1746826171875, 4.719482421875, 17.08167266845703, 29.66228485107422, 169.98065185546875, 55.204193115234375, 8.696952819824219, 20.727935791015625, 172.6920166015625, -23.503570556640625, 159.3466796875, 133.58905029296875, 84.4844970703125, 160.69384765625, 11.030517578125, -22.851577758789062, -27.701919555664062, 19.131271362304688, 147.09906005859375, 1.7077789306640625, 117.62525939941406, 52.46160888671875, 157.12696838378906, -33.383758544921875, -7.1231689453125, 64.10882568359375, 80.0142822265625, 6.7613525390625, 122.308837890625, -26.408309936523438, -21.297348022460938, 29.647064208984375, 148.86373901367188, 13.862152099609375, 84.03298950195312, 116.84140014648438, 83.813232421875, 6.741216659545898, 25.145004272460938, 141.770263671875, 50.65132141113281, -20.707632064819336, -15.932037353515625, 114.42105102539062, -28.037368774414062, 0.0, -7.55975341796875, 5.091087341308594, 7.681346893310547, 184.27597045898438, 43.3614501953125, -13.142814636230469, 89.4373779296875, 130.6936492919922, -17.107101440429688, 3.056354522705078, 48.31426239013672, -106.01339721679688, 30.28790283203125, 23.646976470947266, 118.85983276367188, -0.1465282440185547, 6.679412841796875, -95.77928161621094, -35.99835205078125, 96.09195709228516, -2.114471435546875, 151.21112060546875, 158.86903381347656, 41.96920394897461, 157.76229858398438, 91.993408203125, 198.55374145507812, 38.5423583984375, 44.497467041015625, 82.0953369140625, 131.10643005371094, 158.109130859375, -157.11810302734375, -20.247161865234375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000463.npy"} {"epoch": 0.9696335078534032, "step": 464, "batch_size": 128, "mean": 50.38481140136719, "std": 86.09242248535156, "min": -197.74285888671875, "p10": -39.656491088867185, "median": 30.947059631347656, "p90": 161.62281494140623, "max": 199.8736572265625, "pos_frac": 0.75, "sample": [0.0, 100.14866638183594, 34.24736785888672, 136.69549560546875, -8.773239135742188, 134.71746826171875, 151.62322998046875, -30.891815185546875, 152.73831176757812, -0.37148475646972656, 65.04948425292969, 64.60649108886719, 141.3513946533203, 45.104644775390625, 136.7880401611328, 25.11339569091797, 157.63238525390625, 1.58538818359375, 149.50384521484375, 106.17192077636719, 48.5386962890625, 153.67054748535156, 9.107690811157227, 10.46185302734375, 51.96929931640625, 107.22317504882812, 18.710243225097656, -151.96157836914062, 199.8736572265625, 17.833084106445312, -0.85382080078125, -14.425140380859375, 21.526016235351562, 185.70962524414062, 173.09014892578125, 167.0718231201172, 93.65704345703125, 6.748908996582031, -11.46142578125, 116.48809814453125, -3.722503662109375, 24.78729248046875, 165.0216064453125, 82.47734069824219, 115.78369140625, 3.0668182373046875, -13.804229736328125, 17.053863525390625, 1.336822509765625, 92.14665222167969, 18.520828247070312, 23.589080810546875, 11.826438903808594, 136.31808471679688, 17.19344711303711, 96.75082397460938, 145.13531494140625, 120.59085083007812, -145.2705078125, 114.23480224609375, -1.243072509765625, 145.987548828125, -7.5966033935546875, 151.36184692382812, 179.77279663085938, -10.6927490234375, 94.55084991455078, 26.770278930664062, 31.3760986328125, 127.8560791015625, -110.85736083984375, -197.74285888671875, -137.17330932617188, 2.6165618896484375, 5.39202880859375, 30.518020629882812, -15.019256591796875, -58.26458740234375, 123.06866455078125, 12.09014892578125, 161.274658203125, 191.60528564453125, 153.97454833984375, 176.64959716796875, -43.79412841796875, 162.4351806640625, 1.3544273376464844, -39.068511962890625, 11.8427734375, 3.446260452270508, 41.46113586425781, -140.71505737304688, 158.72555541992188, -8.634796142578125, -78.32147216796875, 6.398231506347656, -1.9965057373046875, 138.92733764648438, -0.9722900390625, 10.063859939575195, 123.37145233154297, 172.36253356933594, 36.88201141357422, 166.71859741210938, 170.3514404296875, -41.0284423828125, 87.340576171875, 3.9373779296875, 139.3229522705078, -115.1492919921875, 13.515289306640625, 8.60440444946289, -13.145954132080078, 70.44223022460938, 169.271728515625, 45.8454475402832, 126.07731628417969, 146.2444610595703, 137.7459259033203, -178.262939453125, 18.019927978515625, 30.31829833984375, -67.373779296875, 41.52970886230469, 19.748233795166016, -24.148635864257812, 41.26708984375, 32.96527099609375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000464.npy"} {"epoch": 0.9717277486910995, "step": 465, "batch_size": 128, "mean": 65.58464813232422, "std": 90.07942199707031, "min": -177.57546997070312, "p10": -31.19387512207031, "median": 42.3311767578125, "p90": 180.4577621459961, "max": 214.64822387695312, "pos_frac": 0.75, "sample": [-2.52581787109375, -7.289459228515625, 12.435150146484375, 32.63140869140625, 165.8131103515625, -24.206697463989258, -31.966400146484375, 183.91445922851562, 9.50213623046875, 23.040145874023438, 2.0221405029296875, 14.67974853515625, -5.036224365234375, 11.7738037109375, -123.71096801757812, -34.92242431640625, 182.22796630859375, 158.74032592773438, 195.21267700195312, 172.66455078125, 156.40249633789062, -47.68145751953125, 46.64752197265625, 161.60995483398438, 35.797821044921875, -0.85833740234375, 2.167755126953125, 165.15887451171875, 180.13479614257812, 137.040283203125, 0.9510345458984375, 141.8291473388672, 43.0736083984375, 130.0440673828125, 2.2576446533203125, -10.0299072265625, -35.22265625, 179.12075805664062, 140.38356018066406, 212.22576904296875, 13.090682983398438, 96.21356201171875, -1.7037353515625, -30.86279296875, -33.232688903808594, 41.5887451171875, -4.365730285644531, 22.980451583862305, -23.05584716796875, 141.168701171875, 132.91139221191406, 123.15213012695312, 119.00181579589844, 135.92420959472656, 46.14207458496094, -91.3993148803711, 201.1162109375, 40.40202331542969, -9.13385009765625, 22.84503173828125, 85.16448974609375, 31.037567138671875, 46.077667236328125, 85.55059814453125, 170.51602172851562, 21.529312133789062, 183.96762084960938, 207.52691650390625, -14.09832763671875, 15.05294418334961, 94.71297454833984, 156.96630859375, 149.94566345214844, 130.14036560058594, 107.20077514648438, -86.21414184570312, -4.085205078125, 45.786346435546875, 27.601272583007812, 102.74821472167969, 41.397552490234375, 10.421417236328125, 5.03253173828125, 11.195159912109375, 7.0140380859375, 136.96023559570312, 185.00009155273438, -88.18511962890625, 63.734458923339844, -14.452728271484375, 177.22113037109375, 200.0130615234375, 214.64822387695312, 53.81793975830078, 134.19625854492188, 14.833038330078125, 190.36151123046875, 170.0621337890625, 176.142333984375, 4.234458923339844, 167.0921630859375, 166.81344604492188, 189.45623779296875, 167.19862365722656, 9.8380126953125, 145.42782592773438, -18.414306640625, 144.73733520507812, -141.69329833984375, 35.63018798828125, -88.63650512695312, 0.216644287109375, 170.78884887695312, 163.98583984375, 156.55238342285156, 181.2113494873047, -132.75799560546875, 150.65899658203125, 41.534088134765625, 10.030778884887695, -25.19482421875, 112.66143798828125, -18.9224853515625, -16.803054809570312, 159.56494140625, -177.57546997070312, -9.809814453125, 81.63276672363281], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000465.npy"} {"epoch": 0.9738219895287958, "step": 466, "batch_size": 128, "mean": 71.20044708251953, "std": 81.57740783691406, "min": -156.23385620117188, "p10": -14.603228759765624, "median": 72.98837280273438, "p90": 166.9453659057617, "max": 218.65853881835938, "pos_frac": 0.8125, "sample": [6.344047546386719, 121.47979736328125, 138.53863525390625, 160.85919189453125, 123.83114624023438, 171.66323852539062, 35.539154052734375, 0.0, 89.9697265625, 107.02143859863281, 156.3321533203125, -25.127613067626953, 127.8333740234375, 113.36529541015625, 107.5482177734375, 209.05322265625, 102.60991668701172, 1.9045467376708984, 141.72225952148438, 83.1588134765625, 127.98886108398438, 27.307479858398438, 57.99101257324219, -95.47018432617188, 15.47052001953125, 48.2064208984375, 129.1691131591797, 25.96380615234375, 135.11956787109375, 113.21376037597656, -3.687274932861328, 155.50802612304688, 5.002449035644531, 129.46441650390625, 138.90121459960938, 0.4431419372558594, 148.4429931640625, -7.9201507568359375, 155.76780700683594, 88.81454467773438, 217.3916778564453, 9.760147094726562, 207.77810668945312, 53.3525390625, 110.48638916015625, 179.40850830078125, 30.294082641601562, -38.087799072265625, -124.23417663574219, 2.444091796875, 9.05596923828125, 143.91220092773438, 7.141387939453125, 147.6988525390625, 129.48892211914062, 161.2484130859375, -15.196502685546875, -137.7646484375, 23.6488037109375, 160.141357421875, 145.99363708496094, 146.26425170898438, 121.25086975097656, 87.9263916015625, 17.686553955078125, 135.8477783203125, 165.20501708984375, 131.54623413085938, 190.17669677734375, 143.68789672851562, 0.0, 150.08912658691406, 206.41015625, -30.655914306640625, -14.348968505859375, 30.92778778076172, 126.21781921386719, -27.454269409179688, 68.64712524414062, 218.65853881835938, 6.648109436035156, 9.5985107421875, 205.0096435546875, 77.32962036132812, 30.163055419921875, -156.23385620117188, 23.15325927734375, 57.75004577636719, 27.90380859375, 113.5952377319336, 149.84609985351562, 20.932418823242188, 20.27393341064453, -21.529129028320312, 203.55377197265625, 3.423100471496582, 85.87037658691406, 147.494873046875, -56.94853973388672, -135.83969116210938, 52.64344787597656, 25.0477294921875, -5.1346435546875, 171.0061798095703, 23.56542205810547, 0.0, 164.39474487304688, 20.796844482421875, -0.13614845275878906, 81.06317138671875, 181.09515380859375, 104.91366577148438, -4.567760467529297, 145.6385955810547, 94.97159576416016, -6.436920166015625, 23.58209228515625, 24.494308471679688, 20.172828674316406, 6.5450439453125, 26.15704345703125, 33.39940643310547, 193.04269409179688, 140.26431274414062, -21.236663818359375, -6.739223480224609, 121.21868896484375, 30.511795043945312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000466.npy"} {"epoch": 0.9759162303664921, "step": 467, "batch_size": 128, "mean": 50.40740203857422, "std": 90.76933288574219, "min": -148.60018920898438, "p10": -59.12401390075683, "median": 30.26438617706299, "p90": 170.94210662841795, "max": 210.48764038085938, "pos_frac": 0.640625, "sample": [166.65310668945312, 125.71044921875, 113.37483215332031, 104.07455444335938, 38.30634307861328, -12.508995056152344, 63.77685546875, 130.83135986328125, 115.91607666015625, -1.09295654296875, -136.95806884765625, -148.60018920898438, -10.99029541015625, -12.022476196289062, 20.264129638671875, 140.38198852539062, -109.91619873046875, 190.3355712890625, -0.2138671875, 86.0066146850586, -19.700424194335938, 132.83514404296875, -17.862655639648438, 117.99656677246094, 165.2675323486328, 49.190032958984375, 22.856536865234375, 135.66390991210938, -1.2463722229003906, 13.41094970703125, 22.972381591796875, 153.85568237304688, 76.56011962890625, 158.58612060546875, 178.13336181640625, 84.32780456542969, 25.279281616210938, 181.33401489257812, 86.53199768066406, -79.32049560546875, 181.58282470703125, -0.177398681640625, 0.05542945861816406, 173.68319702148438, 34.61407470703125, 136.02984619140625, -16.447628021240234, -40.937286376953125, 173.2307891845703, -126.8333511352539, 12.312995910644531, -128.301513671875, -57.87001037597656, 155.31631469726562, -49.5595703125, 137.602783203125, -31.692729949951172, -43.818328857421875, 153.84149169921875, 190.13690185546875, 17.2449951171875, 25.914697647094727, -0.7336883544921875, 7.65325927734375, 137.42776489257812, 16.8251953125, -0.7672958374023438, 74.656005859375, -9.956634521484375, -62.05002212524414, 6.752471923828125, -140.4551239013672, 146.87222290039062, 0.0, -0.1149749755859375, -7.7520904541015625, 83.56658935546875, 186.3032989501953, 4.475494384765625, -1.8669395446777344, 0.0, 57.39271545410156, 153.9466552734375, -9.094085693359375, 121.310302734375, 99.94497680664062, 139.1951904296875, -35.92979431152344, 203.3177490234375, -113.33377075195312, -40.04695129394531, 188.19366455078125, 107.93878173828125, 130.9860382080078, 169.96124267578125, -0.46010398864746094, -18.239059448242188, 23.095428466796875, -8.079010009765625, 65.12803649902344, -36.572235107421875, 161.5052032470703, 155.89553833007812, 40.305938720703125, -33.54151916503906, 19.64026641845703, -105.06959533691406, -75.4451904296875, 13.44840145111084, 95.98077392578125, 136.7376708984375, 132.2631378173828, -117.65199279785156, 123.70780944824219, 8.601642608642578, 2.8330078125, -8.638744354248047, 47.553253173828125, 210.48764038085938, 58.7275505065918, 161.97088623046875, 92.59019470214844, 35.55342102050781, 165.08790588378906, 181.2176513671875, 186.96435546875, -101.7813720703125, -22.21685791015625], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000467.npy"} {"epoch": 0.9780104712041885, "step": 468, "batch_size": 128, "mean": 45.302711486816406, "std": 82.92472076416016, "min": -140.30572509765625, "p10": -50.164782714843746, "median": 22.977319717407227, "p90": 159.65784912109373, "max": 239.3275146484375, "pos_frac": 0.6953125, "sample": [37.28363037109375, 187.63299560546875, 140.9084930419922, -51.21478271484375, 59.51148986816406, 20.49462890625, -17.12506103515625, 156.93789672851562, 70.42822265625, 32.99238586425781, 32.83929443359375, -40.734588623046875, 22.47802734375, 180.87637329101562, -70.90353393554688, -13.323509216308594, -0.925018310546875, 169.385498046875, -117.05751037597656, -21.126556396484375, 59.07496643066406, 23.476612091064453, -111.43504333496094, 125.6484375, 36.375640869140625, 205.97607421875, 51.62213134765625, 81.34390258789062, 8.27264404296875, -34.859737396240234, 5.68548583984375, 27.4224853515625, 179.74472045898438, -59.34801483154297, 33.57513427734375, -26.217838287353516, -24.993148803710938, 158.23138427734375, -116.94580078125, 21.147369384765625, 5.400230407714844, 25.135162353515625, -35.266204833984375, 144.11123657226562, 10.818893432617188, 8.9488525390625, 69.8408203125, -77.45948791503906, 52.01994323730469, 5.0061492919921875, -6.5057525634765625, 3.5140304565429688, 19.836456298828125, -0.36004638671875, 13.6512451171875, 44.344696044921875, 12.25750732421875, 139.92880249023438, 134.58258056640625, 149.45359802246094, 165.35635375976562, -20.136749267578125, -140.30572509765625, 10.043357849121094, 239.3275146484375, 6.9930572509765625, 12.991935729980469, 176.37088012695312, -4.2244873046875, 96.89492797851562, 50.926849365234375, 110.14340209960938, -26.51861572265625, 72.15740966796875, 146.18218994140625, 153.17205810546875, -49.41081237792969, 16.623443603515625, 7.8833770751953125, 120.21142578125, -54.764705657958984, 192.79254150390625, 30.310771942138672, 168.499267578125, -8.85321044921875, 126.0343246459961, 136.26123046875, -20.294984817504883, 138.56246948242188, 112.21378326416016, -16.42822265625, 6.289745330810547, 3.2059326171875, 190.20806884765625, 1.02099609375, 34.56150817871094, 124.30078125, 3.7129364013671875, 51.57281494140625, -0.570526123046875, 149.7574462890625, 159.25869750976562, 114.29875183105469, 6.06500244140625, -53.67584228515625, -29.825546264648438, -49.71478271484375, 67.53472900390625, 53.3128662109375, 133.10504150390625, -12.889907836914062, -5.25335693359375, 127.70135498046875, -98.40086364746094, -56.283721923828125, 15.19775390625, 146.72740173339844, 7.293388366699219, -4.8984375, -116.0102767944336, 141.41075134277344, 141.5728759765625, 71.28927612304688, 163.50885009765625, -47.25775146484375, 160.58920288085938, -40.291839599609375, 148.8943328857422], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000468.npy"} {"epoch": 0.9801047120418848, "step": 469, "batch_size": 128, "mean": 59.72987747192383, "std": 92.98295593261719, "min": -219.66940307617188, "p10": -28.645101928710936, "median": 43.72969055175781, "p90": 171.90650939941406, "max": 217.9805908203125, "pos_frac": 0.7578125, "sample": [19.5948486328125, 170.44076538085938, 0.0, 13.706550598144531, 31.50543212890625, 50.91319274902344, -39.31666564941406, 21.04541015625, 0.38262939453125, 67.79222106933594, 125.30726623535156, 36.144500732421875, 135.75418090820312, -219.66940307617188, 19.986785888671875, -2.02276611328125, 192.25180053710938, 152.98699951171875, 22.157684326171875, 132.73593139648438, -9.669418334960938, 96.23513793945312, 6.715904235839844, 0.8456954956054688, -153.30169677734375, 166.94873046875, 18.11505126953125, 154.45648193359375, 172.97622680664062, 29.36968994140625, 139.48797607421875, 70.606201171875, 211.21148681640625, 167.16702270507812, 1.3862190246582031, 108.04481506347656, 50.4000244140625, 85.5570068359375, 165.8783416748047, 71.43084716796875, -1.15155029296875, -27.89263916015625, 30.77374267578125, 3.5650634765625, 141.20188903808594, -2.721923828125, 23.435531616210938, 15.949996948242188, 47.929473876953125, 36.40478515625, 131.2132568359375, 171.44805908203125, 160.0203399658203, 176.65203857421875, -143.13253784179688, 122.2957534790039, 149.64788818359375, 197.87301635742188, 157.1395263671875, -65.7379150390625, 5.174243927001953, -12.99359130859375, 186.75119018554688, -153.33465576171875, -114.44721221923828, 65.2369384765625, 21.167877197265625, -3.51654052734375, -30.400848388671875, 162.7449493408203, 12.663009643554688, 139.98399353027344, 75.95834350585938, 160.1047821044922, -200.55740356445312, 8.290151596069336, 57.58784484863281, 150.3004913330078, -6.93109130859375, 217.9805908203125, 161.6623077392578, 98.58355712890625, 40.614959716796875, 1.305023193359375, 166.35986328125, -14.442474365234375, -198.91998291015625, 5.089349746704102, 89.037353515625, 21.073768615722656, 60.391693115234375, 182.3377685546875, -6.1764984130859375, 164.60427856445312, -35.910552978515625, -19.337249755859375, 0.545562744140625, -0.7854461669921875, -13.987030029296875, 150.71849060058594, -16.843368530273438, 4.065666198730469, -25.93499755859375, 202.37844848632812, 8.152250289916992, -59.8704833984375, 170.06378173828125, 111.6604995727539, 145.47723388671875, 182.95538330078125, 146.0851593017578, 8.13726806640625, 167.19302368164062, 46.84442138671875, -7.690639495849609, 64.1072998046875, 33.81964874267578, 153.2562713623047, 131.38705444335938, 29.08045196533203, 176.0194091796875, 173.96112060546875, 137.59764099121094, -15.113189697265625, 75.29380798339844, 34.48809814453125, 198.29519653320312, -34.442718505859375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000469.npy"} {"epoch": 0.9821989528795811, "step": 470, "batch_size": 128, "mean": 66.35189056396484, "std": 90.85882568359375, "min": -167.71636962890625, "p10": -51.8224754333496, "median": 58.4384765625, "p90": 179.5221435546875, "max": 221.35401916503906, "pos_frac": 0.8046875, "sample": [-18.5029296875, 20.93084716796875, 173.95013427734375, 69.64532470703125, -4.0852813720703125, 65.85086059570312, 6.06170654296875, 151.99917602539062, 164.6741943359375, 23.93182373046875, -31.1240234375, 4.256378173828125, 0.5901336669921875, 157.84481811523438, 144.5601806640625, 144.29855346679688, 171.39901733398438, 1.5161056518554688, -12.936874389648438, 50.0423583984375, 124.16852569580078, -81.1268310546875, -98.49386596679688, -126.00494384765625, 207.2315673828125, 135.4146728515625, 75.88604736328125, 16.4278564453125, 158.11322021484375, 164.29763793945312, 210.86883544921875, 132.6197509765625, 4.446052551269531, -2.13623046875, 116.5979995727539, -84.62713623046875, 141.53945922851562, 143.36532592773438, 27.12139892578125, -113.95606994628906, 11.27154541015625, 200.16583251953125, 131.07440185546875, 19.1527099609375, 9.962554931640625, 152.76229858398438, 189.491943359375, 43.498046875, 54.26712417602539, 114.38104248046875, 12.006362915039062, 149.7664794921875, 27.196533203125, 33.4095458984375, -19.42535400390625, 146.0858154296875, 124.76899719238281, 4.320343017578125, 146.71592712402344, 17.74346923828125, 68.16936492919922, 3.9957332611083984, 155.12196350097656, 58.482391357421875, 23.37884521484375, 34.58331298828125, 4.266685485839844, -167.71636962890625, 18.51654052734375, 112.33163452148438, 12.240325927734375, 134.39315795898438, 124.74417114257812, -31.968368530273438, -118.1341552734375, -55.960487365722656, -69.38165283203125, 129.997802734375, 173.8439178466797, 150.72073364257812, 36.56976318359375, 57.64739990234375, 158.01132202148438, 201.71197509765625, 106.76228332519531, -20.938011169433594, 7.6333770751953125, 185.44793701171875, 184.99606323242188, 185.22805786132812, -25.29157257080078, 155.47360229492188, 133.80975341796875, 147.80545043945312, -1.5548324584960938, 58.394561767578125, 12.386543273925781, 130.8794403076172, 99.32342529296875, 3.76025390625, 170.31857299804688, 177.17617797851562, 124.26443481445312, 221.35401916503906, 9.0465087890625, 113.1168212890625, 1.0768280029296875, 7.689483642578125, -15.05609130859375, -151.14068603515625, 28.924713134765625, -69.08415222167969, 194.03817749023438, -50.049041748046875, 105.7040786743164, 202.9990234375, 4.9879150390625, 54.964508056640625, 73.0379638671875, 63.283599853515625, 193.2349853515625, -114.71127319335938, 60.65240478515625, 35.921417236328125, 194.55490112304688, 164.71804809570312, 170.78521728515625, -63.720428466796875], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000470.npy"} {"epoch": 0.9842931937172775, "step": 471, "batch_size": 128, "mean": 46.79279708862305, "std": 84.24169921875, "min": -160.46202087402344, "p10": -35.89033470153808, "median": 30.46747589111328, "p90": 166.22205505371093, "max": 235.181640625, "pos_frac": 0.6796875, "sample": [-102.56761169433594, 154.17031860351562, 30.553436279296875, 73.47579956054688, 126.73677062988281, 204.020751953125, 151.85658264160156, -121.77685546875, -11.793855667114258, 147.6343231201172, -49.84294128417969, -19.884521484375, -5.22393798828125, 83.9716796875, -1.3221588134765625, 46.92291259765625, 19.956497192382812, 50.31427001953125, -23.93553924560547, -25.58160400390625, 25.946990966796875, 26.500579833984375, 19.030609130859375, 44.192352294921875, -52.043785095214844, -7.701805114746094, 189.48382568359375, -0.4471435546875, -26.036407470703125, 35.59271240234375, 124.64657592773438, 48.172760009765625, 61.61322021484375, 10.80157470703125, -4.908843994140625, 59.168975830078125, 114.70457458496094, 118.12141418457031, 30.739974975585938, 16.568511962890625, 24.7921142578125, -4.46142578125, 159.46408081054688, 168.93438720703125, -37.61003494262695, 90.08015441894531, 26.237045288085938, 198.4450225830078, 141.95065307617188, -7.660514831542969, 5.5600433349609375, 10.76318359375, 163.52310180664062, 114.41290283203125, -115.80746459960938, 135.9300537109375, 122.59596252441406, 3.1767406463623047, -1.7315673828125, 31.573509216308594, 209.18328857421875, 235.181640625, 14.064765930175781, -4.127685546875, -33.325565338134766, -6.676177978515625, -6.2471160888671875, 7.7173004150390625, -160.46202087402344, 25.201416015625, 111.97482299804688, 30.381515502929688, 144.22178649902344, 46.310760498046875, -134.63841247558594, 33.620361328125, 9.54376220703125, 83.65776824951172, 48.52928924560547, -33.9263916015625, 97.897705078125, -19.10003662109375, -2.05218505859375, -9.506156921386719, 54.5732421875, 131.20062255859375, 9.459293365478516, 4.533866882324219, 125.91018676757812, -2.703125, 133.72894287109375, 152.07977294921875, -7.9737091064453125, 54.82438659667969, 16.329910278320312, 165.96981811523438, 172.01226806640625, -63.96477508544922, -9.050537109375, -25.453842163085938, 32.21843338012695, 17.279930114746094, 10.157684326171875, 86.5391845703125, 141.79505920410156, 168.0092315673828, 43.63079833984375, 221.59536743164062, 46.55012512207031, 6.974090576171875, 85.22607421875, 148.39837646484375, 178.5535888671875, -120.6079330444336, 166.81060791015625, 42.31623840332031, 120.83749389648438, 0.0, 46.467864990234375, -88.19125366210938, 182.5235595703125, 152.21759033203125, 179.23486328125, -15.50042724609375, 14.56121826171875, -35.1533203125, -105.68363952636719, -128.18243408203125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000471.npy"} {"epoch": 0.9863874345549738, "step": 472, "batch_size": 128, "mean": 63.051971435546875, "std": 88.46087646484375, "min": -152.13409423828125, "p10": -33.48979797363281, "median": 45.450927734375, "p90": 171.51930236816406, "max": 317.200927734375, "pos_frac": 0.7421875, "sample": [317.200927734375, -8.8670654296875, 0.7995204925537109, -67.73027038574219, 4.91047477722168, 1.943328857421875, -2.4516830444335938, 140.56649780273438, -1.19805908203125, 44.304412841796875, 98.78704833984375, 91.00132751464844, 169.86325073242188, 55.323974609375, -50.71941375732422, 148.70513916015625, -15.214698791503906, 8.32073974609375, -34.379302978515625, 75.20603942871094, 2.58245849609375, 29.93414306640625, 122.06622314453125, 44.91607666015625, -15.511665344238281, 180.54837036132812, 132.35418701171875, -129.30322265625, 3.7364540100097656, 148.05624389648438, 45.686370849609375, -62.27557373046875, 20.2235107421875, 204.28607177734375, 10.755035400390625, -28.6046142578125, 67.24613952636719, 162.7257080078125, -35.34075927734375, 4.714439392089844, 23.0108642578125, 117.7698974609375, 114.43572998046875, 66.59869384765625, -27.915374755859375, 19.464950561523438, 15.8077392578125, -4.011077880859375, 23.22857666015625, 14.290863037109375, 139.11563110351562, 56.28460693359375, 171.34091186523438, -98.02862548828125, 144.331298828125, 143.52081298828125, 164.30044555664062, 128.53305053710938, 36.13700866699219, 28.126220703125, -1.9590911865234375, 96.49627685546875, 3.216217041015625, -1.5819587707519531, 141.8133544921875, 171.0545654296875, 119.6140365600586, 82.17535400390625, 118.81356811523438, -2.40362548828125, 59.3002815246582, 34.160118103027344, 26.640960693359375, 162.52369689941406, 135.1945037841797, 201.99362182617188, 171.935546875, -9.219741821289062, 165.3044891357422, 23.90070343017578, -17.924774169921875, 231.0855712890625, 159.96620178222656, 203.2999267578125, 27.61468505859375, 204.81503295898438, 197.3212890625, 169.15225219726562, 143.76898193359375, -152.13409423828125, -17.224639892578125, -90.44921875, 127.2357177734375, 89.14447021484375, 0.7532958984375, 129.88665771484375, 140.87753295898438, 10.632125854492188, -33.10858154296875, 116.23345947265625, 161.26397705078125, -12.275421142578125, 168.55865478515625, 131.65203857421875, 52.265113830566406, 72.22354125976562, -3.3748779296875, 12.976226806640625, -3.2938804626464844, 3.37432861328125, 187.51602172851562, 176.2434844970703, 5.752176284790039, 18.857696533203125, 45.215484619140625, -121.29643249511719, -55.96528625488281, 234.3795166015625, -15.200103759765625, 48.19061279296875, 179.7401123046875, 99.54144287109375, -125.11395263671875, 106.03579711914062, 129.85150146484375, 132.29183959960938, -46.16609191894531, -14.01956558227539], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000472.npy"} {"epoch": 0.9884816753926702, "step": 473, "batch_size": 128, "mean": 62.650901794433594, "std": 88.21563720703125, "min": -179.16212463378906, "p10": -41.06092071533203, "median": 46.78654479980469, "p90": 177.53214111328126, "max": 214.00424194335938, "pos_frac": 0.75, "sample": [10.957183837890625, 177.91204833984375, 47.11799621582031, -2.8848819732666016, 129.2112579345703, 6.066864013671875, 19.56048583984375, 18.91082763671875, 58.28192138671875, -12.205230712890625, 103.96775817871094, 44.45702362060547, -82.2869873046875, 42.06896209716797, -110.41461944580078, 53.42995071411133, 64.64862060546875, 9.976608276367188, 105.16555786132812, 148.98870849609375, 109.76431274414062, 142.70640563964844, 189.75790405273438, 214.00424194335938, 64.52337646484375, -116.92486572265625, -15.15325927734375, 125.38822937011719, 98.3683090209961, 21.170791625976562, 180.763671875, 174.0849609375, 154.54153442382812, 36.2359619140625, -15.036903381347656, -40.82940673828125, 47.500762939453125, -105.9435043334961, 202.23690795898438, -12.373367309570312, 45.452552795410156, 141.32861328125, 39.068267822265625, -19.537002563476562, -14.030349731445312, -41.60111999511719, -45.0040283203125, 183.3095703125, 52.610626220703125, 25.475830078125, 28.819534301757812, 15.793319702148438, -0.10560798645019531, 188.83212280273438, -16.930419921875, 177.0595703125, -18.042694091796875, 137.40386962890625, 194.5665283203125, -165.35812377929688, 14.612457275390625, 183.19308471679688, 183.38568115234375, 3.0753936767578125, 18.346206665039062, -63.746337890625, 210.61968994140625, 169.82595825195312, 107.46994018554688, 188.88311767578125, 70.2891845703125, 136.80636596679688, 110.6221923828125, 17.090530395507812, 4.207225799560547, 150.01388549804688, 144.91390991210938, 17.639137268066406, -64.36196899414062, 163.34783935546875, -3.7016067504882812, -32.356163024902344, 168.056396484375, 122.02607727050781, 138.77438354492188, 172.46913146972656, -179.16212463378906, -118.52812957763672, 151.0462646484375, -16.76812744140625, 86.04150390625, 133.01950073242188, 4.216758728027344, 46.45509338378906, -0.2703857421875, -61.24139404296875, 35.27687454223633, 18.29376220703125, 25.789352416992188, -7.3251495361328125, 177.36932373046875, 142.12335205078125, 136.56800842285156, 0.0, 149.7247314453125, 97.112548828125, -75.62289428710938, 139.16795349121094, 27.940567016601562, 133.35617065429688, 146.24609375, 17.62889862060547, 0.8977546691894531, 171.47177124023438, 36.29022216796875, -4.7371826171875, 2.455810546875, -18.000839233398438, 28.344146728515625, 46.276824951171875, 156.67807006835938, 96.0330810546875, 160.76718139648438, 61.326622009277344, 206.535400390625, 138.2052001953125, 142.68507385253906, 57.2991943359375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000473.npy"} {"epoch": 0.9905759162303664, "step": 474, "batch_size": 128, "mean": 46.212501525878906, "std": 82.97400665283203, "min": -174.45849609375, "p10": -42.61291160583495, "median": 35.921817779541016, "p90": 154.29430847167967, "max": 255.2861328125, "pos_frac": 0.7578125, "sample": [-31.204177856445312, 0.741455078125, 255.2861328125, -3.482574462890625, -9.926956176757812, -48.37701416015625, 136.8149871826172, 183.13510131835938, 75.13038635253906, 150.69857788085938, 7.284759521484375, 36.34173583984375, -6.9296875, -122.17041015625, 7.533359527587891, -108.92549133300781, 6.10186767578125, 4.624168395996094, 70.97968292236328, 35.50189971923828, 98.53546905517578, 170.41531372070312, 4.9730224609375, -66.71383666992188, 138.07479858398438, -40.142581939697266, -92.6044921875, 52.080718994140625, -17.84575653076172, 54.38714599609375, 135.86618041992188, 47.6761474609375, 3.4049835205078125, 136.35350036621094, 51.38427734375, 130.6661376953125, 220.18218994140625, 21.996898651123047, 2.3259658813476562, 181.947265625, 129.228271484375, 170.30743408203125, 181.09967041015625, 169.5784149169922, 62.6298828125, 3.893817901611328, 0.0, 5.0807037353515625, 22.029083251953125, 10.538841247558594, -1.6328659057617188, -11.1583251953125, -103.74937438964844, 18.111465454101562, 30.810497283935547, 116.99138641357422, 18.911705017089844, -56.18335723876953, 160.97283935546875, 139.6025390625, 1.0199432373046875, -8.473907470703125, -7.4194183349609375, 138.05294799804688, 138.69371032714844, 23.090789794921875, 152.78265380859375, 68.53936767578125, -36.2564697265625, 12.398834228515625, 113.00082397460938, 54.77341079711914, -137.64044189453125, 37.16165542602539, 44.319915771484375, -28.19366455078125, 180.51687622070312, 126.51893615722656, 88.63453674316406, -6.74749755859375, -29.3824462890625, 22.339019775390625, 0.3190193176269531, 10.890901565551758, -0.4847373962402344, -174.45849609375, -90.3121566772461, 119.67576599121094, -123.81501770019531, 144.8543243408203, 65.98475646972656, -171.66842651367188, -1.909027099609375, 2.158203125, 128.95997619628906, 9.391036987304688, 41.37876892089844, -14.35986328125, 31.494293212890625, 122.9471435546875, 79.95230102539062, 157.82150268554688, 149.27801513671875, 69.62527465820312, 203.3367919921875, 83.94438171386719, 135.69729614257812, 18.098480224609375, 50.177978515625, 21.3243408203125, 72.76324462890625, 36.83927917480469, 116.8250732421875, 66.69393920898438, 38.00201416015625, 50.49066162109375, 144.9800262451172, 119.00628662109375, 25.492538452148438, 38.82611083984375, 47.056182861328125, 166.8232879638672, 28.9722900390625, -108.40972900390625, 27.718292236328125, 5.373779296875, 30.5745849609375, 49.956298828125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000474.npy"} {"epoch": 0.9926701570680628, "step": 475, "batch_size": 128, "mean": 50.49064636230469, "std": 83.92864990234375, "min": -188.98956298828125, "p10": -51.41510162353515, "median": 32.64087677001953, "p90": 155.35251159667968, "max": 240.33505249023438, "pos_frac": 0.7109375, "sample": [185.586181640625, 153.6234130859375, -56.67158508300781, 152.38853454589844, 117.48395538330078, 15.75567626953125, -188.98956298828125, 204.07562255859375, -73.15078735351562, -49.162322998046875, -4.8653411865234375, -27.635589599609375, -66.8951416015625, -40.64350128173828, 6.613525390625, 121.69436645507812, 106.50630187988281, 152.6983184814453, 156.67660522460938, -11.200557708740234, 184.14120483398438, -15.3919677734375, 82.23016357421875, 72.2713623046875, 169.37669372558594, 153.9671173095703, 141.28515625, -9.430267333984375, 219.69512939453125, -2.004404067993164, 95.89158630371094, 84.88728332519531, -76.341064453125, 153.08580017089844, 40.17604064941406, 20.31244659423828, -74.6791763305664, 28.944808959960938, 1.0321178436279297, 156.49661254882812, 130.90130615234375, 30.76415252685547, -131.02215576171875, 93.61521911621094, 84.51527404785156, -16.052734375, 119.26203155517578, -11.333251953125, 148.14422607421875, -25.909149169921875, 14.844249725341797, -87.49138641357422, 154.8621826171875, 151.53872680664062, -101.98699951171875, 141.06744384765625, 179.40115356445312, -2.1254024505615234, 20.80129051208496, 35.560150146484375, 148.23619079589844, 56.01170349121094, 7.389930725097656, -24.51532745361328, 25.7432861328125, 58.89678955078125, 20.391311645507812, 44.45098876953125, 9.673416137695312, 29.03131103515625, 14.6728515625, 161.38912963867188, 85.14521789550781, -25.76873779296875, 122.400390625, -37.396942138671875, 1.2562065124511719, 240.33505249023438, 36.09033203125, -3.118499755859375, 29.871261596679688, 191.00006103515625, 148.02734375, -20.76519775390625, 147.69009399414062, -41.80361557006836, 58.09360122680664, 10.491943359375, 52.45587158203125, 31.075759887695312, -1.7311172485351562, 39.75551986694336, 109.45394897460938, 129.3543701171875, -1.824310302734375, 154.35784912109375, 140.91542053222656, 160.23016357421875, 31.735733032226562, 88.49551391601562, 11.876365661621094, -96.2545166015625, 33.669403076171875, 0.0601654052734375, -7.26727294921875, -31.0003662109375, -89.43511962890625, 26.77197265625, 126.81840515136719, 67.38516235351562, 102.71977233886719, 109.53179931640625, 0.2916717529296875, 13.208282470703125, 141.99472045898438, -25.4969482421875, 72.22511291503906, -65.579833984375, -82.7298583984375, 55.828948974609375, 143.74154663085938, 33.5460205078125, 19.09020233154297, 192.66043090820312, 70.72308349609375, 11.37091064453125, -13.870758056640625, 4.53753662109375], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000475.npy"} {"epoch": 0.9947643979057592, "step": 476, "batch_size": 128, "mean": 57.88644790649414, "std": 87.15155029296875, "min": -158.59657287597656, "p10": -36.62553253173828, "median": 34.81964111328125, "p90": 170.40921173095703, "max": 225.74356079101562, "pos_frac": 0.7109375, "sample": [-89.35830688476562, -120.57662963867188, -3.8664093017578125, -36.8277587890625, 20.3057861328125, 137.81930541992188, -1.9541473388671875, -8.149860382080078, 17.001129150390625, 96.44007873535156, 130.42433166503906, 55.43792724609375, 163.92076110839844, 189.647705078125, 127.7612533569336, 203.266357421875, -29.053955078125, 178.9924774169922, 142.23094177246094, -4.286994934082031, -6.011024475097656, 35.46124267578125, 19.10614013671875, 33.992828369140625, 128.55050659179688, -19.524269104003906, 159.68499755859375, -2.242778778076172, -134.31838989257812, 184.82470703125, 17.24090576171875, 23.634979248046875, -36.53886413574219, 181.7476806640625, 160.48892211914062, 16.03521728515625, 173.37789916992188, -78.77640533447266, 27.18707275390625, -16.83544921875, -1.08251953125, -154.27890014648438, 127.89907836914062, 19.525009155273438, 102.4364013671875, 36.370086669921875, 135.36038208007812, 187.36807250976562, 29.50928497314453, -66.85525512695312, 104.67420959472656, 164.054443359375, 10.1304931640625, 10.524085998535156, -1.4584197998046875, 192.44515991210938, 12.148101806640625, -17.11920166015625, 166.6267852783203, 29.645172119140625, 148.16064453125, 104.94072723388672, -58.8292236328125, 168.40237426757812, 26.84979248046875, 182.9320068359375, 165.92034912109375, 10.2889404296875, 138.33499145507812, 63.23095703125, 28.3809814453125, -38.689666748046875, 211.34263610839844, 39.26025390625, 74.65819549560547, -0.9745407104492188, 139.79348754882812, -129.04281616210938, 150.95864868164062, 63.85418701171875, 145.58119201660156, 136.91748046875, 136.0405731201172, 0.44677734375, 10.560348510742188, 4.3966827392578125, -4.206520080566406, 135.7757568359375, -158.59657287597656, 117.93378448486328, -10.759368896484375, 101.223876953125, 176.37091064453125, 2.8944244384765625, 41.58833312988281, -18.978927612304688, -15.58721923828125, 65.09452056884766, 1.21783447265625, 140.2194366455078, -19.05419921875, 34.17803955078125, -48.0474853515625, 225.74356079101562, 18.380775451660156, 17.810470581054688, -10.334808349609375, 58.697906494140625, 104.57212829589844, 105.18325805664062, 75.9462890625, 152.42929077148438, 142.605712890625, 14.481231689453125, 5.236541748046875, -19.98095703125, 102.4647216796875, -19.638824462890625, -6.7124176025390625, 106.8653564453125, 164.42276000976562, 132.96913146972656, -90.59979248046875, 169.1369171142578, -25.59088134765625, 174.719970703125, 141.73275756835938, 79.7589111328125], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000476.npy"} {"epoch": 0.9968586387434555, "step": 477, "batch_size": 128, "mean": 61.34270477294922, "std": 93.09144592285156, "min": -189.37411499023438, "p10": -48.5469970703125, "median": 53.601234436035156, "p90": 170.80095520019532, "max": 227.63836669921875, "pos_frac": 0.703125, "sample": [-99.6036376953125, 69.3853759765625, 147.3457489013672, -108.0699462890625, 159.50732421875, 37.098388671875, -94.1456298828125, 147.31358337402344, -28.074569702148438, 112.13954162597656, -8.26705551147461, 174.66876220703125, -0.3417816162109375, 154.47354125976562, 64.69100952148438, -11.147247314453125, 199.80731201171875, 97.32821655273438, -189.37411499023438, -50.27606201171875, -2.56134033203125, -37.592071533203125, 54.46656799316406, 172.15103149414062, 134.6440887451172, 180.35400390625, 132.40567016601562, -8.488075256347656, 14.698028564453125, 32.53057861328125, 145.4540557861328, 167.47198486328125, 13.2484130859375, 211.50350952148438, 123.86883544921875, 45.459625244140625, 206.3153076171875, 36.882179260253906, -9.665489196777344, 0.0, 17.58258056640625, 153.83761596679688, 15.44122314453125, -29.938720703125, 116.29186248779297, -12.808746337890625, 15.3917236328125, 144.62353515625, 155.144287109375, -74.37957763671875, -29.611167907714844, 1.963958740234375, -29.312606811523438, 160.47262573242188, 26.15625, 14.0328369140625, 144.12722778320312, 29.385223388671875, 149.8319091796875, 64.15293884277344, 138.7811279296875, 163.90362548828125, 227.63836669921875, -7.91827392578125, 78.7109375, 123.30817413330078, 186.77471923828125, -91.93634033203125, 84.337890625, -42.618255615234375, 119.29422760009766, 156.74302673339844, 119.49270629882812, 140.50421142578125, 94.11212158203125, 170.22235107421875, -39.18572998046875, 13.961044311523438, 1.4691619873046875, 151.66531372070312, -24.532745361328125, 162.41322326660156, 64.23858642578125, -7.784721374511719, -56.252105712890625, -154.52218627929688, 158.218017578125, 93.73104858398438, 38.100467681884766, 119.90325927734375, -3.050570487976074, 9.89178466796875, -10.027633666992188, 195.50283813476562, -5.86151123046875, 141.88983154296875, -116.76295471191406, 126.49313354492188, 32.34527587890625, 154.38723754882812, -62.87095642089844, 209.00274658203125, -47.80596923828125, -162.05380249023438, 18.1905517578125, -20.846527099609375, -45.455810546875, -5.983306884765625, 192.25479125976562, 77.39804077148438, 41.738494873046875, 218.27587890625, 166.68765258789062, 179.3626708984375, 1.2354278564453125, 159.53036499023438, 162.95126342773438, -65.0588150024414, 165.7292022705078, 16.03466796875, 54.586669921875, 134.21978759765625, 142.42640686035156, 18.456031799316406, 3.6445236206054688, 111.60774993896484, 52.73590087890625, 8.301345825195312], "npy": "/workspace/dynamic-dpo-v4/outputs/qwen3-8b-base-new-dpo-ultrafeedback-4xh200-batch-128-q_t-0.45-s_star-0.35-20260430-143919/margin_logs/step_0000477.npy"}